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/ioctl.h>
  56 #include <asm/uaccess.h>
  57 #include <asm/pgtable.h>
  58
  59 #include "coalesced_mmio.h"
  60 #include "async_pf.h"
  61 #include "vfio.h"
  62
  63 #define CREATE_TRACE_POINTS
  64 #include <trace/events/kvm.h>
  65
  66 MODULE_AUTHOR("Qumranet");
  67 MODULE_LICENSE("GPL");
  68
  69 /*
  70  * Ordering of locks:
  71  *
  72  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  73  */
  74
  75 DEFINE_SPINLOCK(kvm_lock);
  76 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
  77 LIST_HEAD(vm_list);
  78
  79 static cpumask_var_t cpus_hardware_enabled;
  80 static int kvm_usage_count = 0;
  81 static atomic_t hardware_enable_failed;
  82
  83 struct kmem_cache *kvm_vcpu_cache;
  84 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  85
  86 static __read_mostly struct preempt_ops kvm_preempt_ops;
  87
  88 struct dentry *kvm_debugfs_dir;
  89
  90 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  91                            unsigned long arg);
  92 #ifdef CONFIG_COMPAT
  93 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  94                                   unsigned long arg);
  95 #endif
  96 static int hardware_enable_all(void);
  97 static void hardware_disable_all(void);
  98
  99 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
 100
 101 static void kvm_release_pfn_dirty(pfn_t pfn);
 102 static void mark_page_dirty_in_slot(struct kvm *kvm,
 103                                     struct kvm_memory_slot *memslot, gfn_t gfn);
 104
 105 __visible bool kvm_rebooting;
 106 EXPORT_SYMBOL_GPL(kvm_rebooting);
 107
 108 static bool largepages_enabled = true;
 109
 110 bool kvm_is_mmio_pfn(pfn_t pfn)
 111 {
 112         if (pfn_valid(pfn))
 113                 return PageReserved(pfn_to_page(pfn));
 114
 115         return true;
 116 }
 117
 118 /*
 119  * Switches to specified vcpu, until a matching vcpu_put()
 120  */
 121 int vcpu_load(struct kvm_vcpu *vcpu)
 122 {
 123         int cpu;
 124
 125         if (mutex_lock_killable(&vcpu->mutex))
 126                 return -EINTR;
 127         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 128                 /* The thread running this VCPU changed. */
 129                 struct pid *oldpid = vcpu->pid;
 130                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 131                 rcu_assign_pointer(vcpu->pid, newpid);
 132                 if (oldpid)
 133                         synchronize_rcu();
 134                 put_pid(oldpid);
 135         }
 136         cpu = get_cpu();
 137         preempt_notifier_register(&vcpu->preempt_notifier);
 138         kvm_arch_vcpu_load(vcpu, cpu);
 139         put_cpu();
 140         return 0;
 141 }
 142
 143 void vcpu_put(struct kvm_vcpu *vcpu)
 144 {
 145         preempt_disable();
 146         kvm_arch_vcpu_put(vcpu);
 147         preempt_notifier_unregister(&vcpu->preempt_notifier);
 148         preempt_enable();
 149         mutex_unlock(&vcpu->mutex);
 150 }
 151
 152 static void ack_flush(void *_completed)
 153 {
 154 }
 155
 156 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 157 {
 158         int i, cpu, me;
 159         cpumask_var_t cpus;
 160         bool called = true;
 161         struct kvm_vcpu *vcpu;
 162
 163         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 164
 165         me = get_cpu();
 166         kvm_for_each_vcpu(i, vcpu, kvm) {
 167                 kvm_make_request(req, vcpu);
 168                 cpu = vcpu->cpu;
 169
 170                 /* Set ->requests bit before we read ->mode */
 171                 smp_mb();
 172
 173                 if (cpus != NULL && cpu != -1 && cpu != me &&
 174                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 175                         cpumask_set_cpu(cpu, cpus);
 176         }
 177         if (unlikely(cpus == NULL))
 178                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 179         else if (!cpumask_empty(cpus))
 180                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 181         else
 182                 called = false;
 183         put_cpu();
 184         free_cpumask_var(cpus);
 185         return called;
 186 }
 187
 188 void kvm_flush_remote_tlbs(struct kvm *kvm)
 189 {
 190         long dirty_count = kvm->tlbs_dirty;
 191
 192         smp_mb();
 193         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 194                 ++kvm->stat.remote_tlb_flush;
 195         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 196 }
 197 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 198
 199 void kvm_reload_remote_mmus(struct kvm *kvm)
 200 {
 201         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 202 }
 203
 204 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 205 {
 206         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 207 }
 208
 209 void kvm_make_scan_ioapic_request(struct kvm *kvm)
 210 {
 211         make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
 212 }
 213
 214 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 215 {
 216         struct page *page;
 217         int r;
 218
 219         mutex_init(&vcpu->mutex);
 220         vcpu->cpu = -1;
 221         vcpu->kvm = kvm;
 222         vcpu->vcpu_id = id;
 223         vcpu->pid = NULL;
 224         init_waitqueue_head(&vcpu->wq);
 225         kvm_async_pf_vcpu_init(vcpu);
 226
 227         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 228         if (!page) {
 229                 r = -ENOMEM;
 230                 goto fail;
 231         }
 232         vcpu->run = page_address(page);
 233
 234         kvm_vcpu_set_in_spin_loop(vcpu, false);
 235         kvm_vcpu_set_dy_eligible(vcpu, false);
 236         vcpu->preempted = false;
 237
 238         r = kvm_arch_vcpu_init(vcpu);
 239         if (r < 0)
 240                 goto fail_free_run;
 241         return 0;
 242
 243 fail_free_run:
 244         free_page((unsigned long)vcpu->run);
 245 fail:
 246         return r;
 247 }
 248 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 249
 250 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 251 {
 252         put_pid(vcpu->pid);
 253         kvm_arch_vcpu_uninit(vcpu);
 254         free_page((unsigned long)vcpu->run);
 255 }
 256 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 257
 258 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 259 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 260 {
 261         return container_of(mn, struct kvm, mmu_notifier);
 262 }
 263
 264 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 265                                              struct mm_struct *mm,
 266                                              unsigned long address)
 267 {
 268         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 269         int need_tlb_flush, idx;
 270
 271         /*
 272          * When ->invalidate_page runs, the linux pte has been zapped
 273          * already but the page is still allocated until
 274          * ->invalidate_page returns. So if we increase the sequence
 275          * here the kvm page fault will notice if the spte can't be
 276          * established because the page is going to be freed. If
 277          * instead the kvm page fault establishes the spte before
 278          * ->invalidate_page runs, kvm_unmap_hva will release it
 279          * before returning.
 280          *
 281          * The sequence increase only need to be seen at spin_unlock
 282          * time, and not at spin_lock time.
 283          *
 284          * Increasing the sequence after the spin_unlock would be
 285          * unsafe because the kvm page fault could then establish the
 286          * pte after kvm_unmap_hva returned, without noticing the page
 287          * is going to be freed.
 288          */
 289         idx = srcu_read_lock(&kvm->srcu);
 290         spin_lock(&kvm->mmu_lock);
 291
 292         kvm->mmu_notifier_seq++;
 293         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 294         /* we've to flush the tlb before the pages can be freed */
 295         if (need_tlb_flush)
 296                 kvm_flush_remote_tlbs(kvm);
 297
 298         spin_unlock(&kvm->mmu_lock);
 299         srcu_read_unlock(&kvm->srcu, idx);
 300 }
 301
 302 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 303                                         struct mm_struct *mm,
 304                                         unsigned long address,
 305                                         pte_t pte)
 306 {
 307         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 308         int idx;
 309
 310         idx = srcu_read_lock(&kvm->srcu);
 311         spin_lock(&kvm->mmu_lock);
 312         kvm->mmu_notifier_seq++;
 313         kvm_set_spte_hva(kvm, address, pte);
 314         spin_unlock(&kvm->mmu_lock);
 315         srcu_read_unlock(&kvm->srcu, idx);
 316 }
 317
 318 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 319                                                     struct mm_struct *mm,
 320                                                     unsigned long start,
 321                                                     unsigned long end)
 322 {
 323         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 324         int need_tlb_flush = 0, idx;
 325
 326         idx = srcu_read_lock(&kvm->srcu);
 327         spin_lock(&kvm->mmu_lock);
 328         /*
 329          * The count increase must become visible at unlock time as no
 330          * spte can be established without taking the mmu_lock and
 331          * count is also read inside the mmu_lock critical section.
 332          */
 333         kvm->mmu_notifier_count++;
 334         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 335         need_tlb_flush |= kvm->tlbs_dirty;
 336         /* we've to flush the tlb before the pages can be freed */
 337         if (need_tlb_flush)
 338                 kvm_flush_remote_tlbs(kvm);
 339
 340         spin_unlock(&kvm->mmu_lock);
 341         srcu_read_unlock(&kvm->srcu, idx);
 342 }
 343
 344 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 345                                                   struct mm_struct *mm,
 346                                                   unsigned long start,
 347                                                   unsigned long end)
 348 {
 349         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 350
 351         spin_lock(&kvm->mmu_lock);
 352         /*
 353          * This sequence increase will notify the kvm page fault that
 354          * the page that is going to be mapped in the spte could have
 355          * been freed.
 356          */
 357         kvm->mmu_notifier_seq++;
 358         smp_wmb();
 359         /*
 360          * The above sequence increase must be visible before the
 361          * below count decrease, which is ensured by the smp_wmb above
 362          * in conjunction with the smp_rmb in mmu_notifier_retry().
 363          */
 364         kvm->mmu_notifier_count--;
 365         spin_unlock(&kvm->mmu_lock);
 366
 367         BUG_ON(kvm->mmu_notifier_count < 0);
 368 }
 369
 370 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 371                                               struct mm_struct *mm,
 372                                               unsigned long start,
 373                                               unsigned long end)
 374 {
 375         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 376         int young, idx;
 377
 378         idx = srcu_read_lock(&kvm->srcu);
 379         spin_lock(&kvm->mmu_lock);
 380
 381         young = kvm_age_hva(kvm, start, end);
 382         if (young)
 383                 kvm_flush_remote_tlbs(kvm);
 384
 385         spin_unlock(&kvm->mmu_lock);
 386         srcu_read_unlock(&kvm->srcu, idx);
 387
 388         return young;
 389 }
 390
 391 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 392                                        struct mm_struct *mm,
 393                                        unsigned long address)
 394 {
 395         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 396         int young, idx;
 397
 398         idx = srcu_read_lock(&kvm->srcu);
 399         spin_lock(&kvm->mmu_lock);
 400         young = kvm_test_age_hva(kvm, address);
 401         spin_unlock(&kvm->mmu_lock);
 402         srcu_read_unlock(&kvm->srcu, idx);
 403
 404         return young;
 405 }
 406
 407 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 408                                      struct mm_struct *mm)
 409 {
 410         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 411         int idx;
 412
 413         idx = srcu_read_lock(&kvm->srcu);
 414         kvm_arch_flush_shadow_all(kvm);
 415         srcu_read_unlock(&kvm->srcu, idx);
 416 }
 417
 418 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 419         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 420         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 421         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 422         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 423         .test_young             = kvm_mmu_notifier_test_young,
 424         .change_pte             = kvm_mmu_notifier_change_pte,
 425         .release                = kvm_mmu_notifier_release,
 426 };
 427
 428 static int kvm_init_mmu_notifier(struct kvm *kvm)
 429 {
 430         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 431         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 432 }
 433
 434 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 435
 436 static int kvm_init_mmu_notifier(struct kvm *kvm)
 437 {
 438         return 0;
 439 }
 440
 441 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 442
 443 static void kvm_init_memslots_id(struct kvm *kvm)
 444 {
 445         int i;
 446         struct kvm_memslots *slots = kvm->memslots;
 447
 448         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 449                 slots->id_to_index[i] = slots->memslots[i].id = i;
 450 }
 451
 452 static struct kvm *kvm_create_vm(unsigned long type)
 453 {
 454         int r, i;
 455         struct kvm *kvm = kvm_arch_alloc_vm();
 456
 457         if (!kvm)
 458                 return ERR_PTR(-ENOMEM);
 459
 460         r = kvm_arch_init_vm(kvm, type);
 461         if (r)
 462                 goto out_err_no_disable;
 463
 464         r = hardware_enable_all();
 465         if (r)
 466                 goto out_err_no_disable;
 467
 468 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 469         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 470 #endif
 471 #ifdef CONFIG_HAVE_KVM_IRQFD
 472         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 473 #endif
 474
 475         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 476
 477         r = -ENOMEM;
 478         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 479         if (!kvm->memslots)
 480                 goto out_err_no_srcu;
 481
 482         /*
 483          * Init kvm generation close to the maximum to easily test the
 484          * code of handling generation number wrap-around.
 485          */
 486         kvm->memslots->generation = -150;
 487
 488         kvm_init_memslots_id(kvm);
 489         if (init_srcu_struct(&kvm->srcu))
 490                 goto out_err_no_srcu;
 491         if (init_srcu_struct(&kvm->irq_srcu))
 492                 goto out_err_no_irq_srcu;
 493         for (i = 0; i < KVM_NR_BUSES; i++) {
 494                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 495                                         GFP_KERNEL);
 496                 if (!kvm->buses[i])
 497                         goto out_err;
 498         }
 499
 500         spin_lock_init(&kvm->mmu_lock);
 501         kvm->mm = current->mm;
 502         atomic_inc(&kvm->mm->mm_count);
 503         kvm_eventfd_init(kvm);
 504         mutex_init(&kvm->lock);
 505         mutex_init(&kvm->irq_lock);
 506         mutex_init(&kvm->slots_lock);
 507         atomic_set(&kvm->users_count, 1);
 508         INIT_LIST_HEAD(&kvm->devices);
 509
 510         r = kvm_init_mmu_notifier(kvm);
 511         if (r)
 512                 goto out_err;
 513
 514         spin_lock(&kvm_lock);
 515         list_add(&kvm->vm_list, &vm_list);
 516         spin_unlock(&kvm_lock);
 517
 518         return kvm;
 519
 520 out_err:
 521         cleanup_srcu_struct(&kvm->irq_srcu);
 522 out_err_no_irq_srcu:
 523         cleanup_srcu_struct(&kvm->srcu);
 524 out_err_no_srcu:
 525         hardware_disable_all();
 526 out_err_no_disable:
 527         for (i = 0; i < KVM_NR_BUSES; i++)
 528                 kfree(kvm->buses[i]);
 529         kfree(kvm->memslots);
 530         kvm_arch_free_vm(kvm);
 531         return ERR_PTR(r);
 532 }
 533
 534 /*
 535  * Avoid using vmalloc for a small buffer.
 536  * Should not be used when the size is statically known.
 537  */
 538 void *kvm_kvzalloc(unsigned long size)
 539 {
 540         if (size > PAGE_SIZE)
 541                 return vzalloc(size);
 542         else
 543                 return kzalloc(size, GFP_KERNEL);
 544 }
 545
 546 void kvm_kvfree(const void *addr)
 547 {
 548         if (is_vmalloc_addr(addr))
 549                 vfree(addr);
 550         else
 551                 kfree(addr);
 552 }
 553
 554 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 555 {
 556         if (!memslot->dirty_bitmap)
 557                 return;
 558
 559         kvm_kvfree(memslot->dirty_bitmap);
 560         memslot->dirty_bitmap = NULL;
 561 }
 562
 563 /*
 564  * Free any memory in @free but not in @dont.
 565  */
 566 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
 567                                   struct kvm_memory_slot *dont)
 568 {
 569         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 570                 kvm_destroy_dirty_bitmap(free);
 571
 572         kvm_arch_free_memslot(kvm, free, dont);
 573
 574         free->npages = 0;
 575 }
 576
 577 static void kvm_free_physmem(struct kvm *kvm)
 578 {
 579         struct kvm_memslots *slots = kvm->memslots;
 580         struct kvm_memory_slot *memslot;
 581
 582         kvm_for_each_memslot(memslot, slots)
 583                 kvm_free_physmem_slot(kvm, memslot, NULL);
 584
 585         kfree(kvm->memslots);
 586 }
 587
 588 static void kvm_destroy_devices(struct kvm *kvm)
 589 {
 590         struct list_head *node, *tmp;
 591
 592         list_for_each_safe(node, tmp, &kvm->devices) {
 593                 struct kvm_device *dev =
 594                         list_entry(node, struct kvm_device, vm_node);
 595
 596                 list_del(node);
 597                 dev->ops->destroy(dev);
 598         }
 599 }
 600
 601 static void kvm_destroy_vm(struct kvm *kvm)
 602 {
 603         int i;
 604         struct mm_struct *mm = kvm->mm;
 605
 606         kvm_arch_sync_events(kvm);
 607         spin_lock(&kvm_lock);
 608         list_del(&kvm->vm_list);
 609         spin_unlock(&kvm_lock);
 610         kvm_free_irq_routing(kvm);
 611         for (i = 0; i < KVM_NR_BUSES; i++)
 612                 kvm_io_bus_destroy(kvm->buses[i]);
 613         kvm_coalesced_mmio_free(kvm);
 614 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 615         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 616 #else
 617         kvm_arch_flush_shadow_all(kvm);
 618 #endif
 619         kvm_arch_destroy_vm(kvm);
 620         kvm_destroy_devices(kvm);
 621         kvm_free_physmem(kvm);
 622         cleanup_srcu_struct(&kvm->irq_srcu);
 623         cleanup_srcu_struct(&kvm->srcu);
 624         kvm_arch_free_vm(kvm);
 625         hardware_disable_all();
 626         mmdrop(mm);
 627 }
 628
 629 void kvm_get_kvm(struct kvm *kvm)
 630 {
 631         atomic_inc(&kvm->users_count);
 632 }
 633 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 634
 635 void kvm_put_kvm(struct kvm *kvm)
 636 {
 637         if (atomic_dec_and_test(&kvm->users_count))
 638                 kvm_destroy_vm(kvm);
 639 }
 640 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 641
 642
 643 static int kvm_vm_release(struct inode *inode, struct file *filp)
 644 {
 645         struct kvm *kvm = filp->private_data;
 646
 647         kvm_irqfd_release(kvm);
 648
 649         kvm_put_kvm(kvm);
 650         return 0;
 651 }
 652
 653 /*
 654  * Allocation size is twice as large as the actual dirty bitmap size.
 655  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 656  */
 657 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 658 {
 659         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 660
 661         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 662         if (!memslot->dirty_bitmap)
 663                 return -ENOMEM;
 664
 665         return 0;
 666 }
 667
 668 static int cmp_memslot(const void *slot1, const void *slot2)
 669 {
 670         struct kvm_memory_slot *s1, *s2;
 671
 672         s1 = (struct kvm_memory_slot *)slot1;
 673         s2 = (struct kvm_memory_slot *)slot2;
 674
 675         if (s1->npages < s2->npages)
 676                 return 1;
 677         if (s1->npages > s2->npages)
 678                 return -1;
 679
 680         return 0;
 681 }
 682
 683 /*
 684  * Sort the memslots base on its size, so the larger slots
 685  * will get better fit.
 686  */
 687 static void sort_memslots(struct kvm_memslots *slots)
 688 {
 689         int i;
 690
 691         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 692               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 693
 694         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 695                 slots->id_to_index[slots->memslots[i].id] = i;
 696 }
 697
 698 static void update_memslots(struct kvm_memslots *slots,
 699                             struct kvm_memory_slot *new)
 700 {
 701         if (new) {
 702                 int id = new->id;
 703                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 704                 unsigned long npages = old->npages;
 705
 706                 *old = *new;
 707                 if (new->npages != npages)
 708                         sort_memslots(slots);
 709         }
 710 }
 711
 712 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 713 {
 714         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 715
 716 #ifdef __KVM_HAVE_READONLY_MEM
 717         valid_flags |= KVM_MEM_READONLY;
 718 #endif
 719
 720         if (mem->flags & ~valid_flags)
 721                 return -EINVAL;
 722
 723         return 0;
 724 }
 725
 726 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 727                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 728 {
 729         struct kvm_memslots *old_memslots = kvm->memslots;
 730
 731         /*
 732          * Set the low bit in the generation, which disables SPTE caching
 733          * until the end of synchronize_srcu_expedited.
 734          */
 735         WARN_ON(old_memslots->generation & 1);
 736         slots->generation = old_memslots->generation + 1;
 737
 738         update_memslots(slots, new);
 739         rcu_assign_pointer(kvm->memslots, slots);
 740         synchronize_srcu_expedited(&kvm->srcu);
 741
 742         /*
 743          * Increment the new memslot generation a second time. This prevents
 744          * vm exits that race with memslot updates from caching a memslot
 745          * generation that will (potentially) be valid forever.
 746          */
 747         slots->generation++;
 748
 749         kvm_arch_memslots_updated(kvm);
 750
 751         return old_memslots;
 752 }
 753
 754 /*
 755  * Allocate some memory and give it an address in the guest physical address
 756  * space.
 757  *
 758  * Discontiguous memory is allowed, mostly for framebuffers.
 759  *
 760  * Must be called holding mmap_sem for write.
 761  */
 762 int __kvm_set_memory_region(struct kvm *kvm,
 763                             struct kvm_userspace_memory_region *mem)
 764 {
 765         int r;
 766         gfn_t base_gfn;
 767         unsigned long npages;
 768         struct kvm_memory_slot *slot;
 769         struct kvm_memory_slot old, new;
 770         struct kvm_memslots *slots = NULL, *old_memslots;
 771         enum kvm_mr_change change;
 772
 773         r = check_memory_region_flags(mem);
 774         if (r)
 775                 goto out;
 776
 777         r = -EINVAL;
 778         /* General sanity checks */
 779         if (mem->memory_size & (PAGE_SIZE - 1))
 780                 goto out;
 781         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 782                 goto out;
 783         /* We can read the guest memory with __xxx_user() later on. */
 784         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 785             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 786              !access_ok(VERIFY_WRITE,
 787                         (void __user *)(unsigned long)mem->userspace_addr,
 788                         mem->memory_size)))
 789                 goto out;
 790         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 791                 goto out;
 792         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 793                 goto out;
 794
 795         slot = id_to_memslot(kvm->memslots, mem->slot);
 796         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 797         npages = mem->memory_size >> PAGE_SHIFT;
 798
 799         if (npages > KVM_MEM_MAX_NR_PAGES)
 800                 goto out;
 801
 802         if (!npages)
 803                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 804
 805         new = old = *slot;
 806
 807         new.id = mem->slot;
 808         new.base_gfn = base_gfn;
 809         new.npages = npages;
 810         new.flags = mem->flags;
 811
 812         if (npages) {
 813                 if (!old.npages)
 814                         change = KVM_MR_CREATE;
 815                 else { /* Modify an existing slot. */
 816                         if ((mem->userspace_addr != old.userspace_addr) ||
 817                             (npages != old.npages) ||
 818                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 819                                 goto out;
 820
 821                         if (base_gfn != old.base_gfn)
 822                                 change = KVM_MR_MOVE;
 823                         else if (new.flags != old.flags)
 824                                 change = KVM_MR_FLAGS_ONLY;
 825                         else { /* Nothing to change. */
 826                                 r = 0;
 827                                 goto out;
 828                         }
 829                 }
 830         } else if (old.npages) {
 831                 change = KVM_MR_DELETE;
 832         } else /* Modify a non-existent slot: disallowed. */
 833                 goto out;
 834
 835         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 836                 /* Check for overlaps */
 837                 r = -EEXIST;
 838                 kvm_for_each_memslot(slot, kvm->memslots) {
 839                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 840                             (slot->id == mem->slot))
 841                                 continue;
 842                         if (!((base_gfn + npages <= slot->base_gfn) ||
 843                               (base_gfn >= slot->base_gfn + slot->npages)))
 844                                 goto out;
 845                 }
 846         }
 847
 848         /* Free page dirty bitmap if unneeded */
 849         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 850                 new.dirty_bitmap = NULL;
 851
 852         r = -ENOMEM;
 853         if (change == KVM_MR_CREATE) {
 854                 new.userspace_addr = mem->userspace_addr;
 855
 856                 if (kvm_arch_create_memslot(kvm, &new, npages))
 857                         goto out_free;
 858         }
 859
 860         /* Allocate page dirty bitmap if needed */
 861         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 862                 if (kvm_create_dirty_bitmap(&new) < 0)
 863                         goto out_free;
 864         }
 865
 866         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 867                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 868                                 GFP_KERNEL);
 869                 if (!slots)
 870                         goto out_free;
 871                 slot = id_to_memslot(slots, mem->slot);
 872                 slot->flags |= KVM_MEMSLOT_INVALID;
 873
 874                 old_memslots = install_new_memslots(kvm, slots, NULL);
 875
 876                 /* slot was deleted or moved, clear iommu mapping */
 877                 kvm_iommu_unmap_pages(kvm, &old);
 878                 /* From this point no new shadow pages pointing to a deleted,
 879                  * or moved, memslot will be created.
 880                  *
 881                  * validation of sp->gfn happens in:
 882                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 883                  *      - kvm_is_visible_gfn (mmu_check_roots)
 884                  */
 885                 kvm_arch_flush_shadow_memslot(kvm, slot);
 886                 slots = old_memslots;
 887         }
 888
 889         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 890         if (r)
 891                 goto out_slots;
 892
 893         r = -ENOMEM;
 894         /*
 895          * We can re-use the old_memslots from above, the only difference
 896          * from the currently installed memslots is the invalid flag.  This
 897          * will get overwritten by update_memslots anyway.
 898          */
 899         if (!slots) {
 900                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 901                                 GFP_KERNEL);
 902                 if (!slots)
 903                         goto out_free;
 904         }
 905
 906         /* actual memory is freed via old in kvm_free_physmem_slot below */
 907         if (change == KVM_MR_DELETE) {
 908                 new.dirty_bitmap = NULL;
 909                 memset(&new.arch, 0, sizeof(new.arch));
 910         }
 911
 912         old_memslots = install_new_memslots(kvm, slots, &new);
 913
 914         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 915
 916         kvm_free_physmem_slot(kvm, &old, &new);
 917         kfree(old_memslots);
 918
 919         /*
 920          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 921          * un-mapped and re-mapped if their base changes.  Since base change
 922          * unmapping is handled above with slot deletion, mapping alone is
 923          * needed here.  Anything else the iommu might care about for existing
 924          * slots (size changes, userspace addr changes and read-only flag
 925          * changes) is disallowed above, so any other attribute changes getting
 926          * here can be skipped.
 927          */
 928         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 929                 r = kvm_iommu_map_pages(kvm, &new);
 930                 return r;
 931         }
 932
 933         return 0;
 934
 935 out_slots:
 936         kfree(slots);
 937 out_free:
 938         kvm_free_physmem_slot(kvm, &new, &old);
 939 out:
 940         return r;
 941 }
 942 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 943
 944 int kvm_set_memory_region(struct kvm *kvm,
 945                           struct kvm_userspace_memory_region *mem)
 946 {
 947         int r;
 948
 949         mutex_lock(&kvm->slots_lock);
 950         r = __kvm_set_memory_region(kvm, mem);
 951         mutex_unlock(&kvm->slots_lock);
 952         return r;
 953 }
 954 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 955
 956 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 957                                           struct kvm_userspace_memory_region *mem)
 958 {
 959         if (mem->slot >= KVM_USER_MEM_SLOTS)
 960                 return -EINVAL;
 961         return kvm_set_memory_region(kvm, mem);
 962 }
 963
 964 int kvm_get_dirty_log(struct kvm *kvm,
 965                         struct kvm_dirty_log *log, int *is_dirty)
 966 {
 967         struct kvm_memory_slot *memslot;
 968         int r, i;
 969         unsigned long n;
 970         unsigned long any = 0;
 971
 972         r = -EINVAL;
 973         if (log->slot >= KVM_USER_MEM_SLOTS)
 974                 goto out;
 975
 976         memslot = id_to_memslot(kvm->memslots, log->slot);
 977         r = -ENOENT;
 978         if (!memslot->dirty_bitmap)
 979                 goto out;
 980
 981         n = kvm_dirty_bitmap_bytes(memslot);
 982
 983         for (i = 0; !any && i < n/sizeof(long); ++i)
 984                 any = memslot->dirty_bitmap[i];
 985
 986         r = -EFAULT;
 987         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 988                 goto out;
 989
 990         if (any)
 991                 *is_dirty = 1;
 992
 993         r = 0;
 994 out:
 995         return r;
 996 }
 997 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
 998
 999 bool kvm_largepages_enabled(void)
1000 {
1001         return largepages_enabled;
1002 }
1003
1004 void kvm_disable_largepages(void)
1005 {
1006         largepages_enabled = false;
1007 }
1008 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1009
1010 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1011 {
1012         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1013 }
1014 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1015
1016 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1017 {
1018         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1019
1020         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1021               memslot->flags & KVM_MEMSLOT_INVALID)
1022                 return 0;
1023
1024         return 1;
1025 }
1026 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1027
1028 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1029 {
1030         struct vm_area_struct *vma;
1031         unsigned long addr, size;
1032
1033         size = PAGE_SIZE;
1034
1035         addr = gfn_to_hva(kvm, gfn);
1036         if (kvm_is_error_hva(addr))
1037                 return PAGE_SIZE;
1038
1039         down_read(&current->mm->mmap_sem);
1040         vma = find_vma(current->mm, addr);
1041         if (!vma)
1042                 goto out;
1043
1044         size = vma_kernel_pagesize(vma);
1045
1046 out:
1047         up_read(&current->mm->mmap_sem);
1048
1049         return size;
1050 }
1051
1052 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1053 {
1054         return slot->flags & KVM_MEM_READONLY;
1055 }
1056
1057 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1058                                        gfn_t *nr_pages, bool write)
1059 {
1060         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1061                 return KVM_HVA_ERR_BAD;
1062
1063         if (memslot_is_readonly(slot) && write)
1064                 return KVM_HVA_ERR_RO_BAD;
1065
1066         if (nr_pages)
1067                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1068
1069         return __gfn_to_hva_memslot(slot, gfn);
1070 }
1071
1072 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1073                                      gfn_t *nr_pages)
1074 {
1075         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1076 }
1077
1078 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1079                                         gfn_t gfn)
1080 {
1081         return gfn_to_hva_many(slot, gfn, NULL);
1082 }
1083 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1084
1085 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1086 {
1087         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1088 }
1089 EXPORT_SYMBOL_GPL(gfn_to_hva);
1090
1091 /*
1092  * If writable is set to false, the hva returned by this function is only
1093  * allowed to be read.
1094  */
1095 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1096 {
1097         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1098         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1099
1100         if (!kvm_is_error_hva(hva) && writable)
1101                 *writable = !memslot_is_readonly(slot);
1102
1103         return hva;
1104 }
1105
1106 static int kvm_read_hva(void *data, void __user *hva, int len)
1107 {
1108         return __copy_from_user(data, hva, len);
1109 }
1110
1111 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1112 {
1113         return __copy_from_user_inatomic(data, hva, len);
1114 }
1115
1116 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1117         unsigned long start, int write, struct page **page)
1118 {
1119         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1120
1121         if (write)
1122                 flags |= FOLL_WRITE;
1123
1124         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1125 }
1126
1127 int kvm_get_user_page_io(struct task_struct *tsk, struct mm_struct *mm,
1128                          unsigned long addr, bool write_fault,
1129                          struct page **pagep)
1130 {
1131         int npages;
1132         int locked = 1;
1133         int flags = FOLL_TOUCH | FOLL_HWPOISON |
1134                     (pagep ? FOLL_GET : 0) |
1135                     (write_fault ? FOLL_WRITE : 0);
1136
1137         /*
1138          * If retrying the fault, we get here *not* having allowed the filemap
1139          * to wait on the page lock. We should now allow waiting on the IO with
1140          * the mmap semaphore released.
1141          */
1142         down_read(&mm->mmap_sem);
1143         npages = __get_user_pages(tsk, mm, addr, 1, flags, pagep, NULL,
1144                                   &locked);
1145         if (!locked) {
1146                 VM_BUG_ON(npages != -EBUSY);
1147
1148                 if (!pagep)
1149                         return 0;
1150
1151                 /*
1152                  * The previous call has now waited on the IO. Now we can
1153                  * retry and complete. Pass TRIED to ensure we do not re
1154                  * schedule async IO (see e.g. filemap_fault).
1155                  */
1156                 down_read(&mm->mmap_sem);
1157                 npages = __get_user_pages(tsk, mm, addr, 1, flags | FOLL_TRIED,
1158                                           pagep, NULL, NULL);
1159         }
1160         up_read(&mm->mmap_sem);
1161         return npages;
1162 }
1163
1164 static inline int check_user_page_hwpoison(unsigned long addr)
1165 {
1166         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1167
1168         rc = __get_user_pages(current, current->mm, addr, 1,
1169                               flags, NULL, NULL, NULL);
1170         return rc == -EHWPOISON;
1171 }
1172
1173 /*
1174  * The atomic path to get the writable pfn which will be stored in @pfn,
1175  * true indicates success, otherwise false is returned.
1176  */
1177 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1178                             bool write_fault, bool *writable, pfn_t *pfn)
1179 {
1180         struct page *page[1];
1181         int npages;
1182
1183         if (!(async || atomic))
1184                 return false;
1185
1186         /*
1187          * Fast pin a writable pfn only if it is a write fault request
1188          * or the caller allows to map a writable pfn for a read fault
1189          * request.
1190          */
1191         if (!(write_fault || writable))
1192                 return false;
1193
1194         npages = __get_user_pages_fast(addr, 1, 1, page);
1195         if (npages == 1) {
1196                 *pfn = page_to_pfn(page[0]);
1197
1198                 if (writable)
1199                         *writable = true;
1200                 return true;
1201         }
1202
1203         return false;
1204 }
1205
1206 /*
1207  * The slow path to get the pfn of the specified host virtual address,
1208  * 1 indicates success, -errno is returned if error is detected.
1209  */
1210 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1211                            bool *writable, pfn_t *pfn)
1212 {
1213         struct page *page[1];
1214         int npages = 0;
1215
1216         might_sleep();
1217
1218         if (writable)
1219                 *writable = write_fault;
1220
1221         if (async) {
1222                 down_read(&current->mm->mmap_sem);
1223                 npages = get_user_page_nowait(current, current->mm,
1224                                               addr, write_fault, page);
1225                 up_read(&current->mm->mmap_sem);
1226         } else {
1227                 /*
1228                  * By now we have tried gup_fast, and possibly async_pf, and we
1229                  * are certainly not atomic. Time to retry the gup, allowing
1230                  * mmap semaphore to be relinquished in the case of IO.
1231                  */
1232                 npages = kvm_get_user_page_io(current, current->mm, addr,
1233                                               write_fault, page);
1234         }
1235         if (npages != 1)
1236                 return npages;
1237
1238         /* map read fault as writable if possible */
1239         if (unlikely(!write_fault) && writable) {
1240                 struct page *wpage[1];
1241
1242                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1243                 if (npages == 1) {
1244                         *writable = true;
1245                         put_page(page[0]);
1246                         page[0] = wpage[0];
1247                 }
1248
1249                 npages = 1;
1250         }
1251         *pfn = page_to_pfn(page[0]);
1252         return npages;
1253 }
1254
1255 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1256 {
1257         if (unlikely(!(vma->vm_flags & VM_READ)))
1258                 return false;
1259
1260         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1261                 return false;
1262
1263         return true;
1264 }
1265
1266 /*
1267  * Pin guest page in memory and return its pfn.
1268  * @addr: host virtual address which maps memory to the guest
1269  * @atomic: whether this function can sleep
1270  * @async: whether this function need to wait IO complete if the
1271  *         host page is not in the memory
1272  * @write_fault: whether we should get a writable host page
1273  * @writable: whether it allows to map a writable host page for !@write_fault
1274  *
1275  * The function will map a writable host page for these two cases:
1276  * 1): @write_fault = true
1277  * 2): @write_fault = false && @writable, @writable will tell the caller
1278  *     whether the mapping is writable.
1279  */
1280 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1281                         bool write_fault, bool *writable)
1282 {
1283         struct vm_area_struct *vma;
1284         pfn_t pfn = 0;
1285         int npages;
1286
1287         /* we can do it either atomically or asynchronously, not both */
1288         BUG_ON(atomic && async);
1289
1290         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1291                 return pfn;
1292
1293         if (atomic)
1294                 return KVM_PFN_ERR_FAULT;
1295
1296         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1297         if (npages == 1)
1298                 return pfn;
1299
1300         down_read(&current->mm->mmap_sem);
1301         if (npages == -EHWPOISON ||
1302               (!async && check_user_page_hwpoison(addr))) {
1303                 pfn = KVM_PFN_ERR_HWPOISON;
1304                 goto exit;
1305         }
1306
1307         vma = find_vma_intersection(current->mm, addr, addr + 1);
1308
1309         if (vma == NULL)
1310                 pfn = KVM_PFN_ERR_FAULT;
1311         else if ((vma->vm_flags & VM_PFNMAP)) {
1312                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1313                         vma->vm_pgoff;
1314                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1315         } else {
1316                 if (async && vma_is_valid(vma, write_fault))
1317                         *async = true;
1318                 pfn = KVM_PFN_ERR_FAULT;
1319         }
1320 exit:
1321         up_read(&current->mm->mmap_sem);
1322         return pfn;
1323 }
1324
1325 static pfn_t
1326 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1327                      bool *async, bool write_fault, bool *writable)
1328 {
1329         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1330
1331         if (addr == KVM_HVA_ERR_RO_BAD)
1332                 return KVM_PFN_ERR_RO_FAULT;
1333
1334         if (kvm_is_error_hva(addr))
1335                 return KVM_PFN_NOSLOT;
1336
1337         /* Do not map writable pfn in the readonly memslot. */
1338         if (writable && memslot_is_readonly(slot)) {
1339                 *writable = false;
1340                 writable = NULL;
1341         }
1342
1343         return hva_to_pfn(addr, atomic, async, write_fault,
1344                           writable);
1345 }
1346
1347 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1348                           bool write_fault, bool *writable)
1349 {
1350         struct kvm_memory_slot *slot;
1351
1352         if (async)
1353                 *async = false;
1354
1355         slot = gfn_to_memslot(kvm, gfn);
1356
1357         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1358                                     writable);
1359 }
1360
1361 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1362 {
1363         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1364 }
1365 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1366
1367 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1368                        bool write_fault, bool *writable)
1369 {
1370         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1371 }
1372 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1373
1374 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1375 {
1376         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1377 }
1378 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1379
1380 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1381                       bool *writable)
1382 {
1383         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1384 }
1385 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1386
1387 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1388 {
1389         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1390 }
1391
1392 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1393 {
1394         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1395 }
1396 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1397
1398 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1399                                                                   int nr_pages)
1400 {
1401         unsigned long addr;
1402         gfn_t entry;
1403
1404         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1405         if (kvm_is_error_hva(addr))
1406                 return -1;
1407
1408         if (entry < nr_pages)
1409                 return 0;
1410
1411         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1412 }
1413 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1414
1415 static struct page *kvm_pfn_to_page(pfn_t pfn)
1416 {
1417         if (is_error_noslot_pfn(pfn))
1418                 return KVM_ERR_PTR_BAD_PAGE;
1419
1420         if (kvm_is_mmio_pfn(pfn)) {
1421                 WARN_ON(1);
1422                 return KVM_ERR_PTR_BAD_PAGE;
1423         }
1424
1425         return pfn_to_page(pfn);
1426 }
1427
1428 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1429 {
1430         pfn_t pfn;
1431
1432         pfn = gfn_to_pfn(kvm, gfn);
1433
1434         return kvm_pfn_to_page(pfn);
1435 }
1436
1437 EXPORT_SYMBOL_GPL(gfn_to_page);
1438
1439 void kvm_release_page_clean(struct page *page)
1440 {
1441         WARN_ON(is_error_page(page));
1442
1443         kvm_release_pfn_clean(page_to_pfn(page));
1444 }
1445 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1446
1447 void kvm_release_pfn_clean(pfn_t pfn)
1448 {
1449         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1450                 put_page(pfn_to_page(pfn));
1451 }
1452 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1453
1454 void kvm_release_page_dirty(struct page *page)
1455 {
1456         WARN_ON(is_error_page(page));
1457
1458         kvm_release_pfn_dirty(page_to_pfn(page));
1459 }
1460 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1461
1462 static void kvm_release_pfn_dirty(pfn_t pfn)
1463 {
1464         kvm_set_pfn_dirty(pfn);
1465         kvm_release_pfn_clean(pfn);
1466 }
1467
1468 void kvm_set_pfn_dirty(pfn_t pfn)
1469 {
1470         if (!kvm_is_mmio_pfn(pfn)) {
1471                 struct page *page = pfn_to_page(pfn);
1472                 if (!PageReserved(page))
1473                         SetPageDirty(page);
1474         }
1475 }
1476 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1477
1478 void kvm_set_pfn_accessed(pfn_t pfn)
1479 {
1480         if (!kvm_is_mmio_pfn(pfn))
1481                 mark_page_accessed(pfn_to_page(pfn));
1482 }
1483 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1484
1485 void kvm_get_pfn(pfn_t pfn)
1486 {
1487         if (!kvm_is_mmio_pfn(pfn))
1488                 get_page(pfn_to_page(pfn));
1489 }
1490 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1491
1492 static int next_segment(unsigned long len, int offset)
1493 {
1494         if (len > PAGE_SIZE - offset)
1495                 return PAGE_SIZE - offset;
1496         else
1497                 return len;
1498 }
1499
1500 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1501                         int len)
1502 {
1503         int r;
1504         unsigned long addr;
1505
1506         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1507         if (kvm_is_error_hva(addr))
1508                 return -EFAULT;
1509         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1510         if (r)
1511                 return -EFAULT;
1512         return 0;
1513 }
1514 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1515
1516 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1517 {
1518         gfn_t gfn = gpa >> PAGE_SHIFT;
1519         int seg;
1520         int offset = offset_in_page(gpa);
1521         int ret;
1522
1523         while ((seg = next_segment(len, offset)) != 0) {
1524                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1525                 if (ret < 0)
1526                         return ret;
1527                 offset = 0;
1528                 len -= seg;
1529                 data += seg;
1530                 ++gfn;
1531         }
1532         return 0;
1533 }
1534 EXPORT_SYMBOL_GPL(kvm_read_guest);
1535
1536 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1537                           unsigned long len)
1538 {
1539         int r;
1540         unsigned long addr;
1541         gfn_t gfn = gpa >> PAGE_SHIFT;
1542         int offset = offset_in_page(gpa);
1543
1544         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1545         if (kvm_is_error_hva(addr))
1546                 return -EFAULT;
1547         pagefault_disable();
1548         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1549         pagefault_enable();
1550         if (r)
1551                 return -EFAULT;
1552         return 0;
1553 }
1554 EXPORT_SYMBOL(kvm_read_guest_atomic);
1555
1556 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1557                          int offset, int len)
1558 {
1559         int r;
1560         unsigned long addr;
1561
1562         addr = gfn_to_hva(kvm, gfn);
1563         if (kvm_is_error_hva(addr))
1564                 return -EFAULT;
1565         r = __copy_to_user((void __user *)addr + offset, data, len);
1566         if (r)
1567                 return -EFAULT;
1568         mark_page_dirty(kvm, gfn);
1569         return 0;
1570 }
1571 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1572
1573 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1574                     unsigned long len)
1575 {
1576         gfn_t gfn = gpa >> PAGE_SHIFT;
1577         int seg;
1578         int offset = offset_in_page(gpa);
1579         int ret;
1580
1581         while ((seg = next_segment(len, offset)) != 0) {
1582                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1583                 if (ret < 0)
1584                         return ret;
1585                 offset = 0;
1586                 len -= seg;
1587                 data += seg;
1588                 ++gfn;
1589         }
1590         return 0;
1591 }
1592
1593 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1594                               gpa_t gpa, unsigned long len)
1595 {
1596         struct kvm_memslots *slots = kvm_memslots(kvm);
1597         int offset = offset_in_page(gpa);
1598         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1599         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1600         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1601         gfn_t nr_pages_avail;
1602
1603         ghc->gpa = gpa;
1604         ghc->generation = slots->generation;
1605         ghc->len = len;
1606         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1607         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1608         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1609                 ghc->hva += offset;
1610         } else {
1611                 /*
1612                  * If the requested region crosses two memslots, we still
1613                  * verify that the entire region is valid here.
1614                  */
1615                 while (start_gfn <= end_gfn) {
1616                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1617                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1618                                                    &nr_pages_avail);
1619                         if (kvm_is_error_hva(ghc->hva))
1620                                 return -EFAULT;
1621                         start_gfn += nr_pages_avail;
1622                 }
1623                 /* Use the slow path for cross page reads and writes. */
1624                 ghc->memslot = NULL;
1625         }
1626         return 0;
1627 }
1628 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1629
1630 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1631                            void *data, unsigned long len)
1632 {
1633         struct kvm_memslots *slots = kvm_memslots(kvm);
1634         int r;
1635
1636         BUG_ON(len > ghc->len);
1637
1638         if (slots->generation != ghc->generation)
1639                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1640
1641         if (unlikely(!ghc->memslot))
1642                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1643
1644         if (kvm_is_error_hva(ghc->hva))
1645                 return -EFAULT;
1646
1647         r = __copy_to_user((void __user *)ghc->hva, data, len);
1648         if (r)
1649                 return -EFAULT;
1650         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1651
1652         return 0;
1653 }
1654 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1655
1656 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1657                            void *data, unsigned long len)
1658 {
1659         struct kvm_memslots *slots = kvm_memslots(kvm);
1660         int r;
1661
1662         BUG_ON(len > ghc->len);
1663
1664         if (slots->generation != ghc->generation)
1665                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1666
1667         if (unlikely(!ghc->memslot))
1668                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1669
1670         if (kvm_is_error_hva(ghc->hva))
1671                 return -EFAULT;
1672
1673         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1674         if (r)
1675                 return -EFAULT;
1676
1677         return 0;
1678 }
1679 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1680
1681 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1682 {
1683         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1684
1685         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1686 }
1687 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1688
1689 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1690 {
1691         gfn_t gfn = gpa >> PAGE_SHIFT;
1692         int seg;
1693         int offset = offset_in_page(gpa);
1694         int ret;
1695
1696         while ((seg = next_segment(len, offset)) != 0) {
1697                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1698                 if (ret < 0)
1699                         return ret;
1700                 offset = 0;
1701                 len -= seg;
1702                 ++gfn;
1703         }
1704         return 0;
1705 }
1706 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1707
1708 static void mark_page_dirty_in_slot(struct kvm *kvm,
1709                                     struct kvm_memory_slot *memslot,
1710                                     gfn_t gfn)
1711 {
1712         if (memslot && memslot->dirty_bitmap) {
1713                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1714
1715                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1716         }
1717 }
1718
1719 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1720 {
1721         struct kvm_memory_slot *memslot;
1722
1723         memslot = gfn_to_memslot(kvm, gfn);
1724         mark_page_dirty_in_slot(kvm, memslot, gfn);
1725 }
1726 EXPORT_SYMBOL_GPL(mark_page_dirty);
1727
1728 /*
1729  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1730  */
1731 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1732 {
1733         DEFINE_WAIT(wait);
1734
1735         for (;;) {
1736                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1737
1738                 if (kvm_arch_vcpu_runnable(vcpu)) {
1739                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1740                         break;
1741                 }
1742                 if (kvm_cpu_has_pending_timer(vcpu))
1743                         break;
1744                 if (signal_pending(current))
1745                         break;
1746
1747                 schedule();
1748         }
1749
1750         finish_wait(&vcpu->wq, &wait);
1751 }
1752 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1753
1754 #ifndef CONFIG_S390
1755 /*
1756  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1757  */
1758 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1759 {
1760         int me;
1761         int cpu = vcpu->cpu;
1762         wait_queue_head_t *wqp;
1763
1764         wqp = kvm_arch_vcpu_wq(vcpu);
1765         if (waitqueue_active(wqp)) {
1766                 wake_up_interruptible(wqp);
1767                 ++vcpu->stat.halt_wakeup;
1768         }
1769
1770         me = get_cpu();
1771         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1772                 if (kvm_arch_vcpu_should_kick(vcpu))
1773                         smp_send_reschedule(cpu);
1774         put_cpu();
1775 }
1776 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1777 #endif /* !CONFIG_S390 */
1778
1779 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1780 {
1781         struct pid *pid;
1782         struct task_struct *task = NULL;
1783         int ret = 0;
1784
1785         rcu_read_lock();
1786         pid = rcu_dereference(target->pid);
1787         if (pid)
1788                 task = get_pid_task(target->pid, PIDTYPE_PID);
1789         rcu_read_unlock();
1790         if (!task)
1791                 return ret;
1792         if (task->flags & PF_VCPU) {
1793                 put_task_struct(task);
1794                 return ret;
1795         }
1796         ret = yield_to(task, 1);
1797         put_task_struct(task);
1798
1799         return ret;
1800 }
1801 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1802
1803 /*
1804  * Helper that checks whether a VCPU is eligible for directed yield.
1805  * Most eligible candidate to yield is decided by following heuristics:
1806  *
1807  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1808  *  (preempted lock holder), indicated by @in_spin_loop.
1809  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1810  *
1811  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1812  *  chance last time (mostly it has become eligible now since we have probably
1813  *  yielded to lockholder in last iteration. This is done by toggling
1814  *  @dy_eligible each time a VCPU checked for eligibility.)
1815  *
1816  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1817  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1818  *  burning. Giving priority for a potential lock-holder increases lock
1819  *  progress.
1820  *
1821  *  Since algorithm is based on heuristics, accessing another VCPU data without
1822  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1823  *  and continue with next VCPU and so on.
1824  */
1825 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1826 {
1827 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1828         bool eligible;
1829
1830         eligible = !vcpu->spin_loop.in_spin_loop ||
1831                     vcpu->spin_loop.dy_eligible;
1832
1833         if (vcpu->spin_loop.in_spin_loop)
1834                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1835
1836         return eligible;
1837 #else
1838         return true;
1839 #endif
1840 }
1841
1842 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1843 {
1844         struct kvm *kvm = me->kvm;
1845         struct kvm_vcpu *vcpu;
1846         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1847         int yielded = 0;
1848         int try = 3;
1849         int pass;
1850         int i;
1851
1852         kvm_vcpu_set_in_spin_loop(me, true);
1853         /*
1854          * We boost the priority of a VCPU that is runnable but not
1855          * currently running, because it got preempted by something
1856          * else and called schedule in __vcpu_run.  Hopefully that
1857          * VCPU is holding the lock that we need and will release it.
1858          * We approximate round-robin by starting at the last boosted VCPU.
1859          */
1860         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1861                 kvm_for_each_vcpu(i, vcpu, kvm) {
1862                         if (!pass && i <= last_boosted_vcpu) {
1863                                 i = last_boosted_vcpu;
1864                                 continue;
1865                         } else if (pass && i > last_boosted_vcpu)
1866                                 break;
1867                         if (!ACCESS_ONCE(vcpu->preempted))
1868                                 continue;
1869                         if (vcpu == me)
1870                                 continue;
1871                         if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1872                                 continue;
1873                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1874                                 continue;
1875
1876                         yielded = kvm_vcpu_yield_to(vcpu);
1877                         if (yielded > 0) {
1878                                 kvm->last_boosted_vcpu = i;
1879                                 break;
1880                         } else if (yielded < 0) {
1881                                 try--;
1882                                 if (!try)
1883                                         break;
1884                         }
1885                 }
1886         }
1887         kvm_vcpu_set_in_spin_loop(me, false);
1888
1889         /* Ensure vcpu is not eligible during next spinloop */
1890         kvm_vcpu_set_dy_eligible(me, false);
1891 }
1892 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1893
1894 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1895 {
1896         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1897         struct page *page;
1898
1899         if (vmf->pgoff == 0)
1900                 page = virt_to_page(vcpu->run);
1901 #ifdef CONFIG_X86
1902         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1903                 page = virt_to_page(vcpu->arch.pio_data);
1904 #endif
1905 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1906         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1907                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1908 #endif
1909         else
1910                 return kvm_arch_vcpu_fault(vcpu, vmf);
1911         get_page(page);
1912         vmf->page = page;
1913         return 0;
1914 }
1915
1916 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1917         .fault = kvm_vcpu_fault,
1918 };
1919
1920 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1921 {
1922         vma->vm_ops = &kvm_vcpu_vm_ops;
1923         return 0;
1924 }
1925
1926 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1927 {
1928         struct kvm_vcpu *vcpu = filp->private_data;
1929
1930         kvm_put_kvm(vcpu->kvm);
1931         return 0;
1932 }
1933
1934 static struct file_operations kvm_vcpu_fops = {
1935         .release        = kvm_vcpu_release,
1936         .unlocked_ioctl = kvm_vcpu_ioctl,
1937 #ifdef CONFIG_COMPAT
1938         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1939 #endif
1940         .mmap           = kvm_vcpu_mmap,
1941         .llseek         = noop_llseek,
1942 };
1943
1944 /*
1945  * Allocates an inode for the vcpu.
1946  */
1947 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1948 {
1949         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1950 }
1951
1952 /*
1953  * Creates some virtual cpus.  Good luck creating more than one.
1954  */
1955 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1956 {
1957         int r;
1958         struct kvm_vcpu *vcpu, *v;
1959
1960         if (id >= KVM_MAX_VCPUS)
1961                 return -EINVAL;
1962
1963         vcpu = kvm_arch_vcpu_create(kvm, id);
1964         if (IS_ERR(vcpu))
1965                 return PTR_ERR(vcpu);
1966
1967         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1968
1969         r = kvm_arch_vcpu_setup(vcpu);
1970         if (r)
1971                 goto vcpu_destroy;
1972
1973         mutex_lock(&kvm->lock);
1974         if (!kvm_vcpu_compatible(vcpu)) {
1975                 r = -EINVAL;
1976                 goto unlock_vcpu_destroy;
1977         }
1978         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1979                 r = -EINVAL;
1980                 goto unlock_vcpu_destroy;
1981         }
1982
1983         kvm_for_each_vcpu(r, v, kvm)
1984                 if (v->vcpu_id == id) {
1985                         r = -EEXIST;
1986                         goto unlock_vcpu_destroy;
1987                 }
1988
1989         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1990
1991         /* Now it's all set up, let userspace reach it */
1992         kvm_get_kvm(kvm);
1993         r = create_vcpu_fd(vcpu);
1994         if (r < 0) {
1995                 kvm_put_kvm(kvm);
1996                 goto unlock_vcpu_destroy;
1997         }
1998
1999         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2000         smp_wmb();
2001         atomic_inc(&kvm->online_vcpus);
2002
2003         mutex_unlock(&kvm->lock);
2004         kvm_arch_vcpu_postcreate(vcpu);
2005         return r;
2006
2007 unlock_vcpu_destroy:
2008         mutex_unlock(&kvm->lock);
2009 vcpu_destroy:
2010         kvm_arch_vcpu_destroy(vcpu);
2011         return r;
2012 }
2013
2014 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2015 {
2016         if (sigset) {
2017                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2018                 vcpu->sigset_active = 1;
2019                 vcpu->sigset = *sigset;
2020         } else
2021                 vcpu->sigset_active = 0;
2022         return 0;
2023 }
2024
2025 static long kvm_vcpu_ioctl(struct file *filp,
2026                            unsigned int ioctl, unsigned long arg)
2027 {
2028         struct kvm_vcpu *vcpu = filp->private_data;
2029         void __user *argp = (void __user *)arg;
2030         int r;
2031         struct kvm_fpu *fpu = NULL;
2032         struct kvm_sregs *kvm_sregs = NULL;
2033
2034         if (vcpu->kvm->mm != current->mm)
2035                 return -EIO;
2036
2037         if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2038                 return -EINVAL;
2039
2040 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2041         /*
2042          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2043          * so vcpu_load() would break it.
2044          */
2045         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2046                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2047 #endif
2048
2049
2050         r = vcpu_load(vcpu);
2051         if (r)
2052                 return r;
2053         switch (ioctl) {
2054         case KVM_RUN:
2055                 r = -EINVAL;
2056                 if (arg)
2057                         goto out;
2058                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2059                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2060                 break;
2061         case KVM_GET_REGS: {
2062                 struct kvm_regs *kvm_regs;
2063
2064                 r = -ENOMEM;
2065                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2066                 if (!kvm_regs)
2067                         goto out;
2068                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2069                 if (r)
2070                         goto out_free1;
2071                 r = -EFAULT;
2072                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2073                         goto out_free1;
2074                 r = 0;
2075 out_free1:
2076                 kfree(kvm_regs);
2077                 break;
2078         }
2079         case KVM_SET_REGS: {
2080                 struct kvm_regs *kvm_regs;
2081
2082                 r = -ENOMEM;
2083                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2084                 if (IS_ERR(kvm_regs)) {
2085                         r = PTR_ERR(kvm_regs);
2086                         goto out;
2087                 }
2088                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2089                 kfree(kvm_regs);
2090                 break;
2091         }
2092         case KVM_GET_SREGS: {
2093                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2094                 r = -ENOMEM;
2095                 if (!kvm_sregs)
2096                         goto out;
2097                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2098                 if (r)
2099                         goto out;
2100                 r = -EFAULT;
2101                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2102                         goto out;
2103                 r = 0;
2104                 break;
2105         }
2106         case KVM_SET_SREGS: {
2107                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2108                 if (IS_ERR(kvm_sregs)) {
2109                         r = PTR_ERR(kvm_sregs);
2110                         kvm_sregs = NULL;
2111                         goto out;
2112                 }
2113                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2114                 break;
2115         }
2116         case KVM_GET_MP_STATE: {
2117                 struct kvm_mp_state mp_state;
2118
2119                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2120                 if (r)
2121                         goto out;
2122                 r = -EFAULT;
2123                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2124                         goto out;
2125                 r = 0;
2126                 break;
2127         }
2128         case KVM_SET_MP_STATE: {
2129                 struct kvm_mp_state mp_state;
2130
2131                 r = -EFAULT;
2132                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2133                         goto out;
2134                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2135                 break;
2136         }
2137         case KVM_TRANSLATE: {
2138                 struct kvm_translation tr;
2139
2140                 r = -EFAULT;
2141                 if (copy_from_user(&tr, argp, sizeof tr))
2142                         goto out;
2143                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2144                 if (r)
2145                         goto out;
2146                 r = -EFAULT;
2147                 if (copy_to_user(argp, &tr, sizeof tr))
2148                         goto out;
2149                 r = 0;
2150                 break;
2151         }
2152         case KVM_SET_GUEST_DEBUG: {
2153                 struct kvm_guest_debug dbg;
2154
2155                 r = -EFAULT;
2156                 if (copy_from_user(&dbg, argp, sizeof dbg))
2157                         goto out;
2158                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2159                 break;
2160         }
2161         case KVM_SET_SIGNAL_MASK: {
2162                 struct kvm_signal_mask __user *sigmask_arg = argp;
2163                 struct kvm_signal_mask kvm_sigmask;
2164                 sigset_t sigset, *p;
2165
2166                 p = NULL;
2167                 if (argp) {
2168                         r = -EFAULT;
2169                         if (copy_from_user(&kvm_sigmask, argp,
2170                                            sizeof kvm_sigmask))
2171                                 goto out;
2172                         r = -EINVAL;
2173                         if (kvm_sigmask.len != sizeof sigset)
2174                                 goto out;
2175                         r = -EFAULT;
2176                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2177                                            sizeof sigset))
2178                                 goto out;
2179                         p = &sigset;
2180                 }
2181                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2182                 break;
2183         }
2184         case KVM_GET_FPU: {
2185                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2186                 r = -ENOMEM;
2187                 if (!fpu)
2188                         goto out;
2189                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2190                 if (r)
2191                         goto out;
2192                 r = -EFAULT;
2193                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2194                         goto out;
2195                 r = 0;
2196                 break;
2197         }
2198         case KVM_SET_FPU: {
2199                 fpu = memdup_user(argp, sizeof(*fpu));
2200                 if (IS_ERR(fpu)) {
2201                         r = PTR_ERR(fpu);
2202                         fpu = NULL;
2203                         goto out;
2204                 }
2205                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2206                 break;
2207         }
2208         default:
2209                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2210         }
2211 out:
2212         vcpu_put(vcpu);
2213         kfree(fpu);
2214         kfree(kvm_sregs);
2215         return r;
2216 }
2217
2218 #ifdef CONFIG_COMPAT
2219 static long kvm_vcpu_compat_ioctl(struct file *filp,
2220                                   unsigned int ioctl, unsigned long arg)
2221 {
2222         struct kvm_vcpu *vcpu = filp->private_data;
2223         void __user *argp = compat_ptr(arg);
2224         int r;
2225
2226         if (vcpu->kvm->mm != current->mm)
2227                 return -EIO;
2228
2229         switch (ioctl) {
2230         case KVM_SET_SIGNAL_MASK: {
2231                 struct kvm_signal_mask __user *sigmask_arg = argp;
2232                 struct kvm_signal_mask kvm_sigmask;
2233                 compat_sigset_t csigset;
2234                 sigset_t sigset;
2235
2236                 if (argp) {
2237                         r = -EFAULT;
2238                         if (copy_from_user(&kvm_sigmask, argp,
2239                                            sizeof kvm_sigmask))
2240                                 goto out;
2241                         r = -EINVAL;
2242                         if (kvm_sigmask.len != sizeof csigset)
2243                                 goto out;
2244                         r = -EFAULT;
2245                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2246                                            sizeof csigset))
2247                                 goto out;
2248                         sigset_from_compat(&sigset, &csigset);
2249                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2250                 } else
2251                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2252                 break;
2253         }
2254         default:
2255                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2256         }
2257
2258 out:
2259         return r;
2260 }
2261 #endif
2262
2263 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2264                                  int (*accessor)(struct kvm_device *dev,
2265                                                  struct kvm_device_attr *attr),
2266                                  unsigned long arg)
2267 {
2268         struct kvm_device_attr attr;
2269
2270         if (!accessor)
2271                 return -EPERM;
2272
2273         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2274                 return -EFAULT;
2275
2276         return accessor(dev, &attr);
2277 }
2278
2279 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2280                              unsigned long arg)
2281 {
2282         struct kvm_device *dev = filp->private_data;
2283
2284         switch (ioctl) {
2285         case KVM_SET_DEVICE_ATTR:
2286                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2287         case KVM_GET_DEVICE_ATTR:
2288                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2289         case KVM_HAS_DEVICE_ATTR:
2290                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2291         default:
2292                 if (dev->ops->ioctl)
2293                         return dev->ops->ioctl(dev, ioctl, arg);
2294
2295                 return -ENOTTY;
2296         }
2297 }
2298
2299 static int kvm_device_release(struct inode *inode, struct file *filp)
2300 {
2301         struct kvm_device *dev = filp->private_data;
2302         struct kvm *kvm = dev->kvm;
2303
2304         kvm_put_kvm(kvm);
2305         return 0;
2306 }
2307
2308 static const struct file_operations kvm_device_fops = {
2309         .unlocked_ioctl = kvm_device_ioctl,
2310 #ifdef CONFIG_COMPAT
2311         .compat_ioctl = kvm_device_ioctl,
2312 #endif
2313         .release = kvm_device_release,
2314 };
2315
2316 struct kvm_device *kvm_device_from_filp(struct file *filp)
2317 {
2318         if (filp->f_op != &kvm_device_fops)
2319                 return NULL;
2320
2321         return filp->private_data;
2322 }
2323
2324 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2325 #ifdef CONFIG_KVM_MPIC
2326         [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
2327         [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
2328 #endif
2329
2330 #ifdef CONFIG_KVM_XICS
2331         [KVM_DEV_TYPE_XICS]             = &kvm_xics_ops,
2332 #endif
2333 };
2334
2335 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2336 {
2337         if (type >= ARRAY_SIZE(kvm_device_ops_table))
2338                 return -ENOSPC;
2339
2340         if (kvm_device_ops_table[type] != NULL)
2341                 return -EEXIST;
2342
2343         kvm_device_ops_table[type] = ops;
2344         return 0;
2345 }
2346
2347 static int kvm_ioctl_create_device(struct kvm *kvm,
2348                                    struct kvm_create_device *cd)
2349 {
2350         struct kvm_device_ops *ops = NULL;
2351         struct kvm_device *dev;
2352         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2353         int ret;
2354
2355         if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2356                 return -ENODEV;
2357
2358         ops = kvm_device_ops_table[cd->type];
2359         if (ops == NULL)
2360                 return -ENODEV;
2361
2362         if (test)
2363                 return 0;
2364
2365         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2366         if (!dev)
2367                 return -ENOMEM;
2368
2369         dev->ops = ops;
2370         dev->kvm = kvm;
2371
2372         ret = ops->create(dev, cd->type);
2373         if (ret < 0) {
2374                 kfree(dev);
2375                 return ret;
2376         }
2377
2378         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2379         if (ret < 0) {
2380                 ops->destroy(dev);
2381                 return ret;
2382         }
2383
2384         list_add(&dev->vm_node, &kvm->devices);
2385         kvm_get_kvm(kvm);
2386         cd->fd = ret;
2387         return 0;
2388 }
2389
2390 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2391 {
2392         switch (arg) {
2393         case KVM_CAP_USER_MEMORY:
2394         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2395         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2396 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2397         case KVM_CAP_SET_BOOT_CPU_ID:
2398 #endif
2399         case KVM_CAP_INTERNAL_ERROR_DATA:
2400 #ifdef CONFIG_HAVE_KVM_MSI
2401         case KVM_CAP_SIGNAL_MSI:
2402 #endif
2403 #ifdef CONFIG_HAVE_KVM_IRQFD
2404         case KVM_CAP_IRQFD_RESAMPLE:
2405 #endif
2406         case KVM_CAP_CHECK_EXTENSION_VM:
2407                 return 1;
2408 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2409         case KVM_CAP_IRQ_ROUTING:
2410                 return KVM_MAX_IRQ_ROUTES;
2411 #endif
2412         default:
2413                 break;
2414         }
2415         return kvm_vm_ioctl_check_extension(kvm, arg);
2416 }
2417
2418 static long kvm_vm_ioctl(struct file *filp,
2419                            unsigned int ioctl, unsigned long arg)
2420 {
2421         struct kvm *kvm = filp->private_data;
2422         void __user *argp = (void __user *)arg;
2423         int r;
2424
2425         if (kvm->mm != current->mm)
2426                 return -EIO;
2427         switch (ioctl) {
2428         case KVM_CREATE_VCPU:
2429                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2430                 break;
2431         case KVM_SET_USER_MEMORY_REGION: {
2432                 struct kvm_userspace_memory_region kvm_userspace_mem;
2433
2434                 r = -EFAULT;
2435                 if (copy_from_user(&kvm_userspace_mem, argp,
2436                                                 sizeof kvm_userspace_mem))
2437                         goto out;
2438
2439                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2440                 break;
2441         }
2442         case KVM_GET_DIRTY_LOG: {
2443                 struct kvm_dirty_log log;
2444
2445                 r = -EFAULT;
2446                 if (copy_from_user(&log, argp, sizeof log))
2447                         goto out;
2448                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2449                 break;
2450         }
2451 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2452         case KVM_REGISTER_COALESCED_MMIO: {
2453                 struct kvm_coalesced_mmio_zone zone;
2454                 r = -EFAULT;
2455                 if (copy_from_user(&zone, argp, sizeof zone))
2456                         goto out;
2457                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2458                 break;
2459         }
2460         case KVM_UNREGISTER_COALESCED_MMIO: {
2461                 struct kvm_coalesced_mmio_zone zone;
2462                 r = -EFAULT;
2463                 if (copy_from_user(&zone, argp, sizeof zone))
2464                         goto out;
2465                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2466                 break;
2467         }
2468 #endif
2469         case KVM_IRQFD: {
2470                 struct kvm_irqfd data;
2471
2472                 r = -EFAULT;
2473                 if (copy_from_user(&data, argp, sizeof data))
2474                         goto out;
2475                 r = kvm_irqfd(kvm, &data);
2476                 break;
2477         }
2478         case KVM_IOEVENTFD: {
2479                 struct kvm_ioeventfd data;
2480
2481                 r = -EFAULT;
2482                 if (copy_from_user(&data, argp, sizeof data))
2483                         goto out;
2484                 r = kvm_ioeventfd(kvm, &data);
2485                 break;
2486         }
2487 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2488         case KVM_SET_BOOT_CPU_ID:
2489                 r = 0;
2490                 mutex_lock(&kvm->lock);
2491                 if (atomic_read(&kvm->online_vcpus) != 0)
2492                         r = -EBUSY;
2493                 else
2494                         kvm->bsp_vcpu_id = arg;
2495                 mutex_unlock(&kvm->lock);
2496                 break;
2497 #endif
2498 #ifdef CONFIG_HAVE_KVM_MSI
2499         case KVM_SIGNAL_MSI: {
2500                 struct kvm_msi msi;
2501
2502                 r = -EFAULT;
2503                 if (copy_from_user(&msi, argp, sizeof msi))
2504                         goto out;
2505                 r = kvm_send_userspace_msi(kvm, &msi);
2506                 break;
2507         }
2508 #endif
2509 #ifdef __KVM_HAVE_IRQ_LINE
2510         case KVM_IRQ_LINE_STATUS:
2511         case KVM_IRQ_LINE: {
2512                 struct kvm_irq_level irq_event;
2513
2514                 r = -EFAULT;
2515                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2516                         goto out;
2517
2518                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2519                                         ioctl == KVM_IRQ_LINE_STATUS);
2520                 if (r)
2521                         goto out;
2522
2523                 r = -EFAULT;
2524                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2525                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2526                                 goto out;
2527                 }
2528
2529                 r = 0;
2530                 break;
2531         }
2532 #endif
2533 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2534         case KVM_SET_GSI_ROUTING: {
2535                 struct kvm_irq_routing routing;
2536                 struct kvm_irq_routing __user *urouting;
2537                 struct kvm_irq_routing_entry *entries;
2538
2539                 r = -EFAULT;
2540                 if (copy_from_user(&routing, argp, sizeof(routing)))
2541                         goto out;
2542                 r = -EINVAL;
2543                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2544                         goto out;
2545                 if (routing.flags)
2546                         goto out;
2547                 r = -ENOMEM;
2548                 entries = vmalloc(routing.nr * sizeof(*entries));
2549                 if (!entries)
2550                         goto out;
2551                 r = -EFAULT;
2552                 urouting = argp;
2553                 if (copy_from_user(entries, urouting->entries,
2554                                    routing.nr * sizeof(*entries)))
2555                         goto out_free_irq_routing;
2556                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2557                                         routing.flags);
2558         out_free_irq_routing:
2559                 vfree(entries);
2560                 break;
2561         }
2562 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2563         case KVM_CREATE_DEVICE: {
2564                 struct kvm_create_device cd;
2565
2566                 r = -EFAULT;
2567                 if (copy_from_user(&cd, argp, sizeof(cd)))
2568                         goto out;
2569
2570                 r = kvm_ioctl_create_device(kvm, &cd);
2571                 if (r)
2572                         goto out;
2573
2574                 r = -EFAULT;
2575                 if (copy_to_user(argp, &cd, sizeof(cd)))
2576                         goto out;
2577
2578                 r = 0;
2579                 break;
2580         }
2581         case KVM_CHECK_EXTENSION:
2582                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2583                 break;
2584         default:
2585                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2586                 if (r == -ENOTTY)
2587                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2588         }
2589 out:
2590         return r;
2591 }
2592
2593 #ifdef CONFIG_COMPAT
2594 struct compat_kvm_dirty_log {
2595         __u32 slot;
2596         __u32 padding1;
2597         union {
2598                 compat_uptr_t dirty_bitmap; /* one bit per page */
2599                 __u64 padding2;
2600         };
2601 };
2602
2603 static long kvm_vm_compat_ioctl(struct file *filp,
2604                            unsigned int ioctl, unsigned long arg)
2605 {
2606         struct kvm *kvm = filp->private_data;
2607         int r;
2608
2609         if (kvm->mm != current->mm)
2610                 return -EIO;
2611         switch (ioctl) {
2612         case KVM_GET_DIRTY_LOG: {
2613                 struct compat_kvm_dirty_log compat_log;
2614                 struct kvm_dirty_log log;
2615
2616                 r = -EFAULT;
2617                 if (copy_from_user(&compat_log, (void __user *)arg,
2618                                    sizeof(compat_log)))
2619                         goto out;
2620                 log.slot         = compat_log.slot;
2621                 log.padding1     = compat_log.padding1;
2622                 log.padding2     = compat_log.padding2;
2623                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2624
2625                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2626                 break;
2627         }
2628         default:
2629                 r = kvm_vm_ioctl(filp, ioctl, arg);
2630         }
2631
2632 out:
2633         return r;
2634 }
2635 #endif
2636
2637 static struct file_operations kvm_vm_fops = {
2638         .release        = kvm_vm_release,
2639         .unlocked_ioctl = kvm_vm_ioctl,
2640 #ifdef CONFIG_COMPAT
2641         .compat_ioctl   = kvm_vm_compat_ioctl,
2642 #endif
2643         .llseek         = noop_llseek,
2644 };
2645
2646 static int kvm_dev_ioctl_create_vm(unsigned long type)
2647 {
2648         int r;
2649         struct kvm *kvm;
2650
2651         kvm = kvm_create_vm(type);
2652         if (IS_ERR(kvm))
2653                 return PTR_ERR(kvm);
2654 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2655         r = kvm_coalesced_mmio_init(kvm);
2656         if (r < 0) {
2657                 kvm_put_kvm(kvm);
2658                 return r;
2659         }
2660 #endif
2661         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2662         if (r < 0)
2663                 kvm_put_kvm(kvm);
2664
2665         return r;
2666 }
2667
2668 static long kvm_dev_ioctl(struct file *filp,
2669                           unsigned int ioctl, unsigned long arg)
2670 {
2671         long r = -EINVAL;
2672
2673         switch (ioctl) {
2674         case KVM_GET_API_VERSION:
2675                 if (arg)
2676                         goto out;
2677                 r = KVM_API_VERSION;
2678                 break;
2679         case KVM_CREATE_VM:
2680                 r = kvm_dev_ioctl_create_vm(arg);
2681                 break;
2682         case KVM_CHECK_EXTENSION:
2683                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2684                 break;
2685         case KVM_GET_VCPU_MMAP_SIZE:
2686                 if (arg)
2687                         goto out;
2688                 r = PAGE_SIZE;     /* struct kvm_run */
2689 #ifdef CONFIG_X86
2690                 r += PAGE_SIZE;    /* pio data page */
2691 #endif
2692 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2693                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2694 #endif
2695                 break;
2696         case KVM_TRACE_ENABLE:
2697         case KVM_TRACE_PAUSE:
2698         case KVM_TRACE_DISABLE:
2699                 r = -EOPNOTSUPP;
2700                 break;
2701         default:
2702                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2703         }
2704 out:
2705         return r;
2706 }
2707
2708 static struct file_operations kvm_chardev_ops = {
2709         .unlocked_ioctl = kvm_dev_ioctl,
2710         .compat_ioctl   = kvm_dev_ioctl,
2711         .llseek         = noop_llseek,
2712 };
2713
2714 static struct miscdevice kvm_dev = {
2715         KVM_MINOR,
2716         "kvm",
2717         &kvm_chardev_ops,
2718 };
2719
2720 static void hardware_enable_nolock(void *junk)
2721 {
2722         int cpu = raw_smp_processor_id();
2723         int r;
2724
2725         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2726                 return;
2727
2728         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2729
2730         r = kvm_arch_hardware_enable();
2731
2732         if (r) {
2733                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2734                 atomic_inc(&hardware_enable_failed);
2735                 printk(KERN_INFO "kvm: enabling virtualization on "
2736                                  "CPU%d failed\n", cpu);
2737         }
2738 }
2739
2740 static void hardware_enable(void)
2741 {
2742         raw_spin_lock(&kvm_count_lock);
2743         if (kvm_usage_count)
2744                 hardware_enable_nolock(NULL);
2745         raw_spin_unlock(&kvm_count_lock);
2746 }
2747
2748 static void hardware_disable_nolock(void *junk)
2749 {
2750         int cpu = raw_smp_processor_id();
2751
2752         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2753                 return;
2754         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2755         kvm_arch_hardware_disable();
2756 }
2757
2758 static void hardware_disable(void)
2759 {
2760         raw_spin_lock(&kvm_count_lock);
2761         if (kvm_usage_count)
2762                 hardware_disable_nolock(NULL);
2763         raw_spin_unlock(&kvm_count_lock);
2764 }
2765
2766 static void hardware_disable_all_nolock(void)
2767 {
2768         BUG_ON(!kvm_usage_count);
2769
2770         kvm_usage_count--;
2771         if (!kvm_usage_count)
2772                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2773 }
2774
2775 static void hardware_disable_all(void)
2776 {
2777         raw_spin_lock(&kvm_count_lock);
2778         hardware_disable_all_nolock();
2779         raw_spin_unlock(&kvm_count_lock);
2780 }
2781
2782 static int hardware_enable_all(void)
2783 {
2784         int r = 0;
2785
2786         raw_spin_lock(&kvm_count_lock);
2787
2788         kvm_usage_count++;
2789         if (kvm_usage_count == 1) {
2790                 atomic_set(&hardware_enable_failed, 0);
2791                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2792
2793                 if (atomic_read(&hardware_enable_failed)) {
2794                         hardware_disable_all_nolock();
2795                         r = -EBUSY;
2796                 }
2797         }
2798
2799         raw_spin_unlock(&kvm_count_lock);
2800
2801         return r;
2802 }
2803
2804 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2805                            void *v)
2806 {
2807         int cpu = (long)v;
2808
2809         val &= ~CPU_TASKS_FROZEN;
2810         switch (val) {
2811         case CPU_DYING:
2812                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2813                        cpu);
2814                 hardware_disable();
2815                 break;
2816         case CPU_STARTING:
2817                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2818                        cpu);
2819                 hardware_enable();
2820                 break;
2821         }
2822         return NOTIFY_OK;
2823 }
2824
2825 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2826                       void *v)
2827 {
2828         /*
2829          * Some (well, at least mine) BIOSes hang on reboot if
2830          * in vmx root mode.
2831          *
2832          * And Intel TXT required VMX off for all cpu when system shutdown.
2833          */
2834         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2835         kvm_rebooting = true;
2836         on_each_cpu(hardware_disable_nolock, NULL, 1);
2837         return NOTIFY_OK;
2838 }
2839
2840 static struct notifier_block kvm_reboot_notifier = {
2841         .notifier_call = kvm_reboot,
2842         .priority = 0,
2843 };
2844
2845 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2846 {
2847         int i;
2848
2849         for (i = 0; i < bus->dev_count; i++) {
2850                 struct kvm_io_device *pos = bus->range[i].dev;
2851
2852                 kvm_iodevice_destructor(pos);
2853         }
2854         kfree(bus);
2855 }
2856
2857 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2858                                  const struct kvm_io_range *r2)
2859 {
2860         if (r1->addr < r2->addr)
2861                 return -1;
2862         if (r1->addr + r1->len > r2->addr + r2->len)
2863                 return 1;
2864         return 0;
2865 }
2866
2867 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2868 {
2869         return kvm_io_bus_cmp(p1, p2);
2870 }
2871
2872 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2873                           gpa_t addr, int len)
2874 {
2875         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2876                 .addr = addr,
2877                 .len = len,
2878                 .dev = dev,
2879         };
2880
2881         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2882                 kvm_io_bus_sort_cmp, NULL);
2883
2884         return 0;
2885 }
2886
2887 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2888                              gpa_t addr, int len)
2889 {
2890         struct kvm_io_range *range, key;
2891         int off;
2892
2893         key = (struct kvm_io_range) {
2894                 .addr = addr,
2895                 .len = len,
2896         };
2897
2898         range = bsearch(&key, bus->range, bus->dev_count,
2899                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2900         if (range == NULL)
2901                 return -ENOENT;
2902
2903         off = range - bus->range;
2904
2905         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2906                 off--;
2907
2908         return off;
2909 }
2910
2911 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2912                               struct kvm_io_range *range, const void *val)
2913 {
2914         int idx;
2915
2916         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2917         if (idx < 0)
2918                 return -EOPNOTSUPP;
2919
2920         while (idx < bus->dev_count &&
2921                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2922                 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2923                                         range->len, val))
2924                         return idx;
2925                 idx++;
2926         }
2927
2928         return -EOPNOTSUPP;
2929 }
2930
2931 /* kvm_io_bus_write - called under kvm->slots_lock */
2932 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2933                      int len, const void *val)
2934 {
2935         struct kvm_io_bus *bus;
2936         struct kvm_io_range range;
2937         int r;
2938
2939         range = (struct kvm_io_range) {
2940                 .addr = addr,
2941                 .len = len,
2942         };
2943
2944         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2945         r = __kvm_io_bus_write(bus, &range, val);
2946         return r < 0 ? r : 0;
2947 }
2948
2949 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2950 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2951                             int len, const void *val, long cookie)
2952 {
2953         struct kvm_io_bus *bus;
2954         struct kvm_io_range range;
2955
2956         range = (struct kvm_io_range) {
2957                 .addr = addr,
2958                 .len = len,
2959         };
2960
2961         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2962
2963         /* First try the device referenced by cookie. */
2964         if ((cookie >= 0) && (cookie < bus->dev_count) &&
2965             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2966                 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2967                                         val))
2968                         return cookie;
2969
2970         /*
2971          * cookie contained garbage; fall back to search and return the
2972          * correct cookie value.
2973          */
2974         return __kvm_io_bus_write(bus, &range, val);
2975 }
2976
2977 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2978                              void *val)
2979 {
2980         int idx;
2981
2982         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2983         if (idx < 0)
2984                 return -EOPNOTSUPP;
2985
2986         while (idx < bus->dev_count &&
2987                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2988                 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2989                                        range->len, val))
2990                         return idx;
2991                 idx++;
2992         }
2993
2994         return -EOPNOTSUPP;
2995 }
2996 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2997
2998 /* kvm_io_bus_read - called under kvm->slots_lock */
2999 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3000                     int len, void *val)
3001 {
3002         struct kvm_io_bus *bus;
3003         struct kvm_io_range range;
3004         int r;
3005
3006         range = (struct kvm_io_range) {
3007                 .addr = addr,
3008                 .len = len,
3009         };
3010
3011         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3012         r = __kvm_io_bus_read(bus, &range, val);
3013         return r < 0 ? r : 0;
3014 }
3015
3016
3017 /* Caller must hold slots_lock. */
3018 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3019                             int len, struct kvm_io_device *dev)
3020 {
3021         struct kvm_io_bus *new_bus, *bus;
3022
3023         bus = kvm->buses[bus_idx];
3024         /* exclude ioeventfd which is limited by maximum fd */
3025         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3026                 return -ENOSPC;
3027
3028         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3029                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3030         if (!new_bus)
3031                 return -ENOMEM;
3032         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3033                sizeof(struct kvm_io_range)));
3034         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3035         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3036         synchronize_srcu_expedited(&kvm->srcu);
3037         kfree(bus);
3038
3039         return 0;
3040 }
3041
3042 /* Caller must hold slots_lock. */
3043 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3044                               struct kvm_io_device *dev)
3045 {
3046         int i, r;
3047         struct kvm_io_bus *new_bus, *bus;
3048
3049         bus = kvm->buses[bus_idx];
3050         r = -ENOENT;
3051         for (i = 0; i < bus->dev_count; i++)
3052                 if (bus->range[i].dev == dev) {
3053                         r = 0;
3054                         break;
3055                 }
3056
3057         if (r)
3058                 return r;
3059
3060         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3061                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3062         if (!new_bus)
3063                 return -ENOMEM;
3064
3065         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3066         new_bus->dev_count--;
3067         memcpy(new_bus->range + i, bus->range + i + 1,
3068                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3069
3070         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3071         synchronize_srcu_expedited(&kvm->srcu);
3072         kfree(bus);
3073         return r;
3074 }
3075
3076 static struct notifier_block kvm_cpu_notifier = {
3077         .notifier_call = kvm_cpu_hotplug,
3078 };
3079
3080 static int vm_stat_get(void *_offset, u64 *val)
3081 {
3082         unsigned offset = (long)_offset;
3083         struct kvm *kvm;
3084
3085         *val = 0;
3086         spin_lock(&kvm_lock);
3087         list_for_each_entry(kvm, &vm_list, vm_list)
3088                 *val += *(u32 *)((void *)kvm + offset);
3089         spin_unlock(&kvm_lock);
3090         return 0;
3091 }
3092
3093 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3094
3095 static int vcpu_stat_get(void *_offset, u64 *val)
3096 {
3097         unsigned offset = (long)_offset;
3098         struct kvm *kvm;
3099         struct kvm_vcpu *vcpu;
3100         int i;
3101
3102         *val = 0;
3103         spin_lock(&kvm_lock);
3104         list_for_each_entry(kvm, &vm_list, vm_list)
3105                 kvm_for_each_vcpu(i, vcpu, kvm)
3106                         *val += *(u32 *)((void *)vcpu + offset);
3107
3108         spin_unlock(&kvm_lock);
3109         return 0;
3110 }
3111
3112 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3113
3114 static const struct file_operations *stat_fops[] = {
3115         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3116         [KVM_STAT_VM]   = &vm_stat_fops,
3117 };
3118
3119 static int kvm_init_debug(void)
3120 {
3121         int r = -EEXIST;
3122         struct kvm_stats_debugfs_item *p;
3123
3124         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3125         if (kvm_debugfs_dir == NULL)
3126                 goto out;
3127
3128         for (p = debugfs_entries; p->name; ++p) {
3129                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3130                                                 (void *)(long)p->offset,
3131                                                 stat_fops[p->kind]);
3132                 if (p->dentry == NULL)
3133                         goto out_dir;
3134         }
3135
3136         return 0;
3137
3138 out_dir:
3139         debugfs_remove_recursive(kvm_debugfs_dir);
3140 out:
3141         return r;
3142 }
3143
3144 static void kvm_exit_debug(void)
3145 {
3146         struct kvm_stats_debugfs_item *p;
3147
3148         for (p = debugfs_entries; p->name; ++p)
3149                 debugfs_remove(p->dentry);
3150         debugfs_remove(kvm_debugfs_dir);
3151 }
3152
3153 static int kvm_suspend(void)
3154 {
3155         if (kvm_usage_count)
3156                 hardware_disable_nolock(NULL);
3157         return 0;
3158 }
3159
3160 static void kvm_resume(void)
3161 {
3162         if (kvm_usage_count) {
3163                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3164                 hardware_enable_nolock(NULL);
3165         }
3166 }
3167
3168 static struct syscore_ops kvm_syscore_ops = {
3169         .suspend = kvm_suspend,
3170         .resume = kvm_resume,
3171 };
3172
3173 static inline
3174 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3175 {
3176         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3177 }
3178
3179 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3180 {
3181         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3182         if (vcpu->preempted)
3183                 vcpu->preempted = false;
3184
3185         kvm_arch_sched_in(vcpu, cpu);
3186
3187         kvm_arch_vcpu_load(vcpu, cpu);
3188 }
3189
3190 static void kvm_sched_out(struct preempt_notifier *pn,
3191                           struct task_struct *next)
3192 {
3193         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3194
3195         if (current->state == TASK_RUNNING)
3196                 vcpu->preempted = true;
3197         kvm_arch_vcpu_put(vcpu);
3198 }
3199
3200 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3201                   struct module *module)
3202 {
3203         int r;
3204         int cpu;
3205
3206         r = kvm_arch_init(opaque);
3207         if (r)
3208                 goto out_fail;
3209
3210         /*
3211          * kvm_arch_init makes sure there's at most one caller
3212          * for architectures that support multiple implementations,
3213          * like intel and amd on x86.
3214          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3215          * conflicts in case kvm is already setup for another implementation.
3216          */
3217         r = kvm_irqfd_init();
3218         if (r)
3219                 goto out_irqfd;
3220
3221         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3222                 r = -ENOMEM;
3223                 goto out_free_0;
3224         }
3225
3226         r = kvm_arch_hardware_setup();
3227         if (r < 0)
3228                 goto out_free_0a;
3229
3230         for_each_online_cpu(cpu) {
3231                 smp_call_function_single(cpu,
3232                                 kvm_arch_check_processor_compat,
3233                                 &r, 1);
3234                 if (r < 0)
3235                         goto out_free_1;
3236         }
3237
3238         r = register_cpu_notifier(&kvm_cpu_notifier);
3239         if (r)
3240                 goto out_free_2;
3241         register_reboot_notifier(&kvm_reboot_notifier);
3242
3243         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3244         if (!vcpu_align)
3245                 vcpu_align = __alignof__(struct kvm_vcpu);
3246         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3247                                            0, NULL);
3248         if (!kvm_vcpu_cache) {
3249                 r = -ENOMEM;
3250                 goto out_free_3;
3251         }
3252
3253         r = kvm_async_pf_init();
3254         if (r)
3255                 goto out_free;
3256
3257         kvm_chardev_ops.owner = module;
3258         kvm_vm_fops.owner = module;
3259         kvm_vcpu_fops.owner = module;
3260
3261         r = misc_register(&kvm_dev);
3262         if (r) {
3263                 printk(KERN_ERR "kvm: misc device register failed\n");
3264                 goto out_unreg;
3265         }
3266
3267         register_syscore_ops(&kvm_syscore_ops);
3268
3269         kvm_preempt_ops.sched_in = kvm_sched_in;
3270         kvm_preempt_ops.sched_out = kvm_sched_out;
3271
3272         r = kvm_init_debug();
3273         if (r) {
3274                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3275                 goto out_undebugfs;
3276         }
3277
3278         r = kvm_vfio_ops_init();
3279         WARN_ON(r);
3280
3281         return 0;
3282
3283 out_undebugfs:
3284         unregister_syscore_ops(&kvm_syscore_ops);
3285         misc_deregister(&kvm_dev);
3286 out_unreg:
3287         kvm_async_pf_deinit();
3288 out_free:
3289         kmem_cache_destroy(kvm_vcpu_cache);
3290 out_free_3:
3291         unregister_reboot_notifier(&kvm_reboot_notifier);
3292         unregister_cpu_notifier(&kvm_cpu_notifier);
3293 out_free_2:
3294 out_free_1:
3295         kvm_arch_hardware_unsetup();
3296 out_free_0a:
3297         free_cpumask_var(cpus_hardware_enabled);
3298 out_free_0:
3299         kvm_irqfd_exit();
3300 out_irqfd:
3301         kvm_arch_exit();
3302 out_fail:
3303         return r;
3304 }
3305 EXPORT_SYMBOL_GPL(kvm_init);
3306
3307 void kvm_exit(void)
3308 {
3309         kvm_exit_debug();
3310         misc_deregister(&kvm_dev);
3311         kmem_cache_destroy(kvm_vcpu_cache);
3312         kvm_async_pf_deinit();
3313         unregister_syscore_ops(&kvm_syscore_ops);
3314         unregister_reboot_notifier(&kvm_reboot_notifier);
3315         unregister_cpu_notifier(&kvm_cpu_notifier);
3316         on_each_cpu(hardware_disable_nolock, NULL, 1);
3317         kvm_arch_hardware_unsetup();
3318         kvm_arch_exit();
3319         kvm_irqfd_exit();
3320         free_cpumask_var(cpus_hardware_enabled);
3321 }
3322 EXPORT_SYMBOL_GPL(kvm_exit);