virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/syscore_ops.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/pgtable.h>
  57
  58 #include "coalesced_mmio.h"
  59 #include "async_pf.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/kvm.h>
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 /*
  68  * Ordering of locks:
  69  *
  70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  71  */
  72
  73 DEFINE_SPINLOCK(kvm_lock);
  74 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
  75 LIST_HEAD(vm_list);
  76
  77 static cpumask_var_t cpus_hardware_enabled;
  78 static int kvm_usage_count = 0;
  79 static atomic_t hardware_enable_failed;
  80
  81 struct kmem_cache *kvm_vcpu_cache;
  82 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  83
  84 static __read_mostly struct preempt_ops kvm_preempt_ops;
  85
  86 struct dentry *kvm_debugfs_dir;
  87
  88 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  89                            unsigned long arg);
  90 #ifdef CONFIG_COMPAT
  91 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  92                                   unsigned long arg);
  93 #endif
  94 static int hardware_enable_all(void);
  95 static void hardware_disable_all(void);
  96
  97 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  98
  99 bool kvm_rebooting;
 100 EXPORT_SYMBOL_GPL(kvm_rebooting);
 101
 102 static bool largepages_enabled = true;
 103
 104 bool kvm_is_mmio_pfn(pfn_t pfn)
 105 {
 106         if (pfn_valid(pfn))
 107                 return PageReserved(pfn_to_page(pfn));
 108
 109         return true;
 110 }
 111
 112 /*
 113  * Switches to specified vcpu, until a matching vcpu_put()
 114  */
 115 int vcpu_load(struct kvm_vcpu *vcpu)
 116 {
 117         int cpu;
 118
 119         if (mutex_lock_killable(&vcpu->mutex))
 120                 return -EINTR;
 121         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 122                 /* The thread running this VCPU changed. */
 123                 struct pid *oldpid = vcpu->pid;
 124                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 125                 rcu_assign_pointer(vcpu->pid, newpid);
 126                 synchronize_rcu();
 127                 put_pid(oldpid);
 128         }
 129         cpu = get_cpu();
 130         preempt_notifier_register(&vcpu->preempt_notifier);
 131         kvm_arch_vcpu_load(vcpu, cpu);
 132         put_cpu();
 133         return 0;
 134 }
 135
 136 void vcpu_put(struct kvm_vcpu *vcpu)
 137 {
 138         preempt_disable();
 139         kvm_arch_vcpu_put(vcpu);
 140         preempt_notifier_unregister(&vcpu->preempt_notifier);
 141         preempt_enable();
 142         mutex_unlock(&vcpu->mutex);
 143 }
 144
 145 static void ack_flush(void *_completed)
 146 {
 147 }
 148
 149 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 150 {
 151         int i, cpu, me;
 152         cpumask_var_t cpus;
 153         bool called = true;
 154         struct kvm_vcpu *vcpu;
 155
 156         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 157
 158         me = get_cpu();
 159         kvm_for_each_vcpu(i, vcpu, kvm) {
 160                 kvm_make_request(req, vcpu);
 161                 cpu = vcpu->cpu;
 162
 163                 /* Set ->requests bit before we read ->mode */
 164                 smp_mb();
 165
 166                 if (cpus != NULL && cpu != -1 && cpu != me &&
 167                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 168                         cpumask_set_cpu(cpu, cpus);
 169         }
 170         if (unlikely(cpus == NULL))
 171                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 172         else if (!cpumask_empty(cpus))
 173                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 174         else
 175                 called = false;
 176         put_cpu();
 177         free_cpumask_var(cpus);
 178         return called;
 179 }
 180
 181 void kvm_flush_remote_tlbs(struct kvm *kvm)
 182 {
 183         long dirty_count = kvm->tlbs_dirty;
 184
 185         smp_mb();
 186         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 187                 ++kvm->stat.remote_tlb_flush;
 188         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 189 }
 190 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 191
 192 void kvm_reload_remote_mmus(struct kvm *kvm)
 193 {
 194         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 195 }
 196
 197 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 198 {
 199         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 200 }
 201
 202 void kvm_make_scan_ioapic_request(struct kvm *kvm)
 203 {
 204         make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
 205 }
 206
 207 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 208 {
 209         struct page *page;
 210         int r;
 211
 212         mutex_init(&vcpu->mutex);
 213         vcpu->cpu = -1;
 214         vcpu->kvm = kvm;
 215         vcpu->vcpu_id = id;
 216         vcpu->pid = NULL;
 217         init_waitqueue_head(&vcpu->wq);
 218         kvm_async_pf_vcpu_init(vcpu);
 219
 220         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 221         if (!page) {
 222                 r = -ENOMEM;
 223                 goto fail;
 224         }
 225         vcpu->run = page_address(page);
 226
 227         kvm_vcpu_set_in_spin_loop(vcpu, false);
 228         kvm_vcpu_set_dy_eligible(vcpu, false);
 229         vcpu->preempted = false;
 230
 231         r = kvm_arch_vcpu_init(vcpu);
 232         if (r < 0)
 233                 goto fail_free_run;
 234         return 0;
 235
 236 fail_free_run:
 237         free_page((unsigned long)vcpu->run);
 238 fail:
 239         return r;
 240 }
 241 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 242
 243 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 244 {
 245         put_pid(vcpu->pid);
 246         kvm_arch_vcpu_uninit(vcpu);
 247         free_page((unsigned long)vcpu->run);
 248 }
 249 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 250
 251 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 252 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 253 {
 254         return container_of(mn, struct kvm, mmu_notifier);
 255 }
 256
 257 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 258                                              struct mm_struct *mm,
 259                                              unsigned long address)
 260 {
 261         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 262         int need_tlb_flush, idx;
 263
 264         /*
 265          * When ->invalidate_page runs, the linux pte has been zapped
 266          * already but the page is still allocated until
 267          * ->invalidate_page returns. So if we increase the sequence
 268          * here the kvm page fault will notice if the spte can't be
 269          * established because the page is going to be freed. If
 270          * instead the kvm page fault establishes the spte before
 271          * ->invalidate_page runs, kvm_unmap_hva will release it
 272          * before returning.
 273          *
 274          * The sequence increase only need to be seen at spin_unlock
 275          * time, and not at spin_lock time.
 276          *
 277          * Increasing the sequence after the spin_unlock would be
 278          * unsafe because the kvm page fault could then establish the
 279          * pte after kvm_unmap_hva returned, without noticing the page
 280          * is going to be freed.
 281          */
 282         idx = srcu_read_lock(&kvm->srcu);
 283         spin_lock(&kvm->mmu_lock);
 284
 285         kvm->mmu_notifier_seq++;
 286         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 287         /* we've to flush the tlb before the pages can be freed */
 288         if (need_tlb_flush)
 289                 kvm_flush_remote_tlbs(kvm);
 290
 291         spin_unlock(&kvm->mmu_lock);
 292         srcu_read_unlock(&kvm->srcu, idx);
 293 }
 294
 295 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 296                                         struct mm_struct *mm,
 297                                         unsigned long address,
 298                                         pte_t pte)
 299 {
 300         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 301         int idx;
 302
 303         idx = srcu_read_lock(&kvm->srcu);
 304         spin_lock(&kvm->mmu_lock);
 305         kvm->mmu_notifier_seq++;
 306         kvm_set_spte_hva(kvm, address, pte);
 307         spin_unlock(&kvm->mmu_lock);
 308         srcu_read_unlock(&kvm->srcu, idx);
 309 }
 310
 311 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 312                                                     struct mm_struct *mm,
 313                                                     unsigned long start,
 314                                                     unsigned long end)
 315 {
 316         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 317         int need_tlb_flush = 0, idx;
 318
 319         idx = srcu_read_lock(&kvm->srcu);
 320         spin_lock(&kvm->mmu_lock);
 321         /*
 322          * The count increase must become visible at unlock time as no
 323          * spte can be established without taking the mmu_lock and
 324          * count is also read inside the mmu_lock critical section.
 325          */
 326         kvm->mmu_notifier_count++;
 327         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 328         need_tlb_flush |= kvm->tlbs_dirty;
 329         /* we've to flush the tlb before the pages can be freed */
 330         if (need_tlb_flush)
 331                 kvm_flush_remote_tlbs(kvm);
 332
 333         spin_unlock(&kvm->mmu_lock);
 334         srcu_read_unlock(&kvm->srcu, idx);
 335 }
 336
 337 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 338                                                   struct mm_struct *mm,
 339                                                   unsigned long start,
 340                                                   unsigned long end)
 341 {
 342         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 343
 344         spin_lock(&kvm->mmu_lock);
 345         /*
 346          * This sequence increase will notify the kvm page fault that
 347          * the page that is going to be mapped in the spte could have
 348          * been freed.
 349          */
 350         kvm->mmu_notifier_seq++;
 351         smp_wmb();
 352         /*
 353          * The above sequence increase must be visible before the
 354          * below count decrease, which is ensured by the smp_wmb above
 355          * in conjunction with the smp_rmb in mmu_notifier_retry().
 356          */
 357         kvm->mmu_notifier_count--;
 358         spin_unlock(&kvm->mmu_lock);
 359
 360         BUG_ON(kvm->mmu_notifier_count < 0);
 361 }
 362
 363 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 364                                               struct mm_struct *mm,
 365                                               unsigned long address)
 366 {
 367         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 368         int young, idx;
 369
 370         idx = srcu_read_lock(&kvm->srcu);
 371         spin_lock(&kvm->mmu_lock);
 372
 373         young = kvm_age_hva(kvm, address);
 374         if (young)
 375                 kvm_flush_remote_tlbs(kvm);
 376
 377         spin_unlock(&kvm->mmu_lock);
 378         srcu_read_unlock(&kvm->srcu, idx);
 379
 380         return young;
 381 }
 382
 383 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 384                                        struct mm_struct *mm,
 385                                        unsigned long address)
 386 {
 387         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 388         int young, idx;
 389
 390         idx = srcu_read_lock(&kvm->srcu);
 391         spin_lock(&kvm->mmu_lock);
 392         young = kvm_test_age_hva(kvm, address);
 393         spin_unlock(&kvm->mmu_lock);
 394         srcu_read_unlock(&kvm->srcu, idx);
 395
 396         return young;
 397 }
 398
 399 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 400                                      struct mm_struct *mm)
 401 {
 402         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 403         int idx;
 404
 405         idx = srcu_read_lock(&kvm->srcu);
 406         kvm_arch_flush_shadow_all(kvm);
 407         srcu_read_unlock(&kvm->srcu, idx);
 408 }
 409
 410 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 411         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 412         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 413         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 414         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 415         .test_young             = kvm_mmu_notifier_test_young,
 416         .change_pte             = kvm_mmu_notifier_change_pte,
 417         .release                = kvm_mmu_notifier_release,
 418 };
 419
 420 static int kvm_init_mmu_notifier(struct kvm *kvm)
 421 {
 422         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 423         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 424 }
 425
 426 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 427
 428 static int kvm_init_mmu_notifier(struct kvm *kvm)
 429 {
 430         return 0;
 431 }
 432
 433 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 434
 435 static void kvm_init_memslots_id(struct kvm *kvm)
 436 {
 437         int i;
 438         struct kvm_memslots *slots = kvm->memslots;
 439
 440         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 441                 slots->id_to_index[i] = slots->memslots[i].id = i;
 442 }
 443
 444 static struct kvm *kvm_create_vm(unsigned long type)
 445 {
 446         int r, i;
 447         struct kvm *kvm = kvm_arch_alloc_vm();
 448
 449         if (!kvm)
 450                 return ERR_PTR(-ENOMEM);
 451
 452         r = kvm_arch_init_vm(kvm, type);
 453         if (r)
 454                 goto out_err_nodisable;
 455
 456         r = hardware_enable_all();
 457         if (r)
 458                 goto out_err_nodisable;
 459
 460 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 461         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 462         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 463 #endif
 464
 465         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 466
 467         r = -ENOMEM;
 468         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 469         if (!kvm->memslots)
 470                 goto out_err_nosrcu;
 471         kvm_init_memslots_id(kvm);
 472         if (init_srcu_struct(&kvm->srcu))
 473                 goto out_err_nosrcu;
 474         for (i = 0; i < KVM_NR_BUSES; i++) {
 475                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 476                                         GFP_KERNEL);
 477                 if (!kvm->buses[i])
 478                         goto out_err;
 479         }
 480
 481         spin_lock_init(&kvm->mmu_lock);
 482         kvm->mm = current->mm;
 483         atomic_inc(&kvm->mm->mm_count);
 484         kvm_eventfd_init(kvm);
 485         mutex_init(&kvm->lock);
 486         mutex_init(&kvm->irq_lock);
 487         mutex_init(&kvm->slots_lock);
 488         atomic_set(&kvm->users_count, 1);
 489         INIT_LIST_HEAD(&kvm->devices);
 490
 491         r = kvm_init_mmu_notifier(kvm);
 492         if (r)
 493                 goto out_err;
 494
 495         spin_lock(&kvm_lock);
 496         list_add(&kvm->vm_list, &vm_list);
 497         spin_unlock(&kvm_lock);
 498
 499         return kvm;
 500
 501 out_err:
 502         cleanup_srcu_struct(&kvm->srcu);
 503 out_err_nosrcu:
 504         hardware_disable_all();
 505 out_err_nodisable:
 506         for (i = 0; i < KVM_NR_BUSES; i++)
 507                 kfree(kvm->buses[i]);
 508         kfree(kvm->memslots);
 509         kvm_arch_free_vm(kvm);
 510         return ERR_PTR(r);
 511 }
 512
 513 /*
 514  * Avoid using vmalloc for a small buffer.
 515  * Should not be used when the size is statically known.
 516  */
 517 void *kvm_kvzalloc(unsigned long size)
 518 {
 519         if (size > PAGE_SIZE)
 520                 return vzalloc(size);
 521         else
 522                 return kzalloc(size, GFP_KERNEL);
 523 }
 524
 525 void kvm_kvfree(const void *addr)
 526 {
 527         if (is_vmalloc_addr(addr))
 528                 vfree(addr);
 529         else
 530                 kfree(addr);
 531 }
 532
 533 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 534 {
 535         if (!memslot->dirty_bitmap)
 536                 return;
 537
 538         kvm_kvfree(memslot->dirty_bitmap);
 539         memslot->dirty_bitmap = NULL;
 540 }
 541
 542 /*
 543  * Free any memory in @free but not in @dont.
 544  */
 545 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
 546                                   struct kvm_memory_slot *dont)
 547 {
 548         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 549                 kvm_destroy_dirty_bitmap(free);
 550
 551         kvm_arch_free_memslot(kvm, free, dont);
 552
 553         free->npages = 0;
 554 }
 555
 556 void kvm_free_physmem(struct kvm *kvm)
 557 {
 558         struct kvm_memslots *slots = kvm->memslots;
 559         struct kvm_memory_slot *memslot;
 560
 561         kvm_for_each_memslot(memslot, slots)
 562                 kvm_free_physmem_slot(kvm, memslot, NULL);
 563
 564         kfree(kvm->memslots);
 565 }
 566
 567 static void kvm_destroy_devices(struct kvm *kvm)
 568 {
 569         struct list_head *node, *tmp;
 570
 571         list_for_each_safe(node, tmp, &kvm->devices) {
 572                 struct kvm_device *dev =
 573                         list_entry(node, struct kvm_device, vm_node);
 574
 575                 list_del(node);
 576                 dev->ops->destroy(dev);
 577         }
 578 }
 579
 580 static void kvm_destroy_vm(struct kvm *kvm)
 581 {
 582         int i;
 583         struct mm_struct *mm = kvm->mm;
 584
 585         kvm_arch_sync_events(kvm);
 586         spin_lock(&kvm_lock);
 587         list_del(&kvm->vm_list);
 588         spin_unlock(&kvm_lock);
 589         kvm_free_irq_routing(kvm);
 590         for (i = 0; i < KVM_NR_BUSES; i++)
 591                 kvm_io_bus_destroy(kvm->buses[i]);
 592         kvm_coalesced_mmio_free(kvm);
 593 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 594         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 595 #else
 596         kvm_arch_flush_shadow_all(kvm);
 597 #endif
 598         kvm_arch_destroy_vm(kvm);
 599         kvm_destroy_devices(kvm);
 600         kvm_free_physmem(kvm);
 601         cleanup_srcu_struct(&kvm->srcu);
 602         kvm_arch_free_vm(kvm);
 603         hardware_disable_all();
 604         mmdrop(mm);
 605 }
 606
 607 void kvm_get_kvm(struct kvm *kvm)
 608 {
 609         atomic_inc(&kvm->users_count);
 610 }
 611 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 612
 613 void kvm_put_kvm(struct kvm *kvm)
 614 {
 615         if (atomic_dec_and_test(&kvm->users_count))
 616                 kvm_destroy_vm(kvm);
 617 }
 618 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 619
 620
 621 static int kvm_vm_release(struct inode *inode, struct file *filp)
 622 {
 623         struct kvm *kvm = filp->private_data;
 624
 625         kvm_irqfd_release(kvm);
 626
 627         kvm_put_kvm(kvm);
 628         return 0;
 629 }
 630
 631 /*
 632  * Allocation size is twice as large as the actual dirty bitmap size.
 633  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 634  */
 635 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 636 {
 637 #ifndef CONFIG_S390
 638         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 639
 640         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 641         if (!memslot->dirty_bitmap)
 642                 return -ENOMEM;
 643
 644 #endif /* !CONFIG_S390 */
 645         return 0;
 646 }
 647
 648 static int cmp_memslot(const void *slot1, const void *slot2)
 649 {
 650         struct kvm_memory_slot *s1, *s2;
 651
 652         s1 = (struct kvm_memory_slot *)slot1;
 653         s2 = (struct kvm_memory_slot *)slot2;
 654
 655         if (s1->npages < s2->npages)
 656                 return 1;
 657         if (s1->npages > s2->npages)
 658                 return -1;
 659
 660         return 0;
 661 }
 662
 663 /*
 664  * Sort the memslots base on its size, so the larger slots
 665  * will get better fit.
 666  */
 667 static void sort_memslots(struct kvm_memslots *slots)
 668 {
 669         int i;
 670
 671         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 672               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 673
 674         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 675                 slots->id_to_index[slots->memslots[i].id] = i;
 676 }
 677
 678 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
 679                      u64 last_generation)
 680 {
 681         if (new) {
 682                 int id = new->id;
 683                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 684                 unsigned long npages = old->npages;
 685
 686                 *old = *new;
 687                 if (new->npages != npages)
 688                         sort_memslots(slots);
 689         }
 690
 691         slots->generation = last_generation + 1;
 692 }
 693
 694 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 695 {
 696         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 697
 698 #ifdef KVM_CAP_READONLY_MEM
 699         valid_flags |= KVM_MEM_READONLY;
 700 #endif
 701
 702         if (mem->flags & ~valid_flags)
 703                 return -EINVAL;
 704
 705         return 0;
 706 }
 707
 708 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 709                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 710 {
 711         struct kvm_memslots *old_memslots = kvm->memslots;
 712
 713         update_memslots(slots, new, kvm->memslots->generation);
 714         rcu_assign_pointer(kvm->memslots, slots);
 715         synchronize_srcu_expedited(&kvm->srcu);
 716
 717         kvm_arch_memslots_updated(kvm);
 718
 719         return old_memslots;
 720 }
 721
 722 /*
 723  * Allocate some memory and give it an address in the guest physical address
 724  * space.
 725  *
 726  * Discontiguous memory is allowed, mostly for framebuffers.
 727  *
 728  * Must be called holding mmap_sem for write.
 729  */
 730 int __kvm_set_memory_region(struct kvm *kvm,
 731                             struct kvm_userspace_memory_region *mem)
 732 {
 733         int r;
 734         gfn_t base_gfn;
 735         unsigned long npages;
 736         struct kvm_memory_slot *slot;
 737         struct kvm_memory_slot old, new;
 738         struct kvm_memslots *slots = NULL, *old_memslots;
 739         enum kvm_mr_change change;
 740
 741         r = check_memory_region_flags(mem);
 742         if (r)
 743                 goto out;
 744
 745         r = -EINVAL;
 746         /* General sanity checks */
 747         if (mem->memory_size & (PAGE_SIZE - 1))
 748                 goto out;
 749         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 750                 goto out;
 751         /* We can read the guest memory with __xxx_user() later on. */
 752         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 753             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 754              !access_ok(VERIFY_WRITE,
 755                         (void __user *)(unsigned long)mem->userspace_addr,
 756                         mem->memory_size)))
 757                 goto out;
 758         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 759                 goto out;
 760         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 761                 goto out;
 762
 763         slot = id_to_memslot(kvm->memslots, mem->slot);
 764         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 765         npages = mem->memory_size >> PAGE_SHIFT;
 766
 767         r = -EINVAL;
 768         if (npages > KVM_MEM_MAX_NR_PAGES)
 769                 goto out;
 770
 771         if (!npages)
 772                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 773
 774         new = old = *slot;
 775
 776         new.id = mem->slot;
 777         new.base_gfn = base_gfn;
 778         new.npages = npages;
 779         new.flags = mem->flags;
 780
 781         r = -EINVAL;
 782         if (npages) {
 783                 if (!old.npages)
 784                         change = KVM_MR_CREATE;
 785                 else { /* Modify an existing slot. */
 786                         if ((mem->userspace_addr != old.userspace_addr) ||
 787                             (npages != old.npages) ||
 788                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 789                                 goto out;
 790
 791                         if (base_gfn != old.base_gfn)
 792                                 change = KVM_MR_MOVE;
 793                         else if (new.flags != old.flags)
 794                                 change = KVM_MR_FLAGS_ONLY;
 795                         else { /* Nothing to change. */
 796                                 r = 0;
 797                                 goto out;
 798                         }
 799                 }
 800         } else if (old.npages) {
 801                 change = KVM_MR_DELETE;
 802         } else /* Modify a non-existent slot: disallowed. */
 803                 goto out;
 804
 805         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 806                 /* Check for overlaps */
 807                 r = -EEXIST;
 808                 kvm_for_each_memslot(slot, kvm->memslots) {
 809                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 810                             (slot->id == mem->slot))
 811                                 continue;
 812                         if (!((base_gfn + npages <= slot->base_gfn) ||
 813                               (base_gfn >= slot->base_gfn + slot->npages)))
 814                                 goto out;
 815                 }
 816         }
 817
 818         /* Free page dirty bitmap if unneeded */
 819         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 820                 new.dirty_bitmap = NULL;
 821
 822         r = -ENOMEM;
 823         if (change == KVM_MR_CREATE) {
 824                 new.userspace_addr = mem->userspace_addr;
 825
 826                 if (kvm_arch_create_memslot(kvm, &new, npages))
 827                         goto out_free;
 828         }
 829
 830         /* Allocate page dirty bitmap if needed */
 831         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 832                 if (kvm_create_dirty_bitmap(&new) < 0)
 833                         goto out_free;
 834         }
 835
 836         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 837                 r = -ENOMEM;
 838                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 839                                 GFP_KERNEL);
 840                 if (!slots)
 841                         goto out_free;
 842                 slot = id_to_memslot(slots, mem->slot);
 843                 slot->flags |= KVM_MEMSLOT_INVALID;
 844
 845                 old_memslots = install_new_memslots(kvm, slots, NULL);
 846
 847                 /* slot was deleted or moved, clear iommu mapping */
 848                 kvm_iommu_unmap_pages(kvm, &old);
 849                 /* From this point no new shadow pages pointing to a deleted,
 850                  * or moved, memslot will be created.
 851                  *
 852                  * validation of sp->gfn happens in:
 853                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 854                  *      - kvm_is_visible_gfn (mmu_check_roots)
 855                  */
 856                 kvm_arch_flush_shadow_memslot(kvm, slot);
 857                 slots = old_memslots;
 858         }
 859
 860         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 861         if (r)
 862                 goto out_slots;
 863
 864         r = -ENOMEM;
 865         /*
 866          * We can re-use the old_memslots from above, the only difference
 867          * from the currently installed memslots is the invalid flag.  This
 868          * will get overwritten by update_memslots anyway.
 869          */
 870         if (!slots) {
 871                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 872                                 GFP_KERNEL);
 873                 if (!slots)
 874                         goto out_free;
 875         }
 876
 877         /*
 878          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 879          * un-mapped and re-mapped if their base changes.  Since base change
 880          * unmapping is handled above with slot deletion, mapping alone is
 881          * needed here.  Anything else the iommu might care about for existing
 882          * slots (size changes, userspace addr changes and read-only flag
 883          * changes) is disallowed above, so any other attribute changes getting
 884          * here can be skipped.
 885          */
 886         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 887                 r = kvm_iommu_map_pages(kvm, &new);
 888                 if (r)
 889                         goto out_slots;
 890         }
 891
 892         /* actual memory is freed via old in kvm_free_physmem_slot below */
 893         if (change == KVM_MR_DELETE) {
 894                 new.dirty_bitmap = NULL;
 895                 memset(&new.arch, 0, sizeof(new.arch));
 896         }
 897
 898         old_memslots = install_new_memslots(kvm, slots, &new);
 899
 900         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 901
 902         kvm_free_physmem_slot(kvm, &old, &new);
 903         kfree(old_memslots);
 904
 905         return 0;
 906
 907 out_slots:
 908         kfree(slots);
 909 out_free:
 910         kvm_free_physmem_slot(kvm, &new, &old);
 911 out:
 912         return r;
 913 }
 914 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 915
 916 int kvm_set_memory_region(struct kvm *kvm,
 917                           struct kvm_userspace_memory_region *mem)
 918 {
 919         int r;
 920
 921         mutex_lock(&kvm->slots_lock);
 922         r = __kvm_set_memory_region(kvm, mem);
 923         mutex_unlock(&kvm->slots_lock);
 924         return r;
 925 }
 926 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 927
 928 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 929                                    struct kvm_userspace_memory_region *mem)
 930 {
 931         if (mem->slot >= KVM_USER_MEM_SLOTS)
 932                 return -EINVAL;
 933         return kvm_set_memory_region(kvm, mem);
 934 }
 935
 936 int kvm_get_dirty_log(struct kvm *kvm,
 937                         struct kvm_dirty_log *log, int *is_dirty)
 938 {
 939         struct kvm_memory_slot *memslot;
 940         int r, i;
 941         unsigned long n;
 942         unsigned long any = 0;
 943
 944         r = -EINVAL;
 945         if (log->slot >= KVM_USER_MEM_SLOTS)
 946                 goto out;
 947
 948         memslot = id_to_memslot(kvm->memslots, log->slot);
 949         r = -ENOENT;
 950         if (!memslot->dirty_bitmap)
 951                 goto out;
 952
 953         n = kvm_dirty_bitmap_bytes(memslot);
 954
 955         for (i = 0; !any && i < n/sizeof(long); ++i)
 956                 any = memslot->dirty_bitmap[i];
 957
 958         r = -EFAULT;
 959         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 960                 goto out;
 961
 962         if (any)
 963                 *is_dirty = 1;
 964
 965         r = 0;
 966 out:
 967         return r;
 968 }
 969 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
 970
 971 bool kvm_largepages_enabled(void)
 972 {
 973         return largepages_enabled;
 974 }
 975
 976 void kvm_disable_largepages(void)
 977 {
 978         largepages_enabled = false;
 979 }
 980 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 981
 982 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 983 {
 984         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 985 }
 986 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 987
 988 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 989 {
 990         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 991
 992         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
 993               memslot->flags & KVM_MEMSLOT_INVALID)
 994                 return 0;
 995
 996         return 1;
 997 }
 998 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 999
1000 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1001 {
1002         struct vm_area_struct *vma;
1003         unsigned long addr, size;
1004
1005         size = PAGE_SIZE;
1006
1007         addr = gfn_to_hva(kvm, gfn);
1008         if (kvm_is_error_hva(addr))
1009                 return PAGE_SIZE;
1010
1011         down_read(&current->mm->mmap_sem);
1012         vma = find_vma(current->mm, addr);
1013         if (!vma)
1014                 goto out;
1015
1016         size = vma_kernel_pagesize(vma);
1017
1018 out:
1019         up_read(&current->mm->mmap_sem);
1020
1021         return size;
1022 }
1023
1024 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1025 {
1026         return slot->flags & KVM_MEM_READONLY;
1027 }
1028
1029 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1030                                        gfn_t *nr_pages, bool write)
1031 {
1032         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1033                 return KVM_HVA_ERR_BAD;
1034
1035         if (memslot_is_readonly(slot) && write)
1036                 return KVM_HVA_ERR_RO_BAD;
1037
1038         if (nr_pages)
1039                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1040
1041         return __gfn_to_hva_memslot(slot, gfn);
1042 }
1043
1044 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1045                                      gfn_t *nr_pages)
1046 {
1047         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1048 }
1049
1050 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1051                                  gfn_t gfn)
1052 {
1053         return gfn_to_hva_many(slot, gfn, NULL);
1054 }
1055 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1056
1057 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1058 {
1059         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1060 }
1061 EXPORT_SYMBOL_GPL(gfn_to_hva);
1062
1063 /*
1064  * If writable is set to false, the hva returned by this function is only
1065  * allowed to be read.
1066  */
1067 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1068 {
1069         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1070         if (writable)
1071                 *writable = !memslot_is_readonly(slot);
1072
1073         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1074 }
1075
1076 static int kvm_read_hva(void *data, void __user *hva, int len)
1077 {
1078         return __copy_from_user(data, hva, len);
1079 }
1080
1081 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1082 {
1083         return __copy_from_user_inatomic(data, hva, len);
1084 }
1085
1086 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1087         unsigned long start, int write, struct page **page)
1088 {
1089         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1090
1091         if (write)
1092                 flags |= FOLL_WRITE;
1093
1094         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1095 }
1096
1097 static inline int check_user_page_hwpoison(unsigned long addr)
1098 {
1099         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1100
1101         rc = __get_user_pages(current, current->mm, addr, 1,
1102                               flags, NULL, NULL, NULL);
1103         return rc == -EHWPOISON;
1104 }
1105
1106 /*
1107  * The atomic path to get the writable pfn which will be stored in @pfn,
1108  * true indicates success, otherwise false is returned.
1109  */
1110 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1111                             bool write_fault, bool *writable, pfn_t *pfn)
1112 {
1113         struct page *page[1];
1114         int npages;
1115
1116         if (!(async || atomic))
1117                 return false;
1118
1119         /*
1120          * Fast pin a writable pfn only if it is a write fault request
1121          * or the caller allows to map a writable pfn for a read fault
1122          * request.
1123          */
1124         if (!(write_fault || writable))
1125                 return false;
1126
1127         npages = __get_user_pages_fast(addr, 1, 1, page);
1128         if (npages == 1) {
1129                 *pfn = page_to_pfn(page[0]);
1130
1131                 if (writable)
1132                         *writable = true;
1133                 return true;
1134         }
1135
1136         return false;
1137 }
1138
1139 /*
1140  * The slow path to get the pfn of the specified host virtual address,
1141  * 1 indicates success, -errno is returned if error is detected.
1142  */
1143 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1144                            bool *writable, pfn_t *pfn)
1145 {
1146         struct page *page[1];
1147         int npages = 0;
1148
1149         might_sleep();
1150
1151         if (writable)
1152                 *writable = write_fault;
1153
1154         if (async) {
1155                 down_read(&current->mm->mmap_sem);
1156                 npages = get_user_page_nowait(current, current->mm,
1157                                               addr, write_fault, page);
1158                 up_read(&current->mm->mmap_sem);
1159         } else
1160                 npages = get_user_pages_fast(addr, 1, write_fault,
1161                                              page);
1162         if (npages != 1)
1163                 return npages;
1164
1165         /* map read fault as writable if possible */
1166         if (unlikely(!write_fault) && writable) {
1167                 struct page *wpage[1];
1168
1169                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1170                 if (npages == 1) {
1171                         *writable = true;
1172                         put_page(page[0]);
1173                         page[0] = wpage[0];
1174                 }
1175
1176                 npages = 1;
1177         }
1178         *pfn = page_to_pfn(page[0]);
1179         return npages;
1180 }
1181
1182 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1183 {
1184         if (unlikely(!(vma->vm_flags & VM_READ)))
1185                 return false;
1186
1187         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1188                 return false;
1189
1190         return true;
1191 }
1192
1193 /*
1194  * Pin guest page in memory and return its pfn.
1195  * @addr: host virtual address which maps memory to the guest
1196  * @atomic: whether this function can sleep
1197  * @async: whether this function need to wait IO complete if the
1198  *         host page is not in the memory
1199  * @write_fault: whether we should get a writable host page
1200  * @writable: whether it allows to map a writable host page for !@write_fault
1201  *
1202  * The function will map a writable host page for these two cases:
1203  * 1): @write_fault = true
1204  * 2): @write_fault = false && @writable, @writable will tell the caller
1205  *     whether the mapping is writable.
1206  */
1207 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1208                         bool write_fault, bool *writable)
1209 {
1210         struct vm_area_struct *vma;
1211         pfn_t pfn = 0;
1212         int npages;
1213
1214         /* we can do it either atomically or asynchronously, not both */
1215         BUG_ON(atomic && async);
1216
1217         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1218                 return pfn;
1219
1220         if (atomic)
1221                 return KVM_PFN_ERR_FAULT;
1222
1223         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1224         if (npages == 1)
1225                 return pfn;
1226
1227         down_read(&current->mm->mmap_sem);
1228         if (npages == -EHWPOISON ||
1229               (!async && check_user_page_hwpoison(addr))) {
1230                 pfn = KVM_PFN_ERR_HWPOISON;
1231                 goto exit;
1232         }
1233
1234         vma = find_vma_intersection(current->mm, addr, addr + 1);
1235
1236         if (vma == NULL)
1237                 pfn = KVM_PFN_ERR_FAULT;
1238         else if ((vma->vm_flags & VM_PFNMAP)) {
1239                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1240                         vma->vm_pgoff;
1241                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1242         } else {
1243                 if (async && vma_is_valid(vma, write_fault))
1244                         *async = true;
1245                 pfn = KVM_PFN_ERR_FAULT;
1246         }
1247 exit:
1248         up_read(&current->mm->mmap_sem);
1249         return pfn;
1250 }
1251
1252 static pfn_t
1253 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1254                      bool *async, bool write_fault, bool *writable)
1255 {
1256         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1257
1258         if (addr == KVM_HVA_ERR_RO_BAD)
1259                 return KVM_PFN_ERR_RO_FAULT;
1260
1261         if (kvm_is_error_hva(addr))
1262                 return KVM_PFN_NOSLOT;
1263
1264         /* Do not map writable pfn in the readonly memslot. */
1265         if (writable && memslot_is_readonly(slot)) {
1266                 *writable = false;
1267                 writable = NULL;
1268         }
1269
1270         return hva_to_pfn(addr, atomic, async, write_fault,
1271                           writable);
1272 }
1273
1274 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1275                           bool write_fault, bool *writable)
1276 {
1277         struct kvm_memory_slot *slot;
1278
1279         if (async)
1280                 *async = false;
1281
1282         slot = gfn_to_memslot(kvm, gfn);
1283
1284         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1285                                     writable);
1286 }
1287
1288 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1289 {
1290         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1291 }
1292 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1293
1294 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1295                        bool write_fault, bool *writable)
1296 {
1297         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1298 }
1299 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1300
1301 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1302 {
1303         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1304 }
1305 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1306
1307 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1308                       bool *writable)
1309 {
1310         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1311 }
1312 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1313
1314 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1315 {
1316         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1317 }
1318
1319 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1320 {
1321         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1322 }
1323 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1324
1325 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1326                                                                   int nr_pages)
1327 {
1328         unsigned long addr;
1329         gfn_t entry;
1330
1331         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1332         if (kvm_is_error_hva(addr))
1333                 return -1;
1334
1335         if (entry < nr_pages)
1336                 return 0;
1337
1338         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1339 }
1340 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1341
1342 static struct page *kvm_pfn_to_page(pfn_t pfn)
1343 {
1344         if (is_error_noslot_pfn(pfn))
1345                 return KVM_ERR_PTR_BAD_PAGE;
1346
1347         if (kvm_is_mmio_pfn(pfn)) {
1348                 WARN_ON(1);
1349                 return KVM_ERR_PTR_BAD_PAGE;
1350         }
1351
1352         return pfn_to_page(pfn);
1353 }
1354
1355 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1356 {
1357         pfn_t pfn;
1358
1359         pfn = gfn_to_pfn(kvm, gfn);
1360
1361         return kvm_pfn_to_page(pfn);
1362 }
1363
1364 EXPORT_SYMBOL_GPL(gfn_to_page);
1365
1366 void kvm_release_page_clean(struct page *page)
1367 {
1368         WARN_ON(is_error_page(page));
1369
1370         kvm_release_pfn_clean(page_to_pfn(page));
1371 }
1372 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1373
1374 void kvm_release_pfn_clean(pfn_t pfn)
1375 {
1376         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1377                 put_page(pfn_to_page(pfn));
1378 }
1379 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1380
1381 void kvm_release_page_dirty(struct page *page)
1382 {
1383         WARN_ON(is_error_page(page));
1384
1385         kvm_release_pfn_dirty(page_to_pfn(page));
1386 }
1387 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1388
1389 void kvm_release_pfn_dirty(pfn_t pfn)
1390 {
1391         kvm_set_pfn_dirty(pfn);
1392         kvm_release_pfn_clean(pfn);
1393 }
1394 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1395
1396 void kvm_set_page_dirty(struct page *page)
1397 {
1398         kvm_set_pfn_dirty(page_to_pfn(page));
1399 }
1400 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1401
1402 void kvm_set_pfn_dirty(pfn_t pfn)
1403 {
1404         if (!kvm_is_mmio_pfn(pfn)) {
1405                 struct page *page = pfn_to_page(pfn);
1406                 if (!PageReserved(page))
1407                         SetPageDirty(page);
1408         }
1409 }
1410 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1411
1412 void kvm_set_pfn_accessed(pfn_t pfn)
1413 {
1414         if (!kvm_is_mmio_pfn(pfn))
1415                 mark_page_accessed(pfn_to_page(pfn));
1416 }
1417 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1418
1419 void kvm_get_pfn(pfn_t pfn)
1420 {
1421         if (!kvm_is_mmio_pfn(pfn))
1422                 get_page(pfn_to_page(pfn));
1423 }
1424 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1425
1426 static int next_segment(unsigned long len, int offset)
1427 {
1428         if (len > PAGE_SIZE - offset)
1429                 return PAGE_SIZE - offset;
1430         else
1431                 return len;
1432 }
1433
1434 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1435                         int len)
1436 {
1437         int r;
1438         unsigned long addr;
1439
1440         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1441         if (kvm_is_error_hva(addr))
1442                 return -EFAULT;
1443         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1444         if (r)
1445                 return -EFAULT;
1446         return 0;
1447 }
1448 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1449
1450 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1451 {
1452         gfn_t gfn = gpa >> PAGE_SHIFT;
1453         int seg;
1454         int offset = offset_in_page(gpa);
1455         int ret;
1456
1457         while ((seg = next_segment(len, offset)) != 0) {
1458                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1459                 if (ret < 0)
1460                         return ret;
1461                 offset = 0;
1462                 len -= seg;
1463                 data += seg;
1464                 ++gfn;
1465         }
1466         return 0;
1467 }
1468 EXPORT_SYMBOL_GPL(kvm_read_guest);
1469
1470 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1471                           unsigned long len)
1472 {
1473         int r;
1474         unsigned long addr;
1475         gfn_t gfn = gpa >> PAGE_SHIFT;
1476         int offset = offset_in_page(gpa);
1477
1478         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1479         if (kvm_is_error_hva(addr))
1480                 return -EFAULT;
1481         pagefault_disable();
1482         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1483         pagefault_enable();
1484         if (r)
1485                 return -EFAULT;
1486         return 0;
1487 }
1488 EXPORT_SYMBOL(kvm_read_guest_atomic);
1489
1490 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1491                          int offset, int len)
1492 {
1493         int r;
1494         unsigned long addr;
1495
1496         addr = gfn_to_hva(kvm, gfn);
1497         if (kvm_is_error_hva(addr))
1498                 return -EFAULT;
1499         r = __copy_to_user((void __user *)addr + offset, data, len);
1500         if (r)
1501                 return -EFAULT;
1502         mark_page_dirty(kvm, gfn);
1503         return 0;
1504 }
1505 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1506
1507 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1508                     unsigned long len)
1509 {
1510         gfn_t gfn = gpa >> PAGE_SHIFT;
1511         int seg;
1512         int offset = offset_in_page(gpa);
1513         int ret;
1514
1515         while ((seg = next_segment(len, offset)) != 0) {
1516                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1517                 if (ret < 0)
1518                         return ret;
1519                 offset = 0;
1520                 len -= seg;
1521                 data += seg;
1522                 ++gfn;
1523         }
1524         return 0;
1525 }
1526
1527 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1528                               gpa_t gpa, unsigned long len)
1529 {
1530         struct kvm_memslots *slots = kvm_memslots(kvm);
1531         int offset = offset_in_page(gpa);
1532         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1533         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1534         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1535         gfn_t nr_pages_avail;
1536
1537         ghc->gpa = gpa;
1538         ghc->generation = slots->generation;
1539         ghc->len = len;
1540         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1541         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1542         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1543                 ghc->hva += offset;
1544         } else {
1545                 /*
1546                  * If the requested region crosses two memslots, we still
1547                  * verify that the entire region is valid here.
1548                  */
1549                 while (start_gfn <= end_gfn) {
1550                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1551                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1552                                                    &nr_pages_avail);
1553                         if (kvm_is_error_hva(ghc->hva))
1554                                 return -EFAULT;
1555                         start_gfn += nr_pages_avail;
1556                 }
1557                 /* Use the slow path for cross page reads and writes. */
1558                 ghc->memslot = NULL;
1559         }
1560         return 0;
1561 }
1562 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1563
1564 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1565                            void *data, unsigned long len)
1566 {
1567         struct kvm_memslots *slots = kvm_memslots(kvm);
1568         int r;
1569
1570         BUG_ON(len > ghc->len);
1571
1572         if (slots->generation != ghc->generation)
1573                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1574
1575         if (unlikely(!ghc->memslot))
1576                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1577
1578         if (kvm_is_error_hva(ghc->hva))
1579                 return -EFAULT;
1580
1581         r = __copy_to_user((void __user *)ghc->hva, data, len);
1582         if (r)
1583                 return -EFAULT;
1584         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1585
1586         return 0;
1587 }
1588 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1589
1590 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1591                            void *data, unsigned long len)
1592 {
1593         struct kvm_memslots *slots = kvm_memslots(kvm);
1594         int r;
1595
1596         BUG_ON(len > ghc->len);
1597
1598         if (slots->generation != ghc->generation)
1599                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1600
1601         if (unlikely(!ghc->memslot))
1602                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1603
1604         if (kvm_is_error_hva(ghc->hva))
1605                 return -EFAULT;
1606
1607         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1608         if (r)
1609                 return -EFAULT;
1610
1611         return 0;
1612 }
1613 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1614
1615 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1616 {
1617         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1618                                     offset, len);
1619 }
1620 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1621
1622 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1623 {
1624         gfn_t gfn = gpa >> PAGE_SHIFT;
1625         int seg;
1626         int offset = offset_in_page(gpa);
1627         int ret;
1628
1629         while ((seg = next_segment(len, offset)) != 0) {
1630                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1631                 if (ret < 0)
1632                         return ret;
1633                 offset = 0;
1634                 len -= seg;
1635                 ++gfn;
1636         }
1637         return 0;
1638 }
1639 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1640
1641 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1642                              gfn_t gfn)
1643 {
1644         if (memslot && memslot->dirty_bitmap) {
1645                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1646
1647                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1648         }
1649 }
1650
1651 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1652 {
1653         struct kvm_memory_slot *memslot;
1654
1655         memslot = gfn_to_memslot(kvm, gfn);
1656         mark_page_dirty_in_slot(kvm, memslot, gfn);
1657 }
1658 EXPORT_SYMBOL_GPL(mark_page_dirty);
1659
1660 /*
1661  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1662  */
1663 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1664 {
1665         DEFINE_WAIT(wait);
1666
1667         for (;;) {
1668                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1669
1670                 if (kvm_arch_vcpu_runnable(vcpu)) {
1671                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1672                         break;
1673                 }
1674                 if (kvm_cpu_has_pending_timer(vcpu))
1675                         break;
1676                 if (signal_pending(current))
1677                         break;
1678
1679                 schedule();
1680         }
1681
1682         finish_wait(&vcpu->wq, &wait);
1683 }
1684 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1685
1686 #ifndef CONFIG_S390
1687 /*
1688  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1689  */
1690 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1691 {
1692         int me;
1693         int cpu = vcpu->cpu;
1694         wait_queue_head_t *wqp;
1695
1696         wqp = kvm_arch_vcpu_wq(vcpu);
1697         if (waitqueue_active(wqp)) {
1698                 wake_up_interruptible(wqp);
1699                 ++vcpu->stat.halt_wakeup;
1700         }
1701
1702         me = get_cpu();
1703         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1704                 if (kvm_arch_vcpu_should_kick(vcpu))
1705                         smp_send_reschedule(cpu);
1706         put_cpu();
1707 }
1708 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1709 #endif /* !CONFIG_S390 */
1710
1711 void kvm_resched(struct kvm_vcpu *vcpu)
1712 {
1713         if (!need_resched())
1714                 return;
1715         cond_resched();
1716 }
1717 EXPORT_SYMBOL_GPL(kvm_resched);
1718
1719 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1720 {
1721         struct pid *pid;
1722         struct task_struct *task = NULL;
1723         bool ret = false;
1724
1725         rcu_read_lock();
1726         pid = rcu_dereference(target->pid);
1727         if (pid)
1728                 task = get_pid_task(target->pid, PIDTYPE_PID);
1729         rcu_read_unlock();
1730         if (!task)
1731                 return ret;
1732         if (task->flags & PF_VCPU) {
1733                 put_task_struct(task);
1734                 return ret;
1735         }
1736         ret = yield_to(task, 1);
1737         put_task_struct(task);
1738
1739         return ret;
1740 }
1741 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1742
1743 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1744 /*
1745  * Helper that checks whether a VCPU is eligible for directed yield.
1746  * Most eligible candidate to yield is decided by following heuristics:
1747  *
1748  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1749  *  (preempted lock holder), indicated by @in_spin_loop.
1750  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1751  *
1752  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1753  *  chance last time (mostly it has become eligible now since we have probably
1754  *  yielded to lockholder in last iteration. This is done by toggling
1755  *  @dy_eligible each time a VCPU checked for eligibility.)
1756  *
1757  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1758  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1759  *  burning. Giving priority for a potential lock-holder increases lock
1760  *  progress.
1761  *
1762  *  Since algorithm is based on heuristics, accessing another VCPU data without
1763  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1764  *  and continue with next VCPU and so on.
1765  */
1766 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1767 {
1768         bool eligible;
1769
1770         eligible = !vcpu->spin_loop.in_spin_loop ||
1771                         (vcpu->spin_loop.in_spin_loop &&
1772                          vcpu->spin_loop.dy_eligible);
1773
1774         if (vcpu->spin_loop.in_spin_loop)
1775                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1776
1777         return eligible;
1778 }
1779 #endif
1780
1781 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1782 {
1783         struct kvm *kvm = me->kvm;
1784         struct kvm_vcpu *vcpu;
1785         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1786         int yielded = 0;
1787         int try = 3;
1788         int pass;
1789         int i;
1790
1791         kvm_vcpu_set_in_spin_loop(me, true);
1792         /*
1793          * We boost the priority of a VCPU that is runnable but not
1794          * currently running, because it got preempted by something
1795          * else and called schedule in __vcpu_run.  Hopefully that
1796          * VCPU is holding the lock that we need and will release it.
1797          * We approximate round-robin by starting at the last boosted VCPU.
1798          */
1799         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1800                 kvm_for_each_vcpu(i, vcpu, kvm) {
1801                         if (!pass && i <= last_boosted_vcpu) {
1802                                 i = last_boosted_vcpu;
1803                                 continue;
1804                         } else if (pass && i > last_boosted_vcpu)
1805                                 break;
1806                         if (!ACCESS_ONCE(vcpu->preempted))
1807                                 continue;
1808                         if (vcpu == me)
1809                                 continue;
1810                         if (waitqueue_active(&vcpu->wq))
1811                                 continue;
1812                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1813                                 continue;
1814
1815                         yielded = kvm_vcpu_yield_to(vcpu);
1816                         if (yielded > 0) {
1817                                 kvm->last_boosted_vcpu = i;
1818                                 break;
1819                         } else if (yielded < 0) {
1820                                 try--;
1821                                 if (!try)
1822                                         break;
1823                         }
1824                 }
1825         }
1826         kvm_vcpu_set_in_spin_loop(me, false);
1827
1828         /* Ensure vcpu is not eligible during next spinloop */
1829         kvm_vcpu_set_dy_eligible(me, false);
1830 }
1831 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1832
1833 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1834 {
1835         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1836         struct page *page;
1837
1838         if (vmf->pgoff == 0)
1839                 page = virt_to_page(vcpu->run);
1840 #ifdef CONFIG_X86
1841         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1842                 page = virt_to_page(vcpu->arch.pio_data);
1843 #endif
1844 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1845         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1846                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1847 #endif
1848         else
1849                 return kvm_arch_vcpu_fault(vcpu, vmf);
1850         get_page(page);
1851         vmf->page = page;
1852         return 0;
1853 }
1854
1855 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1856         .fault = kvm_vcpu_fault,
1857 };
1858
1859 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1860 {
1861         vma->vm_ops = &kvm_vcpu_vm_ops;
1862         return 0;
1863 }
1864
1865 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1866 {
1867         struct kvm_vcpu *vcpu = filp->private_data;
1868
1869         kvm_put_kvm(vcpu->kvm);
1870         return 0;
1871 }
1872
1873 static struct file_operations kvm_vcpu_fops = {
1874         .release        = kvm_vcpu_release,
1875         .unlocked_ioctl = kvm_vcpu_ioctl,
1876 #ifdef CONFIG_COMPAT
1877         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1878 #endif
1879         .mmap           = kvm_vcpu_mmap,
1880         .llseek         = noop_llseek,
1881 };
1882
1883 /*
1884  * Allocates an inode for the vcpu.
1885  */
1886 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1887 {
1888         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1889 }
1890
1891 /*
1892  * Creates some virtual cpus.  Good luck creating more than one.
1893  */
1894 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1895 {
1896         int r;
1897         struct kvm_vcpu *vcpu, *v;
1898
1899         vcpu = kvm_arch_vcpu_create(kvm, id);
1900         if (IS_ERR(vcpu))
1901                 return PTR_ERR(vcpu);
1902
1903         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1904
1905         r = kvm_arch_vcpu_setup(vcpu);
1906         if (r)
1907                 goto vcpu_destroy;
1908
1909         mutex_lock(&kvm->lock);
1910         if (!kvm_vcpu_compatible(vcpu)) {
1911                 r = -EINVAL;
1912                 goto unlock_vcpu_destroy;
1913         }
1914         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1915                 r = -EINVAL;
1916                 goto unlock_vcpu_destroy;
1917         }
1918
1919         kvm_for_each_vcpu(r, v, kvm)
1920                 if (v->vcpu_id == id) {
1921                         r = -EEXIST;
1922                         goto unlock_vcpu_destroy;
1923                 }
1924
1925         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1926
1927         /* Now it's all set up, let userspace reach it */
1928         kvm_get_kvm(kvm);
1929         r = create_vcpu_fd(vcpu);
1930         if (r < 0) {
1931                 kvm_put_kvm(kvm);
1932                 goto unlock_vcpu_destroy;
1933         }
1934
1935         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1936         smp_wmb();
1937         atomic_inc(&kvm->online_vcpus);
1938
1939         mutex_unlock(&kvm->lock);
1940         kvm_arch_vcpu_postcreate(vcpu);
1941         return r;
1942
1943 unlock_vcpu_destroy:
1944         mutex_unlock(&kvm->lock);
1945 vcpu_destroy:
1946         kvm_arch_vcpu_destroy(vcpu);
1947         return r;
1948 }
1949
1950 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1951 {
1952         if (sigset) {
1953                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1954                 vcpu->sigset_active = 1;
1955                 vcpu->sigset = *sigset;
1956         } else
1957                 vcpu->sigset_active = 0;
1958         return 0;
1959 }
1960
1961 static long kvm_vcpu_ioctl(struct file *filp,
1962                            unsigned int ioctl, unsigned long arg)
1963 {
1964         struct kvm_vcpu *vcpu = filp->private_data;
1965         void __user *argp = (void __user *)arg;
1966         int r;
1967         struct kvm_fpu *fpu = NULL;
1968         struct kvm_sregs *kvm_sregs = NULL;
1969
1970         if (vcpu->kvm->mm != current->mm)
1971                 return -EIO;
1972
1973 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1974         /*
1975          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1976          * so vcpu_load() would break it.
1977          */
1978         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1979                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1980 #endif
1981
1982
1983         r = vcpu_load(vcpu);
1984         if (r)
1985                 return r;
1986         switch (ioctl) {
1987         case KVM_RUN:
1988                 r = -EINVAL;
1989                 if (arg)
1990                         goto out;
1991                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1992                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1993                 break;
1994         case KVM_GET_REGS: {
1995                 struct kvm_regs *kvm_regs;
1996
1997                 r = -ENOMEM;
1998                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1999                 if (!kvm_regs)
2000                         goto out;
2001                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2002                 if (r)
2003                         goto out_free1;
2004                 r = -EFAULT;
2005                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2006                         goto out_free1;
2007                 r = 0;
2008 out_free1:
2009                 kfree(kvm_regs);
2010                 break;
2011         }
2012         case KVM_SET_REGS: {
2013                 struct kvm_regs *kvm_regs;
2014
2015                 r = -ENOMEM;
2016                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2017                 if (IS_ERR(kvm_regs)) {
2018                         r = PTR_ERR(kvm_regs);
2019                         goto out;
2020                 }
2021                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2022                 kfree(kvm_regs);
2023                 break;
2024         }
2025         case KVM_GET_SREGS: {
2026                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2027                 r = -ENOMEM;
2028                 if (!kvm_sregs)
2029                         goto out;
2030                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2031                 if (r)
2032                         goto out;
2033                 r = -EFAULT;
2034                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2035                         goto out;
2036                 r = 0;
2037                 break;
2038         }
2039         case KVM_SET_SREGS: {
2040                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2041                 if (IS_ERR(kvm_sregs)) {
2042                         r = PTR_ERR(kvm_sregs);
2043                         kvm_sregs = NULL;
2044                         goto out;
2045                 }
2046                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2047                 break;
2048         }
2049         case KVM_GET_MP_STATE: {
2050                 struct kvm_mp_state mp_state;
2051
2052                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2053                 if (r)
2054                         goto out;
2055                 r = -EFAULT;
2056                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2057                         goto out;
2058                 r = 0;
2059                 break;
2060         }
2061         case KVM_SET_MP_STATE: {
2062                 struct kvm_mp_state mp_state;
2063
2064                 r = -EFAULT;
2065                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2066                         goto out;
2067                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2068                 break;
2069         }
2070         case KVM_TRANSLATE: {
2071                 struct kvm_translation tr;
2072
2073                 r = -EFAULT;
2074                 if (copy_from_user(&tr, argp, sizeof tr))
2075                         goto out;
2076                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2077                 if (r)
2078                         goto out;
2079                 r = -EFAULT;
2080                 if (copy_to_user(argp, &tr, sizeof tr))
2081                         goto out;
2082                 r = 0;
2083                 break;
2084         }
2085         case KVM_SET_GUEST_DEBUG: {
2086                 struct kvm_guest_debug dbg;
2087
2088                 r = -EFAULT;
2089                 if (copy_from_user(&dbg, argp, sizeof dbg))
2090                         goto out;
2091                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2092                 break;
2093         }
2094         case KVM_SET_SIGNAL_MASK: {
2095                 struct kvm_signal_mask __user *sigmask_arg = argp;
2096                 struct kvm_signal_mask kvm_sigmask;
2097                 sigset_t sigset, *p;
2098
2099                 p = NULL;
2100                 if (argp) {
2101                         r = -EFAULT;
2102                         if (copy_from_user(&kvm_sigmask, argp,
2103                                            sizeof kvm_sigmask))
2104                                 goto out;
2105                         r = -EINVAL;
2106                         if (kvm_sigmask.len != sizeof sigset)
2107                                 goto out;
2108                         r = -EFAULT;
2109                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2110                                            sizeof sigset))
2111                                 goto out;
2112                         p = &sigset;
2113                 }
2114                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2115                 break;
2116         }
2117         case KVM_GET_FPU: {
2118                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2119                 r = -ENOMEM;
2120                 if (!fpu)
2121                         goto out;
2122                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2123                 if (r)
2124                         goto out;
2125                 r = -EFAULT;
2126                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2127                         goto out;
2128                 r = 0;
2129                 break;
2130         }
2131         case KVM_SET_FPU: {
2132                 fpu = memdup_user(argp, sizeof(*fpu));
2133                 if (IS_ERR(fpu)) {
2134                         r = PTR_ERR(fpu);
2135                         fpu = NULL;
2136                         goto out;
2137                 }
2138                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2139                 break;
2140         }
2141         default:
2142                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2143         }
2144 out:
2145         vcpu_put(vcpu);
2146         kfree(fpu);
2147         kfree(kvm_sregs);
2148         return r;
2149 }
2150
2151 #ifdef CONFIG_COMPAT
2152 static long kvm_vcpu_compat_ioctl(struct file *filp,
2153                                   unsigned int ioctl, unsigned long arg)
2154 {
2155         struct kvm_vcpu *vcpu = filp->private_data;
2156         void __user *argp = compat_ptr(arg);
2157         int r;
2158
2159         if (vcpu->kvm->mm != current->mm)
2160                 return -EIO;
2161
2162         switch (ioctl) {
2163         case KVM_SET_SIGNAL_MASK: {
2164                 struct kvm_signal_mask __user *sigmask_arg = argp;
2165                 struct kvm_signal_mask kvm_sigmask;
2166                 compat_sigset_t csigset;
2167                 sigset_t sigset;
2168
2169                 if (argp) {
2170                         r = -EFAULT;
2171                         if (copy_from_user(&kvm_sigmask, argp,
2172                                            sizeof kvm_sigmask))
2173                                 goto out;
2174                         r = -EINVAL;
2175                         if (kvm_sigmask.len != sizeof csigset)
2176                                 goto out;
2177                         r = -EFAULT;
2178                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2179                                            sizeof csigset))
2180                                 goto out;
2181                         sigset_from_compat(&sigset, &csigset);
2182                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2183                 } else
2184                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2185                 break;
2186         }
2187         default:
2188                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2189         }
2190
2191 out:
2192         return r;
2193 }
2194 #endif
2195
2196 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2197                                  int (*accessor)(struct kvm_device *dev,
2198                                                  struct kvm_device_attr *attr),
2199                                  unsigned long arg)
2200 {
2201         struct kvm_device_attr attr;
2202
2203         if (!accessor)
2204                 return -EPERM;
2205
2206         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2207                 return -EFAULT;
2208
2209         return accessor(dev, &attr);
2210 }
2211
2212 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2213                              unsigned long arg)
2214 {
2215         struct kvm_device *dev = filp->private_data;
2216
2217         switch (ioctl) {
2218         case KVM_SET_DEVICE_ATTR:
2219                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2220         case KVM_GET_DEVICE_ATTR:
2221                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2222         case KVM_HAS_DEVICE_ATTR:
2223                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2224         default:
2225                 if (dev->ops->ioctl)
2226                         return dev->ops->ioctl(dev, ioctl, arg);
2227
2228                 return -ENOTTY;
2229         }
2230 }
2231
2232 static int kvm_device_release(struct inode *inode, struct file *filp)
2233 {
2234         struct kvm_device *dev = filp->private_data;
2235         struct kvm *kvm = dev->kvm;
2236
2237         kvm_put_kvm(kvm);
2238         return 0;
2239 }
2240
2241 static const struct file_operations kvm_device_fops = {
2242         .unlocked_ioctl = kvm_device_ioctl,
2243 #ifdef CONFIG_COMPAT
2244         .compat_ioctl = kvm_device_ioctl,
2245 #endif
2246         .release = kvm_device_release,
2247 };
2248
2249 struct kvm_device *kvm_device_from_filp(struct file *filp)
2250 {
2251         if (filp->f_op != &kvm_device_fops)
2252                 return NULL;
2253
2254         return filp->private_data;
2255 }
2256
2257 static int kvm_ioctl_create_device(struct kvm *kvm,
2258                                    struct kvm_create_device *cd)
2259 {
2260         struct kvm_device_ops *ops = NULL;
2261         struct kvm_device *dev;
2262         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2263         int ret;
2264
2265         switch (cd->type) {
2266 #ifdef CONFIG_KVM_MPIC
2267         case KVM_DEV_TYPE_FSL_MPIC_20:
2268         case KVM_DEV_TYPE_FSL_MPIC_42:
2269                 ops = &kvm_mpic_ops;
2270                 break;
2271 #endif
2272 #ifdef CONFIG_KVM_XICS
2273         case KVM_DEV_TYPE_XICS:
2274                 ops = &kvm_xics_ops;
2275                 break;
2276 #endif
2277         default:
2278                 return -ENODEV;
2279         }
2280
2281         if (test)
2282                 return 0;
2283
2284         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2285         if (!dev)
2286                 return -ENOMEM;
2287
2288         dev->ops = ops;
2289         dev->kvm = kvm;
2290
2291         ret = ops->create(dev, cd->type);
2292         if (ret < 0) {
2293                 kfree(dev);
2294                 return ret;
2295         }
2296
2297         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2298         if (ret < 0) {
2299                 ops->destroy(dev);
2300                 return ret;
2301         }
2302
2303         list_add(&dev->vm_node, &kvm->devices);
2304         kvm_get_kvm(kvm);
2305         cd->fd = ret;
2306         return 0;
2307 }
2308
2309 static long kvm_vm_ioctl(struct file *filp,
2310                            unsigned int ioctl, unsigned long arg)
2311 {
2312         struct kvm *kvm = filp->private_data;
2313         void __user *argp = (void __user *)arg;
2314         int r;
2315
2316         if (kvm->mm != current->mm)
2317                 return -EIO;
2318         switch (ioctl) {
2319         case KVM_CREATE_VCPU:
2320                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2321                 break;
2322         case KVM_SET_USER_MEMORY_REGION: {
2323                 struct kvm_userspace_memory_region kvm_userspace_mem;
2324
2325                 r = -EFAULT;
2326                 if (copy_from_user(&kvm_userspace_mem, argp,
2327                                                 sizeof kvm_userspace_mem))
2328                         goto out;
2329
2330                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2331                 break;
2332         }
2333         case KVM_GET_DIRTY_LOG: {
2334                 struct kvm_dirty_log log;
2335
2336                 r = -EFAULT;
2337                 if (copy_from_user(&log, argp, sizeof log))
2338                         goto out;
2339                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2340                 break;
2341         }
2342 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2343         case KVM_REGISTER_COALESCED_MMIO: {
2344                 struct kvm_coalesced_mmio_zone zone;
2345                 r = -EFAULT;
2346                 if (copy_from_user(&zone, argp, sizeof zone))
2347                         goto out;
2348                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2349                 break;
2350         }
2351         case KVM_UNREGISTER_COALESCED_MMIO: {
2352                 struct kvm_coalesced_mmio_zone zone;
2353                 r = -EFAULT;
2354                 if (copy_from_user(&zone, argp, sizeof zone))
2355                         goto out;
2356                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2357                 break;
2358         }
2359 #endif
2360         case KVM_IRQFD: {
2361                 struct kvm_irqfd data;
2362
2363                 r = -EFAULT;
2364                 if (copy_from_user(&data, argp, sizeof data))
2365                         goto out;
2366                 r = kvm_irqfd(kvm, &data);
2367                 break;
2368         }
2369         case KVM_IOEVENTFD: {
2370                 struct kvm_ioeventfd data;
2371
2372                 r = -EFAULT;
2373                 if (copy_from_user(&data, argp, sizeof data))
2374                         goto out;
2375                 r = kvm_ioeventfd(kvm, &data);
2376                 break;
2377         }
2378 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2379         case KVM_SET_BOOT_CPU_ID:
2380                 r = 0;
2381                 mutex_lock(&kvm->lock);
2382                 if (atomic_read(&kvm->online_vcpus) != 0)
2383                         r = -EBUSY;
2384                 else
2385                         kvm->bsp_vcpu_id = arg;
2386                 mutex_unlock(&kvm->lock);
2387                 break;
2388 #endif
2389 #ifdef CONFIG_HAVE_KVM_MSI
2390         case KVM_SIGNAL_MSI: {
2391                 struct kvm_msi msi;
2392
2393                 r = -EFAULT;
2394                 if (copy_from_user(&msi, argp, sizeof msi))
2395                         goto out;
2396                 r = kvm_send_userspace_msi(kvm, &msi);
2397                 break;
2398         }
2399 #endif
2400 #ifdef __KVM_HAVE_IRQ_LINE
2401         case KVM_IRQ_LINE_STATUS:
2402         case KVM_IRQ_LINE: {
2403                 struct kvm_irq_level irq_event;
2404
2405                 r = -EFAULT;
2406                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2407                         goto out;
2408
2409                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2410                                         ioctl == KVM_IRQ_LINE_STATUS);
2411                 if (r)
2412                         goto out;
2413
2414                 r = -EFAULT;
2415                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2416                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2417                                 goto out;
2418                 }
2419
2420                 r = 0;
2421                 break;
2422         }
2423 #endif
2424 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2425         case KVM_SET_GSI_ROUTING: {
2426                 struct kvm_irq_routing routing;
2427                 struct kvm_irq_routing __user *urouting;
2428                 struct kvm_irq_routing_entry *entries;
2429
2430                 r = -EFAULT;
2431                 if (copy_from_user(&routing, argp, sizeof(routing)))
2432                         goto out;
2433                 r = -EINVAL;
2434                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2435                         goto out;
2436                 if (routing.flags)
2437                         goto out;
2438                 r = -ENOMEM;
2439                 entries = vmalloc(routing.nr * sizeof(*entries));
2440                 if (!entries)
2441                         goto out;
2442                 r = -EFAULT;
2443                 urouting = argp;
2444                 if (copy_from_user(entries, urouting->entries,
2445                                    routing.nr * sizeof(*entries)))
2446                         goto out_free_irq_routing;
2447                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2448                                         routing.flags);
2449         out_free_irq_routing:
2450                 vfree(entries);
2451                 break;
2452         }
2453 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2454         case KVM_CREATE_DEVICE: {
2455                 struct kvm_create_device cd;
2456
2457                 r = -EFAULT;
2458                 if (copy_from_user(&cd, argp, sizeof(cd)))
2459                         goto out;
2460
2461                 r = kvm_ioctl_create_device(kvm, &cd);
2462                 if (r)
2463                         goto out;
2464
2465                 r = -EFAULT;
2466                 if (copy_to_user(argp, &cd, sizeof(cd)))
2467                         goto out;
2468
2469                 r = 0;
2470                 break;
2471         }
2472         default:
2473                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2474                 if (r == -ENOTTY)
2475                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2476         }
2477 out:
2478         return r;
2479 }
2480
2481 #ifdef CONFIG_COMPAT
2482 struct compat_kvm_dirty_log {
2483         __u32 slot;
2484         __u32 padding1;
2485         union {
2486                 compat_uptr_t dirty_bitmap; /* one bit per page */
2487                 __u64 padding2;
2488         };
2489 };
2490
2491 static long kvm_vm_compat_ioctl(struct file *filp,
2492                            unsigned int ioctl, unsigned long arg)
2493 {
2494         struct kvm *kvm = filp->private_data;
2495         int r;
2496
2497         if (kvm->mm != current->mm)
2498                 return -EIO;
2499         switch (ioctl) {
2500         case KVM_GET_DIRTY_LOG: {
2501                 struct compat_kvm_dirty_log compat_log;
2502                 struct kvm_dirty_log log;
2503
2504                 r = -EFAULT;
2505                 if (copy_from_user(&compat_log, (void __user *)arg,
2506                                    sizeof(compat_log)))
2507                         goto out;
2508                 log.slot         = compat_log.slot;
2509                 log.padding1     = compat_log.padding1;
2510                 log.padding2     = compat_log.padding2;
2511                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2512
2513                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2514                 break;
2515         }
2516         default:
2517                 r = kvm_vm_ioctl(filp, ioctl, arg);
2518         }
2519
2520 out:
2521         return r;
2522 }
2523 #endif
2524
2525 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2526 {
2527         struct page *page[1];
2528         unsigned long addr;
2529         int npages;
2530         gfn_t gfn = vmf->pgoff;
2531         struct kvm *kvm = vma->vm_file->private_data;
2532
2533         addr = gfn_to_hva(kvm, gfn);
2534         if (kvm_is_error_hva(addr))
2535                 return VM_FAULT_SIGBUS;
2536
2537         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2538                                 NULL);
2539         if (unlikely(npages != 1))
2540                 return VM_FAULT_SIGBUS;
2541
2542         vmf->page = page[0];
2543         return 0;
2544 }
2545
2546 static const struct vm_operations_struct kvm_vm_vm_ops = {
2547         .fault = kvm_vm_fault,
2548 };
2549
2550 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2551 {
2552         vma->vm_ops = &kvm_vm_vm_ops;
2553         return 0;
2554 }
2555
2556 static struct file_operations kvm_vm_fops = {
2557         .release        = kvm_vm_release,
2558         .unlocked_ioctl = kvm_vm_ioctl,
2559 #ifdef CONFIG_COMPAT
2560         .compat_ioctl   = kvm_vm_compat_ioctl,
2561 #endif
2562         .mmap           = kvm_vm_mmap,
2563         .llseek         = noop_llseek,
2564 };
2565
2566 static int kvm_dev_ioctl_create_vm(unsigned long type)
2567 {
2568         int r;
2569         struct kvm *kvm;
2570
2571         kvm = kvm_create_vm(type);
2572         if (IS_ERR(kvm))
2573                 return PTR_ERR(kvm);
2574 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2575         r = kvm_coalesced_mmio_init(kvm);
2576         if (r < 0) {
2577                 kvm_put_kvm(kvm);
2578                 return r;
2579         }
2580 #endif
2581         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2582         if (r < 0)
2583                 kvm_put_kvm(kvm);
2584
2585         return r;
2586 }
2587
2588 static long kvm_dev_ioctl_check_extension_generic(long arg)
2589 {
2590         switch (arg) {
2591         case KVM_CAP_USER_MEMORY:
2592         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2593         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2594 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2595         case KVM_CAP_SET_BOOT_CPU_ID:
2596 #endif
2597         case KVM_CAP_INTERNAL_ERROR_DATA:
2598 #ifdef CONFIG_HAVE_KVM_MSI
2599         case KVM_CAP_SIGNAL_MSI:
2600 #endif
2601 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2602         case KVM_CAP_IRQFD_RESAMPLE:
2603 #endif
2604                 return 1;
2605 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2606         case KVM_CAP_IRQ_ROUTING:
2607                 return KVM_MAX_IRQ_ROUTES;
2608 #endif
2609         default:
2610                 break;
2611         }
2612         return kvm_dev_ioctl_check_extension(arg);
2613 }
2614
2615 static long kvm_dev_ioctl(struct file *filp,
2616                           unsigned int ioctl, unsigned long arg)
2617 {
2618         long r = -EINVAL;
2619
2620         switch (ioctl) {
2621         case KVM_GET_API_VERSION:
2622                 r = -EINVAL;
2623                 if (arg)
2624                         goto out;
2625                 r = KVM_API_VERSION;
2626                 break;
2627         case KVM_CREATE_VM:
2628                 r = kvm_dev_ioctl_create_vm(arg);
2629                 break;
2630         case KVM_CHECK_EXTENSION:
2631                 r = kvm_dev_ioctl_check_extension_generic(arg);
2632                 break;
2633         case KVM_GET_VCPU_MMAP_SIZE:
2634                 r = -EINVAL;
2635                 if (arg)
2636                         goto out;
2637                 r = PAGE_SIZE;     /* struct kvm_run */
2638 #ifdef CONFIG_X86
2639                 r += PAGE_SIZE;    /* pio data page */
2640 #endif
2641 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2642                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2643 #endif
2644                 break;
2645         case KVM_TRACE_ENABLE:
2646         case KVM_TRACE_PAUSE:
2647         case KVM_TRACE_DISABLE:
2648                 r = -EOPNOTSUPP;
2649                 break;
2650         default:
2651                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2652         }
2653 out:
2654         return r;
2655 }
2656
2657 static struct file_operations kvm_chardev_ops = {
2658         .unlocked_ioctl = kvm_dev_ioctl,
2659         .compat_ioctl   = kvm_dev_ioctl,
2660         .llseek         = noop_llseek,
2661 };
2662
2663 static struct miscdevice kvm_dev = {
2664         KVM_MINOR,
2665         "kvm",
2666         &kvm_chardev_ops,
2667 };
2668
2669 static void hardware_enable_nolock(void *junk)
2670 {
2671         int cpu = raw_smp_processor_id();
2672         int r;
2673
2674         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2675                 return;
2676
2677         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2678
2679         r = kvm_arch_hardware_enable(NULL);
2680
2681         if (r) {
2682                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2683                 atomic_inc(&hardware_enable_failed);
2684                 printk(KERN_INFO "kvm: enabling virtualization on "
2685                                  "CPU%d failed\n", cpu);
2686         }
2687 }
2688
2689 static void hardware_enable(void)
2690 {
2691         raw_spin_lock(&kvm_count_lock);
2692         if (kvm_usage_count)
2693                 hardware_enable_nolock(NULL);
2694         raw_spin_unlock(&kvm_count_lock);
2695 }
2696
2697 static void hardware_disable_nolock(void *junk)
2698 {
2699         int cpu = raw_smp_processor_id();
2700
2701         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2702                 return;
2703         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2704         kvm_arch_hardware_disable(NULL);
2705 }
2706
2707 static void hardware_disable(void)
2708 {
2709         raw_spin_lock(&kvm_count_lock);
2710         if (kvm_usage_count)
2711                 hardware_disable_nolock(NULL);
2712         raw_spin_unlock(&kvm_count_lock);
2713 }
2714
2715 static void hardware_disable_all_nolock(void)
2716 {
2717         BUG_ON(!kvm_usage_count);
2718
2719         kvm_usage_count--;
2720         if (!kvm_usage_count)
2721                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2722 }
2723
2724 static void hardware_disable_all(void)
2725 {
2726         raw_spin_lock(&kvm_count_lock);
2727         hardware_disable_all_nolock();
2728         raw_spin_unlock(&kvm_count_lock);
2729 }
2730
2731 static int hardware_enable_all(void)
2732 {
2733         int r = 0;
2734
2735         raw_spin_lock(&kvm_count_lock);
2736
2737         kvm_usage_count++;
2738         if (kvm_usage_count == 1) {
2739                 atomic_set(&hardware_enable_failed, 0);
2740                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2741
2742                 if (atomic_read(&hardware_enable_failed)) {
2743                         hardware_disable_all_nolock();
2744                         r = -EBUSY;
2745                 }
2746         }
2747
2748         raw_spin_unlock(&kvm_count_lock);
2749
2750         return r;
2751 }
2752
2753 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2754                            void *v)
2755 {
2756         int cpu = (long)v;
2757
2758         val &= ~CPU_TASKS_FROZEN;
2759         switch (val) {
2760         case CPU_DYING:
2761                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2762                        cpu);
2763                 hardware_disable();
2764                 break;
2765         case CPU_STARTING:
2766                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2767                        cpu);
2768                 hardware_enable();
2769                 break;
2770         }
2771         return NOTIFY_OK;
2772 }
2773
2774 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2775                       void *v)
2776 {
2777         /*
2778          * Some (well, at least mine) BIOSes hang on reboot if
2779          * in vmx root mode.
2780          *
2781          * And Intel TXT required VMX off for all cpu when system shutdown.
2782          */
2783         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2784         kvm_rebooting = true;
2785         on_each_cpu(hardware_disable_nolock, NULL, 1);
2786         return NOTIFY_OK;
2787 }
2788
2789 static struct notifier_block kvm_reboot_notifier = {
2790         .notifier_call = kvm_reboot,
2791         .priority = 0,
2792 };
2793
2794 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2795 {
2796         int i;
2797
2798         for (i = 0; i < bus->dev_count; i++) {
2799                 struct kvm_io_device *pos = bus->range[i].dev;
2800
2801                 kvm_iodevice_destructor(pos);
2802         }
2803         kfree(bus);
2804 }
2805
2806 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2807                                  const struct kvm_io_range *r2)
2808 {
2809         if (r1->addr < r2->addr)
2810                 return -1;
2811         if (r1->addr + r1->len > r2->addr + r2->len)
2812                 return 1;
2813         return 0;
2814 }
2815
2816 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2817 {
2818         return kvm_io_bus_cmp(p1, p2);
2819 }
2820
2821 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2822                           gpa_t addr, int len)
2823 {
2824         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2825                 .addr = addr,
2826                 .len = len,
2827                 .dev = dev,
2828         };
2829
2830         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2831                 kvm_io_bus_sort_cmp, NULL);
2832
2833         return 0;
2834 }
2835
2836 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2837                              gpa_t addr, int len)
2838 {
2839         struct kvm_io_range *range, key;
2840         int off;
2841
2842         key = (struct kvm_io_range) {
2843                 .addr = addr,
2844                 .len = len,
2845         };
2846
2847         range = bsearch(&key, bus->range, bus->dev_count,
2848                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2849         if (range == NULL)
2850                 return -ENOENT;
2851
2852         off = range - bus->range;
2853
2854         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2855                 off--;
2856
2857         return off;
2858 }
2859
2860 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2861                               struct kvm_io_range *range, const void *val)
2862 {
2863         int idx;
2864
2865         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2866         if (idx < 0)
2867                 return -EOPNOTSUPP;
2868
2869         while (idx < bus->dev_count &&
2870                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2871                 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2872                                         range->len, val))
2873                         return idx;
2874                 idx++;
2875         }
2876
2877         return -EOPNOTSUPP;
2878 }
2879
2880 /* kvm_io_bus_write - called under kvm->slots_lock */
2881 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2882                      int len, const void *val)
2883 {
2884         struct kvm_io_bus *bus;
2885         struct kvm_io_range range;
2886         int r;
2887
2888         range = (struct kvm_io_range) {
2889                 .addr = addr,
2890                 .len = len,
2891         };
2892
2893         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2894         r = __kvm_io_bus_write(bus, &range, val);
2895         return r < 0 ? r : 0;
2896 }
2897
2898 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2899 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2900                             int len, const void *val, long cookie)
2901 {
2902         struct kvm_io_bus *bus;
2903         struct kvm_io_range range;
2904
2905         range = (struct kvm_io_range) {
2906                 .addr = addr,
2907                 .len = len,
2908         };
2909
2910         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2911
2912         /* First try the device referenced by cookie. */
2913         if ((cookie >= 0) && (cookie < bus->dev_count) &&
2914             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2915                 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2916                                         val))
2917                         return cookie;
2918
2919         /*
2920          * cookie contained garbage; fall back to search and return the
2921          * correct cookie value.
2922          */
2923         return __kvm_io_bus_write(bus, &range, val);
2924 }
2925
2926 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2927                              void *val)
2928 {
2929         int idx;
2930
2931         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2932         if (idx < 0)
2933                 return -EOPNOTSUPP;
2934
2935         while (idx < bus->dev_count &&
2936                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2937                 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2938                                        range->len, val))
2939                         return idx;
2940                 idx++;
2941         }
2942
2943         return -EOPNOTSUPP;
2944 }
2945
2946 /* kvm_io_bus_read - called under kvm->slots_lock */
2947 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2948                     int len, void *val)
2949 {
2950         struct kvm_io_bus *bus;
2951         struct kvm_io_range range;
2952         int r;
2953
2954         range = (struct kvm_io_range) {
2955                 .addr = addr,
2956                 .len = len,
2957         };
2958
2959         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2960         r = __kvm_io_bus_read(bus, &range, val);
2961         return r < 0 ? r : 0;
2962 }
2963
2964 /* kvm_io_bus_read_cookie - called under kvm->slots_lock */
2965 int kvm_io_bus_read_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2966                            int len, void *val, long cookie)
2967 {
2968         struct kvm_io_bus *bus;
2969         struct kvm_io_range range;
2970
2971         range = (struct kvm_io_range) {
2972                 .addr = addr,
2973                 .len = len,
2974         };
2975
2976         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2977
2978         /* First try the device referenced by cookie. */
2979         if ((cookie >= 0) && (cookie < bus->dev_count) &&
2980             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2981                 if (!kvm_iodevice_read(bus->range[cookie].dev, addr, len,
2982                                        val))
2983                         return cookie;
2984
2985         /*
2986          * cookie contained garbage; fall back to search and return the
2987          * correct cookie value.
2988          */
2989         return __kvm_io_bus_read(bus, &range, val);
2990 }
2991
2992 /* Caller must hold slots_lock. */
2993 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2994                             int len, struct kvm_io_device *dev)
2995 {
2996         struct kvm_io_bus *new_bus, *bus;
2997
2998         bus = kvm->buses[bus_idx];
2999         /* exclude ioeventfd which is limited by maximum fd */
3000         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3001                 return -ENOSPC;
3002
3003         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3004                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3005         if (!new_bus)
3006                 return -ENOMEM;
3007         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3008                sizeof(struct kvm_io_range)));
3009         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3010         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3011         synchronize_srcu_expedited(&kvm->srcu);
3012         kfree(bus);
3013
3014         return 0;
3015 }
3016
3017 /* Caller must hold slots_lock. */
3018 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3019                               struct kvm_io_device *dev)
3020 {
3021         int i, r;
3022         struct kvm_io_bus *new_bus, *bus;
3023
3024         bus = kvm->buses[bus_idx];
3025         r = -ENOENT;
3026         for (i = 0; i < bus->dev_count; i++)
3027                 if (bus->range[i].dev == dev) {
3028                         r = 0;
3029                         break;
3030                 }
3031
3032         if (r)
3033                 return r;
3034
3035         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3036                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3037         if (!new_bus)
3038                 return -ENOMEM;
3039
3040         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3041         new_bus->dev_count--;
3042         memcpy(new_bus->range + i, bus->range + i + 1,
3043                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3044
3045         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3046         synchronize_srcu_expedited(&kvm->srcu);
3047         kfree(bus);
3048         return r;
3049 }
3050
3051 static struct notifier_block kvm_cpu_notifier = {
3052         .notifier_call = kvm_cpu_hotplug,
3053 };
3054
3055 static int vm_stat_get(void *_offset, u64 *val)
3056 {
3057         unsigned offset = (long)_offset;
3058         struct kvm *kvm;
3059
3060         *val = 0;
3061         spin_lock(&kvm_lock);
3062         list_for_each_entry(kvm, &vm_list, vm_list)
3063                 *val += *(u32 *)((void *)kvm + offset);
3064         spin_unlock(&kvm_lock);
3065         return 0;
3066 }
3067
3068 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3069
3070 static int vcpu_stat_get(void *_offset, u64 *val)
3071 {
3072         unsigned offset = (long)_offset;
3073         struct kvm *kvm;
3074         struct kvm_vcpu *vcpu;
3075         int i;
3076
3077         *val = 0;
3078         spin_lock(&kvm_lock);
3079         list_for_each_entry(kvm, &vm_list, vm_list)
3080                 kvm_for_each_vcpu(i, vcpu, kvm)
3081                         *val += *(u32 *)((void *)vcpu + offset);
3082
3083         spin_unlock(&kvm_lock);
3084         return 0;
3085 }
3086
3087 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3088
3089 static const struct file_operations *stat_fops[] = {
3090         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3091         [KVM_STAT_VM]   = &vm_stat_fops,
3092 };
3093
3094 static int kvm_init_debug(void)
3095 {
3096         int r = -EFAULT;
3097         struct kvm_stats_debugfs_item *p;
3098
3099         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3100         if (kvm_debugfs_dir == NULL)
3101                 goto out;
3102
3103         for (p = debugfs_entries; p->name; ++p) {
3104                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3105                                                 (void *)(long)p->offset,
3106                                                 stat_fops[p->kind]);
3107                 if (p->dentry == NULL)
3108                         goto out_dir;
3109         }
3110
3111         return 0;
3112
3113 out_dir:
3114         debugfs_remove_recursive(kvm_debugfs_dir);
3115 out:
3116         return r;
3117 }
3118
3119 static void kvm_exit_debug(void)
3120 {
3121         struct kvm_stats_debugfs_item *p;
3122
3123         for (p = debugfs_entries; p->name; ++p)
3124                 debugfs_remove(p->dentry);
3125         debugfs_remove(kvm_debugfs_dir);
3126 }
3127
3128 static int kvm_suspend(void)
3129 {
3130         if (kvm_usage_count)
3131                 hardware_disable_nolock(NULL);
3132         return 0;
3133 }
3134
3135 static void kvm_resume(void)
3136 {
3137         if (kvm_usage_count) {
3138                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3139                 hardware_enable_nolock(NULL);
3140         }
3141 }
3142
3143 static struct syscore_ops kvm_syscore_ops = {
3144         .suspend = kvm_suspend,
3145         .resume = kvm_resume,
3146 };
3147
3148 static inline
3149 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3150 {
3151         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3152 }
3153
3154 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3155 {
3156         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3157         if (vcpu->preempted)
3158                 vcpu->preempted = false;
3159
3160         kvm_arch_vcpu_load(vcpu, cpu);
3161 }
3162
3163 static void kvm_sched_out(struct preempt_notifier *pn,
3164                           struct task_struct *next)
3165 {
3166         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3167
3168         if (current->state == TASK_RUNNING)
3169                 vcpu->preempted = true;
3170         kvm_arch_vcpu_put(vcpu);
3171 }
3172
3173 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3174                   struct module *module)
3175 {
3176         int r;
3177         int cpu;
3178
3179         r = kvm_arch_init(opaque);
3180         if (r)
3181                 goto out_fail;
3182
3183         /*
3184          * kvm_arch_init makes sure there's at most one caller
3185          * for architectures that support multiple implementations,
3186          * like intel and amd on x86.
3187          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3188          * conflicts in case kvm is already setup for another implementation.
3189          */
3190         r = kvm_irqfd_init();
3191         if (r)
3192                 goto out_irqfd;
3193
3194         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3195                 r = -ENOMEM;
3196                 goto out_free_0;
3197         }
3198
3199         r = kvm_arch_hardware_setup();
3200         if (r < 0)
3201                 goto out_free_0a;
3202
3203         for_each_online_cpu(cpu) {
3204                 smp_call_function_single(cpu,
3205                                 kvm_arch_check_processor_compat,
3206                                 &r, 1);
3207                 if (r < 0)
3208                         goto out_free_1;
3209         }
3210
3211         r = register_cpu_notifier(&kvm_cpu_notifier);
3212         if (r)
3213                 goto out_free_2;
3214         register_reboot_notifier(&kvm_reboot_notifier);
3215
3216         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3217         if (!vcpu_align)
3218                 vcpu_align = __alignof__(struct kvm_vcpu);
3219         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3220                                            0, NULL);
3221         if (!kvm_vcpu_cache) {
3222                 r = -ENOMEM;
3223                 goto out_free_3;
3224         }
3225
3226         r = kvm_async_pf_init();
3227         if (r)
3228                 goto out_free;
3229
3230         kvm_chardev_ops.owner = module;
3231         kvm_vm_fops.owner = module;
3232         kvm_vcpu_fops.owner = module;
3233
3234         r = misc_register(&kvm_dev);
3235         if (r) {
3236                 printk(KERN_ERR "kvm: misc device register failed\n");
3237                 goto out_unreg;
3238         }
3239
3240         register_syscore_ops(&kvm_syscore_ops);
3241
3242         kvm_preempt_ops.sched_in = kvm_sched_in;
3243         kvm_preempt_ops.sched_out = kvm_sched_out;
3244
3245         r = kvm_init_debug();
3246         if (r) {
3247                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3248                 goto out_undebugfs;
3249         }
3250
3251         return 0;
3252
3253 out_undebugfs:
3254         unregister_syscore_ops(&kvm_syscore_ops);
3255         misc_deregister(&kvm_dev);
3256 out_unreg:
3257         kvm_async_pf_deinit();
3258 out_free:
3259         kmem_cache_destroy(kvm_vcpu_cache);
3260 out_free_3:
3261         unregister_reboot_notifier(&kvm_reboot_notifier);
3262         unregister_cpu_notifier(&kvm_cpu_notifier);
3263 out_free_2:
3264 out_free_1:
3265         kvm_arch_hardware_unsetup();
3266 out_free_0a:
3267         free_cpumask_var(cpus_hardware_enabled);
3268 out_free_0:
3269         kvm_irqfd_exit();
3270 out_irqfd:
3271         kvm_arch_exit();
3272 out_fail:
3273         return r;
3274 }
3275 EXPORT_SYMBOL_GPL(kvm_init);
3276
3277 void kvm_exit(void)
3278 {
3279         kvm_exit_debug();
3280         misc_deregister(&kvm_dev);
3281         kmem_cache_destroy(kvm_vcpu_cache);
3282         kvm_async_pf_deinit();
3283         unregister_syscore_ops(&kvm_syscore_ops);
3284         unregister_reboot_notifier(&kvm_reboot_notifier);
3285         unregister_cpu_notifier(&kvm_cpu_notifier);
3286         on_each_cpu(hardware_disable_nolock, NULL, 1);
3287         kvm_arch_hardware_unsetup();
3288         kvm_arch_exit();
3289         kvm_irqfd_exit();
3290         free_cpumask_var(cpus_hardware_enabled);
3291 }
3292 EXPORT_SYMBOL_GPL(kvm_exit);