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 !is_zero_pfn(pfn) && 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 bool kvm_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 (kvm_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         kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 202 }
 203
 204 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 205 {
 206         kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 207 }
 208
 209 void kvm_make_scan_ioapic_request(struct kvm *kvm)
 210 {
 211         kvm_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
 300         kvm_arch_mmu_notifier_invalidate_page(kvm, address);
 301
 302         srcu_read_unlock(&kvm->srcu, idx);
 303 }
 304
 305 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 306                                         struct mm_struct *mm,
 307                                         unsigned long address,
 308                                         pte_t pte)
 309 {
 310         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 311         int idx;
 312
 313         idx = srcu_read_lock(&kvm->srcu);
 314         spin_lock(&kvm->mmu_lock);
 315         kvm->mmu_notifier_seq++;
 316         kvm_set_spte_hva(kvm, address, pte);
 317         spin_unlock(&kvm->mmu_lock);
 318         srcu_read_unlock(&kvm->srcu, idx);
 319 }
 320
 321 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 322                                                     struct mm_struct *mm,
 323                                                     unsigned long start,
 324                                                     unsigned long end)
 325 {
 326         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 327         int need_tlb_flush = 0, idx;
 328
 329         idx = srcu_read_lock(&kvm->srcu);
 330         spin_lock(&kvm->mmu_lock);
 331         /*
 332          * The count increase must become visible at unlock time as no
 333          * spte can be established without taking the mmu_lock and
 334          * count is also read inside the mmu_lock critical section.
 335          */
 336         kvm->mmu_notifier_count++;
 337         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 338         need_tlb_flush |= kvm->tlbs_dirty;
 339         /* we've to flush the tlb before the pages can be freed */
 340         if (need_tlb_flush)
 341                 kvm_flush_remote_tlbs(kvm);
 342
 343         spin_unlock(&kvm->mmu_lock);
 344         srcu_read_unlock(&kvm->srcu, idx);
 345 }
 346
 347 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 348                                                   struct mm_struct *mm,
 349                                                   unsigned long start,
 350                                                   unsigned long end)
 351 {
 352         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 353
 354         spin_lock(&kvm->mmu_lock);
 355         /*
 356          * This sequence increase will notify the kvm page fault that
 357          * the page that is going to be mapped in the spte could have
 358          * been freed.
 359          */
 360         kvm->mmu_notifier_seq++;
 361         smp_wmb();
 362         /*
 363          * The above sequence increase must be visible before the
 364          * below count decrease, which is ensured by the smp_wmb above
 365          * in conjunction with the smp_rmb in mmu_notifier_retry().
 366          */
 367         kvm->mmu_notifier_count--;
 368         spin_unlock(&kvm->mmu_lock);
 369
 370         BUG_ON(kvm->mmu_notifier_count < 0);
 371 }
 372
 373 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 374                                               struct mm_struct *mm,
 375                                               unsigned long start,
 376                                               unsigned long end)
 377 {
 378         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 379         int young, idx;
 380
 381         idx = srcu_read_lock(&kvm->srcu);
 382         spin_lock(&kvm->mmu_lock);
 383
 384         young = kvm_age_hva(kvm, start, end);
 385         if (young)
 386                 kvm_flush_remote_tlbs(kvm);
 387
 388         spin_unlock(&kvm->mmu_lock);
 389         srcu_read_unlock(&kvm->srcu, idx);
 390
 391         return young;
 392 }
 393
 394 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 395                                        struct mm_struct *mm,
 396                                        unsigned long address)
 397 {
 398         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 399         int young, idx;
 400
 401         idx = srcu_read_lock(&kvm->srcu);
 402         spin_lock(&kvm->mmu_lock);
 403         young = kvm_test_age_hva(kvm, address);
 404         spin_unlock(&kvm->mmu_lock);
 405         srcu_read_unlock(&kvm->srcu, idx);
 406
 407         return young;
 408 }
 409
 410 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 411                                      struct mm_struct *mm)
 412 {
 413         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 414         int idx;
 415
 416         idx = srcu_read_lock(&kvm->srcu);
 417         kvm_arch_flush_shadow_all(kvm);
 418         srcu_read_unlock(&kvm->srcu, idx);
 419 }
 420
 421 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 422         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 423         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 424         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 425         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 426         .test_young             = kvm_mmu_notifier_test_young,
 427         .change_pte             = kvm_mmu_notifier_change_pte,
 428         .release                = kvm_mmu_notifier_release,
 429 };
 430
 431 static int kvm_init_mmu_notifier(struct kvm *kvm)
 432 {
 433         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 434         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 435 }
 436
 437 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 438
 439 static int kvm_init_mmu_notifier(struct kvm *kvm)
 440 {
 441         return 0;
 442 }
 443
 444 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 445
 446 static void kvm_init_memslots_id(struct kvm *kvm)
 447 {
 448         int i;
 449         struct kvm_memslots *slots = kvm->memslots;
 450
 451         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 452                 slots->id_to_index[i] = slots->memslots[i].id = i;
 453 }
 454
 455 static struct kvm *kvm_create_vm(unsigned long type)
 456 {
 457         int r, i;
 458         struct kvm *kvm = kvm_arch_alloc_vm();
 459
 460         if (!kvm)
 461                 return ERR_PTR(-ENOMEM);
 462
 463         r = kvm_arch_init_vm(kvm, type);
 464         if (r)
 465                 goto out_err_no_disable;
 466
 467         r = hardware_enable_all();
 468         if (r)
 469                 goto out_err_no_disable;
 470
 471 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 472         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 473 #endif
 474 #ifdef CONFIG_HAVE_KVM_IRQFD
 475         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 476 #endif
 477
 478         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 479
 480         r = -ENOMEM;
 481         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 482         if (!kvm->memslots)
 483                 goto out_err_no_srcu;
 484
 485         /*
 486          * Init kvm generation close to the maximum to easily test the
 487          * code of handling generation number wrap-around.
 488          */
 489         kvm->memslots->generation = -150;
 490
 491         kvm_init_memslots_id(kvm);
 492         if (init_srcu_struct(&kvm->srcu))
 493                 goto out_err_no_srcu;
 494         if (init_srcu_struct(&kvm->irq_srcu))
 495                 goto out_err_no_irq_srcu;
 496         for (i = 0; i < KVM_NR_BUSES; i++) {
 497                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 498                                         GFP_KERNEL);
 499                 if (!kvm->buses[i])
 500                         goto out_err;
 501         }
 502
 503         spin_lock_init(&kvm->mmu_lock);
 504         kvm->mm = current->mm;
 505         atomic_inc(&kvm->mm->mm_count);
 506         kvm_eventfd_init(kvm);
 507         mutex_init(&kvm->lock);
 508         mutex_init(&kvm->irq_lock);
 509         mutex_init(&kvm->slots_lock);
 510         atomic_set(&kvm->users_count, 1);
 511         INIT_LIST_HEAD(&kvm->devices);
 512
 513         r = kvm_init_mmu_notifier(kvm);
 514         if (r)
 515                 goto out_err;
 516
 517         spin_lock(&kvm_lock);
 518         list_add(&kvm->vm_list, &vm_list);
 519         spin_unlock(&kvm_lock);
 520
 521         return kvm;
 522
 523 out_err:
 524         cleanup_srcu_struct(&kvm->irq_srcu);
 525 out_err_no_irq_srcu:
 526         cleanup_srcu_struct(&kvm->srcu);
 527 out_err_no_srcu:
 528         hardware_disable_all();
 529 out_err_no_disable:
 530         for (i = 0; i < KVM_NR_BUSES; i++)
 531                 kfree(kvm->buses[i]);
 532         kfree(kvm->memslots);
 533         kvm_arch_free_vm(kvm);
 534         return ERR_PTR(r);
 535 }
 536
 537 /*
 538  * Avoid using vmalloc for a small buffer.
 539  * Should not be used when the size is statically known.
 540  */
 541 void *kvm_kvzalloc(unsigned long size)
 542 {
 543         if (size > PAGE_SIZE)
 544                 return vzalloc(size);
 545         else
 546                 return kzalloc(size, GFP_KERNEL);
 547 }
 548
 549 void kvm_kvfree(const void *addr)
 550 {
 551         if (is_vmalloc_addr(addr))
 552                 vfree(addr);
 553         else
 554                 kfree(addr);
 555 }
 556
 557 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 558 {
 559         if (!memslot->dirty_bitmap)
 560                 return;
 561
 562         kvm_kvfree(memslot->dirty_bitmap);
 563         memslot->dirty_bitmap = NULL;
 564 }
 565
 566 /*
 567  * Free any memory in @free but not in @dont.
 568  */
 569 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
 570                                   struct kvm_memory_slot *dont)
 571 {
 572         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 573                 kvm_destroy_dirty_bitmap(free);
 574
 575         kvm_arch_free_memslot(kvm, free, dont);
 576
 577         free->npages = 0;
 578 }
 579
 580 static void kvm_free_physmem(struct kvm *kvm)
 581 {
 582         struct kvm_memslots *slots = kvm->memslots;
 583         struct kvm_memory_slot *memslot;
 584
 585         kvm_for_each_memslot(memslot, slots)
 586                 kvm_free_physmem_slot(kvm, memslot, NULL);
 587
 588         kfree(kvm->memslots);
 589 }
 590
 591 static void kvm_destroy_devices(struct kvm *kvm)
 592 {
 593         struct list_head *node, *tmp;
 594
 595         list_for_each_safe(node, tmp, &kvm->devices) {
 596                 struct kvm_device *dev =
 597                         list_entry(node, struct kvm_device, vm_node);
 598
 599                 list_del(node);
 600                 dev->ops->destroy(dev);
 601         }
 602 }
 603
 604 static void kvm_destroy_vm(struct kvm *kvm)
 605 {
 606         int i;
 607         struct mm_struct *mm = kvm->mm;
 608
 609         kvm_arch_sync_events(kvm);
 610         spin_lock(&kvm_lock);
 611         list_del(&kvm->vm_list);
 612         spin_unlock(&kvm_lock);
 613         kvm_free_irq_routing(kvm);
 614         for (i = 0; i < KVM_NR_BUSES; i++)
 615                 kvm_io_bus_destroy(kvm->buses[i]);
 616         kvm_coalesced_mmio_free(kvm);
 617 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 618         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 619 #else
 620         kvm_arch_flush_shadow_all(kvm);
 621 #endif
 622         kvm_arch_destroy_vm(kvm);
 623         kvm_destroy_devices(kvm);
 624         kvm_free_physmem(kvm);
 625         cleanup_srcu_struct(&kvm->irq_srcu);
 626         cleanup_srcu_struct(&kvm->srcu);
 627         kvm_arch_free_vm(kvm);
 628         hardware_disable_all();
 629         mmdrop(mm);
 630 }
 631
 632 void kvm_get_kvm(struct kvm *kvm)
 633 {
 634         atomic_inc(&kvm->users_count);
 635 }
 636 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 637
 638 void kvm_put_kvm(struct kvm *kvm)
 639 {
 640         if (atomic_dec_and_test(&kvm->users_count))
 641                 kvm_destroy_vm(kvm);
 642 }
 643 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 644
 645
 646 static int kvm_vm_release(struct inode *inode, struct file *filp)
 647 {
 648         struct kvm *kvm = filp->private_data;
 649
 650         kvm_irqfd_release(kvm);
 651
 652         kvm_put_kvm(kvm);
 653         return 0;
 654 }
 655
 656 /*
 657  * Allocation size is twice as large as the actual dirty bitmap size.
 658  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 659  */
 660 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 661 {
 662         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 663
 664         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 665         if (!memslot->dirty_bitmap)
 666                 return -ENOMEM;
 667
 668         return 0;
 669 }
 670
 671 static int cmp_memslot(const void *slot1, const void *slot2)
 672 {
 673         struct kvm_memory_slot *s1, *s2;
 674
 675         s1 = (struct kvm_memory_slot *)slot1;
 676         s2 = (struct kvm_memory_slot *)slot2;
 677
 678         if (s1->npages < s2->npages)
 679                 return 1;
 680         if (s1->npages > s2->npages)
 681                 return -1;
 682
 683         return 0;
 684 }
 685
 686 /*
 687  * Sort the memslots base on its size, so the larger slots
 688  * will get better fit.
 689  */
 690 static void sort_memslots(struct kvm_memslots *slots)
 691 {
 692         int i;
 693
 694         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 695               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 696
 697         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 698                 slots->id_to_index[slots->memslots[i].id] = i;
 699 }
 700
 701 static void update_memslots(struct kvm_memslots *slots,
 702                             struct kvm_memory_slot *new)
 703 {
 704         if (new) {
 705                 int id = new->id;
 706                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 707                 unsigned long npages = old->npages;
 708
 709                 *old = *new;
 710                 if (new->npages != npages)
 711                         sort_memslots(slots);
 712         }
 713 }
 714
 715 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 716 {
 717         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 718
 719 #ifdef __KVM_HAVE_READONLY_MEM
 720         valid_flags |= KVM_MEM_READONLY;
 721 #endif
 722
 723         if (mem->flags & ~valid_flags)
 724                 return -EINVAL;
 725
 726         return 0;
 727 }
 728
 729 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 730                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 731 {
 732         struct kvm_memslots *old_memslots = kvm->memslots;
 733
 734         /*
 735          * Set the low bit in the generation, which disables SPTE caching
 736          * until the end of synchronize_srcu_expedited.
 737          */
 738         WARN_ON(old_memslots->generation & 1);
 739         slots->generation = old_memslots->generation + 1;
 740
 741         update_memslots(slots, new);
 742         rcu_assign_pointer(kvm->memslots, slots);
 743         synchronize_srcu_expedited(&kvm->srcu);
 744
 745         /*
 746          * Increment the new memslot generation a second time. This prevents
 747          * vm exits that race with memslot updates from caching a memslot
 748          * generation that will (potentially) be valid forever.
 749          */
 750         slots->generation++;
 751
 752         kvm_arch_memslots_updated(kvm);
 753
 754         return old_memslots;
 755 }
 756
 757 /*
 758  * Allocate some memory and give it an address in the guest physical address
 759  * space.
 760  *
 761  * Discontiguous memory is allowed, mostly for framebuffers.
 762  *
 763  * Must be called holding mmap_sem for write.
 764  */
 765 int __kvm_set_memory_region(struct kvm *kvm,
 766                             struct kvm_userspace_memory_region *mem)
 767 {
 768         int r;
 769         gfn_t base_gfn;
 770         unsigned long npages;
 771         struct kvm_memory_slot *slot;
 772         struct kvm_memory_slot old, new;
 773         struct kvm_memslots *slots = NULL, *old_memslots;
 774         enum kvm_mr_change change;
 775
 776         r = check_memory_region_flags(mem);
 777         if (r)
 778                 goto out;
 779
 780         r = -EINVAL;
 781         /* General sanity checks */
 782         if (mem->memory_size & (PAGE_SIZE - 1))
 783                 goto out;
 784         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 785                 goto out;
 786         /* We can read the guest memory with __xxx_user() later on. */
 787         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 788             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 789              !access_ok(VERIFY_WRITE,
 790                         (void __user *)(unsigned long)mem->userspace_addr,
 791                         mem->memory_size)))
 792                 goto out;
 793         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 794                 goto out;
 795         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 796                 goto out;
 797
 798         slot = id_to_memslot(kvm->memslots, mem->slot);
 799         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 800         npages = mem->memory_size >> PAGE_SHIFT;
 801
 802         if (npages > KVM_MEM_MAX_NR_PAGES)
 803                 goto out;
 804
 805         if (!npages)
 806                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 807
 808         new = old = *slot;
 809
 810         new.id = mem->slot;
 811         new.base_gfn = base_gfn;
 812         new.npages = npages;
 813         new.flags = mem->flags;
 814
 815         if (npages) {
 816                 if (!old.npages)
 817                         change = KVM_MR_CREATE;
 818                 else { /* Modify an existing slot. */
 819                         if ((mem->userspace_addr != old.userspace_addr) ||
 820                             (npages != old.npages) ||
 821                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 822                                 goto out;
 823
 824                         if (base_gfn != old.base_gfn)
 825                                 change = KVM_MR_MOVE;
 826                         else if (new.flags != old.flags)
 827                                 change = KVM_MR_FLAGS_ONLY;
 828                         else { /* Nothing to change. */
 829                                 r = 0;
 830                                 goto out;
 831                         }
 832                 }
 833         } else if (old.npages) {
 834                 change = KVM_MR_DELETE;
 835         } else /* Modify a non-existent slot: disallowed. */
 836                 goto out;
 837
 838         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 839                 /* Check for overlaps */
 840                 r = -EEXIST;
 841                 kvm_for_each_memslot(slot, kvm->memslots) {
 842                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 843                             (slot->id == mem->slot))
 844                                 continue;
 845                         if (!((base_gfn + npages <= slot->base_gfn) ||
 846                               (base_gfn >= slot->base_gfn + slot->npages)))
 847                                 goto out;
 848                 }
 849         }
 850
 851         /* Free page dirty bitmap if unneeded */
 852         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 853                 new.dirty_bitmap = NULL;
 854
 855         r = -ENOMEM;
 856         if (change == KVM_MR_CREATE) {
 857                 new.userspace_addr = mem->userspace_addr;
 858
 859                 if (kvm_arch_create_memslot(kvm, &new, npages))
 860                         goto out_free;
 861         }
 862
 863         /* Allocate page dirty bitmap if needed */
 864         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 865                 if (kvm_create_dirty_bitmap(&new) < 0)
 866                         goto out_free;
 867         }
 868
 869         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 870                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 871                                 GFP_KERNEL);
 872                 if (!slots)
 873                         goto out_free;
 874                 slot = id_to_memslot(slots, mem->slot);
 875                 slot->flags |= KVM_MEMSLOT_INVALID;
 876
 877                 old_memslots = install_new_memslots(kvm, slots, NULL);
 878
 879                 /* slot was deleted or moved, clear iommu mapping */
 880                 kvm_iommu_unmap_pages(kvm, &old);
 881                 /* From this point no new shadow pages pointing to a deleted,
 882                  * or moved, memslot will be created.
 883                  *
 884                  * validation of sp->gfn happens in:
 885                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 886                  *      - kvm_is_visible_gfn (mmu_check_roots)
 887                  */
 888                 kvm_arch_flush_shadow_memslot(kvm, slot);
 889                 slots = old_memslots;
 890         }
 891
 892         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 893         if (r)
 894                 goto out_slots;
 895
 896         r = -ENOMEM;
 897         /*
 898          * We can re-use the old_memslots from above, the only difference
 899          * from the currently installed memslots is the invalid flag.  This
 900          * will get overwritten by update_memslots anyway.
 901          */
 902         if (!slots) {
 903                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 904                                 GFP_KERNEL);
 905                 if (!slots)
 906                         goto out_free;
 907         }
 908
 909         /* actual memory is freed via old in kvm_free_physmem_slot below */
 910         if (change == KVM_MR_DELETE) {
 911                 new.dirty_bitmap = NULL;
 912                 memset(&new.arch, 0, sizeof(new.arch));
 913         }
 914
 915         old_memslots = install_new_memslots(kvm, slots, &new);
 916
 917         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 918
 919         kvm_free_physmem_slot(kvm, &old, &new);
 920         kfree(old_memslots);
 921
 922         /*
 923          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 924          * un-mapped and re-mapped if their base changes.  Since base change
 925          * unmapping is handled above with slot deletion, mapping alone is
 926          * needed here.  Anything else the iommu might care about for existing
 927          * slots (size changes, userspace addr changes and read-only flag
 928          * changes) is disallowed above, so any other attribute changes getting
 929          * here can be skipped.
 930          */
 931         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 932                 r = kvm_iommu_map_pages(kvm, &new);
 933                 return r;
 934         }
 935
 936         return 0;
 937
 938 out_slots:
 939         kfree(slots);
 940 out_free:
 941         kvm_free_physmem_slot(kvm, &new, &old);
 942 out:
 943         return r;
 944 }
 945 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 946
 947 int kvm_set_memory_region(struct kvm *kvm,
 948                           struct kvm_userspace_memory_region *mem)
 949 {
 950         int r;
 951
 952         mutex_lock(&kvm->slots_lock);
 953         r = __kvm_set_memory_region(kvm, mem);
 954         mutex_unlock(&kvm->slots_lock);
 955         return r;
 956 }
 957 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 958
 959 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 960                                           struct kvm_userspace_memory_region *mem)
 961 {
 962         if (mem->slot >= KVM_USER_MEM_SLOTS)
 963                 return -EINVAL;
 964         return kvm_set_memory_region(kvm, mem);
 965 }
 966
 967 int kvm_get_dirty_log(struct kvm *kvm,
 968                         struct kvm_dirty_log *log, int *is_dirty)
 969 {
 970         struct kvm_memory_slot *memslot;
 971         int r, i;
 972         unsigned long n;
 973         unsigned long any = 0;
 974
 975         r = -EINVAL;
 976         if (log->slot >= KVM_USER_MEM_SLOTS)
 977                 goto out;
 978
 979         memslot = id_to_memslot(kvm->memslots, log->slot);
 980         r = -ENOENT;
 981         if (!memslot->dirty_bitmap)
 982                 goto out;
 983
 984         n = kvm_dirty_bitmap_bytes(memslot);
 985
 986         for (i = 0; !any && i < n/sizeof(long); ++i)
 987                 any = memslot->dirty_bitmap[i];
 988
 989         r = -EFAULT;
 990         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 991                 goto out;
 992
 993         if (any)
 994                 *is_dirty = 1;
 995
 996         r = 0;
 997 out:
 998         return r;
 999 }
1000 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1001
1002 bool kvm_largepages_enabled(void)
1003 {
1004         return largepages_enabled;
1005 }
1006
1007 void kvm_disable_largepages(void)
1008 {
1009         largepages_enabled = false;
1010 }
1011 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1012
1013 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1014 {
1015         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1016 }
1017 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1018
1019 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1020 {
1021         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1022
1023         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1024               memslot->flags & KVM_MEMSLOT_INVALID)
1025                 return 0;
1026
1027         return 1;
1028 }
1029 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1030
1031 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1032 {
1033         struct vm_area_struct *vma;
1034         unsigned long addr, size;
1035
1036         size = PAGE_SIZE;
1037
1038         addr = gfn_to_hva(kvm, gfn);
1039         if (kvm_is_error_hva(addr))
1040                 return PAGE_SIZE;
1041
1042         down_read(&current->mm->mmap_sem);
1043         vma = find_vma(current->mm, addr);
1044         if (!vma)
1045                 goto out;
1046
1047         size = vma_kernel_pagesize(vma);
1048
1049 out:
1050         up_read(&current->mm->mmap_sem);
1051
1052         return size;
1053 }
1054
1055 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1056 {
1057         return slot->flags & KVM_MEM_READONLY;
1058 }
1059
1060 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1061                                        gfn_t *nr_pages, bool write)
1062 {
1063         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1064                 return KVM_HVA_ERR_BAD;
1065
1066         if (memslot_is_readonly(slot) && write)
1067                 return KVM_HVA_ERR_RO_BAD;
1068
1069         if (nr_pages)
1070                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1071
1072         return __gfn_to_hva_memslot(slot, gfn);
1073 }
1074
1075 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1076                                      gfn_t *nr_pages)
1077 {
1078         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1079 }
1080
1081 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1082                                         gfn_t gfn)
1083 {
1084         return gfn_to_hva_many(slot, gfn, NULL);
1085 }
1086 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1087
1088 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1089 {
1090         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1091 }
1092 EXPORT_SYMBOL_GPL(gfn_to_hva);
1093
1094 /*
1095  * If writable is set to false, the hva returned by this function is only
1096  * allowed to be read.
1097  */
1098 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1099                                       gfn_t gfn, bool *writable)
1100 {
1101         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1102
1103         if (!kvm_is_error_hva(hva) && writable)
1104                 *writable = !memslot_is_readonly(slot);
1105
1106         return hva;
1107 }
1108
1109 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1110 {
1111         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1112
1113         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1114 }
1115
1116 static int kvm_read_hva(void *data, void __user *hva, int len)
1117 {
1118         return __copy_from_user(data, hva, len);
1119 }
1120
1121 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1122 {
1123         return __copy_from_user_inatomic(data, hva, len);
1124 }
1125
1126 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1127         unsigned long start, int write, struct page **page)
1128 {
1129         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1130
1131         if (write)
1132                 flags |= FOLL_WRITE;
1133
1134         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1135 }
1136
1137 int kvm_get_user_page_io(struct task_struct *tsk, struct mm_struct *mm,
1138                          unsigned long addr, bool write_fault,
1139                          struct page **pagep)
1140 {
1141         int npages;
1142         int locked = 1;
1143         int flags = FOLL_TOUCH | FOLL_HWPOISON |
1144                     (pagep ? FOLL_GET : 0) |
1145                     (write_fault ? FOLL_WRITE : 0);
1146
1147         /*
1148          * If retrying the fault, we get here *not* having allowed the filemap
1149          * to wait on the page lock. We should now allow waiting on the IO with
1150          * the mmap semaphore released.
1151          */
1152         down_read(&mm->mmap_sem);
1153         npages = __get_user_pages(tsk, mm, addr, 1, flags, pagep, NULL,
1154                                   &locked);
1155         if (!locked) {
1156                 VM_BUG_ON(npages);
1157
1158                 if (!pagep)
1159                         return 0;
1160
1161                 /*
1162                  * The previous call has now waited on the IO. Now we can
1163                  * retry and complete. Pass TRIED to ensure we do not re
1164                  * schedule async IO (see e.g. filemap_fault).
1165                  */
1166                 down_read(&mm->mmap_sem);
1167                 npages = __get_user_pages(tsk, mm, addr, 1, flags | FOLL_TRIED,
1168                                           pagep, NULL, NULL);
1169         }
1170         up_read(&mm->mmap_sem);
1171         return npages;
1172 }
1173
1174 static inline int check_user_page_hwpoison(unsigned long addr)
1175 {
1176         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1177
1178         rc = __get_user_pages(current, current->mm, addr, 1,
1179                               flags, NULL, NULL, NULL);
1180         return rc == -EHWPOISON;
1181 }
1182
1183 /*
1184  * The atomic path to get the writable pfn which will be stored in @pfn,
1185  * true indicates success, otherwise false is returned.
1186  */
1187 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1188                             bool write_fault, bool *writable, pfn_t *pfn)
1189 {
1190         struct page *page[1];
1191         int npages;
1192
1193         if (!(async || atomic))
1194                 return false;
1195
1196         /*
1197          * Fast pin a writable pfn only if it is a write fault request
1198          * or the caller allows to map a writable pfn for a read fault
1199          * request.
1200          */
1201         if (!(write_fault || writable))
1202                 return false;
1203
1204         npages = __get_user_pages_fast(addr, 1, 1, page);
1205         if (npages == 1) {
1206                 *pfn = page_to_pfn(page[0]);
1207
1208                 if (writable)
1209                         *writable = true;
1210                 return true;
1211         }
1212
1213         return false;
1214 }
1215
1216 /*
1217  * The slow path to get the pfn of the specified host virtual address,
1218  * 1 indicates success, -errno is returned if error is detected.
1219  */
1220 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1221                            bool *writable, pfn_t *pfn)
1222 {
1223         struct page *page[1];
1224         int npages = 0;
1225
1226         might_sleep();
1227
1228         if (writable)
1229                 *writable = write_fault;
1230
1231         if (async) {
1232                 down_read(&current->mm->mmap_sem);
1233                 npages = get_user_page_nowait(current, current->mm,
1234                                               addr, write_fault, page);
1235                 up_read(&current->mm->mmap_sem);
1236         } else {
1237                 /*
1238                  * By now we have tried gup_fast, and possibly async_pf, and we
1239                  * are certainly not atomic. Time to retry the gup, allowing
1240                  * mmap semaphore to be relinquished in the case of IO.
1241                  */
1242                 npages = kvm_get_user_page_io(current, current->mm, addr,
1243                                               write_fault, page);
1244         }
1245         if (npages != 1)
1246                 return npages;
1247
1248         /* map read fault as writable if possible */
1249         if (unlikely(!write_fault) && writable) {
1250                 struct page *wpage[1];
1251
1252                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1253                 if (npages == 1) {
1254                         *writable = true;
1255                         put_page(page[0]);
1256                         page[0] = wpage[0];
1257                 }
1258
1259                 npages = 1;
1260         }
1261         *pfn = page_to_pfn(page[0]);
1262         return npages;
1263 }
1264
1265 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1266 {
1267         if (unlikely(!(vma->vm_flags & VM_READ)))
1268                 return false;
1269
1270         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1271                 return false;
1272
1273         return true;
1274 }
1275
1276 /*
1277  * Pin guest page in memory and return its pfn.
1278  * @addr: host virtual address which maps memory to the guest
1279  * @atomic: whether this function can sleep
1280  * @async: whether this function need to wait IO complete if the
1281  *         host page is not in the memory
1282  * @write_fault: whether we should get a writable host page
1283  * @writable: whether it allows to map a writable host page for !@write_fault
1284  *
1285  * The function will map a writable host page for these two cases:
1286  * 1): @write_fault = true
1287  * 2): @write_fault = false && @writable, @writable will tell the caller
1288  *     whether the mapping is writable.
1289  */
1290 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1291                         bool write_fault, bool *writable)
1292 {
1293         struct vm_area_struct *vma;
1294         pfn_t pfn = 0;
1295         int npages;
1296
1297         /* we can do it either atomically or asynchronously, not both */
1298         BUG_ON(atomic && async);
1299
1300         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1301                 return pfn;
1302
1303         if (atomic)
1304                 return KVM_PFN_ERR_FAULT;
1305
1306         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1307         if (npages == 1)
1308                 return pfn;
1309
1310         down_read(&current->mm->mmap_sem);
1311         if (npages == -EHWPOISON ||
1312               (!async && check_user_page_hwpoison(addr))) {
1313                 pfn = KVM_PFN_ERR_HWPOISON;
1314                 goto exit;
1315         }
1316
1317         vma = find_vma_intersection(current->mm, addr, addr + 1);
1318
1319         if (vma == NULL)
1320                 pfn = KVM_PFN_ERR_FAULT;
1321         else if ((vma->vm_flags & VM_PFNMAP)) {
1322                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1323                         vma->vm_pgoff;
1324                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1325         } else {
1326                 if (async && vma_is_valid(vma, write_fault))
1327                         *async = true;
1328                 pfn = KVM_PFN_ERR_FAULT;
1329         }
1330 exit:
1331         up_read(&current->mm->mmap_sem);
1332         return pfn;
1333 }
1334
1335 static pfn_t
1336 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1337                      bool *async, bool write_fault, bool *writable)
1338 {
1339         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1340
1341         if (addr == KVM_HVA_ERR_RO_BAD)
1342                 return KVM_PFN_ERR_RO_FAULT;
1343
1344         if (kvm_is_error_hva(addr))
1345                 return KVM_PFN_NOSLOT;
1346
1347         /* Do not map writable pfn in the readonly memslot. */
1348         if (writable && memslot_is_readonly(slot)) {
1349                 *writable = false;
1350                 writable = NULL;
1351         }
1352
1353         return hva_to_pfn(addr, atomic, async, write_fault,
1354                           writable);
1355 }
1356
1357 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1358                           bool write_fault, bool *writable)
1359 {
1360         struct kvm_memory_slot *slot;
1361
1362         if (async)
1363                 *async = false;
1364
1365         slot = gfn_to_memslot(kvm, gfn);
1366
1367         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1368                                     writable);
1369 }
1370
1371 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1372 {
1373         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1374 }
1375 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1376
1377 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1378                        bool write_fault, bool *writable)
1379 {
1380         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1381 }
1382 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1383
1384 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1385 {
1386         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1387 }
1388 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1389
1390 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1391                       bool *writable)
1392 {
1393         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1394 }
1395 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1396
1397 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1398 {
1399         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1400 }
1401
1402 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1403 {
1404         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1405 }
1406 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1407
1408 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1409                                                                   int nr_pages)
1410 {
1411         unsigned long addr;
1412         gfn_t entry;
1413
1414         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1415         if (kvm_is_error_hva(addr))
1416                 return -1;
1417
1418         if (entry < nr_pages)
1419                 return 0;
1420
1421         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1422 }
1423 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1424
1425 static struct page *kvm_pfn_to_page(pfn_t pfn)
1426 {
1427         if (is_error_noslot_pfn(pfn))
1428                 return KVM_ERR_PTR_BAD_PAGE;
1429
1430         if (kvm_is_mmio_pfn(pfn)) {
1431                 WARN_ON(1);
1432                 return KVM_ERR_PTR_BAD_PAGE;
1433         }
1434
1435         return pfn_to_page(pfn);
1436 }
1437
1438 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1439 {
1440         pfn_t pfn;
1441
1442         pfn = gfn_to_pfn(kvm, gfn);
1443
1444         return kvm_pfn_to_page(pfn);
1445 }
1446
1447 EXPORT_SYMBOL_GPL(gfn_to_page);
1448
1449 void kvm_release_page_clean(struct page *page)
1450 {
1451         WARN_ON(is_error_page(page));
1452
1453         kvm_release_pfn_clean(page_to_pfn(page));
1454 }
1455 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1456
1457 void kvm_release_pfn_clean(pfn_t pfn)
1458 {
1459         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1460                 put_page(pfn_to_page(pfn));
1461 }
1462 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1463
1464 void kvm_release_page_dirty(struct page *page)
1465 {
1466         WARN_ON(is_error_page(page));
1467
1468         kvm_release_pfn_dirty(page_to_pfn(page));
1469 }
1470 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1471
1472 static void kvm_release_pfn_dirty(pfn_t pfn)
1473 {
1474         kvm_set_pfn_dirty(pfn);
1475         kvm_release_pfn_clean(pfn);
1476 }
1477
1478 void kvm_set_pfn_dirty(pfn_t pfn)
1479 {
1480         if (!kvm_is_mmio_pfn(pfn)) {
1481                 struct page *page = pfn_to_page(pfn);
1482                 if (!PageReserved(page))
1483                         SetPageDirty(page);
1484         }
1485 }
1486 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1487
1488 void kvm_set_pfn_accessed(pfn_t pfn)
1489 {
1490         if (!kvm_is_mmio_pfn(pfn))
1491                 mark_page_accessed(pfn_to_page(pfn));
1492 }
1493 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1494
1495 void kvm_get_pfn(pfn_t pfn)
1496 {
1497         if (!kvm_is_mmio_pfn(pfn))
1498                 get_page(pfn_to_page(pfn));
1499 }
1500 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1501
1502 static int next_segment(unsigned long len, int offset)
1503 {
1504         if (len > PAGE_SIZE - offset)
1505                 return PAGE_SIZE - offset;
1506         else
1507                 return len;
1508 }
1509
1510 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1511                         int len)
1512 {
1513         int r;
1514         unsigned long addr;
1515
1516         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1517         if (kvm_is_error_hva(addr))
1518                 return -EFAULT;
1519         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1520         if (r)
1521                 return -EFAULT;
1522         return 0;
1523 }
1524 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1525
1526 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1527 {
1528         gfn_t gfn = gpa >> PAGE_SHIFT;
1529         int seg;
1530         int offset = offset_in_page(gpa);
1531         int ret;
1532
1533         while ((seg = next_segment(len, offset)) != 0) {
1534                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1535                 if (ret < 0)
1536                         return ret;
1537                 offset = 0;
1538                 len -= seg;
1539                 data += seg;
1540                 ++gfn;
1541         }
1542         return 0;
1543 }
1544 EXPORT_SYMBOL_GPL(kvm_read_guest);
1545
1546 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1547                           unsigned long len)
1548 {
1549         int r;
1550         unsigned long addr;
1551         gfn_t gfn = gpa >> PAGE_SHIFT;
1552         int offset = offset_in_page(gpa);
1553
1554         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1555         if (kvm_is_error_hva(addr))
1556                 return -EFAULT;
1557         pagefault_disable();
1558         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1559         pagefault_enable();
1560         if (r)
1561                 return -EFAULT;
1562         return 0;
1563 }
1564 EXPORT_SYMBOL(kvm_read_guest_atomic);
1565
1566 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1567                          int offset, int len)
1568 {
1569         int r;
1570         unsigned long addr;
1571
1572         addr = gfn_to_hva(kvm, gfn);
1573         if (kvm_is_error_hva(addr))
1574                 return -EFAULT;
1575         r = __copy_to_user((void __user *)addr + offset, data, len);
1576         if (r)
1577                 return -EFAULT;
1578         mark_page_dirty(kvm, gfn);
1579         return 0;
1580 }
1581 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1582
1583 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1584                     unsigned long len)
1585 {
1586         gfn_t gfn = gpa >> PAGE_SHIFT;
1587         int seg;
1588         int offset = offset_in_page(gpa);
1589         int ret;
1590
1591         while ((seg = next_segment(len, offset)) != 0) {
1592                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1593                 if (ret < 0)
1594                         return ret;
1595                 offset = 0;
1596                 len -= seg;
1597                 data += seg;
1598                 ++gfn;
1599         }
1600         return 0;
1601 }
1602
1603 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1604                               gpa_t gpa, unsigned long len)
1605 {
1606         struct kvm_memslots *slots = kvm_memslots(kvm);
1607         int offset = offset_in_page(gpa);
1608         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1609         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1610         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1611         gfn_t nr_pages_avail;
1612
1613         ghc->gpa = gpa;
1614         ghc->generation = slots->generation;
1615         ghc->len = len;
1616         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1617         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1618         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1619                 ghc->hva += offset;
1620         } else {
1621                 /*
1622                  * If the requested region crosses two memslots, we still
1623                  * verify that the entire region is valid here.
1624                  */
1625                 while (start_gfn <= end_gfn) {
1626                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1627                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1628                                                    &nr_pages_avail);
1629                         if (kvm_is_error_hva(ghc->hva))
1630                                 return -EFAULT;
1631                         start_gfn += nr_pages_avail;
1632                 }
1633                 /* Use the slow path for cross page reads and writes. */
1634                 ghc->memslot = NULL;
1635         }
1636         return 0;
1637 }
1638 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1639
1640 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1641                            void *data, unsigned long len)
1642 {
1643         struct kvm_memslots *slots = kvm_memslots(kvm);
1644         int r;
1645
1646         BUG_ON(len > ghc->len);
1647
1648         if (slots->generation != ghc->generation)
1649                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1650
1651         if (unlikely(!ghc->memslot))
1652                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1653
1654         if (kvm_is_error_hva(ghc->hva))
1655                 return -EFAULT;
1656
1657         r = __copy_to_user((void __user *)ghc->hva, data, len);
1658         if (r)
1659                 return -EFAULT;
1660         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1661
1662         return 0;
1663 }
1664 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1665
1666 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1667                            void *data, unsigned long len)
1668 {
1669         struct kvm_memslots *slots = kvm_memslots(kvm);
1670         int r;
1671
1672         BUG_ON(len > ghc->len);
1673
1674         if (slots->generation != ghc->generation)
1675                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1676
1677         if (unlikely(!ghc->memslot))
1678                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1679
1680         if (kvm_is_error_hva(ghc->hva))
1681                 return -EFAULT;
1682
1683         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1684         if (r)
1685                 return -EFAULT;
1686
1687         return 0;
1688 }
1689 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1690
1691 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1692 {
1693         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1694
1695         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1696 }
1697 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1698
1699 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1700 {
1701         gfn_t gfn = gpa >> PAGE_SHIFT;
1702         int seg;
1703         int offset = offset_in_page(gpa);
1704         int ret;
1705
1706         while ((seg = next_segment(len, offset)) != 0) {
1707                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1708                 if (ret < 0)
1709                         return ret;
1710                 offset = 0;
1711                 len -= seg;
1712                 ++gfn;
1713         }
1714         return 0;
1715 }
1716 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1717
1718 static void mark_page_dirty_in_slot(struct kvm *kvm,
1719                                     struct kvm_memory_slot *memslot,
1720                                     gfn_t gfn)
1721 {
1722         if (memslot && memslot->dirty_bitmap) {
1723                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1724
1725                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1726         }
1727 }
1728
1729 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1730 {
1731         struct kvm_memory_slot *memslot;
1732
1733         memslot = gfn_to_memslot(kvm, gfn);
1734         mark_page_dirty_in_slot(kvm, memslot, gfn);
1735 }
1736 EXPORT_SYMBOL_GPL(mark_page_dirty);
1737
1738 /*
1739  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1740  */
1741 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1742 {
1743         DEFINE_WAIT(wait);
1744
1745         for (;;) {
1746                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1747
1748                 if (kvm_arch_vcpu_runnable(vcpu)) {
1749                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1750                         break;
1751                 }
1752                 if (kvm_cpu_has_pending_timer(vcpu))
1753                         break;
1754                 if (signal_pending(current))
1755                         break;
1756
1757                 schedule();
1758         }
1759
1760         finish_wait(&vcpu->wq, &wait);
1761 }
1762 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1763
1764 #ifndef CONFIG_S390
1765 /*
1766  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1767  */
1768 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1769 {
1770         int me;
1771         int cpu = vcpu->cpu;
1772         wait_queue_head_t *wqp;
1773
1774         wqp = kvm_arch_vcpu_wq(vcpu);
1775         if (waitqueue_active(wqp)) {
1776                 wake_up_interruptible(wqp);
1777                 ++vcpu->stat.halt_wakeup;
1778         }
1779
1780         me = get_cpu();
1781         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1782                 if (kvm_arch_vcpu_should_kick(vcpu))
1783                         smp_send_reschedule(cpu);
1784         put_cpu();
1785 }
1786 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1787 #endif /* !CONFIG_S390 */
1788
1789 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1790 {
1791         struct pid *pid;
1792         struct task_struct *task = NULL;
1793         int ret = 0;
1794
1795         rcu_read_lock();
1796         pid = rcu_dereference(target->pid);
1797         if (pid)
1798                 task = get_pid_task(pid, PIDTYPE_PID);
1799         rcu_read_unlock();
1800         if (!task)
1801                 return ret;
1802         if (task->flags & PF_VCPU) {
1803                 put_task_struct(task);
1804                 return ret;
1805         }
1806         ret = yield_to(task, 1);
1807         put_task_struct(task);
1808
1809         return ret;
1810 }
1811 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1812
1813 /*
1814  * Helper that checks whether a VCPU is eligible for directed yield.
1815  * Most eligible candidate to yield is decided by following heuristics:
1816  *
1817  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1818  *  (preempted lock holder), indicated by @in_spin_loop.
1819  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1820  *
1821  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1822  *  chance last time (mostly it has become eligible now since we have probably
1823  *  yielded to lockholder in last iteration. This is done by toggling
1824  *  @dy_eligible each time a VCPU checked for eligibility.)
1825  *
1826  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1827  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1828  *  burning. Giving priority for a potential lock-holder increases lock
1829  *  progress.
1830  *
1831  *  Since algorithm is based on heuristics, accessing another VCPU data without
1832  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1833  *  and continue with next VCPU and so on.
1834  */
1835 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1836 {
1837 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1838         bool eligible;
1839
1840         eligible = !vcpu->spin_loop.in_spin_loop ||
1841                     vcpu->spin_loop.dy_eligible;
1842
1843         if (vcpu->spin_loop.in_spin_loop)
1844                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1845
1846         return eligible;
1847 #else
1848         return true;
1849 #endif
1850 }
1851
1852 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1853 {
1854         struct kvm *kvm = me->kvm;
1855         struct kvm_vcpu *vcpu;
1856         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1857         int yielded = 0;
1858         int try = 3;
1859         int pass;
1860         int i;
1861
1862         kvm_vcpu_set_in_spin_loop(me, true);
1863         /*
1864          * We boost the priority of a VCPU that is runnable but not
1865          * currently running, because it got preempted by something
1866          * else and called schedule in __vcpu_run.  Hopefully that
1867          * VCPU is holding the lock that we need and will release it.
1868          * We approximate round-robin by starting at the last boosted VCPU.
1869          */
1870         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1871                 kvm_for_each_vcpu(i, vcpu, kvm) {
1872                         if (!pass && i <= last_boosted_vcpu) {
1873                                 i = last_boosted_vcpu;
1874                                 continue;
1875                         } else if (pass && i > last_boosted_vcpu)
1876                                 break;
1877                         if (!ACCESS_ONCE(vcpu->preempted))
1878                                 continue;
1879                         if (vcpu == me)
1880                                 continue;
1881                         if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1882                                 continue;
1883                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1884                                 continue;
1885
1886                         yielded = kvm_vcpu_yield_to(vcpu);
1887                         if (yielded > 0) {
1888                                 kvm->last_boosted_vcpu = i;
1889                                 break;
1890                         } else if (yielded < 0) {
1891                                 try--;
1892                                 if (!try)
1893                                         break;
1894                         }
1895                 }
1896         }
1897         kvm_vcpu_set_in_spin_loop(me, false);
1898
1899         /* Ensure vcpu is not eligible during next spinloop */
1900         kvm_vcpu_set_dy_eligible(me, false);
1901 }
1902 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1903
1904 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1905 {
1906         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1907         struct page *page;
1908
1909         if (vmf->pgoff == 0)
1910                 page = virt_to_page(vcpu->run);
1911 #ifdef CONFIG_X86
1912         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1913                 page = virt_to_page(vcpu->arch.pio_data);
1914 #endif
1915 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1916         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1917                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1918 #endif
1919         else
1920                 return kvm_arch_vcpu_fault(vcpu, vmf);
1921         get_page(page);
1922         vmf->page = page;
1923         return 0;
1924 }
1925
1926 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1927         .fault = kvm_vcpu_fault,
1928 };
1929
1930 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1931 {
1932         vma->vm_ops = &kvm_vcpu_vm_ops;
1933         return 0;
1934 }
1935
1936 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1937 {
1938         struct kvm_vcpu *vcpu = filp->private_data;
1939
1940         kvm_put_kvm(vcpu->kvm);
1941         return 0;
1942 }
1943
1944 static struct file_operations kvm_vcpu_fops = {
1945         .release        = kvm_vcpu_release,
1946         .unlocked_ioctl = kvm_vcpu_ioctl,
1947 #ifdef CONFIG_COMPAT
1948         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1949 #endif
1950         .mmap           = kvm_vcpu_mmap,
1951         .llseek         = noop_llseek,
1952 };
1953
1954 /*
1955  * Allocates an inode for the vcpu.
1956  */
1957 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1958 {
1959         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1960 }
1961
1962 /*
1963  * Creates some virtual cpus.  Good luck creating more than one.
1964  */
1965 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1966 {
1967         int r;
1968         struct kvm_vcpu *vcpu, *v;
1969
1970         if (id >= KVM_MAX_VCPUS)
1971                 return -EINVAL;
1972
1973         vcpu = kvm_arch_vcpu_create(kvm, id);
1974         if (IS_ERR(vcpu))
1975                 return PTR_ERR(vcpu);
1976
1977         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1978
1979         r = kvm_arch_vcpu_setup(vcpu);
1980         if (r)
1981                 goto vcpu_destroy;
1982
1983         mutex_lock(&kvm->lock);
1984         if (!kvm_vcpu_compatible(vcpu)) {
1985                 r = -EINVAL;
1986                 goto unlock_vcpu_destroy;
1987         }
1988         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1989                 r = -EINVAL;
1990                 goto unlock_vcpu_destroy;
1991         }
1992
1993         kvm_for_each_vcpu(r, v, kvm)
1994                 if (v->vcpu_id == id) {
1995                         r = -EEXIST;
1996                         goto unlock_vcpu_destroy;
1997                 }
1998
1999         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2000
2001         /* Now it's all set up, let userspace reach it */
2002         kvm_get_kvm(kvm);
2003         r = create_vcpu_fd(vcpu);
2004         if (r < 0) {
2005                 kvm_put_kvm(kvm);
2006                 goto unlock_vcpu_destroy;
2007         }
2008
2009         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2010         smp_wmb();
2011         atomic_inc(&kvm->online_vcpus);
2012
2013         mutex_unlock(&kvm->lock);
2014         kvm_arch_vcpu_postcreate(vcpu);
2015         return r;
2016
2017 unlock_vcpu_destroy:
2018         mutex_unlock(&kvm->lock);
2019 vcpu_destroy:
2020         kvm_arch_vcpu_destroy(vcpu);
2021         return r;
2022 }
2023
2024 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2025 {
2026         if (sigset) {
2027                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2028                 vcpu->sigset_active = 1;
2029                 vcpu->sigset = *sigset;
2030         } else
2031                 vcpu->sigset_active = 0;
2032         return 0;
2033 }
2034
2035 static long kvm_vcpu_ioctl(struct file *filp,
2036                            unsigned int ioctl, unsigned long arg)
2037 {
2038         struct kvm_vcpu *vcpu = filp->private_data;
2039         void __user *argp = (void __user *)arg;
2040         int r;
2041         struct kvm_fpu *fpu = NULL;
2042         struct kvm_sregs *kvm_sregs = NULL;
2043
2044         if (vcpu->kvm->mm != current->mm)
2045                 return -EIO;
2046
2047         if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2048                 return -EINVAL;
2049
2050 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2051         /*
2052          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2053          * so vcpu_load() would break it.
2054          */
2055         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2056                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2057 #endif
2058
2059
2060         r = vcpu_load(vcpu);
2061         if (r)
2062                 return r;
2063         switch (ioctl) {
2064         case KVM_RUN:
2065                 r = -EINVAL;
2066                 if (arg)
2067                         goto out;
2068                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2069                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2070                 break;
2071         case KVM_GET_REGS: {
2072                 struct kvm_regs *kvm_regs;
2073
2074                 r = -ENOMEM;
2075                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2076                 if (!kvm_regs)
2077                         goto out;
2078                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2079                 if (r)
2080                         goto out_free1;
2081                 r = -EFAULT;
2082                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2083                         goto out_free1;
2084                 r = 0;
2085 out_free1:
2086                 kfree(kvm_regs);
2087                 break;
2088         }
2089         case KVM_SET_REGS: {
2090                 struct kvm_regs *kvm_regs;
2091
2092                 r = -ENOMEM;
2093                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2094                 if (IS_ERR(kvm_regs)) {
2095                         r = PTR_ERR(kvm_regs);
2096                         goto out;
2097                 }
2098                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2099                 kfree(kvm_regs);
2100                 break;
2101         }
2102         case KVM_GET_SREGS: {
2103                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2104                 r = -ENOMEM;
2105                 if (!kvm_sregs)
2106                         goto out;
2107                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2108                 if (r)
2109                         goto out;
2110                 r = -EFAULT;
2111                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2112                         goto out;
2113                 r = 0;
2114                 break;
2115         }
2116         case KVM_SET_SREGS: {
2117                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2118                 if (IS_ERR(kvm_sregs)) {
2119                         r = PTR_ERR(kvm_sregs);
2120                         kvm_sregs = NULL;
2121                         goto out;
2122                 }
2123                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2124                 break;
2125         }
2126         case KVM_GET_MP_STATE: {
2127                 struct kvm_mp_state mp_state;
2128
2129                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2130                 if (r)
2131                         goto out;
2132                 r = -EFAULT;
2133                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2134                         goto out;
2135                 r = 0;
2136                 break;
2137         }
2138         case KVM_SET_MP_STATE: {
2139                 struct kvm_mp_state mp_state;
2140
2141                 r = -EFAULT;
2142                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2143                         goto out;
2144                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2145                 break;
2146         }
2147         case KVM_TRANSLATE: {
2148                 struct kvm_translation tr;
2149
2150                 r = -EFAULT;
2151                 if (copy_from_user(&tr, argp, sizeof tr))
2152                         goto out;
2153                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2154                 if (r)
2155                         goto out;
2156                 r = -EFAULT;
2157                 if (copy_to_user(argp, &tr, sizeof tr))
2158                         goto out;
2159                 r = 0;
2160                 break;
2161         }
2162         case KVM_SET_GUEST_DEBUG: {
2163                 struct kvm_guest_debug dbg;
2164
2165                 r = -EFAULT;
2166                 if (copy_from_user(&dbg, argp, sizeof dbg))
2167                         goto out;
2168                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2169                 break;
2170         }
2171         case KVM_SET_SIGNAL_MASK: {
2172                 struct kvm_signal_mask __user *sigmask_arg = argp;
2173                 struct kvm_signal_mask kvm_sigmask;
2174                 sigset_t sigset, *p;
2175
2176                 p = NULL;
2177                 if (argp) {
2178                         r = -EFAULT;
2179                         if (copy_from_user(&kvm_sigmask, argp,
2180                                            sizeof kvm_sigmask))
2181                                 goto out;
2182                         r = -EINVAL;
2183                         if (kvm_sigmask.len != sizeof sigset)
2184                                 goto out;
2185                         r = -EFAULT;
2186                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2187                                            sizeof sigset))
2188                                 goto out;
2189                         p = &sigset;
2190                 }
2191                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2192                 break;
2193         }
2194         case KVM_GET_FPU: {
2195                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2196                 r = -ENOMEM;
2197                 if (!fpu)
2198                         goto out;
2199                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2200                 if (r)
2201                         goto out;
2202                 r = -EFAULT;
2203                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2204                         goto out;
2205                 r = 0;
2206                 break;
2207         }
2208         case KVM_SET_FPU: {
2209                 fpu = memdup_user(argp, sizeof(*fpu));
2210                 if (IS_ERR(fpu)) {
2211                         r = PTR_ERR(fpu);
2212                         fpu = NULL;
2213                         goto out;
2214                 }
2215                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2216                 break;
2217         }
2218         default:
2219                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2220         }
2221 out:
2222         vcpu_put(vcpu);
2223         kfree(fpu);
2224         kfree(kvm_sregs);
2225         return r;
2226 }
2227
2228 #ifdef CONFIG_COMPAT
2229 static long kvm_vcpu_compat_ioctl(struct file *filp,
2230                                   unsigned int ioctl, unsigned long arg)
2231 {
2232         struct kvm_vcpu *vcpu = filp->private_data;
2233         void __user *argp = compat_ptr(arg);
2234         int r;
2235
2236         if (vcpu->kvm->mm != current->mm)
2237                 return -EIO;
2238
2239         switch (ioctl) {
2240         case KVM_SET_SIGNAL_MASK: {
2241                 struct kvm_signal_mask __user *sigmask_arg = argp;
2242                 struct kvm_signal_mask kvm_sigmask;
2243                 compat_sigset_t csigset;
2244                 sigset_t sigset;
2245
2246                 if (argp) {
2247                         r = -EFAULT;
2248                         if (copy_from_user(&kvm_sigmask, argp,
2249                                            sizeof kvm_sigmask))
2250                                 goto out;
2251                         r = -EINVAL;
2252                         if (kvm_sigmask.len != sizeof csigset)
2253                                 goto out;
2254                         r = -EFAULT;
2255                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2256                                            sizeof csigset))
2257                                 goto out;
2258                         sigset_from_compat(&sigset, &csigset);
2259                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2260                 } else
2261                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2262                 break;
2263         }
2264         default:
2265                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2266         }
2267
2268 out:
2269         return r;
2270 }
2271 #endif
2272
2273 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2274                                  int (*accessor)(struct kvm_device *dev,
2275                                                  struct kvm_device_attr *attr),
2276                                  unsigned long arg)
2277 {
2278         struct kvm_device_attr attr;
2279
2280         if (!accessor)
2281                 return -EPERM;
2282
2283         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2284                 return -EFAULT;
2285
2286         return accessor(dev, &attr);
2287 }
2288
2289 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2290                              unsigned long arg)
2291 {
2292         struct kvm_device *dev = filp->private_data;
2293
2294         switch (ioctl) {
2295         case KVM_SET_DEVICE_ATTR:
2296                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2297         case KVM_GET_DEVICE_ATTR:
2298                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2299         case KVM_HAS_DEVICE_ATTR:
2300                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2301         default:
2302                 if (dev->ops->ioctl)
2303                         return dev->ops->ioctl(dev, ioctl, arg);
2304
2305                 return -ENOTTY;
2306         }
2307 }
2308
2309 static int kvm_device_release(struct inode *inode, struct file *filp)
2310 {
2311         struct kvm_device *dev = filp->private_data;
2312         struct kvm *kvm = dev->kvm;
2313
2314         kvm_put_kvm(kvm);
2315         return 0;
2316 }
2317
2318 static const struct file_operations kvm_device_fops = {
2319         .unlocked_ioctl = kvm_device_ioctl,
2320 #ifdef CONFIG_COMPAT
2321         .compat_ioctl = kvm_device_ioctl,
2322 #endif
2323         .release = kvm_device_release,
2324 };
2325
2326 struct kvm_device *kvm_device_from_filp(struct file *filp)
2327 {
2328         if (filp->f_op != &kvm_device_fops)
2329                 return NULL;
2330
2331         return filp->private_data;
2332 }
2333
2334 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2335 #ifdef CONFIG_KVM_MPIC
2336         [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
2337         [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
2338 #endif
2339
2340 #ifdef CONFIG_KVM_XICS
2341         [KVM_DEV_TYPE_XICS]             = &kvm_xics_ops,
2342 #endif
2343 };
2344
2345 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2346 {
2347         if (type >= ARRAY_SIZE(kvm_device_ops_table))
2348                 return -ENOSPC;
2349
2350         if (kvm_device_ops_table[type] != NULL)
2351                 return -EEXIST;
2352
2353         kvm_device_ops_table[type] = ops;
2354         return 0;
2355 }
2356
2357 static int kvm_ioctl_create_device(struct kvm *kvm,
2358                                    struct kvm_create_device *cd)
2359 {
2360         struct kvm_device_ops *ops = NULL;
2361         struct kvm_device *dev;
2362         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2363         int ret;
2364
2365         if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2366                 return -ENODEV;
2367
2368         ops = kvm_device_ops_table[cd->type];
2369         if (ops == NULL)
2370                 return -ENODEV;
2371
2372         if (test)
2373                 return 0;
2374
2375         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2376         if (!dev)
2377                 return -ENOMEM;
2378
2379         dev->ops = ops;
2380         dev->kvm = kvm;
2381
2382         ret = ops->create(dev, cd->type);
2383         if (ret < 0) {
2384                 kfree(dev);
2385                 return ret;
2386         }
2387
2388         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2389         if (ret < 0) {
2390                 ops->destroy(dev);
2391                 return ret;
2392         }
2393
2394         list_add(&dev->vm_node, &kvm->devices);
2395         kvm_get_kvm(kvm);
2396         cd->fd = ret;
2397         return 0;
2398 }
2399
2400 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2401 {
2402         switch (arg) {
2403         case KVM_CAP_USER_MEMORY:
2404         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2405         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2406 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2407         case KVM_CAP_SET_BOOT_CPU_ID:
2408 #endif
2409         case KVM_CAP_INTERNAL_ERROR_DATA:
2410 #ifdef CONFIG_HAVE_KVM_MSI
2411         case KVM_CAP_SIGNAL_MSI:
2412 #endif
2413 #ifdef CONFIG_HAVE_KVM_IRQFD
2414         case KVM_CAP_IRQFD_RESAMPLE:
2415 #endif
2416         case KVM_CAP_CHECK_EXTENSION_VM:
2417                 return 1;
2418 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2419         case KVM_CAP_IRQ_ROUTING:
2420                 return KVM_MAX_IRQ_ROUTES;
2421 #endif
2422         default:
2423                 break;
2424         }
2425         return kvm_vm_ioctl_check_extension(kvm, arg);
2426 }
2427
2428 static long kvm_vm_ioctl(struct file *filp,
2429                            unsigned int ioctl, unsigned long arg)
2430 {
2431         struct kvm *kvm = filp->private_data;
2432         void __user *argp = (void __user *)arg;
2433         int r;
2434
2435         if (kvm->mm != current->mm)
2436                 return -EIO;
2437         switch (ioctl) {
2438         case KVM_CREATE_VCPU:
2439                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2440                 break;
2441         case KVM_SET_USER_MEMORY_REGION: {
2442                 struct kvm_userspace_memory_region kvm_userspace_mem;
2443
2444                 r = -EFAULT;
2445                 if (copy_from_user(&kvm_userspace_mem, argp,
2446                                                 sizeof kvm_userspace_mem))
2447                         goto out;
2448
2449                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2450                 break;
2451         }
2452         case KVM_GET_DIRTY_LOG: {
2453                 struct kvm_dirty_log log;
2454
2455                 r = -EFAULT;
2456                 if (copy_from_user(&log, argp, sizeof log))
2457                         goto out;
2458                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2459                 break;
2460         }
2461 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2462         case KVM_REGISTER_COALESCED_MMIO: {
2463                 struct kvm_coalesced_mmio_zone zone;
2464                 r = -EFAULT;
2465                 if (copy_from_user(&zone, argp, sizeof zone))
2466                         goto out;
2467                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2468                 break;
2469         }
2470         case KVM_UNREGISTER_COALESCED_MMIO: {
2471                 struct kvm_coalesced_mmio_zone zone;
2472                 r = -EFAULT;
2473                 if (copy_from_user(&zone, argp, sizeof zone))
2474                         goto out;
2475                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2476                 break;
2477         }
2478 #endif
2479         case KVM_IRQFD: {
2480                 struct kvm_irqfd data;
2481
2482                 r = -EFAULT;
2483                 if (copy_from_user(&data, argp, sizeof data))
2484                         goto out;
2485                 r = kvm_irqfd(kvm, &data);
2486                 break;
2487         }
2488         case KVM_IOEVENTFD: {
2489                 struct kvm_ioeventfd data;
2490
2491                 r = -EFAULT;
2492                 if (copy_from_user(&data, argp, sizeof data))
2493                         goto out;
2494                 r = kvm_ioeventfd(kvm, &data);
2495                 break;
2496         }
2497 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2498         case KVM_SET_BOOT_CPU_ID:
2499                 r = 0;
2500                 mutex_lock(&kvm->lock);
2501                 if (atomic_read(&kvm->online_vcpus) != 0)
2502                         r = -EBUSY;
2503                 else
2504                         kvm->bsp_vcpu_id = arg;
2505                 mutex_unlock(&kvm->lock);
2506                 break;
2507 #endif
2508 #ifdef CONFIG_HAVE_KVM_MSI
2509         case KVM_SIGNAL_MSI: {
2510                 struct kvm_msi msi;
2511
2512                 r = -EFAULT;
2513                 if (copy_from_user(&msi, argp, sizeof msi))
2514                         goto out;
2515                 r = kvm_send_userspace_msi(kvm, &msi);
2516                 break;
2517         }
2518 #endif
2519 #ifdef __KVM_HAVE_IRQ_LINE
2520         case KVM_IRQ_LINE_STATUS:
2521         case KVM_IRQ_LINE: {
2522                 struct kvm_irq_level irq_event;
2523
2524                 r = -EFAULT;
2525                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2526                         goto out;
2527
2528                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2529                                         ioctl == KVM_IRQ_LINE_STATUS);
2530                 if (r)
2531                         goto out;
2532
2533                 r = -EFAULT;
2534                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2535                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2536                                 goto out;
2537                 }
2538
2539                 r = 0;
2540                 break;
2541         }
2542 #endif
2543 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2544         case KVM_SET_GSI_ROUTING: {
2545                 struct kvm_irq_routing routing;
2546                 struct kvm_irq_routing __user *urouting;
2547                 struct kvm_irq_routing_entry *entries;
2548
2549                 r = -EFAULT;
2550                 if (copy_from_user(&routing, argp, sizeof(routing)))
2551                         goto out;
2552                 r = -EINVAL;
2553                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2554                         goto out;
2555                 if (routing.flags)
2556                         goto out;
2557                 r = -ENOMEM;
2558                 entries = vmalloc(routing.nr * sizeof(*entries));
2559                 if (!entries)
2560                         goto out;
2561                 r = -EFAULT;
2562                 urouting = argp;
2563                 if (copy_from_user(entries, urouting->entries,
2564                                    routing.nr * sizeof(*entries)))
2565                         goto out_free_irq_routing;
2566                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2567                                         routing.flags);
2568         out_free_irq_routing:
2569                 vfree(entries);
2570                 break;
2571         }
2572 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2573         case KVM_CREATE_DEVICE: {
2574                 struct kvm_create_device cd;
2575
2576                 r = -EFAULT;
2577                 if (copy_from_user(&cd, argp, sizeof(cd)))
2578                         goto out;
2579
2580                 r = kvm_ioctl_create_device(kvm, &cd);
2581                 if (r)
2582                         goto out;
2583
2584                 r = -EFAULT;
2585                 if (copy_to_user(argp, &cd, sizeof(cd)))
2586                         goto out;
2587
2588                 r = 0;
2589                 break;
2590         }
2591         case KVM_CHECK_EXTENSION:
2592                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2593                 break;
2594         default:
2595                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2596                 if (r == -ENOTTY)
2597                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2598         }
2599 out:
2600         return r;
2601 }
2602
2603 #ifdef CONFIG_COMPAT
2604 struct compat_kvm_dirty_log {
2605         __u32 slot;
2606         __u32 padding1;
2607         union {
2608                 compat_uptr_t dirty_bitmap; /* one bit per page */
2609                 __u64 padding2;
2610         };
2611 };
2612
2613 static long kvm_vm_compat_ioctl(struct file *filp,
2614                            unsigned int ioctl, unsigned long arg)
2615 {
2616         struct kvm *kvm = filp->private_data;
2617         int r;
2618
2619         if (kvm->mm != current->mm)
2620                 return -EIO;
2621         switch (ioctl) {
2622         case KVM_GET_DIRTY_LOG: {
2623                 struct compat_kvm_dirty_log compat_log;
2624                 struct kvm_dirty_log log;
2625
2626                 r = -EFAULT;
2627                 if (copy_from_user(&compat_log, (void __user *)arg,
2628                                    sizeof(compat_log)))
2629                         goto out;
2630                 log.slot         = compat_log.slot;
2631                 log.padding1     = compat_log.padding1;
2632                 log.padding2     = compat_log.padding2;
2633                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2634
2635                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2636                 break;
2637         }
2638         default:
2639                 r = kvm_vm_ioctl(filp, ioctl, arg);
2640         }
2641
2642 out:
2643         return r;
2644 }
2645 #endif
2646
2647 static struct file_operations kvm_vm_fops = {
2648         .release        = kvm_vm_release,
2649         .unlocked_ioctl = kvm_vm_ioctl,
2650 #ifdef CONFIG_COMPAT
2651         .compat_ioctl   = kvm_vm_compat_ioctl,
2652 #endif
2653         .llseek         = noop_llseek,
2654 };
2655
2656 static int kvm_dev_ioctl_create_vm(unsigned long type)
2657 {
2658         int r;
2659         struct kvm *kvm;
2660
2661         kvm = kvm_create_vm(type);
2662         if (IS_ERR(kvm))
2663                 return PTR_ERR(kvm);
2664 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2665         r = kvm_coalesced_mmio_init(kvm);
2666         if (r < 0) {
2667                 kvm_put_kvm(kvm);
2668                 return r;
2669         }
2670 #endif
2671         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2672         if (r < 0)
2673                 kvm_put_kvm(kvm);
2674
2675         return r;
2676 }
2677
2678 static long kvm_dev_ioctl(struct file *filp,
2679                           unsigned int ioctl, unsigned long arg)
2680 {
2681         long r = -EINVAL;
2682
2683         switch (ioctl) {
2684         case KVM_GET_API_VERSION:
2685                 if (arg)
2686                         goto out;
2687                 r = KVM_API_VERSION;
2688                 break;
2689         case KVM_CREATE_VM:
2690                 r = kvm_dev_ioctl_create_vm(arg);
2691                 break;
2692         case KVM_CHECK_EXTENSION:
2693                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2694                 break;
2695         case KVM_GET_VCPU_MMAP_SIZE:
2696                 if (arg)
2697                         goto out;
2698                 r = PAGE_SIZE;     /* struct kvm_run */
2699 #ifdef CONFIG_X86
2700                 r += PAGE_SIZE;    /* pio data page */
2701 #endif
2702 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2703                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2704 #endif
2705                 break;
2706         case KVM_TRACE_ENABLE:
2707         case KVM_TRACE_PAUSE:
2708         case KVM_TRACE_DISABLE:
2709                 r = -EOPNOTSUPP;
2710                 break;
2711         default:
2712                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2713         }
2714 out:
2715         return r;
2716 }
2717
2718 static struct file_operations kvm_chardev_ops = {
2719         .unlocked_ioctl = kvm_dev_ioctl,
2720         .compat_ioctl   = kvm_dev_ioctl,
2721         .llseek         = noop_llseek,
2722 };
2723
2724 static struct miscdevice kvm_dev = {
2725         KVM_MINOR,
2726         "kvm",
2727         &kvm_chardev_ops,
2728 };
2729
2730 static void hardware_enable_nolock(void *junk)
2731 {
2732         int cpu = raw_smp_processor_id();
2733         int r;
2734
2735         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2736                 return;
2737
2738         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2739
2740         r = kvm_arch_hardware_enable();
2741
2742         if (r) {
2743                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2744                 atomic_inc(&hardware_enable_failed);
2745                 printk(KERN_INFO "kvm: enabling virtualization on "
2746                                  "CPU%d failed\n", cpu);
2747         }
2748 }
2749
2750 static void hardware_enable(void)
2751 {
2752         raw_spin_lock(&kvm_count_lock);
2753         if (kvm_usage_count)
2754                 hardware_enable_nolock(NULL);
2755         raw_spin_unlock(&kvm_count_lock);
2756 }
2757
2758 static void hardware_disable_nolock(void *junk)
2759 {
2760         int cpu = raw_smp_processor_id();
2761
2762         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2763                 return;
2764         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2765         kvm_arch_hardware_disable();
2766 }
2767
2768 static void hardware_disable(void)
2769 {
2770         raw_spin_lock(&kvm_count_lock);
2771         if (kvm_usage_count)
2772                 hardware_disable_nolock(NULL);
2773         raw_spin_unlock(&kvm_count_lock);
2774 }
2775
2776 static void hardware_disable_all_nolock(void)
2777 {
2778         BUG_ON(!kvm_usage_count);
2779
2780         kvm_usage_count--;
2781         if (!kvm_usage_count)
2782                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2783 }
2784
2785 static void hardware_disable_all(void)
2786 {
2787         raw_spin_lock(&kvm_count_lock);
2788         hardware_disable_all_nolock();
2789         raw_spin_unlock(&kvm_count_lock);
2790 }
2791
2792 static int hardware_enable_all(void)
2793 {
2794         int r = 0;
2795
2796         raw_spin_lock(&kvm_count_lock);
2797
2798         kvm_usage_count++;
2799         if (kvm_usage_count == 1) {
2800                 atomic_set(&hardware_enable_failed, 0);
2801                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2802
2803                 if (atomic_read(&hardware_enable_failed)) {
2804                         hardware_disable_all_nolock();
2805                         r = -EBUSY;
2806                 }
2807         }
2808
2809         raw_spin_unlock(&kvm_count_lock);
2810
2811         return r;
2812 }
2813
2814 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2815                            void *v)
2816 {
2817         int cpu = (long)v;
2818
2819         val &= ~CPU_TASKS_FROZEN;
2820         switch (val) {
2821         case CPU_DYING:
2822                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2823                        cpu);
2824                 hardware_disable();
2825                 break;
2826         case CPU_STARTING:
2827                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2828                        cpu);
2829                 hardware_enable();
2830                 break;
2831         }
2832         return NOTIFY_OK;
2833 }
2834
2835 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2836                       void *v)
2837 {
2838         /*
2839          * Some (well, at least mine) BIOSes hang on reboot if
2840          * in vmx root mode.
2841          *
2842          * And Intel TXT required VMX off for all cpu when system shutdown.
2843          */
2844         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2845         kvm_rebooting = true;
2846         on_each_cpu(hardware_disable_nolock, NULL, 1);
2847         return NOTIFY_OK;
2848 }
2849
2850 static struct notifier_block kvm_reboot_notifier = {
2851         .notifier_call = kvm_reboot,
2852         .priority = 0,
2853 };
2854
2855 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2856 {
2857         int i;
2858
2859         for (i = 0; i < bus->dev_count; i++) {
2860                 struct kvm_io_device *pos = bus->range[i].dev;
2861
2862                 kvm_iodevice_destructor(pos);
2863         }
2864         kfree(bus);
2865 }
2866
2867 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2868                                  const struct kvm_io_range *r2)
2869 {
2870         if (r1->addr < r2->addr)
2871                 return -1;
2872         if (r1->addr + r1->len > r2->addr + r2->len)
2873                 return 1;
2874         return 0;
2875 }
2876
2877 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2878 {
2879         return kvm_io_bus_cmp(p1, p2);
2880 }
2881
2882 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2883                           gpa_t addr, int len)
2884 {
2885         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2886                 .addr = addr,
2887                 .len = len,
2888                 .dev = dev,
2889         };
2890
2891         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2892                 kvm_io_bus_sort_cmp, NULL);
2893
2894         return 0;
2895 }
2896
2897 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2898                              gpa_t addr, int len)
2899 {
2900         struct kvm_io_range *range, key;
2901         int off;
2902
2903         key = (struct kvm_io_range) {
2904                 .addr = addr,
2905                 .len = len,
2906         };
2907
2908         range = bsearch(&key, bus->range, bus->dev_count,
2909                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2910         if (range == NULL)
2911                 return -ENOENT;
2912
2913         off = range - bus->range;
2914
2915         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2916                 off--;
2917
2918         return off;
2919 }
2920
2921 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2922                               struct kvm_io_range *range, const void *val)
2923 {
2924         int idx;
2925
2926         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2927         if (idx < 0)
2928                 return -EOPNOTSUPP;
2929
2930         while (idx < bus->dev_count &&
2931                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2932                 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2933                                         range->len, val))
2934                         return idx;
2935                 idx++;
2936         }
2937
2938         return -EOPNOTSUPP;
2939 }
2940
2941 /* kvm_io_bus_write - called under kvm->slots_lock */
2942 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2943                      int len, const void *val)
2944 {
2945         struct kvm_io_bus *bus;
2946         struct kvm_io_range range;
2947         int r;
2948
2949         range = (struct kvm_io_range) {
2950                 .addr = addr,
2951                 .len = len,
2952         };
2953
2954         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2955         r = __kvm_io_bus_write(bus, &range, val);
2956         return r < 0 ? r : 0;
2957 }
2958
2959 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2960 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2961                             int len, const void *val, long cookie)
2962 {
2963         struct kvm_io_bus *bus;
2964         struct kvm_io_range range;
2965
2966         range = (struct kvm_io_range) {
2967                 .addr = addr,
2968                 .len = len,
2969         };
2970
2971         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2972
2973         /* First try the device referenced by cookie. */
2974         if ((cookie >= 0) && (cookie < bus->dev_count) &&
2975             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2976                 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2977                                         val))
2978                         return cookie;
2979
2980         /*
2981          * cookie contained garbage; fall back to search and return the
2982          * correct cookie value.
2983          */
2984         return __kvm_io_bus_write(bus, &range, val);
2985 }
2986
2987 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2988                              void *val)
2989 {
2990         int idx;
2991
2992         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2993         if (idx < 0)
2994                 return -EOPNOTSUPP;
2995
2996         while (idx < bus->dev_count &&
2997                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2998                 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2999                                        range->len, val))
3000                         return idx;
3001                 idx++;
3002         }
3003
3004         return -EOPNOTSUPP;
3005 }
3006 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3007
3008 /* kvm_io_bus_read - called under kvm->slots_lock */
3009 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3010                     int len, void *val)
3011 {
3012         struct kvm_io_bus *bus;
3013         struct kvm_io_range range;
3014         int r;
3015
3016         range = (struct kvm_io_range) {
3017                 .addr = addr,
3018                 .len = len,
3019         };
3020
3021         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3022         r = __kvm_io_bus_read(bus, &range, val);
3023         return r < 0 ? r : 0;
3024 }
3025
3026
3027 /* Caller must hold slots_lock. */
3028 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3029                             int len, struct kvm_io_device *dev)
3030 {
3031         struct kvm_io_bus *new_bus, *bus;
3032
3033         bus = kvm->buses[bus_idx];
3034         /* exclude ioeventfd which is limited by maximum fd */
3035         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3036                 return -ENOSPC;
3037
3038         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3039                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3040         if (!new_bus)
3041                 return -ENOMEM;
3042         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3043                sizeof(struct kvm_io_range)));
3044         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3045         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3046         synchronize_srcu_expedited(&kvm->srcu);
3047         kfree(bus);
3048
3049         return 0;
3050 }
3051
3052 /* Caller must hold slots_lock. */
3053 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3054                               struct kvm_io_device *dev)
3055 {
3056         int i, r;
3057         struct kvm_io_bus *new_bus, *bus;
3058
3059         bus = kvm->buses[bus_idx];
3060         r = -ENOENT;
3061         for (i = 0; i < bus->dev_count; i++)
3062                 if (bus->range[i].dev == dev) {
3063                         r = 0;
3064                         break;
3065                 }
3066
3067         if (r)
3068                 return r;
3069
3070         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3071                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3072         if (!new_bus)
3073                 return -ENOMEM;
3074
3075         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3076         new_bus->dev_count--;
3077         memcpy(new_bus->range + i, bus->range + i + 1,
3078                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3079
3080         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3081         synchronize_srcu_expedited(&kvm->srcu);
3082         kfree(bus);
3083         return r;
3084 }
3085
3086 static struct notifier_block kvm_cpu_notifier = {
3087         .notifier_call = kvm_cpu_hotplug,
3088 };
3089
3090 static int vm_stat_get(void *_offset, u64 *val)
3091 {
3092         unsigned offset = (long)_offset;
3093         struct kvm *kvm;
3094
3095         *val = 0;
3096         spin_lock(&kvm_lock);
3097         list_for_each_entry(kvm, &vm_list, vm_list)
3098                 *val += *(u32 *)((void *)kvm + offset);
3099         spin_unlock(&kvm_lock);
3100         return 0;
3101 }
3102
3103 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3104
3105 static int vcpu_stat_get(void *_offset, u64 *val)
3106 {
3107         unsigned offset = (long)_offset;
3108         struct kvm *kvm;
3109         struct kvm_vcpu *vcpu;
3110         int i;
3111
3112         *val = 0;
3113         spin_lock(&kvm_lock);
3114         list_for_each_entry(kvm, &vm_list, vm_list)
3115                 kvm_for_each_vcpu(i, vcpu, kvm)
3116                         *val += *(u32 *)((void *)vcpu + offset);
3117
3118         spin_unlock(&kvm_lock);
3119         return 0;
3120 }
3121
3122 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3123
3124 static const struct file_operations *stat_fops[] = {
3125         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3126         [KVM_STAT_VM]   = &vm_stat_fops,
3127 };
3128
3129 static int kvm_init_debug(void)
3130 {
3131         int r = -EEXIST;
3132         struct kvm_stats_debugfs_item *p;
3133
3134         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3135         if (kvm_debugfs_dir == NULL)
3136                 goto out;
3137
3138         for (p = debugfs_entries; p->name; ++p) {
3139                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3140                                                 (void *)(long)p->offset,
3141                                                 stat_fops[p->kind]);
3142                 if (p->dentry == NULL)
3143                         goto out_dir;
3144         }
3145
3146         return 0;
3147
3148 out_dir:
3149         debugfs_remove_recursive(kvm_debugfs_dir);
3150 out:
3151         return r;
3152 }
3153
3154 static void kvm_exit_debug(void)
3155 {
3156         struct kvm_stats_debugfs_item *p;
3157
3158         for (p = debugfs_entries; p->name; ++p)
3159                 debugfs_remove(p->dentry);
3160         debugfs_remove(kvm_debugfs_dir);
3161 }
3162
3163 static int kvm_suspend(void)
3164 {
3165         if (kvm_usage_count)
3166                 hardware_disable_nolock(NULL);
3167         return 0;
3168 }
3169
3170 static void kvm_resume(void)
3171 {
3172         if (kvm_usage_count) {
3173                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3174                 hardware_enable_nolock(NULL);
3175         }
3176 }
3177
3178 static struct syscore_ops kvm_syscore_ops = {
3179         .suspend = kvm_suspend,
3180         .resume = kvm_resume,
3181 };
3182
3183 static inline
3184 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3185 {
3186         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3187 }
3188
3189 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3190 {
3191         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3192         if (vcpu->preempted)
3193                 vcpu->preempted = false;
3194
3195         kvm_arch_sched_in(vcpu, cpu);
3196
3197         kvm_arch_vcpu_load(vcpu, cpu);
3198 }
3199
3200 static void kvm_sched_out(struct preempt_notifier *pn,
3201                           struct task_struct *next)
3202 {
3203         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3204
3205         if (current->state == TASK_RUNNING)
3206                 vcpu->preempted = true;
3207         kvm_arch_vcpu_put(vcpu);
3208 }
3209
3210 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3211                   struct module *module)
3212 {
3213         int r;
3214         int cpu;
3215
3216         r = kvm_arch_init(opaque);
3217         if (r)
3218                 goto out_fail;
3219
3220         /*
3221          * kvm_arch_init makes sure there's at most one caller
3222          * for architectures that support multiple implementations,
3223          * like intel and amd on x86.
3224          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3225          * conflicts in case kvm is already setup for another implementation.
3226          */
3227         r = kvm_irqfd_init();
3228         if (r)
3229                 goto out_irqfd;
3230
3231         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3232                 r = -ENOMEM;
3233                 goto out_free_0;
3234         }
3235
3236         r = kvm_arch_hardware_setup();
3237         if (r < 0)
3238                 goto out_free_0a;
3239
3240         for_each_online_cpu(cpu) {
3241                 smp_call_function_single(cpu,
3242                                 kvm_arch_check_processor_compat,
3243                                 &r, 1);
3244                 if (r < 0)
3245                         goto out_free_1;
3246         }
3247
3248         r = register_cpu_notifier(&kvm_cpu_notifier);
3249         if (r)
3250                 goto out_free_2;
3251         register_reboot_notifier(&kvm_reboot_notifier);
3252
3253         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3254         if (!vcpu_align)
3255                 vcpu_align = __alignof__(struct kvm_vcpu);
3256         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3257                                            0, NULL);
3258         if (!kvm_vcpu_cache) {
3259                 r = -ENOMEM;
3260                 goto out_free_3;
3261         }
3262
3263         r = kvm_async_pf_init();
3264         if (r)
3265                 goto out_free;
3266
3267         kvm_chardev_ops.owner = module;
3268         kvm_vm_fops.owner = module;
3269         kvm_vcpu_fops.owner = module;
3270
3271         r = misc_register(&kvm_dev);
3272         if (r) {
3273                 printk(KERN_ERR "kvm: misc device register failed\n");
3274                 goto out_unreg;
3275         }
3276
3277         register_syscore_ops(&kvm_syscore_ops);
3278
3279         kvm_preempt_ops.sched_in = kvm_sched_in;
3280         kvm_preempt_ops.sched_out = kvm_sched_out;
3281
3282         r = kvm_init_debug();
3283         if (r) {
3284                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3285                 goto out_undebugfs;
3286         }
3287
3288         r = kvm_vfio_ops_init();
3289         WARN_ON(r);
3290
3291         return 0;
3292
3293 out_undebugfs:
3294         unregister_syscore_ops(&kvm_syscore_ops);
3295         misc_deregister(&kvm_dev);
3296 out_unreg:
3297         kvm_async_pf_deinit();
3298 out_free:
3299         kmem_cache_destroy(kvm_vcpu_cache);
3300 out_free_3:
3301         unregister_reboot_notifier(&kvm_reboot_notifier);
3302         unregister_cpu_notifier(&kvm_cpu_notifier);
3303 out_free_2:
3304 out_free_1:
3305         kvm_arch_hardware_unsetup();
3306 out_free_0a:
3307         free_cpumask_var(cpus_hardware_enabled);
3308 out_free_0:
3309         kvm_irqfd_exit();
3310 out_irqfd:
3311         kvm_arch_exit();
3312 out_fail:
3313         return r;
3314 }
3315 EXPORT_SYMBOL_GPL(kvm_init);
3316
3317 void kvm_exit(void)
3318 {
3319         kvm_exit_debug();
3320         misc_deregister(&kvm_dev);
3321         kmem_cache_destroy(kvm_vcpu_cache);
3322         kvm_async_pf_deinit();
3323         unregister_syscore_ops(&kvm_syscore_ops);
3324         unregister_reboot_notifier(&kvm_reboot_notifier);
3325         unregister_cpu_notifier(&kvm_cpu_notifier);
3326         on_each_cpu(hardware_disable_nolock, NULL, 1);
3327         kvm_arch_hardware_unsetup();
3328         kvm_arch_exit();
3329         kvm_irqfd_exit();
3330         free_cpumask_var(cpus_hardware_enabled);
3331 }
3332 EXPORT_SYMBOL_GPL(kvm_exit);