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 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/pgtable.h>
  57
  58 #include "coalesced_mmio.h"
  59 #include "async_pf.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/kvm.h>
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 /*
  68  * Ordering of locks:
  69  *
  70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  71  */
  72
  73 DEFINE_SPINLOCK(kvm_lock);
  74 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
  75 LIST_HEAD(vm_list);
  76
  77 static cpumask_var_t cpus_hardware_enabled;
  78 static int kvm_usage_count = 0;
  79 static atomic_t hardware_enable_failed;
  80
  81 struct kmem_cache *kvm_vcpu_cache;
  82 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  83
  84 static __read_mostly struct preempt_ops kvm_preempt_ops;
  85
  86 struct dentry *kvm_debugfs_dir;
  87
  88 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  89                            unsigned long arg);
  90 #ifdef CONFIG_COMPAT
  91 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  92                                   unsigned long arg);
  93 #endif
  94 static int hardware_enable_all(void);
  95 static void hardware_disable_all(void);
  96
  97 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  98 static void update_memslots(struct kvm_memslots *slots,
  99                             struct kvm_memory_slot *new, u64 last_generation);
 100
 101 static void kvm_release_pfn_dirty(pfn_t pfn);
 102 static void mark_page_dirty_in_slot(struct kvm *kvm,
 103                                     struct kvm_memory_slot *memslot, gfn_t gfn);
 104
 105 __visible bool kvm_rebooting;
 106 EXPORT_SYMBOL_GPL(kvm_rebooting);
 107
 108 static bool largepages_enabled = true;
 109
 110 bool kvm_is_mmio_pfn(pfn_t pfn)
 111 {
 112         if (pfn_valid(pfn))
 113                 return PageReserved(pfn_to_page(pfn));
 114
 115         return true;
 116 }
 117
 118 /*
 119  * Switches to specified vcpu, until a matching vcpu_put()
 120  */
 121 int vcpu_load(struct kvm_vcpu *vcpu)
 122 {
 123         int cpu;
 124
 125         if (mutex_lock_killable(&vcpu->mutex))
 126                 return -EINTR;
 127         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 128                 /* The thread running this VCPU changed. */
 129                 struct pid *oldpid = vcpu->pid;
 130                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 131                 rcu_assign_pointer(vcpu->pid, newpid);
 132                 if (oldpid)
 133                         synchronize_rcu();
 134                 put_pid(oldpid);
 135         }
 136         cpu = get_cpu();
 137         preempt_notifier_register(&vcpu->preempt_notifier);
 138         kvm_arch_vcpu_load(vcpu, cpu);
 139         put_cpu();
 140         return 0;
 141 }
 142
 143 void vcpu_put(struct kvm_vcpu *vcpu)
 144 {
 145         preempt_disable();
 146         kvm_arch_vcpu_put(vcpu);
 147         preempt_notifier_unregister(&vcpu->preempt_notifier);
 148         preempt_enable();
 149         mutex_unlock(&vcpu->mutex);
 150 }
 151
 152 static void ack_flush(void *_completed)
 153 {
 154 }
 155
 156 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 157 {
 158         int i, cpu, me;
 159         cpumask_var_t cpus;
 160         bool called = true;
 161         struct kvm_vcpu *vcpu;
 162
 163         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 164
 165         me = get_cpu();
 166         kvm_for_each_vcpu(i, vcpu, kvm) {
 167                 kvm_make_request(req, vcpu);
 168                 cpu = vcpu->cpu;
 169
 170                 /* Set ->requests bit before we read ->mode */
 171                 smp_mb();
 172
 173                 if (cpus != NULL && cpu != -1 && cpu != me &&
 174                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 175                         cpumask_set_cpu(cpu, cpus);
 176         }
 177         if (unlikely(cpus == NULL))
 178                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 179         else if (!cpumask_empty(cpus))
 180                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 181         else
 182                 called = false;
 183         put_cpu();
 184         free_cpumask_var(cpus);
 185         return called;
 186 }
 187
 188 void kvm_flush_remote_tlbs(struct kvm *kvm)
 189 {
 190         long dirty_count = kvm->tlbs_dirty;
 191
 192         smp_mb();
 193         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 194                 ++kvm->stat.remote_tlb_flush;
 195         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 196 }
 197 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 198
 199 void kvm_reload_remote_mmus(struct kvm *kvm)
 200 {
 201         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 202 }
 203
 204 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 205 {
 206         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 207 }
 208
 209 void kvm_make_scan_ioapic_request(struct kvm *kvm)
 210 {
 211         make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
 212 }
 213
 214 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 215 {
 216         struct page *page;
 217         int r;
 218
 219         mutex_init(&vcpu->mutex);
 220         vcpu->cpu = -1;
 221         vcpu->kvm = kvm;
 222         vcpu->vcpu_id = id;
 223         vcpu->pid = NULL;
 224         init_waitqueue_head(&vcpu->wq);
 225         kvm_async_pf_vcpu_init(vcpu);
 226
 227         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 228         if (!page) {
 229                 r = -ENOMEM;
 230                 goto fail;
 231         }
 232         vcpu->run = page_address(page);
 233
 234         kvm_vcpu_set_in_spin_loop(vcpu, false);
 235         kvm_vcpu_set_dy_eligible(vcpu, false);
 236         vcpu->preempted = false;
 237
 238         r = kvm_arch_vcpu_init(vcpu);
 239         if (r < 0)
 240                 goto fail_free_run;
 241         return 0;
 242
 243 fail_free_run:
 244         free_page((unsigned long)vcpu->run);
 245 fail:
 246         return r;
 247 }
 248 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 249
 250 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 251 {
 252         put_pid(vcpu->pid);
 253         kvm_arch_vcpu_uninit(vcpu);
 254         free_page((unsigned long)vcpu->run);
 255 }
 256 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 257
 258 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 259 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 260 {
 261         return container_of(mn, struct kvm, mmu_notifier);
 262 }
 263
 264 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 265                                              struct mm_struct *mm,
 266                                              unsigned long address)
 267 {
 268         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 269         int need_tlb_flush, idx;
 270
 271         /*
 272          * When ->invalidate_page runs, the linux pte has been zapped
 273          * already but the page is still allocated until
 274          * ->invalidate_page returns. So if we increase the sequence
 275          * here the kvm page fault will notice if the spte can't be
 276          * established because the page is going to be freed. If
 277          * instead the kvm page fault establishes the spte before
 278          * ->invalidate_page runs, kvm_unmap_hva will release it
 279          * before returning.
 280          *
 281          * The sequence increase only need to be seen at spin_unlock
 282          * time, and not at spin_lock time.
 283          *
 284          * Increasing the sequence after the spin_unlock would be
 285          * unsafe because the kvm page fault could then establish the
 286          * pte after kvm_unmap_hva returned, without noticing the page
 287          * is going to be freed.
 288          */
 289         idx = srcu_read_lock(&kvm->srcu);
 290         spin_lock(&kvm->mmu_lock);
 291
 292         kvm->mmu_notifier_seq++;
 293         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 294         /* we've to flush the tlb before the pages can be freed */
 295         if (need_tlb_flush)
 296                 kvm_flush_remote_tlbs(kvm);
 297
 298         spin_unlock(&kvm->mmu_lock);
 299         srcu_read_unlock(&kvm->srcu, idx);
 300 }
 301
 302 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 303                                         struct mm_struct *mm,
 304                                         unsigned long address,
 305                                         pte_t pte)
 306 {
 307         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 308         int idx;
 309
 310         idx = srcu_read_lock(&kvm->srcu);
 311         spin_lock(&kvm->mmu_lock);
 312         kvm->mmu_notifier_seq++;
 313         kvm_set_spte_hva(kvm, address, pte);
 314         spin_unlock(&kvm->mmu_lock);
 315         srcu_read_unlock(&kvm->srcu, idx);
 316 }
 317
 318 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 319                                                     struct mm_struct *mm,
 320                                                     unsigned long start,
 321                                                     unsigned long end)
 322 {
 323         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 324         int need_tlb_flush = 0, idx;
 325
 326         idx = srcu_read_lock(&kvm->srcu);
 327         spin_lock(&kvm->mmu_lock);
 328         /*
 329          * The count increase must become visible at unlock time as no
 330          * spte can be established without taking the mmu_lock and
 331          * count is also read inside the mmu_lock critical section.
 332          */
 333         kvm->mmu_notifier_count++;
 334         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 335         need_tlb_flush |= kvm->tlbs_dirty;
 336         /* we've to flush the tlb before the pages can be freed */
 337         if (need_tlb_flush)
 338                 kvm_flush_remote_tlbs(kvm);
 339
 340         spin_unlock(&kvm->mmu_lock);
 341         srcu_read_unlock(&kvm->srcu, idx);
 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         smp_wmb();
 359         /*
 360          * The above sequence increase must be visible before the
 361          * below count decrease, which is ensured by the smp_wmb above
 362          * in conjunction with the smp_rmb in mmu_notifier_retry().
 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
 380         young = kvm_age_hva(kvm, address);
 381         if (young)
 382                 kvm_flush_remote_tlbs(kvm);
 383
 384         spin_unlock(&kvm->mmu_lock);
 385         srcu_read_unlock(&kvm->srcu, idx);
 386
 387         return young;
 388 }
 389
 390 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 391                                        struct mm_struct *mm,
 392                                        unsigned long address)
 393 {
 394         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 395         int young, idx;
 396
 397         idx = srcu_read_lock(&kvm->srcu);
 398         spin_lock(&kvm->mmu_lock);
 399         young = kvm_test_age_hva(kvm, address);
 400         spin_unlock(&kvm->mmu_lock);
 401         srcu_read_unlock(&kvm->srcu, idx);
 402
 403         return young;
 404 }
 405
 406 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 407                                      struct mm_struct *mm)
 408 {
 409         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 410         int idx;
 411
 412         idx = srcu_read_lock(&kvm->srcu);
 413         kvm_arch_flush_shadow_all(kvm);
 414         srcu_read_unlock(&kvm->srcu, idx);
 415 }
 416
 417 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 418         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 419         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 420         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 421         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 422         .test_young             = kvm_mmu_notifier_test_young,
 423         .change_pte             = kvm_mmu_notifier_change_pte,
 424         .release                = kvm_mmu_notifier_release,
 425 };
 426
 427 static int kvm_init_mmu_notifier(struct kvm *kvm)
 428 {
 429         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 430         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 431 }
 432
 433 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 434
 435 static int kvm_init_mmu_notifier(struct kvm *kvm)
 436 {
 437         return 0;
 438 }
 439
 440 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 441
 442 static void kvm_init_memslots_id(struct kvm *kvm)
 443 {
 444         int i;
 445         struct kvm_memslots *slots = kvm->memslots;
 446
 447         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 448                 slots->id_to_index[i] = slots->memslots[i].id = i;
 449 }
 450
 451 static struct kvm *kvm_create_vm(unsigned long type)
 452 {
 453         int r, i;
 454         struct kvm *kvm = kvm_arch_alloc_vm();
 455
 456         if (!kvm)
 457                 return ERR_PTR(-ENOMEM);
 458
 459         r = kvm_arch_init_vm(kvm, type);
 460         if (r)
 461                 goto out_err_no_disable;
 462
 463         r = hardware_enable_all();
 464         if (r)
 465                 goto out_err_no_disable;
 466
 467 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 468         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 469 #endif
 470 #ifdef CONFIG_HAVE_KVM_IRQFD
 471         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 472 #endif
 473
 474         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 475
 476         r = -ENOMEM;
 477         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 478         if (!kvm->memslots)
 479                 goto out_err_no_srcu;
 480         kvm_init_memslots_id(kvm);
 481         if (init_srcu_struct(&kvm->srcu))
 482                 goto out_err_no_srcu;
 483         if (init_srcu_struct(&kvm->irq_srcu))
 484                 goto out_err_no_irq_srcu;
 485         for (i = 0; i < KVM_NR_BUSES; i++) {
 486                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 487                                         GFP_KERNEL);
 488                 if (!kvm->buses[i])
 489                         goto out_err;
 490         }
 491
 492         spin_lock_init(&kvm->mmu_lock);
 493         kvm->mm = current->mm;
 494         atomic_inc(&kvm->mm->mm_count);
 495         kvm_eventfd_init(kvm);
 496         mutex_init(&kvm->lock);
 497         mutex_init(&kvm->irq_lock);
 498         mutex_init(&kvm->slots_lock);
 499         atomic_set(&kvm->users_count, 1);
 500         INIT_LIST_HEAD(&kvm->devices);
 501
 502         r = kvm_init_mmu_notifier(kvm);
 503         if (r)
 504                 goto out_err;
 505
 506         spin_lock(&kvm_lock);
 507         list_add(&kvm->vm_list, &vm_list);
 508         spin_unlock(&kvm_lock);
 509
 510         return kvm;
 511
 512 out_err:
 513         cleanup_srcu_struct(&kvm->irq_srcu);
 514 out_err_no_irq_srcu:
 515         cleanup_srcu_struct(&kvm->srcu);
 516 out_err_no_srcu:
 517         hardware_disable_all();
 518 out_err_no_disable:
 519         for (i = 0; i < KVM_NR_BUSES; i++)
 520                 kfree(kvm->buses[i]);
 521         kfree(kvm->memslots);
 522         kvm_arch_free_vm(kvm);
 523         return ERR_PTR(r);
 524 }
 525
 526 /*
 527  * Avoid using vmalloc for a small buffer.
 528  * Should not be used when the size is statically known.
 529  */
 530 void *kvm_kvzalloc(unsigned long size)
 531 {
 532         if (size > PAGE_SIZE)
 533                 return vzalloc(size);
 534         else
 535                 return kzalloc(size, GFP_KERNEL);
 536 }
 537
 538 void kvm_kvfree(const void *addr)
 539 {
 540         if (is_vmalloc_addr(addr))
 541                 vfree(addr);
 542         else
 543                 kfree(addr);
 544 }
 545
 546 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 547 {
 548         if (!memslot->dirty_bitmap)
 549                 return;
 550
 551         kvm_kvfree(memslot->dirty_bitmap);
 552         memslot->dirty_bitmap = NULL;
 553 }
 554
 555 /*
 556  * Free any memory in @free but not in @dont.
 557  */
 558 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
 559                                   struct kvm_memory_slot *dont)
 560 {
 561         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 562                 kvm_destroy_dirty_bitmap(free);
 563
 564         kvm_arch_free_memslot(kvm, free, dont);
 565
 566         free->npages = 0;
 567 }
 568
 569 static void kvm_free_physmem(struct kvm *kvm)
 570 {
 571         struct kvm_memslots *slots = kvm->memslots;
 572         struct kvm_memory_slot *memslot;
 573
 574         kvm_for_each_memslot(memslot, slots)
 575                 kvm_free_physmem_slot(kvm, memslot, NULL);
 576
 577         kfree(kvm->memslots);
 578 }
 579
 580 static void kvm_destroy_devices(struct kvm *kvm)
 581 {
 582         struct list_head *node, *tmp;
 583
 584         list_for_each_safe(node, tmp, &kvm->devices) {
 585                 struct kvm_device *dev =
 586                         list_entry(node, struct kvm_device, vm_node);
 587
 588                 list_del(node);
 589                 dev->ops->destroy(dev);
 590         }
 591 }
 592
 593 static void kvm_destroy_vm(struct kvm *kvm)
 594 {
 595         int i;
 596         struct mm_struct *mm = kvm->mm;
 597
 598         kvm_arch_sync_events(kvm);
 599         spin_lock(&kvm_lock);
 600         list_del(&kvm->vm_list);
 601         spin_unlock(&kvm_lock);
 602         kvm_free_irq_routing(kvm);
 603         for (i = 0; i < KVM_NR_BUSES; i++)
 604                 kvm_io_bus_destroy(kvm->buses[i]);
 605         kvm_coalesced_mmio_free(kvm);
 606 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 607         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 608 #else
 609         kvm_arch_flush_shadow_all(kvm);
 610 #endif
 611         kvm_arch_destroy_vm(kvm);
 612         kvm_destroy_devices(kvm);
 613         kvm_free_physmem(kvm);
 614         cleanup_srcu_struct(&kvm->irq_srcu);
 615         cleanup_srcu_struct(&kvm->srcu);
 616         kvm_arch_free_vm(kvm);
 617         hardware_disable_all();
 618         mmdrop(mm);
 619 }
 620
 621 void kvm_get_kvm(struct kvm *kvm)
 622 {
 623         atomic_inc(&kvm->users_count);
 624 }
 625 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 626
 627 void kvm_put_kvm(struct kvm *kvm)
 628 {
 629         if (atomic_dec_and_test(&kvm->users_count))
 630                 kvm_destroy_vm(kvm);
 631 }
 632 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 633
 634
 635 static int kvm_vm_release(struct inode *inode, struct file *filp)
 636 {
 637         struct kvm *kvm = filp->private_data;
 638
 639         kvm_irqfd_release(kvm);
 640
 641         kvm_put_kvm(kvm);
 642         return 0;
 643 }
 644
 645 /*
 646  * Allocation size is twice as large as the actual dirty bitmap size.
 647  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 648  */
 649 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 650 {
 651         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 652
 653         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 654         if (!memslot->dirty_bitmap)
 655                 return -ENOMEM;
 656
 657         return 0;
 658 }
 659
 660 static int cmp_memslot(const void *slot1, const void *slot2)
 661 {
 662         struct kvm_memory_slot *s1, *s2;
 663
 664         s1 = (struct kvm_memory_slot *)slot1;
 665         s2 = (struct kvm_memory_slot *)slot2;
 666
 667         if (s1->npages < s2->npages)
 668                 return 1;
 669         if (s1->npages > s2->npages)
 670                 return -1;
 671
 672         return 0;
 673 }
 674
 675 /*
 676  * Sort the memslots base on its size, so the larger slots
 677  * will get better fit.
 678  */
 679 static void sort_memslots(struct kvm_memslots *slots)
 680 {
 681         int i;
 682
 683         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 684               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 685
 686         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 687                 slots->id_to_index[slots->memslots[i].id] = i;
 688 }
 689
 690 static void update_memslots(struct kvm_memslots *slots,
 691                             struct kvm_memory_slot *new,
 692                             u64 last_generation)
 693 {
 694         if (new) {
 695                 int id = new->id;
 696                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 697                 unsigned long npages = old->npages;
 698
 699                 *old = *new;
 700                 if (new->npages != npages)
 701                         sort_memslots(slots);
 702         }
 703
 704         slots->generation = last_generation + 1;
 705 }
 706
 707 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 708 {
 709         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 710
 711 #ifdef KVM_CAP_READONLY_MEM
 712         valid_flags |= KVM_MEM_READONLY;
 713 #endif
 714
 715         if (mem->flags & ~valid_flags)
 716                 return -EINVAL;
 717
 718         return 0;
 719 }
 720
 721 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 722                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 723 {
 724         struct kvm_memslots *old_memslots = kvm->memslots;
 725
 726         update_memslots(slots, new, kvm->memslots->generation);
 727         rcu_assign_pointer(kvm->memslots, slots);
 728         synchronize_srcu_expedited(&kvm->srcu);
 729
 730         kvm_arch_memslots_updated(kvm);
 731
 732         return old_memslots;
 733 }
 734
 735 /*
 736  * Allocate some memory and give it an address in the guest physical address
 737  * space.
 738  *
 739  * Discontiguous memory is allowed, mostly for framebuffers.
 740  *
 741  * Must be called holding mmap_sem for write.
 742  */
 743 int __kvm_set_memory_region(struct kvm *kvm,
 744                             struct kvm_userspace_memory_region *mem)
 745 {
 746         int r;
 747         gfn_t base_gfn;
 748         unsigned long npages;
 749         struct kvm_memory_slot *slot;
 750         struct kvm_memory_slot old, new;
 751         struct kvm_memslots *slots = NULL, *old_memslots;
 752         enum kvm_mr_change change;
 753
 754         r = check_memory_region_flags(mem);
 755         if (r)
 756                 goto out;
 757
 758         r = -EINVAL;
 759         /* General sanity checks */
 760         if (mem->memory_size & (PAGE_SIZE - 1))
 761                 goto out;
 762         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 763                 goto out;
 764         /* We can read the guest memory with __xxx_user() later on. */
 765         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 766             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 767              !access_ok(VERIFY_WRITE,
 768                         (void __user *)(unsigned long)mem->userspace_addr,
 769                         mem->memory_size)))
 770                 goto out;
 771         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 772                 goto out;
 773         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 774                 goto out;
 775
 776         slot = id_to_memslot(kvm->memslots, mem->slot);
 777         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 778         npages = mem->memory_size >> PAGE_SHIFT;
 779
 780         r = -EINVAL;
 781         if (npages > KVM_MEM_MAX_NR_PAGES)
 782                 goto out;
 783
 784         if (!npages)
 785                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 786
 787         new = old = *slot;
 788
 789         new.id = mem->slot;
 790         new.base_gfn = base_gfn;
 791         new.npages = npages;
 792         new.flags = mem->flags;
 793
 794         r = -EINVAL;
 795         if (npages) {
 796                 if (!old.npages)
 797                         change = KVM_MR_CREATE;
 798                 else { /* Modify an existing slot. */
 799                         if ((mem->userspace_addr != old.userspace_addr) ||
 800                             (npages != old.npages) ||
 801                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 802                                 goto out;
 803
 804                         if (base_gfn != old.base_gfn)
 805                                 change = KVM_MR_MOVE;
 806                         else if (new.flags != old.flags)
 807                                 change = KVM_MR_FLAGS_ONLY;
 808                         else { /* Nothing to change. */
 809                                 r = 0;
 810                                 goto out;
 811                         }
 812                 }
 813         } else if (old.npages) {
 814                 change = KVM_MR_DELETE;
 815         } else /* Modify a non-existent slot: disallowed. */
 816                 goto out;
 817
 818         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 819                 /* Check for overlaps */
 820                 r = -EEXIST;
 821                 kvm_for_each_memslot(slot, kvm->memslots) {
 822                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 823                             (slot->id == mem->slot))
 824                                 continue;
 825                         if (!((base_gfn + npages <= slot->base_gfn) ||
 826                               (base_gfn >= slot->base_gfn + slot->npages)))
 827                                 goto out;
 828                 }
 829         }
 830
 831         /* Free page dirty bitmap if unneeded */
 832         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 833                 new.dirty_bitmap = NULL;
 834
 835         r = -ENOMEM;
 836         if (change == KVM_MR_CREATE) {
 837                 new.userspace_addr = mem->userspace_addr;
 838
 839                 if (kvm_arch_create_memslot(kvm, &new, npages))
 840                         goto out_free;
 841         }
 842
 843         /* Allocate page dirty bitmap if needed */
 844         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 845                 if (kvm_create_dirty_bitmap(&new) < 0)
 846                         goto out_free;
 847         }
 848
 849         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 850                 r = -ENOMEM;
 851                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 852                                 GFP_KERNEL);
 853                 if (!slots)
 854                         goto out_free;
 855                 slot = id_to_memslot(slots, mem->slot);
 856                 slot->flags |= KVM_MEMSLOT_INVALID;
 857
 858                 old_memslots = install_new_memslots(kvm, slots, NULL);
 859
 860                 /* slot was deleted or moved, clear iommu mapping */
 861                 kvm_iommu_unmap_pages(kvm, &old);
 862                 /* From this point no new shadow pages pointing to a deleted,
 863                  * or moved, memslot will be created.
 864                  *
 865                  * validation of sp->gfn happens in:
 866                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 867                  *      - kvm_is_visible_gfn (mmu_check_roots)
 868                  */
 869                 kvm_arch_flush_shadow_memslot(kvm, slot);
 870                 slots = old_memslots;
 871         }
 872
 873         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 874         if (r)
 875                 goto out_slots;
 876
 877         r = -ENOMEM;
 878         /*
 879          * We can re-use the old_memslots from above, the only difference
 880          * from the currently installed memslots is the invalid flag.  This
 881          * will get overwritten by update_memslots anyway.
 882          */
 883         if (!slots) {
 884                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 885                                 GFP_KERNEL);
 886                 if (!slots)
 887                         goto out_free;
 888         }
 889
 890         /* actual memory is freed via old in kvm_free_physmem_slot below */
 891         if (change == KVM_MR_DELETE) {
 892                 new.dirty_bitmap = NULL;
 893                 memset(&new.arch, 0, sizeof(new.arch));
 894         }
 895
 896         old_memslots = install_new_memslots(kvm, slots, &new);
 897
 898         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 899
 900         kvm_free_physmem_slot(kvm, &old, &new);
 901         kfree(old_memslots);
 902
 903         /*
 904          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 905          * un-mapped and re-mapped if their base changes.  Since base change
 906          * unmapping is handled above with slot deletion, mapping alone is
 907          * needed here.  Anything else the iommu might care about for existing
 908          * slots (size changes, userspace addr changes and read-only flag
 909          * changes) is disallowed above, so any other attribute changes getting
 910          * here can be skipped.
 911          */
 912         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 913                 r = kvm_iommu_map_pages(kvm, &new);
 914                 return r;
 915         }
 916
 917         return 0;
 918
 919 out_slots:
 920         kfree(slots);
 921 out_free:
 922         kvm_free_physmem_slot(kvm, &new, &old);
 923 out:
 924         return r;
 925 }
 926 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 927
 928 int kvm_set_memory_region(struct kvm *kvm,
 929                           struct kvm_userspace_memory_region *mem)
 930 {
 931         int r;
 932
 933         mutex_lock(&kvm->slots_lock);
 934         r = __kvm_set_memory_region(kvm, mem);
 935         mutex_unlock(&kvm->slots_lock);
 936         return r;
 937 }
 938 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 939
 940 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 941                                           struct kvm_userspace_memory_region *mem)
 942 {
 943         if (mem->slot >= KVM_USER_MEM_SLOTS)
 944                 return -EINVAL;
 945         return kvm_set_memory_region(kvm, mem);
 946 }
 947
 948 int kvm_get_dirty_log(struct kvm *kvm,
 949                         struct kvm_dirty_log *log, int *is_dirty)
 950 {
 951         struct kvm_memory_slot *memslot;
 952         int r, i;
 953         unsigned long n;
 954         unsigned long any = 0;
 955
 956         r = -EINVAL;
 957         if (log->slot >= KVM_USER_MEM_SLOTS)
 958                 goto out;
 959
 960         memslot = id_to_memslot(kvm->memslots, log->slot);
 961         r = -ENOENT;
 962         if (!memslot->dirty_bitmap)
 963                 goto out;
 964
 965         n = kvm_dirty_bitmap_bytes(memslot);
 966
 967         for (i = 0; !any && i < n/sizeof(long); ++i)
 968                 any = memslot->dirty_bitmap[i];
 969
 970         r = -EFAULT;
 971         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 972                 goto out;
 973
 974         if (any)
 975                 *is_dirty = 1;
 976
 977         r = 0;
 978 out:
 979         return r;
 980 }
 981 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
 982
 983 bool kvm_largepages_enabled(void)
 984 {
 985         return largepages_enabled;
 986 }
 987
 988 void kvm_disable_largepages(void)
 989 {
 990         largepages_enabled = false;
 991 }
 992 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 993
 994 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 995 {
 996         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 997 }
 998 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 999
1000 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1001 {
1002         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1003
1004         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1005               memslot->flags & KVM_MEMSLOT_INVALID)
1006                 return 0;
1007
1008         return 1;
1009 }
1010 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1011
1012 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1013 {
1014         struct vm_area_struct *vma;
1015         unsigned long addr, size;
1016
1017         size = PAGE_SIZE;
1018
1019         addr = gfn_to_hva(kvm, gfn);
1020         if (kvm_is_error_hva(addr))
1021                 return PAGE_SIZE;
1022
1023         down_read(&current->mm->mmap_sem);
1024         vma = find_vma(current->mm, addr);
1025         if (!vma)
1026                 goto out;
1027
1028         size = vma_kernel_pagesize(vma);
1029
1030 out:
1031         up_read(&current->mm->mmap_sem);
1032
1033         return size;
1034 }
1035
1036 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1037 {
1038         return slot->flags & KVM_MEM_READONLY;
1039 }
1040
1041 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1042                                        gfn_t *nr_pages, bool write)
1043 {
1044         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1045                 return KVM_HVA_ERR_BAD;
1046
1047         if (memslot_is_readonly(slot) && write)
1048                 return KVM_HVA_ERR_RO_BAD;
1049
1050         if (nr_pages)
1051                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1052
1053         return __gfn_to_hva_memslot(slot, gfn);
1054 }
1055
1056 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1057                                      gfn_t *nr_pages)
1058 {
1059         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1060 }
1061
1062 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1063                                         gfn_t gfn)
1064 {
1065         return gfn_to_hva_many(slot, gfn, NULL);
1066 }
1067 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1068
1069 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1070 {
1071         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1072 }
1073 EXPORT_SYMBOL_GPL(gfn_to_hva);
1074
1075 /*
1076  * If writable is set to false, the hva returned by this function is only
1077  * allowed to be read.
1078  */
1079 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1080                                       gfn_t gfn, bool *writable)
1081 {
1082         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1083
1084         if (!kvm_is_error_hva(hva) && writable)
1085                 *writable = !memslot_is_readonly(slot);
1086
1087         return hva;
1088 }
1089
1090 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1091 {
1092         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1093
1094         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1095 }
1096
1097 static int kvm_read_hva(void *data, void __user *hva, int len)
1098 {
1099         return __copy_from_user(data, hva, len);
1100 }
1101
1102 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1103 {
1104         return __copy_from_user_inatomic(data, hva, len);
1105 }
1106
1107 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1108         unsigned long start, int write, struct page **page)
1109 {
1110         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1111
1112         if (write)
1113                 flags |= FOLL_WRITE;
1114
1115         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1116 }
1117
1118 static inline int check_user_page_hwpoison(unsigned long addr)
1119 {
1120         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1121
1122         rc = __get_user_pages(current, current->mm, addr, 1,
1123                               flags, NULL, NULL, NULL);
1124         return rc == -EHWPOISON;
1125 }
1126
1127 /*
1128  * The atomic path to get the writable pfn which will be stored in @pfn,
1129  * true indicates success, otherwise false is returned.
1130  */
1131 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1132                             bool write_fault, bool *writable, pfn_t *pfn)
1133 {
1134         struct page *page[1];
1135         int npages;
1136
1137         if (!(async || atomic))
1138                 return false;
1139
1140         /*
1141          * Fast pin a writable pfn only if it is a write fault request
1142          * or the caller allows to map a writable pfn for a read fault
1143          * request.
1144          */
1145         if (!(write_fault || writable))
1146                 return false;
1147
1148         npages = __get_user_pages_fast(addr, 1, 1, page);
1149         if (npages == 1) {
1150                 *pfn = page_to_pfn(page[0]);
1151
1152                 if (writable)
1153                         *writable = true;
1154                 return true;
1155         }
1156
1157         return false;
1158 }
1159
1160 /*
1161  * The slow path to get the pfn of the specified host virtual address,
1162  * 1 indicates success, -errno is returned if error is detected.
1163  */
1164 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1165                            bool *writable, pfn_t *pfn)
1166 {
1167         struct page *page[1];
1168         int npages = 0;
1169
1170         might_sleep();
1171
1172         if (writable)
1173                 *writable = write_fault;
1174
1175         if (async) {
1176                 down_read(&current->mm->mmap_sem);
1177                 npages = get_user_page_nowait(current, current->mm,
1178                                               addr, write_fault, page);
1179                 up_read(&current->mm->mmap_sem);
1180         } else
1181                 npages = get_user_pages_fast(addr, 1, write_fault,
1182                                              page);
1183         if (npages != 1)
1184                 return npages;
1185
1186         /* map read fault as writable if possible */
1187         if (unlikely(!write_fault) && writable) {
1188                 struct page *wpage[1];
1189
1190                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1191                 if (npages == 1) {
1192                         *writable = true;
1193                         put_page(page[0]);
1194                         page[0] = wpage[0];
1195                 }
1196
1197                 npages = 1;
1198         }
1199         *pfn = page_to_pfn(page[0]);
1200         return npages;
1201 }
1202
1203 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1204 {
1205         if (unlikely(!(vma->vm_flags & VM_READ)))
1206                 return false;
1207
1208         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1209                 return false;
1210
1211         return true;
1212 }
1213
1214 /*
1215  * Pin guest page in memory and return its pfn.
1216  * @addr: host virtual address which maps memory to the guest
1217  * @atomic: whether this function can sleep
1218  * @async: whether this function need to wait IO complete if the
1219  *         host page is not in the memory
1220  * @write_fault: whether we should get a writable host page
1221  * @writable: whether it allows to map a writable host page for !@write_fault
1222  *
1223  * The function will map a writable host page for these two cases:
1224  * 1): @write_fault = true
1225  * 2): @write_fault = false && @writable, @writable will tell the caller
1226  *     whether the mapping is writable.
1227  */
1228 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1229                         bool write_fault, bool *writable)
1230 {
1231         struct vm_area_struct *vma;
1232         pfn_t pfn = 0;
1233         int npages;
1234
1235         /* we can do it either atomically or asynchronously, not both */
1236         BUG_ON(atomic && async);
1237
1238         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1239                 return pfn;
1240
1241         if (atomic)
1242                 return KVM_PFN_ERR_FAULT;
1243
1244         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1245         if (npages == 1)
1246                 return pfn;
1247
1248         down_read(&current->mm->mmap_sem);
1249         if (npages == -EHWPOISON ||
1250               (!async && check_user_page_hwpoison(addr))) {
1251                 pfn = KVM_PFN_ERR_HWPOISON;
1252                 goto exit;
1253         }
1254
1255         vma = find_vma_intersection(current->mm, addr, addr + 1);
1256
1257         if (vma == NULL)
1258                 pfn = KVM_PFN_ERR_FAULT;
1259         else if ((vma->vm_flags & VM_PFNMAP)) {
1260                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1261                         vma->vm_pgoff;
1262                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1263         } else {
1264                 if (async && vma_is_valid(vma, write_fault))
1265                         *async = true;
1266                 pfn = KVM_PFN_ERR_FAULT;
1267         }
1268 exit:
1269         up_read(&current->mm->mmap_sem);
1270         return pfn;
1271 }
1272
1273 static pfn_t
1274 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1275                      bool *async, bool write_fault, bool *writable)
1276 {
1277         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1278
1279         if (addr == KVM_HVA_ERR_RO_BAD)
1280                 return KVM_PFN_ERR_RO_FAULT;
1281
1282         if (kvm_is_error_hva(addr))
1283                 return KVM_PFN_NOSLOT;
1284
1285         /* Do not map writable pfn in the readonly memslot. */
1286         if (writable && memslot_is_readonly(slot)) {
1287                 *writable = false;
1288                 writable = NULL;
1289         }
1290
1291         return hva_to_pfn(addr, atomic, async, write_fault,
1292                           writable);
1293 }
1294
1295 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1296                           bool write_fault, bool *writable)
1297 {
1298         struct kvm_memory_slot *slot;
1299
1300         if (async)
1301                 *async = false;
1302
1303         slot = gfn_to_memslot(kvm, gfn);
1304
1305         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1306                                     writable);
1307 }
1308
1309 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1310 {
1311         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1312 }
1313 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1314
1315 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1316                        bool write_fault, bool *writable)
1317 {
1318         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1319 }
1320 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1321
1322 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1323 {
1324         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1325 }
1326 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1327
1328 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1329                       bool *writable)
1330 {
1331         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1332 }
1333 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1334
1335 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1336 {
1337         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1338 }
1339
1340 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1341 {
1342         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1343 }
1344 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1345
1346 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1347                                                                   int nr_pages)
1348 {
1349         unsigned long addr;
1350         gfn_t entry;
1351
1352         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1353         if (kvm_is_error_hva(addr))
1354                 return -1;
1355
1356         if (entry < nr_pages)
1357                 return 0;
1358
1359         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1360 }
1361 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1362
1363 static struct page *kvm_pfn_to_page(pfn_t pfn)
1364 {
1365         if (is_error_noslot_pfn(pfn))
1366                 return KVM_ERR_PTR_BAD_PAGE;
1367
1368         if (kvm_is_mmio_pfn(pfn)) {
1369                 WARN_ON(1);
1370                 return KVM_ERR_PTR_BAD_PAGE;
1371         }
1372
1373         return pfn_to_page(pfn);
1374 }
1375
1376 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1377 {
1378         pfn_t pfn;
1379
1380         pfn = gfn_to_pfn(kvm, gfn);
1381
1382         return kvm_pfn_to_page(pfn);
1383 }
1384
1385 EXPORT_SYMBOL_GPL(gfn_to_page);
1386
1387 void kvm_release_page_clean(struct page *page)
1388 {
1389         WARN_ON(is_error_page(page));
1390
1391         kvm_release_pfn_clean(page_to_pfn(page));
1392 }
1393 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1394
1395 void kvm_release_pfn_clean(pfn_t pfn)
1396 {
1397         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1398                 put_page(pfn_to_page(pfn));
1399 }
1400 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1401
1402 void kvm_release_page_dirty(struct page *page)
1403 {
1404         WARN_ON(is_error_page(page));
1405
1406         kvm_release_pfn_dirty(page_to_pfn(page));
1407 }
1408 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1409
1410 static void kvm_release_pfn_dirty(pfn_t pfn)
1411 {
1412         kvm_set_pfn_dirty(pfn);
1413         kvm_release_pfn_clean(pfn);
1414 }
1415
1416 void kvm_set_pfn_dirty(pfn_t pfn)
1417 {
1418         if (!kvm_is_mmio_pfn(pfn)) {
1419                 struct page *page = pfn_to_page(pfn);
1420                 if (!PageReserved(page))
1421                         SetPageDirty(page);
1422         }
1423 }
1424 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1425
1426 void kvm_set_pfn_accessed(pfn_t pfn)
1427 {
1428         if (!kvm_is_mmio_pfn(pfn))
1429                 mark_page_accessed(pfn_to_page(pfn));
1430 }
1431 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1432
1433 void kvm_get_pfn(pfn_t pfn)
1434 {
1435         if (!kvm_is_mmio_pfn(pfn))
1436                 get_page(pfn_to_page(pfn));
1437 }
1438 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1439
1440 static int next_segment(unsigned long len, int offset)
1441 {
1442         if (len > PAGE_SIZE - offset)
1443                 return PAGE_SIZE - offset;
1444         else
1445                 return len;
1446 }
1447
1448 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1449                         int len)
1450 {
1451         int r;
1452         unsigned long addr;
1453
1454         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1455         if (kvm_is_error_hva(addr))
1456                 return -EFAULT;
1457         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1458         if (r)
1459                 return -EFAULT;
1460         return 0;
1461 }
1462 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1463
1464 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1465 {
1466         gfn_t gfn = gpa >> PAGE_SHIFT;
1467         int seg;
1468         int offset = offset_in_page(gpa);
1469         int ret;
1470
1471         while ((seg = next_segment(len, offset)) != 0) {
1472                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1473                 if (ret < 0)
1474                         return ret;
1475                 offset = 0;
1476                 len -= seg;
1477                 data += seg;
1478                 ++gfn;
1479         }
1480         return 0;
1481 }
1482 EXPORT_SYMBOL_GPL(kvm_read_guest);
1483
1484 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1485                           unsigned long len)
1486 {
1487         int r;
1488         unsigned long addr;
1489         gfn_t gfn = gpa >> PAGE_SHIFT;
1490         int offset = offset_in_page(gpa);
1491
1492         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1493         if (kvm_is_error_hva(addr))
1494                 return -EFAULT;
1495         pagefault_disable();
1496         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1497         pagefault_enable();
1498         if (r)
1499                 return -EFAULT;
1500         return 0;
1501 }
1502 EXPORT_SYMBOL(kvm_read_guest_atomic);
1503
1504 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1505                          int offset, int len)
1506 {
1507         int r;
1508         unsigned long addr;
1509
1510         addr = gfn_to_hva(kvm, gfn);
1511         if (kvm_is_error_hva(addr))
1512                 return -EFAULT;
1513         r = __copy_to_user((void __user *)addr + offset, data, len);
1514         if (r)
1515                 return -EFAULT;
1516         mark_page_dirty(kvm, gfn);
1517         return 0;
1518 }
1519 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1520
1521 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1522                     unsigned long len)
1523 {
1524         gfn_t gfn = gpa >> PAGE_SHIFT;
1525         int seg;
1526         int offset = offset_in_page(gpa);
1527         int ret;
1528
1529         while ((seg = next_segment(len, offset)) != 0) {
1530                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1531                 if (ret < 0)
1532                         return ret;
1533                 offset = 0;
1534                 len -= seg;
1535                 data += seg;
1536                 ++gfn;
1537         }
1538         return 0;
1539 }
1540
1541 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1542                               gpa_t gpa, unsigned long len)
1543 {
1544         struct kvm_memslots *slots = kvm_memslots(kvm);
1545         int offset = offset_in_page(gpa);
1546         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1547         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1548         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1549         gfn_t nr_pages_avail;
1550
1551         ghc->gpa = gpa;
1552         ghc->generation = slots->generation;
1553         ghc->len = len;
1554         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1555         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1556         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1557                 ghc->hva += offset;
1558         } else {
1559                 /*
1560                  * If the requested region crosses two memslots, we still
1561                  * verify that the entire region is valid here.
1562                  */
1563                 while (start_gfn <= end_gfn) {
1564                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1565                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1566                                                    &nr_pages_avail);
1567                         if (kvm_is_error_hva(ghc->hva))
1568                                 return -EFAULT;
1569                         start_gfn += nr_pages_avail;
1570                 }
1571                 /* Use the slow path for cross page reads and writes. */
1572                 ghc->memslot = NULL;
1573         }
1574         return 0;
1575 }
1576 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1577
1578 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1579                            void *data, unsigned long len)
1580 {
1581         struct kvm_memslots *slots = kvm_memslots(kvm);
1582         int r;
1583
1584         BUG_ON(len > ghc->len);
1585
1586         if (slots->generation != ghc->generation)
1587                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1588
1589         if (unlikely(!ghc->memslot))
1590                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1591
1592         if (kvm_is_error_hva(ghc->hva))
1593                 return -EFAULT;
1594
1595         r = __copy_to_user((void __user *)ghc->hva, data, len);
1596         if (r)
1597                 return -EFAULT;
1598         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1599
1600         return 0;
1601 }
1602 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1603
1604 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1605                            void *data, unsigned long len)
1606 {
1607         struct kvm_memslots *slots = kvm_memslots(kvm);
1608         int r;
1609
1610         BUG_ON(len > ghc->len);
1611
1612         if (slots->generation != ghc->generation)
1613                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1614
1615         if (unlikely(!ghc->memslot))
1616                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1617
1618         if (kvm_is_error_hva(ghc->hva))
1619                 return -EFAULT;
1620
1621         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1622         if (r)
1623                 return -EFAULT;
1624
1625         return 0;
1626 }
1627 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1628
1629 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1630 {
1631         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1632
1633         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1634 }
1635 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1636
1637 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1638 {
1639         gfn_t gfn = gpa >> PAGE_SHIFT;
1640         int seg;
1641         int offset = offset_in_page(gpa);
1642         int ret;
1643
1644         while ((seg = next_segment(len, offset)) != 0) {
1645                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1646                 if (ret < 0)
1647                         return ret;
1648                 offset = 0;
1649                 len -= seg;
1650                 ++gfn;
1651         }
1652         return 0;
1653 }
1654 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1655
1656 static void mark_page_dirty_in_slot(struct kvm *kvm,
1657                                     struct kvm_memory_slot *memslot,
1658                                     gfn_t gfn)
1659 {
1660         if (memslot && memslot->dirty_bitmap) {
1661                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1662
1663                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1664         }
1665 }
1666
1667 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1668 {
1669         struct kvm_memory_slot *memslot;
1670
1671         memslot = gfn_to_memslot(kvm, gfn);
1672         mark_page_dirty_in_slot(kvm, memslot, gfn);
1673 }
1674 EXPORT_SYMBOL_GPL(mark_page_dirty);
1675
1676 /*
1677  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1678  */
1679 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1680 {
1681         DEFINE_WAIT(wait);
1682
1683         for (;;) {
1684                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1685
1686                 if (kvm_arch_vcpu_runnable(vcpu)) {
1687                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1688                         break;
1689                 }
1690                 if (kvm_cpu_has_pending_timer(vcpu))
1691                         break;
1692                 if (signal_pending(current))
1693                         break;
1694
1695                 schedule();
1696         }
1697
1698         finish_wait(&vcpu->wq, &wait);
1699 }
1700 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1701
1702 #ifndef CONFIG_S390
1703 /*
1704  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1705  */
1706 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1707 {
1708         int me;
1709         int cpu = vcpu->cpu;
1710         wait_queue_head_t *wqp;
1711
1712         wqp = kvm_arch_vcpu_wq(vcpu);
1713         if (waitqueue_active(wqp)) {
1714                 wake_up_interruptible(wqp);
1715                 ++vcpu->stat.halt_wakeup;
1716         }
1717
1718         me = get_cpu();
1719         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1720                 if (kvm_arch_vcpu_should_kick(vcpu))
1721                         smp_send_reschedule(cpu);
1722         put_cpu();
1723 }
1724 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1725 #endif /* !CONFIG_S390 */
1726
1727 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1728 {
1729         struct pid *pid;
1730         struct task_struct *task = NULL;
1731         int ret = 0;
1732
1733         rcu_read_lock();
1734         pid = rcu_dereference(target->pid);
1735         if (pid)
1736                 task = get_pid_task(target->pid, PIDTYPE_PID);
1737         rcu_read_unlock();
1738         if (!task)
1739                 return ret;
1740         if (task->flags & PF_VCPU) {
1741                 put_task_struct(task);
1742                 return ret;
1743         }
1744         ret = yield_to(task, 1);
1745         put_task_struct(task);
1746
1747         return ret;
1748 }
1749 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1750
1751 /*
1752  * Helper that checks whether a VCPU is eligible for directed yield.
1753  * Most eligible candidate to yield is decided by following heuristics:
1754  *
1755  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1756  *  (preempted lock holder), indicated by @in_spin_loop.
1757  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1758  *
1759  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1760  *  chance last time (mostly it has become eligible now since we have probably
1761  *  yielded to lockholder in last iteration. This is done by toggling
1762  *  @dy_eligible each time a VCPU checked for eligibility.)
1763  *
1764  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1765  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1766  *  burning. Giving priority for a potential lock-holder increases lock
1767  *  progress.
1768  *
1769  *  Since algorithm is based on heuristics, accessing another VCPU data without
1770  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1771  *  and continue with next VCPU and so on.
1772  */
1773 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1774 {
1775 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1776         bool eligible;
1777
1778         eligible = !vcpu->spin_loop.in_spin_loop ||
1779                         (vcpu->spin_loop.in_spin_loop &&
1780                          vcpu->spin_loop.dy_eligible);
1781
1782         if (vcpu->spin_loop.in_spin_loop)
1783                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1784
1785         return eligible;
1786 #else
1787         return true;
1788 #endif
1789 }
1790
1791 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1792 {
1793         struct kvm *kvm = me->kvm;
1794         struct kvm_vcpu *vcpu;
1795         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1796         int yielded = 0;
1797         int try = 3;
1798         int pass;
1799         int i;
1800
1801         kvm_vcpu_set_in_spin_loop(me, true);
1802         /*
1803          * We boost the priority of a VCPU that is runnable but not
1804          * currently running, because it got preempted by something
1805          * else and called schedule in __vcpu_run.  Hopefully that
1806          * VCPU is holding the lock that we need and will release it.
1807          * We approximate round-robin by starting at the last boosted VCPU.
1808          */
1809         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1810                 kvm_for_each_vcpu(i, vcpu, kvm) {
1811                         if (!pass && i <= last_boosted_vcpu) {
1812                                 i = last_boosted_vcpu;
1813                                 continue;
1814                         } else if (pass && i > last_boosted_vcpu)
1815                                 break;
1816                         if (!ACCESS_ONCE(vcpu->preempted))
1817                                 continue;
1818                         if (vcpu == me)
1819                                 continue;
1820                         if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1821                                 continue;
1822                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1823                                 continue;
1824
1825                         yielded = kvm_vcpu_yield_to(vcpu);
1826                         if (yielded > 0) {
1827                                 kvm->last_boosted_vcpu = i;
1828                                 break;
1829                         } else if (yielded < 0) {
1830                                 try--;
1831                                 if (!try)
1832                                         break;
1833                         }
1834                 }
1835         }
1836         kvm_vcpu_set_in_spin_loop(me, false);
1837
1838         /* Ensure vcpu is not eligible during next spinloop */
1839         kvm_vcpu_set_dy_eligible(me, false);
1840 }
1841 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1842
1843 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1844 {
1845         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1846         struct page *page;
1847
1848         if (vmf->pgoff == 0)
1849                 page = virt_to_page(vcpu->run);
1850 #ifdef CONFIG_X86
1851         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1852                 page = virt_to_page(vcpu->arch.pio_data);
1853 #endif
1854 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1855         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1856                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1857 #endif
1858         else
1859                 return kvm_arch_vcpu_fault(vcpu, vmf);
1860         get_page(page);
1861         vmf->page = page;
1862         return 0;
1863 }
1864
1865 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1866         .fault = kvm_vcpu_fault,
1867 };
1868
1869 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1870 {
1871         vma->vm_ops = &kvm_vcpu_vm_ops;
1872         return 0;
1873 }
1874
1875 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1876 {
1877         struct kvm_vcpu *vcpu = filp->private_data;
1878
1879         kvm_put_kvm(vcpu->kvm);
1880         return 0;
1881 }
1882
1883 static struct file_operations kvm_vcpu_fops = {
1884         .release        = kvm_vcpu_release,
1885         .unlocked_ioctl = kvm_vcpu_ioctl,
1886 #ifdef CONFIG_COMPAT
1887         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1888 #endif
1889         .mmap           = kvm_vcpu_mmap,
1890         .llseek         = noop_llseek,
1891 };
1892
1893 /*
1894  * Allocates an inode for the vcpu.
1895  */
1896 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1897 {
1898         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1899 }
1900
1901 /*
1902  * Creates some virtual cpus.  Good luck creating more than one.
1903  */
1904 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1905 {
1906         int r;
1907         struct kvm_vcpu *vcpu, *v;
1908
1909         if (id >= KVM_MAX_VCPUS)
1910                 return -EINVAL;
1911
1912         vcpu = kvm_arch_vcpu_create(kvm, id);
1913         if (IS_ERR(vcpu))
1914                 return PTR_ERR(vcpu);
1915
1916         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1917
1918         r = kvm_arch_vcpu_setup(vcpu);
1919         if (r)
1920                 goto vcpu_destroy;
1921
1922         mutex_lock(&kvm->lock);
1923         if (!kvm_vcpu_compatible(vcpu)) {
1924                 r = -EINVAL;
1925                 goto unlock_vcpu_destroy;
1926         }
1927         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1928                 r = -EINVAL;
1929                 goto unlock_vcpu_destroy;
1930         }
1931
1932         kvm_for_each_vcpu(r, v, kvm)
1933                 if (v->vcpu_id == id) {
1934                         r = -EEXIST;
1935                         goto unlock_vcpu_destroy;
1936                 }
1937
1938         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1939
1940         /* Now it's all set up, let userspace reach it */
1941         kvm_get_kvm(kvm);
1942         r = create_vcpu_fd(vcpu);
1943         if (r < 0) {
1944                 kvm_put_kvm(kvm);
1945                 goto unlock_vcpu_destroy;
1946         }
1947
1948         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1949         smp_wmb();
1950         atomic_inc(&kvm->online_vcpus);
1951
1952         mutex_unlock(&kvm->lock);
1953         kvm_arch_vcpu_postcreate(vcpu);
1954         return r;
1955
1956 unlock_vcpu_destroy:
1957         mutex_unlock(&kvm->lock);
1958 vcpu_destroy:
1959         kvm_arch_vcpu_destroy(vcpu);
1960         return r;
1961 }
1962
1963 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1964 {
1965         if (sigset) {
1966                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1967                 vcpu->sigset_active = 1;
1968                 vcpu->sigset = *sigset;
1969         } else
1970                 vcpu->sigset_active = 0;
1971         return 0;
1972 }
1973
1974 static long kvm_vcpu_ioctl(struct file *filp,
1975                            unsigned int ioctl, unsigned long arg)
1976 {
1977         struct kvm_vcpu *vcpu = filp->private_data;
1978         void __user *argp = (void __user *)arg;
1979         int r;
1980         struct kvm_fpu *fpu = NULL;
1981         struct kvm_sregs *kvm_sregs = NULL;
1982
1983         if (vcpu->kvm->mm != current->mm)
1984                 return -EIO;
1985
1986 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1987         /*
1988          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1989          * so vcpu_load() would break it.
1990          */
1991         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1992                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1993 #endif
1994
1995
1996         r = vcpu_load(vcpu);
1997         if (r)
1998                 return r;
1999         switch (ioctl) {
2000         case KVM_RUN:
2001                 r = -EINVAL;
2002                 if (arg)
2003                         goto out;
2004                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2005                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2006                 break;
2007         case KVM_GET_REGS: {
2008                 struct kvm_regs *kvm_regs;
2009
2010                 r = -ENOMEM;
2011                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2012                 if (!kvm_regs)
2013                         goto out;
2014                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2015                 if (r)
2016                         goto out_free1;
2017                 r = -EFAULT;
2018                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2019                         goto out_free1;
2020                 r = 0;
2021 out_free1:
2022                 kfree(kvm_regs);
2023                 break;
2024         }
2025         case KVM_SET_REGS: {
2026                 struct kvm_regs *kvm_regs;
2027
2028                 r = -ENOMEM;
2029                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2030                 if (IS_ERR(kvm_regs)) {
2031                         r = PTR_ERR(kvm_regs);
2032                         goto out;
2033                 }
2034                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2035                 kfree(kvm_regs);
2036                 break;
2037         }
2038         case KVM_GET_SREGS: {
2039                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2040                 r = -ENOMEM;
2041                 if (!kvm_sregs)
2042                         goto out;
2043                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2044                 if (r)
2045                         goto out;
2046                 r = -EFAULT;
2047                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2048                         goto out;
2049                 r = 0;
2050                 break;
2051         }
2052         case KVM_SET_SREGS: {
2053                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2054                 if (IS_ERR(kvm_sregs)) {
2055                         r = PTR_ERR(kvm_sregs);
2056                         kvm_sregs = NULL;
2057                         goto out;
2058                 }
2059                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2060                 break;
2061         }
2062         case KVM_GET_MP_STATE: {
2063                 struct kvm_mp_state mp_state;
2064
2065                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2066                 if (r)
2067                         goto out;
2068                 r = -EFAULT;
2069                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2070                         goto out;
2071                 r = 0;
2072                 break;
2073         }
2074         case KVM_SET_MP_STATE: {
2075                 struct kvm_mp_state mp_state;
2076
2077                 r = -EFAULT;
2078                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2079                         goto out;
2080                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2081                 break;
2082         }
2083         case KVM_TRANSLATE: {
2084                 struct kvm_translation tr;
2085
2086                 r = -EFAULT;
2087                 if (copy_from_user(&tr, argp, sizeof tr))
2088                         goto out;
2089                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2090                 if (r)
2091                         goto out;
2092                 r = -EFAULT;
2093                 if (copy_to_user(argp, &tr, sizeof tr))
2094                         goto out;
2095                 r = 0;
2096                 break;
2097         }
2098         case KVM_SET_GUEST_DEBUG: {
2099                 struct kvm_guest_debug dbg;
2100
2101                 r = -EFAULT;
2102                 if (copy_from_user(&dbg, argp, sizeof dbg))
2103                         goto out;
2104                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2105                 break;
2106         }
2107         case KVM_SET_SIGNAL_MASK: {
2108                 struct kvm_signal_mask __user *sigmask_arg = argp;
2109                 struct kvm_signal_mask kvm_sigmask;
2110                 sigset_t sigset, *p;
2111
2112                 p = NULL;
2113                 if (argp) {
2114                         r = -EFAULT;
2115                         if (copy_from_user(&kvm_sigmask, argp,
2116                                            sizeof kvm_sigmask))
2117                                 goto out;
2118                         r = -EINVAL;
2119                         if (kvm_sigmask.len != sizeof sigset)
2120                                 goto out;
2121                         r = -EFAULT;
2122                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2123                                            sizeof sigset))
2124                                 goto out;
2125                         p = &sigset;
2126                 }
2127                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2128                 break;
2129         }
2130         case KVM_GET_FPU: {
2131                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2132                 r = -ENOMEM;
2133                 if (!fpu)
2134                         goto out;
2135                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2136                 if (r)
2137                         goto out;
2138                 r = -EFAULT;
2139                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2140                         goto out;
2141                 r = 0;
2142                 break;
2143         }
2144         case KVM_SET_FPU: {
2145                 fpu = memdup_user(argp, sizeof(*fpu));
2146                 if (IS_ERR(fpu)) {
2147                         r = PTR_ERR(fpu);
2148                         fpu = NULL;
2149                         goto out;
2150                 }
2151                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2152                 break;
2153         }
2154         default:
2155                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2156         }
2157 out:
2158         vcpu_put(vcpu);
2159         kfree(fpu);
2160         kfree(kvm_sregs);
2161         return r;
2162 }
2163
2164 #ifdef CONFIG_COMPAT
2165 static long kvm_vcpu_compat_ioctl(struct file *filp,
2166                                   unsigned int ioctl, unsigned long arg)
2167 {
2168         struct kvm_vcpu *vcpu = filp->private_data;
2169         void __user *argp = compat_ptr(arg);
2170         int r;
2171
2172         if (vcpu->kvm->mm != current->mm)
2173                 return -EIO;
2174
2175         switch (ioctl) {
2176         case KVM_SET_SIGNAL_MASK: {
2177                 struct kvm_signal_mask __user *sigmask_arg = argp;
2178                 struct kvm_signal_mask kvm_sigmask;
2179                 compat_sigset_t csigset;
2180                 sigset_t sigset;
2181
2182                 if (argp) {
2183                         r = -EFAULT;
2184                         if (copy_from_user(&kvm_sigmask, argp,
2185                                            sizeof kvm_sigmask))
2186                                 goto out;
2187                         r = -EINVAL;
2188                         if (kvm_sigmask.len != sizeof csigset)
2189                                 goto out;
2190                         r = -EFAULT;
2191                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2192                                            sizeof csigset))
2193                                 goto out;
2194                         sigset_from_compat(&sigset, &csigset);
2195                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2196                 } else
2197                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2198                 break;
2199         }
2200         default:
2201                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2202         }
2203
2204 out:
2205         return r;
2206 }
2207 #endif
2208
2209 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2210                                  int (*accessor)(struct kvm_device *dev,
2211                                                  struct kvm_device_attr *attr),
2212                                  unsigned long arg)
2213 {
2214         struct kvm_device_attr attr;
2215
2216         if (!accessor)
2217                 return -EPERM;
2218
2219         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2220                 return -EFAULT;
2221
2222         return accessor(dev, &attr);
2223 }
2224
2225 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2226                              unsigned long arg)
2227 {
2228         struct kvm_device *dev = filp->private_data;
2229
2230         switch (ioctl) {
2231         case KVM_SET_DEVICE_ATTR:
2232                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2233         case KVM_GET_DEVICE_ATTR:
2234                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2235         case KVM_HAS_DEVICE_ATTR:
2236                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2237         default:
2238                 if (dev->ops->ioctl)
2239                         return dev->ops->ioctl(dev, ioctl, arg);
2240
2241                 return -ENOTTY;
2242         }
2243 }
2244
2245 static int kvm_device_release(struct inode *inode, struct file *filp)
2246 {
2247         struct kvm_device *dev = filp->private_data;
2248         struct kvm *kvm = dev->kvm;
2249
2250         kvm_put_kvm(kvm);
2251         return 0;
2252 }
2253
2254 static const struct file_operations kvm_device_fops = {
2255         .unlocked_ioctl = kvm_device_ioctl,
2256 #ifdef CONFIG_COMPAT
2257         .compat_ioctl = kvm_device_ioctl,
2258 #endif
2259         .release = kvm_device_release,
2260 };
2261
2262 struct kvm_device *kvm_device_from_filp(struct file *filp)
2263 {
2264         if (filp->f_op != &kvm_device_fops)
2265                 return NULL;
2266
2267         return filp->private_data;
2268 }
2269
2270 static int kvm_ioctl_create_device(struct kvm *kvm,
2271                                    struct kvm_create_device *cd)
2272 {
2273         struct kvm_device_ops *ops = NULL;
2274         struct kvm_device *dev;
2275         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2276         int ret;
2277
2278         switch (cd->type) {
2279 #ifdef CONFIG_KVM_MPIC
2280         case KVM_DEV_TYPE_FSL_MPIC_20:
2281         case KVM_DEV_TYPE_FSL_MPIC_42:
2282                 ops = &kvm_mpic_ops;
2283                 break;
2284 #endif
2285 #ifdef CONFIG_KVM_XICS
2286         case KVM_DEV_TYPE_XICS:
2287                 ops = &kvm_xics_ops;
2288                 break;
2289 #endif
2290 #ifdef CONFIG_KVM_VFIO
2291         case KVM_DEV_TYPE_VFIO:
2292                 ops = &kvm_vfio_ops;
2293                 break;
2294 #endif
2295 #ifdef CONFIG_KVM_ARM_VGIC
2296         case KVM_DEV_TYPE_ARM_VGIC_V2:
2297                 ops = &kvm_arm_vgic_v2_ops;
2298                 break;
2299 #endif
2300 #ifdef CONFIG_S390
2301         case KVM_DEV_TYPE_FLIC:
2302                 ops = &kvm_flic_ops;
2303                 break;
2304 #endif
2305         default:
2306                 return -ENODEV;
2307         }
2308
2309         if (test)
2310                 return 0;
2311
2312         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2313         if (!dev)
2314                 return -ENOMEM;
2315
2316         dev->ops = ops;
2317         dev->kvm = kvm;
2318
2319         ret = ops->create(dev, cd->type);
2320         if (ret < 0) {
2321                 kfree(dev);
2322                 return ret;
2323         }
2324
2325         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2326         if (ret < 0) {
2327                 ops->destroy(dev);
2328                 return ret;
2329         }
2330
2331         list_add(&dev->vm_node, &kvm->devices);
2332         kvm_get_kvm(kvm);
2333         cd->fd = ret;
2334         return 0;
2335 }
2336
2337 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2338 {
2339         switch (arg) {
2340         case KVM_CAP_USER_MEMORY:
2341         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2342         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2343 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2344         case KVM_CAP_SET_BOOT_CPU_ID:
2345 #endif
2346         case KVM_CAP_INTERNAL_ERROR_DATA:
2347 #ifdef CONFIG_HAVE_KVM_MSI
2348         case KVM_CAP_SIGNAL_MSI:
2349 #endif
2350 #ifdef CONFIG_HAVE_KVM_IRQFD
2351         case KVM_CAP_IRQFD_RESAMPLE:
2352 #endif
2353         case KVM_CAP_CHECK_EXTENSION_VM:
2354                 return 1;
2355 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2356         case KVM_CAP_IRQ_ROUTING:
2357                 return KVM_MAX_IRQ_ROUTES;
2358 #endif
2359         default:
2360                 break;
2361         }
2362         return kvm_vm_ioctl_check_extension(kvm, arg);
2363 }
2364
2365 static long kvm_vm_ioctl(struct file *filp,
2366                            unsigned int ioctl, unsigned long arg)
2367 {
2368         struct kvm *kvm = filp->private_data;
2369         void __user *argp = (void __user *)arg;
2370         int r;
2371
2372         if (kvm->mm != current->mm)
2373                 return -EIO;
2374         switch (ioctl) {
2375         case KVM_CREATE_VCPU:
2376                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2377                 break;
2378         case KVM_SET_USER_MEMORY_REGION: {
2379                 struct kvm_userspace_memory_region kvm_userspace_mem;
2380
2381                 r = -EFAULT;
2382                 if (copy_from_user(&kvm_userspace_mem, argp,
2383                                                 sizeof kvm_userspace_mem))
2384                         goto out;
2385
2386                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2387                 break;
2388         }
2389         case KVM_GET_DIRTY_LOG: {
2390                 struct kvm_dirty_log log;
2391
2392                 r = -EFAULT;
2393                 if (copy_from_user(&log, argp, sizeof log))
2394                         goto out;
2395                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2396                 break;
2397         }
2398 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2399         case KVM_REGISTER_COALESCED_MMIO: {
2400                 struct kvm_coalesced_mmio_zone zone;
2401                 r = -EFAULT;
2402                 if (copy_from_user(&zone, argp, sizeof zone))
2403                         goto out;
2404                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2405                 break;
2406         }
2407         case KVM_UNREGISTER_COALESCED_MMIO: {
2408                 struct kvm_coalesced_mmio_zone zone;
2409                 r = -EFAULT;
2410                 if (copy_from_user(&zone, argp, sizeof zone))
2411                         goto out;
2412                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2413                 break;
2414         }
2415 #endif
2416         case KVM_IRQFD: {
2417                 struct kvm_irqfd data;
2418
2419                 r = -EFAULT;
2420                 if (copy_from_user(&data, argp, sizeof data))
2421                         goto out;
2422                 r = kvm_irqfd(kvm, &data);
2423                 break;
2424         }
2425         case KVM_IOEVENTFD: {
2426                 struct kvm_ioeventfd data;
2427
2428                 r = -EFAULT;
2429                 if (copy_from_user(&data, argp, sizeof data))
2430                         goto out;
2431                 r = kvm_ioeventfd(kvm, &data);
2432                 break;
2433         }
2434 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2435         case KVM_SET_BOOT_CPU_ID:
2436                 r = 0;
2437                 mutex_lock(&kvm->lock);
2438                 if (atomic_read(&kvm->online_vcpus) != 0)
2439                         r = -EBUSY;
2440                 else
2441                         kvm->bsp_vcpu_id = arg;
2442                 mutex_unlock(&kvm->lock);
2443                 break;
2444 #endif
2445 #ifdef CONFIG_HAVE_KVM_MSI
2446         case KVM_SIGNAL_MSI: {
2447                 struct kvm_msi msi;
2448
2449                 r = -EFAULT;
2450                 if (copy_from_user(&msi, argp, sizeof msi))
2451                         goto out;
2452                 r = kvm_send_userspace_msi(kvm, &msi);
2453                 break;
2454         }
2455 #endif
2456 #ifdef __KVM_HAVE_IRQ_LINE
2457         case KVM_IRQ_LINE_STATUS:
2458         case KVM_IRQ_LINE: {
2459                 struct kvm_irq_level irq_event;
2460
2461                 r = -EFAULT;
2462                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2463                         goto out;
2464
2465                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2466                                         ioctl == KVM_IRQ_LINE_STATUS);
2467                 if (r)
2468                         goto out;
2469
2470                 r = -EFAULT;
2471                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2472                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2473                                 goto out;
2474                 }
2475
2476                 r = 0;
2477                 break;
2478         }
2479 #endif
2480 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2481         case KVM_SET_GSI_ROUTING: {
2482                 struct kvm_irq_routing routing;
2483                 struct kvm_irq_routing __user *urouting;
2484                 struct kvm_irq_routing_entry *entries;
2485
2486                 r = -EFAULT;
2487                 if (copy_from_user(&routing, argp, sizeof(routing)))
2488                         goto out;
2489                 r = -EINVAL;
2490                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2491                         goto out;
2492                 if (routing.flags)
2493                         goto out;
2494                 r = -ENOMEM;
2495                 entries = vmalloc(routing.nr * sizeof(*entries));
2496                 if (!entries)
2497                         goto out;
2498                 r = -EFAULT;
2499                 urouting = argp;
2500                 if (copy_from_user(entries, urouting->entries,
2501                                    routing.nr * sizeof(*entries)))
2502                         goto out_free_irq_routing;
2503                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2504                                         routing.flags);
2505         out_free_irq_routing:
2506                 vfree(entries);
2507                 break;
2508         }
2509 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2510         case KVM_CREATE_DEVICE: {
2511                 struct kvm_create_device cd;
2512
2513                 r = -EFAULT;
2514                 if (copy_from_user(&cd, argp, sizeof(cd)))
2515                         goto out;
2516
2517                 r = kvm_ioctl_create_device(kvm, &cd);
2518                 if (r)
2519                         goto out;
2520
2521                 r = -EFAULT;
2522                 if (copy_to_user(argp, &cd, sizeof(cd)))
2523                         goto out;
2524
2525                 r = 0;
2526                 break;
2527         }
2528         case KVM_CHECK_EXTENSION:
2529                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2530                 break;
2531         default:
2532                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2533                 if (r == -ENOTTY)
2534                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2535         }
2536 out:
2537         return r;
2538 }
2539
2540 #ifdef CONFIG_COMPAT
2541 struct compat_kvm_dirty_log {
2542         __u32 slot;
2543         __u32 padding1;
2544         union {
2545                 compat_uptr_t dirty_bitmap; /* one bit per page */
2546                 __u64 padding2;
2547         };
2548 };
2549
2550 static long kvm_vm_compat_ioctl(struct file *filp,
2551                            unsigned int ioctl, unsigned long arg)
2552 {
2553         struct kvm *kvm = filp->private_data;
2554         int r;
2555
2556         if (kvm->mm != current->mm)
2557                 return -EIO;
2558         switch (ioctl) {
2559         case KVM_GET_DIRTY_LOG: {
2560                 struct compat_kvm_dirty_log compat_log;
2561                 struct kvm_dirty_log log;
2562
2563                 r = -EFAULT;
2564                 if (copy_from_user(&compat_log, (void __user *)arg,
2565                                    sizeof(compat_log)))
2566                         goto out;
2567                 log.slot         = compat_log.slot;
2568                 log.padding1     = compat_log.padding1;
2569                 log.padding2     = compat_log.padding2;
2570                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2571
2572                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2573                 break;
2574         }
2575         default:
2576                 r = kvm_vm_ioctl(filp, ioctl, arg);
2577         }
2578
2579 out:
2580         return r;
2581 }
2582 #endif
2583
2584 static struct file_operations kvm_vm_fops = {
2585         .release        = kvm_vm_release,
2586         .unlocked_ioctl = kvm_vm_ioctl,
2587 #ifdef CONFIG_COMPAT
2588         .compat_ioctl   = kvm_vm_compat_ioctl,
2589 #endif
2590         .llseek         = noop_llseek,
2591 };
2592
2593 static int kvm_dev_ioctl_create_vm(unsigned long type)
2594 {
2595         int r;
2596         struct kvm *kvm;
2597
2598         kvm = kvm_create_vm(type);
2599         if (IS_ERR(kvm))
2600                 return PTR_ERR(kvm);
2601 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2602         r = kvm_coalesced_mmio_init(kvm);
2603         if (r < 0) {
2604                 kvm_put_kvm(kvm);
2605                 return r;
2606         }
2607 #endif
2608         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2609         if (r < 0)
2610                 kvm_put_kvm(kvm);
2611
2612         return r;
2613 }
2614
2615 static long kvm_dev_ioctl(struct file *filp,
2616                           unsigned int ioctl, unsigned long arg)
2617 {
2618         long r = -EINVAL;
2619
2620         switch (ioctl) {
2621         case KVM_GET_API_VERSION:
2622                 r = -EINVAL;
2623                 if (arg)
2624                         goto out;
2625                 r = KVM_API_VERSION;
2626                 break;
2627         case KVM_CREATE_VM:
2628                 r = kvm_dev_ioctl_create_vm(arg);
2629                 break;
2630         case KVM_CHECK_EXTENSION:
2631                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2632                 break;
2633         case KVM_GET_VCPU_MMAP_SIZE:
2634                 r = -EINVAL;
2635                 if (arg)
2636                         goto out;
2637                 r = PAGE_SIZE;     /* struct kvm_run */
2638 #ifdef CONFIG_X86
2639                 r += PAGE_SIZE;    /* pio data page */
2640 #endif
2641 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2642                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2643 #endif
2644                 break;
2645         case KVM_TRACE_ENABLE:
2646         case KVM_TRACE_PAUSE:
2647         case KVM_TRACE_DISABLE:
2648                 r = -EOPNOTSUPP;
2649                 break;
2650         default:
2651                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2652         }
2653 out:
2654         return r;
2655 }
2656
2657 static struct file_operations kvm_chardev_ops = {
2658         .unlocked_ioctl = kvm_dev_ioctl,
2659         .compat_ioctl   = kvm_dev_ioctl,
2660         .llseek         = noop_llseek,
2661 };
2662
2663 static struct miscdevice kvm_dev = {
2664         KVM_MINOR,
2665         "kvm",
2666         &kvm_chardev_ops,
2667 };
2668
2669 static void hardware_enable_nolock(void *junk)
2670 {
2671         int cpu = raw_smp_processor_id();
2672         int r;
2673
2674         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2675                 return;
2676
2677         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2678
2679         r = kvm_arch_hardware_enable(NULL);
2680
2681         if (r) {
2682                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2683                 atomic_inc(&hardware_enable_failed);
2684                 printk(KERN_INFO "kvm: enabling virtualization on "
2685                                  "CPU%d failed\n", cpu);
2686         }
2687 }
2688
2689 static void hardware_enable(void)
2690 {
2691         raw_spin_lock(&kvm_count_lock);
2692         if (kvm_usage_count)
2693                 hardware_enable_nolock(NULL);
2694         raw_spin_unlock(&kvm_count_lock);
2695 }
2696
2697 static void hardware_disable_nolock(void *junk)
2698 {
2699         int cpu = raw_smp_processor_id();
2700
2701         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2702                 return;
2703         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2704         kvm_arch_hardware_disable(NULL);
2705 }
2706
2707 static void hardware_disable(void)
2708 {
2709         raw_spin_lock(&kvm_count_lock);
2710         if (kvm_usage_count)
2711                 hardware_disable_nolock(NULL);
2712         raw_spin_unlock(&kvm_count_lock);
2713 }
2714
2715 static void hardware_disable_all_nolock(void)
2716 {
2717         BUG_ON(!kvm_usage_count);
2718
2719         kvm_usage_count--;
2720         if (!kvm_usage_count)
2721                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2722 }
2723
2724 static void hardware_disable_all(void)
2725 {
2726         raw_spin_lock(&kvm_count_lock);
2727         hardware_disable_all_nolock();
2728         raw_spin_unlock(&kvm_count_lock);
2729 }
2730
2731 static int hardware_enable_all(void)
2732 {
2733         int r = 0;
2734
2735         raw_spin_lock(&kvm_count_lock);
2736
2737         kvm_usage_count++;
2738         if (kvm_usage_count == 1) {
2739                 atomic_set(&hardware_enable_failed, 0);
2740                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2741
2742                 if (atomic_read(&hardware_enable_failed)) {
2743                         hardware_disable_all_nolock();
2744                         r = -EBUSY;
2745                 }
2746         }
2747
2748         raw_spin_unlock(&kvm_count_lock);
2749
2750         return r;
2751 }
2752
2753 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2754                            void *v)
2755 {
2756         int cpu = (long)v;
2757
2758         val &= ~CPU_TASKS_FROZEN;
2759         switch (val) {
2760         case CPU_DYING:
2761                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2762                        cpu);
2763                 hardware_disable();
2764                 break;
2765         case CPU_STARTING:
2766                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2767                        cpu);
2768                 hardware_enable();
2769                 break;
2770         }
2771         return NOTIFY_OK;
2772 }
2773
2774 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2775                       void *v)
2776 {
2777         /*
2778          * Some (well, at least mine) BIOSes hang on reboot if
2779          * in vmx root mode.
2780          *
2781          * And Intel TXT required VMX off for all cpu when system shutdown.
2782          */
2783         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2784         kvm_rebooting = true;
2785         on_each_cpu(hardware_disable_nolock, NULL, 1);
2786         return NOTIFY_OK;
2787 }
2788
2789 static struct notifier_block kvm_reboot_notifier = {
2790         .notifier_call = kvm_reboot,
2791         .priority = 0,
2792 };
2793
2794 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2795 {
2796         int i;
2797
2798         for (i = 0; i < bus->dev_count; i++) {
2799                 struct kvm_io_device *pos = bus->range[i].dev;
2800
2801                 kvm_iodevice_destructor(pos);
2802         }
2803         kfree(bus);
2804 }
2805
2806 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2807                                  const struct kvm_io_range *r2)
2808 {
2809         if (r1->addr < r2->addr)
2810                 return -1;
2811         if (r1->addr + r1->len > r2->addr + r2->len)
2812                 return 1;
2813         return 0;
2814 }
2815
2816 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2817 {
2818         return kvm_io_bus_cmp(p1, p2);
2819 }
2820
2821 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2822                           gpa_t addr, int len)
2823 {
2824         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2825                 .addr = addr,
2826                 .len = len,
2827                 .dev = dev,
2828         };
2829
2830         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2831                 kvm_io_bus_sort_cmp, NULL);
2832
2833         return 0;
2834 }
2835
2836 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2837                              gpa_t addr, int len)
2838 {
2839         struct kvm_io_range *range, key;
2840         int off;
2841
2842         key = (struct kvm_io_range) {
2843                 .addr = addr,
2844                 .len = len,
2845         };
2846
2847         range = bsearch(&key, bus->range, bus->dev_count,
2848                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2849         if (range == NULL)
2850                 return -ENOENT;
2851
2852         off = range - bus->range;
2853
2854         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2855                 off--;
2856
2857         return off;
2858 }
2859
2860 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2861                               struct kvm_io_range *range, const void *val)
2862 {
2863         int idx;
2864
2865         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2866         if (idx < 0)
2867                 return -EOPNOTSUPP;
2868
2869         while (idx < bus->dev_count &&
2870                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2871                 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2872                                         range->len, val))
2873                         return idx;
2874                 idx++;
2875         }
2876
2877         return -EOPNOTSUPP;
2878 }
2879
2880 /* kvm_io_bus_write - called under kvm->slots_lock */
2881 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2882                      int len, const void *val)
2883 {
2884         struct kvm_io_bus *bus;
2885         struct kvm_io_range range;
2886         int r;
2887
2888         range = (struct kvm_io_range) {
2889                 .addr = addr,
2890                 .len = len,
2891         };
2892
2893         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2894         r = __kvm_io_bus_write(bus, &range, val);
2895         return r < 0 ? r : 0;
2896 }
2897
2898 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2899 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2900                             int len, const void *val, long cookie)
2901 {
2902         struct kvm_io_bus *bus;
2903         struct kvm_io_range range;
2904
2905         range = (struct kvm_io_range) {
2906                 .addr = addr,
2907                 .len = len,
2908         };
2909
2910         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2911
2912         /* First try the device referenced by cookie. */
2913         if ((cookie >= 0) && (cookie < bus->dev_count) &&
2914             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2915                 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2916                                         val))
2917                         return cookie;
2918
2919         /*
2920          * cookie contained garbage; fall back to search and return the
2921          * correct cookie value.
2922          */
2923         return __kvm_io_bus_write(bus, &range, val);
2924 }
2925
2926 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2927                              void *val)
2928 {
2929         int idx;
2930
2931         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2932         if (idx < 0)
2933                 return -EOPNOTSUPP;
2934
2935         while (idx < bus->dev_count &&
2936                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2937                 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2938                                        range->len, val))
2939                         return idx;
2940                 idx++;
2941         }
2942
2943         return -EOPNOTSUPP;
2944 }
2945 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2946
2947 /* kvm_io_bus_read - called under kvm->slots_lock */
2948 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2949                     int len, void *val)
2950 {
2951         struct kvm_io_bus *bus;
2952         struct kvm_io_range range;
2953         int r;
2954
2955         range = (struct kvm_io_range) {
2956                 .addr = addr,
2957                 .len = len,
2958         };
2959
2960         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2961         r = __kvm_io_bus_read(bus, &range, val);
2962         return r < 0 ? r : 0;
2963 }
2964
2965
2966 /* Caller must hold slots_lock. */
2967 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2968                             int len, struct kvm_io_device *dev)
2969 {
2970         struct kvm_io_bus *new_bus, *bus;
2971
2972         bus = kvm->buses[bus_idx];
2973         /* exclude ioeventfd which is limited by maximum fd */
2974         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2975                 return -ENOSPC;
2976
2977         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2978                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2979         if (!new_bus)
2980                 return -ENOMEM;
2981         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2982                sizeof(struct kvm_io_range)));
2983         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2984         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2985         synchronize_srcu_expedited(&kvm->srcu);
2986         kfree(bus);
2987
2988         return 0;
2989 }
2990
2991 /* Caller must hold slots_lock. */
2992 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2993                               struct kvm_io_device *dev)
2994 {
2995         int i, r;
2996         struct kvm_io_bus *new_bus, *bus;
2997
2998         bus = kvm->buses[bus_idx];
2999         r = -ENOENT;
3000         for (i = 0; i < bus->dev_count; i++)
3001                 if (bus->range[i].dev == dev) {
3002                         r = 0;
3003                         break;
3004                 }
3005
3006         if (r)
3007                 return r;
3008
3009         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3010                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3011         if (!new_bus)
3012                 return -ENOMEM;
3013
3014         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3015         new_bus->dev_count--;
3016         memcpy(new_bus->range + i, bus->range + i + 1,
3017                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3018
3019         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3020         synchronize_srcu_expedited(&kvm->srcu);
3021         kfree(bus);
3022         return r;
3023 }
3024
3025 static struct notifier_block kvm_cpu_notifier = {
3026         .notifier_call = kvm_cpu_hotplug,
3027 };
3028
3029 static int vm_stat_get(void *_offset, u64 *val)
3030 {
3031         unsigned offset = (long)_offset;
3032         struct kvm *kvm;
3033
3034         *val = 0;
3035         spin_lock(&kvm_lock);
3036         list_for_each_entry(kvm, &vm_list, vm_list)
3037                 *val += *(u32 *)((void *)kvm + offset);
3038         spin_unlock(&kvm_lock);
3039         return 0;
3040 }
3041
3042 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3043
3044 static int vcpu_stat_get(void *_offset, u64 *val)
3045 {
3046         unsigned offset = (long)_offset;
3047         struct kvm *kvm;
3048         struct kvm_vcpu *vcpu;
3049         int i;
3050
3051         *val = 0;
3052         spin_lock(&kvm_lock);
3053         list_for_each_entry(kvm, &vm_list, vm_list)
3054                 kvm_for_each_vcpu(i, vcpu, kvm)
3055                         *val += *(u32 *)((void *)vcpu + offset);
3056
3057         spin_unlock(&kvm_lock);
3058         return 0;
3059 }
3060
3061 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3062
3063 static const struct file_operations *stat_fops[] = {
3064         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3065         [KVM_STAT_VM]   = &vm_stat_fops,
3066 };
3067
3068 static int kvm_init_debug(void)
3069 {
3070         int r = -EEXIST;
3071         struct kvm_stats_debugfs_item *p;
3072
3073         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3074         if (kvm_debugfs_dir == NULL)
3075                 goto out;
3076
3077         for (p = debugfs_entries; p->name; ++p) {
3078                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3079                                                 (void *)(long)p->offset,
3080                                                 stat_fops[p->kind]);
3081                 if (p->dentry == NULL)
3082                         goto out_dir;
3083         }
3084
3085         return 0;
3086
3087 out_dir:
3088         debugfs_remove_recursive(kvm_debugfs_dir);
3089 out:
3090         return r;
3091 }
3092
3093 static void kvm_exit_debug(void)
3094 {
3095         struct kvm_stats_debugfs_item *p;
3096
3097         for (p = debugfs_entries; p->name; ++p)
3098                 debugfs_remove(p->dentry);
3099         debugfs_remove(kvm_debugfs_dir);
3100 }
3101
3102 static int kvm_suspend(void)
3103 {
3104         if (kvm_usage_count)
3105                 hardware_disable_nolock(NULL);
3106         return 0;
3107 }
3108
3109 static void kvm_resume(void)
3110 {
3111         if (kvm_usage_count) {
3112                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3113                 hardware_enable_nolock(NULL);
3114         }
3115 }
3116
3117 static struct syscore_ops kvm_syscore_ops = {
3118         .suspend = kvm_suspend,
3119         .resume = kvm_resume,
3120 };
3121
3122 static inline
3123 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3124 {
3125         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3126 }
3127
3128 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3129 {
3130         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3131         if (vcpu->preempted)
3132                 vcpu->preempted = false;
3133
3134         kvm_arch_sched_in(vcpu, cpu);
3135
3136         kvm_arch_vcpu_load(vcpu, cpu);
3137 }
3138
3139 static void kvm_sched_out(struct preempt_notifier *pn,
3140                           struct task_struct *next)
3141 {
3142         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3143
3144         if (current->state == TASK_RUNNING)
3145                 vcpu->preempted = true;
3146         kvm_arch_vcpu_put(vcpu);
3147 }
3148
3149 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3150                   struct module *module)
3151 {
3152         int r;
3153         int cpu;
3154
3155         r = kvm_arch_init(opaque);
3156         if (r)
3157                 goto out_fail;
3158
3159         /*
3160          * kvm_arch_init makes sure there's at most one caller
3161          * for architectures that support multiple implementations,
3162          * like intel and amd on x86.
3163          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3164          * conflicts in case kvm is already setup for another implementation.
3165          */
3166         r = kvm_irqfd_init();
3167         if (r)
3168                 goto out_irqfd;
3169
3170         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3171                 r = -ENOMEM;
3172                 goto out_free_0;
3173         }
3174
3175         r = kvm_arch_hardware_setup();
3176         if (r < 0)
3177                 goto out_free_0a;
3178
3179         for_each_online_cpu(cpu) {
3180                 smp_call_function_single(cpu,
3181                                 kvm_arch_check_processor_compat,
3182                                 &r, 1);
3183                 if (r < 0)
3184                         goto out_free_1;
3185         }
3186
3187         r = register_cpu_notifier(&kvm_cpu_notifier);
3188         if (r)
3189                 goto out_free_2;
3190         register_reboot_notifier(&kvm_reboot_notifier);
3191
3192         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3193         if (!vcpu_align)
3194                 vcpu_align = __alignof__(struct kvm_vcpu);
3195         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3196                                            0, NULL);
3197         if (!kvm_vcpu_cache) {
3198                 r = -ENOMEM;
3199                 goto out_free_3;
3200         }
3201
3202         r = kvm_async_pf_init();
3203         if (r)
3204                 goto out_free;
3205
3206         kvm_chardev_ops.owner = module;
3207         kvm_vm_fops.owner = module;
3208         kvm_vcpu_fops.owner = module;
3209
3210         r = misc_register(&kvm_dev);
3211         if (r) {
3212                 printk(KERN_ERR "kvm: misc device register failed\n");
3213                 goto out_unreg;
3214         }
3215
3216         register_syscore_ops(&kvm_syscore_ops);
3217
3218         kvm_preempt_ops.sched_in = kvm_sched_in;
3219         kvm_preempt_ops.sched_out = kvm_sched_out;
3220
3221         r = kvm_init_debug();
3222         if (r) {
3223                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3224                 goto out_undebugfs;
3225         }
3226
3227         return 0;
3228
3229 out_undebugfs:
3230         unregister_syscore_ops(&kvm_syscore_ops);
3231         misc_deregister(&kvm_dev);
3232 out_unreg:
3233         kvm_async_pf_deinit();
3234 out_free:
3235         kmem_cache_destroy(kvm_vcpu_cache);
3236 out_free_3:
3237         unregister_reboot_notifier(&kvm_reboot_notifier);
3238         unregister_cpu_notifier(&kvm_cpu_notifier);
3239 out_free_2:
3240 out_free_1:
3241         kvm_arch_hardware_unsetup();
3242 out_free_0a:
3243         free_cpumask_var(cpus_hardware_enabled);
3244 out_free_0:
3245         kvm_irqfd_exit();
3246 out_irqfd:
3247         kvm_arch_exit();
3248 out_fail:
3249         return r;
3250 }
3251 EXPORT_SYMBOL_GPL(kvm_init);
3252
3253 void kvm_exit(void)
3254 {
3255         kvm_exit_debug();
3256         misc_deregister(&kvm_dev);
3257         kmem_cache_destroy(kvm_vcpu_cache);
3258         kvm_async_pf_deinit();
3259         unregister_syscore_ops(&kvm_syscore_ops);
3260         unregister_reboot_notifier(&kvm_reboot_notifier);
3261         unregister_cpu_notifier(&kvm_cpu_notifier);
3262         on_each_cpu(hardware_disable_nolock, NULL, 1);
3263         kvm_arch_hardware_unsetup();
3264         kvm_arch_exit();
3265         kvm_irqfd_exit();
3266         free_cpumask_var(cpus_hardware_enabled);
3267 }
3268 EXPORT_SYMBOL_GPL(kvm_exit);