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 #include "vfio.h"
  61
  62 #define CREATE_TRACE_POINTS
  63 #include <trace/events/kvm.h>
  64
  65 MODULE_AUTHOR("Qumranet");
  66 MODULE_LICENSE("GPL");
  67
  68 /*
  69  * Ordering of locks:
  70  *
  71  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  72  */
  73
  74 DEFINE_SPINLOCK(kvm_lock);
  75 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
  76 LIST_HEAD(vm_list);
  77
  78 static cpumask_var_t cpus_hardware_enabled;
  79 static int kvm_usage_count = 0;
  80 static atomic_t hardware_enable_failed;
  81
  82 struct kmem_cache *kvm_vcpu_cache;
  83 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  84
  85 static __read_mostly struct preempt_ops kvm_preempt_ops;
  86
  87 struct dentry *kvm_debugfs_dir;
  88
  89 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  90                            unsigned long arg);
  91 #ifdef CONFIG_COMPAT
  92 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  93                                   unsigned long arg);
  94 #endif
  95 static int hardware_enable_all(void);
  96 static void hardware_disable_all(void);
  97
  98 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  99
 100 static void kvm_release_pfn_dirty(pfn_t pfn);
 101 static void mark_page_dirty_in_slot(struct kvm *kvm,
 102                                     struct kvm_memory_slot *memslot, gfn_t gfn);
 103
 104 __visible bool kvm_rebooting;
 105 EXPORT_SYMBOL_GPL(kvm_rebooting);
 106
 107 static bool largepages_enabled = true;
 108
 109 bool kvm_is_mmio_pfn(pfn_t pfn)
 110 {
 111         if (pfn_valid(pfn))
 112                 return PageReserved(pfn_to_page(pfn));
 113
 114         return true;
 115 }
 116
 117 /*
 118  * Switches to specified vcpu, until a matching vcpu_put()
 119  */
 120 int vcpu_load(struct kvm_vcpu *vcpu)
 121 {
 122         int cpu;
 123
 124         if (mutex_lock_killable(&vcpu->mutex))
 125                 return -EINTR;
 126         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 127                 /* The thread running this VCPU changed. */
 128                 struct pid *oldpid = vcpu->pid;
 129                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 130                 rcu_assign_pointer(vcpu->pid, newpid);
 131                 if (oldpid)
 132                         synchronize_rcu();
 133                 put_pid(oldpid);
 134         }
 135         cpu = get_cpu();
 136         preempt_notifier_register(&vcpu->preempt_notifier);
 137         kvm_arch_vcpu_load(vcpu, cpu);
 138         put_cpu();
 139         return 0;
 140 }
 141
 142 void vcpu_put(struct kvm_vcpu *vcpu)
 143 {
 144         preempt_disable();
 145         kvm_arch_vcpu_put(vcpu);
 146         preempt_notifier_unregister(&vcpu->preempt_notifier);
 147         preempt_enable();
 148         mutex_unlock(&vcpu->mutex);
 149 }
 150
 151 static void ack_flush(void *_completed)
 152 {
 153 }
 154
 155 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 156 {
 157         int i, cpu, me;
 158         cpumask_var_t cpus;
 159         bool called = true;
 160         struct kvm_vcpu *vcpu;
 161
 162         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 163
 164         me = get_cpu();
 165         kvm_for_each_vcpu(i, vcpu, kvm) {
 166                 kvm_make_request(req, vcpu);
 167                 cpu = vcpu->cpu;
 168
 169                 /* Set ->requests bit before we read ->mode */
 170                 smp_mb();
 171
 172                 if (cpus != NULL && cpu != -1 && cpu != me &&
 173                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 174                         cpumask_set_cpu(cpu, cpus);
 175         }
 176         if (unlikely(cpus == NULL))
 177                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 178         else if (!cpumask_empty(cpus))
 179                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 180         else
 181                 called = false;
 182         put_cpu();
 183         free_cpumask_var(cpus);
 184         return called;
 185 }
 186
 187 void kvm_flush_remote_tlbs(struct kvm *kvm)
 188 {
 189         long dirty_count = kvm->tlbs_dirty;
 190
 191         smp_mb();
 192         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 193                 ++kvm->stat.remote_tlb_flush;
 194         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 195 }
 196 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 197
 198 void kvm_reload_remote_mmus(struct kvm *kvm)
 199 {
 200         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 201 }
 202
 203 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 204 {
 205         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 206 }
 207
 208 void kvm_make_scan_ioapic_request(struct kvm *kvm)
 209 {
 210         make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
 211 }
 212
 213 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 214 {
 215         struct page *page;
 216         int r;
 217
 218         mutex_init(&vcpu->mutex);
 219         vcpu->cpu = -1;
 220         vcpu->kvm = kvm;
 221         vcpu->vcpu_id = id;
 222         vcpu->pid = NULL;
 223         init_waitqueue_head(&vcpu->wq);
 224         kvm_async_pf_vcpu_init(vcpu);
 225
 226         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 227         if (!page) {
 228                 r = -ENOMEM;
 229                 goto fail;
 230         }
 231         vcpu->run = page_address(page);
 232
 233         kvm_vcpu_set_in_spin_loop(vcpu, false);
 234         kvm_vcpu_set_dy_eligible(vcpu, false);
 235         vcpu->preempted = false;
 236
 237         r = kvm_arch_vcpu_init(vcpu);
 238         if (r < 0)
 239                 goto fail_free_run;
 240         return 0;
 241
 242 fail_free_run:
 243         free_page((unsigned long)vcpu->run);
 244 fail:
 245         return r;
 246 }
 247 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 248
 249 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 250 {
 251         put_pid(vcpu->pid);
 252         kvm_arch_vcpu_uninit(vcpu);
 253         free_page((unsigned long)vcpu->run);
 254 }
 255 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 256
 257 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 258 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 259 {
 260         return container_of(mn, struct kvm, mmu_notifier);
 261 }
 262
 263 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 264                                              struct mm_struct *mm,
 265                                              unsigned long address)
 266 {
 267         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 268         int need_tlb_flush, idx;
 269
 270         /*
 271          * When ->invalidate_page runs, the linux pte has been zapped
 272          * already but the page is still allocated until
 273          * ->invalidate_page returns. So if we increase the sequence
 274          * here the kvm page fault will notice if the spte can't be
 275          * established because the page is going to be freed. If
 276          * instead the kvm page fault establishes the spte before
 277          * ->invalidate_page runs, kvm_unmap_hva will release it
 278          * before returning.
 279          *
 280          * The sequence increase only need to be seen at spin_unlock
 281          * time, and not at spin_lock time.
 282          *
 283          * Increasing the sequence after the spin_unlock would be
 284          * unsafe because the kvm page fault could then establish the
 285          * pte after kvm_unmap_hva returned, without noticing the page
 286          * is going to be freed.
 287          */
 288         idx = srcu_read_lock(&kvm->srcu);
 289         spin_lock(&kvm->mmu_lock);
 290
 291         kvm->mmu_notifier_seq++;
 292         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 293         /* we've to flush the tlb before the pages can be freed */
 294         if (need_tlb_flush)
 295                 kvm_flush_remote_tlbs(kvm);
 296
 297         spin_unlock(&kvm->mmu_lock);
 298         srcu_read_unlock(&kvm->srcu, idx);
 299 }
 300
 301 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 302                                         struct mm_struct *mm,
 303                                         unsigned long address,
 304                                         pte_t pte)
 305 {
 306         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 307         int idx;
 308
 309         idx = srcu_read_lock(&kvm->srcu);
 310         spin_lock(&kvm->mmu_lock);
 311         kvm->mmu_notifier_seq++;
 312         kvm_set_spte_hva(kvm, address, pte);
 313         spin_unlock(&kvm->mmu_lock);
 314         srcu_read_unlock(&kvm->srcu, idx);
 315 }
 316
 317 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 318                                                     struct mm_struct *mm,
 319                                                     unsigned long start,
 320                                                     unsigned long end)
 321 {
 322         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 323         int need_tlb_flush = 0, idx;
 324
 325         idx = srcu_read_lock(&kvm->srcu);
 326         spin_lock(&kvm->mmu_lock);
 327         /*
 328          * The count increase must become visible at unlock time as no
 329          * spte can be established without taking the mmu_lock and
 330          * count is also read inside the mmu_lock critical section.
 331          */
 332         kvm->mmu_notifier_count++;
 333         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 334         need_tlb_flush |= kvm->tlbs_dirty;
 335         /* we've to flush the tlb before the pages can be freed */
 336         if (need_tlb_flush)
 337                 kvm_flush_remote_tlbs(kvm);
 338
 339         spin_unlock(&kvm->mmu_lock);
 340         srcu_read_unlock(&kvm->srcu, idx);
 341 }
 342
 343 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 344                                                   struct mm_struct *mm,
 345                                                   unsigned long start,
 346                                                   unsigned long end)
 347 {
 348         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 349
 350         spin_lock(&kvm->mmu_lock);
 351         /*
 352          * This sequence increase will notify the kvm page fault that
 353          * the page that is going to be mapped in the spte could have
 354          * been freed.
 355          */
 356         kvm->mmu_notifier_seq++;
 357         smp_wmb();
 358         /*
 359          * The above sequence increase must be visible before the
 360          * below count decrease, which is ensured by the smp_wmb above
 361          * in conjunction with the smp_rmb in mmu_notifier_retry().
 362          */
 363         kvm->mmu_notifier_count--;
 364         spin_unlock(&kvm->mmu_lock);
 365
 366         BUG_ON(kvm->mmu_notifier_count < 0);
 367 }
 368
 369 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 370                                               struct mm_struct *mm,
 371                                               unsigned long address)
 372 {
 373         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 374         int young, idx;
 375
 376         idx = srcu_read_lock(&kvm->srcu);
 377         spin_lock(&kvm->mmu_lock);
 378
 379         young = kvm_age_hva(kvm, address);
 380         if (young)
 381                 kvm_flush_remote_tlbs(kvm);
 382
 383         spin_unlock(&kvm->mmu_lock);
 384         srcu_read_unlock(&kvm->srcu, idx);
 385
 386         return young;
 387 }
 388
 389 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 390                                        struct mm_struct *mm,
 391                                        unsigned long address)
 392 {
 393         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 394         int young, idx;
 395
 396         idx = srcu_read_lock(&kvm->srcu);
 397         spin_lock(&kvm->mmu_lock);
 398         young = kvm_test_age_hva(kvm, address);
 399         spin_unlock(&kvm->mmu_lock);
 400         srcu_read_unlock(&kvm->srcu, idx);
 401
 402         return young;
 403 }
 404
 405 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 406                                      struct mm_struct *mm)
 407 {
 408         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 409         int idx;
 410
 411         idx = srcu_read_lock(&kvm->srcu);
 412         kvm_arch_flush_shadow_all(kvm);
 413         srcu_read_unlock(&kvm->srcu, idx);
 414 }
 415
 416 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 417         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 418         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 419         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 420         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 421         .test_young             = kvm_mmu_notifier_test_young,
 422         .change_pte             = kvm_mmu_notifier_change_pte,
 423         .release                = kvm_mmu_notifier_release,
 424 };
 425
 426 static int kvm_init_mmu_notifier(struct kvm *kvm)
 427 {
 428         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 429         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 430 }
 431
 432 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 433
 434 static int kvm_init_mmu_notifier(struct kvm *kvm)
 435 {
 436         return 0;
 437 }
 438
 439 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 440
 441 static void kvm_init_memslots_id(struct kvm *kvm)
 442 {
 443         int i;
 444         struct kvm_memslots *slots = kvm->memslots;
 445
 446         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 447                 slots->id_to_index[i] = slots->memslots[i].id = i;
 448 }
 449
 450 static struct kvm *kvm_create_vm(unsigned long type)
 451 {
 452         int r, i;
 453         struct kvm *kvm = kvm_arch_alloc_vm();
 454
 455         if (!kvm)
 456                 return ERR_PTR(-ENOMEM);
 457
 458         r = kvm_arch_init_vm(kvm, type);
 459         if (r)
 460                 goto out_err_no_disable;
 461
 462         r = hardware_enable_all();
 463         if (r)
 464                 goto out_err_no_disable;
 465
 466 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 467         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 468 #endif
 469 #ifdef CONFIG_HAVE_KVM_IRQFD
 470         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 471 #endif
 472
 473         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 474
 475         r = -ENOMEM;
 476         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 477         if (!kvm->memslots)
 478                 goto out_err_no_srcu;
 479
 480         /*
 481          * Init kvm generation close to the maximum to easily test the
 482          * code of handling generation number wrap-around.
 483          */
 484         kvm->memslots->generation = -150;
 485
 486         kvm_init_memslots_id(kvm);
 487         if (init_srcu_struct(&kvm->srcu))
 488                 goto out_err_no_srcu;
 489         if (init_srcu_struct(&kvm->irq_srcu))
 490                 goto out_err_no_irq_srcu;
 491         for (i = 0; i < KVM_NR_BUSES; i++) {
 492                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 493                                         GFP_KERNEL);
 494                 if (!kvm->buses[i])
 495                         goto out_err;
 496         }
 497
 498         spin_lock_init(&kvm->mmu_lock);
 499         kvm->mm = current->mm;
 500         atomic_inc(&kvm->mm->mm_count);
 501         kvm_eventfd_init(kvm);
 502         mutex_init(&kvm->lock);
 503         mutex_init(&kvm->irq_lock);
 504         mutex_init(&kvm->slots_lock);
 505         atomic_set(&kvm->users_count, 1);
 506         INIT_LIST_HEAD(&kvm->devices);
 507
 508         r = kvm_init_mmu_notifier(kvm);
 509         if (r)
 510                 goto out_err;
 511
 512         spin_lock(&kvm_lock);
 513         list_add(&kvm->vm_list, &vm_list);
 514         spin_unlock(&kvm_lock);
 515
 516         return kvm;
 517
 518 out_err:
 519         cleanup_srcu_struct(&kvm->irq_srcu);
 520 out_err_no_irq_srcu:
 521         cleanup_srcu_struct(&kvm->srcu);
 522 out_err_no_srcu:
 523         hardware_disable_all();
 524 out_err_no_disable:
 525         for (i = 0; i < KVM_NR_BUSES; i++)
 526                 kfree(kvm->buses[i]);
 527         kfree(kvm->memslots);
 528         kvm_arch_free_vm(kvm);
 529         return ERR_PTR(r);
 530 }
 531
 532 /*
 533  * Avoid using vmalloc for a small buffer.
 534  * Should not be used when the size is statically known.
 535  */
 536 void *kvm_kvzalloc(unsigned long size)
 537 {
 538         if (size > PAGE_SIZE)
 539                 return vzalloc(size);
 540         else
 541                 return kzalloc(size, GFP_KERNEL);
 542 }
 543
 544 void kvm_kvfree(const void *addr)
 545 {
 546         if (is_vmalloc_addr(addr))
 547                 vfree(addr);
 548         else
 549                 kfree(addr);
 550 }
 551
 552 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 553 {
 554         if (!memslot->dirty_bitmap)
 555                 return;
 556
 557         kvm_kvfree(memslot->dirty_bitmap);
 558         memslot->dirty_bitmap = NULL;
 559 }
 560
 561 /*
 562  * Free any memory in @free but not in @dont.
 563  */
 564 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
 565                                   struct kvm_memory_slot *dont)
 566 {
 567         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 568                 kvm_destroy_dirty_bitmap(free);
 569
 570         kvm_arch_free_memslot(kvm, free, dont);
 571
 572         free->npages = 0;
 573 }
 574
 575 static void kvm_free_physmem(struct kvm *kvm)
 576 {
 577         struct kvm_memslots *slots = kvm->memslots;
 578         struct kvm_memory_slot *memslot;
 579
 580         kvm_for_each_memslot(memslot, slots)
 581                 kvm_free_physmem_slot(kvm, memslot, NULL);
 582
 583         kfree(kvm->memslots);
 584 }
 585
 586 static void kvm_destroy_devices(struct kvm *kvm)
 587 {
 588         struct list_head *node, *tmp;
 589
 590         list_for_each_safe(node, tmp, &kvm->devices) {
 591                 struct kvm_device *dev =
 592                         list_entry(node, struct kvm_device, vm_node);
 593
 594                 list_del(node);
 595                 dev->ops->destroy(dev);
 596         }
 597 }
 598
 599 static void kvm_destroy_vm(struct kvm *kvm)
 600 {
 601         int i;
 602         struct mm_struct *mm = kvm->mm;
 603
 604         kvm_arch_sync_events(kvm);
 605         spin_lock(&kvm_lock);
 606         list_del(&kvm->vm_list);
 607         spin_unlock(&kvm_lock);
 608         kvm_free_irq_routing(kvm);
 609         for (i = 0; i < KVM_NR_BUSES; i++)
 610                 kvm_io_bus_destroy(kvm->buses[i]);
 611         kvm_coalesced_mmio_free(kvm);
 612 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 613         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 614 #else
 615         kvm_arch_flush_shadow_all(kvm);
 616 #endif
 617         kvm_arch_destroy_vm(kvm);
 618         kvm_destroy_devices(kvm);
 619         kvm_free_physmem(kvm);
 620         cleanup_srcu_struct(&kvm->irq_srcu);
 621         cleanup_srcu_struct(&kvm->srcu);
 622         kvm_arch_free_vm(kvm);
 623         hardware_disable_all();
 624         mmdrop(mm);
 625 }
 626
 627 void kvm_get_kvm(struct kvm *kvm)
 628 {
 629         atomic_inc(&kvm->users_count);
 630 }
 631 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 632
 633 void kvm_put_kvm(struct kvm *kvm)
 634 {
 635         if (atomic_dec_and_test(&kvm->users_count))
 636                 kvm_destroy_vm(kvm);
 637 }
 638 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 639
 640
 641 static int kvm_vm_release(struct inode *inode, struct file *filp)
 642 {
 643         struct kvm *kvm = filp->private_data;
 644
 645         kvm_irqfd_release(kvm);
 646
 647         kvm_put_kvm(kvm);
 648         return 0;
 649 }
 650
 651 /*
 652  * Allocation size is twice as large as the actual dirty bitmap size.
 653  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 654  */
 655 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 656 {
 657         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 658
 659         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 660         if (!memslot->dirty_bitmap)
 661                 return -ENOMEM;
 662
 663         return 0;
 664 }
 665
 666 static int cmp_memslot(const void *slot1, const void *slot2)
 667 {
 668         struct kvm_memory_slot *s1, *s2;
 669
 670         s1 = (struct kvm_memory_slot *)slot1;
 671         s2 = (struct kvm_memory_slot *)slot2;
 672
 673         if (s1->npages < s2->npages)
 674                 return 1;
 675         if (s1->npages > s2->npages)
 676                 return -1;
 677
 678         return 0;
 679 }
 680
 681 /*
 682  * Sort the memslots base on its size, so the larger slots
 683  * will get better fit.
 684  */
 685 static void sort_memslots(struct kvm_memslots *slots)
 686 {
 687         int i;
 688
 689         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 690               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 691
 692         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 693                 slots->id_to_index[slots->memslots[i].id] = i;
 694 }
 695
 696 static void update_memslots(struct kvm_memslots *slots,
 697                             struct kvm_memory_slot *new)
 698 {
 699         if (new) {
 700                 int id = new->id;
 701                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 702                 unsigned long npages = old->npages;
 703
 704                 *old = *new;
 705                 if (new->npages != npages)
 706                         sort_memslots(slots);
 707         }
 708 }
 709
 710 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 711 {
 712         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 713
 714 #ifdef __KVM_HAVE_READONLY_MEM
 715         valid_flags |= KVM_MEM_READONLY;
 716 #endif
 717
 718         if (mem->flags & ~valid_flags)
 719                 return -EINVAL;
 720
 721         return 0;
 722 }
 723
 724 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 725                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 726 {
 727         struct kvm_memslots *old_memslots = kvm->memslots;
 728
 729         /*
 730          * Set the low bit in the generation, which disables SPTE caching
 731          * until the end of synchronize_srcu_expedited.
 732          */
 733         WARN_ON(old_memslots->generation & 1);
 734         slots->generation = old_memslots->generation + 1;
 735
 736         update_memslots(slots, new);
 737         rcu_assign_pointer(kvm->memslots, slots);
 738         synchronize_srcu_expedited(&kvm->srcu);
 739
 740         /*
 741          * Increment the new memslot generation a second time. This prevents
 742          * vm exits that race with memslot updates from caching a memslot
 743          * generation that will (potentially) be valid forever.
 744          */
 745         slots->generation++;
 746
 747         kvm_arch_memslots_updated(kvm);
 748
 749         return old_memslots;
 750 }
 751
 752 /*
 753  * Allocate some memory and give it an address in the guest physical address
 754  * space.
 755  *
 756  * Discontiguous memory is allowed, mostly for framebuffers.
 757  *
 758  * Must be called holding mmap_sem for write.
 759  */
 760 int __kvm_set_memory_region(struct kvm *kvm,
 761                             struct kvm_userspace_memory_region *mem)
 762 {
 763         int r;
 764         gfn_t base_gfn;
 765         unsigned long npages;
 766         struct kvm_memory_slot *slot;
 767         struct kvm_memory_slot old, new;
 768         struct kvm_memslots *slots = NULL, *old_memslots;
 769         enum kvm_mr_change change;
 770
 771         r = check_memory_region_flags(mem);
 772         if (r)
 773                 goto out;
 774
 775         r = -EINVAL;
 776         /* General sanity checks */
 777         if (mem->memory_size & (PAGE_SIZE - 1))
 778                 goto out;
 779         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 780                 goto out;
 781         /* We can read the guest memory with __xxx_user() later on. */
 782         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 783             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 784              !access_ok(VERIFY_WRITE,
 785                         (void __user *)(unsigned long)mem->userspace_addr,
 786                         mem->memory_size)))
 787                 goto out;
 788         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 789                 goto out;
 790         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 791                 goto out;
 792
 793         slot = id_to_memslot(kvm->memslots, mem->slot);
 794         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 795         npages = mem->memory_size >> PAGE_SHIFT;
 796
 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         if (npages) {
 811                 if (!old.npages)
 812                         change = KVM_MR_CREATE;
 813                 else { /* Modify an existing slot. */
 814                         if ((mem->userspace_addr != old.userspace_addr) ||
 815                             (npages != old.npages) ||
 816                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 817                                 goto out;
 818
 819                         if (base_gfn != old.base_gfn)
 820                                 change = KVM_MR_MOVE;
 821                         else if (new.flags != old.flags)
 822                                 change = KVM_MR_FLAGS_ONLY;
 823                         else { /* Nothing to change. */
 824                                 r = 0;
 825                                 goto out;
 826                         }
 827                 }
 828         } else if (old.npages) {
 829                 change = KVM_MR_DELETE;
 830         } else /* Modify a non-existent slot: disallowed. */
 831                 goto out;
 832
 833         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 834                 /* Check for overlaps */
 835                 r = -EEXIST;
 836                 kvm_for_each_memslot(slot, kvm->memslots) {
 837                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 838                             (slot->id == mem->slot))
 839                                 continue;
 840                         if (!((base_gfn + npages <= slot->base_gfn) ||
 841                               (base_gfn >= slot->base_gfn + slot->npages)))
 842                                 goto out;
 843                 }
 844         }
 845
 846         /* Free page dirty bitmap if unneeded */
 847         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 848                 new.dirty_bitmap = NULL;
 849
 850         r = -ENOMEM;
 851         if (change == KVM_MR_CREATE) {
 852                 new.userspace_addr = mem->userspace_addr;
 853
 854                 if (kvm_arch_create_memslot(kvm, &new, npages))
 855                         goto out_free;
 856         }
 857
 858         /* Allocate page dirty bitmap if needed */
 859         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 860                 if (kvm_create_dirty_bitmap(&new) < 0)
 861                         goto out_free;
 862         }
 863
 864         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 865                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 866                                 GFP_KERNEL);
 867                 if (!slots)
 868                         goto out_free;
 869                 slot = id_to_memslot(slots, mem->slot);
 870                 slot->flags |= KVM_MEMSLOT_INVALID;
 871
 872                 old_memslots = install_new_memslots(kvm, slots, NULL);
 873
 874                 /* slot was deleted or moved, clear iommu mapping */
 875                 kvm_iommu_unmap_pages(kvm, &old);
 876                 /* From this point no new shadow pages pointing to a deleted,
 877                  * or moved, memslot will be created.
 878                  *
 879                  * validation of sp->gfn happens in:
 880                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 881                  *      - kvm_is_visible_gfn (mmu_check_roots)
 882                  */
 883                 kvm_arch_flush_shadow_memslot(kvm, slot);
 884                 slots = old_memslots;
 885         }
 886
 887         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 888         if (r)
 889                 goto out_slots;
 890
 891         r = -ENOMEM;
 892         /*
 893          * We can re-use the old_memslots from above, the only difference
 894          * from the currently installed memslots is the invalid flag.  This
 895          * will get overwritten by update_memslots anyway.
 896          */
 897         if (!slots) {
 898                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 899                                 GFP_KERNEL);
 900                 if (!slots)
 901                         goto out_free;
 902         }
 903
 904         /* actual memory is freed via old in kvm_free_physmem_slot below */
 905         if (change == KVM_MR_DELETE) {
 906                 new.dirty_bitmap = NULL;
 907                 memset(&new.arch, 0, sizeof(new.arch));
 908         }
 909
 910         old_memslots = install_new_memslots(kvm, slots, &new);
 911
 912         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 913
 914         kvm_free_physmem_slot(kvm, &old, &new);
 915         kfree(old_memslots);
 916
 917         /*
 918          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 919          * un-mapped and re-mapped if their base changes.  Since base change
 920          * unmapping is handled above with slot deletion, mapping alone is
 921          * needed here.  Anything else the iommu might care about for existing
 922          * slots (size changes, userspace addr changes and read-only flag
 923          * changes) is disallowed above, so any other attribute changes getting
 924          * here can be skipped.
 925          */
 926         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 927                 r = kvm_iommu_map_pages(kvm, &new);
 928                 return r;
 929         }
 930
 931         return 0;
 932
 933 out_slots:
 934         kfree(slots);
 935 out_free:
 936         kvm_free_physmem_slot(kvm, &new, &old);
 937 out:
 938         return r;
 939 }
 940 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 941
 942 int kvm_set_memory_region(struct kvm *kvm,
 943                           struct kvm_userspace_memory_region *mem)
 944 {
 945         int r;
 946
 947         mutex_lock(&kvm->slots_lock);
 948         r = __kvm_set_memory_region(kvm, mem);
 949         mutex_unlock(&kvm->slots_lock);
 950         return r;
 951 }
 952 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 953
 954 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 955                                           struct kvm_userspace_memory_region *mem)
 956 {
 957         if (mem->slot >= KVM_USER_MEM_SLOTS)
 958                 return -EINVAL;
 959         return kvm_set_memory_region(kvm, mem);
 960 }
 961
 962 int kvm_get_dirty_log(struct kvm *kvm,
 963                         struct kvm_dirty_log *log, int *is_dirty)
 964 {
 965         struct kvm_memory_slot *memslot;
 966         int r, i;
 967         unsigned long n;
 968         unsigned long any = 0;
 969
 970         r = -EINVAL;
 971         if (log->slot >= KVM_USER_MEM_SLOTS)
 972                 goto out;
 973
 974         memslot = id_to_memslot(kvm->memslots, log->slot);
 975         r = -ENOENT;
 976         if (!memslot->dirty_bitmap)
 977                 goto out;
 978
 979         n = kvm_dirty_bitmap_bytes(memslot);
 980
 981         for (i = 0; !any && i < n/sizeof(long); ++i)
 982                 any = memslot->dirty_bitmap[i];
 983
 984         r = -EFAULT;
 985         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 986                 goto out;
 987
 988         if (any)
 989                 *is_dirty = 1;
 990
 991         r = 0;
 992 out:
 993         return r;
 994 }
 995 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
 996
 997 bool kvm_largepages_enabled(void)
 998 {
 999         return largepages_enabled;
1000 }
1001
1002 void kvm_disable_largepages(void)
1003 {
1004         largepages_enabled = false;
1005 }
1006 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1007
1008 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1009 {
1010         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1011 }
1012 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1013
1014 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1015 {
1016         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1017
1018         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1019               memslot->flags & KVM_MEMSLOT_INVALID)
1020                 return 0;
1021
1022         return 1;
1023 }
1024 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1025
1026 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1027 {
1028         struct vm_area_struct *vma;
1029         unsigned long addr, size;
1030
1031         size = PAGE_SIZE;
1032
1033         addr = gfn_to_hva(kvm, gfn);
1034         if (kvm_is_error_hva(addr))
1035                 return PAGE_SIZE;
1036
1037         down_read(&current->mm->mmap_sem);
1038         vma = find_vma(current->mm, addr);
1039         if (!vma)
1040                 goto out;
1041
1042         size = vma_kernel_pagesize(vma);
1043
1044 out:
1045         up_read(&current->mm->mmap_sem);
1046
1047         return size;
1048 }
1049
1050 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1051 {
1052         return slot->flags & KVM_MEM_READONLY;
1053 }
1054
1055 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1056                                        gfn_t *nr_pages, bool write)
1057 {
1058         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1059                 return KVM_HVA_ERR_BAD;
1060
1061         if (memslot_is_readonly(slot) && write)
1062                 return KVM_HVA_ERR_RO_BAD;
1063
1064         if (nr_pages)
1065                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1066
1067         return __gfn_to_hva_memslot(slot, gfn);
1068 }
1069
1070 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1071                                      gfn_t *nr_pages)
1072 {
1073         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1074 }
1075
1076 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1077                                         gfn_t gfn)
1078 {
1079         return gfn_to_hva_many(slot, gfn, NULL);
1080 }
1081 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1082
1083 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1084 {
1085         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1086 }
1087 EXPORT_SYMBOL_GPL(gfn_to_hva);
1088
1089 /*
1090  * If writable is set to false, the hva returned by this function is only
1091  * allowed to be read.
1092  */
1093 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1094 {
1095         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1096         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1097
1098         if (!kvm_is_error_hva(hva) && writable)
1099                 *writable = !memslot_is_readonly(slot);
1100
1101         return hva;
1102 }
1103
1104 static int kvm_read_hva(void *data, void __user *hva, int len)
1105 {
1106         return __copy_from_user(data, hva, len);
1107 }
1108
1109 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1110 {
1111         return __copy_from_user_inatomic(data, hva, len);
1112 }
1113
1114 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1115         unsigned long start, int write, struct page **page)
1116 {
1117         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1118
1119         if (write)
1120                 flags |= FOLL_WRITE;
1121
1122         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1123 }
1124
1125 static inline int check_user_page_hwpoison(unsigned long addr)
1126 {
1127         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1128
1129         rc = __get_user_pages(current, current->mm, addr, 1,
1130                               flags, NULL, NULL, NULL);
1131         return rc == -EHWPOISON;
1132 }
1133
1134 /*
1135  * The atomic path to get the writable pfn which will be stored in @pfn,
1136  * true indicates success, otherwise false is returned.
1137  */
1138 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1139                             bool write_fault, bool *writable, pfn_t *pfn)
1140 {
1141         struct page *page[1];
1142         int npages;
1143
1144         if (!(async || atomic))
1145                 return false;
1146
1147         /*
1148          * Fast pin a writable pfn only if it is a write fault request
1149          * or the caller allows to map a writable pfn for a read fault
1150          * request.
1151          */
1152         if (!(write_fault || writable))
1153                 return false;
1154
1155         npages = __get_user_pages_fast(addr, 1, 1, page);
1156         if (npages == 1) {
1157                 *pfn = page_to_pfn(page[0]);
1158
1159                 if (writable)
1160                         *writable = true;
1161                 return true;
1162         }
1163
1164         return false;
1165 }
1166
1167 /*
1168  * The slow path to get the pfn of the specified host virtual address,
1169  * 1 indicates success, -errno is returned if error is detected.
1170  */
1171 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1172                            bool *writable, pfn_t *pfn)
1173 {
1174         struct page *page[1];
1175         int npages = 0;
1176
1177         might_sleep();
1178
1179         if (writable)
1180                 *writable = write_fault;
1181
1182         if (async) {
1183                 down_read(&current->mm->mmap_sem);
1184                 npages = get_user_page_nowait(current, current->mm,
1185                                               addr, write_fault, page);
1186                 up_read(&current->mm->mmap_sem);
1187         } else
1188                 npages = get_user_pages_fast(addr, 1, write_fault,
1189                                              page);
1190         if (npages != 1)
1191                 return npages;
1192
1193         /* map read fault as writable if possible */
1194         if (unlikely(!write_fault) && writable) {
1195                 struct page *wpage[1];
1196
1197                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1198                 if (npages == 1) {
1199                         *writable = true;
1200                         put_page(page[0]);
1201                         page[0] = wpage[0];
1202                 }
1203
1204                 npages = 1;
1205         }
1206         *pfn = page_to_pfn(page[0]);
1207         return npages;
1208 }
1209
1210 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1211 {
1212         if (unlikely(!(vma->vm_flags & VM_READ)))
1213                 return false;
1214
1215         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1216                 return false;
1217
1218         return true;
1219 }
1220
1221 /*
1222  * Pin guest page in memory and return its pfn.
1223  * @addr: host virtual address which maps memory to the guest
1224  * @atomic: whether this function can sleep
1225  * @async: whether this function need to wait IO complete if the
1226  *         host page is not in the memory
1227  * @write_fault: whether we should get a writable host page
1228  * @writable: whether it allows to map a writable host page for !@write_fault
1229  *
1230  * The function will map a writable host page for these two cases:
1231  * 1): @write_fault = true
1232  * 2): @write_fault = false && @writable, @writable will tell the caller
1233  *     whether the mapping is writable.
1234  */
1235 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1236                         bool write_fault, bool *writable)
1237 {
1238         struct vm_area_struct *vma;
1239         pfn_t pfn = 0;
1240         int npages;
1241
1242         /* we can do it either atomically or asynchronously, not both */
1243         BUG_ON(atomic && async);
1244
1245         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1246                 return pfn;
1247
1248         if (atomic)
1249                 return KVM_PFN_ERR_FAULT;
1250
1251         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1252         if (npages == 1)
1253                 return pfn;
1254
1255         down_read(&current->mm->mmap_sem);
1256         if (npages == -EHWPOISON ||
1257               (!async && check_user_page_hwpoison(addr))) {
1258                 pfn = KVM_PFN_ERR_HWPOISON;
1259                 goto exit;
1260         }
1261
1262         vma = find_vma_intersection(current->mm, addr, addr + 1);
1263
1264         if (vma == NULL)
1265                 pfn = KVM_PFN_ERR_FAULT;
1266         else if ((vma->vm_flags & VM_PFNMAP)) {
1267                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1268                         vma->vm_pgoff;
1269                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1270         } else {
1271                 if (async && vma_is_valid(vma, write_fault))
1272                         *async = true;
1273                 pfn = KVM_PFN_ERR_FAULT;
1274         }
1275 exit:
1276         up_read(&current->mm->mmap_sem);
1277         return pfn;
1278 }
1279
1280 static pfn_t
1281 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1282                      bool *async, bool write_fault, bool *writable)
1283 {
1284         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1285
1286         if (addr == KVM_HVA_ERR_RO_BAD)
1287                 return KVM_PFN_ERR_RO_FAULT;
1288
1289         if (kvm_is_error_hva(addr))
1290                 return KVM_PFN_NOSLOT;
1291
1292         /* Do not map writable pfn in the readonly memslot. */
1293         if (writable && memslot_is_readonly(slot)) {
1294                 *writable = false;
1295                 writable = NULL;
1296         }
1297
1298         return hva_to_pfn(addr, atomic, async, write_fault,
1299                           writable);
1300 }
1301
1302 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1303                           bool write_fault, bool *writable)
1304 {
1305         struct kvm_memory_slot *slot;
1306
1307         if (async)
1308                 *async = false;
1309
1310         slot = gfn_to_memslot(kvm, gfn);
1311
1312         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1313                                     writable);
1314 }
1315
1316 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1317 {
1318         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1319 }
1320 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1321
1322 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1323                        bool write_fault, bool *writable)
1324 {
1325         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1326 }
1327 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1328
1329 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1330 {
1331         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1332 }
1333 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1334
1335 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1336                       bool *writable)
1337 {
1338         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1339 }
1340 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1341
1342 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1343 {
1344         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1345 }
1346
1347 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1348 {
1349         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1350 }
1351 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1352
1353 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1354                                                                   int nr_pages)
1355 {
1356         unsigned long addr;
1357         gfn_t entry;
1358
1359         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1360         if (kvm_is_error_hva(addr))
1361                 return -1;
1362
1363         if (entry < nr_pages)
1364                 return 0;
1365
1366         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1367 }
1368 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1369
1370 static struct page *kvm_pfn_to_page(pfn_t pfn)
1371 {
1372         if (is_error_noslot_pfn(pfn))
1373                 return KVM_ERR_PTR_BAD_PAGE;
1374
1375         if (kvm_is_mmio_pfn(pfn)) {
1376                 WARN_ON(1);
1377                 return KVM_ERR_PTR_BAD_PAGE;
1378         }
1379
1380         return pfn_to_page(pfn);
1381 }
1382
1383 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1384 {
1385         pfn_t pfn;
1386
1387         pfn = gfn_to_pfn(kvm, gfn);
1388
1389         return kvm_pfn_to_page(pfn);
1390 }
1391
1392 EXPORT_SYMBOL_GPL(gfn_to_page);
1393
1394 void kvm_release_page_clean(struct page *page)
1395 {
1396         WARN_ON(is_error_page(page));
1397
1398         kvm_release_pfn_clean(page_to_pfn(page));
1399 }
1400 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1401
1402 void kvm_release_pfn_clean(pfn_t pfn)
1403 {
1404         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1405                 put_page(pfn_to_page(pfn));
1406 }
1407 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1408
1409 void kvm_release_page_dirty(struct page *page)
1410 {
1411         WARN_ON(is_error_page(page));
1412
1413         kvm_release_pfn_dirty(page_to_pfn(page));
1414 }
1415 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1416
1417 static void kvm_release_pfn_dirty(pfn_t pfn)
1418 {
1419         kvm_set_pfn_dirty(pfn);
1420         kvm_release_pfn_clean(pfn);
1421 }
1422
1423 void kvm_set_pfn_dirty(pfn_t pfn)
1424 {
1425         if (!kvm_is_mmio_pfn(pfn)) {
1426                 struct page *page = pfn_to_page(pfn);
1427                 if (!PageReserved(page))
1428                         SetPageDirty(page);
1429         }
1430 }
1431 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1432
1433 void kvm_set_pfn_accessed(pfn_t pfn)
1434 {
1435         if (!kvm_is_mmio_pfn(pfn))
1436                 mark_page_accessed(pfn_to_page(pfn));
1437 }
1438 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1439
1440 void kvm_get_pfn(pfn_t pfn)
1441 {
1442         if (!kvm_is_mmio_pfn(pfn))
1443                 get_page(pfn_to_page(pfn));
1444 }
1445 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1446
1447 static int next_segment(unsigned long len, int offset)
1448 {
1449         if (len > PAGE_SIZE - offset)
1450                 return PAGE_SIZE - offset;
1451         else
1452                 return len;
1453 }
1454
1455 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1456                         int len)
1457 {
1458         int r;
1459         unsigned long addr;
1460
1461         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1462         if (kvm_is_error_hva(addr))
1463                 return -EFAULT;
1464         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1465         if (r)
1466                 return -EFAULT;
1467         return 0;
1468 }
1469 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1470
1471 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1472 {
1473         gfn_t gfn = gpa >> PAGE_SHIFT;
1474         int seg;
1475         int offset = offset_in_page(gpa);
1476         int ret;
1477
1478         while ((seg = next_segment(len, offset)) != 0) {
1479                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1480                 if (ret < 0)
1481                         return ret;
1482                 offset = 0;
1483                 len -= seg;
1484                 data += seg;
1485                 ++gfn;
1486         }
1487         return 0;
1488 }
1489 EXPORT_SYMBOL_GPL(kvm_read_guest);
1490
1491 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1492                           unsigned long len)
1493 {
1494         int r;
1495         unsigned long addr;
1496         gfn_t gfn = gpa >> PAGE_SHIFT;
1497         int offset = offset_in_page(gpa);
1498
1499         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1500         if (kvm_is_error_hva(addr))
1501                 return -EFAULT;
1502         pagefault_disable();
1503         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1504         pagefault_enable();
1505         if (r)
1506                 return -EFAULT;
1507         return 0;
1508 }
1509 EXPORT_SYMBOL(kvm_read_guest_atomic);
1510
1511 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1512                          int offset, int len)
1513 {
1514         int r;
1515         unsigned long addr;
1516
1517         addr = gfn_to_hva(kvm, gfn);
1518         if (kvm_is_error_hva(addr))
1519                 return -EFAULT;
1520         r = __copy_to_user((void __user *)addr + offset, data, len);
1521         if (r)
1522                 return -EFAULT;
1523         mark_page_dirty(kvm, gfn);
1524         return 0;
1525 }
1526 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1527
1528 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1529                     unsigned long len)
1530 {
1531         gfn_t gfn = gpa >> PAGE_SHIFT;
1532         int seg;
1533         int offset = offset_in_page(gpa);
1534         int ret;
1535
1536         while ((seg = next_segment(len, offset)) != 0) {
1537                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1538                 if (ret < 0)
1539                         return ret;
1540                 offset = 0;
1541                 len -= seg;
1542                 data += seg;
1543                 ++gfn;
1544         }
1545         return 0;
1546 }
1547
1548 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1549                               gpa_t gpa, unsigned long len)
1550 {
1551         struct kvm_memslots *slots = kvm_memslots(kvm);
1552         int offset = offset_in_page(gpa);
1553         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1554         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1555         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1556         gfn_t nr_pages_avail;
1557
1558         ghc->gpa = gpa;
1559         ghc->generation = slots->generation;
1560         ghc->len = len;
1561         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1562         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1563         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1564                 ghc->hva += offset;
1565         } else {
1566                 /*
1567                  * If the requested region crosses two memslots, we still
1568                  * verify that the entire region is valid here.
1569                  */
1570                 while (start_gfn <= end_gfn) {
1571                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1572                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1573                                                    &nr_pages_avail);
1574                         if (kvm_is_error_hva(ghc->hva))
1575                                 return -EFAULT;
1576                         start_gfn += nr_pages_avail;
1577                 }
1578                 /* Use the slow path for cross page reads and writes. */
1579                 ghc->memslot = NULL;
1580         }
1581         return 0;
1582 }
1583 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1584
1585 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1586                            void *data, unsigned long len)
1587 {
1588         struct kvm_memslots *slots = kvm_memslots(kvm);
1589         int r;
1590
1591         BUG_ON(len > ghc->len);
1592
1593         if (slots->generation != ghc->generation)
1594                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1595
1596         if (unlikely(!ghc->memslot))
1597                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1598
1599         if (kvm_is_error_hva(ghc->hva))
1600                 return -EFAULT;
1601
1602         r = __copy_to_user((void __user *)ghc->hva, data, len);
1603         if (r)
1604                 return -EFAULT;
1605         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1606
1607         return 0;
1608 }
1609 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1610
1611 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1612                            void *data, unsigned long len)
1613 {
1614         struct kvm_memslots *slots = kvm_memslots(kvm);
1615         int r;
1616
1617         BUG_ON(len > ghc->len);
1618
1619         if (slots->generation != ghc->generation)
1620                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1621
1622         if (unlikely(!ghc->memslot))
1623                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1624
1625         if (kvm_is_error_hva(ghc->hva))
1626                 return -EFAULT;
1627
1628         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1629         if (r)
1630                 return -EFAULT;
1631
1632         return 0;
1633 }
1634 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1635
1636 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1637 {
1638         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1639
1640         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1641 }
1642 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1643
1644 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1645 {
1646         gfn_t gfn = gpa >> PAGE_SHIFT;
1647         int seg;
1648         int offset = offset_in_page(gpa);
1649         int ret;
1650
1651         while ((seg = next_segment(len, offset)) != 0) {
1652                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1653                 if (ret < 0)
1654                         return ret;
1655                 offset = 0;
1656                 len -= seg;
1657                 ++gfn;
1658         }
1659         return 0;
1660 }
1661 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1662
1663 static void mark_page_dirty_in_slot(struct kvm *kvm,
1664                                     struct kvm_memory_slot *memslot,
1665                                     gfn_t gfn)
1666 {
1667         if (memslot && memslot->dirty_bitmap) {
1668                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1669
1670                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1671         }
1672 }
1673
1674 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1675 {
1676         struct kvm_memory_slot *memslot;
1677
1678         memslot = gfn_to_memslot(kvm, gfn);
1679         mark_page_dirty_in_slot(kvm, memslot, gfn);
1680 }
1681 EXPORT_SYMBOL_GPL(mark_page_dirty);
1682
1683 /*
1684  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1685  */
1686 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1687 {
1688         DEFINE_WAIT(wait);
1689
1690         for (;;) {
1691                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1692
1693                 if (kvm_arch_vcpu_runnable(vcpu)) {
1694                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1695                         break;
1696                 }
1697                 if (kvm_cpu_has_pending_timer(vcpu))
1698                         break;
1699                 if (signal_pending(current))
1700                         break;
1701
1702                 schedule();
1703         }
1704
1705         finish_wait(&vcpu->wq, &wait);
1706 }
1707 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1708
1709 #ifndef CONFIG_S390
1710 /*
1711  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1712  */
1713 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1714 {
1715         int me;
1716         int cpu = vcpu->cpu;
1717         wait_queue_head_t *wqp;
1718
1719         wqp = kvm_arch_vcpu_wq(vcpu);
1720         if (waitqueue_active(wqp)) {
1721                 wake_up_interruptible(wqp);
1722                 ++vcpu->stat.halt_wakeup;
1723         }
1724
1725         me = get_cpu();
1726         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1727                 if (kvm_arch_vcpu_should_kick(vcpu))
1728                         smp_send_reschedule(cpu);
1729         put_cpu();
1730 }
1731 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1732 #endif /* !CONFIG_S390 */
1733
1734 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1735 {
1736         struct pid *pid;
1737         struct task_struct *task = NULL;
1738         int ret = 0;
1739
1740         rcu_read_lock();
1741         pid = rcu_dereference(target->pid);
1742         if (pid)
1743                 task = get_pid_task(target->pid, PIDTYPE_PID);
1744         rcu_read_unlock();
1745         if (!task)
1746                 return ret;
1747         if (task->flags & PF_VCPU) {
1748                 put_task_struct(task);
1749                 return ret;
1750         }
1751         ret = yield_to(task, 1);
1752         put_task_struct(task);
1753
1754         return ret;
1755 }
1756 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1757
1758 /*
1759  * Helper that checks whether a VCPU is eligible for directed yield.
1760  * Most eligible candidate to yield is decided by following heuristics:
1761  *
1762  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1763  *  (preempted lock holder), indicated by @in_spin_loop.
1764  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1765  *
1766  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1767  *  chance last time (mostly it has become eligible now since we have probably
1768  *  yielded to lockholder in last iteration. This is done by toggling
1769  *  @dy_eligible each time a VCPU checked for eligibility.)
1770  *
1771  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1772  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1773  *  burning. Giving priority for a potential lock-holder increases lock
1774  *  progress.
1775  *
1776  *  Since algorithm is based on heuristics, accessing another VCPU data without
1777  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1778  *  and continue with next VCPU and so on.
1779  */
1780 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1781 {
1782 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1783         bool eligible;
1784
1785         eligible = !vcpu->spin_loop.in_spin_loop ||
1786                     vcpu->spin_loop.dy_eligible;
1787
1788         if (vcpu->spin_loop.in_spin_loop)
1789                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1790
1791         return eligible;
1792 #else
1793         return true;
1794 #endif
1795 }
1796
1797 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1798 {
1799         struct kvm *kvm = me->kvm;
1800         struct kvm_vcpu *vcpu;
1801         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1802         int yielded = 0;
1803         int try = 3;
1804         int pass;
1805         int i;
1806
1807         kvm_vcpu_set_in_spin_loop(me, true);
1808         /*
1809          * We boost the priority of a VCPU that is runnable but not
1810          * currently running, because it got preempted by something
1811          * else and called schedule in __vcpu_run.  Hopefully that
1812          * VCPU is holding the lock that we need and will release it.
1813          * We approximate round-robin by starting at the last boosted VCPU.
1814          */
1815         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1816                 kvm_for_each_vcpu(i, vcpu, kvm) {
1817                         if (!pass && i <= last_boosted_vcpu) {
1818                                 i = last_boosted_vcpu;
1819                                 continue;
1820                         } else if (pass && i > last_boosted_vcpu)
1821                                 break;
1822                         if (!ACCESS_ONCE(vcpu->preempted))
1823                                 continue;
1824                         if (vcpu == me)
1825                                 continue;
1826                         if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1827                                 continue;
1828                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1829                                 continue;
1830
1831                         yielded = kvm_vcpu_yield_to(vcpu);
1832                         if (yielded > 0) {
1833                                 kvm->last_boosted_vcpu = i;
1834                                 break;
1835                         } else if (yielded < 0) {
1836                                 try--;
1837                                 if (!try)
1838                                         break;
1839                         }
1840                 }
1841         }
1842         kvm_vcpu_set_in_spin_loop(me, false);
1843
1844         /* Ensure vcpu is not eligible during next spinloop */
1845         kvm_vcpu_set_dy_eligible(me, false);
1846 }
1847 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1848
1849 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1850 {
1851         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1852         struct page *page;
1853
1854         if (vmf->pgoff == 0)
1855                 page = virt_to_page(vcpu->run);
1856 #ifdef CONFIG_X86
1857         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1858                 page = virt_to_page(vcpu->arch.pio_data);
1859 #endif
1860 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1861         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1862                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1863 #endif
1864         else
1865                 return kvm_arch_vcpu_fault(vcpu, vmf);
1866         get_page(page);
1867         vmf->page = page;
1868         return 0;
1869 }
1870
1871 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1872         .fault = kvm_vcpu_fault,
1873 };
1874
1875 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1876 {
1877         vma->vm_ops = &kvm_vcpu_vm_ops;
1878         return 0;
1879 }
1880
1881 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1882 {
1883         struct kvm_vcpu *vcpu = filp->private_data;
1884
1885         kvm_put_kvm(vcpu->kvm);
1886         return 0;
1887 }
1888
1889 static struct file_operations kvm_vcpu_fops = {
1890         .release        = kvm_vcpu_release,
1891         .unlocked_ioctl = kvm_vcpu_ioctl,
1892 #ifdef CONFIG_COMPAT
1893         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1894 #endif
1895         .mmap           = kvm_vcpu_mmap,
1896         .llseek         = noop_llseek,
1897 };
1898
1899 /*
1900  * Allocates an inode for the vcpu.
1901  */
1902 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1903 {
1904         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1905 }
1906
1907 /*
1908  * Creates some virtual cpus.  Good luck creating more than one.
1909  */
1910 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1911 {
1912         int r;
1913         struct kvm_vcpu *vcpu, *v;
1914
1915         if (id >= KVM_MAX_VCPUS)
1916                 return -EINVAL;
1917
1918         vcpu = kvm_arch_vcpu_create(kvm, id);
1919         if (IS_ERR(vcpu))
1920                 return PTR_ERR(vcpu);
1921
1922         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1923
1924         r = kvm_arch_vcpu_setup(vcpu);
1925         if (r)
1926                 goto vcpu_destroy;
1927
1928         mutex_lock(&kvm->lock);
1929         if (!kvm_vcpu_compatible(vcpu)) {
1930                 r = -EINVAL;
1931                 goto unlock_vcpu_destroy;
1932         }
1933         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1934                 r = -EINVAL;
1935                 goto unlock_vcpu_destroy;
1936         }
1937
1938         kvm_for_each_vcpu(r, v, kvm)
1939                 if (v->vcpu_id == id) {
1940                         r = -EEXIST;
1941                         goto unlock_vcpu_destroy;
1942                 }
1943
1944         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1945
1946         /* Now it's all set up, let userspace reach it */
1947         kvm_get_kvm(kvm);
1948         r = create_vcpu_fd(vcpu);
1949         if (r < 0) {
1950                 kvm_put_kvm(kvm);
1951                 goto unlock_vcpu_destroy;
1952         }
1953
1954         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1955         smp_wmb();
1956         atomic_inc(&kvm->online_vcpus);
1957
1958         mutex_unlock(&kvm->lock);
1959         kvm_arch_vcpu_postcreate(vcpu);
1960         return r;
1961
1962 unlock_vcpu_destroy:
1963         mutex_unlock(&kvm->lock);
1964 vcpu_destroy:
1965         kvm_arch_vcpu_destroy(vcpu);
1966         return r;
1967 }
1968
1969 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1970 {
1971         if (sigset) {
1972                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1973                 vcpu->sigset_active = 1;
1974                 vcpu->sigset = *sigset;
1975         } else
1976                 vcpu->sigset_active = 0;
1977         return 0;
1978 }
1979
1980 static long kvm_vcpu_ioctl(struct file *filp,
1981                            unsigned int ioctl, unsigned long arg)
1982 {
1983         struct kvm_vcpu *vcpu = filp->private_data;
1984         void __user *argp = (void __user *)arg;
1985         int r;
1986         struct kvm_fpu *fpu = NULL;
1987         struct kvm_sregs *kvm_sregs = NULL;
1988
1989         if (vcpu->kvm->mm != current->mm)
1990                 return -EIO;
1991
1992 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1993         /*
1994          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1995          * so vcpu_load() would break it.
1996          */
1997         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1998                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1999 #endif
2000
2001
2002         r = vcpu_load(vcpu);
2003         if (r)
2004                 return r;
2005         switch (ioctl) {
2006         case KVM_RUN:
2007                 r = -EINVAL;
2008                 if (arg)
2009                         goto out;
2010                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2011                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2012                 break;
2013         case KVM_GET_REGS: {
2014                 struct kvm_regs *kvm_regs;
2015
2016                 r = -ENOMEM;
2017                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2018                 if (!kvm_regs)
2019                         goto out;
2020                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2021                 if (r)
2022                         goto out_free1;
2023                 r = -EFAULT;
2024                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2025                         goto out_free1;
2026                 r = 0;
2027 out_free1:
2028                 kfree(kvm_regs);
2029                 break;
2030         }
2031         case KVM_SET_REGS: {
2032                 struct kvm_regs *kvm_regs;
2033
2034                 r = -ENOMEM;
2035                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2036                 if (IS_ERR(kvm_regs)) {
2037                         r = PTR_ERR(kvm_regs);
2038                         goto out;
2039                 }
2040                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2041                 kfree(kvm_regs);
2042                 break;
2043         }
2044         case KVM_GET_SREGS: {
2045                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2046                 r = -ENOMEM;
2047                 if (!kvm_sregs)
2048                         goto out;
2049                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2050                 if (r)
2051                         goto out;
2052                 r = -EFAULT;
2053                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2054                         goto out;
2055                 r = 0;
2056                 break;
2057         }
2058         case KVM_SET_SREGS: {
2059                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2060                 if (IS_ERR(kvm_sregs)) {
2061                         r = PTR_ERR(kvm_sregs);
2062                         kvm_sregs = NULL;
2063                         goto out;
2064                 }
2065                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2066                 break;
2067         }
2068         case KVM_GET_MP_STATE: {
2069                 struct kvm_mp_state mp_state;
2070
2071                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2072                 if (r)
2073                         goto out;
2074                 r = -EFAULT;
2075                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2076                         goto out;
2077                 r = 0;
2078                 break;
2079         }
2080         case KVM_SET_MP_STATE: {
2081                 struct kvm_mp_state mp_state;
2082
2083                 r = -EFAULT;
2084                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2085                         goto out;
2086                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2087                 break;
2088         }
2089         case KVM_TRANSLATE: {
2090                 struct kvm_translation tr;
2091
2092                 r = -EFAULT;
2093                 if (copy_from_user(&tr, argp, sizeof tr))
2094                         goto out;
2095                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2096                 if (r)
2097                         goto out;
2098                 r = -EFAULT;
2099                 if (copy_to_user(argp, &tr, sizeof tr))
2100                         goto out;
2101                 r = 0;
2102                 break;
2103         }
2104         case KVM_SET_GUEST_DEBUG: {
2105                 struct kvm_guest_debug dbg;
2106
2107                 r = -EFAULT;
2108                 if (copy_from_user(&dbg, argp, sizeof dbg))
2109                         goto out;
2110                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2111                 break;
2112         }
2113         case KVM_SET_SIGNAL_MASK: {
2114                 struct kvm_signal_mask __user *sigmask_arg = argp;
2115                 struct kvm_signal_mask kvm_sigmask;
2116                 sigset_t sigset, *p;
2117
2118                 p = NULL;
2119                 if (argp) {
2120                         r = -EFAULT;
2121                         if (copy_from_user(&kvm_sigmask, argp,
2122                                            sizeof kvm_sigmask))
2123                                 goto out;
2124                         r = -EINVAL;
2125                         if (kvm_sigmask.len != sizeof sigset)
2126                                 goto out;
2127                         r = -EFAULT;
2128                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2129                                            sizeof sigset))
2130                                 goto out;
2131                         p = &sigset;
2132                 }
2133                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2134                 break;
2135         }
2136         case KVM_GET_FPU: {
2137                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2138                 r = -ENOMEM;
2139                 if (!fpu)
2140                         goto out;
2141                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2142                 if (r)
2143                         goto out;
2144                 r = -EFAULT;
2145                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2146                         goto out;
2147                 r = 0;
2148                 break;
2149         }
2150         case KVM_SET_FPU: {
2151                 fpu = memdup_user(argp, sizeof(*fpu));
2152                 if (IS_ERR(fpu)) {
2153                         r = PTR_ERR(fpu);
2154                         fpu = NULL;
2155                         goto out;
2156                 }
2157                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2158                 break;
2159         }
2160         default:
2161                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2162         }
2163 out:
2164         vcpu_put(vcpu);
2165         kfree(fpu);
2166         kfree(kvm_sregs);
2167         return r;
2168 }
2169
2170 #ifdef CONFIG_COMPAT
2171 static long kvm_vcpu_compat_ioctl(struct file *filp,
2172                                   unsigned int ioctl, unsigned long arg)
2173 {
2174         struct kvm_vcpu *vcpu = filp->private_data;
2175         void __user *argp = compat_ptr(arg);
2176         int r;
2177
2178         if (vcpu->kvm->mm != current->mm)
2179                 return -EIO;
2180
2181         switch (ioctl) {
2182         case KVM_SET_SIGNAL_MASK: {
2183                 struct kvm_signal_mask __user *sigmask_arg = argp;
2184                 struct kvm_signal_mask kvm_sigmask;
2185                 compat_sigset_t csigset;
2186                 sigset_t sigset;
2187
2188                 if (argp) {
2189                         r = -EFAULT;
2190                         if (copy_from_user(&kvm_sigmask, argp,
2191                                            sizeof kvm_sigmask))
2192                                 goto out;
2193                         r = -EINVAL;
2194                         if (kvm_sigmask.len != sizeof csigset)
2195                                 goto out;
2196                         r = -EFAULT;
2197                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2198                                            sizeof csigset))
2199                                 goto out;
2200                         sigset_from_compat(&sigset, &csigset);
2201                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2202                 } else
2203                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2204                 break;
2205         }
2206         default:
2207                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2208         }
2209
2210 out:
2211         return r;
2212 }
2213 #endif
2214
2215 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2216                                  int (*accessor)(struct kvm_device *dev,
2217                                                  struct kvm_device_attr *attr),
2218                                  unsigned long arg)
2219 {
2220         struct kvm_device_attr attr;
2221
2222         if (!accessor)
2223                 return -EPERM;
2224
2225         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2226                 return -EFAULT;
2227
2228         return accessor(dev, &attr);
2229 }
2230
2231 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2232                              unsigned long arg)
2233 {
2234         struct kvm_device *dev = filp->private_data;
2235
2236         switch (ioctl) {
2237         case KVM_SET_DEVICE_ATTR:
2238                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2239         case KVM_GET_DEVICE_ATTR:
2240                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2241         case KVM_HAS_DEVICE_ATTR:
2242                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2243         default:
2244                 if (dev->ops->ioctl)
2245                         return dev->ops->ioctl(dev, ioctl, arg);
2246
2247                 return -ENOTTY;
2248         }
2249 }
2250
2251 static int kvm_device_release(struct inode *inode, struct file *filp)
2252 {
2253         struct kvm_device *dev = filp->private_data;
2254         struct kvm *kvm = dev->kvm;
2255
2256         kvm_put_kvm(kvm);
2257         return 0;
2258 }
2259
2260 static const struct file_operations kvm_device_fops = {
2261         .unlocked_ioctl = kvm_device_ioctl,
2262 #ifdef CONFIG_COMPAT
2263         .compat_ioctl = kvm_device_ioctl,
2264 #endif
2265         .release = kvm_device_release,
2266 };
2267
2268 struct kvm_device *kvm_device_from_filp(struct file *filp)
2269 {
2270         if (filp->f_op != &kvm_device_fops)
2271                 return NULL;
2272
2273         return filp->private_data;
2274 }
2275
2276 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2277 #ifdef CONFIG_KVM_MPIC
2278         [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
2279         [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
2280 #endif
2281
2282 #ifdef CONFIG_KVM_XICS
2283         [KVM_DEV_TYPE_XICS]             = &kvm_xics_ops,
2284 #endif
2285 };
2286
2287 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2288 {
2289         if (type >= ARRAY_SIZE(kvm_device_ops_table))
2290                 return -ENOSPC;
2291
2292         if (kvm_device_ops_table[type] != NULL)
2293                 return -EEXIST;
2294
2295         kvm_device_ops_table[type] = ops;
2296         return 0;
2297 }
2298
2299 static int kvm_ioctl_create_device(struct kvm *kvm,
2300                                    struct kvm_create_device *cd)
2301 {
2302         struct kvm_device_ops *ops = NULL;
2303         struct kvm_device *dev;
2304         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2305         int ret;
2306
2307         if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2308                 return -ENODEV;
2309
2310         ops = kvm_device_ops_table[cd->type];
2311         if (ops == NULL)
2312                 return -ENODEV;
2313
2314         if (test)
2315                 return 0;
2316
2317         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2318         if (!dev)
2319                 return -ENOMEM;
2320
2321         dev->ops = ops;
2322         dev->kvm = kvm;
2323
2324         ret = ops->create(dev, cd->type);
2325         if (ret < 0) {
2326                 kfree(dev);
2327                 return ret;
2328         }
2329
2330         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2331         if (ret < 0) {
2332                 ops->destroy(dev);
2333                 return ret;
2334         }
2335
2336         list_add(&dev->vm_node, &kvm->devices);
2337         kvm_get_kvm(kvm);
2338         cd->fd = ret;
2339         return 0;
2340 }
2341
2342 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2343 {
2344         switch (arg) {
2345         case KVM_CAP_USER_MEMORY:
2346         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2347         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2348 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2349         case KVM_CAP_SET_BOOT_CPU_ID:
2350 #endif
2351         case KVM_CAP_INTERNAL_ERROR_DATA:
2352 #ifdef CONFIG_HAVE_KVM_MSI
2353         case KVM_CAP_SIGNAL_MSI:
2354 #endif
2355 #ifdef CONFIG_HAVE_KVM_IRQFD
2356         case KVM_CAP_IRQFD_RESAMPLE:
2357 #endif
2358         case KVM_CAP_CHECK_EXTENSION_VM:
2359                 return 1;
2360 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2361         case KVM_CAP_IRQ_ROUTING:
2362                 return KVM_MAX_IRQ_ROUTES;
2363 #endif
2364         default:
2365                 break;
2366         }
2367         return kvm_vm_ioctl_check_extension(kvm, arg);
2368 }
2369
2370 static long kvm_vm_ioctl(struct file *filp,
2371                            unsigned int ioctl, unsigned long arg)
2372 {
2373         struct kvm *kvm = filp->private_data;
2374         void __user *argp = (void __user *)arg;
2375         int r;
2376
2377         if (kvm->mm != current->mm)
2378                 return -EIO;
2379         switch (ioctl) {
2380         case KVM_CREATE_VCPU:
2381                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2382                 break;
2383         case KVM_SET_USER_MEMORY_REGION: {
2384                 struct kvm_userspace_memory_region kvm_userspace_mem;
2385
2386                 r = -EFAULT;
2387                 if (copy_from_user(&kvm_userspace_mem, argp,
2388                                                 sizeof kvm_userspace_mem))
2389                         goto out;
2390
2391                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2392                 break;
2393         }
2394         case KVM_GET_DIRTY_LOG: {
2395                 struct kvm_dirty_log log;
2396
2397                 r = -EFAULT;
2398                 if (copy_from_user(&log, argp, sizeof log))
2399                         goto out;
2400                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2401                 break;
2402         }
2403 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2404         case KVM_REGISTER_COALESCED_MMIO: {
2405                 struct kvm_coalesced_mmio_zone zone;
2406                 r = -EFAULT;
2407                 if (copy_from_user(&zone, argp, sizeof zone))
2408                         goto out;
2409                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2410                 break;
2411         }
2412         case KVM_UNREGISTER_COALESCED_MMIO: {
2413                 struct kvm_coalesced_mmio_zone zone;
2414                 r = -EFAULT;
2415                 if (copy_from_user(&zone, argp, sizeof zone))
2416                         goto out;
2417                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2418                 break;
2419         }
2420 #endif
2421         case KVM_IRQFD: {
2422                 struct kvm_irqfd data;
2423
2424                 r = -EFAULT;
2425                 if (copy_from_user(&data, argp, sizeof data))
2426                         goto out;
2427                 r = kvm_irqfd(kvm, &data);
2428                 break;
2429         }
2430         case KVM_IOEVENTFD: {
2431                 struct kvm_ioeventfd data;
2432
2433                 r = -EFAULT;
2434                 if (copy_from_user(&data, argp, sizeof data))
2435                         goto out;
2436                 r = kvm_ioeventfd(kvm, &data);
2437                 break;
2438         }
2439 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2440         case KVM_SET_BOOT_CPU_ID:
2441                 r = 0;
2442                 mutex_lock(&kvm->lock);
2443                 if (atomic_read(&kvm->online_vcpus) != 0)
2444                         r = -EBUSY;
2445                 else
2446                         kvm->bsp_vcpu_id = arg;
2447                 mutex_unlock(&kvm->lock);
2448                 break;
2449 #endif
2450 #ifdef CONFIG_HAVE_KVM_MSI
2451         case KVM_SIGNAL_MSI: {
2452                 struct kvm_msi msi;
2453
2454                 r = -EFAULT;
2455                 if (copy_from_user(&msi, argp, sizeof msi))
2456                         goto out;
2457                 r = kvm_send_userspace_msi(kvm, &msi);
2458                 break;
2459         }
2460 #endif
2461 #ifdef __KVM_HAVE_IRQ_LINE
2462         case KVM_IRQ_LINE_STATUS:
2463         case KVM_IRQ_LINE: {
2464                 struct kvm_irq_level irq_event;
2465
2466                 r = -EFAULT;
2467                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2468                         goto out;
2469
2470                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2471                                         ioctl == KVM_IRQ_LINE_STATUS);
2472                 if (r)
2473                         goto out;
2474
2475                 r = -EFAULT;
2476                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2477                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2478                                 goto out;
2479                 }
2480
2481                 r = 0;
2482                 break;
2483         }
2484 #endif
2485 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2486         case KVM_SET_GSI_ROUTING: {
2487                 struct kvm_irq_routing routing;
2488                 struct kvm_irq_routing __user *urouting;
2489                 struct kvm_irq_routing_entry *entries;
2490
2491                 r = -EFAULT;
2492                 if (copy_from_user(&routing, argp, sizeof(routing)))
2493                         goto out;
2494                 r = -EINVAL;
2495                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2496                         goto out;
2497                 if (routing.flags)
2498                         goto out;
2499                 r = -ENOMEM;
2500                 entries = vmalloc(routing.nr * sizeof(*entries));
2501                 if (!entries)
2502                         goto out;
2503                 r = -EFAULT;
2504                 urouting = argp;
2505                 if (copy_from_user(entries, urouting->entries,
2506                                    routing.nr * sizeof(*entries)))
2507                         goto out_free_irq_routing;
2508                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2509                                         routing.flags);
2510         out_free_irq_routing:
2511                 vfree(entries);
2512                 break;
2513         }
2514 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2515         case KVM_CREATE_DEVICE: {
2516                 struct kvm_create_device cd;
2517
2518                 r = -EFAULT;
2519                 if (copy_from_user(&cd, argp, sizeof(cd)))
2520                         goto out;
2521
2522                 r = kvm_ioctl_create_device(kvm, &cd);
2523                 if (r)
2524                         goto out;
2525
2526                 r = -EFAULT;
2527                 if (copy_to_user(argp, &cd, sizeof(cd)))
2528                         goto out;
2529
2530                 r = 0;
2531                 break;
2532         }
2533         case KVM_CHECK_EXTENSION:
2534                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2535                 break;
2536         default:
2537                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2538                 if (r == -ENOTTY)
2539                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2540         }
2541 out:
2542         return r;
2543 }
2544
2545 #ifdef CONFIG_COMPAT
2546 struct compat_kvm_dirty_log {
2547         __u32 slot;
2548         __u32 padding1;
2549         union {
2550                 compat_uptr_t dirty_bitmap; /* one bit per page */
2551                 __u64 padding2;
2552         };
2553 };
2554
2555 static long kvm_vm_compat_ioctl(struct file *filp,
2556                            unsigned int ioctl, unsigned long arg)
2557 {
2558         struct kvm *kvm = filp->private_data;
2559         int r;
2560
2561         if (kvm->mm != current->mm)
2562                 return -EIO;
2563         switch (ioctl) {
2564         case KVM_GET_DIRTY_LOG: {
2565                 struct compat_kvm_dirty_log compat_log;
2566                 struct kvm_dirty_log log;
2567
2568                 r = -EFAULT;
2569                 if (copy_from_user(&compat_log, (void __user *)arg,
2570                                    sizeof(compat_log)))
2571                         goto out;
2572                 log.slot         = compat_log.slot;
2573                 log.padding1     = compat_log.padding1;
2574                 log.padding2     = compat_log.padding2;
2575                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2576
2577                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2578                 break;
2579         }
2580         default:
2581                 r = kvm_vm_ioctl(filp, ioctl, arg);
2582         }
2583
2584 out:
2585         return r;
2586 }
2587 #endif
2588
2589 static struct file_operations kvm_vm_fops = {
2590         .release        = kvm_vm_release,
2591         .unlocked_ioctl = kvm_vm_ioctl,
2592 #ifdef CONFIG_COMPAT
2593         .compat_ioctl   = kvm_vm_compat_ioctl,
2594 #endif
2595         .llseek         = noop_llseek,
2596 };
2597
2598 static int kvm_dev_ioctl_create_vm(unsigned long type)
2599 {
2600         int r;
2601         struct kvm *kvm;
2602
2603         kvm = kvm_create_vm(type);
2604         if (IS_ERR(kvm))
2605                 return PTR_ERR(kvm);
2606 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2607         r = kvm_coalesced_mmio_init(kvm);
2608         if (r < 0) {
2609                 kvm_put_kvm(kvm);
2610                 return r;
2611         }
2612 #endif
2613         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2614         if (r < 0)
2615                 kvm_put_kvm(kvm);
2616
2617         return r;
2618 }
2619
2620 static long kvm_dev_ioctl(struct file *filp,
2621                           unsigned int ioctl, unsigned long arg)
2622 {
2623         long r = -EINVAL;
2624
2625         switch (ioctl) {
2626         case KVM_GET_API_VERSION:
2627                 if (arg)
2628                         goto out;
2629                 r = KVM_API_VERSION;
2630                 break;
2631         case KVM_CREATE_VM:
2632                 r = kvm_dev_ioctl_create_vm(arg);
2633                 break;
2634         case KVM_CHECK_EXTENSION:
2635                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2636                 break;
2637         case KVM_GET_VCPU_MMAP_SIZE:
2638                 if (arg)
2639                         goto out;
2640                 r = PAGE_SIZE;     /* struct kvm_run */
2641 #ifdef CONFIG_X86
2642                 r += PAGE_SIZE;    /* pio data page */
2643 #endif
2644 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2645                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2646 #endif
2647                 break;
2648         case KVM_TRACE_ENABLE:
2649         case KVM_TRACE_PAUSE:
2650         case KVM_TRACE_DISABLE:
2651                 r = -EOPNOTSUPP;
2652                 break;
2653         default:
2654                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2655         }
2656 out:
2657         return r;
2658 }
2659
2660 static struct file_operations kvm_chardev_ops = {
2661         .unlocked_ioctl = kvm_dev_ioctl,
2662         .compat_ioctl   = kvm_dev_ioctl,
2663         .llseek         = noop_llseek,
2664 };
2665
2666 static struct miscdevice kvm_dev = {
2667         KVM_MINOR,
2668         "kvm",
2669         &kvm_chardev_ops,
2670 };
2671
2672 static void hardware_enable_nolock(void *junk)
2673 {
2674         int cpu = raw_smp_processor_id();
2675         int r;
2676
2677         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2678                 return;
2679
2680         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2681
2682         r = kvm_arch_hardware_enable();
2683
2684         if (r) {
2685                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2686                 atomic_inc(&hardware_enable_failed);
2687                 printk(KERN_INFO "kvm: enabling virtualization on "
2688                                  "CPU%d failed\n", cpu);
2689         }
2690 }
2691
2692 static void hardware_enable(void)
2693 {
2694         raw_spin_lock(&kvm_count_lock);
2695         if (kvm_usage_count)
2696                 hardware_enable_nolock(NULL);
2697         raw_spin_unlock(&kvm_count_lock);
2698 }
2699
2700 static void hardware_disable_nolock(void *junk)
2701 {
2702         int cpu = raw_smp_processor_id();
2703
2704         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2705                 return;
2706         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2707         kvm_arch_hardware_disable();
2708 }
2709
2710 static void hardware_disable(void)
2711 {
2712         raw_spin_lock(&kvm_count_lock);
2713         if (kvm_usage_count)
2714                 hardware_disable_nolock(NULL);
2715         raw_spin_unlock(&kvm_count_lock);
2716 }
2717
2718 static void hardware_disable_all_nolock(void)
2719 {
2720         BUG_ON(!kvm_usage_count);
2721
2722         kvm_usage_count--;
2723         if (!kvm_usage_count)
2724                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2725 }
2726
2727 static void hardware_disable_all(void)
2728 {
2729         raw_spin_lock(&kvm_count_lock);
2730         hardware_disable_all_nolock();
2731         raw_spin_unlock(&kvm_count_lock);
2732 }
2733
2734 static int hardware_enable_all(void)
2735 {
2736         int r = 0;
2737
2738         raw_spin_lock(&kvm_count_lock);
2739
2740         kvm_usage_count++;
2741         if (kvm_usage_count == 1) {
2742                 atomic_set(&hardware_enable_failed, 0);
2743                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2744
2745                 if (atomic_read(&hardware_enable_failed)) {
2746                         hardware_disable_all_nolock();
2747                         r = -EBUSY;
2748                 }
2749         }
2750
2751         raw_spin_unlock(&kvm_count_lock);
2752
2753         return r;
2754 }
2755
2756 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2757                            void *v)
2758 {
2759         int cpu = (long)v;
2760
2761         val &= ~CPU_TASKS_FROZEN;
2762         switch (val) {
2763         case CPU_DYING:
2764                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2765                        cpu);
2766                 hardware_disable();
2767                 break;
2768         case CPU_STARTING:
2769                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2770                        cpu);
2771                 hardware_enable();
2772                 break;
2773         }
2774         return NOTIFY_OK;
2775 }
2776
2777 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2778                       void *v)
2779 {
2780         /*
2781          * Some (well, at least mine) BIOSes hang on reboot if
2782          * in vmx root mode.
2783          *
2784          * And Intel TXT required VMX off for all cpu when system shutdown.
2785          */
2786         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2787         kvm_rebooting = true;
2788         on_each_cpu(hardware_disable_nolock, NULL, 1);
2789         return NOTIFY_OK;
2790 }
2791
2792 static struct notifier_block kvm_reboot_notifier = {
2793         .notifier_call = kvm_reboot,
2794         .priority = 0,
2795 };
2796
2797 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2798 {
2799         int i;
2800
2801         for (i = 0; i < bus->dev_count; i++) {
2802                 struct kvm_io_device *pos = bus->range[i].dev;
2803
2804                 kvm_iodevice_destructor(pos);
2805         }
2806         kfree(bus);
2807 }
2808
2809 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2810                                  const struct kvm_io_range *r2)
2811 {
2812         if (r1->addr < r2->addr)
2813                 return -1;
2814         if (r1->addr + r1->len > r2->addr + r2->len)
2815                 return 1;
2816         return 0;
2817 }
2818
2819 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2820 {
2821         return kvm_io_bus_cmp(p1, p2);
2822 }
2823
2824 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2825                           gpa_t addr, int len)
2826 {
2827         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2828                 .addr = addr,
2829                 .len = len,
2830                 .dev = dev,
2831         };
2832
2833         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2834                 kvm_io_bus_sort_cmp, NULL);
2835
2836         return 0;
2837 }
2838
2839 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2840                              gpa_t addr, int len)
2841 {
2842         struct kvm_io_range *range, key;
2843         int off;
2844
2845         key = (struct kvm_io_range) {
2846                 .addr = addr,
2847                 .len = len,
2848         };
2849
2850         range = bsearch(&key, bus->range, bus->dev_count,
2851                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2852         if (range == NULL)
2853                 return -ENOENT;
2854
2855         off = range - bus->range;
2856
2857         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2858                 off--;
2859
2860         return off;
2861 }
2862
2863 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2864                               struct kvm_io_range *range, const void *val)
2865 {
2866         int idx;
2867
2868         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2869         if (idx < 0)
2870                 return -EOPNOTSUPP;
2871
2872         while (idx < bus->dev_count &&
2873                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2874                 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2875                                         range->len, val))
2876                         return idx;
2877                 idx++;
2878         }
2879
2880         return -EOPNOTSUPP;
2881 }
2882
2883 /* kvm_io_bus_write - called under kvm->slots_lock */
2884 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2885                      int len, const void *val)
2886 {
2887         struct kvm_io_bus *bus;
2888         struct kvm_io_range range;
2889         int r;
2890
2891         range = (struct kvm_io_range) {
2892                 .addr = addr,
2893                 .len = len,
2894         };
2895
2896         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2897         r = __kvm_io_bus_write(bus, &range, val);
2898         return r < 0 ? r : 0;
2899 }
2900
2901 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2902 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2903                             int len, const void *val, long cookie)
2904 {
2905         struct kvm_io_bus *bus;
2906         struct kvm_io_range range;
2907
2908         range = (struct kvm_io_range) {
2909                 .addr = addr,
2910                 .len = len,
2911         };
2912
2913         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2914
2915         /* First try the device referenced by cookie. */
2916         if ((cookie >= 0) && (cookie < bus->dev_count) &&
2917             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2918                 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2919                                         val))
2920                         return cookie;
2921
2922         /*
2923          * cookie contained garbage; fall back to search and return the
2924          * correct cookie value.
2925          */
2926         return __kvm_io_bus_write(bus, &range, val);
2927 }
2928
2929 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2930                              void *val)
2931 {
2932         int idx;
2933
2934         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2935         if (idx < 0)
2936                 return -EOPNOTSUPP;
2937
2938         while (idx < bus->dev_count &&
2939                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2940                 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2941                                        range->len, val))
2942                         return idx;
2943                 idx++;
2944         }
2945
2946         return -EOPNOTSUPP;
2947 }
2948 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2949
2950 /* kvm_io_bus_read - called under kvm->slots_lock */
2951 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2952                     int len, void *val)
2953 {
2954         struct kvm_io_bus *bus;
2955         struct kvm_io_range range;
2956         int r;
2957
2958         range = (struct kvm_io_range) {
2959                 .addr = addr,
2960                 .len = len,
2961         };
2962
2963         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2964         r = __kvm_io_bus_read(bus, &range, val);
2965         return r < 0 ? r : 0;
2966 }
2967
2968
2969 /* Caller must hold slots_lock. */
2970 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2971                             int len, struct kvm_io_device *dev)
2972 {
2973         struct kvm_io_bus *new_bus, *bus;
2974
2975         bus = kvm->buses[bus_idx];
2976         /* exclude ioeventfd which is limited by maximum fd */
2977         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2978                 return -ENOSPC;
2979
2980         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2981                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2982         if (!new_bus)
2983                 return -ENOMEM;
2984         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2985                sizeof(struct kvm_io_range)));
2986         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2987         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2988         synchronize_srcu_expedited(&kvm->srcu);
2989         kfree(bus);
2990
2991         return 0;
2992 }
2993
2994 /* Caller must hold slots_lock. */
2995 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2996                               struct kvm_io_device *dev)
2997 {
2998         int i, r;
2999         struct kvm_io_bus *new_bus, *bus;
3000
3001         bus = kvm->buses[bus_idx];
3002         r = -ENOENT;
3003         for (i = 0; i < bus->dev_count; i++)
3004                 if (bus->range[i].dev == dev) {
3005                         r = 0;
3006                         break;
3007                 }
3008
3009         if (r)
3010                 return r;
3011
3012         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3013                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3014         if (!new_bus)
3015                 return -ENOMEM;
3016
3017         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3018         new_bus->dev_count--;
3019         memcpy(new_bus->range + i, bus->range + i + 1,
3020                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3021
3022         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3023         synchronize_srcu_expedited(&kvm->srcu);
3024         kfree(bus);
3025         return r;
3026 }
3027
3028 static struct notifier_block kvm_cpu_notifier = {
3029         .notifier_call = kvm_cpu_hotplug,
3030 };
3031
3032 static int vm_stat_get(void *_offset, u64 *val)
3033 {
3034         unsigned offset = (long)_offset;
3035         struct kvm *kvm;
3036
3037         *val = 0;
3038         spin_lock(&kvm_lock);
3039         list_for_each_entry(kvm, &vm_list, vm_list)
3040                 *val += *(u32 *)((void *)kvm + offset);
3041         spin_unlock(&kvm_lock);
3042         return 0;
3043 }
3044
3045 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3046
3047 static int vcpu_stat_get(void *_offset, u64 *val)
3048 {
3049         unsigned offset = (long)_offset;
3050         struct kvm *kvm;
3051         struct kvm_vcpu *vcpu;
3052         int i;
3053
3054         *val = 0;
3055         spin_lock(&kvm_lock);
3056         list_for_each_entry(kvm, &vm_list, vm_list)
3057                 kvm_for_each_vcpu(i, vcpu, kvm)
3058                         *val += *(u32 *)((void *)vcpu + offset);
3059
3060         spin_unlock(&kvm_lock);
3061         return 0;
3062 }
3063
3064 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3065
3066 static const struct file_operations *stat_fops[] = {
3067         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3068         [KVM_STAT_VM]   = &vm_stat_fops,
3069 };
3070
3071 static int kvm_init_debug(void)
3072 {
3073         int r = -EEXIST;
3074         struct kvm_stats_debugfs_item *p;
3075
3076         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3077         if (kvm_debugfs_dir == NULL)
3078                 goto out;
3079
3080         for (p = debugfs_entries; p->name; ++p) {
3081                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3082                                                 (void *)(long)p->offset,
3083                                                 stat_fops[p->kind]);
3084                 if (p->dentry == NULL)
3085                         goto out_dir;
3086         }
3087
3088         return 0;
3089
3090 out_dir:
3091         debugfs_remove_recursive(kvm_debugfs_dir);
3092 out:
3093         return r;
3094 }
3095
3096 static void kvm_exit_debug(void)
3097 {
3098         struct kvm_stats_debugfs_item *p;
3099
3100         for (p = debugfs_entries; p->name; ++p)
3101                 debugfs_remove(p->dentry);
3102         debugfs_remove(kvm_debugfs_dir);
3103 }
3104
3105 static int kvm_suspend(void)
3106 {
3107         if (kvm_usage_count)
3108                 hardware_disable_nolock(NULL);
3109         return 0;
3110 }
3111
3112 static void kvm_resume(void)
3113 {
3114         if (kvm_usage_count) {
3115                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3116                 hardware_enable_nolock(NULL);
3117         }
3118 }
3119
3120 static struct syscore_ops kvm_syscore_ops = {
3121         .suspend = kvm_suspend,
3122         .resume = kvm_resume,
3123 };
3124
3125 static inline
3126 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3127 {
3128         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3129 }
3130
3131 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3132 {
3133         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3134         if (vcpu->preempted)
3135                 vcpu->preempted = false;
3136
3137         kvm_arch_sched_in(vcpu, cpu);
3138
3139         kvm_arch_vcpu_load(vcpu, cpu);
3140 }
3141
3142 static void kvm_sched_out(struct preempt_notifier *pn,
3143                           struct task_struct *next)
3144 {
3145         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3146
3147         if (current->state == TASK_RUNNING)
3148                 vcpu->preempted = true;
3149         kvm_arch_vcpu_put(vcpu);
3150 }
3151
3152 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3153                   struct module *module)
3154 {
3155         int r;
3156         int cpu;
3157
3158         r = kvm_arch_init(opaque);
3159         if (r)
3160                 goto out_fail;
3161
3162         /*
3163          * kvm_arch_init makes sure there's at most one caller
3164          * for architectures that support multiple implementations,
3165          * like intel and amd on x86.
3166          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3167          * conflicts in case kvm is already setup for another implementation.
3168          */
3169         r = kvm_irqfd_init();
3170         if (r)
3171                 goto out_irqfd;
3172
3173         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3174                 r = -ENOMEM;
3175                 goto out_free_0;
3176         }
3177
3178         r = kvm_arch_hardware_setup();
3179         if (r < 0)
3180                 goto out_free_0a;
3181
3182         for_each_online_cpu(cpu) {
3183                 smp_call_function_single(cpu,
3184                                 kvm_arch_check_processor_compat,
3185                                 &r, 1);
3186                 if (r < 0)
3187                         goto out_free_1;
3188         }
3189
3190         r = register_cpu_notifier(&kvm_cpu_notifier);
3191         if (r)
3192                 goto out_free_2;
3193         register_reboot_notifier(&kvm_reboot_notifier);
3194
3195         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3196         if (!vcpu_align)
3197                 vcpu_align = __alignof__(struct kvm_vcpu);
3198         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3199                                            0, NULL);
3200         if (!kvm_vcpu_cache) {
3201                 r = -ENOMEM;
3202                 goto out_free_3;
3203         }
3204
3205         r = kvm_async_pf_init();
3206         if (r)
3207                 goto out_free;
3208
3209         kvm_chardev_ops.owner = module;
3210         kvm_vm_fops.owner = module;
3211         kvm_vcpu_fops.owner = module;
3212
3213         r = misc_register(&kvm_dev);
3214         if (r) {
3215                 printk(KERN_ERR "kvm: misc device register failed\n");
3216                 goto out_unreg;
3217         }
3218
3219         register_syscore_ops(&kvm_syscore_ops);
3220
3221         kvm_preempt_ops.sched_in = kvm_sched_in;
3222         kvm_preempt_ops.sched_out = kvm_sched_out;
3223
3224         r = kvm_init_debug();
3225         if (r) {
3226                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3227                 goto out_undebugfs;
3228         }
3229
3230         r = kvm_vfio_ops_init();
3231         WARN_ON(r);
3232
3233         return 0;
3234
3235 out_undebugfs:
3236         unregister_syscore_ops(&kvm_syscore_ops);
3237         misc_deregister(&kvm_dev);
3238 out_unreg:
3239         kvm_async_pf_deinit();
3240 out_free:
3241         kmem_cache_destroy(kvm_vcpu_cache);
3242 out_free_3:
3243         unregister_reboot_notifier(&kvm_reboot_notifier);
3244         unregister_cpu_notifier(&kvm_cpu_notifier);
3245 out_free_2:
3246 out_free_1:
3247         kvm_arch_hardware_unsetup();
3248 out_free_0a:
3249         free_cpumask_var(cpus_hardware_enabled);
3250 out_free_0:
3251         kvm_irqfd_exit();
3252 out_irqfd:
3253         kvm_arch_exit();
3254 out_fail:
3255         return r;
3256 }
3257 EXPORT_SYMBOL_GPL(kvm_init);
3258
3259 void kvm_exit(void)
3260 {
3261         kvm_exit_debug();
3262         misc_deregister(&kvm_dev);
3263         kmem_cache_destroy(kvm_vcpu_cache);
3264         kvm_async_pf_deinit();
3265         unregister_syscore_ops(&kvm_syscore_ops);
3266         unregister_reboot_notifier(&kvm_reboot_notifier);
3267         unregister_cpu_notifier(&kvm_cpu_notifier);
3268         on_each_cpu(hardware_disable_nolock, NULL, 1);
3269         kvm_arch_hardware_unsetup();
3270         kvm_arch_exit();
3271         kvm_irqfd_exit();
3272         free_cpumask_var(cpus_hardware_enabled);
3273 }
3274 EXPORT_SYMBOL_GPL(kvm_exit);