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 #ifdef CONFIG_COMPAT
  88 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  89                                   unsigned long arg);
  90 #endif
  91 static int hardware_enable_all(void);
  92 static void hardware_disable_all(void);
  93
  94 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  95
  96 bool kvm_rebooting;
  97 EXPORT_SYMBOL_GPL(kvm_rebooting);
  98
  99 static bool largepages_enabled = true;
 100
 101 static struct page *hwpoison_page;
 102 static pfn_t hwpoison_pfn;
 103
 104 struct page *fault_page;
 105 pfn_t fault_pfn;
 106
 107 inline int kvm_is_mmio_pfn(pfn_t pfn)
 108 {
 109         if (pfn_valid(pfn)) {
 110                 int reserved;
 111                 struct page *tail = pfn_to_page(pfn);
 112                 struct page *head = compound_trans_head(tail);
 113                 reserved = PageReserved(head);
 114                 if (head != tail) {
 115                         /*
 116                          * "head" is not a dangling pointer
 117                          * (compound_trans_head takes care of that)
 118                          * but the hugepage may have been splitted
 119                          * from under us (and we may not hold a
 120                          * reference count on the head page so it can
 121                          * be reused before we run PageReferenced), so
 122                          * we've to check PageTail before returning
 123                          * what we just read.
 124                          */
 125                         smp_rmb();
 126                         if (PageTail(tail))
 127                                 return reserved;
 128                 }
 129                 return PageReserved(tail);
 130         }
 131
 132         return true;
 133 }
 134
 135 /*
 136  * Switches to specified vcpu, until a matching vcpu_put()
 137  */
 138 void vcpu_load(struct kvm_vcpu *vcpu)
 139 {
 140         int cpu;
 141
 142         mutex_lock(&vcpu->mutex);
 143         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 144                 /* The thread running this VCPU changed. */
 145                 struct pid *oldpid = vcpu->pid;
 146                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 147                 rcu_assign_pointer(vcpu->pid, newpid);
 148                 synchronize_rcu();
 149                 put_pid(oldpid);
 150         }
 151         cpu = get_cpu();
 152         preempt_notifier_register(&vcpu->preempt_notifier);
 153         kvm_arch_vcpu_load(vcpu, cpu);
 154         put_cpu();
 155 }
 156
 157 void vcpu_put(struct kvm_vcpu *vcpu)
 158 {
 159         preempt_disable();
 160         kvm_arch_vcpu_put(vcpu);
 161         preempt_notifier_unregister(&vcpu->preempt_notifier);
 162         preempt_enable();
 163         mutex_unlock(&vcpu->mutex);
 164 }
 165
 166 static void ack_flush(void *_completed)
 167 {
 168 }
 169
 170 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 171 {
 172         int i, cpu, me;
 173         cpumask_var_t cpus;
 174         bool called = true;
 175         struct kvm_vcpu *vcpu;
 176
 177         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 178
 179         me = get_cpu();
 180         kvm_for_each_vcpu(i, vcpu, kvm) {
 181                 kvm_make_request(req, vcpu);
 182                 cpu = vcpu->cpu;
 183
 184                 /* Set ->requests bit before we read ->mode */
 185                 smp_mb();
 186
 187                 if (cpus != NULL && cpu != -1 && cpu != me &&
 188                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 189                         cpumask_set_cpu(cpu, cpus);
 190         }
 191         if (unlikely(cpus == NULL))
 192                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 193         else if (!cpumask_empty(cpus))
 194                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 195         else
 196                 called = false;
 197         put_cpu();
 198         free_cpumask_var(cpus);
 199         return called;
 200 }
 201
 202 void kvm_flush_remote_tlbs(struct kvm *kvm)
 203 {
 204         int dirty_count = kvm->tlbs_dirty;
 205
 206         smp_mb();
 207         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 208                 ++kvm->stat.remote_tlb_flush;
 209         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 210 }
 211
 212 void kvm_reload_remote_mmus(struct kvm *kvm)
 213 {
 214         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 215 }
 216
 217 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 218 {
 219         struct page *page;
 220         int r;
 221
 222         mutex_init(&vcpu->mutex);
 223         vcpu->cpu = -1;
 224         vcpu->kvm = kvm;
 225         vcpu->vcpu_id = id;
 226         vcpu->pid = NULL;
 227         init_waitqueue_head(&vcpu->wq);
 228         kvm_async_pf_vcpu_init(vcpu);
 229
 230         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 231         if (!page) {
 232                 r = -ENOMEM;
 233                 goto fail;
 234         }
 235         vcpu->run = page_address(page);
 236
 237         r = kvm_arch_vcpu_init(vcpu);
 238         if (r < 0)
 239                 goto fail_free_run;
 240         return 0;
 241
 242 fail_free_run:
 243         free_page((unsigned long)vcpu->run);
 244 fail:
 245         return r;
 246 }
 247 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 248
 249 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 250 {
 251         put_pid(vcpu->pid);
 252         kvm_arch_vcpu_uninit(vcpu);
 253         free_page((unsigned long)vcpu->run);
 254 }
 255 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 256
 257 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 258 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 259 {
 260         return container_of(mn, struct kvm, mmu_notifier);
 261 }
 262
 263 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 264                                              struct mm_struct *mm,
 265                                              unsigned long address)
 266 {
 267         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 268         int need_tlb_flush, idx;
 269
 270         /*
 271          * When ->invalidate_page runs, the linux pte has been zapped
 272          * already but the page is still allocated until
 273          * ->invalidate_page returns. So if we increase the sequence
 274          * here the kvm page fault will notice if the spte can't be
 275          * established because the page is going to be freed. If
 276          * instead the kvm page fault establishes the spte before
 277          * ->invalidate_page runs, kvm_unmap_hva will release it
 278          * before returning.
 279          *
 280          * The sequence increase only need to be seen at spin_unlock
 281          * time, and not at spin_lock time.
 282          *
 283          * Increasing the sequence after the spin_unlock would be
 284          * unsafe because the kvm page fault could then establish the
 285          * pte after kvm_unmap_hva returned, without noticing the page
 286          * is going to be freed.
 287          */
 288         idx = srcu_read_lock(&kvm->srcu);
 289         spin_lock(&kvm->mmu_lock);
 290         kvm->mmu_notifier_seq++;
 291         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 292         spin_unlock(&kvm->mmu_lock);
 293         srcu_read_unlock(&kvm->srcu, idx);
 294
 295         /* we've to flush the tlb before the pages can be freed */
 296         if (need_tlb_flush)
 297                 kvm_flush_remote_tlbs(kvm);
 298
 299 }
 300
 301 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 302                                         struct mm_struct *mm,
 303                                         unsigned long address,
 304                                         pte_t pte)
 305 {
 306         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 307         int idx;
 308
 309         idx = srcu_read_lock(&kvm->srcu);
 310         spin_lock(&kvm->mmu_lock);
 311         kvm->mmu_notifier_seq++;
 312         kvm_set_spte_hva(kvm, address, pte);
 313         spin_unlock(&kvm->mmu_lock);
 314         srcu_read_unlock(&kvm->srcu, idx);
 315 }
 316
 317 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 318                                                     struct mm_struct *mm,
 319                                                     unsigned long start,
 320                                                     unsigned long end)
 321 {
 322         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 323         int need_tlb_flush = 0, idx;
 324
 325         idx = srcu_read_lock(&kvm->srcu);
 326         spin_lock(&kvm->mmu_lock);
 327         /*
 328          * The count increase must become visible at unlock time as no
 329          * spte can be established without taking the mmu_lock and
 330          * count is also read inside the mmu_lock critical section.
 331          */
 332         kvm->mmu_notifier_count++;
 333         for (; start < end; start += PAGE_SIZE)
 334                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 335         need_tlb_flush |= kvm->tlbs_dirty;
 336         spin_unlock(&kvm->mmu_lock);
 337         srcu_read_unlock(&kvm->srcu, idx);
 338
 339         /* we've to flush the tlb before the pages can be freed */
 340         if (need_tlb_flush)
 341                 kvm_flush_remote_tlbs(kvm);
 342 }
 343
 344 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 345                                                   struct mm_struct *mm,
 346                                                   unsigned long start,
 347                                                   unsigned long end)
 348 {
 349         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 350
 351         spin_lock(&kvm->mmu_lock);
 352         /*
 353          * This sequence increase will notify the kvm page fault that
 354          * the page that is going to be mapped in the spte could have
 355          * been freed.
 356          */
 357         kvm->mmu_notifier_seq++;
 358         /*
 359          * The above sequence increase must be visible before the
 360          * below count decrease but both values are read by the kvm
 361          * page fault under mmu_lock spinlock so we don't need to add
 362          * a smb_wmb() here in between the two.
 363          */
 364         kvm->mmu_notifier_count--;
 365         spin_unlock(&kvm->mmu_lock);
 366
 367         BUG_ON(kvm->mmu_notifier_count < 0);
 368 }
 369
 370 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 371                                               struct mm_struct *mm,
 372                                               unsigned long address)
 373 {
 374         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 375         int young, idx;
 376
 377         idx = srcu_read_lock(&kvm->srcu);
 378         spin_lock(&kvm->mmu_lock);
 379         young = kvm_age_hva(kvm, address);
 380         spin_unlock(&kvm->mmu_lock);
 381         srcu_read_unlock(&kvm->srcu, idx);
 382
 383         if (young)
 384                 kvm_flush_remote_tlbs(kvm);
 385
 386         return young;
 387 }
 388
 389 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 390                                        struct mm_struct *mm,
 391                                        unsigned long address)
 392 {
 393         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 394         int young, idx;
 395
 396         idx = srcu_read_lock(&kvm->srcu);
 397         spin_lock(&kvm->mmu_lock);
 398         young = kvm_test_age_hva(kvm, address);
 399         spin_unlock(&kvm->mmu_lock);
 400         srcu_read_unlock(&kvm->srcu, idx);
 401
 402         return young;
 403 }
 404
 405 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 406                                      struct mm_struct *mm)
 407 {
 408         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 409         int idx;
 410
 411         idx = srcu_read_lock(&kvm->srcu);
 412         kvm_arch_flush_shadow(kvm);
 413         srcu_read_unlock(&kvm->srcu, idx);
 414 }
 415
 416 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 417         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 418         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 419         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 420         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 421         .test_young             = kvm_mmu_notifier_test_young,
 422         .change_pte             = kvm_mmu_notifier_change_pte,
 423         .release                = kvm_mmu_notifier_release,
 424 };
 425
 426 static int kvm_init_mmu_notifier(struct kvm *kvm)
 427 {
 428         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 429         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 430 }
 431
 432 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 433
 434 static int kvm_init_mmu_notifier(struct kvm *kvm)
 435 {
 436         return 0;
 437 }
 438
 439 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 440
 441 static struct kvm *kvm_create_vm(void)
 442 {
 443         int r, i;
 444         struct kvm *kvm = kvm_arch_alloc_vm();
 445
 446         if (!kvm)
 447                 return ERR_PTR(-ENOMEM);
 448
 449         r = kvm_arch_init_vm(kvm);
 450         if (r)
 451                 goto out_err_nodisable;
 452
 453         r = hardware_enable_all();
 454         if (r)
 455                 goto out_err_nodisable;
 456
 457 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 458         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 459         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 460 #endif
 461
 462         r = -ENOMEM;
 463         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 464         if (!kvm->memslots)
 465                 goto out_err_nosrcu;
 466         if (init_srcu_struct(&kvm->srcu))
 467                 goto out_err_nosrcu;
 468         for (i = 0; i < KVM_NR_BUSES; i++) {
 469                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 470                                         GFP_KERNEL);
 471                 if (!kvm->buses[i])
 472                         goto out_err;
 473         }
 474
 475         spin_lock_init(&kvm->mmu_lock);
 476         kvm->mm = current->mm;
 477         atomic_inc(&kvm->mm->mm_count);
 478         kvm_eventfd_init(kvm);
 479         mutex_init(&kvm->lock);
 480         mutex_init(&kvm->irq_lock);
 481         mutex_init(&kvm->slots_lock);
 482         atomic_set(&kvm->users_count, 1);
 483
 484         r = kvm_init_mmu_notifier(kvm);
 485         if (r)
 486                 goto out_err;
 487
 488         raw_spin_lock(&kvm_lock);
 489         list_add(&kvm->vm_list, &vm_list);
 490         raw_spin_unlock(&kvm_lock);
 491
 492         return kvm;
 493
 494 out_err:
 495         cleanup_srcu_struct(&kvm->srcu);
 496 out_err_nosrcu:
 497         hardware_disable_all();
 498 out_err_nodisable:
 499         for (i = 0; i < KVM_NR_BUSES; i++)
 500                 kfree(kvm->buses[i]);
 501         kfree(kvm->memslots);
 502         kvm_arch_free_vm(kvm);
 503         return ERR_PTR(r);
 504 }
 505
 506 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 507 {
 508         if (!memslot->dirty_bitmap)
 509                 return;
 510
 511         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
 512                 vfree(memslot->dirty_bitmap_head);
 513         else
 514                 kfree(memslot->dirty_bitmap_head);
 515
 516         memslot->dirty_bitmap = NULL;
 517         memslot->dirty_bitmap_head = NULL;
 518 }
 519
 520 /*
 521  * Free any memory in @free but not in @dont.
 522  */
 523 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 524                                   struct kvm_memory_slot *dont)
 525 {
 526         int i;
 527
 528         if (!dont || free->rmap != dont->rmap)
 529                 vfree(free->rmap);
 530
 531         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 532                 kvm_destroy_dirty_bitmap(free);
 533
 534
 535         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 536                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 537                         vfree(free->lpage_info[i]);
 538                         free->lpage_info[i] = NULL;
 539                 }
 540         }
 541
 542         free->npages = 0;
 543         free->rmap = NULL;
 544 }
 545
 546 void kvm_free_physmem(struct kvm *kvm)
 547 {
 548         int i;
 549         struct kvm_memslots *slots = kvm->memslots;
 550
 551         for (i = 0; i < slots->nmemslots; ++i)
 552                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 553
 554         kfree(kvm->memslots);
 555 }
 556
 557 static void kvm_destroy_vm(struct kvm *kvm)
 558 {
 559         int i;
 560         struct mm_struct *mm = kvm->mm;
 561
 562         kvm_arch_sync_events(kvm);
 563         raw_spin_lock(&kvm_lock);
 564         list_del(&kvm->vm_list);
 565         raw_spin_unlock(&kvm_lock);
 566         kvm_free_irq_routing(kvm);
 567         for (i = 0; i < KVM_NR_BUSES; i++)
 568                 kvm_io_bus_destroy(kvm->buses[i]);
 569         kvm_coalesced_mmio_free(kvm);
 570 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 571         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 572 #else
 573         kvm_arch_flush_shadow(kvm);
 574 #endif
 575         kvm_arch_destroy_vm(kvm);
 576         kvm_free_physmem(kvm);
 577         cleanup_srcu_struct(&kvm->srcu);
 578         kvm_arch_free_vm(kvm);
 579         hardware_disable_all();
 580         mmdrop(mm);
 581 }
 582
 583 void kvm_get_kvm(struct kvm *kvm)
 584 {
 585         atomic_inc(&kvm->users_count);
 586 }
 587 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 588
 589 void kvm_put_kvm(struct kvm *kvm)
 590 {
 591         if (atomic_dec_and_test(&kvm->users_count))
 592                 kvm_destroy_vm(kvm);
 593 }
 594 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 595
 596
 597 static int kvm_vm_release(struct inode *inode, struct file *filp)
 598 {
 599         struct kvm *kvm = filp->private_data;
 600
 601         kvm_irqfd_release(kvm);
 602
 603         kvm_put_kvm(kvm);
 604         return 0;
 605 }
 606
 607 #ifndef CONFIG_S390
 608 /*
 609  * Allocation size is twice as large as the actual dirty bitmap size.
 610  * This makes it possible to do double buffering: see x86's
 611  * kvm_vm_ioctl_get_dirty_log().
 612  */
 613 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 614 {
 615         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 616
 617         if (dirty_bytes > PAGE_SIZE)
 618                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
 619         else
 620                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
 621
 622         if (!memslot->dirty_bitmap)
 623                 return -ENOMEM;
 624
 625         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
 626         return 0;
 627 }
 628 #endif /* !CONFIG_S390 */
 629
 630 /*
 631  * Allocate some memory and give it an address in the guest physical address
 632  * space.
 633  *
 634  * Discontiguous memory is allowed, mostly for framebuffers.
 635  *
 636  * Must be called holding mmap_sem for write.
 637  */
 638 int __kvm_set_memory_region(struct kvm *kvm,
 639                             struct kvm_userspace_memory_region *mem,
 640                             int user_alloc)
 641 {
 642         int r;
 643         gfn_t base_gfn;
 644         unsigned long npages;
 645         unsigned long i;
 646         struct kvm_memory_slot *memslot;
 647         struct kvm_memory_slot old, new;
 648         struct kvm_memslots *slots, *old_memslots;
 649
 650         r = -EINVAL;
 651         /* General sanity checks */
 652         if (mem->memory_size & (PAGE_SIZE - 1))
 653                 goto out;
 654         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 655                 goto out;
 656         /* We can read the guest memory with __xxx_user() later on. */
 657         if (user_alloc &&
 658             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 659              !access_ok(VERIFY_WRITE,
 660                         (void __user *)(unsigned long)mem->userspace_addr,
 661                         mem->memory_size)))
 662                 goto out;
 663         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 664                 goto out;
 665         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 666                 goto out;
 667
 668         memslot = &kvm->memslots->memslots[mem->slot];
 669         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 670         npages = mem->memory_size >> PAGE_SHIFT;
 671
 672         r = -EINVAL;
 673         if (npages > KVM_MEM_MAX_NR_PAGES)
 674                 goto out;
 675
 676         if (!npages)
 677                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 678
 679         new = old = *memslot;
 680
 681         new.id = mem->slot;
 682         new.base_gfn = base_gfn;
 683         new.npages = npages;
 684         new.flags = mem->flags;
 685
 686         /* Disallow changing a memory slot's size. */
 687         r = -EINVAL;
 688         if (npages && old.npages && npages != old.npages)
 689                 goto out_free;
 690
 691         /* Check for overlaps */
 692         r = -EEXIST;
 693         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 694                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 695
 696                 if (s == memslot || !s->npages)
 697                         continue;
 698                 if (!((base_gfn + npages <= s->base_gfn) ||
 699                       (base_gfn >= s->base_gfn + s->npages)))
 700                         goto out_free;
 701         }
 702
 703         /* Free page dirty bitmap if unneeded */
 704         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 705                 new.dirty_bitmap = NULL;
 706
 707         r = -ENOMEM;
 708
 709         /* Allocate if a slot is being created */
 710 #ifndef CONFIG_S390
 711         if (npages && !new.rmap) {
 712                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
 713
 714                 if (!new.rmap)
 715                         goto out_free;
 716
 717                 new.user_alloc = user_alloc;
 718                 new.userspace_addr = mem->userspace_addr;
 719         }
 720         if (!npages)
 721                 goto skip_lpage;
 722
 723         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 724                 unsigned long ugfn;
 725                 unsigned long j;
 726                 int lpages;
 727                 int level = i + 2;
 728
 729                 /* Avoid unused variable warning if no large pages */
 730                 (void)level;
 731
 732                 if (new.lpage_info[i])
 733                         continue;
 734
 735                 lpages = 1 + ((base_gfn + npages - 1)
 736                              >> KVM_HPAGE_GFN_SHIFT(level));
 737                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
 738
 739                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
 740
 741                 if (!new.lpage_info[i])
 742                         goto out_free;
 743
 744                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
 745                         new.lpage_info[i][0].write_count = 1;
 746                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
 747                         new.lpage_info[i][lpages - 1].write_count = 1;
 748                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 749                 /*
 750                  * If the gfn and userspace address are not aligned wrt each
 751                  * other, or if explicitly asked to, disable large page
 752                  * support for this slot
 753                  */
 754                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 755                     !largepages_enabled)
 756                         for (j = 0; j < lpages; ++j)
 757                                 new.lpage_info[i][j].write_count = 1;
 758         }
 759
 760 skip_lpage:
 761
 762         /* Allocate page dirty bitmap if needed */
 763         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 764                 if (kvm_create_dirty_bitmap(&new) < 0)
 765                         goto out_free;
 766                 /* destroy any largepage mappings for dirty tracking */
 767         }
 768 #else  /* not defined CONFIG_S390 */
 769         new.user_alloc = user_alloc;
 770         if (user_alloc)
 771                 new.userspace_addr = mem->userspace_addr;
 772 #endif /* not defined CONFIG_S390 */
 773
 774         if (!npages) {
 775                 r = -ENOMEM;
 776                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 777                 if (!slots)
 778                         goto out_free;
 779                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 780                 if (mem->slot >= slots->nmemslots)
 781                         slots->nmemslots = mem->slot + 1;
 782                 slots->generation++;
 783                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 784
 785                 old_memslots = kvm->memslots;
 786                 rcu_assign_pointer(kvm->memslots, slots);
 787                 synchronize_srcu_expedited(&kvm->srcu);
 788                 /* From this point no new shadow pages pointing to a deleted
 789                  * memslot will be created.
 790                  *
 791                  * validation of sp->gfn happens in:
 792                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 793                  *      - kvm_is_visible_gfn (mmu_check_roots)
 794                  */
 795                 kvm_arch_flush_shadow(kvm);
 796                 kfree(old_memslots);
 797         }
 798
 799         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 800         if (r)
 801                 goto out_free;
 802
 803         /* map the pages in iommu page table */
 804         if (npages) {
 805                 r = kvm_iommu_map_pages(kvm, &new);
 806                 if (r)
 807                         goto out_free;
 808         }
 809
 810         r = -ENOMEM;
 811         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 812         if (!slots)
 813                 goto out_free;
 814         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 815         if (mem->slot >= slots->nmemslots)
 816                 slots->nmemslots = mem->slot + 1;
 817         slots->generation++;
 818
 819         /* actual memory is freed via old in kvm_free_physmem_slot below */
 820         if (!npages) {
 821                 new.rmap = NULL;
 822                 new.dirty_bitmap = NULL;
 823                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 824                         new.lpage_info[i] = NULL;
 825         }
 826
 827         slots->memslots[mem->slot] = new;
 828         old_memslots = kvm->memslots;
 829         rcu_assign_pointer(kvm->memslots, slots);
 830         synchronize_srcu_expedited(&kvm->srcu);
 831
 832         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 833
 834         kvm_free_physmem_slot(&old, &new);
 835         kfree(old_memslots);
 836
 837         return 0;
 838
 839 out_free:
 840         kvm_free_physmem_slot(&new, &old);
 841 out:
 842         return r;
 843
 844 }
 845 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 846
 847 int kvm_set_memory_region(struct kvm *kvm,
 848                           struct kvm_userspace_memory_region *mem,
 849                           int user_alloc)
 850 {
 851         int r;
 852
 853         mutex_lock(&kvm->slots_lock);
 854         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 855         mutex_unlock(&kvm->slots_lock);
 856         return r;
 857 }
 858 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 859
 860 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 861                                    struct
 862                                    kvm_userspace_memory_region *mem,
 863                                    int user_alloc)
 864 {
 865         if (mem->slot >= KVM_MEMORY_SLOTS)
 866                 return -EINVAL;
 867         return kvm_set_memory_region(kvm, mem, user_alloc);
 868 }
 869
 870 int kvm_get_dirty_log(struct kvm *kvm,
 871                         struct kvm_dirty_log *log, int *is_dirty)
 872 {
 873         struct kvm_memory_slot *memslot;
 874         int r, i;
 875         unsigned long n;
 876         unsigned long any = 0;
 877
 878         r = -EINVAL;
 879         if (log->slot >= KVM_MEMORY_SLOTS)
 880                 goto out;
 881
 882         memslot = &kvm->memslots->memslots[log->slot];
 883         r = -ENOENT;
 884         if (!memslot->dirty_bitmap)
 885                 goto out;
 886
 887         n = kvm_dirty_bitmap_bytes(memslot);
 888
 889         for (i = 0; !any && i < n/sizeof(long); ++i)
 890                 any = memslot->dirty_bitmap[i];
 891
 892         r = -EFAULT;
 893         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 894                 goto out;
 895
 896         if (any)
 897                 *is_dirty = 1;
 898
 899         r = 0;
 900 out:
 901         return r;
 902 }
 903
 904 void kvm_disable_largepages(void)
 905 {
 906         largepages_enabled = false;
 907 }
 908 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 909
 910 int is_error_page(struct page *page)
 911 {
 912         return page == bad_page || page == hwpoison_page || page == fault_page;
 913 }
 914 EXPORT_SYMBOL_GPL(is_error_page);
 915
 916 int is_error_pfn(pfn_t pfn)
 917 {
 918         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 919 }
 920 EXPORT_SYMBOL_GPL(is_error_pfn);
 921
 922 int is_hwpoison_pfn(pfn_t pfn)
 923 {
 924         return pfn == hwpoison_pfn;
 925 }
 926 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 927
 928 int is_fault_pfn(pfn_t pfn)
 929 {
 930         return pfn == fault_pfn;
 931 }
 932 EXPORT_SYMBOL_GPL(is_fault_pfn);
 933
 934 int is_noslot_pfn(pfn_t pfn)
 935 {
 936         return pfn == bad_pfn;
 937 }
 938 EXPORT_SYMBOL_GPL(is_noslot_pfn);
 939
 940 int is_invalid_pfn(pfn_t pfn)
 941 {
 942         return pfn == hwpoison_pfn || pfn == fault_pfn;
 943 }
 944 EXPORT_SYMBOL_GPL(is_invalid_pfn);
 945
 946 static inline unsigned long bad_hva(void)
 947 {
 948         return PAGE_OFFSET;
 949 }
 950
 951 int kvm_is_error_hva(unsigned long addr)
 952 {
 953         return addr == bad_hva();
 954 }
 955 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 956
 957 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
 958                                                 gfn_t gfn)
 959 {
 960         int i;
 961
 962         for (i = 0; i < slots->nmemslots; ++i) {
 963                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 964
 965                 if (gfn >= memslot->base_gfn
 966                     && gfn < memslot->base_gfn + memslot->npages)
 967                         return memslot;
 968         }
 969         return NULL;
 970 }
 971
 972 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 973 {
 974         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 975 }
 976 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 977
 978 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 979 {
 980         int i;
 981         struct kvm_memslots *slots = kvm_memslots(kvm);
 982
 983         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 984                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 985
 986                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 987                         continue;
 988
 989                 if (gfn >= memslot->base_gfn
 990                     && gfn < memslot->base_gfn + memslot->npages)
 991                         return 1;
 992         }
 993         return 0;
 994 }
 995 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 996
 997 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 998 {
 999         struct vm_area_struct *vma;
1000         unsigned long addr, size;
1001
1002         size = PAGE_SIZE;
1003
1004         addr = gfn_to_hva(kvm, gfn);
1005         if (kvm_is_error_hva(addr))
1006                 return PAGE_SIZE;
1007
1008         down_read(&current->mm->mmap_sem);
1009         vma = find_vma(current->mm, addr);
1010         if (!vma)
1011                 goto out;
1012
1013         size = vma_kernel_pagesize(vma);
1014
1015 out:
1016         up_read(&current->mm->mmap_sem);
1017
1018         return size;
1019 }
1020
1021 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1022                                      gfn_t *nr_pages)
1023 {
1024         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1025                 return bad_hva();
1026
1027         if (nr_pages)
1028                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1029
1030         return gfn_to_hva_memslot(slot, gfn);
1031 }
1032
1033 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1034 {
1035         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1036 }
1037 EXPORT_SYMBOL_GPL(gfn_to_hva);
1038
1039 static pfn_t get_fault_pfn(void)
1040 {
1041         get_page(fault_page);
1042         return fault_pfn;
1043 }
1044
1045 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1046         unsigned long start, int write, struct page **page)
1047 {
1048         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1049
1050         if (write)
1051                 flags |= FOLL_WRITE;
1052
1053         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1054 }
1055
1056 static inline int check_user_page_hwpoison(unsigned long addr)
1057 {
1058         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1059
1060         rc = __get_user_pages(current, current->mm, addr, 1,
1061                               flags, NULL, NULL, NULL);
1062         return rc == -EHWPOISON;
1063 }
1064
1065 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1066                         bool *async, bool write_fault, bool *writable)
1067 {
1068         struct page *page[1];
1069         int npages = 0;
1070         pfn_t pfn;
1071
1072         /* we can do it either atomically or asynchronously, not both */
1073         BUG_ON(atomic && async);
1074
1075         BUG_ON(!write_fault && !writable);
1076
1077         if (writable)
1078                 *writable = true;
1079
1080         if (atomic || async)
1081                 npages = __get_user_pages_fast(addr, 1, 1, page);
1082
1083         if (unlikely(npages != 1) && !atomic) {
1084                 might_sleep();
1085
1086                 if (writable)
1087                         *writable = write_fault;
1088
1089                 if (async) {
1090                         down_read(&current->mm->mmap_sem);
1091                         npages = get_user_page_nowait(current, current->mm,
1092                                                      addr, write_fault, page);
1093                         up_read(&current->mm->mmap_sem);
1094                 } else
1095                         npages = get_user_pages_fast(addr, 1, write_fault,
1096                                                      page);
1097
1098                 /* map read fault as writable if possible */
1099                 if (unlikely(!write_fault) && npages == 1) {
1100                         struct page *wpage[1];
1101
1102                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1103                         if (npages == 1) {
1104                                 *writable = true;
1105                                 put_page(page[0]);
1106                                 page[0] = wpage[0];
1107                         }
1108                         npages = 1;
1109                 }
1110         }
1111
1112         if (unlikely(npages != 1)) {
1113                 struct vm_area_struct *vma;
1114
1115                 if (atomic)
1116                         return get_fault_pfn();
1117
1118                 down_read(&current->mm->mmap_sem);
1119                 if (npages == -EHWPOISON ||
1120                         (!async && check_user_page_hwpoison(addr))) {
1121                         up_read(&current->mm->mmap_sem);
1122                         get_page(hwpoison_page);
1123                         return page_to_pfn(hwpoison_page);
1124                 }
1125
1126                 vma = find_vma_intersection(current->mm, addr, addr+1);
1127
1128                 if (vma == NULL)
1129                         pfn = get_fault_pfn();
1130                 else if ((vma->vm_flags & VM_PFNMAP)) {
1131                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1132                                 vma->vm_pgoff;
1133                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1134                 } else {
1135                         if (async && (vma->vm_flags & VM_WRITE))
1136                                 *async = true;
1137                         pfn = get_fault_pfn();
1138                 }
1139                 up_read(&current->mm->mmap_sem);
1140         } else
1141                 pfn = page_to_pfn(page[0]);
1142
1143         return pfn;
1144 }
1145
1146 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1147 {
1148         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1149 }
1150 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1151
1152 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1153                           bool write_fault, bool *writable)
1154 {
1155         unsigned long addr;
1156
1157         if (async)
1158                 *async = false;
1159
1160         addr = gfn_to_hva(kvm, gfn);
1161         if (kvm_is_error_hva(addr)) {
1162                 get_page(bad_page);
1163                 return page_to_pfn(bad_page);
1164         }
1165
1166         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1167 }
1168
1169 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1170 {
1171         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1172 }
1173 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1174
1175 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1176                        bool write_fault, bool *writable)
1177 {
1178         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1179 }
1180 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1181
1182 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1183 {
1184         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1185 }
1186 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1187
1188 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1189                       bool *writable)
1190 {
1191         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1192 }
1193 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1194
1195 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1196                          struct kvm_memory_slot *slot, gfn_t gfn)
1197 {
1198         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1199         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1200 }
1201
1202 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1203                                                                   int nr_pages)
1204 {
1205         unsigned long addr;
1206         gfn_t entry;
1207
1208         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1209         if (kvm_is_error_hva(addr))
1210                 return -1;
1211
1212         if (entry < nr_pages)
1213                 return 0;
1214
1215         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1216 }
1217 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1218
1219 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1220 {
1221         pfn_t pfn;
1222
1223         pfn = gfn_to_pfn(kvm, gfn);
1224         if (!kvm_is_mmio_pfn(pfn))
1225                 return pfn_to_page(pfn);
1226
1227         WARN_ON(kvm_is_mmio_pfn(pfn));
1228
1229         get_page(bad_page);
1230         return bad_page;
1231 }
1232
1233 EXPORT_SYMBOL_GPL(gfn_to_page);
1234
1235 void kvm_release_page_clean(struct page *page)
1236 {
1237         kvm_release_pfn_clean(page_to_pfn(page));
1238 }
1239 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1240
1241 void kvm_release_pfn_clean(pfn_t pfn)
1242 {
1243         if (!kvm_is_mmio_pfn(pfn))
1244                 put_page(pfn_to_page(pfn));
1245 }
1246 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1247
1248 void kvm_release_page_dirty(struct page *page)
1249 {
1250         kvm_release_pfn_dirty(page_to_pfn(page));
1251 }
1252 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1253
1254 void kvm_release_pfn_dirty(pfn_t pfn)
1255 {
1256         kvm_set_pfn_dirty(pfn);
1257         kvm_release_pfn_clean(pfn);
1258 }
1259 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1260
1261 void kvm_set_page_dirty(struct page *page)
1262 {
1263         kvm_set_pfn_dirty(page_to_pfn(page));
1264 }
1265 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1266
1267 void kvm_set_pfn_dirty(pfn_t pfn)
1268 {
1269         if (!kvm_is_mmio_pfn(pfn)) {
1270                 struct page *page = pfn_to_page(pfn);
1271                 if (!PageReserved(page))
1272                         SetPageDirty(page);
1273         }
1274 }
1275 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1276
1277 void kvm_set_pfn_accessed(pfn_t pfn)
1278 {
1279         if (!kvm_is_mmio_pfn(pfn))
1280                 mark_page_accessed(pfn_to_page(pfn));
1281 }
1282 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1283
1284 void kvm_get_pfn(pfn_t pfn)
1285 {
1286         if (!kvm_is_mmio_pfn(pfn))
1287                 get_page(pfn_to_page(pfn));
1288 }
1289 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1290
1291 static int next_segment(unsigned long len, int offset)
1292 {
1293         if (len > PAGE_SIZE - offset)
1294                 return PAGE_SIZE - offset;
1295         else
1296                 return len;
1297 }
1298
1299 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1300                         int len)
1301 {
1302         int r;
1303         unsigned long addr;
1304
1305         addr = gfn_to_hva(kvm, gfn);
1306         if (kvm_is_error_hva(addr))
1307                 return -EFAULT;
1308         r = __copy_from_user(data, (void __user *)addr + offset, len);
1309         if (r)
1310                 return -EFAULT;
1311         return 0;
1312 }
1313 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1314
1315 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1316 {
1317         gfn_t gfn = gpa >> PAGE_SHIFT;
1318         int seg;
1319         int offset = offset_in_page(gpa);
1320         int ret;
1321
1322         while ((seg = next_segment(len, offset)) != 0) {
1323                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1324                 if (ret < 0)
1325                         return ret;
1326                 offset = 0;
1327                 len -= seg;
1328                 data += seg;
1329                 ++gfn;
1330         }
1331         return 0;
1332 }
1333 EXPORT_SYMBOL_GPL(kvm_read_guest);
1334
1335 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1336                           unsigned long len)
1337 {
1338         int r;
1339         unsigned long addr;
1340         gfn_t gfn = gpa >> PAGE_SHIFT;
1341         int offset = offset_in_page(gpa);
1342
1343         addr = gfn_to_hva(kvm, gfn);
1344         if (kvm_is_error_hva(addr))
1345                 return -EFAULT;
1346         pagefault_disable();
1347         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1348         pagefault_enable();
1349         if (r)
1350                 return -EFAULT;
1351         return 0;
1352 }
1353 EXPORT_SYMBOL(kvm_read_guest_atomic);
1354
1355 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1356                          int offset, int len)
1357 {
1358         int r;
1359         unsigned long addr;
1360
1361         addr = gfn_to_hva(kvm, gfn);
1362         if (kvm_is_error_hva(addr))
1363                 return -EFAULT;
1364         r = __copy_to_user((void __user *)addr + offset, data, len);
1365         if (r)
1366                 return -EFAULT;
1367         mark_page_dirty(kvm, gfn);
1368         return 0;
1369 }
1370 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1371
1372 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1373                     unsigned long len)
1374 {
1375         gfn_t gfn = gpa >> PAGE_SHIFT;
1376         int seg;
1377         int offset = offset_in_page(gpa);
1378         int ret;
1379
1380         while ((seg = next_segment(len, offset)) != 0) {
1381                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1382                 if (ret < 0)
1383                         return ret;
1384                 offset = 0;
1385                 len -= seg;
1386                 data += seg;
1387                 ++gfn;
1388         }
1389         return 0;
1390 }
1391
1392 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1393                               gpa_t gpa)
1394 {
1395         struct kvm_memslots *slots = kvm_memslots(kvm);
1396         int offset = offset_in_page(gpa);
1397         gfn_t gfn = gpa >> PAGE_SHIFT;
1398
1399         ghc->gpa = gpa;
1400         ghc->generation = slots->generation;
1401         ghc->memslot = __gfn_to_memslot(slots, gfn);
1402         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1403         if (!kvm_is_error_hva(ghc->hva))
1404                 ghc->hva += offset;
1405         else
1406                 return -EFAULT;
1407
1408         return 0;
1409 }
1410 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1411
1412 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1413                            void *data, unsigned long len)
1414 {
1415         struct kvm_memslots *slots = kvm_memslots(kvm);
1416         int r;
1417
1418         if (slots->generation != ghc->generation)
1419                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1420
1421         if (kvm_is_error_hva(ghc->hva))
1422                 return -EFAULT;
1423
1424         r = __copy_to_user((void __user *)ghc->hva, data, len);
1425         if (r)
1426                 return -EFAULT;
1427         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1428
1429         return 0;
1430 }
1431 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1432
1433 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1434                            void *data, unsigned long len)
1435 {
1436         struct kvm_memslots *slots = kvm_memslots(kvm);
1437         int r;
1438
1439         if (slots->generation != ghc->generation)
1440                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1441
1442         if (kvm_is_error_hva(ghc->hva))
1443                 return -EFAULT;
1444
1445         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1446         if (r)
1447                 return -EFAULT;
1448
1449         return 0;
1450 }
1451 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1452
1453 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1454 {
1455         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1456                                     offset, len);
1457 }
1458 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1459
1460 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1461 {
1462         gfn_t gfn = gpa >> PAGE_SHIFT;
1463         int seg;
1464         int offset = offset_in_page(gpa);
1465         int ret;
1466
1467         while ((seg = next_segment(len, offset)) != 0) {
1468                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1469                 if (ret < 0)
1470                         return ret;
1471                 offset = 0;
1472                 len -= seg;
1473                 ++gfn;
1474         }
1475         return 0;
1476 }
1477 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1478
1479 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1480                              gfn_t gfn)
1481 {
1482         if (memslot && memslot->dirty_bitmap) {
1483                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1484
1485                 __set_bit_le(rel_gfn, memslot->dirty_bitmap);
1486         }
1487 }
1488
1489 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1490 {
1491         struct kvm_memory_slot *memslot;
1492
1493         memslot = gfn_to_memslot(kvm, gfn);
1494         mark_page_dirty_in_slot(kvm, memslot, gfn);
1495 }
1496
1497 /*
1498  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1499  */
1500 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1501 {
1502         DEFINE_WAIT(wait);
1503
1504         for (;;) {
1505                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1506
1507                 if (kvm_arch_vcpu_runnable(vcpu)) {
1508                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1509                         break;
1510                 }
1511                 if (kvm_cpu_has_pending_timer(vcpu))
1512                         break;
1513                 if (signal_pending(current))
1514                         break;
1515
1516                 schedule();
1517         }
1518
1519         finish_wait(&vcpu->wq, &wait);
1520 }
1521
1522 void kvm_resched(struct kvm_vcpu *vcpu)
1523 {
1524         if (!need_resched())
1525                 return;
1526         cond_resched();
1527 }
1528 EXPORT_SYMBOL_GPL(kvm_resched);
1529
1530 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1531 {
1532         struct kvm *kvm = me->kvm;
1533         struct kvm_vcpu *vcpu;
1534         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1535         int yielded = 0;
1536         int pass;
1537         int i;
1538
1539         /*
1540          * We boost the priority of a VCPU that is runnable but not
1541          * currently running, because it got preempted by something
1542          * else and called schedule in __vcpu_run.  Hopefully that
1543          * VCPU is holding the lock that we need and will release it.
1544          * We approximate round-robin by starting at the last boosted VCPU.
1545          */
1546         for (pass = 0; pass < 2 && !yielded; pass++) {
1547                 kvm_for_each_vcpu(i, vcpu, kvm) {
1548                         struct task_struct *task = NULL;
1549                         struct pid *pid;
1550                         if (!pass && i < last_boosted_vcpu) {
1551                                 i = last_boosted_vcpu;
1552                                 continue;
1553                         } else if (pass && i > last_boosted_vcpu)
1554                                 break;
1555                         if (vcpu == me)
1556                                 continue;
1557                         if (waitqueue_active(&vcpu->wq))
1558                                 continue;
1559                         rcu_read_lock();
1560                         pid = rcu_dereference(vcpu->pid);
1561                         if (pid)
1562                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1563                         rcu_read_unlock();
1564                         if (!task)
1565                                 continue;
1566                         if (task->flags & PF_VCPU) {
1567                                 put_task_struct(task);
1568                                 continue;
1569                         }
1570                         if (yield_to(task, 1)) {
1571                                 put_task_struct(task);
1572                                 kvm->last_boosted_vcpu = i;
1573                                 yielded = 1;
1574                                 break;
1575                         }
1576                         put_task_struct(task);
1577                 }
1578         }
1579 }
1580 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1581
1582 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1583 {
1584         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1585         struct page *page;
1586
1587         if (vmf->pgoff == 0)
1588                 page = virt_to_page(vcpu->run);
1589 #ifdef CONFIG_X86
1590         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1591                 page = virt_to_page(vcpu->arch.pio_data);
1592 #endif
1593 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1594         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1595                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1596 #endif
1597         else
1598                 return VM_FAULT_SIGBUS;
1599         get_page(page);
1600         vmf->page = page;
1601         return 0;
1602 }
1603
1604 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1605         .fault = kvm_vcpu_fault,
1606 };
1607
1608 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1609 {
1610         vma->vm_ops = &kvm_vcpu_vm_ops;
1611         return 0;
1612 }
1613
1614 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1615 {
1616         struct kvm_vcpu *vcpu = filp->private_data;
1617
1618         kvm_put_kvm(vcpu->kvm);
1619         return 0;
1620 }
1621
1622 static struct file_operations kvm_vcpu_fops = {
1623         .release        = kvm_vcpu_release,
1624         .unlocked_ioctl = kvm_vcpu_ioctl,
1625 #ifdef CONFIG_COMPAT
1626         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1627 #endif
1628         .mmap           = kvm_vcpu_mmap,
1629         .llseek         = noop_llseek,
1630 };
1631
1632 /*
1633  * Allocates an inode for the vcpu.
1634  */
1635 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1636 {
1637         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1638 }
1639
1640 /*
1641  * Creates some virtual cpus.  Good luck creating more than one.
1642  */
1643 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1644 {
1645         int r;
1646         struct kvm_vcpu *vcpu, *v;
1647
1648         vcpu = kvm_arch_vcpu_create(kvm, id);
1649         if (IS_ERR(vcpu))
1650                 return PTR_ERR(vcpu);
1651
1652         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1653
1654         r = kvm_arch_vcpu_setup(vcpu);
1655         if (r)
1656                 goto vcpu_destroy;
1657
1658         mutex_lock(&kvm->lock);
1659         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1660                 r = -EINVAL;
1661                 goto unlock_vcpu_destroy;
1662         }
1663
1664         kvm_for_each_vcpu(r, v, kvm)
1665                 if (v->vcpu_id == id) {
1666                         r = -EEXIST;
1667                         goto unlock_vcpu_destroy;
1668                 }
1669
1670         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1671
1672         /* Now it's all set up, let userspace reach it */
1673         kvm_get_kvm(kvm);
1674         r = create_vcpu_fd(vcpu);
1675         if (r < 0) {
1676                 kvm_put_kvm(kvm);
1677                 goto unlock_vcpu_destroy;
1678         }
1679
1680         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1681         smp_wmb();
1682         atomic_inc(&kvm->online_vcpus);
1683
1684 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1685         if (kvm->bsp_vcpu_id == id)
1686                 kvm->bsp_vcpu = vcpu;
1687 #endif
1688         mutex_unlock(&kvm->lock);
1689         return r;
1690
1691 unlock_vcpu_destroy:
1692         mutex_unlock(&kvm->lock);
1693 vcpu_destroy:
1694         kvm_arch_vcpu_destroy(vcpu);
1695         return r;
1696 }
1697
1698 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1699 {
1700         if (sigset) {
1701                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1702                 vcpu->sigset_active = 1;
1703                 vcpu->sigset = *sigset;
1704         } else
1705                 vcpu->sigset_active = 0;
1706         return 0;
1707 }
1708
1709 static long kvm_vcpu_ioctl(struct file *filp,
1710                            unsigned int ioctl, unsigned long arg)
1711 {
1712         struct kvm_vcpu *vcpu = filp->private_data;
1713         void __user *argp = (void __user *)arg;
1714         int r;
1715         struct kvm_fpu *fpu = NULL;
1716         struct kvm_sregs *kvm_sregs = NULL;
1717
1718         if (vcpu->kvm->mm != current->mm)
1719                 return -EIO;
1720
1721 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1722         /*
1723          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1724          * so vcpu_load() would break it.
1725          */
1726         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1727                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1728 #endif
1729
1730
1731         vcpu_load(vcpu);
1732         switch (ioctl) {
1733         case KVM_RUN:
1734                 r = -EINVAL;
1735                 if (arg)
1736                         goto out;
1737                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1738                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1739                 break;
1740         case KVM_GET_REGS: {
1741                 struct kvm_regs *kvm_regs;
1742
1743                 r = -ENOMEM;
1744                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1745                 if (!kvm_regs)
1746                         goto out;
1747                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1748                 if (r)
1749                         goto out_free1;
1750                 r = -EFAULT;
1751                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1752                         goto out_free1;
1753                 r = 0;
1754 out_free1:
1755                 kfree(kvm_regs);
1756                 break;
1757         }
1758         case KVM_SET_REGS: {
1759                 struct kvm_regs *kvm_regs;
1760
1761                 r = -ENOMEM;
1762                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1763                 if (!kvm_regs)
1764                         goto out;
1765                 r = -EFAULT;
1766                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1767                         goto out_free2;
1768                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1769                 if (r)
1770                         goto out_free2;
1771                 r = 0;
1772 out_free2:
1773                 kfree(kvm_regs);
1774                 break;
1775         }
1776         case KVM_GET_SREGS: {
1777                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1778                 r = -ENOMEM;
1779                 if (!kvm_sregs)
1780                         goto out;
1781                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1782                 if (r)
1783                         goto out;
1784                 r = -EFAULT;
1785                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1786                         goto out;
1787                 r = 0;
1788                 break;
1789         }
1790         case KVM_SET_SREGS: {
1791                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1792                 r = -ENOMEM;
1793                 if (!kvm_sregs)
1794                         goto out;
1795                 r = -EFAULT;
1796                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1797                         goto out;
1798                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1799                 if (r)
1800                         goto out;
1801                 r = 0;
1802                 break;
1803         }
1804         case KVM_GET_MP_STATE: {
1805                 struct kvm_mp_state mp_state;
1806
1807                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1808                 if (r)
1809                         goto out;
1810                 r = -EFAULT;
1811                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1812                         goto out;
1813                 r = 0;
1814                 break;
1815         }
1816         case KVM_SET_MP_STATE: {
1817                 struct kvm_mp_state mp_state;
1818
1819                 r = -EFAULT;
1820                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1821                         goto out;
1822                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1823                 if (r)
1824                         goto out;
1825                 r = 0;
1826                 break;
1827         }
1828         case KVM_TRANSLATE: {
1829                 struct kvm_translation tr;
1830
1831                 r = -EFAULT;
1832                 if (copy_from_user(&tr, argp, sizeof tr))
1833                         goto out;
1834                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1835                 if (r)
1836                         goto out;
1837                 r = -EFAULT;
1838                 if (copy_to_user(argp, &tr, sizeof tr))
1839                         goto out;
1840                 r = 0;
1841                 break;
1842         }
1843         case KVM_SET_GUEST_DEBUG: {
1844                 struct kvm_guest_debug dbg;
1845
1846                 r = -EFAULT;
1847                 if (copy_from_user(&dbg, argp, sizeof dbg))
1848                         goto out;
1849                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1850                 if (r)
1851                         goto out;
1852                 r = 0;
1853                 break;
1854         }
1855         case KVM_SET_SIGNAL_MASK: {
1856                 struct kvm_signal_mask __user *sigmask_arg = argp;
1857                 struct kvm_signal_mask kvm_sigmask;
1858                 sigset_t sigset, *p;
1859
1860                 p = NULL;
1861                 if (argp) {
1862                         r = -EFAULT;
1863                         if (copy_from_user(&kvm_sigmask, argp,
1864                                            sizeof kvm_sigmask))
1865                                 goto out;
1866                         r = -EINVAL;
1867                         if (kvm_sigmask.len != sizeof sigset)
1868                                 goto out;
1869                         r = -EFAULT;
1870                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1871                                            sizeof sigset))
1872                                 goto out;
1873                         p = &sigset;
1874                 }
1875                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1876                 break;
1877         }
1878         case KVM_GET_FPU: {
1879                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1880                 r = -ENOMEM;
1881                 if (!fpu)
1882                         goto out;
1883                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1884                 if (r)
1885                         goto out;
1886                 r = -EFAULT;
1887                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1888                         goto out;
1889                 r = 0;
1890                 break;
1891         }
1892         case KVM_SET_FPU: {
1893                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1894                 r = -ENOMEM;
1895                 if (!fpu)
1896                         goto out;
1897                 r = -EFAULT;
1898                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1899                         goto out;
1900                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1901                 if (r)
1902                         goto out;
1903                 r = 0;
1904                 break;
1905         }
1906         default:
1907                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1908         }
1909 out:
1910         vcpu_put(vcpu);
1911         kfree(fpu);
1912         kfree(kvm_sregs);
1913         return r;
1914 }
1915
1916 #ifdef CONFIG_COMPAT
1917 static long kvm_vcpu_compat_ioctl(struct file *filp,
1918                                   unsigned int ioctl, unsigned long arg)
1919 {
1920         struct kvm_vcpu *vcpu = filp->private_data;
1921         void __user *argp = compat_ptr(arg);
1922         int r;
1923
1924         if (vcpu->kvm->mm != current->mm)
1925                 return -EIO;
1926
1927         switch (ioctl) {
1928         case KVM_SET_SIGNAL_MASK: {
1929                 struct kvm_signal_mask __user *sigmask_arg = argp;
1930                 struct kvm_signal_mask kvm_sigmask;
1931                 compat_sigset_t csigset;
1932                 sigset_t sigset;
1933
1934                 if (argp) {
1935                         r = -EFAULT;
1936                         if (copy_from_user(&kvm_sigmask, argp,
1937                                            sizeof kvm_sigmask))
1938                                 goto out;
1939                         r = -EINVAL;
1940                         if (kvm_sigmask.len != sizeof csigset)
1941                                 goto out;
1942                         r = -EFAULT;
1943                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1944                                            sizeof csigset))
1945                                 goto out;
1946                 }
1947                 sigset_from_compat(&sigset, &csigset);
1948                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1949                 break;
1950         }
1951         default:
1952                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1953         }
1954
1955 out:
1956         return r;
1957 }
1958 #endif
1959
1960 static long kvm_vm_ioctl(struct file *filp,
1961                            unsigned int ioctl, unsigned long arg)
1962 {
1963         struct kvm *kvm = filp->private_data;
1964         void __user *argp = (void __user *)arg;
1965         int r;
1966
1967         if (kvm->mm != current->mm)
1968                 return -EIO;
1969         switch (ioctl) {
1970         case KVM_CREATE_VCPU:
1971                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1972                 if (r < 0)
1973                         goto out;
1974                 break;
1975         case KVM_SET_USER_MEMORY_REGION: {
1976                 struct kvm_userspace_memory_region kvm_userspace_mem;
1977
1978                 r = -EFAULT;
1979                 if (copy_from_user(&kvm_userspace_mem, argp,
1980                                                 sizeof kvm_userspace_mem))
1981                         goto out;
1982
1983                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1984                 if (r)
1985                         goto out;
1986                 break;
1987         }
1988         case KVM_GET_DIRTY_LOG: {
1989                 struct kvm_dirty_log log;
1990
1991                 r = -EFAULT;
1992                 if (copy_from_user(&log, argp, sizeof log))
1993                         goto out;
1994                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1995                 if (r)
1996                         goto out;
1997                 break;
1998         }
1999 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2000         case KVM_REGISTER_COALESCED_MMIO: {
2001                 struct kvm_coalesced_mmio_zone zone;
2002                 r = -EFAULT;
2003                 if (copy_from_user(&zone, argp, sizeof zone))
2004                         goto out;
2005                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2006                 if (r)
2007                         goto out;
2008                 r = 0;
2009                 break;
2010         }
2011         case KVM_UNREGISTER_COALESCED_MMIO: {
2012                 struct kvm_coalesced_mmio_zone zone;
2013                 r = -EFAULT;
2014                 if (copy_from_user(&zone, argp, sizeof zone))
2015                         goto out;
2016                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2017                 if (r)
2018                         goto out;
2019                 r = 0;
2020                 break;
2021         }
2022 #endif
2023         case KVM_IRQFD: {
2024                 struct kvm_irqfd data;
2025
2026                 r = -EFAULT;
2027                 if (copy_from_user(&data, argp, sizeof data))
2028                         goto out;
2029                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2030                 break;
2031         }
2032         case KVM_IOEVENTFD: {
2033                 struct kvm_ioeventfd data;
2034
2035                 r = -EFAULT;
2036                 if (copy_from_user(&data, argp, sizeof data))
2037                         goto out;
2038                 r = kvm_ioeventfd(kvm, &data);
2039                 break;
2040         }
2041 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2042         case KVM_SET_BOOT_CPU_ID:
2043                 r = 0;
2044                 mutex_lock(&kvm->lock);
2045                 if (atomic_read(&kvm->online_vcpus) != 0)
2046                         r = -EBUSY;
2047                 else
2048                         kvm->bsp_vcpu_id = arg;
2049                 mutex_unlock(&kvm->lock);
2050                 break;
2051 #endif
2052         default:
2053                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2054                 if (r == -ENOTTY)
2055                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2056         }
2057 out:
2058         return r;
2059 }
2060
2061 #ifdef CONFIG_COMPAT
2062 struct compat_kvm_dirty_log {
2063         __u32 slot;
2064         __u32 padding1;
2065         union {
2066                 compat_uptr_t dirty_bitmap; /* one bit per page */
2067                 __u64 padding2;
2068         };
2069 };
2070
2071 static long kvm_vm_compat_ioctl(struct file *filp,
2072                            unsigned int ioctl, unsigned long arg)
2073 {
2074         struct kvm *kvm = filp->private_data;
2075         int r;
2076
2077         if (kvm->mm != current->mm)
2078                 return -EIO;
2079         switch (ioctl) {
2080         case KVM_GET_DIRTY_LOG: {
2081                 struct compat_kvm_dirty_log compat_log;
2082                 struct kvm_dirty_log log;
2083
2084                 r = -EFAULT;
2085                 if (copy_from_user(&compat_log, (void __user *)arg,
2086                                    sizeof(compat_log)))
2087                         goto out;
2088                 log.slot         = compat_log.slot;
2089                 log.padding1     = compat_log.padding1;
2090                 log.padding2     = compat_log.padding2;
2091                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2092
2093                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2094                 if (r)
2095                         goto out;
2096                 break;
2097         }
2098         default:
2099                 r = kvm_vm_ioctl(filp, ioctl, arg);
2100         }
2101
2102 out:
2103         return r;
2104 }
2105 #endif
2106
2107 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2108 {
2109         struct page *page[1];
2110         unsigned long addr;
2111         int npages;
2112         gfn_t gfn = vmf->pgoff;
2113         struct kvm *kvm = vma->vm_file->private_data;
2114
2115         addr = gfn_to_hva(kvm, gfn);
2116         if (kvm_is_error_hva(addr))
2117                 return VM_FAULT_SIGBUS;
2118
2119         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2120                                 NULL);
2121         if (unlikely(npages != 1))
2122                 return VM_FAULT_SIGBUS;
2123
2124         vmf->page = page[0];
2125         return 0;
2126 }
2127
2128 static const struct vm_operations_struct kvm_vm_vm_ops = {
2129         .fault = kvm_vm_fault,
2130 };
2131
2132 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2133 {
2134         vma->vm_ops = &kvm_vm_vm_ops;
2135         return 0;
2136 }
2137
2138 static struct file_operations kvm_vm_fops = {
2139         .release        = kvm_vm_release,
2140         .unlocked_ioctl = kvm_vm_ioctl,
2141 #ifdef CONFIG_COMPAT
2142         .compat_ioctl   = kvm_vm_compat_ioctl,
2143 #endif
2144         .mmap           = kvm_vm_mmap,
2145         .llseek         = noop_llseek,
2146 };
2147
2148 static int kvm_dev_ioctl_create_vm(void)
2149 {
2150         int r;
2151         struct kvm *kvm;
2152
2153         kvm = kvm_create_vm();
2154         if (IS_ERR(kvm))
2155                 return PTR_ERR(kvm);
2156 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2157         r = kvm_coalesced_mmio_init(kvm);
2158         if (r < 0) {
2159                 kvm_put_kvm(kvm);
2160                 return r;
2161         }
2162 #endif
2163         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2164         if (r < 0)
2165                 kvm_put_kvm(kvm);
2166
2167         return r;
2168 }
2169
2170 static long kvm_dev_ioctl_check_extension_generic(long arg)
2171 {
2172         switch (arg) {
2173         case KVM_CAP_USER_MEMORY:
2174         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2175         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2176 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2177         case KVM_CAP_SET_BOOT_CPU_ID:
2178 #endif
2179         case KVM_CAP_INTERNAL_ERROR_DATA:
2180                 return 1;
2181 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2182         case KVM_CAP_IRQ_ROUTING:
2183                 return KVM_MAX_IRQ_ROUTES;
2184 #endif
2185         default:
2186                 break;
2187         }
2188         return kvm_dev_ioctl_check_extension(arg);
2189 }
2190
2191 static long kvm_dev_ioctl(struct file *filp,
2192                           unsigned int ioctl, unsigned long arg)
2193 {
2194         long r = -EINVAL;
2195
2196         switch (ioctl) {
2197         case KVM_GET_API_VERSION:
2198                 r = -EINVAL;
2199                 if (arg)
2200                         goto out;
2201                 r = KVM_API_VERSION;
2202                 break;
2203         case KVM_CREATE_VM:
2204                 r = -EINVAL;
2205                 if (arg)
2206                         goto out;
2207                 r = kvm_dev_ioctl_create_vm();
2208                 break;
2209         case KVM_CHECK_EXTENSION:
2210                 r = kvm_dev_ioctl_check_extension_generic(arg);
2211                 break;
2212         case KVM_GET_VCPU_MMAP_SIZE:
2213                 r = -EINVAL;
2214                 if (arg)
2215                         goto out;
2216                 r = PAGE_SIZE;     /* struct kvm_run */
2217 #ifdef CONFIG_X86
2218                 r += PAGE_SIZE;    /* pio data page */
2219 #endif
2220 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2221                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2222 #endif
2223                 break;
2224         case KVM_TRACE_ENABLE:
2225         case KVM_TRACE_PAUSE:
2226         case KVM_TRACE_DISABLE:
2227                 r = -EOPNOTSUPP;
2228                 break;
2229         default:
2230                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2231         }
2232 out:
2233         return r;
2234 }
2235
2236 static struct file_operations kvm_chardev_ops = {
2237         .unlocked_ioctl = kvm_dev_ioctl,
2238         .compat_ioctl   = kvm_dev_ioctl,
2239         .llseek         = noop_llseek,
2240 };
2241
2242 static struct miscdevice kvm_dev = {
2243         KVM_MINOR,
2244         "kvm",
2245         &kvm_chardev_ops,
2246 };
2247
2248 static void hardware_enable_nolock(void *junk)
2249 {
2250         int cpu = raw_smp_processor_id();
2251         int r;
2252
2253         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2254                 return;
2255
2256         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2257
2258         r = kvm_arch_hardware_enable(NULL);
2259
2260         if (r) {
2261                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2262                 atomic_inc(&hardware_enable_failed);
2263                 printk(KERN_INFO "kvm: enabling virtualization on "
2264                                  "CPU%d failed\n", cpu);
2265         }
2266 }
2267
2268 static void hardware_enable(void *junk)
2269 {
2270         raw_spin_lock(&kvm_lock);
2271         hardware_enable_nolock(junk);
2272         raw_spin_unlock(&kvm_lock);
2273 }
2274
2275 static void hardware_disable_nolock(void *junk)
2276 {
2277         int cpu = raw_smp_processor_id();
2278
2279         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2280                 return;
2281         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2282         kvm_arch_hardware_disable(NULL);
2283 }
2284
2285 static void hardware_disable(void *junk)
2286 {
2287         raw_spin_lock(&kvm_lock);
2288         hardware_disable_nolock(junk);
2289         raw_spin_unlock(&kvm_lock);
2290 }
2291
2292 static void hardware_disable_all_nolock(void)
2293 {
2294         BUG_ON(!kvm_usage_count);
2295
2296         kvm_usage_count--;
2297         if (!kvm_usage_count)
2298                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2299 }
2300
2301 static void hardware_disable_all(void)
2302 {
2303         raw_spin_lock(&kvm_lock);
2304         hardware_disable_all_nolock();
2305         raw_spin_unlock(&kvm_lock);
2306 }
2307
2308 static int hardware_enable_all(void)
2309 {
2310         int r = 0;
2311
2312         raw_spin_lock(&kvm_lock);
2313
2314         kvm_usage_count++;
2315         if (kvm_usage_count == 1) {
2316                 atomic_set(&hardware_enable_failed, 0);
2317                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2318
2319                 if (atomic_read(&hardware_enable_failed)) {
2320                         hardware_disable_all_nolock();
2321                         r = -EBUSY;
2322                 }
2323         }
2324
2325         raw_spin_unlock(&kvm_lock);
2326
2327         return r;
2328 }
2329
2330 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2331                            void *v)
2332 {
2333         int cpu = (long)v;
2334
2335         if (!kvm_usage_count)
2336                 return NOTIFY_OK;
2337
2338         val &= ~CPU_TASKS_FROZEN;
2339         switch (val) {
2340         case CPU_DYING:
2341                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2342                        cpu);
2343                 hardware_disable(NULL);
2344                 break;
2345         case CPU_STARTING:
2346                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2347                        cpu);
2348                 hardware_enable(NULL);
2349                 break;
2350         }
2351         return NOTIFY_OK;
2352 }
2353
2354
2355 asmlinkage void kvm_spurious_fault(void)
2356 {
2357         /* Fault while not rebooting.  We want the trace. */
2358         BUG();
2359 }
2360 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2361
2362 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2363                       void *v)
2364 {
2365         /*
2366          * Some (well, at least mine) BIOSes hang on reboot if
2367          * in vmx root mode.
2368          *
2369          * And Intel TXT required VMX off for all cpu when system shutdown.
2370          */
2371         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2372         kvm_rebooting = true;
2373         on_each_cpu(hardware_disable_nolock, NULL, 1);
2374         return NOTIFY_OK;
2375 }
2376
2377 static struct notifier_block kvm_reboot_notifier = {
2378         .notifier_call = kvm_reboot,
2379         .priority = 0,
2380 };
2381
2382 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2383 {
2384         int i;
2385
2386         for (i = 0; i < bus->dev_count; i++) {
2387                 struct kvm_io_device *pos = bus->devs[i];
2388
2389                 kvm_iodevice_destructor(pos);
2390         }
2391         kfree(bus);
2392 }
2393
2394 /* kvm_io_bus_write - called under kvm->slots_lock */
2395 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2396                      int len, const void *val)
2397 {
2398         int i;
2399         struct kvm_io_bus *bus;
2400
2401         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2402         for (i = 0; i < bus->dev_count; i++)
2403                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2404                         return 0;
2405         return -EOPNOTSUPP;
2406 }
2407
2408 /* kvm_io_bus_read - called under kvm->slots_lock */
2409 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2410                     int len, void *val)
2411 {
2412         int i;
2413         struct kvm_io_bus *bus;
2414
2415         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2416         for (i = 0; i < bus->dev_count; i++)
2417                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2418                         return 0;
2419         return -EOPNOTSUPP;
2420 }
2421
2422 /* Caller must hold slots_lock. */
2423 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2424                             struct kvm_io_device *dev)
2425 {
2426         struct kvm_io_bus *new_bus, *bus;
2427
2428         bus = kvm->buses[bus_idx];
2429         if (bus->dev_count > NR_IOBUS_DEVS-1)
2430                 return -ENOSPC;
2431
2432         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2433         if (!new_bus)
2434                 return -ENOMEM;
2435         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2436         new_bus->devs[new_bus->dev_count++] = dev;
2437         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2438         synchronize_srcu_expedited(&kvm->srcu);
2439         kfree(bus);
2440
2441         return 0;
2442 }
2443
2444 /* Caller must hold slots_lock. */
2445 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2446                               struct kvm_io_device *dev)
2447 {
2448         int i, r;
2449         struct kvm_io_bus *new_bus, *bus;
2450
2451         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2452         if (!new_bus)
2453                 return -ENOMEM;
2454
2455         bus = kvm->buses[bus_idx];
2456         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2457
2458         r = -ENOENT;
2459         for (i = 0; i < new_bus->dev_count; i++)
2460                 if (new_bus->devs[i] == dev) {
2461                         r = 0;
2462                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2463                         break;
2464                 }
2465
2466         if (r) {
2467                 kfree(new_bus);
2468                 return r;
2469         }
2470
2471         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2472         synchronize_srcu_expedited(&kvm->srcu);
2473         kfree(bus);
2474         return r;
2475 }
2476
2477 static struct notifier_block kvm_cpu_notifier = {
2478         .notifier_call = kvm_cpu_hotplug,
2479 };
2480
2481 static int vm_stat_get(void *_offset, u64 *val)
2482 {
2483         unsigned offset = (long)_offset;
2484         struct kvm *kvm;
2485
2486         *val = 0;
2487         raw_spin_lock(&kvm_lock);
2488         list_for_each_entry(kvm, &vm_list, vm_list)
2489                 *val += *(u32 *)((void *)kvm + offset);
2490         raw_spin_unlock(&kvm_lock);
2491         return 0;
2492 }
2493
2494 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2495
2496 static int vcpu_stat_get(void *_offset, u64 *val)
2497 {
2498         unsigned offset = (long)_offset;
2499         struct kvm *kvm;
2500         struct kvm_vcpu *vcpu;
2501         int i;
2502
2503         *val = 0;
2504         raw_spin_lock(&kvm_lock);
2505         list_for_each_entry(kvm, &vm_list, vm_list)
2506                 kvm_for_each_vcpu(i, vcpu, kvm)
2507                         *val += *(u32 *)((void *)vcpu + offset);
2508
2509         raw_spin_unlock(&kvm_lock);
2510         return 0;
2511 }
2512
2513 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2514
2515 static const struct file_operations *stat_fops[] = {
2516         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2517         [KVM_STAT_VM]   = &vm_stat_fops,
2518 };
2519
2520 static void kvm_init_debug(void)
2521 {
2522         struct kvm_stats_debugfs_item *p;
2523
2524         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2525         for (p = debugfs_entries; p->name; ++p)
2526                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2527                                                 (void *)(long)p->offset,
2528                                                 stat_fops[p->kind]);
2529 }
2530
2531 static void kvm_exit_debug(void)
2532 {
2533         struct kvm_stats_debugfs_item *p;
2534
2535         for (p = debugfs_entries; p->name; ++p)
2536                 debugfs_remove(p->dentry);
2537         debugfs_remove(kvm_debugfs_dir);
2538 }
2539
2540 static int kvm_suspend(void)
2541 {
2542         if (kvm_usage_count)
2543                 hardware_disable_nolock(NULL);
2544         return 0;
2545 }
2546
2547 static void kvm_resume(void)
2548 {
2549         if (kvm_usage_count) {
2550                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2551                 hardware_enable_nolock(NULL);
2552         }
2553 }
2554
2555 static struct syscore_ops kvm_syscore_ops = {
2556         .suspend = kvm_suspend,
2557         .resume = kvm_resume,
2558 };
2559
2560 struct page *bad_page;
2561 pfn_t bad_pfn;
2562
2563 static inline
2564 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2565 {
2566         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2567 }
2568
2569 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2570 {
2571         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2572
2573         kvm_arch_vcpu_load(vcpu, cpu);
2574 }
2575
2576 static void kvm_sched_out(struct preempt_notifier *pn,
2577                           struct task_struct *next)
2578 {
2579         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2580
2581         kvm_arch_vcpu_put(vcpu);
2582 }
2583
2584 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2585                   struct module *module)
2586 {
2587         int r;
2588         int cpu;
2589
2590         r = kvm_arch_init(opaque);
2591         if (r)
2592                 goto out_fail;
2593
2594         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2595
2596         if (bad_page == NULL) {
2597                 r = -ENOMEM;
2598                 goto out;
2599         }
2600
2601         bad_pfn = page_to_pfn(bad_page);
2602
2603         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2604
2605         if (hwpoison_page == NULL) {
2606                 r = -ENOMEM;
2607                 goto out_free_0;
2608         }
2609
2610         hwpoison_pfn = page_to_pfn(hwpoison_page);
2611
2612         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2613
2614         if (fault_page == NULL) {
2615                 r = -ENOMEM;
2616                 goto out_free_0;
2617         }
2618
2619         fault_pfn = page_to_pfn(fault_page);
2620
2621         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2622                 r = -ENOMEM;
2623                 goto out_free_0;
2624         }
2625
2626         r = kvm_arch_hardware_setup();
2627         if (r < 0)
2628                 goto out_free_0a;
2629
2630         for_each_online_cpu(cpu) {
2631                 smp_call_function_single(cpu,
2632                                 kvm_arch_check_processor_compat,
2633                                 &r, 1);
2634                 if (r < 0)
2635                         goto out_free_1;
2636         }
2637
2638         r = register_cpu_notifier(&kvm_cpu_notifier);
2639         if (r)
2640                 goto out_free_2;
2641         register_reboot_notifier(&kvm_reboot_notifier);
2642
2643         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2644         if (!vcpu_align)
2645                 vcpu_align = __alignof__(struct kvm_vcpu);
2646         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2647                                            0, NULL);
2648         if (!kvm_vcpu_cache) {
2649                 r = -ENOMEM;
2650                 goto out_free_3;
2651         }
2652
2653         r = kvm_async_pf_init();
2654         if (r)
2655                 goto out_free;
2656
2657         kvm_chardev_ops.owner = module;
2658         kvm_vm_fops.owner = module;
2659         kvm_vcpu_fops.owner = module;
2660
2661         r = misc_register(&kvm_dev);
2662         if (r) {
2663                 printk(KERN_ERR "kvm: misc device register failed\n");
2664                 goto out_unreg;
2665         }
2666
2667         register_syscore_ops(&kvm_syscore_ops);
2668
2669         kvm_preempt_ops.sched_in = kvm_sched_in;
2670         kvm_preempt_ops.sched_out = kvm_sched_out;
2671
2672         kvm_init_debug();
2673
2674         return 0;
2675
2676 out_unreg:
2677         kvm_async_pf_deinit();
2678 out_free:
2679         kmem_cache_destroy(kvm_vcpu_cache);
2680 out_free_3:
2681         unregister_reboot_notifier(&kvm_reboot_notifier);
2682         unregister_cpu_notifier(&kvm_cpu_notifier);
2683 out_free_2:
2684 out_free_1:
2685         kvm_arch_hardware_unsetup();
2686 out_free_0a:
2687         free_cpumask_var(cpus_hardware_enabled);
2688 out_free_0:
2689         if (fault_page)
2690                 __free_page(fault_page);
2691         if (hwpoison_page)
2692                 __free_page(hwpoison_page);
2693         __free_page(bad_page);
2694 out:
2695         kvm_arch_exit();
2696 out_fail:
2697         return r;
2698 }
2699 EXPORT_SYMBOL_GPL(kvm_init);
2700
2701 void kvm_exit(void)
2702 {
2703         kvm_exit_debug();
2704         misc_deregister(&kvm_dev);
2705         kmem_cache_destroy(kvm_vcpu_cache);
2706         kvm_async_pf_deinit();
2707         unregister_syscore_ops(&kvm_syscore_ops);
2708         unregister_reboot_notifier(&kvm_reboot_notifier);
2709         unregister_cpu_notifier(&kvm_cpu_notifier);
2710         on_each_cpu(hardware_disable_nolock, NULL, 1);
2711         kvm_arch_hardware_unsetup();
2712         kvm_arch_exit();
2713         free_cpumask_var(cpus_hardware_enabled);
2714         __free_page(hwpoison_page);
2715         __free_page(bad_page);
2716 }
2717 EXPORT_SYMBOL_GPL(kvm_exit);