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         /*
 835          * If the new memory slot is created, we need to clear all
 836          * mmio sptes.
 837          */
 838         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 839                 kvm_arch_flush_shadow(kvm);
 840
 841         kvm_free_physmem_slot(&old, &new);
 842         kfree(old_memslots);
 843
 844         return 0;
 845
 846 out_free:
 847         kvm_free_physmem_slot(&new, &old);
 848 out:
 849         return r;
 850
 851 }
 852 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 853
 854 int kvm_set_memory_region(struct kvm *kvm,
 855                           struct kvm_userspace_memory_region *mem,
 856                           int user_alloc)
 857 {
 858         int r;
 859
 860         mutex_lock(&kvm->slots_lock);
 861         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 862         mutex_unlock(&kvm->slots_lock);
 863         return r;
 864 }
 865 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 866
 867 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 868                                    struct
 869                                    kvm_userspace_memory_region *mem,
 870                                    int user_alloc)
 871 {
 872         if (mem->slot >= KVM_MEMORY_SLOTS)
 873                 return -EINVAL;
 874         return kvm_set_memory_region(kvm, mem, user_alloc);
 875 }
 876
 877 int kvm_get_dirty_log(struct kvm *kvm,
 878                         struct kvm_dirty_log *log, int *is_dirty)
 879 {
 880         struct kvm_memory_slot *memslot;
 881         int r, i;
 882         unsigned long n;
 883         unsigned long any = 0;
 884
 885         r = -EINVAL;
 886         if (log->slot >= KVM_MEMORY_SLOTS)
 887                 goto out;
 888
 889         memslot = &kvm->memslots->memslots[log->slot];
 890         r = -ENOENT;
 891         if (!memslot->dirty_bitmap)
 892                 goto out;
 893
 894         n = kvm_dirty_bitmap_bytes(memslot);
 895
 896         for (i = 0; !any && i < n/sizeof(long); ++i)
 897                 any = memslot->dirty_bitmap[i];
 898
 899         r = -EFAULT;
 900         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 901                 goto out;
 902
 903         if (any)
 904                 *is_dirty = 1;
 905
 906         r = 0;
 907 out:
 908         return r;
 909 }
 910
 911 void kvm_disable_largepages(void)
 912 {
 913         largepages_enabled = false;
 914 }
 915 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 916
 917 int is_error_page(struct page *page)
 918 {
 919         return page == bad_page || page == hwpoison_page || page == fault_page;
 920 }
 921 EXPORT_SYMBOL_GPL(is_error_page);
 922
 923 int is_error_pfn(pfn_t pfn)
 924 {
 925         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 926 }
 927 EXPORT_SYMBOL_GPL(is_error_pfn);
 928
 929 int is_hwpoison_pfn(pfn_t pfn)
 930 {
 931         return pfn == hwpoison_pfn;
 932 }
 933 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 934
 935 int is_fault_pfn(pfn_t pfn)
 936 {
 937         return pfn == fault_pfn;
 938 }
 939 EXPORT_SYMBOL_GPL(is_fault_pfn);
 940
 941 int is_noslot_pfn(pfn_t pfn)
 942 {
 943         return pfn == bad_pfn;
 944 }
 945 EXPORT_SYMBOL_GPL(is_noslot_pfn);
 946
 947 int is_invalid_pfn(pfn_t pfn)
 948 {
 949         return pfn == hwpoison_pfn || pfn == fault_pfn;
 950 }
 951 EXPORT_SYMBOL_GPL(is_invalid_pfn);
 952
 953 static inline unsigned long bad_hva(void)
 954 {
 955         return PAGE_OFFSET;
 956 }
 957
 958 int kvm_is_error_hva(unsigned long addr)
 959 {
 960         return addr == bad_hva();
 961 }
 962 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 963
 964 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
 965                                                 gfn_t gfn)
 966 {
 967         int i;
 968
 969         for (i = 0; i < slots->nmemslots; ++i) {
 970                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 971
 972                 if (gfn >= memslot->base_gfn
 973                     && gfn < memslot->base_gfn + memslot->npages)
 974                         return memslot;
 975         }
 976         return NULL;
 977 }
 978
 979 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 980 {
 981         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 982 }
 983 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 984
 985 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 986 {
 987         int i;
 988         struct kvm_memslots *slots = kvm_memslots(kvm);
 989
 990         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 991                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 992
 993                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 994                         continue;
 995
 996                 if (gfn >= memslot->base_gfn
 997                     && gfn < memslot->base_gfn + memslot->npages)
 998                         return 1;
 999         }
1000         return 0;
1001 }
1002 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1003
1004 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1005 {
1006         struct vm_area_struct *vma;
1007         unsigned long addr, size;
1008
1009         size = PAGE_SIZE;
1010
1011         addr = gfn_to_hva(kvm, gfn);
1012         if (kvm_is_error_hva(addr))
1013                 return PAGE_SIZE;
1014
1015         down_read(&current->mm->mmap_sem);
1016         vma = find_vma(current->mm, addr);
1017         if (!vma)
1018                 goto out;
1019
1020         size = vma_kernel_pagesize(vma);
1021
1022 out:
1023         up_read(&current->mm->mmap_sem);
1024
1025         return size;
1026 }
1027
1028 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1029                                      gfn_t *nr_pages)
1030 {
1031         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1032                 return bad_hva();
1033
1034         if (nr_pages)
1035                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1036
1037         return gfn_to_hva_memslot(slot, gfn);
1038 }
1039
1040 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1041 {
1042         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1043 }
1044 EXPORT_SYMBOL_GPL(gfn_to_hva);
1045
1046 static pfn_t get_fault_pfn(void)
1047 {
1048         get_page(fault_page);
1049         return fault_pfn;
1050 }
1051
1052 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1053         unsigned long start, int write, struct page **page)
1054 {
1055         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1056
1057         if (write)
1058                 flags |= FOLL_WRITE;
1059
1060         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1061 }
1062
1063 static inline int check_user_page_hwpoison(unsigned long addr)
1064 {
1065         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1066
1067         rc = __get_user_pages(current, current->mm, addr, 1,
1068                               flags, NULL, NULL, NULL);
1069         return rc == -EHWPOISON;
1070 }
1071
1072 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1073                         bool *async, bool write_fault, bool *writable)
1074 {
1075         struct page *page[1];
1076         int npages = 0;
1077         pfn_t pfn;
1078
1079         /* we can do it either atomically or asynchronously, not both */
1080         BUG_ON(atomic && async);
1081
1082         BUG_ON(!write_fault && !writable);
1083
1084         if (writable)
1085                 *writable = true;
1086
1087         if (atomic || async)
1088                 npages = __get_user_pages_fast(addr, 1, 1, page);
1089
1090         if (unlikely(npages != 1) && !atomic) {
1091                 might_sleep();
1092
1093                 if (writable)
1094                         *writable = write_fault;
1095
1096                 if (async) {
1097                         down_read(&current->mm->mmap_sem);
1098                         npages = get_user_page_nowait(current, current->mm,
1099                                                      addr, write_fault, page);
1100                         up_read(&current->mm->mmap_sem);
1101                 } else
1102                         npages = get_user_pages_fast(addr, 1, write_fault,
1103                                                      page);
1104
1105                 /* map read fault as writable if possible */
1106                 if (unlikely(!write_fault) && npages == 1) {
1107                         struct page *wpage[1];
1108
1109                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1110                         if (npages == 1) {
1111                                 *writable = true;
1112                                 put_page(page[0]);
1113                                 page[0] = wpage[0];
1114                         }
1115                         npages = 1;
1116                 }
1117         }
1118
1119         if (unlikely(npages != 1)) {
1120                 struct vm_area_struct *vma;
1121
1122                 if (atomic)
1123                         return get_fault_pfn();
1124
1125                 down_read(&current->mm->mmap_sem);
1126                 if (npages == -EHWPOISON ||
1127                         (!async && check_user_page_hwpoison(addr))) {
1128                         up_read(&current->mm->mmap_sem);
1129                         get_page(hwpoison_page);
1130                         return page_to_pfn(hwpoison_page);
1131                 }
1132
1133                 vma = find_vma_intersection(current->mm, addr, addr+1);
1134
1135                 if (vma == NULL)
1136                         pfn = get_fault_pfn();
1137                 else if ((vma->vm_flags & VM_PFNMAP)) {
1138                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1139                                 vma->vm_pgoff;
1140                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1141                 } else {
1142                         if (async && (vma->vm_flags & VM_WRITE))
1143                                 *async = true;
1144                         pfn = get_fault_pfn();
1145                 }
1146                 up_read(&current->mm->mmap_sem);
1147         } else
1148                 pfn = page_to_pfn(page[0]);
1149
1150         return pfn;
1151 }
1152
1153 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1154 {
1155         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1156 }
1157 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1158
1159 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1160                           bool write_fault, bool *writable)
1161 {
1162         unsigned long addr;
1163
1164         if (async)
1165                 *async = false;
1166
1167         addr = gfn_to_hva(kvm, gfn);
1168         if (kvm_is_error_hva(addr)) {
1169                 get_page(bad_page);
1170                 return page_to_pfn(bad_page);
1171         }
1172
1173         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1174 }
1175
1176 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1177 {
1178         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1179 }
1180 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1181
1182 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1183                        bool write_fault, bool *writable)
1184 {
1185         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1186 }
1187 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1188
1189 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1190 {
1191         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1192 }
1193 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1194
1195 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1196                       bool *writable)
1197 {
1198         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1199 }
1200 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1201
1202 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1203                          struct kvm_memory_slot *slot, gfn_t gfn)
1204 {
1205         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1206         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1207 }
1208
1209 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1210                                                                   int nr_pages)
1211 {
1212         unsigned long addr;
1213         gfn_t entry;
1214
1215         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1216         if (kvm_is_error_hva(addr))
1217                 return -1;
1218
1219         if (entry < nr_pages)
1220                 return 0;
1221
1222         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1223 }
1224 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1225
1226 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1227 {
1228         pfn_t pfn;
1229
1230         pfn = gfn_to_pfn(kvm, gfn);
1231         if (!kvm_is_mmio_pfn(pfn))
1232                 return pfn_to_page(pfn);
1233
1234         WARN_ON(kvm_is_mmio_pfn(pfn));
1235
1236         get_page(bad_page);
1237         return bad_page;
1238 }
1239
1240 EXPORT_SYMBOL_GPL(gfn_to_page);
1241
1242 void kvm_release_page_clean(struct page *page)
1243 {
1244         kvm_release_pfn_clean(page_to_pfn(page));
1245 }
1246 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1247
1248 void kvm_release_pfn_clean(pfn_t pfn)
1249 {
1250         if (!kvm_is_mmio_pfn(pfn))
1251                 put_page(pfn_to_page(pfn));
1252 }
1253 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1254
1255 void kvm_release_page_dirty(struct page *page)
1256 {
1257         kvm_release_pfn_dirty(page_to_pfn(page));
1258 }
1259 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1260
1261 void kvm_release_pfn_dirty(pfn_t pfn)
1262 {
1263         kvm_set_pfn_dirty(pfn);
1264         kvm_release_pfn_clean(pfn);
1265 }
1266 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1267
1268 void kvm_set_page_dirty(struct page *page)
1269 {
1270         kvm_set_pfn_dirty(page_to_pfn(page));
1271 }
1272 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1273
1274 void kvm_set_pfn_dirty(pfn_t pfn)
1275 {
1276         if (!kvm_is_mmio_pfn(pfn)) {
1277                 struct page *page = pfn_to_page(pfn);
1278                 if (!PageReserved(page))
1279                         SetPageDirty(page);
1280         }
1281 }
1282 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1283
1284 void kvm_set_pfn_accessed(pfn_t pfn)
1285 {
1286         if (!kvm_is_mmio_pfn(pfn))
1287                 mark_page_accessed(pfn_to_page(pfn));
1288 }
1289 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1290
1291 void kvm_get_pfn(pfn_t pfn)
1292 {
1293         if (!kvm_is_mmio_pfn(pfn))
1294                 get_page(pfn_to_page(pfn));
1295 }
1296 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1297
1298 static int next_segment(unsigned long len, int offset)
1299 {
1300         if (len > PAGE_SIZE - offset)
1301                 return PAGE_SIZE - offset;
1302         else
1303                 return len;
1304 }
1305
1306 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1307                         int len)
1308 {
1309         int r;
1310         unsigned long addr;
1311
1312         addr = gfn_to_hva(kvm, gfn);
1313         if (kvm_is_error_hva(addr))
1314                 return -EFAULT;
1315         r = __copy_from_user(data, (void __user *)addr + offset, len);
1316         if (r)
1317                 return -EFAULT;
1318         return 0;
1319 }
1320 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1321
1322 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1323 {
1324         gfn_t gfn = gpa >> PAGE_SHIFT;
1325         int seg;
1326         int offset = offset_in_page(gpa);
1327         int ret;
1328
1329         while ((seg = next_segment(len, offset)) != 0) {
1330                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1331                 if (ret < 0)
1332                         return ret;
1333                 offset = 0;
1334                 len -= seg;
1335                 data += seg;
1336                 ++gfn;
1337         }
1338         return 0;
1339 }
1340 EXPORT_SYMBOL_GPL(kvm_read_guest);
1341
1342 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1343                           unsigned long len)
1344 {
1345         int r;
1346         unsigned long addr;
1347         gfn_t gfn = gpa >> PAGE_SHIFT;
1348         int offset = offset_in_page(gpa);
1349
1350         addr = gfn_to_hva(kvm, gfn);
1351         if (kvm_is_error_hva(addr))
1352                 return -EFAULT;
1353         pagefault_disable();
1354         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1355         pagefault_enable();
1356         if (r)
1357                 return -EFAULT;
1358         return 0;
1359 }
1360 EXPORT_SYMBOL(kvm_read_guest_atomic);
1361
1362 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1363                          int offset, int len)
1364 {
1365         int r;
1366         unsigned long addr;
1367
1368         addr = gfn_to_hva(kvm, gfn);
1369         if (kvm_is_error_hva(addr))
1370                 return -EFAULT;
1371         r = __copy_to_user((void __user *)addr + offset, data, len);
1372         if (r)
1373                 return -EFAULT;
1374         mark_page_dirty(kvm, gfn);
1375         return 0;
1376 }
1377 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1378
1379 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1380                     unsigned long len)
1381 {
1382         gfn_t gfn = gpa >> PAGE_SHIFT;
1383         int seg;
1384         int offset = offset_in_page(gpa);
1385         int ret;
1386
1387         while ((seg = next_segment(len, offset)) != 0) {
1388                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1389                 if (ret < 0)
1390                         return ret;
1391                 offset = 0;
1392                 len -= seg;
1393                 data += seg;
1394                 ++gfn;
1395         }
1396         return 0;
1397 }
1398
1399 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1400                               gpa_t gpa)
1401 {
1402         struct kvm_memslots *slots = kvm_memslots(kvm);
1403         int offset = offset_in_page(gpa);
1404         gfn_t gfn = gpa >> PAGE_SHIFT;
1405
1406         ghc->gpa = gpa;
1407         ghc->generation = slots->generation;
1408         ghc->memslot = __gfn_to_memslot(slots, gfn);
1409         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1410         if (!kvm_is_error_hva(ghc->hva))
1411                 ghc->hva += offset;
1412         else
1413                 return -EFAULT;
1414
1415         return 0;
1416 }
1417 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1418
1419 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1420                            void *data, unsigned long len)
1421 {
1422         struct kvm_memslots *slots = kvm_memslots(kvm);
1423         int r;
1424
1425         if (slots->generation != ghc->generation)
1426                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1427
1428         if (kvm_is_error_hva(ghc->hva))
1429                 return -EFAULT;
1430
1431         r = __copy_to_user((void __user *)ghc->hva, data, len);
1432         if (r)
1433                 return -EFAULT;
1434         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1435
1436         return 0;
1437 }
1438 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1439
1440 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1441                            void *data, unsigned long len)
1442 {
1443         struct kvm_memslots *slots = kvm_memslots(kvm);
1444         int r;
1445
1446         if (slots->generation != ghc->generation)
1447                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1448
1449         if (kvm_is_error_hva(ghc->hva))
1450                 return -EFAULT;
1451
1452         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1453         if (r)
1454                 return -EFAULT;
1455
1456         return 0;
1457 }
1458 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1459
1460 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1461 {
1462         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1463                                     offset, len);
1464 }
1465 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1466
1467 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1468 {
1469         gfn_t gfn = gpa >> PAGE_SHIFT;
1470         int seg;
1471         int offset = offset_in_page(gpa);
1472         int ret;
1473
1474         while ((seg = next_segment(len, offset)) != 0) {
1475                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1476                 if (ret < 0)
1477                         return ret;
1478                 offset = 0;
1479                 len -= seg;
1480                 ++gfn;
1481         }
1482         return 0;
1483 }
1484 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1485
1486 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1487                              gfn_t gfn)
1488 {
1489         if (memslot && memslot->dirty_bitmap) {
1490                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1491
1492                 __set_bit_le(rel_gfn, memslot->dirty_bitmap);
1493         }
1494 }
1495
1496 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1497 {
1498         struct kvm_memory_slot *memslot;
1499
1500         memslot = gfn_to_memslot(kvm, gfn);
1501         mark_page_dirty_in_slot(kvm, memslot, gfn);
1502 }
1503
1504 /*
1505  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1506  */
1507 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1508 {
1509         DEFINE_WAIT(wait);
1510
1511         for (;;) {
1512                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1513
1514                 if (kvm_arch_vcpu_runnable(vcpu)) {
1515                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1516                         break;
1517                 }
1518                 if (kvm_cpu_has_pending_timer(vcpu))
1519                         break;
1520                 if (signal_pending(current))
1521                         break;
1522
1523                 schedule();
1524         }
1525
1526         finish_wait(&vcpu->wq, &wait);
1527 }
1528
1529 void kvm_resched(struct kvm_vcpu *vcpu)
1530 {
1531         if (!need_resched())
1532                 return;
1533         cond_resched();
1534 }
1535 EXPORT_SYMBOL_GPL(kvm_resched);
1536
1537 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1538 {
1539         struct kvm *kvm = me->kvm;
1540         struct kvm_vcpu *vcpu;
1541         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1542         int yielded = 0;
1543         int pass;
1544         int i;
1545
1546         /*
1547          * We boost the priority of a VCPU that is runnable but not
1548          * currently running, because it got preempted by something
1549          * else and called schedule in __vcpu_run.  Hopefully that
1550          * VCPU is holding the lock that we need and will release it.
1551          * We approximate round-robin by starting at the last boosted VCPU.
1552          */
1553         for (pass = 0; pass < 2 && !yielded; pass++) {
1554                 kvm_for_each_vcpu(i, vcpu, kvm) {
1555                         struct task_struct *task = NULL;
1556                         struct pid *pid;
1557                         if (!pass && i < last_boosted_vcpu) {
1558                                 i = last_boosted_vcpu;
1559                                 continue;
1560                         } else if (pass && i > last_boosted_vcpu)
1561                                 break;
1562                         if (vcpu == me)
1563                                 continue;
1564                         if (waitqueue_active(&vcpu->wq))
1565                                 continue;
1566                         rcu_read_lock();
1567                         pid = rcu_dereference(vcpu->pid);
1568                         if (pid)
1569                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1570                         rcu_read_unlock();
1571                         if (!task)
1572                                 continue;
1573                         if (task->flags & PF_VCPU) {
1574                                 put_task_struct(task);
1575                                 continue;
1576                         }
1577                         if (yield_to(task, 1)) {
1578                                 put_task_struct(task);
1579                                 kvm->last_boosted_vcpu = i;
1580                                 yielded = 1;
1581                                 break;
1582                         }
1583                         put_task_struct(task);
1584                 }
1585         }
1586 }
1587 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1588
1589 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1590 {
1591         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1592         struct page *page;
1593
1594         if (vmf->pgoff == 0)
1595                 page = virt_to_page(vcpu->run);
1596 #ifdef CONFIG_X86
1597         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1598                 page = virt_to_page(vcpu->arch.pio_data);
1599 #endif
1600 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1601         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1602                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1603 #endif
1604         else
1605                 return VM_FAULT_SIGBUS;
1606         get_page(page);
1607         vmf->page = page;
1608         return 0;
1609 }
1610
1611 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1612         .fault = kvm_vcpu_fault,
1613 };
1614
1615 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1616 {
1617         vma->vm_ops = &kvm_vcpu_vm_ops;
1618         return 0;
1619 }
1620
1621 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1622 {
1623         struct kvm_vcpu *vcpu = filp->private_data;
1624
1625         kvm_put_kvm(vcpu->kvm);
1626         return 0;
1627 }
1628
1629 static struct file_operations kvm_vcpu_fops = {
1630         .release        = kvm_vcpu_release,
1631         .unlocked_ioctl = kvm_vcpu_ioctl,
1632 #ifdef CONFIG_COMPAT
1633         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1634 #endif
1635         .mmap           = kvm_vcpu_mmap,
1636         .llseek         = noop_llseek,
1637 };
1638
1639 /*
1640  * Allocates an inode for the vcpu.
1641  */
1642 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1643 {
1644         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1645 }
1646
1647 /*
1648  * Creates some virtual cpus.  Good luck creating more than one.
1649  */
1650 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1651 {
1652         int r;
1653         struct kvm_vcpu *vcpu, *v;
1654
1655         vcpu = kvm_arch_vcpu_create(kvm, id);
1656         if (IS_ERR(vcpu))
1657                 return PTR_ERR(vcpu);
1658
1659         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1660
1661         r = kvm_arch_vcpu_setup(vcpu);
1662         if (r)
1663                 goto vcpu_destroy;
1664
1665         mutex_lock(&kvm->lock);
1666         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1667                 r = -EINVAL;
1668                 goto unlock_vcpu_destroy;
1669         }
1670
1671         kvm_for_each_vcpu(r, v, kvm)
1672                 if (v->vcpu_id == id) {
1673                         r = -EEXIST;
1674                         goto unlock_vcpu_destroy;
1675                 }
1676
1677         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1678
1679         /* Now it's all set up, let userspace reach it */
1680         kvm_get_kvm(kvm);
1681         r = create_vcpu_fd(vcpu);
1682         if (r < 0) {
1683                 kvm_put_kvm(kvm);
1684                 goto unlock_vcpu_destroy;
1685         }
1686
1687         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1688         smp_wmb();
1689         atomic_inc(&kvm->online_vcpus);
1690
1691 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1692         if (kvm->bsp_vcpu_id == id)
1693                 kvm->bsp_vcpu = vcpu;
1694 #endif
1695         mutex_unlock(&kvm->lock);
1696         return r;
1697
1698 unlock_vcpu_destroy:
1699         mutex_unlock(&kvm->lock);
1700 vcpu_destroy:
1701         kvm_arch_vcpu_destroy(vcpu);
1702         return r;
1703 }
1704
1705 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1706 {
1707         if (sigset) {
1708                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1709                 vcpu->sigset_active = 1;
1710                 vcpu->sigset = *sigset;
1711         } else
1712                 vcpu->sigset_active = 0;
1713         return 0;
1714 }
1715
1716 static long kvm_vcpu_ioctl(struct file *filp,
1717                            unsigned int ioctl, unsigned long arg)
1718 {
1719         struct kvm_vcpu *vcpu = filp->private_data;
1720         void __user *argp = (void __user *)arg;
1721         int r;
1722         struct kvm_fpu *fpu = NULL;
1723         struct kvm_sregs *kvm_sregs = NULL;
1724
1725         if (vcpu->kvm->mm != current->mm)
1726                 return -EIO;
1727
1728 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1729         /*
1730          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1731          * so vcpu_load() would break it.
1732          */
1733         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1734                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1735 #endif
1736
1737
1738         vcpu_load(vcpu);
1739         switch (ioctl) {
1740         case KVM_RUN:
1741                 r = -EINVAL;
1742                 if (arg)
1743                         goto out;
1744                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1745                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1746                 break;
1747         case KVM_GET_REGS: {
1748                 struct kvm_regs *kvm_regs;
1749
1750                 r = -ENOMEM;
1751                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1752                 if (!kvm_regs)
1753                         goto out;
1754                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1755                 if (r)
1756                         goto out_free1;
1757                 r = -EFAULT;
1758                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1759                         goto out_free1;
1760                 r = 0;
1761 out_free1:
1762                 kfree(kvm_regs);
1763                 break;
1764         }
1765         case KVM_SET_REGS: {
1766                 struct kvm_regs *kvm_regs;
1767
1768                 r = -ENOMEM;
1769                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1770                 if (!kvm_regs)
1771                         goto out;
1772                 r = -EFAULT;
1773                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1774                         goto out_free2;
1775                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1776                 if (r)
1777                         goto out_free2;
1778                 r = 0;
1779 out_free2:
1780                 kfree(kvm_regs);
1781                 break;
1782         }
1783         case KVM_GET_SREGS: {
1784                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1785                 r = -ENOMEM;
1786                 if (!kvm_sregs)
1787                         goto out;
1788                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1789                 if (r)
1790                         goto out;
1791                 r = -EFAULT;
1792                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1793                         goto out;
1794                 r = 0;
1795                 break;
1796         }
1797         case KVM_SET_SREGS: {
1798                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1799                 r = -ENOMEM;
1800                 if (!kvm_sregs)
1801                         goto out;
1802                 r = -EFAULT;
1803                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1804                         goto out;
1805                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1806                 if (r)
1807                         goto out;
1808                 r = 0;
1809                 break;
1810         }
1811         case KVM_GET_MP_STATE: {
1812                 struct kvm_mp_state mp_state;
1813
1814                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1815                 if (r)
1816                         goto out;
1817                 r = -EFAULT;
1818                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1819                         goto out;
1820                 r = 0;
1821                 break;
1822         }
1823         case KVM_SET_MP_STATE: {
1824                 struct kvm_mp_state mp_state;
1825
1826                 r = -EFAULT;
1827                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1828                         goto out;
1829                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1830                 if (r)
1831                         goto out;
1832                 r = 0;
1833                 break;
1834         }
1835         case KVM_TRANSLATE: {
1836                 struct kvm_translation tr;
1837
1838                 r = -EFAULT;
1839                 if (copy_from_user(&tr, argp, sizeof tr))
1840                         goto out;
1841                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1842                 if (r)
1843                         goto out;
1844                 r = -EFAULT;
1845                 if (copy_to_user(argp, &tr, sizeof tr))
1846                         goto out;
1847                 r = 0;
1848                 break;
1849         }
1850         case KVM_SET_GUEST_DEBUG: {
1851                 struct kvm_guest_debug dbg;
1852
1853                 r = -EFAULT;
1854                 if (copy_from_user(&dbg, argp, sizeof dbg))
1855                         goto out;
1856                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1857                 if (r)
1858                         goto out;
1859                 r = 0;
1860                 break;
1861         }
1862         case KVM_SET_SIGNAL_MASK: {
1863                 struct kvm_signal_mask __user *sigmask_arg = argp;
1864                 struct kvm_signal_mask kvm_sigmask;
1865                 sigset_t sigset, *p;
1866
1867                 p = NULL;
1868                 if (argp) {
1869                         r = -EFAULT;
1870                         if (copy_from_user(&kvm_sigmask, argp,
1871                                            sizeof kvm_sigmask))
1872                                 goto out;
1873                         r = -EINVAL;
1874                         if (kvm_sigmask.len != sizeof sigset)
1875                                 goto out;
1876                         r = -EFAULT;
1877                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1878                                            sizeof sigset))
1879                                 goto out;
1880                         p = &sigset;
1881                 }
1882                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1883                 break;
1884         }
1885         case KVM_GET_FPU: {
1886                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1887                 r = -ENOMEM;
1888                 if (!fpu)
1889                         goto out;
1890                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1891                 if (r)
1892                         goto out;
1893                 r = -EFAULT;
1894                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1895                         goto out;
1896                 r = 0;
1897                 break;
1898         }
1899         case KVM_SET_FPU: {
1900                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1901                 r = -ENOMEM;
1902                 if (!fpu)
1903                         goto out;
1904                 r = -EFAULT;
1905                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1906                         goto out;
1907                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1908                 if (r)
1909                         goto out;
1910                 r = 0;
1911                 break;
1912         }
1913         default:
1914                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1915         }
1916 out:
1917         vcpu_put(vcpu);
1918         kfree(fpu);
1919         kfree(kvm_sregs);
1920         return r;
1921 }
1922
1923 #ifdef CONFIG_COMPAT
1924 static long kvm_vcpu_compat_ioctl(struct file *filp,
1925                                   unsigned int ioctl, unsigned long arg)
1926 {
1927         struct kvm_vcpu *vcpu = filp->private_data;
1928         void __user *argp = compat_ptr(arg);
1929         int r;
1930
1931         if (vcpu->kvm->mm != current->mm)
1932                 return -EIO;
1933
1934         switch (ioctl) {
1935         case KVM_SET_SIGNAL_MASK: {
1936                 struct kvm_signal_mask __user *sigmask_arg = argp;
1937                 struct kvm_signal_mask kvm_sigmask;
1938                 compat_sigset_t csigset;
1939                 sigset_t sigset;
1940
1941                 if (argp) {
1942                         r = -EFAULT;
1943                         if (copy_from_user(&kvm_sigmask, argp,
1944                                            sizeof kvm_sigmask))
1945                                 goto out;
1946                         r = -EINVAL;
1947                         if (kvm_sigmask.len != sizeof csigset)
1948                                 goto out;
1949                         r = -EFAULT;
1950                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1951                                            sizeof csigset))
1952                                 goto out;
1953                 }
1954                 sigset_from_compat(&sigset, &csigset);
1955                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1956                 break;
1957         }
1958         default:
1959                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1960         }
1961
1962 out:
1963         return r;
1964 }
1965 #endif
1966
1967 static long kvm_vm_ioctl(struct file *filp,
1968                            unsigned int ioctl, unsigned long arg)
1969 {
1970         struct kvm *kvm = filp->private_data;
1971         void __user *argp = (void __user *)arg;
1972         int r;
1973
1974         if (kvm->mm != current->mm)
1975                 return -EIO;
1976         switch (ioctl) {
1977         case KVM_CREATE_VCPU:
1978                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1979                 if (r < 0)
1980                         goto out;
1981                 break;
1982         case KVM_SET_USER_MEMORY_REGION: {
1983                 struct kvm_userspace_memory_region kvm_userspace_mem;
1984
1985                 r = -EFAULT;
1986                 if (copy_from_user(&kvm_userspace_mem, argp,
1987                                                 sizeof kvm_userspace_mem))
1988                         goto out;
1989
1990                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1991                 if (r)
1992                         goto out;
1993                 break;
1994         }
1995         case KVM_GET_DIRTY_LOG: {
1996                 struct kvm_dirty_log log;
1997
1998                 r = -EFAULT;
1999                 if (copy_from_user(&log, argp, sizeof log))
2000                         goto out;
2001                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2002                 if (r)
2003                         goto out;
2004                 break;
2005         }
2006 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2007         case KVM_REGISTER_COALESCED_MMIO: {
2008                 struct kvm_coalesced_mmio_zone zone;
2009                 r = -EFAULT;
2010                 if (copy_from_user(&zone, argp, sizeof zone))
2011                         goto out;
2012                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2013                 if (r)
2014                         goto out;
2015                 r = 0;
2016                 break;
2017         }
2018         case KVM_UNREGISTER_COALESCED_MMIO: {
2019                 struct kvm_coalesced_mmio_zone zone;
2020                 r = -EFAULT;
2021                 if (copy_from_user(&zone, argp, sizeof zone))
2022                         goto out;
2023                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2024                 if (r)
2025                         goto out;
2026                 r = 0;
2027                 break;
2028         }
2029 #endif
2030         case KVM_IRQFD: {
2031                 struct kvm_irqfd data;
2032
2033                 r = -EFAULT;
2034                 if (copy_from_user(&data, argp, sizeof data))
2035                         goto out;
2036                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2037                 break;
2038         }
2039         case KVM_IOEVENTFD: {
2040                 struct kvm_ioeventfd data;
2041
2042                 r = -EFAULT;
2043                 if (copy_from_user(&data, argp, sizeof data))
2044                         goto out;
2045                 r = kvm_ioeventfd(kvm, &data);
2046                 break;
2047         }
2048 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2049         case KVM_SET_BOOT_CPU_ID:
2050                 r = 0;
2051                 mutex_lock(&kvm->lock);
2052                 if (atomic_read(&kvm->online_vcpus) != 0)
2053                         r = -EBUSY;
2054                 else
2055                         kvm->bsp_vcpu_id = arg;
2056                 mutex_unlock(&kvm->lock);
2057                 break;
2058 #endif
2059         default:
2060                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2061                 if (r == -ENOTTY)
2062                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2063         }
2064 out:
2065         return r;
2066 }
2067
2068 #ifdef CONFIG_COMPAT
2069 struct compat_kvm_dirty_log {
2070         __u32 slot;
2071         __u32 padding1;
2072         union {
2073                 compat_uptr_t dirty_bitmap; /* one bit per page */
2074                 __u64 padding2;
2075         };
2076 };
2077
2078 static long kvm_vm_compat_ioctl(struct file *filp,
2079                            unsigned int ioctl, unsigned long arg)
2080 {
2081         struct kvm *kvm = filp->private_data;
2082         int r;
2083
2084         if (kvm->mm != current->mm)
2085                 return -EIO;
2086         switch (ioctl) {
2087         case KVM_GET_DIRTY_LOG: {
2088                 struct compat_kvm_dirty_log compat_log;
2089                 struct kvm_dirty_log log;
2090
2091                 r = -EFAULT;
2092                 if (copy_from_user(&compat_log, (void __user *)arg,
2093                                    sizeof(compat_log)))
2094                         goto out;
2095                 log.slot         = compat_log.slot;
2096                 log.padding1     = compat_log.padding1;
2097                 log.padding2     = compat_log.padding2;
2098                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2099
2100                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2101                 if (r)
2102                         goto out;
2103                 break;
2104         }
2105         default:
2106                 r = kvm_vm_ioctl(filp, ioctl, arg);
2107         }
2108
2109 out:
2110         return r;
2111 }
2112 #endif
2113
2114 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2115 {
2116         struct page *page[1];
2117         unsigned long addr;
2118         int npages;
2119         gfn_t gfn = vmf->pgoff;
2120         struct kvm *kvm = vma->vm_file->private_data;
2121
2122         addr = gfn_to_hva(kvm, gfn);
2123         if (kvm_is_error_hva(addr))
2124                 return VM_FAULT_SIGBUS;
2125
2126         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2127                                 NULL);
2128         if (unlikely(npages != 1))
2129                 return VM_FAULT_SIGBUS;
2130
2131         vmf->page = page[0];
2132         return 0;
2133 }
2134
2135 static const struct vm_operations_struct kvm_vm_vm_ops = {
2136         .fault = kvm_vm_fault,
2137 };
2138
2139 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2140 {
2141         vma->vm_ops = &kvm_vm_vm_ops;
2142         return 0;
2143 }
2144
2145 static struct file_operations kvm_vm_fops = {
2146         .release        = kvm_vm_release,
2147         .unlocked_ioctl = kvm_vm_ioctl,
2148 #ifdef CONFIG_COMPAT
2149         .compat_ioctl   = kvm_vm_compat_ioctl,
2150 #endif
2151         .mmap           = kvm_vm_mmap,
2152         .llseek         = noop_llseek,
2153 };
2154
2155 static int kvm_dev_ioctl_create_vm(void)
2156 {
2157         int r;
2158         struct kvm *kvm;
2159
2160         kvm = kvm_create_vm();
2161         if (IS_ERR(kvm))
2162                 return PTR_ERR(kvm);
2163 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2164         r = kvm_coalesced_mmio_init(kvm);
2165         if (r < 0) {
2166                 kvm_put_kvm(kvm);
2167                 return r;
2168         }
2169 #endif
2170         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2171         if (r < 0)
2172                 kvm_put_kvm(kvm);
2173
2174         return r;
2175 }
2176
2177 static long kvm_dev_ioctl_check_extension_generic(long arg)
2178 {
2179         switch (arg) {
2180         case KVM_CAP_USER_MEMORY:
2181         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2182         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2183 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2184         case KVM_CAP_SET_BOOT_CPU_ID:
2185 #endif
2186         case KVM_CAP_INTERNAL_ERROR_DATA:
2187                 return 1;
2188 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2189         case KVM_CAP_IRQ_ROUTING:
2190                 return KVM_MAX_IRQ_ROUTES;
2191 #endif
2192         default:
2193                 break;
2194         }
2195         return kvm_dev_ioctl_check_extension(arg);
2196 }
2197
2198 static long kvm_dev_ioctl(struct file *filp,
2199                           unsigned int ioctl, unsigned long arg)
2200 {
2201         long r = -EINVAL;
2202
2203         switch (ioctl) {
2204         case KVM_GET_API_VERSION:
2205                 r = -EINVAL;
2206                 if (arg)
2207                         goto out;
2208                 r = KVM_API_VERSION;
2209                 break;
2210         case KVM_CREATE_VM:
2211                 r = -EINVAL;
2212                 if (arg)
2213                         goto out;
2214                 r = kvm_dev_ioctl_create_vm();
2215                 break;
2216         case KVM_CHECK_EXTENSION:
2217                 r = kvm_dev_ioctl_check_extension_generic(arg);
2218                 break;
2219         case KVM_GET_VCPU_MMAP_SIZE:
2220                 r = -EINVAL;
2221                 if (arg)
2222                         goto out;
2223                 r = PAGE_SIZE;     /* struct kvm_run */
2224 #ifdef CONFIG_X86
2225                 r += PAGE_SIZE;    /* pio data page */
2226 #endif
2227 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2228                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2229 #endif
2230                 break;
2231         case KVM_TRACE_ENABLE:
2232         case KVM_TRACE_PAUSE:
2233         case KVM_TRACE_DISABLE:
2234                 r = -EOPNOTSUPP;
2235                 break;
2236         default:
2237                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2238         }
2239 out:
2240         return r;
2241 }
2242
2243 static struct file_operations kvm_chardev_ops = {
2244         .unlocked_ioctl = kvm_dev_ioctl,
2245         .compat_ioctl   = kvm_dev_ioctl,
2246         .llseek         = noop_llseek,
2247 };
2248
2249 static struct miscdevice kvm_dev = {
2250         KVM_MINOR,
2251         "kvm",
2252         &kvm_chardev_ops,
2253 };
2254
2255 static void hardware_enable_nolock(void *junk)
2256 {
2257         int cpu = raw_smp_processor_id();
2258         int r;
2259
2260         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2261                 return;
2262
2263         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2264
2265         r = kvm_arch_hardware_enable(NULL);
2266
2267         if (r) {
2268                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2269                 atomic_inc(&hardware_enable_failed);
2270                 printk(KERN_INFO "kvm: enabling virtualization on "
2271                                  "CPU%d failed\n", cpu);
2272         }
2273 }
2274
2275 static void hardware_enable(void *junk)
2276 {
2277         raw_spin_lock(&kvm_lock);
2278         hardware_enable_nolock(junk);
2279         raw_spin_unlock(&kvm_lock);
2280 }
2281
2282 static void hardware_disable_nolock(void *junk)
2283 {
2284         int cpu = raw_smp_processor_id();
2285
2286         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2287                 return;
2288         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2289         kvm_arch_hardware_disable(NULL);
2290 }
2291
2292 static void hardware_disable(void *junk)
2293 {
2294         raw_spin_lock(&kvm_lock);
2295         hardware_disable_nolock(junk);
2296         raw_spin_unlock(&kvm_lock);
2297 }
2298
2299 static void hardware_disable_all_nolock(void)
2300 {
2301         BUG_ON(!kvm_usage_count);
2302
2303         kvm_usage_count--;
2304         if (!kvm_usage_count)
2305                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2306 }
2307
2308 static void hardware_disable_all(void)
2309 {
2310         raw_spin_lock(&kvm_lock);
2311         hardware_disable_all_nolock();
2312         raw_spin_unlock(&kvm_lock);
2313 }
2314
2315 static int hardware_enable_all(void)
2316 {
2317         int r = 0;
2318
2319         raw_spin_lock(&kvm_lock);
2320
2321         kvm_usage_count++;
2322         if (kvm_usage_count == 1) {
2323                 atomic_set(&hardware_enable_failed, 0);
2324                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2325
2326                 if (atomic_read(&hardware_enable_failed)) {
2327                         hardware_disable_all_nolock();
2328                         r = -EBUSY;
2329                 }
2330         }
2331
2332         raw_spin_unlock(&kvm_lock);
2333
2334         return r;
2335 }
2336
2337 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2338                            void *v)
2339 {
2340         int cpu = (long)v;
2341
2342         if (!kvm_usage_count)
2343                 return NOTIFY_OK;
2344
2345         val &= ~CPU_TASKS_FROZEN;
2346         switch (val) {
2347         case CPU_DYING:
2348                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2349                        cpu);
2350                 hardware_disable(NULL);
2351                 break;
2352         case CPU_STARTING:
2353                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2354                        cpu);
2355                 hardware_enable(NULL);
2356                 break;
2357         }
2358         return NOTIFY_OK;
2359 }
2360
2361
2362 asmlinkage void kvm_spurious_fault(void)
2363 {
2364         /* Fault while not rebooting.  We want the trace. */
2365         BUG();
2366 }
2367 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2368
2369 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2370                       void *v)
2371 {
2372         /*
2373          * Some (well, at least mine) BIOSes hang on reboot if
2374          * in vmx root mode.
2375          *
2376          * And Intel TXT required VMX off for all cpu when system shutdown.
2377          */
2378         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2379         kvm_rebooting = true;
2380         on_each_cpu(hardware_disable_nolock, NULL, 1);
2381         return NOTIFY_OK;
2382 }
2383
2384 static struct notifier_block kvm_reboot_notifier = {
2385         .notifier_call = kvm_reboot,
2386         .priority = 0,
2387 };
2388
2389 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2390 {
2391         int i;
2392
2393         for (i = 0; i < bus->dev_count; i++) {
2394                 struct kvm_io_device *pos = bus->devs[i];
2395
2396                 kvm_iodevice_destructor(pos);
2397         }
2398         kfree(bus);
2399 }
2400
2401 /* kvm_io_bus_write - called under kvm->slots_lock */
2402 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2403                      int len, const void *val)
2404 {
2405         int i;
2406         struct kvm_io_bus *bus;
2407
2408         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2409         for (i = 0; i < bus->dev_count; i++)
2410                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2411                         return 0;
2412         return -EOPNOTSUPP;
2413 }
2414
2415 /* kvm_io_bus_read - called under kvm->slots_lock */
2416 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2417                     int len, void *val)
2418 {
2419         int i;
2420         struct kvm_io_bus *bus;
2421
2422         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2423         for (i = 0; i < bus->dev_count; i++)
2424                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2425                         return 0;
2426         return -EOPNOTSUPP;
2427 }
2428
2429 /* Caller must hold slots_lock. */
2430 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2431                             struct kvm_io_device *dev)
2432 {
2433         struct kvm_io_bus *new_bus, *bus;
2434
2435         bus = kvm->buses[bus_idx];
2436         if (bus->dev_count > NR_IOBUS_DEVS-1)
2437                 return -ENOSPC;
2438
2439         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2440         if (!new_bus)
2441                 return -ENOMEM;
2442         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2443         new_bus->devs[new_bus->dev_count++] = dev;
2444         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2445         synchronize_srcu_expedited(&kvm->srcu);
2446         kfree(bus);
2447
2448         return 0;
2449 }
2450
2451 /* Caller must hold slots_lock. */
2452 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2453                               struct kvm_io_device *dev)
2454 {
2455         int i, r;
2456         struct kvm_io_bus *new_bus, *bus;
2457
2458         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2459         if (!new_bus)
2460                 return -ENOMEM;
2461
2462         bus = kvm->buses[bus_idx];
2463         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2464
2465         r = -ENOENT;
2466         for (i = 0; i < new_bus->dev_count; i++)
2467                 if (new_bus->devs[i] == dev) {
2468                         r = 0;
2469                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2470                         break;
2471                 }
2472
2473         if (r) {
2474                 kfree(new_bus);
2475                 return r;
2476         }
2477
2478         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2479         synchronize_srcu_expedited(&kvm->srcu);
2480         kfree(bus);
2481         return r;
2482 }
2483
2484 static struct notifier_block kvm_cpu_notifier = {
2485         .notifier_call = kvm_cpu_hotplug,
2486 };
2487
2488 static int vm_stat_get(void *_offset, u64 *val)
2489 {
2490         unsigned offset = (long)_offset;
2491         struct kvm *kvm;
2492
2493         *val = 0;
2494         raw_spin_lock(&kvm_lock);
2495         list_for_each_entry(kvm, &vm_list, vm_list)
2496                 *val += *(u32 *)((void *)kvm + offset);
2497         raw_spin_unlock(&kvm_lock);
2498         return 0;
2499 }
2500
2501 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2502
2503 static int vcpu_stat_get(void *_offset, u64 *val)
2504 {
2505         unsigned offset = (long)_offset;
2506         struct kvm *kvm;
2507         struct kvm_vcpu *vcpu;
2508         int i;
2509
2510         *val = 0;
2511         raw_spin_lock(&kvm_lock);
2512         list_for_each_entry(kvm, &vm_list, vm_list)
2513                 kvm_for_each_vcpu(i, vcpu, kvm)
2514                         *val += *(u32 *)((void *)vcpu + offset);
2515
2516         raw_spin_unlock(&kvm_lock);
2517         return 0;
2518 }
2519
2520 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2521
2522 static const struct file_operations *stat_fops[] = {
2523         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2524         [KVM_STAT_VM]   = &vm_stat_fops,
2525 };
2526
2527 static void kvm_init_debug(void)
2528 {
2529         struct kvm_stats_debugfs_item *p;
2530
2531         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2532         for (p = debugfs_entries; p->name; ++p)
2533                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2534                                                 (void *)(long)p->offset,
2535                                                 stat_fops[p->kind]);
2536 }
2537
2538 static void kvm_exit_debug(void)
2539 {
2540         struct kvm_stats_debugfs_item *p;
2541
2542         for (p = debugfs_entries; p->name; ++p)
2543                 debugfs_remove(p->dentry);
2544         debugfs_remove(kvm_debugfs_dir);
2545 }
2546
2547 static int kvm_suspend(void)
2548 {
2549         if (kvm_usage_count)
2550                 hardware_disable_nolock(NULL);
2551         return 0;
2552 }
2553
2554 static void kvm_resume(void)
2555 {
2556         if (kvm_usage_count) {
2557                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2558                 hardware_enable_nolock(NULL);
2559         }
2560 }
2561
2562 static struct syscore_ops kvm_syscore_ops = {
2563         .suspend = kvm_suspend,
2564         .resume = kvm_resume,
2565 };
2566
2567 struct page *bad_page;
2568 pfn_t bad_pfn;
2569
2570 static inline
2571 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2572 {
2573         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2574 }
2575
2576 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2577 {
2578         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2579
2580         kvm_arch_vcpu_load(vcpu, cpu);
2581 }
2582
2583 static void kvm_sched_out(struct preempt_notifier *pn,
2584                           struct task_struct *next)
2585 {
2586         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2587
2588         kvm_arch_vcpu_put(vcpu);
2589 }
2590
2591 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2592                   struct module *module)
2593 {
2594         int r;
2595         int cpu;
2596
2597         r = kvm_arch_init(opaque);
2598         if (r)
2599                 goto out_fail;
2600
2601         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2602
2603         if (bad_page == NULL) {
2604                 r = -ENOMEM;
2605                 goto out;
2606         }
2607
2608         bad_pfn = page_to_pfn(bad_page);
2609
2610         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2611
2612         if (hwpoison_page == NULL) {
2613                 r = -ENOMEM;
2614                 goto out_free_0;
2615         }
2616
2617         hwpoison_pfn = page_to_pfn(hwpoison_page);
2618
2619         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2620
2621         if (fault_page == NULL) {
2622                 r = -ENOMEM;
2623                 goto out_free_0;
2624         }
2625
2626         fault_pfn = page_to_pfn(fault_page);
2627
2628         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2629                 r = -ENOMEM;
2630                 goto out_free_0;
2631         }
2632
2633         r = kvm_arch_hardware_setup();
2634         if (r < 0)
2635                 goto out_free_0a;
2636
2637         for_each_online_cpu(cpu) {
2638                 smp_call_function_single(cpu,
2639                                 kvm_arch_check_processor_compat,
2640                                 &r, 1);
2641                 if (r < 0)
2642                         goto out_free_1;
2643         }
2644
2645         r = register_cpu_notifier(&kvm_cpu_notifier);
2646         if (r)
2647                 goto out_free_2;
2648         register_reboot_notifier(&kvm_reboot_notifier);
2649
2650         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2651         if (!vcpu_align)
2652                 vcpu_align = __alignof__(struct kvm_vcpu);
2653         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2654                                            0, NULL);
2655         if (!kvm_vcpu_cache) {
2656                 r = -ENOMEM;
2657                 goto out_free_3;
2658         }
2659
2660         r = kvm_async_pf_init();
2661         if (r)
2662                 goto out_free;
2663
2664         kvm_chardev_ops.owner = module;
2665         kvm_vm_fops.owner = module;
2666         kvm_vcpu_fops.owner = module;
2667
2668         r = misc_register(&kvm_dev);
2669         if (r) {
2670                 printk(KERN_ERR "kvm: misc device register failed\n");
2671                 goto out_unreg;
2672         }
2673
2674         register_syscore_ops(&kvm_syscore_ops);
2675
2676         kvm_preempt_ops.sched_in = kvm_sched_in;
2677         kvm_preempt_ops.sched_out = kvm_sched_out;
2678
2679         kvm_init_debug();
2680
2681         return 0;
2682
2683 out_unreg:
2684         kvm_async_pf_deinit();
2685 out_free:
2686         kmem_cache_destroy(kvm_vcpu_cache);
2687 out_free_3:
2688         unregister_reboot_notifier(&kvm_reboot_notifier);
2689         unregister_cpu_notifier(&kvm_cpu_notifier);
2690 out_free_2:
2691 out_free_1:
2692         kvm_arch_hardware_unsetup();
2693 out_free_0a:
2694         free_cpumask_var(cpus_hardware_enabled);
2695 out_free_0:
2696         if (fault_page)
2697                 __free_page(fault_page);
2698         if (hwpoison_page)
2699                 __free_page(hwpoison_page);
2700         __free_page(bad_page);
2701 out:
2702         kvm_arch_exit();
2703 out_fail:
2704         return r;
2705 }
2706 EXPORT_SYMBOL_GPL(kvm_init);
2707
2708 void kvm_exit(void)
2709 {
2710         kvm_exit_debug();
2711         misc_deregister(&kvm_dev);
2712         kmem_cache_destroy(kvm_vcpu_cache);
2713         kvm_async_pf_deinit();
2714         unregister_syscore_ops(&kvm_syscore_ops);
2715         unregister_reboot_notifier(&kvm_reboot_notifier);
2716         unregister_cpu_notifier(&kvm_cpu_notifier);
2717         on_each_cpu(hardware_disable_nolock, NULL, 1);
2718         kvm_arch_hardware_unsetup();
2719         kvm_arch_exit();
2720         free_cpumask_var(cpus_hardware_enabled);
2721         __free_page(hwpoison_page);
2722         __free_page(bad_page);
2723 }
2724 EXPORT_SYMBOL_GPL(kvm_exit);