virt/kvm/kvm_main.c

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