virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/syscore_ops.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/ioctl.h>
  56 #include <asm/uaccess.h>
  57 #include <asm/pgtable.h>
  58
  59 #include "coalesced_mmio.h"
  60 #include "async_pf.h"
  61 #include "vfio.h"
  62
  63 #define CREATE_TRACE_POINTS
  64 #include <trace/events/kvm.h>
  65
  66 MODULE_AUTHOR("Qumranet");
  67 MODULE_LICENSE("GPL");
  68
  69 /*
  70  * Ordering of locks:
  71  *
  72  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  73  */
  74
  75 DEFINE_SPINLOCK(kvm_lock);
  76 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
  77 LIST_HEAD(vm_list);
  78
  79 static cpumask_var_t cpus_hardware_enabled;
  80 static int kvm_usage_count = 0;
  81 static atomic_t hardware_enable_failed;
  82
  83 struct kmem_cache *kvm_vcpu_cache;
  84 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  85
  86 static __read_mostly struct preempt_ops kvm_preempt_ops;
  87
  88 struct dentry *kvm_debugfs_dir;
  89
  90 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  91                            unsigned long arg);
  92 #ifdef CONFIG_COMPAT
  93 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  94                                   unsigned long arg);
  95 #endif
  96 static int hardware_enable_all(void);
  97 static void hardware_disable_all(void);
  98
  99 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
 100
 101 static void kvm_release_pfn_dirty(pfn_t pfn);
 102 static void mark_page_dirty_in_slot(struct kvm *kvm,
 103                                     struct kvm_memory_slot *memslot, gfn_t gfn);
 104
 105 __visible bool kvm_rebooting;
 106 EXPORT_SYMBOL_GPL(kvm_rebooting);
 107
 108 static bool largepages_enabled = true;
 109
 110 bool kvm_is_reserved_pfn(pfn_t pfn)
 111 {
 112         if (pfn_valid(pfn))
 113                 return PageReserved(pfn_to_page(pfn));
 114
 115         return true;
 116 }
 117
 118 /*
 119  * Switches to specified vcpu, until a matching vcpu_put()
 120  */
 121 int vcpu_load(struct kvm_vcpu *vcpu)
 122 {
 123         int cpu;
 124
 125         if (mutex_lock_killable(&vcpu->mutex))
 126                 return -EINTR;
 127         cpu = get_cpu();
 128         preempt_notifier_register(&vcpu->preempt_notifier);
 129         kvm_arch_vcpu_load(vcpu, cpu);
 130         put_cpu();
 131         return 0;
 132 }
 133
 134 void vcpu_put(struct kvm_vcpu *vcpu)
 135 {
 136         preempt_disable();
 137         kvm_arch_vcpu_put(vcpu);
 138         preempt_notifier_unregister(&vcpu->preempt_notifier);
 139         preempt_enable();
 140         mutex_unlock(&vcpu->mutex);
 141 }
 142
 143 static void ack_flush(void *_completed)
 144 {
 145 }
 146
 147 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
 148 {
 149         int i, cpu, me;
 150         cpumask_var_t cpus;
 151         bool called = true;
 152         struct kvm_vcpu *vcpu;
 153
 154         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 155
 156         me = get_cpu();
 157         kvm_for_each_vcpu(i, vcpu, kvm) {
 158                 kvm_make_request(req, vcpu);
 159                 cpu = vcpu->cpu;
 160
 161                 /* Set ->requests bit before we read ->mode */
 162                 smp_mb();
 163
 164                 if (cpus != NULL && cpu != -1 && cpu != me &&
 165                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 166                         cpumask_set_cpu(cpu, cpus);
 167         }
 168         if (unlikely(cpus == NULL))
 169                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 170         else if (!cpumask_empty(cpus))
 171                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 172         else
 173                 called = false;
 174         put_cpu();
 175         free_cpumask_var(cpus);
 176         return called;
 177 }
 178
 179 void kvm_flush_remote_tlbs(struct kvm *kvm)
 180 {
 181         long dirty_count = kvm->tlbs_dirty;
 182
 183         smp_mb();
 184         if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 185                 ++kvm->stat.remote_tlb_flush;
 186         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 187 }
 188 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 189
 190 void kvm_reload_remote_mmus(struct kvm *kvm)
 191 {
 192         kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 193 }
 194
 195 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 196 {
 197         kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 198 }
 199
 200 void kvm_make_scan_ioapic_request(struct kvm *kvm)
 201 {
 202         kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
 203 }
 204
 205 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 206 {
 207         struct page *page;
 208         int r;
 209
 210         mutex_init(&vcpu->mutex);
 211         vcpu->cpu = -1;
 212         vcpu->kvm = kvm;
 213         vcpu->vcpu_id = id;
 214         vcpu->pid = NULL;
 215         init_waitqueue_head(&vcpu->wq);
 216         kvm_async_pf_vcpu_init(vcpu);
 217
 218         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 219         if (!page) {
 220                 r = -ENOMEM;
 221                 goto fail;
 222         }
 223         vcpu->run = page_address(page);
 224
 225         kvm_vcpu_set_in_spin_loop(vcpu, false);
 226         kvm_vcpu_set_dy_eligible(vcpu, false);
 227         vcpu->preempted = false;
 228
 229         r = kvm_arch_vcpu_init(vcpu);
 230         if (r < 0)
 231                 goto fail_free_run;
 232         return 0;
 233
 234 fail_free_run:
 235         free_page((unsigned long)vcpu->run);
 236 fail:
 237         return r;
 238 }
 239 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 240
 241 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 242 {
 243         put_pid(vcpu->pid);
 244         kvm_arch_vcpu_uninit(vcpu);
 245         free_page((unsigned long)vcpu->run);
 246 }
 247 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 248
 249 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 250 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 251 {
 252         return container_of(mn, struct kvm, mmu_notifier);
 253 }
 254
 255 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 256                                              struct mm_struct *mm,
 257                                              unsigned long address)
 258 {
 259         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 260         int need_tlb_flush, idx;
 261
 262         /*
 263          * When ->invalidate_page runs, the linux pte has been zapped
 264          * already but the page is still allocated until
 265          * ->invalidate_page returns. So if we increase the sequence
 266          * here the kvm page fault will notice if the spte can't be
 267          * established because the page is going to be freed. If
 268          * instead the kvm page fault establishes the spte before
 269          * ->invalidate_page runs, kvm_unmap_hva will release it
 270          * before returning.
 271          *
 272          * The sequence increase only need to be seen at spin_unlock
 273          * time, and not at spin_lock time.
 274          *
 275          * Increasing the sequence after the spin_unlock would be
 276          * unsafe because the kvm page fault could then establish the
 277          * pte after kvm_unmap_hva returned, without noticing the page
 278          * is going to be freed.
 279          */
 280         idx = srcu_read_lock(&kvm->srcu);
 281         spin_lock(&kvm->mmu_lock);
 282
 283         kvm->mmu_notifier_seq++;
 284         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 285         /* we've to flush the tlb before the pages can be freed */
 286         if (need_tlb_flush)
 287                 kvm_flush_remote_tlbs(kvm);
 288
 289         spin_unlock(&kvm->mmu_lock);
 290
 291         kvm_arch_mmu_notifier_invalidate_page(kvm, address);
 292
 293         srcu_read_unlock(&kvm->srcu, idx);
 294 }
 295
 296 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 297                                         struct mm_struct *mm,
 298                                         unsigned long address,
 299                                         pte_t pte)
 300 {
 301         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 302         int idx;
 303
 304         idx = srcu_read_lock(&kvm->srcu);
 305         spin_lock(&kvm->mmu_lock);
 306         kvm->mmu_notifier_seq++;
 307         kvm_set_spte_hva(kvm, address, pte);
 308         spin_unlock(&kvm->mmu_lock);
 309         srcu_read_unlock(&kvm->srcu, idx);
 310 }
 311
 312 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 313                                                     struct mm_struct *mm,
 314                                                     unsigned long start,
 315                                                     unsigned long end)
 316 {
 317         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 318         int need_tlb_flush = 0, idx;
 319
 320         idx = srcu_read_lock(&kvm->srcu);
 321         spin_lock(&kvm->mmu_lock);
 322         /*
 323          * The count increase must become visible at unlock time as no
 324          * spte can be established without taking the mmu_lock and
 325          * count is also read inside the mmu_lock critical section.
 326          */
 327         kvm->mmu_notifier_count++;
 328         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 329         need_tlb_flush |= kvm->tlbs_dirty;
 330         /* we've to flush the tlb before the pages can be freed */
 331         if (need_tlb_flush)
 332                 kvm_flush_remote_tlbs(kvm);
 333
 334         spin_unlock(&kvm->mmu_lock);
 335         srcu_read_unlock(&kvm->srcu, idx);
 336 }
 337
 338 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 339                                                   struct mm_struct *mm,
 340                                                   unsigned long start,
 341                                                   unsigned long end)
 342 {
 343         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 344
 345         spin_lock(&kvm->mmu_lock);
 346         /*
 347          * This sequence increase will notify the kvm page fault that
 348          * the page that is going to be mapped in the spte could have
 349          * been freed.
 350          */
 351         kvm->mmu_notifier_seq++;
 352         smp_wmb();
 353         /*
 354          * The above sequence increase must be visible before the
 355          * below count decrease, which is ensured by the smp_wmb above
 356          * in conjunction with the smp_rmb in mmu_notifier_retry().
 357          */
 358         kvm->mmu_notifier_count--;
 359         spin_unlock(&kvm->mmu_lock);
 360
 361         BUG_ON(kvm->mmu_notifier_count < 0);
 362 }
 363
 364 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 365                                               struct mm_struct *mm,
 366                                               unsigned long start,
 367                                               unsigned long end)
 368 {
 369         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 370         int young, idx;
 371
 372         idx = srcu_read_lock(&kvm->srcu);
 373         spin_lock(&kvm->mmu_lock);
 374
 375         young = kvm_age_hva(kvm, start, end);
 376         if (young)
 377                 kvm_flush_remote_tlbs(kvm);
 378
 379         spin_unlock(&kvm->mmu_lock);
 380         srcu_read_unlock(&kvm->srcu, idx);
 381
 382         return young;
 383 }
 384
 385 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 386                                        struct mm_struct *mm,
 387                                        unsigned long address)
 388 {
 389         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 390         int young, idx;
 391
 392         idx = srcu_read_lock(&kvm->srcu);
 393         spin_lock(&kvm->mmu_lock);
 394         young = kvm_test_age_hva(kvm, address);
 395         spin_unlock(&kvm->mmu_lock);
 396         srcu_read_unlock(&kvm->srcu, idx);
 397
 398         return young;
 399 }
 400
 401 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 402                                      struct mm_struct *mm)
 403 {
 404         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 405         int idx;
 406
 407         idx = srcu_read_lock(&kvm->srcu);
 408         kvm_arch_flush_shadow_all(kvm);
 409         srcu_read_unlock(&kvm->srcu, idx);
 410 }
 411
 412 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 413         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 414         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 415         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 416         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 417         .test_young             = kvm_mmu_notifier_test_young,
 418         .change_pte             = kvm_mmu_notifier_change_pte,
 419         .release                = kvm_mmu_notifier_release,
 420 };
 421
 422 static int kvm_init_mmu_notifier(struct kvm *kvm)
 423 {
 424         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 425         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 426 }
 427
 428 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 429
 430 static int kvm_init_mmu_notifier(struct kvm *kvm)
 431 {
 432         return 0;
 433 }
 434
 435 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 436
 437 static void kvm_init_memslots_id(struct kvm *kvm)
 438 {
 439         int i;
 440         struct kvm_memslots *slots = kvm->memslots;
 441
 442         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 443                 slots->id_to_index[i] = slots->memslots[i].id = i;
 444 }
 445
 446 static struct kvm *kvm_create_vm(unsigned long type)
 447 {
 448         int r, i;
 449         struct kvm *kvm = kvm_arch_alloc_vm();
 450
 451         if (!kvm)
 452                 return ERR_PTR(-ENOMEM);
 453
 454         r = kvm_arch_init_vm(kvm, type);
 455         if (r)
 456                 goto out_err_no_disable;
 457
 458         r = hardware_enable_all();
 459         if (r)
 460                 goto out_err_no_disable;
 461
 462 #ifdef CONFIG_HAVE_KVM_IRQFD
 463         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 464 #endif
 465
 466         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 467
 468         r = -ENOMEM;
 469         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 470         if (!kvm->memslots)
 471                 goto out_err_no_srcu;
 472
 473         /*
 474          * Init kvm generation close to the maximum to easily test the
 475          * code of handling generation number wrap-around.
 476          */
 477         kvm->memslots->generation = -150;
 478
 479         kvm_init_memslots_id(kvm);
 480         if (init_srcu_struct(&kvm->srcu))
 481                 goto out_err_no_srcu;
 482         if (init_srcu_struct(&kvm->irq_srcu))
 483                 goto out_err_no_irq_srcu;
 484         for (i = 0; i < KVM_NR_BUSES; i++) {
 485                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 486                                         GFP_KERNEL);
 487                 if (!kvm->buses[i])
 488                         goto out_err;
 489         }
 490
 491         spin_lock_init(&kvm->mmu_lock);
 492         kvm->mm = current->mm;
 493         atomic_inc(&kvm->mm->mm_count);
 494         kvm_eventfd_init(kvm);
 495         mutex_init(&kvm->lock);
 496         mutex_init(&kvm->irq_lock);
 497         mutex_init(&kvm->slots_lock);
 498         atomic_set(&kvm->users_count, 1);
 499         INIT_LIST_HEAD(&kvm->devices);
 500
 501         r = kvm_init_mmu_notifier(kvm);
 502         if (r)
 503                 goto out_err;
 504
 505         spin_lock(&kvm_lock);
 506         list_add(&kvm->vm_list, &vm_list);
 507         spin_unlock(&kvm_lock);
 508
 509         return kvm;
 510
 511 out_err:
 512         cleanup_srcu_struct(&kvm->irq_srcu);
 513 out_err_no_irq_srcu:
 514         cleanup_srcu_struct(&kvm->srcu);
 515 out_err_no_srcu:
 516         hardware_disable_all();
 517 out_err_no_disable:
 518         for (i = 0; i < KVM_NR_BUSES; i++)
 519                 kfree(kvm->buses[i]);
 520         kfree(kvm->memslots);
 521         kvm_arch_free_vm(kvm);
 522         return ERR_PTR(r);
 523 }
 524
 525 /*
 526  * Avoid using vmalloc for a small buffer.
 527  * Should not be used when the size is statically known.
 528  */
 529 void *kvm_kvzalloc(unsigned long size)
 530 {
 531         if (size > PAGE_SIZE)
 532                 return vzalloc(size);
 533         else
 534                 return kzalloc(size, GFP_KERNEL);
 535 }
 536
 537 void kvm_kvfree(const void *addr)
 538 {
 539         if (is_vmalloc_addr(addr))
 540                 vfree(addr);
 541         else
 542                 kfree(addr);
 543 }
 544
 545 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 546 {
 547         if (!memslot->dirty_bitmap)
 548                 return;
 549
 550         kvm_kvfree(memslot->dirty_bitmap);
 551         memslot->dirty_bitmap = NULL;
 552 }
 553
 554 /*
 555  * Free any memory in @free but not in @dont.
 556  */
 557 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
 558                                   struct kvm_memory_slot *dont)
 559 {
 560         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 561                 kvm_destroy_dirty_bitmap(free);
 562
 563         kvm_arch_free_memslot(kvm, free, dont);
 564
 565         free->npages = 0;
 566 }
 567
 568 static void kvm_free_physmem(struct kvm *kvm)
 569 {
 570         struct kvm_memslots *slots = kvm->memslots;
 571         struct kvm_memory_slot *memslot;
 572
 573         kvm_for_each_memslot(memslot, slots)
 574                 kvm_free_physmem_slot(kvm, memslot, NULL);
 575
 576         kfree(kvm->memslots);
 577 }
 578
 579 static void kvm_destroy_devices(struct kvm *kvm)
 580 {
 581         struct list_head *node, *tmp;
 582
 583         list_for_each_safe(node, tmp, &kvm->devices) {
 584                 struct kvm_device *dev =
 585                         list_entry(node, struct kvm_device, vm_node);
 586
 587                 list_del(node);
 588                 dev->ops->destroy(dev);
 589         }
 590 }
 591
 592 static void kvm_destroy_vm(struct kvm *kvm)
 593 {
 594         int i;
 595         struct mm_struct *mm = kvm->mm;
 596
 597         kvm_arch_sync_events(kvm);
 598         spin_lock(&kvm_lock);
 599         list_del(&kvm->vm_list);
 600         spin_unlock(&kvm_lock);
 601         kvm_free_irq_routing(kvm);
 602         for (i = 0; i < KVM_NR_BUSES; i++)
 603                 kvm_io_bus_destroy(kvm->buses[i]);
 604         kvm_coalesced_mmio_free(kvm);
 605 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 606         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 607 #else
 608         kvm_arch_flush_shadow_all(kvm);
 609 #endif
 610         kvm_arch_destroy_vm(kvm);
 611         kvm_destroy_devices(kvm);
 612         kvm_free_physmem(kvm);
 613         cleanup_srcu_struct(&kvm->irq_srcu);
 614         cleanup_srcu_struct(&kvm->srcu);
 615         kvm_arch_free_vm(kvm);
 616         hardware_disable_all();
 617         mmdrop(mm);
 618 }
 619
 620 void kvm_get_kvm(struct kvm *kvm)
 621 {
 622         atomic_inc(&kvm->users_count);
 623 }
 624 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 625
 626 void kvm_put_kvm(struct kvm *kvm)
 627 {
 628         if (atomic_dec_and_test(&kvm->users_count))
 629                 kvm_destroy_vm(kvm);
 630 }
 631 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 632
 633
 634 static int kvm_vm_release(struct inode *inode, struct file *filp)
 635 {
 636         struct kvm *kvm = filp->private_data;
 637
 638         kvm_irqfd_release(kvm);
 639
 640         kvm_put_kvm(kvm);
 641         return 0;
 642 }
 643
 644 /*
 645  * Allocation size is twice as large as the actual dirty bitmap size.
 646  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 647  */
 648 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 649 {
 650         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 651
 652         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 653         if (!memslot->dirty_bitmap)
 654                 return -ENOMEM;
 655
 656         return 0;
 657 }
 658
 659 /*
 660  * Insert memslot and re-sort memslots based on their GFN,
 661  * so binary search could be used to lookup GFN.
 662  * Sorting algorithm takes advantage of having initially
 663  * sorted array and known changed memslot position.
 664  */
 665 static void update_memslots(struct kvm_memslots *slots,
 666                             struct kvm_memory_slot *new)
 667 {
 668         int id = new->id;
 669         int i = slots->id_to_index[id];
 670         struct kvm_memory_slot *mslots = slots->memslots;
 671
 672         WARN_ON(mslots[i].id != id);
 673         if (!new->npages) {
 674                 WARN_ON(!mslots[i].npages);
 675                 new->base_gfn = 0;
 676                 if (mslots[i].npages)
 677                         slots->used_slots--;
 678         } else {
 679                 if (!mslots[i].npages)
 680                         slots->used_slots++;
 681         }
 682
 683         while (i < KVM_MEM_SLOTS_NUM - 1 &&
 684                new->base_gfn <= mslots[i + 1].base_gfn) {
 685                 if (!mslots[i + 1].npages)
 686                         break;
 687                 mslots[i] = mslots[i + 1];
 688                 slots->id_to_index[mslots[i].id] = i;
 689                 i++;
 690         }
 691
 692         /*
 693          * The ">=" is needed when creating a slot with base_gfn == 0,
 694          * so that it moves before all those with base_gfn == npages == 0.
 695          *
 696          * On the other hand, if new->npages is zero, the above loop has
 697          * already left i pointing to the beginning of the empty part of
 698          * mslots, and the ">=" would move the hole backwards in this
 699          * case---which is wrong.  So skip the loop when deleting a slot.
 700          */
 701         if (new->npages) {
 702                 while (i > 0 &&
 703                        new->base_gfn >= mslots[i - 1].base_gfn) {
 704                         mslots[i] = mslots[i - 1];
 705                         slots->id_to_index[mslots[i].id] = i;
 706                         i--;
 707                 }
 708         } else
 709                 WARN_ON_ONCE(i != slots->used_slots);
 710
 711         mslots[i] = *new;
 712         slots->id_to_index[mslots[i].id] = i;
 713 }
 714
 715 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 716 {
 717         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 718
 719 #ifdef __KVM_HAVE_READONLY_MEM
 720         valid_flags |= KVM_MEM_READONLY;
 721 #endif
 722
 723         if (mem->flags & ~valid_flags)
 724                 return -EINVAL;
 725
 726         return 0;
 727 }
 728
 729 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 730                 struct kvm_memslots *slots)
 731 {
 732         struct kvm_memslots *old_memslots = kvm->memslots;
 733
 734         /*
 735          * Set the low bit in the generation, which disables SPTE caching
 736          * until the end of synchronize_srcu_expedited.
 737          */
 738         WARN_ON(old_memslots->generation & 1);
 739         slots->generation = old_memslots->generation + 1;
 740
 741         rcu_assign_pointer(kvm->memslots, slots);
 742         synchronize_srcu_expedited(&kvm->srcu);
 743
 744         /*
 745          * Increment the new memslot generation a second time. This prevents
 746          * vm exits that race with memslot updates from caching a memslot
 747          * generation that will (potentially) be valid forever.
 748          */
 749         slots->generation++;
 750
 751         kvm_arch_memslots_updated(kvm);
 752
 753         return old_memslots;
 754 }
 755
 756 /*
 757  * Allocate some memory and give it an address in the guest physical address
 758  * space.
 759  *
 760  * Discontiguous memory is allowed, mostly for framebuffers.
 761  *
 762  * Must be called holding kvm->slots_lock for write.
 763  */
 764 int __kvm_set_memory_region(struct kvm *kvm,
 765                             struct kvm_userspace_memory_region *mem)
 766 {
 767         int r;
 768         gfn_t base_gfn;
 769         unsigned long npages;
 770         struct kvm_memory_slot *slot;
 771         struct kvm_memory_slot old, new;
 772         struct kvm_memslots *slots = NULL, *old_memslots;
 773         enum kvm_mr_change change;
 774
 775         r = check_memory_region_flags(mem);
 776         if (r)
 777                 goto out;
 778
 779         r = -EINVAL;
 780         /* General sanity checks */
 781         if (mem->memory_size & (PAGE_SIZE - 1))
 782                 goto out;
 783         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 784                 goto out;
 785         /* We can read the guest memory with __xxx_user() later on. */
 786         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 787             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 788              !access_ok(VERIFY_WRITE,
 789                         (void __user *)(unsigned long)mem->userspace_addr,
 790                         mem->memory_size)))
 791                 goto out;
 792         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 793                 goto out;
 794         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 795                 goto out;
 796
 797         slot = id_to_memslot(kvm->memslots, mem->slot);
 798         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 799         npages = mem->memory_size >> PAGE_SHIFT;
 800
 801         if (npages > KVM_MEM_MAX_NR_PAGES)
 802                 goto out;
 803
 804         if (!npages)
 805                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 806
 807         new = old = *slot;
 808
 809         new.id = mem->slot;
 810         new.base_gfn = base_gfn;
 811         new.npages = npages;
 812         new.flags = mem->flags;
 813
 814         if (npages) {
 815                 if (!old.npages)
 816                         change = KVM_MR_CREATE;
 817                 else { /* Modify an existing slot. */
 818                         if ((mem->userspace_addr != old.userspace_addr) ||
 819                             (npages != old.npages) ||
 820                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 821                                 goto out;
 822
 823                         if (base_gfn != old.base_gfn)
 824                                 change = KVM_MR_MOVE;
 825                         else if (new.flags != old.flags)
 826                                 change = KVM_MR_FLAGS_ONLY;
 827                         else { /* Nothing to change. */
 828                                 r = 0;
 829                                 goto out;
 830                         }
 831                 }
 832         } else if (old.npages) {
 833                 change = KVM_MR_DELETE;
 834         } else /* Modify a non-existent slot: disallowed. */
 835                 goto out;
 836
 837         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 838                 /* Check for overlaps */
 839                 r = -EEXIST;
 840                 kvm_for_each_memslot(slot, kvm->memslots) {
 841                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 842                             (slot->id == mem->slot))
 843                                 continue;
 844                         if (!((base_gfn + npages <= slot->base_gfn) ||
 845                               (base_gfn >= slot->base_gfn + slot->npages)))
 846                                 goto out;
 847                 }
 848         }
 849
 850         /* Free page dirty bitmap if unneeded */
 851         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 852                 new.dirty_bitmap = NULL;
 853
 854         r = -ENOMEM;
 855         if (change == KVM_MR_CREATE) {
 856                 new.userspace_addr = mem->userspace_addr;
 857
 858                 if (kvm_arch_create_memslot(kvm, &new, npages))
 859                         goto out_free;
 860         }
 861
 862         /* Allocate page dirty bitmap if needed */
 863         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 864                 if (kvm_create_dirty_bitmap(&new) < 0)
 865                         goto out_free;
 866         }
 867
 868         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 869                         GFP_KERNEL);
 870         if (!slots)
 871                 goto out_free;
 872
 873         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 874                 slot = id_to_memslot(slots, mem->slot);
 875                 slot->flags |= KVM_MEMSLOT_INVALID;
 876
 877                 old_memslots = install_new_memslots(kvm, slots);
 878
 879                 /* slot was deleted or moved, clear iommu mapping */
 880                 kvm_iommu_unmap_pages(kvm, &old);
 881                 /* From this point no new shadow pages pointing to a deleted,
 882                  * or moved, memslot will be created.
 883                  *
 884                  * validation of sp->gfn happens in:
 885                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 886                  *      - kvm_is_visible_gfn (mmu_check_roots)
 887                  */
 888                 kvm_arch_flush_shadow_memslot(kvm, slot);
 889
 890                 /*
 891                  * We can re-use the old_memslots from above, the only difference
 892                  * from the currently installed memslots is the invalid flag.  This
 893                  * will get overwritten by update_memslots anyway.
 894                  */
 895                 slots = old_memslots;
 896         }
 897
 898         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 899         if (r)
 900                 goto out_slots;
 901
 902         /* actual memory is freed via old in kvm_free_physmem_slot below */
 903         if (change == KVM_MR_DELETE) {
 904                 new.dirty_bitmap = NULL;
 905                 memset(&new.arch, 0, sizeof(new.arch));
 906         }
 907
 908         update_memslots(slots, &new);
 909         old_memslots = install_new_memslots(kvm, slots);
 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_memslot_prot(struct kvm_memory_slot *slot,
1093                                       gfn_t gfn, bool *writable)
1094 {
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 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1104 {
1105         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1106
1107         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1108 }
1109
1110 static int kvm_read_hva(void *data, void __user *hva, int len)
1111 {
1112         return __copy_from_user(data, hva, len);
1113 }
1114
1115 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1116 {
1117         return __copy_from_user_inatomic(data, hva, len);
1118 }
1119
1120 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1121         unsigned long start, int write, struct page **page)
1122 {
1123         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1124
1125         if (write)
1126                 flags |= FOLL_WRITE;
1127
1128         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1129 }
1130
1131 static inline int check_user_page_hwpoison(unsigned long addr)
1132 {
1133         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1134
1135         rc = __get_user_pages(current, current->mm, addr, 1,
1136                               flags, NULL, NULL, NULL);
1137         return rc == -EHWPOISON;
1138 }
1139
1140 /*
1141  * The atomic path to get the writable pfn which will be stored in @pfn,
1142  * true indicates success, otherwise false is returned.
1143  */
1144 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1145                             bool write_fault, bool *writable, pfn_t *pfn)
1146 {
1147         struct page *page[1];
1148         int npages;
1149
1150         if (!(async || atomic))
1151                 return false;
1152
1153         /*
1154          * Fast pin a writable pfn only if it is a write fault request
1155          * or the caller allows to map a writable pfn for a read fault
1156          * request.
1157          */
1158         if (!(write_fault || writable))
1159                 return false;
1160
1161         npages = __get_user_pages_fast(addr, 1, 1, page);
1162         if (npages == 1) {
1163                 *pfn = page_to_pfn(page[0]);
1164
1165                 if (writable)
1166                         *writable = true;
1167                 return true;
1168         }
1169
1170         return false;
1171 }
1172
1173 /*
1174  * The slow path to get the pfn of the specified host virtual address,
1175  * 1 indicates success, -errno is returned if error is detected.
1176  */
1177 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1178                            bool *writable, pfn_t *pfn)
1179 {
1180         struct page *page[1];
1181         int npages = 0;
1182
1183         might_sleep();
1184
1185         if (writable)
1186                 *writable = write_fault;
1187
1188         if (async) {
1189                 down_read(&current->mm->mmap_sem);
1190                 npages = get_user_page_nowait(current, current->mm,
1191                                               addr, write_fault, page);
1192                 up_read(&current->mm->mmap_sem);
1193         } else
1194                 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1195                                                    write_fault, 0, page,
1196                                                    FOLL_TOUCH|FOLL_HWPOISON);
1197         if (npages != 1)
1198                 return npages;
1199
1200         /* map read fault as writable if possible */
1201         if (unlikely(!write_fault) && writable) {
1202                 struct page *wpage[1];
1203
1204                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1205                 if (npages == 1) {
1206                         *writable = true;
1207                         put_page(page[0]);
1208                         page[0] = wpage[0];
1209                 }
1210
1211                 npages = 1;
1212         }
1213         *pfn = page_to_pfn(page[0]);
1214         return npages;
1215 }
1216
1217 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1218 {
1219         if (unlikely(!(vma->vm_flags & VM_READ)))
1220                 return false;
1221
1222         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1223                 return false;
1224
1225         return true;
1226 }
1227
1228 /*
1229  * Pin guest page in memory and return its pfn.
1230  * @addr: host virtual address which maps memory to the guest
1231  * @atomic: whether this function can sleep
1232  * @async: whether this function need to wait IO complete if the
1233  *         host page is not in the memory
1234  * @write_fault: whether we should get a writable host page
1235  * @writable: whether it allows to map a writable host page for !@write_fault
1236  *
1237  * The function will map a writable host page for these two cases:
1238  * 1): @write_fault = true
1239  * 2): @write_fault = false && @writable, @writable will tell the caller
1240  *     whether the mapping is writable.
1241  */
1242 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1243                         bool write_fault, bool *writable)
1244 {
1245         struct vm_area_struct *vma;
1246         pfn_t pfn = 0;
1247         int npages;
1248
1249         /* we can do it either atomically or asynchronously, not both */
1250         BUG_ON(atomic && async);
1251
1252         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1253                 return pfn;
1254
1255         if (atomic)
1256                 return KVM_PFN_ERR_FAULT;
1257
1258         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1259         if (npages == 1)
1260                 return pfn;
1261
1262         down_read(&current->mm->mmap_sem);
1263         if (npages == -EHWPOISON ||
1264               (!async && check_user_page_hwpoison(addr))) {
1265                 pfn = KVM_PFN_ERR_HWPOISON;
1266                 goto exit;
1267         }
1268
1269         vma = find_vma_intersection(current->mm, addr, addr + 1);
1270
1271         if (vma == NULL)
1272                 pfn = KVM_PFN_ERR_FAULT;
1273         else if ((vma->vm_flags & VM_PFNMAP)) {
1274                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1275                         vma->vm_pgoff;
1276                 BUG_ON(!kvm_is_reserved_pfn(pfn));
1277         } else {
1278                 if (async && vma_is_valid(vma, write_fault))
1279                         *async = true;
1280                 pfn = KVM_PFN_ERR_FAULT;
1281         }
1282 exit:
1283         up_read(&current->mm->mmap_sem);
1284         return pfn;
1285 }
1286
1287 static pfn_t
1288 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1289                      bool *async, bool write_fault, bool *writable)
1290 {
1291         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1292
1293         if (addr == KVM_HVA_ERR_RO_BAD)
1294                 return KVM_PFN_ERR_RO_FAULT;
1295
1296         if (kvm_is_error_hva(addr))
1297                 return KVM_PFN_NOSLOT;
1298
1299         /* Do not map writable pfn in the readonly memslot. */
1300         if (writable && memslot_is_readonly(slot)) {
1301                 *writable = false;
1302                 writable = NULL;
1303         }
1304
1305         return hva_to_pfn(addr, atomic, async, write_fault,
1306                           writable);
1307 }
1308
1309 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1310                           bool write_fault, bool *writable)
1311 {
1312         struct kvm_memory_slot *slot;
1313
1314         if (async)
1315                 *async = false;
1316
1317         slot = gfn_to_memslot(kvm, gfn);
1318
1319         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1320                                     writable);
1321 }
1322
1323 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1324 {
1325         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1326 }
1327 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1328
1329 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1330                        bool write_fault, bool *writable)
1331 {
1332         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1333 }
1334 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1335
1336 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1337 {
1338         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1339 }
1340 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1341
1342 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1343                       bool *writable)
1344 {
1345         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1346 }
1347 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1348
1349 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1350 {
1351         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1352 }
1353
1354 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1355 {
1356         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1357 }
1358 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1359
1360 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1361                                                                   int nr_pages)
1362 {
1363         unsigned long addr;
1364         gfn_t entry;
1365
1366         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1367         if (kvm_is_error_hva(addr))
1368                 return -1;
1369
1370         if (entry < nr_pages)
1371                 return 0;
1372
1373         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1374 }
1375 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1376
1377 static struct page *kvm_pfn_to_page(pfn_t pfn)
1378 {
1379         if (is_error_noslot_pfn(pfn))
1380                 return KVM_ERR_PTR_BAD_PAGE;
1381
1382         if (kvm_is_reserved_pfn(pfn)) {
1383                 WARN_ON(1);
1384                 return KVM_ERR_PTR_BAD_PAGE;
1385         }
1386
1387         return pfn_to_page(pfn);
1388 }
1389
1390 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1391 {
1392         pfn_t pfn;
1393
1394         pfn = gfn_to_pfn(kvm, gfn);
1395
1396         return kvm_pfn_to_page(pfn);
1397 }
1398
1399 EXPORT_SYMBOL_GPL(gfn_to_page);
1400
1401 void kvm_release_page_clean(struct page *page)
1402 {
1403         WARN_ON(is_error_page(page));
1404
1405         kvm_release_pfn_clean(page_to_pfn(page));
1406 }
1407 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1408
1409 void kvm_release_pfn_clean(pfn_t pfn)
1410 {
1411         if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1412                 put_page(pfn_to_page(pfn));
1413 }
1414 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1415
1416 void kvm_release_page_dirty(struct page *page)
1417 {
1418         WARN_ON(is_error_page(page));
1419
1420         kvm_release_pfn_dirty(page_to_pfn(page));
1421 }
1422 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1423
1424 static void kvm_release_pfn_dirty(pfn_t pfn)
1425 {
1426         kvm_set_pfn_dirty(pfn);
1427         kvm_release_pfn_clean(pfn);
1428 }
1429
1430 void kvm_set_pfn_dirty(pfn_t pfn)
1431 {
1432         if (!kvm_is_reserved_pfn(pfn)) {
1433                 struct page *page = pfn_to_page(pfn);
1434                 if (!PageReserved(page))
1435                         SetPageDirty(page);
1436         }
1437 }
1438 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1439
1440 void kvm_set_pfn_accessed(pfn_t pfn)
1441 {
1442         if (!kvm_is_reserved_pfn(pfn))
1443                 mark_page_accessed(pfn_to_page(pfn));
1444 }
1445 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1446
1447 void kvm_get_pfn(pfn_t pfn)
1448 {
1449         if (!kvm_is_reserved_pfn(pfn))
1450                 get_page(pfn_to_page(pfn));
1451 }
1452 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1453
1454 static int next_segment(unsigned long len, int offset)
1455 {
1456         if (len > PAGE_SIZE - offset)
1457                 return PAGE_SIZE - offset;
1458         else
1459                 return len;
1460 }
1461
1462 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1463                         int len)
1464 {
1465         int r;
1466         unsigned long addr;
1467
1468         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1469         if (kvm_is_error_hva(addr))
1470                 return -EFAULT;
1471         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1472         if (r)
1473                 return -EFAULT;
1474         return 0;
1475 }
1476 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1477
1478 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1479 {
1480         gfn_t gfn = gpa >> PAGE_SHIFT;
1481         int seg;
1482         int offset = offset_in_page(gpa);
1483         int ret;
1484
1485         while ((seg = next_segment(len, offset)) != 0) {
1486                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1487                 if (ret < 0)
1488                         return ret;
1489                 offset = 0;
1490                 len -= seg;
1491                 data += seg;
1492                 ++gfn;
1493         }
1494         return 0;
1495 }
1496 EXPORT_SYMBOL_GPL(kvm_read_guest);
1497
1498 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1499                           unsigned long len)
1500 {
1501         int r;
1502         unsigned long addr;
1503         gfn_t gfn = gpa >> PAGE_SHIFT;
1504         int offset = offset_in_page(gpa);
1505
1506         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1507         if (kvm_is_error_hva(addr))
1508                 return -EFAULT;
1509         pagefault_disable();
1510         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1511         pagefault_enable();
1512         if (r)
1513                 return -EFAULT;
1514         return 0;
1515 }
1516 EXPORT_SYMBOL(kvm_read_guest_atomic);
1517
1518 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1519                          int offset, int len)
1520 {
1521         int r;
1522         unsigned long addr;
1523
1524         addr = gfn_to_hva(kvm, gfn);
1525         if (kvm_is_error_hva(addr))
1526                 return -EFAULT;
1527         r = __copy_to_user((void __user *)addr + offset, data, len);
1528         if (r)
1529                 return -EFAULT;
1530         mark_page_dirty(kvm, gfn);
1531         return 0;
1532 }
1533 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1534
1535 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1536                     unsigned long len)
1537 {
1538         gfn_t gfn = gpa >> PAGE_SHIFT;
1539         int seg;
1540         int offset = offset_in_page(gpa);
1541         int ret;
1542
1543         while ((seg = next_segment(len, offset)) != 0) {
1544                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1545                 if (ret < 0)
1546                         return ret;
1547                 offset = 0;
1548                 len -= seg;
1549                 data += seg;
1550                 ++gfn;
1551         }
1552         return 0;
1553 }
1554
1555 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1556                               gpa_t gpa, unsigned long len)
1557 {
1558         struct kvm_memslots *slots = kvm_memslots(kvm);
1559         int offset = offset_in_page(gpa);
1560         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1561         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1562         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1563         gfn_t nr_pages_avail;
1564
1565         ghc->gpa = gpa;
1566         ghc->generation = slots->generation;
1567         ghc->len = len;
1568         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1569         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1570         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1571                 ghc->hva += offset;
1572         } else {
1573                 /*
1574                  * If the requested region crosses two memslots, we still
1575                  * verify that the entire region is valid here.
1576                  */
1577                 while (start_gfn <= end_gfn) {
1578                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1579                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1580                                                    &nr_pages_avail);
1581                         if (kvm_is_error_hva(ghc->hva))
1582                                 return -EFAULT;
1583                         start_gfn += nr_pages_avail;
1584                 }
1585                 /* Use the slow path for cross page reads and writes. */
1586                 ghc->memslot = NULL;
1587         }
1588         return 0;
1589 }
1590 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1591
1592 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1593                            void *data, unsigned long len)
1594 {
1595         struct kvm_memslots *slots = kvm_memslots(kvm);
1596         int r;
1597
1598         BUG_ON(len > ghc->len);
1599
1600         if (slots->generation != ghc->generation)
1601                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1602
1603         if (unlikely(!ghc->memslot))
1604                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1605
1606         if (kvm_is_error_hva(ghc->hva))
1607                 return -EFAULT;
1608
1609         r = __copy_to_user((void __user *)ghc->hva, data, len);
1610         if (r)
1611                 return -EFAULT;
1612         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1613
1614         return 0;
1615 }
1616 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1617
1618 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1619                            void *data, unsigned long len)
1620 {
1621         struct kvm_memslots *slots = kvm_memslots(kvm);
1622         int r;
1623
1624         BUG_ON(len > ghc->len);
1625
1626         if (slots->generation != ghc->generation)
1627                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1628
1629         if (unlikely(!ghc->memslot))
1630                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1631
1632         if (kvm_is_error_hva(ghc->hva))
1633                 return -EFAULT;
1634
1635         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1636         if (r)
1637                 return -EFAULT;
1638
1639         return 0;
1640 }
1641 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1642
1643 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1644 {
1645         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1646
1647         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1648 }
1649 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1650
1651 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1652 {
1653         gfn_t gfn = gpa >> PAGE_SHIFT;
1654         int seg;
1655         int offset = offset_in_page(gpa);
1656         int ret;
1657
1658         while ((seg = next_segment(len, offset)) != 0) {
1659                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1660                 if (ret < 0)
1661                         return ret;
1662                 offset = 0;
1663                 len -= seg;
1664                 ++gfn;
1665         }
1666         return 0;
1667 }
1668 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1669
1670 static void mark_page_dirty_in_slot(struct kvm *kvm,
1671                                     struct kvm_memory_slot *memslot,
1672                                     gfn_t gfn)
1673 {
1674         if (memslot && memslot->dirty_bitmap) {
1675                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1676
1677                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1678         }
1679 }
1680
1681 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1682 {
1683         struct kvm_memory_slot *memslot;
1684
1685         memslot = gfn_to_memslot(kvm, gfn);
1686         mark_page_dirty_in_slot(kvm, memslot, gfn);
1687 }
1688 EXPORT_SYMBOL_GPL(mark_page_dirty);
1689
1690 /*
1691  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1692  */
1693 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1694 {
1695         DEFINE_WAIT(wait);
1696
1697         for (;;) {
1698                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1699
1700                 if (kvm_arch_vcpu_runnable(vcpu)) {
1701                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1702                         break;
1703                 }
1704                 if (kvm_cpu_has_pending_timer(vcpu))
1705                         break;
1706                 if (signal_pending(current))
1707                         break;
1708
1709                 schedule();
1710         }
1711
1712         finish_wait(&vcpu->wq, &wait);
1713 }
1714 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1715
1716 #ifndef CONFIG_S390
1717 /*
1718  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1719  */
1720 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1721 {
1722         int me;
1723         int cpu = vcpu->cpu;
1724         wait_queue_head_t *wqp;
1725
1726         wqp = kvm_arch_vcpu_wq(vcpu);
1727         if (waitqueue_active(wqp)) {
1728                 wake_up_interruptible(wqp);
1729                 ++vcpu->stat.halt_wakeup;
1730         }
1731
1732         me = get_cpu();
1733         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1734                 if (kvm_arch_vcpu_should_kick(vcpu))
1735                         smp_send_reschedule(cpu);
1736         put_cpu();
1737 }
1738 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1739 #endif /* !CONFIG_S390 */
1740
1741 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1742 {
1743         struct pid *pid;
1744         struct task_struct *task = NULL;
1745         int ret = 0;
1746
1747         rcu_read_lock();
1748         pid = rcu_dereference(target->pid);
1749         if (pid)
1750                 task = get_pid_task(pid, PIDTYPE_PID);
1751         rcu_read_unlock();
1752         if (!task)
1753                 return ret;
1754         ret = yield_to(task, 1);
1755         put_task_struct(task);
1756
1757         return ret;
1758 }
1759 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1760
1761 /*
1762  * Helper that checks whether a VCPU is eligible for directed yield.
1763  * Most eligible candidate to yield is decided by following heuristics:
1764  *
1765  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1766  *  (preempted lock holder), indicated by @in_spin_loop.
1767  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1768  *
1769  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1770  *  chance last time (mostly it has become eligible now since we have probably
1771  *  yielded to lockholder in last iteration. This is done by toggling
1772  *  @dy_eligible each time a VCPU checked for eligibility.)
1773  *
1774  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1775  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1776  *  burning. Giving priority for a potential lock-holder increases lock
1777  *  progress.
1778  *
1779  *  Since algorithm is based on heuristics, accessing another VCPU data without
1780  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1781  *  and continue with next VCPU and so on.
1782  */
1783 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1784 {
1785 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1786         bool eligible;
1787
1788         eligible = !vcpu->spin_loop.in_spin_loop ||
1789                     vcpu->spin_loop.dy_eligible;
1790
1791         if (vcpu->spin_loop.in_spin_loop)
1792                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1793
1794         return eligible;
1795 #else
1796         return true;
1797 #endif
1798 }
1799
1800 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1801 {
1802         struct kvm *kvm = me->kvm;
1803         struct kvm_vcpu *vcpu;
1804         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1805         int yielded = 0;
1806         int try = 3;
1807         int pass;
1808         int i;
1809
1810         kvm_vcpu_set_in_spin_loop(me, true);
1811         /*
1812          * We boost the priority of a VCPU that is runnable but not
1813          * currently running, because it got preempted by something
1814          * else and called schedule in __vcpu_run.  Hopefully that
1815          * VCPU is holding the lock that we need and will release it.
1816          * We approximate round-robin by starting at the last boosted VCPU.
1817          */
1818         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1819                 kvm_for_each_vcpu(i, vcpu, kvm) {
1820                         if (!pass && i <= last_boosted_vcpu) {
1821                                 i = last_boosted_vcpu;
1822                                 continue;
1823                         } else if (pass && i > last_boosted_vcpu)
1824                                 break;
1825                         if (!ACCESS_ONCE(vcpu->preempted))
1826                                 continue;
1827                         if (vcpu == me)
1828                                 continue;
1829                         if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1830                                 continue;
1831                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1832                                 continue;
1833
1834                         yielded = kvm_vcpu_yield_to(vcpu);
1835                         if (yielded > 0) {
1836                                 kvm->last_boosted_vcpu = i;
1837                                 break;
1838                         } else if (yielded < 0) {
1839                                 try--;
1840                                 if (!try)
1841                                         break;
1842                         }
1843                 }
1844         }
1845         kvm_vcpu_set_in_spin_loop(me, false);
1846
1847         /* Ensure vcpu is not eligible during next spinloop */
1848         kvm_vcpu_set_dy_eligible(me, false);
1849 }
1850 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1851
1852 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1853 {
1854         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1855         struct page *page;
1856
1857         if (vmf->pgoff == 0)
1858                 page = virt_to_page(vcpu->run);
1859 #ifdef CONFIG_X86
1860         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1861                 page = virt_to_page(vcpu->arch.pio_data);
1862 #endif
1863 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1864         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1865                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1866 #endif
1867         else
1868                 return kvm_arch_vcpu_fault(vcpu, vmf);
1869         get_page(page);
1870         vmf->page = page;
1871         return 0;
1872 }
1873
1874 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1875         .fault = kvm_vcpu_fault,
1876 };
1877
1878 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1879 {
1880         vma->vm_ops = &kvm_vcpu_vm_ops;
1881         return 0;
1882 }
1883
1884 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1885 {
1886         struct kvm_vcpu *vcpu = filp->private_data;
1887
1888         kvm_put_kvm(vcpu->kvm);
1889         return 0;
1890 }
1891
1892 static struct file_operations kvm_vcpu_fops = {
1893         .release        = kvm_vcpu_release,
1894         .unlocked_ioctl = kvm_vcpu_ioctl,
1895 #ifdef CONFIG_COMPAT
1896         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1897 #endif
1898         .mmap           = kvm_vcpu_mmap,
1899         .llseek         = noop_llseek,
1900 };
1901
1902 /*
1903  * Allocates an inode for the vcpu.
1904  */
1905 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1906 {
1907         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1908 }
1909
1910 /*
1911  * Creates some virtual cpus.  Good luck creating more than one.
1912  */
1913 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1914 {
1915         int r;
1916         struct kvm_vcpu *vcpu, *v;
1917
1918         if (id >= KVM_MAX_VCPUS)
1919                 return -EINVAL;
1920
1921         vcpu = kvm_arch_vcpu_create(kvm, id);
1922         if (IS_ERR(vcpu))
1923                 return PTR_ERR(vcpu);
1924
1925         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1926
1927         r = kvm_arch_vcpu_setup(vcpu);
1928         if (r)
1929                 goto vcpu_destroy;
1930
1931         mutex_lock(&kvm->lock);
1932         if (!kvm_vcpu_compatible(vcpu)) {
1933                 r = -EINVAL;
1934                 goto unlock_vcpu_destroy;
1935         }
1936         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1937                 r = -EINVAL;
1938                 goto unlock_vcpu_destroy;
1939         }
1940
1941         kvm_for_each_vcpu(r, v, kvm)
1942                 if (v->vcpu_id == id) {
1943                         r = -EEXIST;
1944                         goto unlock_vcpu_destroy;
1945                 }
1946
1947         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1948
1949         /* Now it's all set up, let userspace reach it */
1950         kvm_get_kvm(kvm);
1951         r = create_vcpu_fd(vcpu);
1952         if (r < 0) {
1953                 kvm_put_kvm(kvm);
1954                 goto unlock_vcpu_destroy;
1955         }
1956
1957         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1958         smp_wmb();
1959         atomic_inc(&kvm->online_vcpus);
1960
1961         mutex_unlock(&kvm->lock);
1962         kvm_arch_vcpu_postcreate(vcpu);
1963         return r;
1964
1965 unlock_vcpu_destroy:
1966         mutex_unlock(&kvm->lock);
1967 vcpu_destroy:
1968         kvm_arch_vcpu_destroy(vcpu);
1969         return r;
1970 }
1971
1972 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1973 {
1974         if (sigset) {
1975                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1976                 vcpu->sigset_active = 1;
1977                 vcpu->sigset = *sigset;
1978         } else
1979                 vcpu->sigset_active = 0;
1980         return 0;
1981 }
1982
1983 static long kvm_vcpu_ioctl(struct file *filp,
1984                            unsigned int ioctl, unsigned long arg)
1985 {
1986         struct kvm_vcpu *vcpu = filp->private_data;
1987         void __user *argp = (void __user *)arg;
1988         int r;
1989         struct kvm_fpu *fpu = NULL;
1990         struct kvm_sregs *kvm_sregs = NULL;
1991
1992         if (vcpu->kvm->mm != current->mm)
1993                 return -EIO;
1994
1995         if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
1996                 return -EINVAL;
1997
1998 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1999         /*
2000          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2001          * so vcpu_load() would break it.
2002          */
2003         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2004                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2005 #endif
2006
2007
2008         r = vcpu_load(vcpu);
2009         if (r)
2010                 return r;
2011         switch (ioctl) {
2012         case KVM_RUN:
2013                 r = -EINVAL;
2014                 if (arg)
2015                         goto out;
2016                 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2017                         /* The thread running this VCPU changed. */
2018                         struct pid *oldpid = vcpu->pid;
2019                         struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2020                         rcu_assign_pointer(vcpu->pid, newpid);
2021                         if (oldpid)
2022                                 synchronize_rcu();
2023                         put_pid(oldpid);
2024                 }
2025                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2026                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2027                 break;
2028         case KVM_GET_REGS: {
2029                 struct kvm_regs *kvm_regs;
2030
2031                 r = -ENOMEM;
2032                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2033                 if (!kvm_regs)
2034                         goto out;
2035                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2036                 if (r)
2037                         goto out_free1;
2038                 r = -EFAULT;
2039                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2040                         goto out_free1;
2041                 r = 0;
2042 out_free1:
2043                 kfree(kvm_regs);
2044                 break;
2045         }
2046         case KVM_SET_REGS: {
2047                 struct kvm_regs *kvm_regs;
2048
2049                 r = -ENOMEM;
2050                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2051                 if (IS_ERR(kvm_regs)) {
2052                         r = PTR_ERR(kvm_regs);
2053                         goto out;
2054                 }
2055                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2056                 kfree(kvm_regs);
2057                 break;
2058         }
2059         case KVM_GET_SREGS: {
2060                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2061                 r = -ENOMEM;
2062                 if (!kvm_sregs)
2063                         goto out;
2064                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2065                 if (r)
2066                         goto out;
2067                 r = -EFAULT;
2068                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2069                         goto out;
2070                 r = 0;
2071                 break;
2072         }
2073         case KVM_SET_SREGS: {
2074                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2075                 if (IS_ERR(kvm_sregs)) {
2076                         r = PTR_ERR(kvm_sregs);
2077                         kvm_sregs = NULL;
2078                         goto out;
2079                 }
2080                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2081                 break;
2082         }
2083         case KVM_GET_MP_STATE: {
2084                 struct kvm_mp_state mp_state;
2085
2086                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2087                 if (r)
2088                         goto out;
2089                 r = -EFAULT;
2090                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2091                         goto out;
2092                 r = 0;
2093                 break;
2094         }
2095         case KVM_SET_MP_STATE: {
2096                 struct kvm_mp_state mp_state;
2097
2098                 r = -EFAULT;
2099                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2100                         goto out;
2101                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2102                 break;
2103         }
2104         case KVM_TRANSLATE: {
2105                 struct kvm_translation tr;
2106
2107                 r = -EFAULT;
2108                 if (copy_from_user(&tr, argp, sizeof tr))
2109                         goto out;
2110                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2111                 if (r)
2112                         goto out;
2113                 r = -EFAULT;
2114                 if (copy_to_user(argp, &tr, sizeof tr))
2115                         goto out;
2116                 r = 0;
2117                 break;
2118         }
2119         case KVM_SET_GUEST_DEBUG: {
2120                 struct kvm_guest_debug dbg;
2121
2122                 r = -EFAULT;
2123                 if (copy_from_user(&dbg, argp, sizeof dbg))
2124                         goto out;
2125                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2126                 break;
2127         }
2128         case KVM_SET_SIGNAL_MASK: {
2129                 struct kvm_signal_mask __user *sigmask_arg = argp;
2130                 struct kvm_signal_mask kvm_sigmask;
2131                 sigset_t sigset, *p;
2132
2133                 p = NULL;
2134                 if (argp) {
2135                         r = -EFAULT;
2136                         if (copy_from_user(&kvm_sigmask, argp,
2137                                            sizeof kvm_sigmask))
2138                                 goto out;
2139                         r = -EINVAL;
2140                         if (kvm_sigmask.len != sizeof sigset)
2141                                 goto out;
2142                         r = -EFAULT;
2143                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2144                                            sizeof sigset))
2145                                 goto out;
2146                         p = &sigset;
2147                 }
2148                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2149                 break;
2150         }
2151         case KVM_GET_FPU: {
2152                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2153                 r = -ENOMEM;
2154                 if (!fpu)
2155                         goto out;
2156                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2157                 if (r)
2158                         goto out;
2159                 r = -EFAULT;
2160                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2161                         goto out;
2162                 r = 0;
2163                 break;
2164         }
2165         case KVM_SET_FPU: {
2166                 fpu = memdup_user(argp, sizeof(*fpu));
2167                 if (IS_ERR(fpu)) {
2168                         r = PTR_ERR(fpu);
2169                         fpu = NULL;
2170                         goto out;
2171                 }
2172                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2173                 break;
2174         }
2175         default:
2176                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2177         }
2178 out:
2179         vcpu_put(vcpu);
2180         kfree(fpu);
2181         kfree(kvm_sregs);
2182         return r;
2183 }
2184
2185 #ifdef CONFIG_COMPAT
2186 static long kvm_vcpu_compat_ioctl(struct file *filp,
2187                                   unsigned int ioctl, unsigned long arg)
2188 {
2189         struct kvm_vcpu *vcpu = filp->private_data;
2190         void __user *argp = compat_ptr(arg);
2191         int r;
2192
2193         if (vcpu->kvm->mm != current->mm)
2194                 return -EIO;
2195
2196         switch (ioctl) {
2197         case KVM_SET_SIGNAL_MASK: {
2198                 struct kvm_signal_mask __user *sigmask_arg = argp;
2199                 struct kvm_signal_mask kvm_sigmask;
2200                 compat_sigset_t csigset;
2201                 sigset_t sigset;
2202
2203                 if (argp) {
2204                         r = -EFAULT;
2205                         if (copy_from_user(&kvm_sigmask, argp,
2206                                            sizeof kvm_sigmask))
2207                                 goto out;
2208                         r = -EINVAL;
2209                         if (kvm_sigmask.len != sizeof csigset)
2210                                 goto out;
2211                         r = -EFAULT;
2212                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2213                                            sizeof csigset))
2214                                 goto out;
2215                         sigset_from_compat(&sigset, &csigset);
2216                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2217                 } else
2218                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2219                 break;
2220         }
2221         default:
2222                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2223         }
2224
2225 out:
2226         return r;
2227 }
2228 #endif
2229
2230 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2231                                  int (*accessor)(struct kvm_device *dev,
2232                                                  struct kvm_device_attr *attr),
2233                                  unsigned long arg)
2234 {
2235         struct kvm_device_attr attr;
2236
2237         if (!accessor)
2238                 return -EPERM;
2239
2240         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2241                 return -EFAULT;
2242
2243         return accessor(dev, &attr);
2244 }
2245
2246 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2247                              unsigned long arg)
2248 {
2249         struct kvm_device *dev = filp->private_data;
2250
2251         switch (ioctl) {
2252         case KVM_SET_DEVICE_ATTR:
2253                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2254         case KVM_GET_DEVICE_ATTR:
2255                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2256         case KVM_HAS_DEVICE_ATTR:
2257                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2258         default:
2259                 if (dev->ops->ioctl)
2260                         return dev->ops->ioctl(dev, ioctl, arg);
2261
2262                 return -ENOTTY;
2263         }
2264 }
2265
2266 static int kvm_device_release(struct inode *inode, struct file *filp)
2267 {
2268         struct kvm_device *dev = filp->private_data;
2269         struct kvm *kvm = dev->kvm;
2270
2271         kvm_put_kvm(kvm);
2272         return 0;
2273 }
2274
2275 static const struct file_operations kvm_device_fops = {
2276         .unlocked_ioctl = kvm_device_ioctl,
2277 #ifdef CONFIG_COMPAT
2278         .compat_ioctl = kvm_device_ioctl,
2279 #endif
2280         .release = kvm_device_release,
2281 };
2282
2283 struct kvm_device *kvm_device_from_filp(struct file *filp)
2284 {
2285         if (filp->f_op != &kvm_device_fops)
2286                 return NULL;
2287
2288         return filp->private_data;
2289 }
2290
2291 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2292 #ifdef CONFIG_KVM_MPIC
2293         [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
2294         [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
2295 #endif
2296
2297 #ifdef CONFIG_KVM_XICS
2298         [KVM_DEV_TYPE_XICS]             = &kvm_xics_ops,
2299 #endif
2300 };
2301
2302 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2303 {
2304         if (type >= ARRAY_SIZE(kvm_device_ops_table))
2305                 return -ENOSPC;
2306
2307         if (kvm_device_ops_table[type] != NULL)
2308                 return -EEXIST;
2309
2310         kvm_device_ops_table[type] = ops;
2311         return 0;
2312 }
2313
2314 void kvm_unregister_device_ops(u32 type)
2315 {
2316         if (kvm_device_ops_table[type] != NULL)
2317                 kvm_device_ops_table[type] = NULL;
2318 }
2319
2320 static int kvm_ioctl_create_device(struct kvm *kvm,
2321                                    struct kvm_create_device *cd)
2322 {
2323         struct kvm_device_ops *ops = NULL;
2324         struct kvm_device *dev;
2325         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2326         int ret;
2327
2328         if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2329                 return -ENODEV;
2330
2331         ops = kvm_device_ops_table[cd->type];
2332         if (ops == NULL)
2333                 return -ENODEV;
2334
2335         if (test)
2336                 return 0;
2337
2338         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2339         if (!dev)
2340                 return -ENOMEM;
2341
2342         dev->ops = ops;
2343         dev->kvm = kvm;
2344
2345         ret = ops->create(dev, cd->type);
2346         if (ret < 0) {
2347                 kfree(dev);
2348                 return ret;
2349         }
2350
2351         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2352         if (ret < 0) {
2353                 ops->destroy(dev);
2354                 return ret;
2355         }
2356
2357         list_add(&dev->vm_node, &kvm->devices);
2358         kvm_get_kvm(kvm);
2359         cd->fd = ret;
2360         return 0;
2361 }
2362
2363 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2364 {
2365         switch (arg) {
2366         case KVM_CAP_USER_MEMORY:
2367         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2368         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2369 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2370         case KVM_CAP_SET_BOOT_CPU_ID:
2371 #endif
2372         case KVM_CAP_INTERNAL_ERROR_DATA:
2373 #ifdef CONFIG_HAVE_KVM_MSI
2374         case KVM_CAP_SIGNAL_MSI:
2375 #endif
2376 #ifdef CONFIG_HAVE_KVM_IRQFD
2377         case KVM_CAP_IRQFD_RESAMPLE:
2378 #endif
2379         case KVM_CAP_CHECK_EXTENSION_VM:
2380                 return 1;
2381 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2382         case KVM_CAP_IRQ_ROUTING:
2383                 return KVM_MAX_IRQ_ROUTES;
2384 #endif
2385         default:
2386                 break;
2387         }
2388         return kvm_vm_ioctl_check_extension(kvm, arg);
2389 }
2390
2391 static long kvm_vm_ioctl(struct file *filp,
2392                            unsigned int ioctl, unsigned long arg)
2393 {
2394         struct kvm *kvm = filp->private_data;
2395         void __user *argp = (void __user *)arg;
2396         int r;
2397
2398         if (kvm->mm != current->mm)
2399                 return -EIO;
2400         switch (ioctl) {
2401         case KVM_CREATE_VCPU:
2402                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2403                 break;
2404         case KVM_SET_USER_MEMORY_REGION: {
2405                 struct kvm_userspace_memory_region kvm_userspace_mem;
2406
2407                 r = -EFAULT;
2408                 if (copy_from_user(&kvm_userspace_mem, argp,
2409                                                 sizeof kvm_userspace_mem))
2410                         goto out;
2411
2412                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2413                 break;
2414         }
2415         case KVM_GET_DIRTY_LOG: {
2416                 struct kvm_dirty_log log;
2417
2418                 r = -EFAULT;
2419                 if (copy_from_user(&log, argp, sizeof log))
2420                         goto out;
2421                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2422                 break;
2423         }
2424 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2425         case KVM_REGISTER_COALESCED_MMIO: {
2426                 struct kvm_coalesced_mmio_zone zone;
2427                 r = -EFAULT;
2428                 if (copy_from_user(&zone, argp, sizeof zone))
2429                         goto out;
2430                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2431                 break;
2432         }
2433         case KVM_UNREGISTER_COALESCED_MMIO: {
2434                 struct kvm_coalesced_mmio_zone zone;
2435                 r = -EFAULT;
2436                 if (copy_from_user(&zone, argp, sizeof zone))
2437                         goto out;
2438                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2439                 break;
2440         }
2441 #endif
2442         case KVM_IRQFD: {
2443                 struct kvm_irqfd data;
2444
2445                 r = -EFAULT;
2446                 if (copy_from_user(&data, argp, sizeof data))
2447                         goto out;
2448                 r = kvm_irqfd(kvm, &data);
2449                 break;
2450         }
2451         case KVM_IOEVENTFD: {
2452                 struct kvm_ioeventfd data;
2453
2454                 r = -EFAULT;
2455                 if (copy_from_user(&data, argp, sizeof data))
2456                         goto out;
2457                 r = kvm_ioeventfd(kvm, &data);
2458                 break;
2459         }
2460 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2461         case KVM_SET_BOOT_CPU_ID:
2462                 r = 0;
2463                 mutex_lock(&kvm->lock);
2464                 if (atomic_read(&kvm->online_vcpus) != 0)
2465                         r = -EBUSY;
2466                 else
2467                         kvm->bsp_vcpu_id = arg;
2468                 mutex_unlock(&kvm->lock);
2469                 break;
2470 #endif
2471 #ifdef CONFIG_HAVE_KVM_MSI
2472         case KVM_SIGNAL_MSI: {
2473                 struct kvm_msi msi;
2474
2475                 r = -EFAULT;
2476                 if (copy_from_user(&msi, argp, sizeof msi))
2477                         goto out;
2478                 r = kvm_send_userspace_msi(kvm, &msi);
2479                 break;
2480         }
2481 #endif
2482 #ifdef __KVM_HAVE_IRQ_LINE
2483         case KVM_IRQ_LINE_STATUS:
2484         case KVM_IRQ_LINE: {
2485                 struct kvm_irq_level irq_event;
2486
2487                 r = -EFAULT;
2488                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2489                         goto out;
2490
2491                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2492                                         ioctl == KVM_IRQ_LINE_STATUS);
2493                 if (r)
2494                         goto out;
2495
2496                 r = -EFAULT;
2497                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2498                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2499                                 goto out;
2500                 }
2501
2502                 r = 0;
2503                 break;
2504         }
2505 #endif
2506 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2507         case KVM_SET_GSI_ROUTING: {
2508                 struct kvm_irq_routing routing;
2509                 struct kvm_irq_routing __user *urouting;
2510                 struct kvm_irq_routing_entry *entries;
2511
2512                 r = -EFAULT;
2513                 if (copy_from_user(&routing, argp, sizeof(routing)))
2514                         goto out;
2515                 r = -EINVAL;
2516                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2517                         goto out;
2518                 if (routing.flags)
2519                         goto out;
2520                 r = -ENOMEM;
2521                 entries = vmalloc(routing.nr * sizeof(*entries));
2522                 if (!entries)
2523                         goto out;
2524                 r = -EFAULT;
2525                 urouting = argp;
2526                 if (copy_from_user(entries, urouting->entries,
2527                                    routing.nr * sizeof(*entries)))
2528                         goto out_free_irq_routing;
2529                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2530                                         routing.flags);
2531         out_free_irq_routing:
2532                 vfree(entries);
2533                 break;
2534         }
2535 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2536         case KVM_CREATE_DEVICE: {
2537                 struct kvm_create_device cd;
2538
2539                 r = -EFAULT;
2540                 if (copy_from_user(&cd, argp, sizeof(cd)))
2541                         goto out;
2542
2543                 r = kvm_ioctl_create_device(kvm, &cd);
2544                 if (r)
2545                         goto out;
2546
2547                 r = -EFAULT;
2548                 if (copy_to_user(argp, &cd, sizeof(cd)))
2549                         goto out;
2550
2551                 r = 0;
2552                 break;
2553         }
2554         case KVM_CHECK_EXTENSION:
2555                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2556                 break;
2557         default:
2558                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2559         }
2560 out:
2561         return r;
2562 }
2563
2564 #ifdef CONFIG_COMPAT
2565 struct compat_kvm_dirty_log {
2566         __u32 slot;
2567         __u32 padding1;
2568         union {
2569                 compat_uptr_t dirty_bitmap; /* one bit per page */
2570                 __u64 padding2;
2571         };
2572 };
2573
2574 static long kvm_vm_compat_ioctl(struct file *filp,
2575                            unsigned int ioctl, unsigned long arg)
2576 {
2577         struct kvm *kvm = filp->private_data;
2578         int r;
2579
2580         if (kvm->mm != current->mm)
2581                 return -EIO;
2582         switch (ioctl) {
2583         case KVM_GET_DIRTY_LOG: {
2584                 struct compat_kvm_dirty_log compat_log;
2585                 struct kvm_dirty_log log;
2586
2587                 r = -EFAULT;
2588                 if (copy_from_user(&compat_log, (void __user *)arg,
2589                                    sizeof(compat_log)))
2590                         goto out;
2591                 log.slot         = compat_log.slot;
2592                 log.padding1     = compat_log.padding1;
2593                 log.padding2     = compat_log.padding2;
2594                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2595
2596                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2597                 break;
2598         }
2599         default:
2600                 r = kvm_vm_ioctl(filp, ioctl, arg);
2601         }
2602
2603 out:
2604         return r;
2605 }
2606 #endif
2607
2608 static struct file_operations kvm_vm_fops = {
2609         .release        = kvm_vm_release,
2610         .unlocked_ioctl = kvm_vm_ioctl,
2611 #ifdef CONFIG_COMPAT
2612         .compat_ioctl   = kvm_vm_compat_ioctl,
2613 #endif
2614         .llseek         = noop_llseek,
2615 };
2616
2617 static int kvm_dev_ioctl_create_vm(unsigned long type)
2618 {
2619         int r;
2620         struct kvm *kvm;
2621
2622         kvm = kvm_create_vm(type);
2623         if (IS_ERR(kvm))
2624                 return PTR_ERR(kvm);
2625 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2626         r = kvm_coalesced_mmio_init(kvm);
2627         if (r < 0) {
2628                 kvm_put_kvm(kvm);
2629                 return r;
2630         }
2631 #endif
2632         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2633         if (r < 0)
2634                 kvm_put_kvm(kvm);
2635
2636         return r;
2637 }
2638
2639 static long kvm_dev_ioctl(struct file *filp,
2640                           unsigned int ioctl, unsigned long arg)
2641 {
2642         long r = -EINVAL;
2643
2644         switch (ioctl) {
2645         case KVM_GET_API_VERSION:
2646                 if (arg)
2647                         goto out;
2648                 r = KVM_API_VERSION;
2649                 break;
2650         case KVM_CREATE_VM:
2651                 r = kvm_dev_ioctl_create_vm(arg);
2652                 break;
2653         case KVM_CHECK_EXTENSION:
2654                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2655                 break;
2656         case KVM_GET_VCPU_MMAP_SIZE:
2657                 if (arg)
2658                         goto out;
2659                 r = PAGE_SIZE;     /* struct kvm_run */
2660 #ifdef CONFIG_X86
2661                 r += PAGE_SIZE;    /* pio data page */
2662 #endif
2663 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2664                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2665 #endif
2666                 break;
2667         case KVM_TRACE_ENABLE:
2668         case KVM_TRACE_PAUSE:
2669         case KVM_TRACE_DISABLE:
2670                 r = -EOPNOTSUPP;
2671                 break;
2672         default:
2673                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2674         }
2675 out:
2676         return r;
2677 }
2678
2679 static struct file_operations kvm_chardev_ops = {
2680         .unlocked_ioctl = kvm_dev_ioctl,
2681         .compat_ioctl   = kvm_dev_ioctl,
2682         .llseek         = noop_llseek,
2683 };
2684
2685 static struct miscdevice kvm_dev = {
2686         KVM_MINOR,
2687         "kvm",
2688         &kvm_chardev_ops,
2689 };
2690
2691 static void hardware_enable_nolock(void *junk)
2692 {
2693         int cpu = raw_smp_processor_id();
2694         int r;
2695
2696         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2697                 return;
2698
2699         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2700
2701         r = kvm_arch_hardware_enable();
2702
2703         if (r) {
2704                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2705                 atomic_inc(&hardware_enable_failed);
2706                 printk(KERN_INFO "kvm: enabling virtualization on "
2707                                  "CPU%d failed\n", cpu);
2708         }
2709 }
2710
2711 static void hardware_enable(void)
2712 {
2713         raw_spin_lock(&kvm_count_lock);
2714         if (kvm_usage_count)
2715                 hardware_enable_nolock(NULL);
2716         raw_spin_unlock(&kvm_count_lock);
2717 }
2718
2719 static void hardware_disable_nolock(void *junk)
2720 {
2721         int cpu = raw_smp_processor_id();
2722
2723         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2724                 return;
2725         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2726         kvm_arch_hardware_disable();
2727 }
2728
2729 static void hardware_disable(void)
2730 {
2731         raw_spin_lock(&kvm_count_lock);
2732         if (kvm_usage_count)
2733                 hardware_disable_nolock(NULL);
2734         raw_spin_unlock(&kvm_count_lock);
2735 }
2736
2737 static void hardware_disable_all_nolock(void)
2738 {
2739         BUG_ON(!kvm_usage_count);
2740
2741         kvm_usage_count--;
2742         if (!kvm_usage_count)
2743                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2744 }
2745
2746 static void hardware_disable_all(void)
2747 {
2748         raw_spin_lock(&kvm_count_lock);
2749         hardware_disable_all_nolock();
2750         raw_spin_unlock(&kvm_count_lock);
2751 }
2752
2753 static int hardware_enable_all(void)
2754 {
2755         int r = 0;
2756
2757         raw_spin_lock(&kvm_count_lock);
2758
2759         kvm_usage_count++;
2760         if (kvm_usage_count == 1) {
2761                 atomic_set(&hardware_enable_failed, 0);
2762                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2763
2764                 if (atomic_read(&hardware_enable_failed)) {
2765                         hardware_disable_all_nolock();
2766                         r = -EBUSY;
2767                 }
2768         }
2769
2770         raw_spin_unlock(&kvm_count_lock);
2771
2772         return r;
2773 }
2774
2775 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2776                            void *v)
2777 {
2778         int cpu = (long)v;
2779
2780         val &= ~CPU_TASKS_FROZEN;
2781         switch (val) {
2782         case CPU_DYING:
2783                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2784                        cpu);
2785                 hardware_disable();
2786                 break;
2787         case CPU_STARTING:
2788                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2789                        cpu);
2790                 hardware_enable();
2791                 break;
2792         }
2793         return NOTIFY_OK;
2794 }
2795
2796 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2797                       void *v)
2798 {
2799         /*
2800          * Some (well, at least mine) BIOSes hang on reboot if
2801          * in vmx root mode.
2802          *
2803          * And Intel TXT required VMX off for all cpu when system shutdown.
2804          */
2805         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2806         kvm_rebooting = true;
2807         on_each_cpu(hardware_disable_nolock, NULL, 1);
2808         return NOTIFY_OK;
2809 }
2810
2811 static struct notifier_block kvm_reboot_notifier = {
2812         .notifier_call = kvm_reboot,
2813         .priority = 0,
2814 };
2815
2816 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2817 {
2818         int i;
2819
2820         for (i = 0; i < bus->dev_count; i++) {
2821                 struct kvm_io_device *pos = bus->range[i].dev;
2822
2823                 kvm_iodevice_destructor(pos);
2824         }
2825         kfree(bus);
2826 }
2827
2828 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2829                                  const struct kvm_io_range *r2)
2830 {
2831         if (r1->addr < r2->addr)
2832                 return -1;
2833         if (r1->addr + r1->len > r2->addr + r2->len)
2834                 return 1;
2835         return 0;
2836 }
2837
2838 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2839 {
2840         return kvm_io_bus_cmp(p1, p2);
2841 }
2842
2843 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2844                           gpa_t addr, int len)
2845 {
2846         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2847                 .addr = addr,
2848                 .len = len,
2849                 .dev = dev,
2850         };
2851
2852         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2853                 kvm_io_bus_sort_cmp, NULL);
2854
2855         return 0;
2856 }
2857
2858 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2859                              gpa_t addr, int len)
2860 {
2861         struct kvm_io_range *range, key;
2862         int off;
2863
2864         key = (struct kvm_io_range) {
2865                 .addr = addr,
2866                 .len = len,
2867         };
2868
2869         range = bsearch(&key, bus->range, bus->dev_count,
2870                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2871         if (range == NULL)
2872                 return -ENOENT;
2873
2874         off = range - bus->range;
2875
2876         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2877                 off--;
2878
2879         return off;
2880 }
2881
2882 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2883                               struct kvm_io_range *range, const void *val)
2884 {
2885         int idx;
2886
2887         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2888         if (idx < 0)
2889                 return -EOPNOTSUPP;
2890
2891         while (idx < bus->dev_count &&
2892                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2893                 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2894                                         range->len, val))
2895                         return idx;
2896                 idx++;
2897         }
2898
2899         return -EOPNOTSUPP;
2900 }
2901
2902 /* kvm_io_bus_write - called under kvm->slots_lock */
2903 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2904                      int len, const void *val)
2905 {
2906         struct kvm_io_bus *bus;
2907         struct kvm_io_range range;
2908         int r;
2909
2910         range = (struct kvm_io_range) {
2911                 .addr = addr,
2912                 .len = len,
2913         };
2914
2915         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2916         r = __kvm_io_bus_write(bus, &range, val);
2917         return r < 0 ? r : 0;
2918 }
2919
2920 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2921 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2922                             int len, const void *val, long cookie)
2923 {
2924         struct kvm_io_bus *bus;
2925         struct kvm_io_range range;
2926
2927         range = (struct kvm_io_range) {
2928                 .addr = addr,
2929                 .len = len,
2930         };
2931
2932         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2933
2934         /* First try the device referenced by cookie. */
2935         if ((cookie >= 0) && (cookie < bus->dev_count) &&
2936             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2937                 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2938                                         val))
2939                         return cookie;
2940
2941         /*
2942          * cookie contained garbage; fall back to search and return the
2943          * correct cookie value.
2944          */
2945         return __kvm_io_bus_write(bus, &range, val);
2946 }
2947
2948 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2949                              void *val)
2950 {
2951         int idx;
2952
2953         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2954         if (idx < 0)
2955                 return -EOPNOTSUPP;
2956
2957         while (idx < bus->dev_count &&
2958                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2959                 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2960                                        range->len, val))
2961                         return idx;
2962                 idx++;
2963         }
2964
2965         return -EOPNOTSUPP;
2966 }
2967 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
2968
2969 /* kvm_io_bus_read - called under kvm->slots_lock */
2970 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2971                     int len, void *val)
2972 {
2973         struct kvm_io_bus *bus;
2974         struct kvm_io_range range;
2975         int r;
2976
2977         range = (struct kvm_io_range) {
2978                 .addr = addr,
2979                 .len = len,
2980         };
2981
2982         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2983         r = __kvm_io_bus_read(bus, &range, val);
2984         return r < 0 ? r : 0;
2985 }
2986
2987
2988 /* Caller must hold slots_lock. */
2989 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2990                             int len, struct kvm_io_device *dev)
2991 {
2992         struct kvm_io_bus *new_bus, *bus;
2993
2994         bus = kvm->buses[bus_idx];
2995         /* exclude ioeventfd which is limited by maximum fd */
2996         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2997                 return -ENOSPC;
2998
2999         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3000                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3001         if (!new_bus)
3002                 return -ENOMEM;
3003         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3004                sizeof(struct kvm_io_range)));
3005         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3006         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3007         synchronize_srcu_expedited(&kvm->srcu);
3008         kfree(bus);
3009
3010         return 0;
3011 }
3012
3013 /* Caller must hold slots_lock. */
3014 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3015                               struct kvm_io_device *dev)
3016 {
3017         int i, r;
3018         struct kvm_io_bus *new_bus, *bus;
3019
3020         bus = kvm->buses[bus_idx];
3021         r = -ENOENT;
3022         for (i = 0; i < bus->dev_count; i++)
3023                 if (bus->range[i].dev == dev) {
3024                         r = 0;
3025                         break;
3026                 }
3027
3028         if (r)
3029                 return r;
3030
3031         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3032                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3033         if (!new_bus)
3034                 return -ENOMEM;
3035
3036         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3037         new_bus->dev_count--;
3038         memcpy(new_bus->range + i, bus->range + i + 1,
3039                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3040
3041         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3042         synchronize_srcu_expedited(&kvm->srcu);
3043         kfree(bus);
3044         return r;
3045 }
3046
3047 static struct notifier_block kvm_cpu_notifier = {
3048         .notifier_call = kvm_cpu_hotplug,
3049 };
3050
3051 static int vm_stat_get(void *_offset, u64 *val)
3052 {
3053         unsigned offset = (long)_offset;
3054         struct kvm *kvm;
3055
3056         *val = 0;
3057         spin_lock(&kvm_lock);
3058         list_for_each_entry(kvm, &vm_list, vm_list)
3059                 *val += *(u32 *)((void *)kvm + offset);
3060         spin_unlock(&kvm_lock);
3061         return 0;
3062 }
3063
3064 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3065
3066 static int vcpu_stat_get(void *_offset, u64 *val)
3067 {
3068         unsigned offset = (long)_offset;
3069         struct kvm *kvm;
3070         struct kvm_vcpu *vcpu;
3071         int i;
3072
3073         *val = 0;
3074         spin_lock(&kvm_lock);
3075         list_for_each_entry(kvm, &vm_list, vm_list)
3076                 kvm_for_each_vcpu(i, vcpu, kvm)
3077                         *val += *(u32 *)((void *)vcpu + offset);
3078
3079         spin_unlock(&kvm_lock);
3080         return 0;
3081 }
3082
3083 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3084
3085 static const struct file_operations *stat_fops[] = {
3086         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3087         [KVM_STAT_VM]   = &vm_stat_fops,
3088 };
3089
3090 static int kvm_init_debug(void)
3091 {
3092         int r = -EEXIST;
3093         struct kvm_stats_debugfs_item *p;
3094
3095         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3096         if (kvm_debugfs_dir == NULL)
3097                 goto out;
3098
3099         for (p = debugfs_entries; p->name; ++p) {
3100                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3101                                                 (void *)(long)p->offset,
3102                                                 stat_fops[p->kind]);
3103                 if (p->dentry == NULL)
3104                         goto out_dir;
3105         }
3106
3107         return 0;
3108
3109 out_dir:
3110         debugfs_remove_recursive(kvm_debugfs_dir);
3111 out:
3112         return r;
3113 }
3114
3115 static void kvm_exit_debug(void)
3116 {
3117         struct kvm_stats_debugfs_item *p;
3118
3119         for (p = debugfs_entries; p->name; ++p)
3120                 debugfs_remove(p->dentry);
3121         debugfs_remove(kvm_debugfs_dir);
3122 }
3123
3124 static int kvm_suspend(void)
3125 {
3126         if (kvm_usage_count)
3127                 hardware_disable_nolock(NULL);
3128         return 0;
3129 }
3130
3131 static void kvm_resume(void)
3132 {
3133         if (kvm_usage_count) {
3134                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3135                 hardware_enable_nolock(NULL);
3136         }
3137 }
3138
3139 static struct syscore_ops kvm_syscore_ops = {
3140         .suspend = kvm_suspend,
3141         .resume = kvm_resume,
3142 };
3143
3144 static inline
3145 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3146 {
3147         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3148 }
3149
3150 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3151 {
3152         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3153         if (vcpu->preempted)
3154                 vcpu->preempted = false;
3155
3156         kvm_arch_sched_in(vcpu, cpu);
3157
3158         kvm_arch_vcpu_load(vcpu, cpu);
3159 }
3160
3161 static void kvm_sched_out(struct preempt_notifier *pn,
3162                           struct task_struct *next)
3163 {
3164         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3165
3166         if (current->state == TASK_RUNNING)
3167                 vcpu->preempted = true;
3168         kvm_arch_vcpu_put(vcpu);
3169 }
3170
3171 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3172                   struct module *module)
3173 {
3174         int r;
3175         int cpu;
3176
3177         r = kvm_arch_init(opaque);
3178         if (r)
3179                 goto out_fail;
3180
3181         /*
3182          * kvm_arch_init makes sure there's at most one caller
3183          * for architectures that support multiple implementations,
3184          * like intel and amd on x86.
3185          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3186          * conflicts in case kvm is already setup for another implementation.
3187          */
3188         r = kvm_irqfd_init();
3189         if (r)
3190                 goto out_irqfd;
3191
3192         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3193                 r = -ENOMEM;
3194                 goto out_free_0;
3195         }
3196
3197         r = kvm_arch_hardware_setup();
3198         if (r < 0)
3199                 goto out_free_0a;
3200
3201         for_each_online_cpu(cpu) {
3202                 smp_call_function_single(cpu,
3203                                 kvm_arch_check_processor_compat,
3204                                 &r, 1);
3205                 if (r < 0)
3206                         goto out_free_1;
3207         }
3208
3209         r = register_cpu_notifier(&kvm_cpu_notifier);
3210         if (r)
3211                 goto out_free_2;
3212         register_reboot_notifier(&kvm_reboot_notifier);
3213
3214         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3215         if (!vcpu_align)
3216                 vcpu_align = __alignof__(struct kvm_vcpu);
3217         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3218                                            0, NULL);
3219         if (!kvm_vcpu_cache) {
3220                 r = -ENOMEM;
3221                 goto out_free_3;
3222         }
3223
3224         r = kvm_async_pf_init();
3225         if (r)
3226                 goto out_free;
3227
3228         kvm_chardev_ops.owner = module;
3229         kvm_vm_fops.owner = module;
3230         kvm_vcpu_fops.owner = module;
3231
3232         r = misc_register(&kvm_dev);
3233         if (r) {
3234                 printk(KERN_ERR "kvm: misc device register failed\n");
3235                 goto out_unreg;
3236         }
3237
3238         register_syscore_ops(&kvm_syscore_ops);
3239
3240         kvm_preempt_ops.sched_in = kvm_sched_in;
3241         kvm_preempt_ops.sched_out = kvm_sched_out;
3242
3243         r = kvm_init_debug();
3244         if (r) {
3245                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3246                 goto out_undebugfs;
3247         }
3248
3249         r = kvm_vfio_ops_init();
3250         WARN_ON(r);
3251
3252         return 0;
3253
3254 out_undebugfs:
3255         unregister_syscore_ops(&kvm_syscore_ops);
3256         misc_deregister(&kvm_dev);
3257 out_unreg:
3258         kvm_async_pf_deinit();
3259 out_free:
3260         kmem_cache_destroy(kvm_vcpu_cache);
3261 out_free_3:
3262         unregister_reboot_notifier(&kvm_reboot_notifier);
3263         unregister_cpu_notifier(&kvm_cpu_notifier);
3264 out_free_2:
3265 out_free_1:
3266         kvm_arch_hardware_unsetup();
3267 out_free_0a:
3268         free_cpumask_var(cpus_hardware_enabled);
3269 out_free_0:
3270         kvm_irqfd_exit();
3271 out_irqfd:
3272         kvm_arch_exit();
3273 out_fail:
3274         return r;
3275 }
3276 EXPORT_SYMBOL_GPL(kvm_init);
3277
3278 void kvm_exit(void)
3279 {
3280         kvm_exit_debug();
3281         misc_deregister(&kvm_dev);
3282         kmem_cache_destroy(kvm_vcpu_cache);
3283         kvm_async_pf_deinit();
3284         unregister_syscore_ops(&kvm_syscore_ops);
3285         unregister_reboot_notifier(&kvm_reboot_notifier);
3286         unregister_cpu_notifier(&kvm_cpu_notifier);
3287         on_each_cpu(hardware_disable_nolock, NULL, 1);
3288         kvm_arch_hardware_unsetup();
3289         kvm_arch_exit();
3290         kvm_irqfd_exit();
3291         free_cpumask_var(cpus_hardware_enabled);
3292         kvm_vfio_ops_exit();
3293 }
3294 EXPORT_SYMBOL_GPL(kvm_exit);