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 affiliates.
   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/syscore_ops.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
  56 #include "coalesced_mmio.h"
  57 #include "async_pf.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_RAW_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 bool kvm_rebooting;
  93 EXPORT_SYMBOL_GPL(kvm_rebooting);
  94
  95 static bool largepages_enabled = true;
  96
  97 static struct page *hwpoison_page;
  98 static pfn_t hwpoison_pfn;
  99
 100 static struct page *fault_page;
 101 static pfn_t fault_pfn;
 102
 103 inline int kvm_is_mmio_pfn(pfn_t pfn)
 104 {
 105         if (pfn_valid(pfn)) {
 106                 int reserved;
 107                 struct page *tail = pfn_to_page(pfn);
 108                 struct page *head = compound_trans_head(tail);
 109                 reserved = PageReserved(head);
 110                 if (head != tail) {
 111                         /*
 112                          * "head" is not a dangling pointer
 113                          * (compound_trans_head takes care of that)
 114                          * but the hugepage may have been splitted
 115                          * from under us (and we may not hold a
 116                          * reference count on the head page so it can
 117                          * be reused before we run PageReferenced), so
 118                          * we've to check PageTail before returning
 119                          * what we just read.
 120                          */
 121                         smp_rmb();
 122                         if (PageTail(tail))
 123                                 return reserved;
 124                 }
 125                 return PageReserved(tail);
 126         }
 127
 128         return true;
 129 }
 130
 131 /*
 132  * Switches to specified vcpu, until a matching vcpu_put()
 133  */
 134 void vcpu_load(struct kvm_vcpu *vcpu)
 135 {
 136         int cpu;
 137
 138         mutex_lock(&vcpu->mutex);
 139         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 140                 /* The thread running this VCPU changed. */
 141                 struct pid *oldpid = vcpu->pid;
 142                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 143                 rcu_assign_pointer(vcpu->pid, newpid);
 144                 synchronize_rcu();
 145                 put_pid(oldpid);
 146         }
 147         cpu = get_cpu();
 148         preempt_notifier_register(&vcpu->preempt_notifier);
 149         kvm_arch_vcpu_load(vcpu, cpu);
 150         put_cpu();
 151 }
 152
 153 void vcpu_put(struct kvm_vcpu *vcpu)
 154 {
 155         preempt_disable();
 156         kvm_arch_vcpu_put(vcpu);
 157         preempt_notifier_unregister(&vcpu->preempt_notifier);
 158         preempt_enable();
 159         mutex_unlock(&vcpu->mutex);
 160 }
 161
 162 static void ack_flush(void *_completed)
 163 {
 164 }
 165
 166 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 167 {
 168         int i, cpu, me;
 169         cpumask_var_t cpus;
 170         bool called = true;
 171         struct kvm_vcpu *vcpu;
 172
 173         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 174
 175         me = get_cpu();
 176         kvm_for_each_vcpu(i, vcpu, kvm) {
 177                 kvm_make_request(req, vcpu);
 178                 cpu = vcpu->cpu;
 179
 180                 /* Set ->requests bit before we read ->mode */
 181                 smp_mb();
 182
 183                 if (cpus != NULL && cpu != -1 && cpu != me &&
 184                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 185                         cpumask_set_cpu(cpu, cpus);
 186         }
 187         if (unlikely(cpus == NULL))
 188                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 189         else if (!cpumask_empty(cpus))
 190                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 191         else
 192                 called = false;
 193         put_cpu();
 194         free_cpumask_var(cpus);
 195         return called;
 196 }
 197
 198 void kvm_flush_remote_tlbs(struct kvm *kvm)
 199 {
 200         int dirty_count = kvm->tlbs_dirty;
 201
 202         smp_mb();
 203         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 204                 ++kvm->stat.remote_tlb_flush;
 205         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 206 }
 207
 208 void kvm_reload_remote_mmus(struct kvm *kvm)
 209 {
 210         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 211 }
 212
 213 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 214 {
 215         struct page *page;
 216         int r;
 217
 218         mutex_init(&vcpu->mutex);
 219         vcpu->cpu = -1;
 220         vcpu->kvm = kvm;
 221         vcpu->vcpu_id = id;
 222         vcpu->pid = NULL;
 223         init_waitqueue_head(&vcpu->wq);
 224         kvm_async_pf_vcpu_init(vcpu);
 225
 226         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 227         if (!page) {
 228                 r = -ENOMEM;
 229                 goto fail;
 230         }
 231         vcpu->run = page_address(page);
 232
 233         r = kvm_arch_vcpu_init(vcpu);
 234         if (r < 0)
 235                 goto fail_free_run;
 236         return 0;
 237
 238 fail_free_run:
 239         free_page((unsigned long)vcpu->run);
 240 fail:
 241         return r;
 242 }
 243 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 244
 245 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 246 {
 247         put_pid(vcpu->pid);
 248         kvm_arch_vcpu_uninit(vcpu);
 249         free_page((unsigned long)vcpu->run);
 250 }
 251 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 252
 253 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 254 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 255 {
 256         return container_of(mn, struct kvm, mmu_notifier);
 257 }
 258
 259 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 260                                              struct mm_struct *mm,
 261                                              unsigned long address)
 262 {
 263         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 264         int need_tlb_flush, idx;
 265
 266         /*
 267          * When ->invalidate_page runs, the linux pte has been zapped
 268          * already but the page is still allocated until
 269          * ->invalidate_page returns. So if we increase the sequence
 270          * here the kvm page fault will notice if the spte can't be
 271          * established because the page is going to be freed. If
 272          * instead the kvm page fault establishes the spte before
 273          * ->invalidate_page runs, kvm_unmap_hva will release it
 274          * before returning.
 275          *
 276          * The sequence increase only need to be seen at spin_unlock
 277          * time, and not at spin_lock time.
 278          *
 279          * Increasing the sequence after the spin_unlock would be
 280          * unsafe because the kvm page fault could then establish the
 281          * pte after kvm_unmap_hva returned, without noticing the page
 282          * is going to be freed.
 283          */
 284         idx = srcu_read_lock(&kvm->srcu);
 285         spin_lock(&kvm->mmu_lock);
 286         kvm->mmu_notifier_seq++;
 287         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 288         spin_unlock(&kvm->mmu_lock);
 289         srcu_read_unlock(&kvm->srcu, idx);
 290
 291         /* we've to flush the tlb before the pages can be freed */
 292         if (need_tlb_flush)
 293                 kvm_flush_remote_tlbs(kvm);
 294
 295 }
 296
 297 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 298                                         struct mm_struct *mm,
 299                                         unsigned long address,
 300                                         pte_t pte)
 301 {
 302         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 303         int idx;
 304
 305         idx = srcu_read_lock(&kvm->srcu);
 306         spin_lock(&kvm->mmu_lock);
 307         kvm->mmu_notifier_seq++;
 308         kvm_set_spte_hva(kvm, address, pte);
 309         spin_unlock(&kvm->mmu_lock);
 310         srcu_read_unlock(&kvm->srcu, idx);
 311 }
 312
 313 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 314                                                     struct mm_struct *mm,
 315                                                     unsigned long start,
 316                                                     unsigned long end)
 317 {
 318         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 319         int need_tlb_flush = 0, idx;
 320
 321         idx = srcu_read_lock(&kvm->srcu);
 322         spin_lock(&kvm->mmu_lock);
 323         /*
 324          * The count increase must become visible at unlock time as no
 325          * spte can be established without taking the mmu_lock and
 326          * count is also read inside the mmu_lock critical section.
 327          */
 328         kvm->mmu_notifier_count++;
 329         for (; start < end; start += PAGE_SIZE)
 330                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 331         need_tlb_flush |= kvm->tlbs_dirty;
 332         spin_unlock(&kvm->mmu_lock);
 333         srcu_read_unlock(&kvm->srcu, idx);
 334
 335         /* we've to flush the tlb before the pages can be freed */
 336         if (need_tlb_flush)
 337                 kvm_flush_remote_tlbs(kvm);
 338 }
 339
 340 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 341                                                   struct mm_struct *mm,
 342                                                   unsigned long start,
 343                                                   unsigned long end)
 344 {
 345         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 346
 347         spin_lock(&kvm->mmu_lock);
 348         /*
 349          * This sequence increase will notify the kvm page fault that
 350          * the page that is going to be mapped in the spte could have
 351          * been freed.
 352          */
 353         kvm->mmu_notifier_seq++;
 354         /*
 355          * The above sequence increase must be visible before the
 356          * below count decrease but both values are read by the kvm
 357          * page fault under mmu_lock spinlock so we don't need to add
 358          * a smb_wmb() here in between the two.
 359          */
 360         kvm->mmu_notifier_count--;
 361         spin_unlock(&kvm->mmu_lock);
 362
 363         BUG_ON(kvm->mmu_notifier_count < 0);
 364 }
 365
 366 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 367                                               struct mm_struct *mm,
 368                                               unsigned long address)
 369 {
 370         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 371         int young, idx;
 372
 373         idx = srcu_read_lock(&kvm->srcu);
 374         spin_lock(&kvm->mmu_lock);
 375         young = kvm_age_hva(kvm, address);
 376         spin_unlock(&kvm->mmu_lock);
 377         srcu_read_unlock(&kvm->srcu, idx);
 378
 379         if (young)
 380                 kvm_flush_remote_tlbs(kvm);
 381
 382         return young;
 383 }
 384
 385 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 386                                        struct mm_struct *mm,
 387                                        unsigned long address)
 388 {
 389         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 390         int young, idx;
 391
 392         idx = srcu_read_lock(&kvm->srcu);
 393         spin_lock(&kvm->mmu_lock);
 394         young = kvm_test_age_hva(kvm, address);
 395         spin_unlock(&kvm->mmu_lock);
 396         srcu_read_unlock(&kvm->srcu, idx);
 397
 398         return young;
 399 }
 400
 401 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 402                                      struct mm_struct *mm)
 403 {
 404         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 405         int idx;
 406
 407         idx = srcu_read_lock(&kvm->srcu);
 408         kvm_arch_flush_shadow(kvm);
 409         srcu_read_unlock(&kvm->srcu, idx);
 410 }
 411
 412 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 413         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 414         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 415         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 416         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 417         .test_young             = kvm_mmu_notifier_test_young,
 418         .change_pte             = kvm_mmu_notifier_change_pte,
 419         .release                = kvm_mmu_notifier_release,
 420 };
 421
 422 static int kvm_init_mmu_notifier(struct kvm *kvm)
 423 {
 424         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 425         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 426 }
 427
 428 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 429
 430 static int kvm_init_mmu_notifier(struct kvm *kvm)
 431 {
 432         return 0;
 433 }
 434
 435 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 436
 437 static struct kvm *kvm_create_vm(void)
 438 {
 439         int r, i;
 440         struct kvm *kvm = kvm_arch_alloc_vm();
 441
 442         if (!kvm)
 443                 return ERR_PTR(-ENOMEM);
 444
 445         r = kvm_arch_init_vm(kvm);
 446         if (r)
 447                 goto out_err_nodisable;
 448
 449         r = hardware_enable_all();
 450         if (r)
 451                 goto out_err_nodisable;
 452
 453 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 454         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 455         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 456 #endif
 457
 458         r = -ENOMEM;
 459         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 460         if (!kvm->memslots)
 461                 goto out_err_nosrcu;
 462         if (init_srcu_struct(&kvm->srcu))
 463                 goto out_err_nosrcu;
 464         for (i = 0; i < KVM_NR_BUSES; i++) {
 465                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 466                                         GFP_KERNEL);
 467                 if (!kvm->buses[i])
 468                         goto out_err;
 469         }
 470
 471         spin_lock_init(&kvm->mmu_lock);
 472         kvm->mm = current->mm;
 473         atomic_inc(&kvm->mm->mm_count);
 474         kvm_eventfd_init(kvm);
 475         mutex_init(&kvm->lock);
 476         mutex_init(&kvm->irq_lock);
 477         mutex_init(&kvm->slots_lock);
 478         atomic_set(&kvm->users_count, 1);
 479
 480         r = kvm_init_mmu_notifier(kvm);
 481         if (r)
 482                 goto out_err;
 483
 484         raw_spin_lock(&kvm_lock);
 485         list_add(&kvm->vm_list, &vm_list);
 486         raw_spin_unlock(&kvm_lock);
 487
 488         return kvm;
 489
 490 out_err:
 491         cleanup_srcu_struct(&kvm->srcu);
 492 out_err_nosrcu:
 493         hardware_disable_all();
 494 out_err_nodisable:
 495         for (i = 0; i < KVM_NR_BUSES; i++)
 496                 kfree(kvm->buses[i]);
 497         kfree(kvm->memslots);
 498         kvm_arch_free_vm(kvm);
 499         return ERR_PTR(r);
 500 }
 501
 502 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 503 {
 504         if (!memslot->dirty_bitmap)
 505                 return;
 506
 507         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
 508                 vfree(memslot->dirty_bitmap_head);
 509         else
 510                 kfree(memslot->dirty_bitmap_head);
 511
 512         memslot->dirty_bitmap = NULL;
 513         memslot->dirty_bitmap_head = NULL;
 514 }
 515
 516 /*
 517  * Free any memory in @free but not in @dont.
 518  */
 519 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 520                                   struct kvm_memory_slot *dont)
 521 {
 522         int i;
 523
 524         if (!dont || free->rmap != dont->rmap)
 525                 vfree(free->rmap);
 526
 527         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 528                 kvm_destroy_dirty_bitmap(free);
 529
 530
 531         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 532                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 533                         vfree(free->lpage_info[i]);
 534                         free->lpage_info[i] = NULL;
 535                 }
 536         }
 537
 538         free->npages = 0;
 539         free->rmap = NULL;
 540 }
 541
 542 void kvm_free_physmem(struct kvm *kvm)
 543 {
 544         int i;
 545         struct kvm_memslots *slots = kvm->memslots;
 546
 547         for (i = 0; i < slots->nmemslots; ++i)
 548                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 549
 550         kfree(kvm->memslots);
 551 }
 552
 553 static void kvm_destroy_vm(struct kvm *kvm)
 554 {
 555         int i;
 556         struct mm_struct *mm = kvm->mm;
 557
 558         kvm_arch_sync_events(kvm);
 559         raw_spin_lock(&kvm_lock);
 560         list_del(&kvm->vm_list);
 561         raw_spin_unlock(&kvm_lock);
 562         kvm_free_irq_routing(kvm);
 563         for (i = 0; i < KVM_NR_BUSES; i++)
 564                 kvm_io_bus_destroy(kvm->buses[i]);
 565         kvm_coalesced_mmio_free(kvm);
 566 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 567         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 568 #else
 569         kvm_arch_flush_shadow(kvm);
 570 #endif
 571         kvm_arch_destroy_vm(kvm);
 572         kvm_free_physmem(kvm);
 573         cleanup_srcu_struct(&kvm->srcu);
 574         kvm_arch_free_vm(kvm);
 575         hardware_disable_all();
 576         mmdrop(mm);
 577 }
 578
 579 void kvm_get_kvm(struct kvm *kvm)
 580 {
 581         atomic_inc(&kvm->users_count);
 582 }
 583 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 584
 585 void kvm_put_kvm(struct kvm *kvm)
 586 {
 587         if (atomic_dec_and_test(&kvm->users_count))
 588                 kvm_destroy_vm(kvm);
 589 }
 590 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 591
 592
 593 static int kvm_vm_release(struct inode *inode, struct file *filp)
 594 {
 595         struct kvm *kvm = filp->private_data;
 596
 597         kvm_irqfd_release(kvm);
 598
 599         kvm_put_kvm(kvm);
 600         return 0;
 601 }
 602
 603 #ifndef CONFIG_S390
 604 /*
 605  * Allocation size is twice as large as the actual dirty bitmap size.
 606  * This makes it possible to do double buffering: see x86's
 607  * kvm_vm_ioctl_get_dirty_log().
 608  */
 609 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 610 {
 611         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 612
 613         if (dirty_bytes > PAGE_SIZE)
 614                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
 615         else
 616                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
 617
 618         if (!memslot->dirty_bitmap)
 619                 return -ENOMEM;
 620
 621         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
 622         return 0;
 623 }
 624 #endif /* !CONFIG_S390 */
 625
 626 /*
 627  * Allocate some memory and give it an address in the guest physical address
 628  * space.
 629  *
 630  * Discontiguous memory is allowed, mostly for framebuffers.
 631  *
 632  * Must be called holding mmap_sem for write.
 633  */
 634 int __kvm_set_memory_region(struct kvm *kvm,
 635                             struct kvm_userspace_memory_region *mem,
 636                             int user_alloc)
 637 {
 638         int r;
 639         gfn_t base_gfn;
 640         unsigned long npages;
 641         unsigned long i;
 642         struct kvm_memory_slot *memslot;
 643         struct kvm_memory_slot old, new;
 644         struct kvm_memslots *slots, *old_memslots;
 645
 646         r = -EINVAL;
 647         /* General sanity checks */
 648         if (mem->memory_size & (PAGE_SIZE - 1))
 649                 goto out;
 650         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 651                 goto out;
 652         /* We can read the guest memory with __xxx_user() later on. */
 653         if (user_alloc &&
 654             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 655              !access_ok(VERIFY_WRITE,
 656                         (void __user *)(unsigned long)mem->userspace_addr,
 657                         mem->memory_size)))
 658                 goto out;
 659         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 660                 goto out;
 661         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 662                 goto out;
 663
 664         memslot = &kvm->memslots->memslots[mem->slot];
 665         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 666         npages = mem->memory_size >> PAGE_SHIFT;
 667
 668         r = -EINVAL;
 669         if (npages > KVM_MEM_MAX_NR_PAGES)
 670                 goto out;
 671
 672         if (!npages)
 673                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 674
 675         new = old = *memslot;
 676
 677         new.id = mem->slot;
 678         new.base_gfn = base_gfn;
 679         new.npages = npages;
 680         new.flags = mem->flags;
 681
 682         /* Disallow changing a memory slot's size. */
 683         r = -EINVAL;
 684         if (npages && old.npages && npages != old.npages)
 685                 goto out_free;
 686
 687         /* Check for overlaps */
 688         r = -EEXIST;
 689         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 690                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 691
 692                 if (s == memslot || !s->npages)
 693                         continue;
 694                 if (!((base_gfn + npages <= s->base_gfn) ||
 695                       (base_gfn >= s->base_gfn + s->npages)))
 696                         goto out_free;
 697         }
 698
 699         /* Free page dirty bitmap if unneeded */
 700         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 701                 new.dirty_bitmap = NULL;
 702
 703         r = -ENOMEM;
 704
 705         /* Allocate if a slot is being created */
 706 #ifndef CONFIG_S390
 707         if (npages && !new.rmap) {
 708                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
 709
 710                 if (!new.rmap)
 711                         goto out_free;
 712
 713                 new.user_alloc = user_alloc;
 714                 new.userspace_addr = mem->userspace_addr;
 715         }
 716         if (!npages)
 717                 goto skip_lpage;
 718
 719         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 720                 unsigned long ugfn;
 721                 unsigned long j;
 722                 int lpages;
 723                 int level = i + 2;
 724
 725                 /* Avoid unused variable warning if no large pages */
 726                 (void)level;
 727
 728                 if (new.lpage_info[i])
 729                         continue;
 730
 731                 lpages = 1 + ((base_gfn + npages - 1)
 732                              >> KVM_HPAGE_GFN_SHIFT(level));
 733                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
 734
 735                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
 736
 737                 if (!new.lpage_info[i])
 738                         goto out_free;
 739
 740                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
 741                         new.lpage_info[i][0].write_count = 1;
 742                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
 743                         new.lpage_info[i][lpages - 1].write_count = 1;
 744                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 745                 /*
 746                  * If the gfn and userspace address are not aligned wrt each
 747                  * other, or if explicitly asked to, disable large page
 748                  * support for this slot
 749                  */
 750                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 751                     !largepages_enabled)
 752                         for (j = 0; j < lpages; ++j)
 753                                 new.lpage_info[i][j].write_count = 1;
 754         }
 755
 756 skip_lpage:
 757
 758         /* Allocate page dirty bitmap if needed */
 759         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 760                 if (kvm_create_dirty_bitmap(&new) < 0)
 761                         goto out_free;
 762                 /* destroy any largepage mappings for dirty tracking */
 763         }
 764 #else  /* not defined CONFIG_S390 */
 765         new.user_alloc = user_alloc;
 766         if (user_alloc)
 767                 new.userspace_addr = mem->userspace_addr;
 768 #endif /* not defined CONFIG_S390 */
 769
 770         if (!npages) {
 771                 r = -ENOMEM;
 772                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 773                 if (!slots)
 774                         goto out_free;
 775                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 776                 if (mem->slot >= slots->nmemslots)
 777                         slots->nmemslots = mem->slot + 1;
 778                 slots->generation++;
 779                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 780
 781                 old_memslots = kvm->memslots;
 782                 rcu_assign_pointer(kvm->memslots, slots);
 783                 synchronize_srcu_expedited(&kvm->srcu);
 784                 /* From this point no new shadow pages pointing to a deleted
 785                  * memslot will be created.
 786                  *
 787                  * validation of sp->gfn happens in:
 788                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 789                  *      - kvm_is_visible_gfn (mmu_check_roots)
 790                  */
 791                 kvm_arch_flush_shadow(kvm);
 792                 kfree(old_memslots);
 793         }
 794
 795         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 796         if (r)
 797                 goto out_free;
 798
 799         /* map/unmap the pages in iommu page table */
 800         if (npages) {
 801                 r = kvm_iommu_map_pages(kvm, &new);
 802                 if (r)
 803                         goto out_free;
 804         } else
 805                 kvm_iommu_unmap_pages(kvm, &old);
 806
 807         r = -ENOMEM;
 808         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 809         if (!slots)
 810                 goto out_free;
 811         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 812         if (mem->slot >= slots->nmemslots)
 813                 slots->nmemslots = mem->slot + 1;
 814         slots->generation++;
 815
 816         /* actual memory is freed via old in kvm_free_physmem_slot below */
 817         if (!npages) {
 818                 new.rmap = NULL;
 819                 new.dirty_bitmap = NULL;
 820                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 821                         new.lpage_info[i] = NULL;
 822         }
 823
 824         slots->memslots[mem->slot] = new;
 825         old_memslots = kvm->memslots;
 826         rcu_assign_pointer(kvm->memslots, slots);
 827         synchronize_srcu_expedited(&kvm->srcu);
 828
 829         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 830
 831         kvm_free_physmem_slot(&old, &new);
 832         kfree(old_memslots);
 833
 834         return 0;
 835
 836 out_free:
 837         kvm_free_physmem_slot(&new, &old);
 838 out:
 839         return r;
 840
 841 }
 842 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 843
 844 int kvm_set_memory_region(struct kvm *kvm,
 845                           struct kvm_userspace_memory_region *mem,
 846                           int user_alloc)
 847 {
 848         int r;
 849
 850         mutex_lock(&kvm->slots_lock);
 851         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 852         mutex_unlock(&kvm->slots_lock);
 853         return r;
 854 }
 855 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 856
 857 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 858                                    struct
 859                                    kvm_userspace_memory_region *mem,
 860                                    int user_alloc)
 861 {
 862         if (mem->slot >= KVM_MEMORY_SLOTS)
 863                 return -EINVAL;
 864         return kvm_set_memory_region(kvm, mem, user_alloc);
 865 }
 866
 867 int kvm_get_dirty_log(struct kvm *kvm,
 868                         struct kvm_dirty_log *log, int *is_dirty)
 869 {
 870         struct kvm_memory_slot *memslot;
 871         int r, i;
 872         unsigned long n;
 873         unsigned long any = 0;
 874
 875         r = -EINVAL;
 876         if (log->slot >= KVM_MEMORY_SLOTS)
 877                 goto out;
 878
 879         memslot = &kvm->memslots->memslots[log->slot];
 880         r = -ENOENT;
 881         if (!memslot->dirty_bitmap)
 882                 goto out;
 883
 884         n = kvm_dirty_bitmap_bytes(memslot);
 885
 886         for (i = 0; !any && i < n/sizeof(long); ++i)
 887                 any = memslot->dirty_bitmap[i];
 888
 889         r = -EFAULT;
 890         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 891                 goto out;
 892
 893         if (any)
 894                 *is_dirty = 1;
 895
 896         r = 0;
 897 out:
 898         return r;
 899 }
 900
 901 void kvm_disable_largepages(void)
 902 {
 903         largepages_enabled = false;
 904 }
 905 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 906
 907 int is_error_page(struct page *page)
 908 {
 909         return page == bad_page || page == hwpoison_page || page == fault_page;
 910 }
 911 EXPORT_SYMBOL_GPL(is_error_page);
 912
 913 int is_error_pfn(pfn_t pfn)
 914 {
 915         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 916 }
 917 EXPORT_SYMBOL_GPL(is_error_pfn);
 918
 919 int is_hwpoison_pfn(pfn_t pfn)
 920 {
 921         return pfn == hwpoison_pfn;
 922 }
 923 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 924
 925 int is_fault_pfn(pfn_t pfn)
 926 {
 927         return pfn == fault_pfn;
 928 }
 929 EXPORT_SYMBOL_GPL(is_fault_pfn);
 930
 931 static inline unsigned long bad_hva(void)
 932 {
 933         return PAGE_OFFSET;
 934 }
 935
 936 int kvm_is_error_hva(unsigned long addr)
 937 {
 938         return addr == bad_hva();
 939 }
 940 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 941
 942 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
 943                                                 gfn_t gfn)
 944 {
 945         int i;
 946
 947         for (i = 0; i < slots->nmemslots; ++i) {
 948                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 949
 950                 if (gfn >= memslot->base_gfn
 951                     && gfn < memslot->base_gfn + memslot->npages)
 952                         return memslot;
 953         }
 954         return NULL;
 955 }
 956
 957 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 958 {
 959         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 960 }
 961 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 962
 963 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 964 {
 965         int i;
 966         struct kvm_memslots *slots = kvm_memslots(kvm);
 967
 968         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 969                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 970
 971                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 972                         continue;
 973
 974                 if (gfn >= memslot->base_gfn
 975                     && gfn < memslot->base_gfn + memslot->npages)
 976                         return 1;
 977         }
 978         return 0;
 979 }
 980 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 981
 982 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 983 {
 984         struct vm_area_struct *vma;
 985         unsigned long addr, size;
 986
 987         size = PAGE_SIZE;
 988
 989         addr = gfn_to_hva(kvm, gfn);
 990         if (kvm_is_error_hva(addr))
 991                 return PAGE_SIZE;
 992
 993         down_read(&current->mm->mmap_sem);
 994         vma = find_vma(current->mm, addr);
 995         if (!vma)
 996                 goto out;
 997
 998         size = vma_kernel_pagesize(vma);
 999
1000 out:
1001         up_read(&current->mm->mmap_sem);
1002
1003         return size;
1004 }
1005
1006 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1007                                      gfn_t *nr_pages)
1008 {
1009         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1010                 return bad_hva();
1011
1012         if (nr_pages)
1013                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1014
1015         return gfn_to_hva_memslot(slot, gfn);
1016 }
1017
1018 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1019 {
1020         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1021 }
1022 EXPORT_SYMBOL_GPL(gfn_to_hva);
1023
1024 static pfn_t get_fault_pfn(void)
1025 {
1026         get_page(fault_page);
1027         return fault_pfn;
1028 }
1029
1030 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1031         unsigned long start, int write, struct page **page)
1032 {
1033         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1034
1035         if (write)
1036                 flags |= FOLL_WRITE;
1037
1038         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1039 }
1040
1041 static inline int check_user_page_hwpoison(unsigned long addr)
1042 {
1043         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1044
1045         rc = __get_user_pages(current, current->mm, addr, 1,
1046                               flags, NULL, NULL, NULL);
1047         return rc == -EHWPOISON;
1048 }
1049
1050 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1051                         bool *async, bool write_fault, bool *writable)
1052 {
1053         struct page *page[1];
1054         int npages = 0;
1055         pfn_t pfn;
1056
1057         /* we can do it either atomically or asynchronously, not both */
1058         BUG_ON(atomic && async);
1059
1060         BUG_ON(!write_fault && !writable);
1061
1062         if (writable)
1063                 *writable = true;
1064
1065         if (atomic || async)
1066                 npages = __get_user_pages_fast(addr, 1, 1, page);
1067
1068         if (unlikely(npages != 1) && !atomic) {
1069                 might_sleep();
1070
1071                 if (writable)
1072                         *writable = write_fault;
1073
1074                 if (async) {
1075                         down_read(&current->mm->mmap_sem);
1076                         npages = get_user_page_nowait(current, current->mm,
1077                                                      addr, write_fault, page);
1078                         up_read(&current->mm->mmap_sem);
1079                 } else
1080                         npages = get_user_pages_fast(addr, 1, write_fault,
1081                                                      page);
1082
1083                 /* map read fault as writable if possible */
1084                 if (unlikely(!write_fault) && npages == 1) {
1085                         struct page *wpage[1];
1086
1087                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1088                         if (npages == 1) {
1089                                 *writable = true;
1090                                 put_page(page[0]);
1091                                 page[0] = wpage[0];
1092                         }
1093                         npages = 1;
1094                 }
1095         }
1096
1097         if (unlikely(npages != 1)) {
1098                 struct vm_area_struct *vma;
1099
1100                 if (atomic)
1101                         return get_fault_pfn();
1102
1103                 down_read(&current->mm->mmap_sem);
1104                 if (npages == -EHWPOISON ||
1105                         (!async && check_user_page_hwpoison(addr))) {
1106                         up_read(&current->mm->mmap_sem);
1107                         get_page(hwpoison_page);
1108                         return page_to_pfn(hwpoison_page);
1109                 }
1110
1111                 vma = find_vma_intersection(current->mm, addr, addr+1);
1112
1113                 if (vma == NULL)
1114                         pfn = get_fault_pfn();
1115                 else if ((vma->vm_flags & VM_PFNMAP)) {
1116                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1117                                 vma->vm_pgoff;
1118                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1119                 } else {
1120                         if (async && (vma->vm_flags & VM_WRITE))
1121                                 *async = true;
1122                         pfn = get_fault_pfn();
1123                 }
1124                 up_read(&current->mm->mmap_sem);
1125         } else
1126                 pfn = page_to_pfn(page[0]);
1127
1128         return pfn;
1129 }
1130
1131 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1132 {
1133         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1134 }
1135 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1136
1137 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1138                           bool write_fault, bool *writable)
1139 {
1140         unsigned long addr;
1141
1142         if (async)
1143                 *async = false;
1144
1145         addr = gfn_to_hva(kvm, gfn);
1146         if (kvm_is_error_hva(addr)) {
1147                 get_page(bad_page);
1148                 return page_to_pfn(bad_page);
1149         }
1150
1151         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1152 }
1153
1154 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1155 {
1156         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1157 }
1158 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1159
1160 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1161                        bool write_fault, bool *writable)
1162 {
1163         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1164 }
1165 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1166
1167 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1168 {
1169         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1170 }
1171 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1172
1173 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1174                       bool *writable)
1175 {
1176         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1177 }
1178 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1179
1180 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1181                          struct kvm_memory_slot *slot, gfn_t gfn)
1182 {
1183         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1184         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1185 }
1186
1187 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1188                                                                   int nr_pages)
1189 {
1190         unsigned long addr;
1191         gfn_t entry;
1192
1193         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1194         if (kvm_is_error_hva(addr))
1195                 return -1;
1196
1197         if (entry < nr_pages)
1198                 return 0;
1199
1200         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1201 }
1202 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1203
1204 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1205 {
1206         pfn_t pfn;
1207
1208         pfn = gfn_to_pfn(kvm, gfn);
1209         if (!kvm_is_mmio_pfn(pfn))
1210                 return pfn_to_page(pfn);
1211
1212         WARN_ON(kvm_is_mmio_pfn(pfn));
1213
1214         get_page(bad_page);
1215         return bad_page;
1216 }
1217
1218 EXPORT_SYMBOL_GPL(gfn_to_page);
1219
1220 void kvm_release_page_clean(struct page *page)
1221 {
1222         kvm_release_pfn_clean(page_to_pfn(page));
1223 }
1224 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1225
1226 void kvm_release_pfn_clean(pfn_t pfn)
1227 {
1228         if (!kvm_is_mmio_pfn(pfn))
1229                 put_page(pfn_to_page(pfn));
1230 }
1231 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1232
1233 void kvm_release_page_dirty(struct page *page)
1234 {
1235         kvm_release_pfn_dirty(page_to_pfn(page));
1236 }
1237 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1238
1239 void kvm_release_pfn_dirty(pfn_t pfn)
1240 {
1241         kvm_set_pfn_dirty(pfn);
1242         kvm_release_pfn_clean(pfn);
1243 }
1244 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1245
1246 void kvm_set_page_dirty(struct page *page)
1247 {
1248         kvm_set_pfn_dirty(page_to_pfn(page));
1249 }
1250 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1251
1252 void kvm_set_pfn_dirty(pfn_t pfn)
1253 {
1254         if (!kvm_is_mmio_pfn(pfn)) {
1255                 struct page *page = pfn_to_page(pfn);
1256                 if (!PageReserved(page))
1257                         SetPageDirty(page);
1258         }
1259 }
1260 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1261
1262 void kvm_set_pfn_accessed(pfn_t pfn)
1263 {
1264         if (!kvm_is_mmio_pfn(pfn))
1265                 mark_page_accessed(pfn_to_page(pfn));
1266 }
1267 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1268
1269 void kvm_get_pfn(pfn_t pfn)
1270 {
1271         if (!kvm_is_mmio_pfn(pfn))
1272                 get_page(pfn_to_page(pfn));
1273 }
1274 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1275
1276 static int next_segment(unsigned long len, int offset)
1277 {
1278         if (len > PAGE_SIZE - offset)
1279                 return PAGE_SIZE - offset;
1280         else
1281                 return len;
1282 }
1283
1284 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1285                         int len)
1286 {
1287         int r;
1288         unsigned long addr;
1289
1290         addr = gfn_to_hva(kvm, gfn);
1291         if (kvm_is_error_hva(addr))
1292                 return -EFAULT;
1293         r = __copy_from_user(data, (void __user *)addr + offset, len);
1294         if (r)
1295                 return -EFAULT;
1296         return 0;
1297 }
1298 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1299
1300 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1301 {
1302         gfn_t gfn = gpa >> PAGE_SHIFT;
1303         int seg;
1304         int offset = offset_in_page(gpa);
1305         int ret;
1306
1307         while ((seg = next_segment(len, offset)) != 0) {
1308                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1309                 if (ret < 0)
1310                         return ret;
1311                 offset = 0;
1312                 len -= seg;
1313                 data += seg;
1314                 ++gfn;
1315         }
1316         return 0;
1317 }
1318 EXPORT_SYMBOL_GPL(kvm_read_guest);
1319
1320 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1321                           unsigned long len)
1322 {
1323         int r;
1324         unsigned long addr;
1325         gfn_t gfn = gpa >> PAGE_SHIFT;
1326         int offset = offset_in_page(gpa);
1327
1328         addr = gfn_to_hva(kvm, gfn);
1329         if (kvm_is_error_hva(addr))
1330                 return -EFAULT;
1331         pagefault_disable();
1332         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1333         pagefault_enable();
1334         if (r)
1335                 return -EFAULT;
1336         return 0;
1337 }
1338 EXPORT_SYMBOL(kvm_read_guest_atomic);
1339
1340 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1341                          int offset, int len)
1342 {
1343         int r;
1344         unsigned long addr;
1345
1346         addr = gfn_to_hva(kvm, gfn);
1347         if (kvm_is_error_hva(addr))
1348                 return -EFAULT;
1349         r = copy_to_user((void __user *)addr + offset, data, len);
1350         if (r)
1351                 return -EFAULT;
1352         mark_page_dirty(kvm, gfn);
1353         return 0;
1354 }
1355 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1356
1357 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1358                     unsigned long len)
1359 {
1360         gfn_t gfn = gpa >> PAGE_SHIFT;
1361         int seg;
1362         int offset = offset_in_page(gpa);
1363         int ret;
1364
1365         while ((seg = next_segment(len, offset)) != 0) {
1366                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1367                 if (ret < 0)
1368                         return ret;
1369                 offset = 0;
1370                 len -= seg;
1371                 data += seg;
1372                 ++gfn;
1373         }
1374         return 0;
1375 }
1376
1377 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1378                               gpa_t gpa)
1379 {
1380         struct kvm_memslots *slots = kvm_memslots(kvm);
1381         int offset = offset_in_page(gpa);
1382         gfn_t gfn = gpa >> PAGE_SHIFT;
1383
1384         ghc->gpa = gpa;
1385         ghc->generation = slots->generation;
1386         ghc->memslot = __gfn_to_memslot(slots, gfn);
1387         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1388         if (!kvm_is_error_hva(ghc->hva))
1389                 ghc->hva += offset;
1390         else
1391                 return -EFAULT;
1392
1393         return 0;
1394 }
1395 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1396
1397 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1398                            void *data, unsigned long len)
1399 {
1400         struct kvm_memslots *slots = kvm_memslots(kvm);
1401         int r;
1402
1403         if (slots->generation != ghc->generation)
1404                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1405
1406         if (kvm_is_error_hva(ghc->hva))
1407                 return -EFAULT;
1408
1409         r = copy_to_user((void __user *)ghc->hva, data, len);
1410         if (r)
1411                 return -EFAULT;
1412         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1413
1414         return 0;
1415 }
1416 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1417
1418 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1419 {
1420         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1421                                     offset, len);
1422 }
1423 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1424
1425 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1426 {
1427         gfn_t gfn = gpa >> PAGE_SHIFT;
1428         int seg;
1429         int offset = offset_in_page(gpa);
1430         int ret;
1431
1432         while ((seg = next_segment(len, offset)) != 0) {
1433                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1434                 if (ret < 0)
1435                         return ret;
1436                 offset = 0;
1437                 len -= seg;
1438                 ++gfn;
1439         }
1440         return 0;
1441 }
1442 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1443
1444 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1445                              gfn_t gfn)
1446 {
1447         if (memslot && memslot->dirty_bitmap) {
1448                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1449
1450                 __set_bit_le(rel_gfn, memslot->dirty_bitmap);
1451         }
1452 }
1453
1454 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1455 {
1456         struct kvm_memory_slot *memslot;
1457
1458         memslot = gfn_to_memslot(kvm, gfn);
1459         mark_page_dirty_in_slot(kvm, memslot, gfn);
1460 }
1461
1462 /*
1463  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1464  */
1465 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1466 {
1467         DEFINE_WAIT(wait);
1468
1469         for (;;) {
1470                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1471
1472                 if (kvm_arch_vcpu_runnable(vcpu)) {
1473                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1474                         break;
1475                 }
1476                 if (kvm_cpu_has_pending_timer(vcpu))
1477                         break;
1478                 if (signal_pending(current))
1479                         break;
1480
1481                 schedule();
1482         }
1483
1484         finish_wait(&vcpu->wq, &wait);
1485 }
1486
1487 void kvm_resched(struct kvm_vcpu *vcpu)
1488 {
1489         if (!need_resched())
1490                 return;
1491         cond_resched();
1492 }
1493 EXPORT_SYMBOL_GPL(kvm_resched);
1494
1495 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1496 {
1497         struct kvm *kvm = me->kvm;
1498         struct kvm_vcpu *vcpu;
1499         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1500         int yielded = 0;
1501         int pass;
1502         int i;
1503
1504         /*
1505          * We boost the priority of a VCPU that is runnable but not
1506          * currently running, because it got preempted by something
1507          * else and called schedule in __vcpu_run.  Hopefully that
1508          * VCPU is holding the lock that we need and will release it.
1509          * We approximate round-robin by starting at the last boosted VCPU.
1510          */
1511         for (pass = 0; pass < 2 && !yielded; pass++) {
1512                 kvm_for_each_vcpu(i, vcpu, kvm) {
1513                         struct task_struct *task = NULL;
1514                         struct pid *pid;
1515                         if (!pass && i < last_boosted_vcpu) {
1516                                 i = last_boosted_vcpu;
1517                                 continue;
1518                         } else if (pass && i > last_boosted_vcpu)
1519                                 break;
1520                         if (vcpu == me)
1521                                 continue;
1522                         if (waitqueue_active(&vcpu->wq))
1523                                 continue;
1524                         rcu_read_lock();
1525                         pid = rcu_dereference(vcpu->pid);
1526                         if (pid)
1527                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1528                         rcu_read_unlock();
1529                         if (!task)
1530                                 continue;
1531                         if (task->flags & PF_VCPU) {
1532                                 put_task_struct(task);
1533                                 continue;
1534                         }
1535                         if (yield_to(task, 1)) {
1536                                 put_task_struct(task);
1537                                 kvm->last_boosted_vcpu = i;
1538                                 yielded = 1;
1539                                 break;
1540                         }
1541                         put_task_struct(task);
1542                 }
1543         }
1544 }
1545 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1546
1547 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1548 {
1549         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1550         struct page *page;
1551
1552         if (vmf->pgoff == 0)
1553                 page = virt_to_page(vcpu->run);
1554 #ifdef CONFIG_X86
1555         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1556                 page = virt_to_page(vcpu->arch.pio_data);
1557 #endif
1558 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1559         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1560                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1561 #endif
1562         else
1563                 return VM_FAULT_SIGBUS;
1564         get_page(page);
1565         vmf->page = page;
1566         return 0;
1567 }
1568
1569 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1570         .fault = kvm_vcpu_fault,
1571 };
1572
1573 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1574 {
1575         vma->vm_ops = &kvm_vcpu_vm_ops;
1576         return 0;
1577 }
1578
1579 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1580 {
1581         struct kvm_vcpu *vcpu = filp->private_data;
1582
1583         kvm_put_kvm(vcpu->kvm);
1584         return 0;
1585 }
1586
1587 static struct file_operations kvm_vcpu_fops = {
1588         .release        = kvm_vcpu_release,
1589         .unlocked_ioctl = kvm_vcpu_ioctl,
1590         .compat_ioctl   = kvm_vcpu_ioctl,
1591         .mmap           = kvm_vcpu_mmap,
1592         .llseek         = noop_llseek,
1593 };
1594
1595 /*
1596  * Allocates an inode for the vcpu.
1597  */
1598 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1599 {
1600         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1601 }
1602
1603 /*
1604  * Creates some virtual cpus.  Good luck creating more than one.
1605  */
1606 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1607 {
1608         int r;
1609         struct kvm_vcpu *vcpu, *v;
1610
1611         vcpu = kvm_arch_vcpu_create(kvm, id);
1612         if (IS_ERR(vcpu))
1613                 return PTR_ERR(vcpu);
1614
1615         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1616
1617         r = kvm_arch_vcpu_setup(vcpu);
1618         if (r)
1619                 goto vcpu_destroy;
1620
1621         mutex_lock(&kvm->lock);
1622         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1623                 r = -EINVAL;
1624                 goto unlock_vcpu_destroy;
1625         }
1626
1627         kvm_for_each_vcpu(r, v, kvm)
1628                 if (v->vcpu_id == id) {
1629                         r = -EEXIST;
1630                         goto unlock_vcpu_destroy;
1631                 }
1632
1633         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1634
1635         /* Now it's all set up, let userspace reach it */
1636         kvm_get_kvm(kvm);
1637         r = create_vcpu_fd(vcpu);
1638         if (r < 0) {
1639                 kvm_put_kvm(kvm);
1640                 goto unlock_vcpu_destroy;
1641         }
1642
1643         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1644         smp_wmb();
1645         atomic_inc(&kvm->online_vcpus);
1646
1647 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1648         if (kvm->bsp_vcpu_id == id)
1649                 kvm->bsp_vcpu = vcpu;
1650 #endif
1651         mutex_unlock(&kvm->lock);
1652         return r;
1653
1654 unlock_vcpu_destroy:
1655         mutex_unlock(&kvm->lock);
1656 vcpu_destroy:
1657         kvm_arch_vcpu_destroy(vcpu);
1658         return r;
1659 }
1660
1661 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1662 {
1663         if (sigset) {
1664                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1665                 vcpu->sigset_active = 1;
1666                 vcpu->sigset = *sigset;
1667         } else
1668                 vcpu->sigset_active = 0;
1669         return 0;
1670 }
1671
1672 static long kvm_vcpu_ioctl(struct file *filp,
1673                            unsigned int ioctl, unsigned long arg)
1674 {
1675         struct kvm_vcpu *vcpu = filp->private_data;
1676         void __user *argp = (void __user *)arg;
1677         int r;
1678         struct kvm_fpu *fpu = NULL;
1679         struct kvm_sregs *kvm_sregs = NULL;
1680
1681         if (vcpu->kvm->mm != current->mm)
1682                 return -EIO;
1683
1684 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1685         /*
1686          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1687          * so vcpu_load() would break it.
1688          */
1689         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1690                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1691 #endif
1692
1693
1694         vcpu_load(vcpu);
1695         switch (ioctl) {
1696         case KVM_RUN:
1697                 r = -EINVAL;
1698                 if (arg)
1699                         goto out;
1700                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1701                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1702                 break;
1703         case KVM_GET_REGS: {
1704                 struct kvm_regs *kvm_regs;
1705
1706                 r = -ENOMEM;
1707                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1708                 if (!kvm_regs)
1709                         goto out;
1710                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1711                 if (r)
1712                         goto out_free1;
1713                 r = -EFAULT;
1714                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1715                         goto out_free1;
1716                 r = 0;
1717 out_free1:
1718                 kfree(kvm_regs);
1719                 break;
1720         }
1721         case KVM_SET_REGS: {
1722                 struct kvm_regs *kvm_regs;
1723
1724                 r = -ENOMEM;
1725                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1726                 if (!kvm_regs)
1727                         goto out;
1728                 r = -EFAULT;
1729                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1730                         goto out_free2;
1731                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1732                 if (r)
1733                         goto out_free2;
1734                 r = 0;
1735 out_free2:
1736                 kfree(kvm_regs);
1737                 break;
1738         }
1739         case KVM_GET_SREGS: {
1740                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1741                 r = -ENOMEM;
1742                 if (!kvm_sregs)
1743                         goto out;
1744                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1745                 if (r)
1746                         goto out;
1747                 r = -EFAULT;
1748                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1749                         goto out;
1750                 r = 0;
1751                 break;
1752         }
1753         case KVM_SET_SREGS: {
1754                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1755                 r = -ENOMEM;
1756                 if (!kvm_sregs)
1757                         goto out;
1758                 r = -EFAULT;
1759                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1760                         goto out;
1761                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1762                 if (r)
1763                         goto out;
1764                 r = 0;
1765                 break;
1766         }
1767         case KVM_GET_MP_STATE: {
1768                 struct kvm_mp_state mp_state;
1769
1770                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1771                 if (r)
1772                         goto out;
1773                 r = -EFAULT;
1774                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1775                         goto out;
1776                 r = 0;
1777                 break;
1778         }
1779         case KVM_SET_MP_STATE: {
1780                 struct kvm_mp_state mp_state;
1781
1782                 r = -EFAULT;
1783                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1784                         goto out;
1785                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1786                 if (r)
1787                         goto out;
1788                 r = 0;
1789                 break;
1790         }
1791         case KVM_TRANSLATE: {
1792                 struct kvm_translation tr;
1793
1794                 r = -EFAULT;
1795                 if (copy_from_user(&tr, argp, sizeof tr))
1796                         goto out;
1797                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1798                 if (r)
1799                         goto out;
1800                 r = -EFAULT;
1801                 if (copy_to_user(argp, &tr, sizeof tr))
1802                         goto out;
1803                 r = 0;
1804                 break;
1805         }
1806         case KVM_SET_GUEST_DEBUG: {
1807                 struct kvm_guest_debug dbg;
1808
1809                 r = -EFAULT;
1810                 if (copy_from_user(&dbg, argp, sizeof dbg))
1811                         goto out;
1812                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1813                 if (r)
1814                         goto out;
1815                 r = 0;
1816                 break;
1817         }
1818         case KVM_SET_SIGNAL_MASK: {
1819                 struct kvm_signal_mask __user *sigmask_arg = argp;
1820                 struct kvm_signal_mask kvm_sigmask;
1821                 sigset_t sigset, *p;
1822
1823                 p = NULL;
1824                 if (argp) {
1825                         r = -EFAULT;
1826                         if (copy_from_user(&kvm_sigmask, argp,
1827                                            sizeof kvm_sigmask))
1828                                 goto out;
1829                         r = -EINVAL;
1830                         if (kvm_sigmask.len != sizeof sigset)
1831                                 goto out;
1832                         r = -EFAULT;
1833                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1834                                            sizeof sigset))
1835                                 goto out;
1836                         p = &sigset;
1837                 }
1838                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1839                 break;
1840         }
1841         case KVM_GET_FPU: {
1842                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1843                 r = -ENOMEM;
1844                 if (!fpu)
1845                         goto out;
1846                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1847                 if (r)
1848                         goto out;
1849                 r = -EFAULT;
1850                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1851                         goto out;
1852                 r = 0;
1853                 break;
1854         }
1855         case KVM_SET_FPU: {
1856                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1857                 r = -ENOMEM;
1858                 if (!fpu)
1859                         goto out;
1860                 r = -EFAULT;
1861                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1862                         goto out;
1863                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1864                 if (r)
1865                         goto out;
1866                 r = 0;
1867                 break;
1868         }
1869         default:
1870                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1871         }
1872 out:
1873         vcpu_put(vcpu);
1874         kfree(fpu);
1875         kfree(kvm_sregs);
1876         return r;
1877 }
1878
1879 static long kvm_vm_ioctl(struct file *filp,
1880                            unsigned int ioctl, unsigned long arg)
1881 {
1882         struct kvm *kvm = filp->private_data;
1883         void __user *argp = (void __user *)arg;
1884         int r;
1885
1886         if (kvm->mm != current->mm)
1887                 return -EIO;
1888         switch (ioctl) {
1889         case KVM_CREATE_VCPU:
1890                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1891                 if (r < 0)
1892                         goto out;
1893                 break;
1894         case KVM_SET_USER_MEMORY_REGION: {
1895                 struct kvm_userspace_memory_region kvm_userspace_mem;
1896
1897                 r = -EFAULT;
1898                 if (copy_from_user(&kvm_userspace_mem, argp,
1899                                                 sizeof kvm_userspace_mem))
1900                         goto out;
1901
1902                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1903                 if (r)
1904                         goto out;
1905                 break;
1906         }
1907         case KVM_GET_DIRTY_LOG: {
1908                 struct kvm_dirty_log log;
1909
1910                 r = -EFAULT;
1911                 if (copy_from_user(&log, argp, sizeof log))
1912                         goto out;
1913                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1914                 if (r)
1915                         goto out;
1916                 break;
1917         }
1918 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1919         case KVM_REGISTER_COALESCED_MMIO: {
1920                 struct kvm_coalesced_mmio_zone zone;
1921                 r = -EFAULT;
1922                 if (copy_from_user(&zone, argp, sizeof zone))
1923                         goto out;
1924                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1925                 if (r)
1926                         goto out;
1927                 r = 0;
1928                 break;
1929         }
1930         case KVM_UNREGISTER_COALESCED_MMIO: {
1931                 struct kvm_coalesced_mmio_zone zone;
1932                 r = -EFAULT;
1933                 if (copy_from_user(&zone, argp, sizeof zone))
1934                         goto out;
1935                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1936                 if (r)
1937                         goto out;
1938                 r = 0;
1939                 break;
1940         }
1941 #endif
1942         case KVM_IRQFD: {
1943                 struct kvm_irqfd data;
1944
1945                 r = -EFAULT;
1946                 if (copy_from_user(&data, argp, sizeof data))
1947                         goto out;
1948                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1949                 break;
1950         }
1951         case KVM_IOEVENTFD: {
1952                 struct kvm_ioeventfd data;
1953
1954                 r = -EFAULT;
1955                 if (copy_from_user(&data, argp, sizeof data))
1956                         goto out;
1957                 r = kvm_ioeventfd(kvm, &data);
1958                 break;
1959         }
1960 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1961         case KVM_SET_BOOT_CPU_ID:
1962                 r = 0;
1963                 mutex_lock(&kvm->lock);
1964                 if (atomic_read(&kvm->online_vcpus) != 0)
1965                         r = -EBUSY;
1966                 else
1967                         kvm->bsp_vcpu_id = arg;
1968                 mutex_unlock(&kvm->lock);
1969                 break;
1970 #endif
1971         default:
1972                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1973                 if (r == -ENOTTY)
1974                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1975         }
1976 out:
1977         return r;
1978 }
1979
1980 #ifdef CONFIG_COMPAT
1981 struct compat_kvm_dirty_log {
1982         __u32 slot;
1983         __u32 padding1;
1984         union {
1985                 compat_uptr_t dirty_bitmap; /* one bit per page */
1986                 __u64 padding2;
1987         };
1988 };
1989
1990 static long kvm_vm_compat_ioctl(struct file *filp,
1991                            unsigned int ioctl, unsigned long arg)
1992 {
1993         struct kvm *kvm = filp->private_data;
1994         int r;
1995
1996         if (kvm->mm != current->mm)
1997                 return -EIO;
1998         switch (ioctl) {
1999         case KVM_GET_DIRTY_LOG: {
2000                 struct compat_kvm_dirty_log compat_log;
2001                 struct kvm_dirty_log log;
2002
2003                 r = -EFAULT;
2004                 if (copy_from_user(&compat_log, (void __user *)arg,
2005                                    sizeof(compat_log)))
2006                         goto out;
2007                 log.slot         = compat_log.slot;
2008                 log.padding1     = compat_log.padding1;
2009                 log.padding2     = compat_log.padding2;
2010                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2011
2012                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2013                 if (r)
2014                         goto out;
2015                 break;
2016         }
2017         default:
2018                 r = kvm_vm_ioctl(filp, ioctl, arg);
2019         }
2020
2021 out:
2022         return r;
2023 }
2024 #endif
2025
2026 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2027 {
2028         struct page *page[1];
2029         unsigned long addr;
2030         int npages;
2031         gfn_t gfn = vmf->pgoff;
2032         struct kvm *kvm = vma->vm_file->private_data;
2033
2034         addr = gfn_to_hva(kvm, gfn);
2035         if (kvm_is_error_hva(addr))
2036                 return VM_FAULT_SIGBUS;
2037
2038         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2039                                 NULL);
2040         if (unlikely(npages != 1))
2041                 return VM_FAULT_SIGBUS;
2042
2043         vmf->page = page[0];
2044         return 0;
2045 }
2046
2047 static const struct vm_operations_struct kvm_vm_vm_ops = {
2048         .fault = kvm_vm_fault,
2049 };
2050
2051 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2052 {
2053         vma->vm_ops = &kvm_vm_vm_ops;
2054         return 0;
2055 }
2056
2057 static struct file_operations kvm_vm_fops = {
2058         .release        = kvm_vm_release,
2059         .unlocked_ioctl = kvm_vm_ioctl,
2060 #ifdef CONFIG_COMPAT
2061         .compat_ioctl   = kvm_vm_compat_ioctl,
2062 #endif
2063         .mmap           = kvm_vm_mmap,
2064         .llseek         = noop_llseek,
2065 };
2066
2067 static int kvm_dev_ioctl_create_vm(void)
2068 {
2069         int r;
2070         struct kvm *kvm;
2071
2072         kvm = kvm_create_vm();
2073         if (IS_ERR(kvm))
2074                 return PTR_ERR(kvm);
2075 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2076         r = kvm_coalesced_mmio_init(kvm);
2077         if (r < 0) {
2078                 kvm_put_kvm(kvm);
2079                 return r;
2080         }
2081 #endif
2082         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2083         if (r < 0)
2084                 kvm_put_kvm(kvm);
2085
2086         return r;
2087 }
2088
2089 static long kvm_dev_ioctl_check_extension_generic(long arg)
2090 {
2091         switch (arg) {
2092         case KVM_CAP_USER_MEMORY:
2093         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2094         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2095 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2096         case KVM_CAP_SET_BOOT_CPU_ID:
2097 #endif
2098         case KVM_CAP_INTERNAL_ERROR_DATA:
2099                 return 1;
2100 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2101         case KVM_CAP_IRQ_ROUTING:
2102                 return KVM_MAX_IRQ_ROUTES;
2103 #endif
2104         default:
2105                 break;
2106         }
2107         return kvm_dev_ioctl_check_extension(arg);
2108 }
2109
2110 static long kvm_dev_ioctl(struct file *filp,
2111                           unsigned int ioctl, unsigned long arg)
2112 {
2113         long r = -EINVAL;
2114
2115         switch (ioctl) {
2116         case KVM_GET_API_VERSION:
2117                 r = -EINVAL;
2118                 if (arg)
2119                         goto out;
2120                 r = KVM_API_VERSION;
2121                 break;
2122         case KVM_CREATE_VM:
2123                 r = -EINVAL;
2124                 if (arg)
2125                         goto out;
2126                 r = kvm_dev_ioctl_create_vm();
2127                 break;
2128         case KVM_CHECK_EXTENSION:
2129                 r = kvm_dev_ioctl_check_extension_generic(arg);
2130                 break;
2131         case KVM_GET_VCPU_MMAP_SIZE:
2132                 r = -EINVAL;
2133                 if (arg)
2134                         goto out;
2135                 r = PAGE_SIZE;     /* struct kvm_run */
2136 #ifdef CONFIG_X86
2137                 r += PAGE_SIZE;    /* pio data page */
2138 #endif
2139 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2140                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2141 #endif
2142                 break;
2143         case KVM_TRACE_ENABLE:
2144         case KVM_TRACE_PAUSE:
2145         case KVM_TRACE_DISABLE:
2146                 r = -EOPNOTSUPP;
2147                 break;
2148         default:
2149                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2150         }
2151 out:
2152         return r;
2153 }
2154
2155 static struct file_operations kvm_chardev_ops = {
2156         .unlocked_ioctl = kvm_dev_ioctl,
2157         .compat_ioctl   = kvm_dev_ioctl,
2158         .llseek         = noop_llseek,
2159 };
2160
2161 static struct miscdevice kvm_dev = {
2162         KVM_MINOR,
2163         "kvm",
2164         &kvm_chardev_ops,
2165 };
2166
2167 static void hardware_enable_nolock(void *junk)
2168 {
2169         int cpu = raw_smp_processor_id();
2170         int r;
2171
2172         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2173                 return;
2174
2175         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2176
2177         r = kvm_arch_hardware_enable(NULL);
2178
2179         if (r) {
2180                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2181                 atomic_inc(&hardware_enable_failed);
2182                 printk(KERN_INFO "kvm: enabling virtualization on "
2183                                  "CPU%d failed\n", cpu);
2184         }
2185 }
2186
2187 static void hardware_enable(void *junk)
2188 {
2189         raw_spin_lock(&kvm_lock);
2190         hardware_enable_nolock(junk);
2191         raw_spin_unlock(&kvm_lock);
2192 }
2193
2194 static void hardware_disable_nolock(void *junk)
2195 {
2196         int cpu = raw_smp_processor_id();
2197
2198         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2199                 return;
2200         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2201         kvm_arch_hardware_disable(NULL);
2202 }
2203
2204 static void hardware_disable(void *junk)
2205 {
2206         raw_spin_lock(&kvm_lock);
2207         hardware_disable_nolock(junk);
2208         raw_spin_unlock(&kvm_lock);
2209 }
2210
2211 static void hardware_disable_all_nolock(void)
2212 {
2213         BUG_ON(!kvm_usage_count);
2214
2215         kvm_usage_count--;
2216         if (!kvm_usage_count)
2217                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2218 }
2219
2220 static void hardware_disable_all(void)
2221 {
2222         raw_spin_lock(&kvm_lock);
2223         hardware_disable_all_nolock();
2224         raw_spin_unlock(&kvm_lock);
2225 }
2226
2227 static int hardware_enable_all(void)
2228 {
2229         int r = 0;
2230
2231         raw_spin_lock(&kvm_lock);
2232
2233         kvm_usage_count++;
2234         if (kvm_usage_count == 1) {
2235                 atomic_set(&hardware_enable_failed, 0);
2236                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2237
2238                 if (atomic_read(&hardware_enable_failed)) {
2239                         hardware_disable_all_nolock();
2240                         r = -EBUSY;
2241                 }
2242         }
2243
2244         raw_spin_unlock(&kvm_lock);
2245
2246         return r;
2247 }
2248
2249 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2250                            void *v)
2251 {
2252         int cpu = (long)v;
2253
2254         if (!kvm_usage_count)
2255                 return NOTIFY_OK;
2256
2257         val &= ~CPU_TASKS_FROZEN;
2258         switch (val) {
2259         case CPU_DYING:
2260                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2261                        cpu);
2262                 hardware_disable(NULL);
2263                 break;
2264         case CPU_STARTING:
2265                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2266                        cpu);
2267                 hardware_enable(NULL);
2268                 break;
2269         }
2270         return NOTIFY_OK;
2271 }
2272
2273
2274 asmlinkage void kvm_spurious_fault(void)
2275 {
2276         /* Fault while not rebooting.  We want the trace. */
2277         BUG();
2278 }
2279 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2280
2281 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2282                       void *v)
2283 {
2284         /*
2285          * Some (well, at least mine) BIOSes hang on reboot if
2286          * in vmx root mode.
2287          *
2288          * And Intel TXT required VMX off for all cpu when system shutdown.
2289          */
2290         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2291         kvm_rebooting = true;
2292         on_each_cpu(hardware_disable_nolock, NULL, 1);
2293         return NOTIFY_OK;
2294 }
2295
2296 static struct notifier_block kvm_reboot_notifier = {
2297         .notifier_call = kvm_reboot,
2298         .priority = 0,
2299 };
2300
2301 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2302 {
2303         int i;
2304
2305         for (i = 0; i < bus->dev_count; i++) {
2306                 struct kvm_io_device *pos = bus->devs[i];
2307
2308                 kvm_iodevice_destructor(pos);
2309         }
2310         kfree(bus);
2311 }
2312
2313 /* kvm_io_bus_write - called under kvm->slots_lock */
2314 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2315                      int len, const void *val)
2316 {
2317         int i;
2318         struct kvm_io_bus *bus;
2319
2320         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2321         for (i = 0; i < bus->dev_count; i++)
2322                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2323                         return 0;
2324         return -EOPNOTSUPP;
2325 }
2326
2327 /* kvm_io_bus_read - called under kvm->slots_lock */
2328 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2329                     int len, void *val)
2330 {
2331         int i;
2332         struct kvm_io_bus *bus;
2333
2334         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2335         for (i = 0; i < bus->dev_count; i++)
2336                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2337                         return 0;
2338         return -EOPNOTSUPP;
2339 }
2340
2341 /* Caller must hold slots_lock. */
2342 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2343                             struct kvm_io_device *dev)
2344 {
2345         struct kvm_io_bus *new_bus, *bus;
2346
2347         bus = kvm->buses[bus_idx];
2348         if (bus->dev_count > NR_IOBUS_DEVS-1)
2349                 return -ENOSPC;
2350
2351         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2352         if (!new_bus)
2353                 return -ENOMEM;
2354         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2355         new_bus->devs[new_bus->dev_count++] = dev;
2356         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2357         synchronize_srcu_expedited(&kvm->srcu);
2358         kfree(bus);
2359
2360         return 0;
2361 }
2362
2363 /* Caller must hold slots_lock. */
2364 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2365                               struct kvm_io_device *dev)
2366 {
2367         int i, r;
2368         struct kvm_io_bus *new_bus, *bus;
2369
2370         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2371         if (!new_bus)
2372                 return -ENOMEM;
2373
2374         bus = kvm->buses[bus_idx];
2375         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2376
2377         r = -ENOENT;
2378         for (i = 0; i < new_bus->dev_count; i++)
2379                 if (new_bus->devs[i] == dev) {
2380                         r = 0;
2381                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2382                         break;
2383                 }
2384
2385         if (r) {
2386                 kfree(new_bus);
2387                 return r;
2388         }
2389
2390         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2391         synchronize_srcu_expedited(&kvm->srcu);
2392         kfree(bus);
2393         return r;
2394 }
2395
2396 static struct notifier_block kvm_cpu_notifier = {
2397         .notifier_call = kvm_cpu_hotplug,
2398 };
2399
2400 static int vm_stat_get(void *_offset, u64 *val)
2401 {
2402         unsigned offset = (long)_offset;
2403         struct kvm *kvm;
2404
2405         *val = 0;
2406         raw_spin_lock(&kvm_lock);
2407         list_for_each_entry(kvm, &vm_list, vm_list)
2408                 *val += *(u32 *)((void *)kvm + offset);
2409         raw_spin_unlock(&kvm_lock);
2410         return 0;
2411 }
2412
2413 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2414
2415 static int vcpu_stat_get(void *_offset, u64 *val)
2416 {
2417         unsigned offset = (long)_offset;
2418         struct kvm *kvm;
2419         struct kvm_vcpu *vcpu;
2420         int i;
2421
2422         *val = 0;
2423         raw_spin_lock(&kvm_lock);
2424         list_for_each_entry(kvm, &vm_list, vm_list)
2425                 kvm_for_each_vcpu(i, vcpu, kvm)
2426                         *val += *(u32 *)((void *)vcpu + offset);
2427
2428         raw_spin_unlock(&kvm_lock);
2429         return 0;
2430 }
2431
2432 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2433
2434 static const struct file_operations *stat_fops[] = {
2435         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2436         [KVM_STAT_VM]   = &vm_stat_fops,
2437 };
2438
2439 static void kvm_init_debug(void)
2440 {
2441         struct kvm_stats_debugfs_item *p;
2442
2443         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2444         for (p = debugfs_entries; p->name; ++p)
2445                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2446                                                 (void *)(long)p->offset,
2447                                                 stat_fops[p->kind]);
2448 }
2449
2450 static void kvm_exit_debug(void)
2451 {
2452         struct kvm_stats_debugfs_item *p;
2453
2454         for (p = debugfs_entries; p->name; ++p)
2455                 debugfs_remove(p->dentry);
2456         debugfs_remove(kvm_debugfs_dir);
2457 }
2458
2459 static int kvm_suspend(void)
2460 {
2461         if (kvm_usage_count)
2462                 hardware_disable_nolock(NULL);
2463         return 0;
2464 }
2465
2466 static void kvm_resume(void)
2467 {
2468         if (kvm_usage_count) {
2469                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2470                 hardware_enable_nolock(NULL);
2471         }
2472 }
2473
2474 static struct syscore_ops kvm_syscore_ops = {
2475         .suspend = kvm_suspend,
2476         .resume = kvm_resume,
2477 };
2478
2479 struct page *bad_page;
2480 pfn_t bad_pfn;
2481
2482 static inline
2483 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2484 {
2485         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2486 }
2487
2488 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2489 {
2490         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2491
2492         kvm_arch_vcpu_load(vcpu, cpu);
2493 }
2494
2495 static void kvm_sched_out(struct preempt_notifier *pn,
2496                           struct task_struct *next)
2497 {
2498         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2499
2500         kvm_arch_vcpu_put(vcpu);
2501 }
2502
2503 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2504                   struct module *module)
2505 {
2506         int r;
2507         int cpu;
2508
2509         r = kvm_arch_init(opaque);
2510         if (r)
2511                 goto out_fail;
2512
2513         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2514
2515         if (bad_page == NULL) {
2516                 r = -ENOMEM;
2517                 goto out;
2518         }
2519
2520         bad_pfn = page_to_pfn(bad_page);
2521
2522         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2523
2524         if (hwpoison_page == NULL) {
2525                 r = -ENOMEM;
2526                 goto out_free_0;
2527         }
2528
2529         hwpoison_pfn = page_to_pfn(hwpoison_page);
2530
2531         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2532
2533         if (fault_page == NULL) {
2534                 r = -ENOMEM;
2535                 goto out_free_0;
2536         }
2537
2538         fault_pfn = page_to_pfn(fault_page);
2539
2540         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2541                 r = -ENOMEM;
2542                 goto out_free_0;
2543         }
2544
2545         r = kvm_arch_hardware_setup();
2546         if (r < 0)
2547                 goto out_free_0a;
2548
2549         for_each_online_cpu(cpu) {
2550                 smp_call_function_single(cpu,
2551                                 kvm_arch_check_processor_compat,
2552                                 &r, 1);
2553                 if (r < 0)
2554                         goto out_free_1;
2555         }
2556
2557         r = register_cpu_notifier(&kvm_cpu_notifier);
2558         if (r)
2559                 goto out_free_2;
2560         register_reboot_notifier(&kvm_reboot_notifier);
2561
2562         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2563         if (!vcpu_align)
2564                 vcpu_align = __alignof__(struct kvm_vcpu);
2565         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2566                                            0, NULL);
2567         if (!kvm_vcpu_cache) {
2568                 r = -ENOMEM;
2569                 goto out_free_3;
2570         }
2571
2572         r = kvm_async_pf_init();
2573         if (r)
2574                 goto out_free;
2575
2576         kvm_chardev_ops.owner = module;
2577         kvm_vm_fops.owner = module;
2578         kvm_vcpu_fops.owner = module;
2579
2580         r = misc_register(&kvm_dev);
2581         if (r) {
2582                 printk(KERN_ERR "kvm: misc device register failed\n");
2583                 goto out_unreg;
2584         }
2585
2586         register_syscore_ops(&kvm_syscore_ops);
2587
2588         kvm_preempt_ops.sched_in = kvm_sched_in;
2589         kvm_preempt_ops.sched_out = kvm_sched_out;
2590
2591         kvm_init_debug();
2592
2593         return 0;
2594
2595 out_unreg:
2596         kvm_async_pf_deinit();
2597 out_free:
2598         kmem_cache_destroy(kvm_vcpu_cache);
2599 out_free_3:
2600         unregister_reboot_notifier(&kvm_reboot_notifier);
2601         unregister_cpu_notifier(&kvm_cpu_notifier);
2602 out_free_2:
2603 out_free_1:
2604         kvm_arch_hardware_unsetup();
2605 out_free_0a:
2606         free_cpumask_var(cpus_hardware_enabled);
2607 out_free_0:
2608         if (fault_page)
2609                 __free_page(fault_page);
2610         if (hwpoison_page)
2611                 __free_page(hwpoison_page);
2612         __free_page(bad_page);
2613 out:
2614         kvm_arch_exit();
2615 out_fail:
2616         return r;
2617 }
2618 EXPORT_SYMBOL_GPL(kvm_init);
2619
2620 void kvm_exit(void)
2621 {
2622         kvm_exit_debug();
2623         misc_deregister(&kvm_dev);
2624         kmem_cache_destroy(kvm_vcpu_cache);
2625         kvm_async_pf_deinit();
2626         unregister_syscore_ops(&kvm_syscore_ops);
2627         unregister_reboot_notifier(&kvm_reboot_notifier);
2628         unregister_cpu_notifier(&kvm_cpu_notifier);
2629         on_each_cpu(hardware_disable_nolock, NULL, 1);
2630         kvm_arch_hardware_unsetup();
2631         kvm_arch_exit();
2632         free_cpumask_var(cpus_hardware_enabled);
2633         __free_page(hwpoison_page);
2634         __free_page(bad_page);
2635 }
2636 EXPORT_SYMBOL_GPL(kvm_exit);