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
  99 static void kvm_release_pfn_dirty(pfn_t pfn);
 100 static void mark_page_dirty_in_slot(struct kvm *kvm,
 101                                     struct kvm_memory_slot *memslot, gfn_t gfn);
 102
 103 __visible bool kvm_rebooting;
 104 EXPORT_SYMBOL_GPL(kvm_rebooting);
 105
 106 static bool largepages_enabled = true;
 107
 108 bool kvm_is_mmio_pfn(pfn_t pfn)
 109 {
 110         if (pfn_valid(pfn))
 111                 return PageReserved(pfn_to_page(pfn));
 112
 113         return true;
 114 }
 115
 116 /*
 117  * Switches to specified vcpu, until a matching vcpu_put()
 118  */
 119 int vcpu_load(struct kvm_vcpu *vcpu)
 120 {
 121         int cpu;
 122
 123         if (mutex_lock_killable(&vcpu->mutex))
 124                 return -EINTR;
 125         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 126                 /* The thread running this VCPU changed. */
 127                 struct pid *oldpid = vcpu->pid;
 128                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 129                 rcu_assign_pointer(vcpu->pid, newpid);
 130                 if (oldpid)
 131                         synchronize_rcu();
 132                 put_pid(oldpid);
 133         }
 134         cpu = get_cpu();
 135         preempt_notifier_register(&vcpu->preempt_notifier);
 136         kvm_arch_vcpu_load(vcpu, cpu);
 137         put_cpu();
 138         return 0;
 139 }
 140
 141 void vcpu_put(struct kvm_vcpu *vcpu)
 142 {
 143         preempt_disable();
 144         kvm_arch_vcpu_put(vcpu);
 145         preempt_notifier_unregister(&vcpu->preempt_notifier);
 146         preempt_enable();
 147         mutex_unlock(&vcpu->mutex);
 148 }
 149
 150 static void ack_flush(void *_completed)
 151 {
 152 }
 153
 154 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 155 {
 156         int i, cpu, me;
 157         cpumask_var_t cpus;
 158         bool called = true;
 159         struct kvm_vcpu *vcpu;
 160
 161         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 162
 163         me = get_cpu();
 164         kvm_for_each_vcpu(i, vcpu, kvm) {
 165                 kvm_make_request(req, vcpu);
 166                 cpu = vcpu->cpu;
 167
 168                 /* Set ->requests bit before we read ->mode */
 169                 smp_mb();
 170
 171                 if (cpus != NULL && cpu != -1 && cpu != me &&
 172                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 173                         cpumask_set_cpu(cpu, cpus);
 174         }
 175         if (unlikely(cpus == NULL))
 176                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 177         else if (!cpumask_empty(cpus))
 178                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 179         else
 180                 called = false;
 181         put_cpu();
 182         free_cpumask_var(cpus);
 183         return called;
 184 }
 185
 186 void kvm_flush_remote_tlbs(struct kvm *kvm)
 187 {
 188         long dirty_count = kvm->tlbs_dirty;
 189
 190         smp_mb();
 191         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 192                 ++kvm->stat.remote_tlb_flush;
 193         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 194 }
 195 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 196
 197 void kvm_reload_remote_mmus(struct kvm *kvm)
 198 {
 199         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 200 }
 201
 202 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 203 {
 204         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 205 }
 206
 207 void kvm_make_scan_ioapic_request(struct kvm *kvm)
 208 {
 209         make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
 210 }
 211
 212 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 213 {
 214         struct page *page;
 215         int r;
 216
 217         mutex_init(&vcpu->mutex);
 218         vcpu->cpu = -1;
 219         vcpu->kvm = kvm;
 220         vcpu->vcpu_id = id;
 221         vcpu->pid = NULL;
 222         init_waitqueue_head(&vcpu->wq);
 223         kvm_async_pf_vcpu_init(vcpu);
 224
 225         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 226         if (!page) {
 227                 r = -ENOMEM;
 228                 goto fail;
 229         }
 230         vcpu->run = page_address(page);
 231
 232         kvm_vcpu_set_in_spin_loop(vcpu, false);
 233         kvm_vcpu_set_dy_eligible(vcpu, false);
 234         vcpu->preempted = false;
 235
 236         r = kvm_arch_vcpu_init(vcpu);
 237         if (r < 0)
 238                 goto fail_free_run;
 239         return 0;
 240
 241 fail_free_run:
 242         free_page((unsigned long)vcpu->run);
 243 fail:
 244         return r;
 245 }
 246 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 247
 248 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 249 {
 250         put_pid(vcpu->pid);
 251         kvm_arch_vcpu_uninit(vcpu);
 252         free_page((unsigned long)vcpu->run);
 253 }
 254 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 255
 256 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 257 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 258 {
 259         return container_of(mn, struct kvm, mmu_notifier);
 260 }
 261
 262 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 263                                              struct mm_struct *mm,
 264                                              unsigned long address)
 265 {
 266         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 267         int need_tlb_flush, idx;
 268
 269         /*
 270          * When ->invalidate_page runs, the linux pte has been zapped
 271          * already but the page is still allocated until
 272          * ->invalidate_page returns. So if we increase the sequence
 273          * here the kvm page fault will notice if the spte can't be
 274          * established because the page is going to be freed. If
 275          * instead the kvm page fault establishes the spte before
 276          * ->invalidate_page runs, kvm_unmap_hva will release it
 277          * before returning.
 278          *
 279          * The sequence increase only need to be seen at spin_unlock
 280          * time, and not at spin_lock time.
 281          *
 282          * Increasing the sequence after the spin_unlock would be
 283          * unsafe because the kvm page fault could then establish the
 284          * pte after kvm_unmap_hva returned, without noticing the page
 285          * is going to be freed.
 286          */
 287         idx = srcu_read_lock(&kvm->srcu);
 288         spin_lock(&kvm->mmu_lock);
 289
 290         kvm->mmu_notifier_seq++;
 291         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 292         /* we've to flush the tlb before the pages can be freed */
 293         if (need_tlb_flush)
 294                 kvm_flush_remote_tlbs(kvm);
 295
 296         spin_unlock(&kvm->mmu_lock);
 297         srcu_read_unlock(&kvm->srcu, idx);
 298 }
 299
 300 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 301                                         struct mm_struct *mm,
 302                                         unsigned long address,
 303                                         pte_t pte)
 304 {
 305         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 306         int idx;
 307
 308         idx = srcu_read_lock(&kvm->srcu);
 309         spin_lock(&kvm->mmu_lock);
 310         kvm->mmu_notifier_seq++;
 311         kvm_set_spte_hva(kvm, address, pte);
 312         spin_unlock(&kvm->mmu_lock);
 313         srcu_read_unlock(&kvm->srcu, idx);
 314 }
 315
 316 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 317                                                     struct mm_struct *mm,
 318                                                     unsigned long start,
 319                                                     unsigned long end)
 320 {
 321         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 322         int need_tlb_flush = 0, idx;
 323
 324         idx = srcu_read_lock(&kvm->srcu);
 325         spin_lock(&kvm->mmu_lock);
 326         /*
 327          * The count increase must become visible at unlock time as no
 328          * spte can be established without taking the mmu_lock and
 329          * count is also read inside the mmu_lock critical section.
 330          */
 331         kvm->mmu_notifier_count++;
 332         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 333         need_tlb_flush |= kvm->tlbs_dirty;
 334         /* we've to flush the tlb before the pages can be freed */
 335         if (need_tlb_flush)
 336                 kvm_flush_remote_tlbs(kvm);
 337
 338         spin_unlock(&kvm->mmu_lock);
 339         srcu_read_unlock(&kvm->srcu, idx);
 340 }
 341
 342 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 343                                                   struct mm_struct *mm,
 344                                                   unsigned long start,
 345                                                   unsigned long end)
 346 {
 347         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 348
 349         spin_lock(&kvm->mmu_lock);
 350         /*
 351          * This sequence increase will notify the kvm page fault that
 352          * the page that is going to be mapped in the spte could have
 353          * been freed.
 354          */
 355         kvm->mmu_notifier_seq++;
 356         smp_wmb();
 357         /*
 358          * The above sequence increase must be visible before the
 359          * below count decrease, which is ensured by the smp_wmb above
 360          * in conjunction with the smp_rmb in mmu_notifier_retry().
 361          */
 362         kvm->mmu_notifier_count--;
 363         spin_unlock(&kvm->mmu_lock);
 364
 365         BUG_ON(kvm->mmu_notifier_count < 0);
 366 }
 367
 368 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 369                                               struct mm_struct *mm,
 370                                               unsigned long address)
 371 {
 372         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 373         int young, idx;
 374
 375         idx = srcu_read_lock(&kvm->srcu);
 376         spin_lock(&kvm->mmu_lock);
 377
 378         young = kvm_age_hva(kvm, address);
 379         if (young)
 380                 kvm_flush_remote_tlbs(kvm);
 381
 382         spin_unlock(&kvm->mmu_lock);
 383         srcu_read_unlock(&kvm->srcu, idx);
 384
 385         return young;
 386 }
 387
 388 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 389                                        struct mm_struct *mm,
 390                                        unsigned long address)
 391 {
 392         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 393         int young, idx;
 394
 395         idx = srcu_read_lock(&kvm->srcu);
 396         spin_lock(&kvm->mmu_lock);
 397         young = kvm_test_age_hva(kvm, address);
 398         spin_unlock(&kvm->mmu_lock);
 399         srcu_read_unlock(&kvm->srcu, idx);
 400
 401         return young;
 402 }
 403
 404 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 405                                      struct mm_struct *mm)
 406 {
 407         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 408         int idx;
 409
 410         idx = srcu_read_lock(&kvm->srcu);
 411         kvm_arch_flush_shadow_all(kvm);
 412         srcu_read_unlock(&kvm->srcu, idx);
 413 }
 414
 415 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 416         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 417         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 418         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 419         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 420         .test_young             = kvm_mmu_notifier_test_young,
 421         .change_pte             = kvm_mmu_notifier_change_pte,
 422         .release                = kvm_mmu_notifier_release,
 423 };
 424
 425 static int kvm_init_mmu_notifier(struct kvm *kvm)
 426 {
 427         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 428         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 429 }
 430
 431 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 432
 433 static int kvm_init_mmu_notifier(struct kvm *kvm)
 434 {
 435         return 0;
 436 }
 437
 438 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 439
 440 static void kvm_init_memslots_id(struct kvm *kvm)
 441 {
 442         int i;
 443         struct kvm_memslots *slots = kvm->memslots;
 444
 445         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 446                 slots->id_to_index[i] = slots->memslots[i].id = i;
 447 }
 448
 449 static struct kvm *kvm_create_vm(unsigned long type)
 450 {
 451         int r, i;
 452         struct kvm *kvm = kvm_arch_alloc_vm();
 453
 454         if (!kvm)
 455                 return ERR_PTR(-ENOMEM);
 456
 457         r = kvm_arch_init_vm(kvm, type);
 458         if (r)
 459                 goto out_err_no_disable;
 460
 461         r = hardware_enable_all();
 462         if (r)
 463                 goto out_err_no_disable;
 464
 465 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 466         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 467 #endif
 468 #ifdef CONFIG_HAVE_KVM_IRQFD
 469         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 470 #endif
 471
 472         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 473
 474         r = -ENOMEM;
 475         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 476         if (!kvm->memslots)
 477                 goto out_err_no_srcu;
 478
 479         /*
 480          * Init kvm generation close to the maximum to easily test the
 481          * code of handling generation number wrap-around.
 482          */
 483         kvm->memslots->generation = -150;
 484
 485         kvm_init_memslots_id(kvm);
 486         if (init_srcu_struct(&kvm->srcu))
 487                 goto out_err_no_srcu;
 488         if (init_srcu_struct(&kvm->irq_srcu))
 489                 goto out_err_no_irq_srcu;
 490         for (i = 0; i < KVM_NR_BUSES; i++) {
 491                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 492                                         GFP_KERNEL);
 493                 if (!kvm->buses[i])
 494                         goto out_err;
 495         }
 496
 497         spin_lock_init(&kvm->mmu_lock);
 498         kvm->mm = current->mm;
 499         atomic_inc(&kvm->mm->mm_count);
 500         kvm_eventfd_init(kvm);
 501         mutex_init(&kvm->lock);
 502         mutex_init(&kvm->irq_lock);
 503         mutex_init(&kvm->slots_lock);
 504         atomic_set(&kvm->users_count, 1);
 505         INIT_LIST_HEAD(&kvm->devices);
 506
 507         r = kvm_init_mmu_notifier(kvm);
 508         if (r)
 509                 goto out_err;
 510
 511         spin_lock(&kvm_lock);
 512         list_add(&kvm->vm_list, &vm_list);
 513         spin_unlock(&kvm_lock);
 514
 515         return kvm;
 516
 517 out_err:
 518         cleanup_srcu_struct(&kvm->irq_srcu);
 519 out_err_no_irq_srcu:
 520         cleanup_srcu_struct(&kvm->srcu);
 521 out_err_no_srcu:
 522         hardware_disable_all();
 523 out_err_no_disable:
 524         for (i = 0; i < KVM_NR_BUSES; i++)
 525                 kfree(kvm->buses[i]);
 526         kfree(kvm->memslots);
 527         kvm_arch_free_vm(kvm);
 528         return ERR_PTR(r);
 529 }
 530
 531 /*
 532  * Avoid using vmalloc for a small buffer.
 533  * Should not be used when the size is statically known.
 534  */
 535 void *kvm_kvzalloc(unsigned long size)
 536 {
 537         if (size > PAGE_SIZE)
 538                 return vzalloc(size);
 539         else
 540                 return kzalloc(size, GFP_KERNEL);
 541 }
 542
 543 void kvm_kvfree(const void *addr)
 544 {
 545         if (is_vmalloc_addr(addr))
 546                 vfree(addr);
 547         else
 548                 kfree(addr);
 549 }
 550
 551 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 552 {
 553         if (!memslot->dirty_bitmap)
 554                 return;
 555
 556         kvm_kvfree(memslot->dirty_bitmap);
 557         memslot->dirty_bitmap = NULL;
 558 }
 559
 560 /*
 561  * Free any memory in @free but not in @dont.
 562  */
 563 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
 564                                   struct kvm_memory_slot *dont)
 565 {
 566         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 567                 kvm_destroy_dirty_bitmap(free);
 568
 569         kvm_arch_free_memslot(kvm, free, dont);
 570
 571         free->npages = 0;
 572 }
 573
 574 static void kvm_free_physmem(struct kvm *kvm)
 575 {
 576         struct kvm_memslots *slots = kvm->memslots;
 577         struct kvm_memory_slot *memslot;
 578
 579         kvm_for_each_memslot(memslot, slots)
 580                 kvm_free_physmem_slot(kvm, memslot, NULL);
 581
 582         kfree(kvm->memslots);
 583 }
 584
 585 static void kvm_destroy_devices(struct kvm *kvm)
 586 {
 587         struct list_head *node, *tmp;
 588
 589         list_for_each_safe(node, tmp, &kvm->devices) {
 590                 struct kvm_device *dev =
 591                         list_entry(node, struct kvm_device, vm_node);
 592
 593                 list_del(node);
 594                 dev->ops->destroy(dev);
 595         }
 596 }
 597
 598 static void kvm_destroy_vm(struct kvm *kvm)
 599 {
 600         int i;
 601         struct mm_struct *mm = kvm->mm;
 602
 603         kvm_arch_sync_events(kvm);
 604         spin_lock(&kvm_lock);
 605         list_del(&kvm->vm_list);
 606         spin_unlock(&kvm_lock);
 607         kvm_free_irq_routing(kvm);
 608         for (i = 0; i < KVM_NR_BUSES; i++)
 609                 kvm_io_bus_destroy(kvm->buses[i]);
 610         kvm_coalesced_mmio_free(kvm);
 611 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 612         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 613 #else
 614         kvm_arch_flush_shadow_all(kvm);
 615 #endif
 616         kvm_arch_destroy_vm(kvm);
 617         kvm_destroy_devices(kvm);
 618         kvm_free_physmem(kvm);
 619         cleanup_srcu_struct(&kvm->irq_srcu);
 620         cleanup_srcu_struct(&kvm->srcu);
 621         kvm_arch_free_vm(kvm);
 622         hardware_disable_all();
 623         mmdrop(mm);
 624 }
 625
 626 void kvm_get_kvm(struct kvm *kvm)
 627 {
 628         atomic_inc(&kvm->users_count);
 629 }
 630 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 631
 632 void kvm_put_kvm(struct kvm *kvm)
 633 {
 634         if (atomic_dec_and_test(&kvm->users_count))
 635                 kvm_destroy_vm(kvm);
 636 }
 637 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 638
 639
 640 static int kvm_vm_release(struct inode *inode, struct file *filp)
 641 {
 642         struct kvm *kvm = filp->private_data;
 643
 644         kvm_irqfd_release(kvm);
 645
 646         kvm_put_kvm(kvm);
 647         return 0;
 648 }
 649
 650 /*
 651  * Allocation size is twice as large as the actual dirty bitmap size.
 652  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 653  */
 654 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 655 {
 656         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 657
 658         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 659         if (!memslot->dirty_bitmap)
 660                 return -ENOMEM;
 661
 662         return 0;
 663 }
 664
 665 static int cmp_memslot(const void *slot1, const void *slot2)
 666 {
 667         struct kvm_memory_slot *s1, *s2;
 668
 669         s1 = (struct kvm_memory_slot *)slot1;
 670         s2 = (struct kvm_memory_slot *)slot2;
 671
 672         if (s1->npages < s2->npages)
 673                 return 1;
 674         if (s1->npages > s2->npages)
 675                 return -1;
 676
 677         return 0;
 678 }
 679
 680 /*
 681  * Sort the memslots base on its size, so the larger slots
 682  * will get better fit.
 683  */
 684 static void sort_memslots(struct kvm_memslots *slots)
 685 {
 686         int i;
 687
 688         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 689               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 690
 691         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 692                 slots->id_to_index[slots->memslots[i].id] = i;
 693 }
 694
 695 static void update_memslots(struct kvm_memslots *slots,
 696                             struct kvm_memory_slot *new)
 697 {
 698         if (new) {
 699                 int id = new->id;
 700                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 701                 unsigned long npages = old->npages;
 702
 703                 *old = *new;
 704                 if (new->npages != npages)
 705                         sort_memslots(slots);
 706         }
 707 }
 708
 709 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 710 {
 711         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 712
 713 #ifdef __KVM_HAVE_READONLY_MEM
 714         valid_flags |= KVM_MEM_READONLY;
 715 #endif
 716
 717         if (mem->flags & ~valid_flags)
 718                 return -EINVAL;
 719
 720         return 0;
 721 }
 722
 723 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 724                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 725 {
 726         struct kvm_memslots *old_memslots = kvm->memslots;
 727
 728         /*
 729          * Set the low bit in the generation, which disables SPTE caching
 730          * until the end of synchronize_srcu_expedited.
 731          */
 732         WARN_ON(old_memslots->generation & 1);
 733         slots->generation = old_memslots->generation + 1;
 734
 735         update_memslots(slots, new);
 736         rcu_assign_pointer(kvm->memslots, slots);
 737         synchronize_srcu_expedited(&kvm->srcu);
 738
 739         /*
 740          * Increment the new memslot generation a second time. This prevents
 741          * vm exits that race with memslot updates from caching a memslot
 742          * generation that will (potentially) be valid forever.
 743          */
 744         slots->generation++;
 745
 746         kvm_arch_memslots_updated(kvm);
 747
 748         return old_memslots;
 749 }
 750
 751 /*
 752  * Allocate some memory and give it an address in the guest physical address
 753  * space.
 754  *
 755  * Discontiguous memory is allowed, mostly for framebuffers.
 756  *
 757  * Must be called holding mmap_sem for write.
 758  */
 759 int __kvm_set_memory_region(struct kvm *kvm,
 760                             struct kvm_userspace_memory_region *mem)
 761 {
 762         int r;
 763         gfn_t base_gfn;
 764         unsigned long npages;
 765         struct kvm_memory_slot *slot;
 766         struct kvm_memory_slot old, new;
 767         struct kvm_memslots *slots = NULL, *old_memslots;
 768         enum kvm_mr_change change;
 769
 770         r = check_memory_region_flags(mem);
 771         if (r)
 772                 goto out;
 773
 774         r = -EINVAL;
 775         /* General sanity checks */
 776         if (mem->memory_size & (PAGE_SIZE - 1))
 777                 goto out;
 778         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 779                 goto out;
 780         /* We can read the guest memory with __xxx_user() later on. */
 781         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 782             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 783              !access_ok(VERIFY_WRITE,
 784                         (void __user *)(unsigned long)mem->userspace_addr,
 785                         mem->memory_size)))
 786                 goto out;
 787         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 788                 goto out;
 789         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 790                 goto out;
 791
 792         slot = id_to_memslot(kvm->memslots, mem->slot);
 793         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 794         npages = mem->memory_size >> PAGE_SHIFT;
 795
 796         r = -EINVAL;
 797         if (npages > KVM_MEM_MAX_NR_PAGES)
 798                 goto out;
 799
 800         if (!npages)
 801                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 802
 803         new = old = *slot;
 804
 805         new.id = mem->slot;
 806         new.base_gfn = base_gfn;
 807         new.npages = npages;
 808         new.flags = mem->flags;
 809
 810         r = -EINVAL;
 811         if (npages) {
 812                 if (!old.npages)
 813                         change = KVM_MR_CREATE;
 814                 else { /* Modify an existing slot. */
 815                         if ((mem->userspace_addr != old.userspace_addr) ||
 816                             (npages != old.npages) ||
 817                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 818                                 goto out;
 819
 820                         if (base_gfn != old.base_gfn)
 821                                 change = KVM_MR_MOVE;
 822                         else if (new.flags != old.flags)
 823                                 change = KVM_MR_FLAGS_ONLY;
 824                         else { /* Nothing to change. */
 825                                 r = 0;
 826                                 goto out;
 827                         }
 828                 }
 829         } else if (old.npages) {
 830                 change = KVM_MR_DELETE;
 831         } else /* Modify a non-existent slot: disallowed. */
 832                 goto out;
 833
 834         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 835                 /* Check for overlaps */
 836                 r = -EEXIST;
 837                 kvm_for_each_memslot(slot, kvm->memslots) {
 838                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 839                             (slot->id == mem->slot))
 840                                 continue;
 841                         if (!((base_gfn + npages <= slot->base_gfn) ||
 842                               (base_gfn >= slot->base_gfn + slot->npages)))
 843                                 goto out;
 844                 }
 845         }
 846
 847         /* Free page dirty bitmap if unneeded */
 848         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 849                 new.dirty_bitmap = NULL;
 850
 851         r = -ENOMEM;
 852         if (change == KVM_MR_CREATE) {
 853                 new.userspace_addr = mem->userspace_addr;
 854
 855                 if (kvm_arch_create_memslot(kvm, &new, npages))
 856                         goto out_free;
 857         }
 858
 859         /* Allocate page dirty bitmap if needed */
 860         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 861                 if (kvm_create_dirty_bitmap(&new) < 0)
 862                         goto out_free;
 863         }
 864
 865         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 866                 r = -ENOMEM;
 867                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 868                                 GFP_KERNEL);
 869                 if (!slots)
 870                         goto out_free;
 871                 slot = id_to_memslot(slots, mem->slot);
 872                 slot->flags |= KVM_MEMSLOT_INVALID;
 873
 874                 old_memslots = install_new_memslots(kvm, slots, NULL);
 875
 876                 /* slot was deleted or moved, clear iommu mapping */
 877                 kvm_iommu_unmap_pages(kvm, &old);
 878                 /* From this point no new shadow pages pointing to a deleted,
 879                  * or moved, memslot will be created.
 880                  *
 881                  * validation of sp->gfn happens in:
 882                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 883                  *      - kvm_is_visible_gfn (mmu_check_roots)
 884                  */
 885                 kvm_arch_flush_shadow_memslot(kvm, slot);
 886                 slots = old_memslots;
 887         }
 888
 889         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 890         if (r)
 891                 goto out_slots;
 892
 893         r = -ENOMEM;
 894         /*
 895          * We can re-use the old_memslots from above, the only difference
 896          * from the currently installed memslots is the invalid flag.  This
 897          * will get overwritten by update_memslots anyway.
 898          */
 899         if (!slots) {
 900                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 901                                 GFP_KERNEL);
 902                 if (!slots)
 903                         goto out_free;
 904         }
 905
 906         /* actual memory is freed via old in kvm_free_physmem_slot below */
 907         if (change == KVM_MR_DELETE) {
 908                 new.dirty_bitmap = NULL;
 909                 memset(&new.arch, 0, sizeof(new.arch));
 910         }
 911
 912         old_memslots = install_new_memslots(kvm, slots, &new);
 913
 914         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 915
 916         kvm_free_physmem_slot(kvm, &old, &new);
 917         kfree(old_memslots);
 918
 919         /*
 920          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 921          * un-mapped and re-mapped if their base changes.  Since base change
 922          * unmapping is handled above with slot deletion, mapping alone is
 923          * needed here.  Anything else the iommu might care about for existing
 924          * slots (size changes, userspace addr changes and read-only flag
 925          * changes) is disallowed above, so any other attribute changes getting
 926          * here can be skipped.
 927          */
 928         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 929                 r = kvm_iommu_map_pages(kvm, &new);
 930                 return r;
 931         }
 932
 933         return 0;
 934
 935 out_slots:
 936         kfree(slots);
 937 out_free:
 938         kvm_free_physmem_slot(kvm, &new, &old);
 939 out:
 940         return r;
 941 }
 942 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 943
 944 int kvm_set_memory_region(struct kvm *kvm,
 945                           struct kvm_userspace_memory_region *mem)
 946 {
 947         int r;
 948
 949         mutex_lock(&kvm->slots_lock);
 950         r = __kvm_set_memory_region(kvm, mem);
 951         mutex_unlock(&kvm->slots_lock);
 952         return r;
 953 }
 954 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 955
 956 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 957                                           struct kvm_userspace_memory_region *mem)
 958 {
 959         if (mem->slot >= KVM_USER_MEM_SLOTS)
 960                 return -EINVAL;
 961         return kvm_set_memory_region(kvm, mem);
 962 }
 963
 964 int kvm_get_dirty_log(struct kvm *kvm,
 965                         struct kvm_dirty_log *log, int *is_dirty)
 966 {
 967         struct kvm_memory_slot *memslot;
 968         int r, i;
 969         unsigned long n;
 970         unsigned long any = 0;
 971
 972         r = -EINVAL;
 973         if (log->slot >= KVM_USER_MEM_SLOTS)
 974                 goto out;
 975
 976         memslot = id_to_memslot(kvm->memslots, log->slot);
 977         r = -ENOENT;
 978         if (!memslot->dirty_bitmap)
 979                 goto out;
 980
 981         n = kvm_dirty_bitmap_bytes(memslot);
 982
 983         for (i = 0; !any && i < n/sizeof(long); ++i)
 984                 any = memslot->dirty_bitmap[i];
 985
 986         r = -EFAULT;
 987         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 988                 goto out;
 989
 990         if (any)
 991                 *is_dirty = 1;
 992
 993         r = 0;
 994 out:
 995         return r;
 996 }
 997 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
 998
 999 bool kvm_largepages_enabled(void)
1000 {
1001         return largepages_enabled;
1002 }
1003
1004 void kvm_disable_largepages(void)
1005 {
1006         largepages_enabled = false;
1007 }
1008 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1009
1010 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1011 {
1012         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1013 }
1014 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1015
1016 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1017 {
1018         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1019
1020         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1021               memslot->flags & KVM_MEMSLOT_INVALID)
1022                 return 0;
1023
1024         return 1;
1025 }
1026 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1027
1028 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1029 {
1030         struct vm_area_struct *vma;
1031         unsigned long addr, size;
1032
1033         size = PAGE_SIZE;
1034
1035         addr = gfn_to_hva(kvm, gfn);
1036         if (kvm_is_error_hva(addr))
1037                 return PAGE_SIZE;
1038
1039         down_read(&current->mm->mmap_sem);
1040         vma = find_vma(current->mm, addr);
1041         if (!vma)
1042                 goto out;
1043
1044         size = vma_kernel_pagesize(vma);
1045
1046 out:
1047         up_read(&current->mm->mmap_sem);
1048
1049         return size;
1050 }
1051
1052 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1053 {
1054         return slot->flags & KVM_MEM_READONLY;
1055 }
1056
1057 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1058                                        gfn_t *nr_pages, bool write)
1059 {
1060         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1061                 return KVM_HVA_ERR_BAD;
1062
1063         if (memslot_is_readonly(slot) && write)
1064                 return KVM_HVA_ERR_RO_BAD;
1065
1066         if (nr_pages)
1067                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1068
1069         return __gfn_to_hva_memslot(slot, gfn);
1070 }
1071
1072 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1073                                      gfn_t *nr_pages)
1074 {
1075         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1076 }
1077
1078 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1079                                         gfn_t gfn)
1080 {
1081         return gfn_to_hva_many(slot, gfn, NULL);
1082 }
1083 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1084
1085 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1086 {
1087         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1088 }
1089 EXPORT_SYMBOL_GPL(gfn_to_hva);
1090
1091 /*
1092  * If writable is set to false, the hva returned by this function is only
1093  * allowed to be read.
1094  */
1095 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1096 {
1097         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1098         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1099
1100         if (!kvm_is_error_hva(hva) && writable)
1101                 *writable = !memslot_is_readonly(slot);
1102
1103         return hva;
1104 }
1105
1106 static int kvm_read_hva(void *data, void __user *hva, int len)
1107 {
1108         return __copy_from_user(data, hva, len);
1109 }
1110
1111 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1112 {
1113         return __copy_from_user_inatomic(data, hva, len);
1114 }
1115
1116 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1117         unsigned long start, int write, struct page **page)
1118 {
1119         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1120
1121         if (write)
1122                 flags |= FOLL_WRITE;
1123
1124         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1125 }
1126
1127 static inline int check_user_page_hwpoison(unsigned long addr)
1128 {
1129         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1130
1131         rc = __get_user_pages(current, current->mm, addr, 1,
1132                               flags, NULL, NULL, NULL);
1133         return rc == -EHWPOISON;
1134 }
1135
1136 /*
1137  * The atomic path to get the writable pfn which will be stored in @pfn,
1138  * true indicates success, otherwise false is returned.
1139  */
1140 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1141                             bool write_fault, bool *writable, pfn_t *pfn)
1142 {
1143         struct page *page[1];
1144         int npages;
1145
1146         if (!(async || atomic))
1147                 return false;
1148
1149         /*
1150          * Fast pin a writable pfn only if it is a write fault request
1151          * or the caller allows to map a writable pfn for a read fault
1152          * request.
1153          */
1154         if (!(write_fault || writable))
1155                 return false;
1156
1157         npages = __get_user_pages_fast(addr, 1, 1, page);
1158         if (npages == 1) {
1159                 *pfn = page_to_pfn(page[0]);
1160
1161                 if (writable)
1162                         *writable = true;
1163                 return true;
1164         }
1165
1166         return false;
1167 }
1168
1169 /*
1170  * The slow path to get the pfn of the specified host virtual address,
1171  * 1 indicates success, -errno is returned if error is detected.
1172  */
1173 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1174                            bool *writable, pfn_t *pfn)
1175 {
1176         struct page *page[1];
1177         int npages = 0;
1178
1179         might_sleep();
1180
1181         if (writable)
1182                 *writable = write_fault;
1183
1184         if (async) {
1185                 down_read(&current->mm->mmap_sem);
1186                 npages = get_user_page_nowait(current, current->mm,
1187                                               addr, write_fault, page);
1188                 up_read(&current->mm->mmap_sem);
1189         } else
1190                 npages = get_user_pages_fast(addr, 1, write_fault,
1191                                              page);
1192         if (npages != 1)
1193                 return npages;
1194
1195         /* map read fault as writable if possible */
1196         if (unlikely(!write_fault) && writable) {
1197                 struct page *wpage[1];
1198
1199                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1200                 if (npages == 1) {
1201                         *writable = true;
1202                         put_page(page[0]);
1203                         page[0] = wpage[0];
1204                 }
1205
1206                 npages = 1;
1207         }
1208         *pfn = page_to_pfn(page[0]);
1209         return npages;
1210 }
1211
1212 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1213 {
1214         if (unlikely(!(vma->vm_flags & VM_READ)))
1215                 return false;
1216
1217         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1218                 return false;
1219
1220         return true;
1221 }
1222
1223 /*
1224  * Pin guest page in memory and return its pfn.
1225  * @addr: host virtual address which maps memory to the guest
1226  * @atomic: whether this function can sleep
1227  * @async: whether this function need to wait IO complete if the
1228  *         host page is not in the memory
1229  * @write_fault: whether we should get a writable host page
1230  * @writable: whether it allows to map a writable host page for !@write_fault
1231  *
1232  * The function will map a writable host page for these two cases:
1233  * 1): @write_fault = true
1234  * 2): @write_fault = false && @writable, @writable will tell the caller
1235  *     whether the mapping is writable.
1236  */
1237 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1238                         bool write_fault, bool *writable)
1239 {
1240         struct vm_area_struct *vma;
1241         pfn_t pfn = 0;
1242         int npages;
1243
1244         /* we can do it either atomically or asynchronously, not both */
1245         BUG_ON(atomic && async);
1246
1247         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1248                 return pfn;
1249
1250         if (atomic)
1251                 return KVM_PFN_ERR_FAULT;
1252
1253         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1254         if (npages == 1)
1255                 return pfn;
1256
1257         down_read(&current->mm->mmap_sem);
1258         if (npages == -EHWPOISON ||
1259               (!async && check_user_page_hwpoison(addr))) {
1260                 pfn = KVM_PFN_ERR_HWPOISON;
1261                 goto exit;
1262         }
1263
1264         vma = find_vma_intersection(current->mm, addr, addr + 1);
1265
1266         if (vma == NULL)
1267                 pfn = KVM_PFN_ERR_FAULT;
1268         else if ((vma->vm_flags & VM_PFNMAP)) {
1269                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1270                         vma->vm_pgoff;
1271                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1272         } else {
1273                 if (async && vma_is_valid(vma, write_fault))
1274                         *async = true;
1275                 pfn = KVM_PFN_ERR_FAULT;
1276         }
1277 exit:
1278         up_read(&current->mm->mmap_sem);
1279         return pfn;
1280 }
1281
1282 static pfn_t
1283 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1284                      bool *async, bool write_fault, bool *writable)
1285 {
1286         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1287
1288         if (addr == KVM_HVA_ERR_RO_BAD)
1289                 return KVM_PFN_ERR_RO_FAULT;
1290
1291         if (kvm_is_error_hva(addr))
1292                 return KVM_PFN_NOSLOT;
1293
1294         /* Do not map writable pfn in the readonly memslot. */
1295         if (writable && memslot_is_readonly(slot)) {
1296                 *writable = false;
1297                 writable = NULL;
1298         }
1299
1300         return hva_to_pfn(addr, atomic, async, write_fault,
1301                           writable);
1302 }
1303
1304 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1305                           bool write_fault, bool *writable)
1306 {
1307         struct kvm_memory_slot *slot;
1308
1309         if (async)
1310                 *async = false;
1311
1312         slot = gfn_to_memslot(kvm, gfn);
1313
1314         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1315                                     writable);
1316 }
1317
1318 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1319 {
1320         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1321 }
1322 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1323
1324 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1325                        bool write_fault, bool *writable)
1326 {
1327         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1328 }
1329 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1330
1331 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1332 {
1333         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1334 }
1335 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1336
1337 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1338                       bool *writable)
1339 {
1340         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1341 }
1342 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1343
1344 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1345 {
1346         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1347 }
1348
1349 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1350 {
1351         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1352 }
1353 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1354
1355 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1356                                                                   int nr_pages)
1357 {
1358         unsigned long addr;
1359         gfn_t entry;
1360
1361         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1362         if (kvm_is_error_hva(addr))
1363                 return -1;
1364
1365         if (entry < nr_pages)
1366                 return 0;
1367
1368         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1369 }
1370 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1371
1372 static struct page *kvm_pfn_to_page(pfn_t pfn)
1373 {
1374         if (is_error_noslot_pfn(pfn))
1375                 return KVM_ERR_PTR_BAD_PAGE;
1376
1377         if (kvm_is_mmio_pfn(pfn)) {
1378                 WARN_ON(1);
1379                 return KVM_ERR_PTR_BAD_PAGE;
1380         }
1381
1382         return pfn_to_page(pfn);
1383 }
1384
1385 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1386 {
1387         pfn_t pfn;
1388
1389         pfn = gfn_to_pfn(kvm, gfn);
1390
1391         return kvm_pfn_to_page(pfn);
1392 }
1393
1394 EXPORT_SYMBOL_GPL(gfn_to_page);
1395
1396 void kvm_release_page_clean(struct page *page)
1397 {
1398         WARN_ON(is_error_page(page));
1399
1400         kvm_release_pfn_clean(page_to_pfn(page));
1401 }
1402 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1403
1404 void kvm_release_pfn_clean(pfn_t pfn)
1405 {
1406         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1407                 put_page(pfn_to_page(pfn));
1408 }
1409 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1410
1411 void kvm_release_page_dirty(struct page *page)
1412 {
1413         WARN_ON(is_error_page(page));
1414
1415         kvm_release_pfn_dirty(page_to_pfn(page));
1416 }
1417 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1418
1419 static void kvm_release_pfn_dirty(pfn_t pfn)
1420 {
1421         kvm_set_pfn_dirty(pfn);
1422         kvm_release_pfn_clean(pfn);
1423 }
1424
1425 void kvm_set_pfn_dirty(pfn_t pfn)
1426 {
1427         if (!kvm_is_mmio_pfn(pfn)) {
1428                 struct page *page = pfn_to_page(pfn);
1429                 if (!PageReserved(page))
1430                         SetPageDirty(page);
1431         }
1432 }
1433 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1434
1435 void kvm_set_pfn_accessed(pfn_t pfn)
1436 {
1437         if (!kvm_is_mmio_pfn(pfn))
1438                 mark_page_accessed(pfn_to_page(pfn));
1439 }
1440 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1441
1442 void kvm_get_pfn(pfn_t pfn)
1443 {
1444         if (!kvm_is_mmio_pfn(pfn))
1445                 get_page(pfn_to_page(pfn));
1446 }
1447 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1448
1449 static int next_segment(unsigned long len, int offset)
1450 {
1451         if (len > PAGE_SIZE - offset)
1452                 return PAGE_SIZE - offset;
1453         else
1454                 return len;
1455 }
1456
1457 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1458                         int len)
1459 {
1460         int r;
1461         unsigned long addr;
1462
1463         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1464         if (kvm_is_error_hva(addr))
1465                 return -EFAULT;
1466         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1467         if (r)
1468                 return -EFAULT;
1469         return 0;
1470 }
1471 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1472
1473 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1474 {
1475         gfn_t gfn = gpa >> PAGE_SHIFT;
1476         int seg;
1477         int offset = offset_in_page(gpa);
1478         int ret;
1479
1480         while ((seg = next_segment(len, offset)) != 0) {
1481                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1482                 if (ret < 0)
1483                         return ret;
1484                 offset = 0;
1485                 len -= seg;
1486                 data += seg;
1487                 ++gfn;
1488         }
1489         return 0;
1490 }
1491 EXPORT_SYMBOL_GPL(kvm_read_guest);
1492
1493 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1494                           unsigned long len)
1495 {
1496         int r;
1497         unsigned long addr;
1498         gfn_t gfn = gpa >> PAGE_SHIFT;
1499         int offset = offset_in_page(gpa);
1500
1501         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1502         if (kvm_is_error_hva(addr))
1503                 return -EFAULT;
1504         pagefault_disable();
1505         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1506         pagefault_enable();
1507         if (r)
1508                 return -EFAULT;
1509         return 0;
1510 }
1511 EXPORT_SYMBOL(kvm_read_guest_atomic);
1512
1513 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1514                          int offset, int len)
1515 {
1516         int r;
1517         unsigned long addr;
1518
1519         addr = gfn_to_hva(kvm, gfn);
1520         if (kvm_is_error_hva(addr))
1521                 return -EFAULT;
1522         r = __copy_to_user((void __user *)addr + offset, data, len);
1523         if (r)
1524                 return -EFAULT;
1525         mark_page_dirty(kvm, gfn);
1526         return 0;
1527 }
1528 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1529
1530 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1531                     unsigned long len)
1532 {
1533         gfn_t gfn = gpa >> PAGE_SHIFT;
1534         int seg;
1535         int offset = offset_in_page(gpa);
1536         int ret;
1537
1538         while ((seg = next_segment(len, offset)) != 0) {
1539                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1540                 if (ret < 0)
1541                         return ret;
1542                 offset = 0;
1543                 len -= seg;
1544                 data += seg;
1545                 ++gfn;
1546         }
1547         return 0;
1548 }
1549
1550 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1551                               gpa_t gpa, unsigned long len)
1552 {
1553         struct kvm_memslots *slots = kvm_memslots(kvm);
1554         int offset = offset_in_page(gpa);
1555         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1556         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1557         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1558         gfn_t nr_pages_avail;
1559
1560         ghc->gpa = gpa;
1561         ghc->generation = slots->generation;
1562         ghc->len = len;
1563         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1564         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1565         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1566                 ghc->hva += offset;
1567         } else {
1568                 /*
1569                  * If the requested region crosses two memslots, we still
1570                  * verify that the entire region is valid here.
1571                  */
1572                 while (start_gfn <= end_gfn) {
1573                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1574                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1575                                                    &nr_pages_avail);
1576                         if (kvm_is_error_hva(ghc->hva))
1577                                 return -EFAULT;
1578                         start_gfn += nr_pages_avail;
1579                 }
1580                 /* Use the slow path for cross page reads and writes. */
1581                 ghc->memslot = NULL;
1582         }
1583         return 0;
1584 }
1585 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1586
1587 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1588                            void *data, unsigned long len)
1589 {
1590         struct kvm_memslots *slots = kvm_memslots(kvm);
1591         int r;
1592
1593         BUG_ON(len > ghc->len);
1594
1595         if (slots->generation != ghc->generation)
1596                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1597
1598         if (unlikely(!ghc->memslot))
1599                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1600
1601         if (kvm_is_error_hva(ghc->hva))
1602                 return -EFAULT;
1603
1604         r = __copy_to_user((void __user *)ghc->hva, data, len);
1605         if (r)
1606                 return -EFAULT;
1607         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1608
1609         return 0;
1610 }
1611 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1612
1613 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1614                            void *data, unsigned long len)
1615 {
1616         struct kvm_memslots *slots = kvm_memslots(kvm);
1617         int r;
1618
1619         BUG_ON(len > ghc->len);
1620
1621         if (slots->generation != ghc->generation)
1622                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1623
1624         if (unlikely(!ghc->memslot))
1625                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1626
1627         if (kvm_is_error_hva(ghc->hva))
1628                 return -EFAULT;
1629
1630         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1631         if (r)
1632                 return -EFAULT;
1633
1634         return 0;
1635 }
1636 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1637
1638 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1639 {
1640         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1641
1642         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1643 }
1644 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1645
1646 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1647 {
1648         gfn_t gfn = gpa >> PAGE_SHIFT;
1649         int seg;
1650         int offset = offset_in_page(gpa);
1651         int ret;
1652
1653         while ((seg = next_segment(len, offset)) != 0) {
1654                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1655                 if (ret < 0)
1656                         return ret;
1657                 offset = 0;
1658                 len -= seg;
1659                 ++gfn;
1660         }
1661         return 0;
1662 }
1663 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1664
1665 static void mark_page_dirty_in_slot(struct kvm *kvm,
1666                                     struct kvm_memory_slot *memslot,
1667                                     gfn_t gfn)
1668 {
1669         if (memslot && memslot->dirty_bitmap) {
1670                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1671
1672                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1673         }
1674 }
1675
1676 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1677 {
1678         struct kvm_memory_slot *memslot;
1679
1680         memslot = gfn_to_memslot(kvm, gfn);
1681         mark_page_dirty_in_slot(kvm, memslot, gfn);
1682 }
1683 EXPORT_SYMBOL_GPL(mark_page_dirty);
1684
1685 /*
1686  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1687  */
1688 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1689 {
1690         DEFINE_WAIT(wait);
1691
1692         for (;;) {
1693                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1694
1695                 if (kvm_arch_vcpu_runnable(vcpu)) {
1696                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1697                         break;
1698                 }
1699                 if (kvm_cpu_has_pending_timer(vcpu))
1700                         break;
1701                 if (signal_pending(current))
1702                         break;
1703
1704                 schedule();
1705         }
1706
1707         finish_wait(&vcpu->wq, &wait);
1708 }
1709 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1710
1711 #ifndef CONFIG_S390
1712 /*
1713  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1714  */
1715 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1716 {
1717         int me;
1718         int cpu = vcpu->cpu;
1719         wait_queue_head_t *wqp;
1720
1721         wqp = kvm_arch_vcpu_wq(vcpu);
1722         if (waitqueue_active(wqp)) {
1723                 wake_up_interruptible(wqp);
1724                 ++vcpu->stat.halt_wakeup;
1725         }
1726
1727         me = get_cpu();
1728         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1729                 if (kvm_arch_vcpu_should_kick(vcpu))
1730                         smp_send_reschedule(cpu);
1731         put_cpu();
1732 }
1733 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1734 #endif /* !CONFIG_S390 */
1735
1736 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1737 {
1738         struct pid *pid;
1739         struct task_struct *task = NULL;
1740         int ret = 0;
1741
1742         rcu_read_lock();
1743         pid = rcu_dereference(target->pid);
1744         if (pid)
1745                 task = get_pid_task(target->pid, PIDTYPE_PID);
1746         rcu_read_unlock();
1747         if (!task)
1748                 return ret;
1749         if (task->flags & PF_VCPU) {
1750                 put_task_struct(task);
1751                 return ret;
1752         }
1753         ret = yield_to(task, 1);
1754         put_task_struct(task);
1755
1756         return ret;
1757 }
1758 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1759
1760 /*
1761  * Helper that checks whether a VCPU is eligible for directed yield.
1762  * Most eligible candidate to yield is decided by following heuristics:
1763  *
1764  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1765  *  (preempted lock holder), indicated by @in_spin_loop.
1766  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1767  *
1768  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1769  *  chance last time (mostly it has become eligible now since we have probably
1770  *  yielded to lockholder in last iteration. This is done by toggling
1771  *  @dy_eligible each time a VCPU checked for eligibility.)
1772  *
1773  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1774  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1775  *  burning. Giving priority for a potential lock-holder increases lock
1776  *  progress.
1777  *
1778  *  Since algorithm is based on heuristics, accessing another VCPU data without
1779  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1780  *  and continue with next VCPU and so on.
1781  */
1782 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1783 {
1784 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1785         bool eligible;
1786
1787         eligible = !vcpu->spin_loop.in_spin_loop ||
1788                     vcpu->spin_loop.dy_eligible;
1789
1790         if (vcpu->spin_loop.in_spin_loop)
1791                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1792
1793         return eligible;
1794 #else
1795         return true;
1796 #endif
1797 }
1798
1799 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1800 {
1801         struct kvm *kvm = me->kvm;
1802         struct kvm_vcpu *vcpu;
1803         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1804         int yielded = 0;
1805         int try = 3;
1806         int pass;
1807         int i;
1808
1809         kvm_vcpu_set_in_spin_loop(me, true);
1810         /*
1811          * We boost the priority of a VCPU that is runnable but not
1812          * currently running, because it got preempted by something
1813          * else and called schedule in __vcpu_run.  Hopefully that
1814          * VCPU is holding the lock that we need and will release it.
1815          * We approximate round-robin by starting at the last boosted VCPU.
1816          */
1817         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1818                 kvm_for_each_vcpu(i, vcpu, kvm) {
1819                         if (!pass && i <= last_boosted_vcpu) {
1820                                 i = last_boosted_vcpu;
1821                                 continue;
1822                         } else if (pass && i > last_boosted_vcpu)
1823                                 break;
1824                         if (!ACCESS_ONCE(vcpu->preempted))
1825                                 continue;
1826                         if (vcpu == me)
1827                                 continue;
1828                         if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1829                                 continue;
1830                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1831                                 continue;
1832
1833                         yielded = kvm_vcpu_yield_to(vcpu);
1834                         if (yielded > 0) {
1835                                 kvm->last_boosted_vcpu = i;
1836                                 break;
1837                         } else if (yielded < 0) {
1838                                 try--;
1839                                 if (!try)
1840                                         break;
1841                         }
1842                 }
1843         }
1844         kvm_vcpu_set_in_spin_loop(me, false);
1845
1846         /* Ensure vcpu is not eligible during next spinloop */
1847         kvm_vcpu_set_dy_eligible(me, false);
1848 }
1849 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1850
1851 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1852 {
1853         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1854         struct page *page;
1855
1856         if (vmf->pgoff == 0)
1857                 page = virt_to_page(vcpu->run);
1858 #ifdef CONFIG_X86
1859         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1860                 page = virt_to_page(vcpu->arch.pio_data);
1861 #endif
1862 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1863         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1864                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1865 #endif
1866         else
1867                 return kvm_arch_vcpu_fault(vcpu, vmf);
1868         get_page(page);
1869         vmf->page = page;
1870         return 0;
1871 }
1872
1873 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1874         .fault = kvm_vcpu_fault,
1875 };
1876
1877 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1878 {
1879         vma->vm_ops = &kvm_vcpu_vm_ops;
1880         return 0;
1881 }
1882
1883 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1884 {
1885         struct kvm_vcpu *vcpu = filp->private_data;
1886
1887         kvm_put_kvm(vcpu->kvm);
1888         return 0;
1889 }
1890
1891 static struct file_operations kvm_vcpu_fops = {
1892         .release        = kvm_vcpu_release,
1893         .unlocked_ioctl = kvm_vcpu_ioctl,
1894 #ifdef CONFIG_COMPAT
1895         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1896 #endif
1897         .mmap           = kvm_vcpu_mmap,
1898         .llseek         = noop_llseek,
1899 };
1900
1901 /*
1902  * Allocates an inode for the vcpu.
1903  */
1904 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1905 {
1906         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1907 }
1908
1909 /*
1910  * Creates some virtual cpus.  Good luck creating more than one.
1911  */
1912 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1913 {
1914         int r;
1915         struct kvm_vcpu *vcpu, *v;
1916
1917         if (id >= KVM_MAX_VCPUS)
1918                 return -EINVAL;
1919
1920         vcpu = kvm_arch_vcpu_create(kvm, id);
1921         if (IS_ERR(vcpu))
1922                 return PTR_ERR(vcpu);
1923
1924         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1925
1926         r = kvm_arch_vcpu_setup(vcpu);
1927         if (r)
1928                 goto vcpu_destroy;
1929
1930         mutex_lock(&kvm->lock);
1931         if (!kvm_vcpu_compatible(vcpu)) {
1932                 r = -EINVAL;
1933                 goto unlock_vcpu_destroy;
1934         }
1935         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1936                 r = -EINVAL;
1937                 goto unlock_vcpu_destroy;
1938         }
1939
1940         kvm_for_each_vcpu(r, v, kvm)
1941                 if (v->vcpu_id == id) {
1942                         r = -EEXIST;
1943                         goto unlock_vcpu_destroy;
1944                 }
1945
1946         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1947
1948         /* Now it's all set up, let userspace reach it */
1949         kvm_get_kvm(kvm);
1950         r = create_vcpu_fd(vcpu);
1951         if (r < 0) {
1952                 kvm_put_kvm(kvm);
1953                 goto unlock_vcpu_destroy;
1954         }
1955
1956         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1957         smp_wmb();
1958         atomic_inc(&kvm->online_vcpus);
1959
1960         mutex_unlock(&kvm->lock);
1961         kvm_arch_vcpu_postcreate(vcpu);
1962         return r;
1963
1964 unlock_vcpu_destroy:
1965         mutex_unlock(&kvm->lock);
1966 vcpu_destroy:
1967         kvm_arch_vcpu_destroy(vcpu);
1968         return r;
1969 }
1970
1971 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1972 {
1973         if (sigset) {
1974                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1975                 vcpu->sigset_active = 1;
1976                 vcpu->sigset = *sigset;
1977         } else
1978                 vcpu->sigset_active = 0;
1979         return 0;
1980 }
1981
1982 static long kvm_vcpu_ioctl(struct file *filp,
1983                            unsigned int ioctl, unsigned long arg)
1984 {
1985         struct kvm_vcpu *vcpu = filp->private_data;
1986         void __user *argp = (void __user *)arg;
1987         int r;
1988         struct kvm_fpu *fpu = NULL;
1989         struct kvm_sregs *kvm_sregs = NULL;
1990
1991         if (vcpu->kvm->mm != current->mm)
1992                 return -EIO;
1993
1994 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1995         /*
1996          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1997          * so vcpu_load() would break it.
1998          */
1999         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2000                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2001 #endif
2002
2003
2004         r = vcpu_load(vcpu);
2005         if (r)
2006                 return r;
2007         switch (ioctl) {
2008         case KVM_RUN:
2009                 r = -EINVAL;
2010                 if (arg)
2011                         goto out;
2012                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2013                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2014                 break;
2015         case KVM_GET_REGS: {
2016                 struct kvm_regs *kvm_regs;
2017
2018                 r = -ENOMEM;
2019                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2020                 if (!kvm_regs)
2021                         goto out;
2022                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2023                 if (r)
2024                         goto out_free1;
2025                 r = -EFAULT;
2026                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2027                         goto out_free1;
2028                 r = 0;
2029 out_free1:
2030                 kfree(kvm_regs);
2031                 break;
2032         }
2033         case KVM_SET_REGS: {
2034                 struct kvm_regs *kvm_regs;
2035
2036                 r = -ENOMEM;
2037                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2038                 if (IS_ERR(kvm_regs)) {
2039                         r = PTR_ERR(kvm_regs);
2040                         goto out;
2041                 }
2042                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2043                 kfree(kvm_regs);
2044                 break;
2045         }
2046         case KVM_GET_SREGS: {
2047                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2048                 r = -ENOMEM;
2049                 if (!kvm_sregs)
2050                         goto out;
2051                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2052                 if (r)
2053                         goto out;
2054                 r = -EFAULT;
2055                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2056                         goto out;
2057                 r = 0;
2058                 break;
2059         }
2060         case KVM_SET_SREGS: {
2061                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2062                 if (IS_ERR(kvm_sregs)) {
2063                         r = PTR_ERR(kvm_sregs);
2064                         kvm_sregs = NULL;
2065                         goto out;
2066                 }
2067                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2068                 break;
2069         }
2070         case KVM_GET_MP_STATE: {
2071                 struct kvm_mp_state mp_state;
2072
2073                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2074                 if (r)
2075                         goto out;
2076                 r = -EFAULT;
2077                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2078                         goto out;
2079                 r = 0;
2080                 break;
2081         }
2082         case KVM_SET_MP_STATE: {
2083                 struct kvm_mp_state mp_state;
2084
2085                 r = -EFAULT;
2086                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2087                         goto out;
2088                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2089                 break;
2090         }
2091         case KVM_TRANSLATE: {
2092                 struct kvm_translation tr;
2093
2094                 r = -EFAULT;
2095                 if (copy_from_user(&tr, argp, sizeof tr))
2096                         goto out;
2097                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2098                 if (r)
2099                         goto out;
2100                 r = -EFAULT;
2101                 if (copy_to_user(argp, &tr, sizeof tr))
2102                         goto out;
2103                 r = 0;
2104                 break;
2105         }
2106         case KVM_SET_GUEST_DEBUG: {
2107                 struct kvm_guest_debug dbg;
2108
2109                 r = -EFAULT;
2110                 if (copy_from_user(&dbg, argp, sizeof dbg))
2111                         goto out;
2112                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2113                 break;
2114         }
2115         case KVM_SET_SIGNAL_MASK: {
2116                 struct kvm_signal_mask __user *sigmask_arg = argp;
2117                 struct kvm_signal_mask kvm_sigmask;
2118                 sigset_t sigset, *p;
2119
2120                 p = NULL;
2121                 if (argp) {
2122                         r = -EFAULT;
2123                         if (copy_from_user(&kvm_sigmask, argp,
2124                                            sizeof kvm_sigmask))
2125                                 goto out;
2126                         r = -EINVAL;
2127                         if (kvm_sigmask.len != sizeof sigset)
2128                                 goto out;
2129                         r = -EFAULT;
2130                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2131                                            sizeof sigset))
2132                                 goto out;
2133                         p = &sigset;
2134                 }
2135                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2136                 break;
2137         }
2138         case KVM_GET_FPU: {
2139                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2140                 r = -ENOMEM;
2141                 if (!fpu)
2142                         goto out;
2143                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2144                 if (r)
2145                         goto out;
2146                 r = -EFAULT;
2147                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2148                         goto out;
2149                 r = 0;
2150                 break;
2151         }
2152         case KVM_SET_FPU: {
2153                 fpu = memdup_user(argp, sizeof(*fpu));
2154                 if (IS_ERR(fpu)) {
2155                         r = PTR_ERR(fpu);
2156                         fpu = NULL;
2157                         goto out;
2158                 }
2159                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2160                 break;
2161         }
2162         default:
2163                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2164         }
2165 out:
2166         vcpu_put(vcpu);
2167         kfree(fpu);
2168         kfree(kvm_sregs);
2169         return r;
2170 }
2171
2172 #ifdef CONFIG_COMPAT
2173 static long kvm_vcpu_compat_ioctl(struct file *filp,
2174                                   unsigned int ioctl, unsigned long arg)
2175 {
2176         struct kvm_vcpu *vcpu = filp->private_data;
2177         void __user *argp = compat_ptr(arg);
2178         int r;
2179
2180         if (vcpu->kvm->mm != current->mm)
2181                 return -EIO;
2182
2183         switch (ioctl) {
2184         case KVM_SET_SIGNAL_MASK: {
2185                 struct kvm_signal_mask __user *sigmask_arg = argp;
2186                 struct kvm_signal_mask kvm_sigmask;
2187                 compat_sigset_t csigset;
2188                 sigset_t sigset;
2189
2190                 if (argp) {
2191                         r = -EFAULT;
2192                         if (copy_from_user(&kvm_sigmask, argp,
2193                                            sizeof kvm_sigmask))
2194                                 goto out;
2195                         r = -EINVAL;
2196                         if (kvm_sigmask.len != sizeof csigset)
2197                                 goto out;
2198                         r = -EFAULT;
2199                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2200                                            sizeof csigset))
2201                                 goto out;
2202                         sigset_from_compat(&sigset, &csigset);
2203                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2204                 } else
2205                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2206                 break;
2207         }
2208         default:
2209                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2210         }
2211
2212 out:
2213         return r;
2214 }
2215 #endif
2216
2217 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2218                                  int (*accessor)(struct kvm_device *dev,
2219                                                  struct kvm_device_attr *attr),
2220                                  unsigned long arg)
2221 {
2222         struct kvm_device_attr attr;
2223
2224         if (!accessor)
2225                 return -EPERM;
2226
2227         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2228                 return -EFAULT;
2229
2230         return accessor(dev, &attr);
2231 }
2232
2233 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2234                              unsigned long arg)
2235 {
2236         struct kvm_device *dev = filp->private_data;
2237
2238         switch (ioctl) {
2239         case KVM_SET_DEVICE_ATTR:
2240                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2241         case KVM_GET_DEVICE_ATTR:
2242                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2243         case KVM_HAS_DEVICE_ATTR:
2244                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2245         default:
2246                 if (dev->ops->ioctl)
2247                         return dev->ops->ioctl(dev, ioctl, arg);
2248
2249                 return -ENOTTY;
2250         }
2251 }
2252
2253 static int kvm_device_release(struct inode *inode, struct file *filp)
2254 {
2255         struct kvm_device *dev = filp->private_data;
2256         struct kvm *kvm = dev->kvm;
2257
2258         kvm_put_kvm(kvm);
2259         return 0;
2260 }
2261
2262 static const struct file_operations kvm_device_fops = {
2263         .unlocked_ioctl = kvm_device_ioctl,
2264 #ifdef CONFIG_COMPAT
2265         .compat_ioctl = kvm_device_ioctl,
2266 #endif
2267         .release = kvm_device_release,
2268 };
2269
2270 struct kvm_device *kvm_device_from_filp(struct file *filp)
2271 {
2272         if (filp->f_op != &kvm_device_fops)
2273                 return NULL;
2274
2275         return filp->private_data;
2276 }
2277
2278 static int kvm_ioctl_create_device(struct kvm *kvm,
2279                                    struct kvm_create_device *cd)
2280 {
2281         struct kvm_device_ops *ops = NULL;
2282         struct kvm_device *dev;
2283         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2284         int ret;
2285
2286         switch (cd->type) {
2287 #ifdef CONFIG_KVM_MPIC
2288         case KVM_DEV_TYPE_FSL_MPIC_20:
2289         case KVM_DEV_TYPE_FSL_MPIC_42:
2290                 ops = &kvm_mpic_ops;
2291                 break;
2292 #endif
2293 #ifdef CONFIG_KVM_XICS
2294         case KVM_DEV_TYPE_XICS:
2295                 ops = &kvm_xics_ops;
2296                 break;
2297 #endif
2298 #ifdef CONFIG_KVM_VFIO
2299         case KVM_DEV_TYPE_VFIO:
2300                 ops = &kvm_vfio_ops;
2301                 break;
2302 #endif
2303 #ifdef CONFIG_KVM_ARM_VGIC
2304         case KVM_DEV_TYPE_ARM_VGIC_V2:
2305                 ops = &kvm_arm_vgic_v2_ops;
2306                 break;
2307 #endif
2308 #ifdef CONFIG_S390
2309         case KVM_DEV_TYPE_FLIC:
2310                 ops = &kvm_flic_ops;
2311                 break;
2312 #endif
2313         default:
2314                 return -ENODEV;
2315         }
2316
2317         if (test)
2318                 return 0;
2319
2320         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2321         if (!dev)
2322                 return -ENOMEM;
2323
2324         dev->ops = ops;
2325         dev->kvm = kvm;
2326
2327         ret = ops->create(dev, cd->type);
2328         if (ret < 0) {
2329                 kfree(dev);
2330                 return ret;
2331         }
2332
2333         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2334         if (ret < 0) {
2335                 ops->destroy(dev);
2336                 return ret;
2337         }
2338
2339         list_add(&dev->vm_node, &kvm->devices);
2340         kvm_get_kvm(kvm);
2341         cd->fd = ret;
2342         return 0;
2343 }
2344
2345 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2346 {
2347         switch (arg) {
2348         case KVM_CAP_USER_MEMORY:
2349         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2350         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2351 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2352         case KVM_CAP_SET_BOOT_CPU_ID:
2353 #endif
2354         case KVM_CAP_INTERNAL_ERROR_DATA:
2355 #ifdef CONFIG_HAVE_KVM_MSI
2356         case KVM_CAP_SIGNAL_MSI:
2357 #endif
2358 #ifdef CONFIG_HAVE_KVM_IRQFD
2359         case KVM_CAP_IRQFD_RESAMPLE:
2360 #endif
2361         case KVM_CAP_CHECK_EXTENSION_VM:
2362                 return 1;
2363 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2364         case KVM_CAP_IRQ_ROUTING:
2365                 return KVM_MAX_IRQ_ROUTES;
2366 #endif
2367         default:
2368                 break;
2369         }
2370         return kvm_vm_ioctl_check_extension(kvm, arg);
2371 }
2372
2373 static long kvm_vm_ioctl(struct file *filp,
2374                            unsigned int ioctl, unsigned long arg)
2375 {
2376         struct kvm *kvm = filp->private_data;
2377         void __user *argp = (void __user *)arg;
2378         int r;
2379
2380         if (kvm->mm != current->mm)
2381                 return -EIO;
2382         switch (ioctl) {
2383         case KVM_CREATE_VCPU:
2384                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2385                 break;
2386         case KVM_SET_USER_MEMORY_REGION: {
2387                 struct kvm_userspace_memory_region kvm_userspace_mem;
2388
2389                 r = -EFAULT;
2390                 if (copy_from_user(&kvm_userspace_mem, argp,
2391                                                 sizeof kvm_userspace_mem))
2392                         goto out;
2393
2394                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2395                 break;
2396         }
2397         case KVM_GET_DIRTY_LOG: {
2398                 struct kvm_dirty_log log;
2399
2400                 r = -EFAULT;
2401                 if (copy_from_user(&log, argp, sizeof log))
2402                         goto out;
2403                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2404                 break;
2405         }
2406 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2407         case KVM_REGISTER_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_register_coalesced_mmio(kvm, &zone);
2413                 break;
2414         }
2415         case KVM_UNREGISTER_COALESCED_MMIO: {
2416                 struct kvm_coalesced_mmio_zone zone;
2417                 r = -EFAULT;
2418                 if (copy_from_user(&zone, argp, sizeof zone))
2419                         goto out;
2420                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2421                 break;
2422         }
2423 #endif
2424         case KVM_IRQFD: {
2425                 struct kvm_irqfd data;
2426
2427                 r = -EFAULT;
2428                 if (copy_from_user(&data, argp, sizeof data))
2429                         goto out;
2430                 r = kvm_irqfd(kvm, &data);
2431                 break;
2432         }
2433         case KVM_IOEVENTFD: {
2434                 struct kvm_ioeventfd data;
2435
2436                 r = -EFAULT;
2437                 if (copy_from_user(&data, argp, sizeof data))
2438                         goto out;
2439                 r = kvm_ioeventfd(kvm, &data);
2440                 break;
2441         }
2442 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2443         case KVM_SET_BOOT_CPU_ID:
2444                 r = 0;
2445                 mutex_lock(&kvm->lock);
2446                 if (atomic_read(&kvm->online_vcpus) != 0)
2447                         r = -EBUSY;
2448                 else
2449                         kvm->bsp_vcpu_id = arg;
2450                 mutex_unlock(&kvm->lock);
2451                 break;
2452 #endif
2453 #ifdef CONFIG_HAVE_KVM_MSI
2454         case KVM_SIGNAL_MSI: {
2455                 struct kvm_msi msi;
2456
2457                 r = -EFAULT;
2458                 if (copy_from_user(&msi, argp, sizeof msi))
2459                         goto out;
2460                 r = kvm_send_userspace_msi(kvm, &msi);
2461                 break;
2462         }
2463 #endif
2464 #ifdef __KVM_HAVE_IRQ_LINE
2465         case KVM_IRQ_LINE_STATUS:
2466         case KVM_IRQ_LINE: {
2467                 struct kvm_irq_level irq_event;
2468
2469                 r = -EFAULT;
2470                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2471                         goto out;
2472
2473                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2474                                         ioctl == KVM_IRQ_LINE_STATUS);
2475                 if (r)
2476                         goto out;
2477
2478                 r = -EFAULT;
2479                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2480                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2481                                 goto out;
2482                 }
2483
2484                 r = 0;
2485                 break;
2486         }
2487 #endif
2488 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2489         case KVM_SET_GSI_ROUTING: {
2490                 struct kvm_irq_routing routing;
2491                 struct kvm_irq_routing __user *urouting;
2492                 struct kvm_irq_routing_entry *entries;
2493
2494                 r = -EFAULT;
2495                 if (copy_from_user(&routing, argp, sizeof(routing)))
2496                         goto out;
2497                 r = -EINVAL;
2498                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2499                         goto out;
2500                 if (routing.flags)
2501                         goto out;
2502                 r = -ENOMEM;
2503                 entries = vmalloc(routing.nr * sizeof(*entries));
2504                 if (!entries)
2505                         goto out;
2506                 r = -EFAULT;
2507                 urouting = argp;
2508                 if (copy_from_user(entries, urouting->entries,
2509                                    routing.nr * sizeof(*entries)))
2510                         goto out_free_irq_routing;
2511                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2512                                         routing.flags);
2513         out_free_irq_routing:
2514                 vfree(entries);
2515                 break;
2516         }
2517 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2518         case KVM_CREATE_DEVICE: {
2519                 struct kvm_create_device cd;
2520
2521                 r = -EFAULT;
2522                 if (copy_from_user(&cd, argp, sizeof(cd)))
2523                         goto out;
2524
2525                 r = kvm_ioctl_create_device(kvm, &cd);
2526                 if (r)
2527                         goto out;
2528
2529                 r = -EFAULT;
2530                 if (copy_to_user(argp, &cd, sizeof(cd)))
2531                         goto out;
2532
2533                 r = 0;
2534                 break;
2535         }
2536         case KVM_CHECK_EXTENSION:
2537                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2538                 break;
2539         default:
2540                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2541                 if (r == -ENOTTY)
2542                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2543         }
2544 out:
2545         return r;
2546 }
2547
2548 #ifdef CONFIG_COMPAT
2549 struct compat_kvm_dirty_log {
2550         __u32 slot;
2551         __u32 padding1;
2552         union {
2553                 compat_uptr_t dirty_bitmap; /* one bit per page */
2554                 __u64 padding2;
2555         };
2556 };
2557
2558 static long kvm_vm_compat_ioctl(struct file *filp,
2559                            unsigned int ioctl, unsigned long arg)
2560 {
2561         struct kvm *kvm = filp->private_data;
2562         int r;
2563
2564         if (kvm->mm != current->mm)
2565                 return -EIO;
2566         switch (ioctl) {
2567         case KVM_GET_DIRTY_LOG: {
2568                 struct compat_kvm_dirty_log compat_log;
2569                 struct kvm_dirty_log log;
2570
2571                 r = -EFAULT;
2572                 if (copy_from_user(&compat_log, (void __user *)arg,
2573                                    sizeof(compat_log)))
2574                         goto out;
2575                 log.slot         = compat_log.slot;
2576                 log.padding1     = compat_log.padding1;
2577                 log.padding2     = compat_log.padding2;
2578                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2579
2580                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2581                 break;
2582         }
2583         default:
2584                 r = kvm_vm_ioctl(filp, ioctl, arg);
2585         }
2586
2587 out:
2588         return r;
2589 }
2590 #endif
2591
2592 static struct file_operations kvm_vm_fops = {
2593         .release        = kvm_vm_release,
2594         .unlocked_ioctl = kvm_vm_ioctl,
2595 #ifdef CONFIG_COMPAT
2596         .compat_ioctl   = kvm_vm_compat_ioctl,
2597 #endif
2598         .llseek         = noop_llseek,
2599 };
2600
2601 static int kvm_dev_ioctl_create_vm(unsigned long type)
2602 {
2603         int r;
2604         struct kvm *kvm;
2605
2606         kvm = kvm_create_vm(type);
2607         if (IS_ERR(kvm))
2608                 return PTR_ERR(kvm);
2609 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2610         r = kvm_coalesced_mmio_init(kvm);
2611         if (r < 0) {
2612                 kvm_put_kvm(kvm);
2613                 return r;
2614         }
2615 #endif
2616         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2617         if (r < 0)
2618                 kvm_put_kvm(kvm);
2619
2620         return r;
2621 }
2622
2623 static long kvm_dev_ioctl(struct file *filp,
2624                           unsigned int ioctl, unsigned long arg)
2625 {
2626         long r = -EINVAL;
2627
2628         switch (ioctl) {
2629         case KVM_GET_API_VERSION:
2630                 r = -EINVAL;
2631                 if (arg)
2632                         goto out;
2633                 r = KVM_API_VERSION;
2634                 break;
2635         case KVM_CREATE_VM:
2636                 r = kvm_dev_ioctl_create_vm(arg);
2637                 break;
2638         case KVM_CHECK_EXTENSION:
2639                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2640                 break;
2641         case KVM_GET_VCPU_MMAP_SIZE:
2642                 r = -EINVAL;
2643                 if (arg)
2644                         goto out;
2645                 r = PAGE_SIZE;     /* struct kvm_run */
2646 #ifdef CONFIG_X86
2647                 r += PAGE_SIZE;    /* pio data page */
2648 #endif
2649 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2650                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2651 #endif
2652                 break;
2653         case KVM_TRACE_ENABLE:
2654         case KVM_TRACE_PAUSE:
2655         case KVM_TRACE_DISABLE:
2656                 r = -EOPNOTSUPP;
2657                 break;
2658         default:
2659                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2660         }
2661 out:
2662         return r;
2663 }
2664
2665 static struct file_operations kvm_chardev_ops = {
2666         .unlocked_ioctl = kvm_dev_ioctl,
2667         .compat_ioctl   = kvm_dev_ioctl,
2668         .llseek         = noop_llseek,
2669 };
2670
2671 static struct miscdevice kvm_dev = {
2672         KVM_MINOR,
2673         "kvm",
2674         &kvm_chardev_ops,
2675 };
2676
2677 static void hardware_enable_nolock(void *junk)
2678 {
2679         int cpu = raw_smp_processor_id();
2680         int r;
2681
2682         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2683                 return;
2684
2685         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2686
2687         r = kvm_arch_hardware_enable();
2688
2689         if (r) {
2690                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2691                 atomic_inc(&hardware_enable_failed);
2692                 printk(KERN_INFO "kvm: enabling virtualization on "
2693                                  "CPU%d failed\n", cpu);
2694         }
2695 }
2696
2697 static void hardware_enable(void)
2698 {
2699         raw_spin_lock(&kvm_count_lock);
2700         if (kvm_usage_count)
2701                 hardware_enable_nolock(NULL);
2702         raw_spin_unlock(&kvm_count_lock);
2703 }
2704
2705 static void hardware_disable_nolock(void *junk)
2706 {
2707         int cpu = raw_smp_processor_id();
2708
2709         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2710                 return;
2711         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2712         kvm_arch_hardware_disable();
2713 }
2714
2715 static void hardware_disable(void)
2716 {
2717         raw_spin_lock(&kvm_count_lock);
2718         if (kvm_usage_count)
2719                 hardware_disable_nolock(NULL);
2720         raw_spin_unlock(&kvm_count_lock);
2721 }
2722
2723 static void hardware_disable_all_nolock(void)
2724 {
2725         BUG_ON(!kvm_usage_count);
2726
2727         kvm_usage_count--;
2728         if (!kvm_usage_count)
2729                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2730 }
2731
2732 static void hardware_disable_all(void)
2733 {
2734         raw_spin_lock(&kvm_count_lock);
2735         hardware_disable_all_nolock();
2736         raw_spin_unlock(&kvm_count_lock);
2737 }
2738
2739 static int hardware_enable_all(void)
2740 {
2741         int r = 0;
2742
2743         raw_spin_lock(&kvm_count_lock);
2744
2745         kvm_usage_count++;
2746         if (kvm_usage_count == 1) {
2747                 atomic_set(&hardware_enable_failed, 0);
2748                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2749
2750                 if (atomic_read(&hardware_enable_failed)) {
2751                         hardware_disable_all_nolock();
2752                         r = -EBUSY;
2753                 }
2754         }
2755
2756         raw_spin_unlock(&kvm_count_lock);
2757
2758         return r;
2759 }
2760
2761 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2762                            void *v)
2763 {
2764         int cpu = (long)v;
2765
2766         val &= ~CPU_TASKS_FROZEN;
2767         switch (val) {
2768         case CPU_DYING:
2769                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2770                        cpu);
2771                 hardware_disable();
2772                 break;
2773         case CPU_STARTING:
2774                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2775                        cpu);
2776                 hardware_enable();
2777                 break;
2778         }
2779         return NOTIFY_OK;
2780 }
2781
2782 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2783                       void *v)
2784 {
2785         /*
2786          * Some (well, at least mine) BIOSes hang on reboot if
2787          * in vmx root mode.
2788          *
2789          * And Intel TXT required VMX off for all cpu when system shutdown.
2790          */
2791         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2792         kvm_rebooting = true;
2793         on_each_cpu(hardware_disable_nolock, NULL, 1);
2794         return NOTIFY_OK;
2795 }
2796
2797 static struct notifier_block kvm_reboot_notifier = {
2798         .notifier_call = kvm_reboot,
2799         .priority = 0,
2800 };
2801
2802 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2803 {
2804         int i;
2805
2806         for (i = 0; i < bus->dev_count; i++) {
2807                 struct kvm_io_device *pos = bus->range[i].dev;
2808
2809                 kvm_iodevice_destructor(pos);
2810         }
2811         kfree(bus);
2812 }
2813
2814 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2815                                  const struct kvm_io_range *r2)
2816 {
2817         if (r1->addr < r2->addr)
2818                 return -1;
2819         if (r1->addr + r1->len > r2->addr + r2->len)
2820                 return 1;
2821         return 0;
2822 }
2823
2824 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2825 {
2826         return kvm_io_bus_cmp(p1, p2);
2827 }
2828
2829 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2830                           gpa_t addr, int len)
2831 {
2832         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2833                 .addr = addr,
2834                 .len = len,
2835                 .dev = dev,
2836         };
2837
2838         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2839                 kvm_io_bus_sort_cmp, NULL);
2840
2841         return 0;
2842 }
2843
2844 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2845                              gpa_t addr, int len)
2846 {
2847         struct kvm_io_range *range, key;
2848         int off;
2849
2850         key = (struct kvm_io_range) {
2851                 .addr = addr,
2852                 .len = len,
2853         };
2854
2855         range = bsearch(&key, bus->range, bus->dev_count,
2856                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2857         if (range == NULL)
2858                 return -ENOENT;
2859
2860         off = range - bus->range;
2861
2862         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2863                 off--;
2864
2865         return off;
2866 }
2867
2868 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2869                               struct kvm_io_range *range, const void *val)
2870 {
2871         int idx;
2872
2873         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2874         if (idx < 0)
2875                 return -EOPNOTSUPP;
2876
2877         while (idx < bus->dev_count &&
2878                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2879                 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2880                                         range->len, val))
2881                         return idx;
2882                 idx++;
2883         }
2884
2885         return -EOPNOTSUPP;
2886 }
2887
2888 /* kvm_io_bus_write - called under kvm->slots_lock */
2889 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2890                      int len, const void *val)
2891 {
2892         struct kvm_io_bus *bus;
2893         struct kvm_io_range range;
2894         int r;
2895
2896         range = (struct kvm_io_range) {
2897                 .addr = addr,
2898                 .len = len,
2899         };
2900
2901         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2902         r = __kvm_io_bus_write(bus, &range, val);
2903         return r < 0 ? r : 0;
2904 }
2905
2906 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2907 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2908                             int len, const void *val, long cookie)
2909 {
2910         struct kvm_io_bus *bus;
2911         struct kvm_io_range range;
2912
2913         range = (struct kvm_io_range) {
2914                 .addr = addr,
2915                 .len = len,
2916         };
2917
2918         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2919
2920         /* First try the device referenced by cookie. */
2921         if ((cookie >= 0) && (cookie < bus->dev_count) &&
2922             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2923                 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2924                                         val))
2925                         return cookie;
2926
2927         /*
2928          * cookie contained garbage; fall back to search and return the
2929          * correct cookie value.
2930          */
2931         return __kvm_io_bus_write(bus, &range, val);
2932 }
2933
2934 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2935                              void *val)
2936 {
2937         int idx;
2938
2939         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2940         if (idx < 0)
2941                 return -EOPNOTSUPP;
2942
2943         while (idx < bus->dev_count &&
2944                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2945                 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2946                                        range->len, val))
2947                         return idx;
2948                 idx++;
2949         }
2950
2951         return -EOPNOTSUPP;
2952 }
2953 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2954
2955 /* kvm_io_bus_read - called under kvm->slots_lock */
2956 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2957                     int len, void *val)
2958 {
2959         struct kvm_io_bus *bus;
2960         struct kvm_io_range range;
2961         int r;
2962
2963         range = (struct kvm_io_range) {
2964                 .addr = addr,
2965                 .len = len,
2966         };
2967
2968         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2969         r = __kvm_io_bus_read(bus, &range, val);
2970         return r < 0 ? r : 0;
2971 }
2972
2973
2974 /* Caller must hold slots_lock. */
2975 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2976                             int len, struct kvm_io_device *dev)
2977 {
2978         struct kvm_io_bus *new_bus, *bus;
2979
2980         bus = kvm->buses[bus_idx];
2981         /* exclude ioeventfd which is limited by maximum fd */
2982         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2983                 return -ENOSPC;
2984
2985         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2986                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2987         if (!new_bus)
2988                 return -ENOMEM;
2989         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2990                sizeof(struct kvm_io_range)));
2991         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2992         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2993         synchronize_srcu_expedited(&kvm->srcu);
2994         kfree(bus);
2995
2996         return 0;
2997 }
2998
2999 /* Caller must hold slots_lock. */
3000 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3001                               struct kvm_io_device *dev)
3002 {
3003         int i, r;
3004         struct kvm_io_bus *new_bus, *bus;
3005
3006         bus = kvm->buses[bus_idx];
3007         r = -ENOENT;
3008         for (i = 0; i < bus->dev_count; i++)
3009                 if (bus->range[i].dev == dev) {
3010                         r = 0;
3011                         break;
3012                 }
3013
3014         if (r)
3015                 return r;
3016
3017         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3018                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3019         if (!new_bus)
3020                 return -ENOMEM;
3021
3022         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3023         new_bus->dev_count--;
3024         memcpy(new_bus->range + i, bus->range + i + 1,
3025                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3026
3027         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3028         synchronize_srcu_expedited(&kvm->srcu);
3029         kfree(bus);
3030         return r;
3031 }
3032
3033 static struct notifier_block kvm_cpu_notifier = {
3034         .notifier_call = kvm_cpu_hotplug,
3035 };
3036
3037 static int vm_stat_get(void *_offset, u64 *val)
3038 {
3039         unsigned offset = (long)_offset;
3040         struct kvm *kvm;
3041
3042         *val = 0;
3043         spin_lock(&kvm_lock);
3044         list_for_each_entry(kvm, &vm_list, vm_list)
3045                 *val += *(u32 *)((void *)kvm + offset);
3046         spin_unlock(&kvm_lock);
3047         return 0;
3048 }
3049
3050 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3051
3052 static int vcpu_stat_get(void *_offset, u64 *val)
3053 {
3054         unsigned offset = (long)_offset;
3055         struct kvm *kvm;
3056         struct kvm_vcpu *vcpu;
3057         int i;
3058
3059         *val = 0;
3060         spin_lock(&kvm_lock);
3061         list_for_each_entry(kvm, &vm_list, vm_list)
3062                 kvm_for_each_vcpu(i, vcpu, kvm)
3063                         *val += *(u32 *)((void *)vcpu + offset);
3064
3065         spin_unlock(&kvm_lock);
3066         return 0;
3067 }
3068
3069 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3070
3071 static const struct file_operations *stat_fops[] = {
3072         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3073         [KVM_STAT_VM]   = &vm_stat_fops,
3074 };
3075
3076 static int kvm_init_debug(void)
3077 {
3078         int r = -EEXIST;
3079         struct kvm_stats_debugfs_item *p;
3080
3081         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3082         if (kvm_debugfs_dir == NULL)
3083                 goto out;
3084
3085         for (p = debugfs_entries; p->name; ++p) {
3086                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3087                                                 (void *)(long)p->offset,
3088                                                 stat_fops[p->kind]);
3089                 if (p->dentry == NULL)
3090                         goto out_dir;
3091         }
3092
3093         return 0;
3094
3095 out_dir:
3096         debugfs_remove_recursive(kvm_debugfs_dir);
3097 out:
3098         return r;
3099 }
3100
3101 static void kvm_exit_debug(void)
3102 {
3103         struct kvm_stats_debugfs_item *p;
3104
3105         for (p = debugfs_entries; p->name; ++p)
3106                 debugfs_remove(p->dentry);
3107         debugfs_remove(kvm_debugfs_dir);
3108 }
3109
3110 static int kvm_suspend(void)
3111 {
3112         if (kvm_usage_count)
3113                 hardware_disable_nolock(NULL);
3114         return 0;
3115 }
3116
3117 static void kvm_resume(void)
3118 {
3119         if (kvm_usage_count) {
3120                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3121                 hardware_enable_nolock(NULL);
3122         }
3123 }
3124
3125 static struct syscore_ops kvm_syscore_ops = {
3126         .suspend = kvm_suspend,
3127         .resume = kvm_resume,
3128 };
3129
3130 static inline
3131 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3132 {
3133         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3134 }
3135
3136 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3137 {
3138         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3139         if (vcpu->preempted)
3140                 vcpu->preempted = false;
3141
3142         kvm_arch_sched_in(vcpu, cpu);
3143
3144         kvm_arch_vcpu_load(vcpu, cpu);
3145 }
3146
3147 static void kvm_sched_out(struct preempt_notifier *pn,
3148                           struct task_struct *next)
3149 {
3150         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3151
3152         if (current->state == TASK_RUNNING)
3153                 vcpu->preempted = true;
3154         kvm_arch_vcpu_put(vcpu);
3155 }
3156
3157 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3158                   struct module *module)
3159 {
3160         int r;
3161         int cpu;
3162
3163         r = kvm_arch_init(opaque);
3164         if (r)
3165                 goto out_fail;
3166
3167         /*
3168          * kvm_arch_init makes sure there's at most one caller
3169          * for architectures that support multiple implementations,
3170          * like intel and amd on x86.
3171          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3172          * conflicts in case kvm is already setup for another implementation.
3173          */
3174         r = kvm_irqfd_init();
3175         if (r)
3176                 goto out_irqfd;
3177
3178         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3179                 r = -ENOMEM;
3180                 goto out_free_0;
3181         }
3182
3183         r = kvm_arch_hardware_setup();
3184         if (r < 0)
3185                 goto out_free_0a;
3186
3187         for_each_online_cpu(cpu) {
3188                 smp_call_function_single(cpu,
3189                                 kvm_arch_check_processor_compat,
3190                                 &r, 1);
3191                 if (r < 0)
3192                         goto out_free_1;
3193         }
3194
3195         r = register_cpu_notifier(&kvm_cpu_notifier);
3196         if (r)
3197                 goto out_free_2;
3198         register_reboot_notifier(&kvm_reboot_notifier);
3199
3200         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3201         if (!vcpu_align)
3202                 vcpu_align = __alignof__(struct kvm_vcpu);
3203         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3204                                            0, NULL);
3205         if (!kvm_vcpu_cache) {
3206                 r = -ENOMEM;
3207                 goto out_free_3;
3208         }
3209
3210         r = kvm_async_pf_init();
3211         if (r)
3212                 goto out_free;
3213
3214         kvm_chardev_ops.owner = module;
3215         kvm_vm_fops.owner = module;
3216         kvm_vcpu_fops.owner = module;
3217
3218         r = misc_register(&kvm_dev);
3219         if (r) {
3220                 printk(KERN_ERR "kvm: misc device register failed\n");
3221                 goto out_unreg;
3222         }
3223
3224         register_syscore_ops(&kvm_syscore_ops);
3225
3226         kvm_preempt_ops.sched_in = kvm_sched_in;
3227         kvm_preempt_ops.sched_out = kvm_sched_out;
3228
3229         r = kvm_init_debug();
3230         if (r) {
3231                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3232                 goto out_undebugfs;
3233         }
3234
3235         return 0;
3236
3237 out_undebugfs:
3238         unregister_syscore_ops(&kvm_syscore_ops);
3239         misc_deregister(&kvm_dev);
3240 out_unreg:
3241         kvm_async_pf_deinit();
3242 out_free:
3243         kmem_cache_destroy(kvm_vcpu_cache);
3244 out_free_3:
3245         unregister_reboot_notifier(&kvm_reboot_notifier);
3246         unregister_cpu_notifier(&kvm_cpu_notifier);
3247 out_free_2:
3248 out_free_1:
3249         kvm_arch_hardware_unsetup();
3250 out_free_0a:
3251         free_cpumask_var(cpus_hardware_enabled);
3252 out_free_0:
3253         kvm_irqfd_exit();
3254 out_irqfd:
3255         kvm_arch_exit();
3256 out_fail:
3257         return r;
3258 }
3259 EXPORT_SYMBOL_GPL(kvm_init);
3260
3261 void kvm_exit(void)
3262 {
3263         kvm_exit_debug();
3264         misc_deregister(&kvm_dev);
3265         kmem_cache_destroy(kvm_vcpu_cache);
3266         kvm_async_pf_deinit();
3267         unregister_syscore_ops(&kvm_syscore_ops);
3268         unregister_reboot_notifier(&kvm_reboot_notifier);
3269         unregister_cpu_notifier(&kvm_cpu_notifier);
3270         on_each_cpu(hardware_disable_nolock, NULL, 1);
3271         kvm_arch_hardware_unsetup();
3272         kvm_arch_exit();
3273         kvm_irqfd_exit();
3274         free_cpumask_var(cpus_hardware_enabled);
3275 }
3276 EXPORT_SYMBOL_GPL(kvm_exit);