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 int kvm_get_user_page_io(struct task_struct *tsk, struct mm_struct *mm,
1132                          unsigned long addr, bool write_fault,
1133                          struct page **pagep)
1134 {
1135         int npages;
1136         int locked = 1;
1137         int flags = FOLL_TOUCH | FOLL_HWPOISON |
1138                     (pagep ? FOLL_GET : 0) |
1139                     (write_fault ? FOLL_WRITE : 0);
1140
1141         /*
1142          * If retrying the fault, we get here *not* having allowed the filemap
1143          * to wait on the page lock. We should now allow waiting on the IO with
1144          * the mmap semaphore released.
1145          */
1146         down_read(&mm->mmap_sem);
1147         npages = __get_user_pages(tsk, mm, addr, 1, flags, pagep, NULL,
1148                                   &locked);
1149         if (!locked) {
1150                 VM_BUG_ON(npages);
1151
1152                 if (!pagep)
1153                         return 0;
1154
1155                 /*
1156                  * The previous call has now waited on the IO. Now we can
1157                  * retry and complete. Pass TRIED to ensure we do not re
1158                  * schedule async IO (see e.g. filemap_fault).
1159                  */
1160                 down_read(&mm->mmap_sem);
1161                 npages = __get_user_pages(tsk, mm, addr, 1, flags | FOLL_TRIED,
1162                                           pagep, NULL, NULL);
1163         }
1164         up_read(&mm->mmap_sem);
1165         return npages;
1166 }
1167
1168 static inline int check_user_page_hwpoison(unsigned long addr)
1169 {
1170         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1171
1172         rc = __get_user_pages(current, current->mm, addr, 1,
1173                               flags, NULL, NULL, NULL);
1174         return rc == -EHWPOISON;
1175 }
1176
1177 /*
1178  * The atomic path to get the writable pfn which will be stored in @pfn,
1179  * true indicates success, otherwise false is returned.
1180  */
1181 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1182                             bool write_fault, bool *writable, pfn_t *pfn)
1183 {
1184         struct page *page[1];
1185         int npages;
1186
1187         if (!(async || atomic))
1188                 return false;
1189
1190         /*
1191          * Fast pin a writable pfn only if it is a write fault request
1192          * or the caller allows to map a writable pfn for a read fault
1193          * request.
1194          */
1195         if (!(write_fault || writable))
1196                 return false;
1197
1198         npages = __get_user_pages_fast(addr, 1, 1, page);
1199         if (npages == 1) {
1200                 *pfn = page_to_pfn(page[0]);
1201
1202                 if (writable)
1203                         *writable = true;
1204                 return true;
1205         }
1206
1207         return false;
1208 }
1209
1210 /*
1211  * The slow path to get the pfn of the specified host virtual address,
1212  * 1 indicates success, -errno is returned if error is detected.
1213  */
1214 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1215                            bool *writable, pfn_t *pfn)
1216 {
1217         struct page *page[1];
1218         int npages = 0;
1219
1220         might_sleep();
1221
1222         if (writable)
1223                 *writable = write_fault;
1224
1225         if (async) {
1226                 down_read(&current->mm->mmap_sem);
1227                 npages = get_user_page_nowait(current, current->mm,
1228                                               addr, write_fault, page);
1229                 up_read(&current->mm->mmap_sem);
1230         } else {
1231                 /*
1232                  * By now we have tried gup_fast, and possibly async_pf, and we
1233                  * are certainly not atomic. Time to retry the gup, allowing
1234                  * mmap semaphore to be relinquished in the case of IO.
1235                  */
1236                 npages = kvm_get_user_page_io(current, current->mm, addr,
1237                                               write_fault, page);
1238         }
1239         if (npages != 1)
1240                 return npages;
1241
1242         /* map read fault as writable if possible */
1243         if (unlikely(!write_fault) && writable) {
1244                 struct page *wpage[1];
1245
1246                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1247                 if (npages == 1) {
1248                         *writable = true;
1249                         put_page(page[0]);
1250                         page[0] = wpage[0];
1251                 }
1252
1253                 npages = 1;
1254         }
1255         *pfn = page_to_pfn(page[0]);
1256         return npages;
1257 }
1258
1259 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1260 {
1261         if (unlikely(!(vma->vm_flags & VM_READ)))
1262                 return false;
1263
1264         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1265                 return false;
1266
1267         return true;
1268 }
1269
1270 /*
1271  * Pin guest page in memory and return its pfn.
1272  * @addr: host virtual address which maps memory to the guest
1273  * @atomic: whether this function can sleep
1274  * @async: whether this function need to wait IO complete if the
1275  *         host page is not in the memory
1276  * @write_fault: whether we should get a writable host page
1277  * @writable: whether it allows to map a writable host page for !@write_fault
1278  *
1279  * The function will map a writable host page for these two cases:
1280  * 1): @write_fault = true
1281  * 2): @write_fault = false && @writable, @writable will tell the caller
1282  *     whether the mapping is writable.
1283  */
1284 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1285                         bool write_fault, bool *writable)
1286 {
1287         struct vm_area_struct *vma;
1288         pfn_t pfn = 0;
1289         int npages;
1290
1291         /* we can do it either atomically or asynchronously, not both */
1292         BUG_ON(atomic && async);
1293
1294         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1295                 return pfn;
1296
1297         if (atomic)
1298                 return KVM_PFN_ERR_FAULT;
1299
1300         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1301         if (npages == 1)
1302                 return pfn;
1303
1304         down_read(&current->mm->mmap_sem);
1305         if (npages == -EHWPOISON ||
1306               (!async && check_user_page_hwpoison(addr))) {
1307                 pfn = KVM_PFN_ERR_HWPOISON;
1308                 goto exit;
1309         }
1310
1311         vma = find_vma_intersection(current->mm, addr, addr + 1);
1312
1313         if (vma == NULL)
1314                 pfn = KVM_PFN_ERR_FAULT;
1315         else if ((vma->vm_flags & VM_PFNMAP)) {
1316                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1317                         vma->vm_pgoff;
1318                 BUG_ON(!kvm_is_reserved_pfn(pfn));
1319         } else {
1320                 if (async && vma_is_valid(vma, write_fault))
1321                         *async = true;
1322                 pfn = KVM_PFN_ERR_FAULT;
1323         }
1324 exit:
1325         up_read(&current->mm->mmap_sem);
1326         return pfn;
1327 }
1328
1329 static pfn_t
1330 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1331                      bool *async, bool write_fault, bool *writable)
1332 {
1333         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1334
1335         if (addr == KVM_HVA_ERR_RO_BAD)
1336                 return KVM_PFN_ERR_RO_FAULT;
1337
1338         if (kvm_is_error_hva(addr))
1339                 return KVM_PFN_NOSLOT;
1340
1341         /* Do not map writable pfn in the readonly memslot. */
1342         if (writable && memslot_is_readonly(slot)) {
1343                 *writable = false;
1344                 writable = NULL;
1345         }
1346
1347         return hva_to_pfn(addr, atomic, async, write_fault,
1348                           writable);
1349 }
1350
1351 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1352                           bool write_fault, bool *writable)
1353 {
1354         struct kvm_memory_slot *slot;
1355
1356         if (async)
1357                 *async = false;
1358
1359         slot = gfn_to_memslot(kvm, gfn);
1360
1361         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1362                                     writable);
1363 }
1364
1365 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1366 {
1367         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1368 }
1369 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1370
1371 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1372                        bool write_fault, bool *writable)
1373 {
1374         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1375 }
1376 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1377
1378 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1379 {
1380         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1381 }
1382 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1383
1384 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1385                       bool *writable)
1386 {
1387         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1388 }
1389 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1390
1391 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1392 {
1393         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1394 }
1395
1396 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1397 {
1398         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1399 }
1400 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1401
1402 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1403                                                                   int nr_pages)
1404 {
1405         unsigned long addr;
1406         gfn_t entry;
1407
1408         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1409         if (kvm_is_error_hva(addr))
1410                 return -1;
1411
1412         if (entry < nr_pages)
1413                 return 0;
1414
1415         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1416 }
1417 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1418
1419 static struct page *kvm_pfn_to_page(pfn_t pfn)
1420 {
1421         if (is_error_noslot_pfn(pfn))
1422                 return KVM_ERR_PTR_BAD_PAGE;
1423
1424         if (kvm_is_reserved_pfn(pfn)) {
1425                 WARN_ON(1);
1426                 return KVM_ERR_PTR_BAD_PAGE;
1427         }
1428
1429         return pfn_to_page(pfn);
1430 }
1431
1432 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1433 {
1434         pfn_t pfn;
1435
1436         pfn = gfn_to_pfn(kvm, gfn);
1437
1438         return kvm_pfn_to_page(pfn);
1439 }
1440
1441 EXPORT_SYMBOL_GPL(gfn_to_page);
1442
1443 void kvm_release_page_clean(struct page *page)
1444 {
1445         WARN_ON(is_error_page(page));
1446
1447         kvm_release_pfn_clean(page_to_pfn(page));
1448 }
1449 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1450
1451 void kvm_release_pfn_clean(pfn_t pfn)
1452 {
1453         if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1454                 put_page(pfn_to_page(pfn));
1455 }
1456 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1457
1458 void kvm_release_page_dirty(struct page *page)
1459 {
1460         WARN_ON(is_error_page(page));
1461
1462         kvm_release_pfn_dirty(page_to_pfn(page));
1463 }
1464 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1465
1466 static void kvm_release_pfn_dirty(pfn_t pfn)
1467 {
1468         kvm_set_pfn_dirty(pfn);
1469         kvm_release_pfn_clean(pfn);
1470 }
1471
1472 void kvm_set_pfn_dirty(pfn_t pfn)
1473 {
1474         if (!kvm_is_reserved_pfn(pfn)) {
1475                 struct page *page = pfn_to_page(pfn);
1476                 if (!PageReserved(page))
1477                         SetPageDirty(page);
1478         }
1479 }
1480 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1481
1482 void kvm_set_pfn_accessed(pfn_t pfn)
1483 {
1484         if (!kvm_is_reserved_pfn(pfn))
1485                 mark_page_accessed(pfn_to_page(pfn));
1486 }
1487 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1488
1489 void kvm_get_pfn(pfn_t pfn)
1490 {
1491         if (!kvm_is_reserved_pfn(pfn))
1492                 get_page(pfn_to_page(pfn));
1493 }
1494 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1495
1496 static int next_segment(unsigned long len, int offset)
1497 {
1498         if (len > PAGE_SIZE - offset)
1499                 return PAGE_SIZE - offset;
1500         else
1501                 return len;
1502 }
1503
1504 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1505                         int len)
1506 {
1507         int r;
1508         unsigned long addr;
1509
1510         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1511         if (kvm_is_error_hva(addr))
1512                 return -EFAULT;
1513         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1514         if (r)
1515                 return -EFAULT;
1516         return 0;
1517 }
1518 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1519
1520 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1521 {
1522         gfn_t gfn = gpa >> PAGE_SHIFT;
1523         int seg;
1524         int offset = offset_in_page(gpa);
1525         int ret;
1526
1527         while ((seg = next_segment(len, offset)) != 0) {
1528                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1529                 if (ret < 0)
1530                         return ret;
1531                 offset = 0;
1532                 len -= seg;
1533                 data += seg;
1534                 ++gfn;
1535         }
1536         return 0;
1537 }
1538 EXPORT_SYMBOL_GPL(kvm_read_guest);
1539
1540 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1541                           unsigned long len)
1542 {
1543         int r;
1544         unsigned long addr;
1545         gfn_t gfn = gpa >> PAGE_SHIFT;
1546         int offset = offset_in_page(gpa);
1547
1548         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1549         if (kvm_is_error_hva(addr))
1550                 return -EFAULT;
1551         pagefault_disable();
1552         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1553         pagefault_enable();
1554         if (r)
1555                 return -EFAULT;
1556         return 0;
1557 }
1558 EXPORT_SYMBOL(kvm_read_guest_atomic);
1559
1560 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1561                          int offset, int len)
1562 {
1563         int r;
1564         unsigned long addr;
1565
1566         addr = gfn_to_hva(kvm, gfn);
1567         if (kvm_is_error_hva(addr))
1568                 return -EFAULT;
1569         r = __copy_to_user((void __user *)addr + offset, data, len);
1570         if (r)
1571                 return -EFAULT;
1572         mark_page_dirty(kvm, gfn);
1573         return 0;
1574 }
1575 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1576
1577 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1578                     unsigned long len)
1579 {
1580         gfn_t gfn = gpa >> PAGE_SHIFT;
1581         int seg;
1582         int offset = offset_in_page(gpa);
1583         int ret;
1584
1585         while ((seg = next_segment(len, offset)) != 0) {
1586                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1587                 if (ret < 0)
1588                         return ret;
1589                 offset = 0;
1590                 len -= seg;
1591                 data += seg;
1592                 ++gfn;
1593         }
1594         return 0;
1595 }
1596
1597 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1598                               gpa_t gpa, unsigned long len)
1599 {
1600         struct kvm_memslots *slots = kvm_memslots(kvm);
1601         int offset = offset_in_page(gpa);
1602         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1603         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1604         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1605         gfn_t nr_pages_avail;
1606
1607         ghc->gpa = gpa;
1608         ghc->generation = slots->generation;
1609         ghc->len = len;
1610         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1611         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1612         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1613                 ghc->hva += offset;
1614         } else {
1615                 /*
1616                  * If the requested region crosses two memslots, we still
1617                  * verify that the entire region is valid here.
1618                  */
1619                 while (start_gfn <= end_gfn) {
1620                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1621                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1622                                                    &nr_pages_avail);
1623                         if (kvm_is_error_hva(ghc->hva))
1624                                 return -EFAULT;
1625                         start_gfn += nr_pages_avail;
1626                 }
1627                 /* Use the slow path for cross page reads and writes. */
1628                 ghc->memslot = NULL;
1629         }
1630         return 0;
1631 }
1632 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1633
1634 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1635                            void *data, unsigned long len)
1636 {
1637         struct kvm_memslots *slots = kvm_memslots(kvm);
1638         int r;
1639
1640         BUG_ON(len > ghc->len);
1641
1642         if (slots->generation != ghc->generation)
1643                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1644
1645         if (unlikely(!ghc->memslot))
1646                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1647
1648         if (kvm_is_error_hva(ghc->hva))
1649                 return -EFAULT;
1650
1651         r = __copy_to_user((void __user *)ghc->hva, data, len);
1652         if (r)
1653                 return -EFAULT;
1654         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1655
1656         return 0;
1657 }
1658 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1659
1660 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1661                            void *data, unsigned long len)
1662 {
1663         struct kvm_memslots *slots = kvm_memslots(kvm);
1664         int r;
1665
1666         BUG_ON(len > ghc->len);
1667
1668         if (slots->generation != ghc->generation)
1669                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1670
1671         if (unlikely(!ghc->memslot))
1672                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1673
1674         if (kvm_is_error_hva(ghc->hva))
1675                 return -EFAULT;
1676
1677         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1678         if (r)
1679                 return -EFAULT;
1680
1681         return 0;
1682 }
1683 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1684
1685 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1686 {
1687         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1688
1689         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1690 }
1691 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1692
1693 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1694 {
1695         gfn_t gfn = gpa >> PAGE_SHIFT;
1696         int seg;
1697         int offset = offset_in_page(gpa);
1698         int ret;
1699
1700         while ((seg = next_segment(len, offset)) != 0) {
1701                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1702                 if (ret < 0)
1703                         return ret;
1704                 offset = 0;
1705                 len -= seg;
1706                 ++gfn;
1707         }
1708         return 0;
1709 }
1710 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1711
1712 static void mark_page_dirty_in_slot(struct kvm *kvm,
1713                                     struct kvm_memory_slot *memslot,
1714                                     gfn_t gfn)
1715 {
1716         if (memslot && memslot->dirty_bitmap) {
1717                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1718
1719                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1720         }
1721 }
1722
1723 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1724 {
1725         struct kvm_memory_slot *memslot;
1726
1727         memslot = gfn_to_memslot(kvm, gfn);
1728         mark_page_dirty_in_slot(kvm, memslot, gfn);
1729 }
1730 EXPORT_SYMBOL_GPL(mark_page_dirty);
1731
1732 /*
1733  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1734  */
1735 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1736 {
1737         DEFINE_WAIT(wait);
1738
1739         for (;;) {
1740                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1741
1742                 if (kvm_arch_vcpu_runnable(vcpu)) {
1743                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1744                         break;
1745                 }
1746                 if (kvm_cpu_has_pending_timer(vcpu))
1747                         break;
1748                 if (signal_pending(current))
1749                         break;
1750
1751                 schedule();
1752         }
1753
1754         finish_wait(&vcpu->wq, &wait);
1755 }
1756 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1757
1758 #ifndef CONFIG_S390
1759 /*
1760  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1761  */
1762 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1763 {
1764         int me;
1765         int cpu = vcpu->cpu;
1766         wait_queue_head_t *wqp;
1767
1768         wqp = kvm_arch_vcpu_wq(vcpu);
1769         if (waitqueue_active(wqp)) {
1770                 wake_up_interruptible(wqp);
1771                 ++vcpu->stat.halt_wakeup;
1772         }
1773
1774         me = get_cpu();
1775         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1776                 if (kvm_arch_vcpu_should_kick(vcpu))
1777                         smp_send_reschedule(cpu);
1778         put_cpu();
1779 }
1780 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1781 #endif /* !CONFIG_S390 */
1782
1783 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1784 {
1785         struct pid *pid;
1786         struct task_struct *task = NULL;
1787         int ret = 0;
1788
1789         rcu_read_lock();
1790         pid = rcu_dereference(target->pid);
1791         if (pid)
1792                 task = get_pid_task(pid, PIDTYPE_PID);
1793         rcu_read_unlock();
1794         if (!task)
1795                 return ret;
1796         ret = yield_to(task, 1);
1797         put_task_struct(task);
1798
1799         return ret;
1800 }
1801 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1802
1803 /*
1804  * Helper that checks whether a VCPU is eligible for directed yield.
1805  * Most eligible candidate to yield is decided by following heuristics:
1806  *
1807  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1808  *  (preempted lock holder), indicated by @in_spin_loop.
1809  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1810  *
1811  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1812  *  chance last time (mostly it has become eligible now since we have probably
1813  *  yielded to lockholder in last iteration. This is done by toggling
1814  *  @dy_eligible each time a VCPU checked for eligibility.)
1815  *
1816  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1817  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1818  *  burning. Giving priority for a potential lock-holder increases lock
1819  *  progress.
1820  *
1821  *  Since algorithm is based on heuristics, accessing another VCPU data without
1822  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1823  *  and continue with next VCPU and so on.
1824  */
1825 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1826 {
1827 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1828         bool eligible;
1829
1830         eligible = !vcpu->spin_loop.in_spin_loop ||
1831                     vcpu->spin_loop.dy_eligible;
1832
1833         if (vcpu->spin_loop.in_spin_loop)
1834                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1835
1836         return eligible;
1837 #else
1838         return true;
1839 #endif
1840 }
1841
1842 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1843 {
1844         struct kvm *kvm = me->kvm;
1845         struct kvm_vcpu *vcpu;
1846         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1847         int yielded = 0;
1848         int try = 3;
1849         int pass;
1850         int i;
1851
1852         kvm_vcpu_set_in_spin_loop(me, true);
1853         /*
1854          * We boost the priority of a VCPU that is runnable but not
1855          * currently running, because it got preempted by something
1856          * else and called schedule in __vcpu_run.  Hopefully that
1857          * VCPU is holding the lock that we need and will release it.
1858          * We approximate round-robin by starting at the last boosted VCPU.
1859          */
1860         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1861                 kvm_for_each_vcpu(i, vcpu, kvm) {
1862                         if (!pass && i <= last_boosted_vcpu) {
1863                                 i = last_boosted_vcpu;
1864                                 continue;
1865                         } else if (pass && i > last_boosted_vcpu)
1866                                 break;
1867                         if (!ACCESS_ONCE(vcpu->preempted))
1868                                 continue;
1869                         if (vcpu == me)
1870                                 continue;
1871                         if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1872                                 continue;
1873                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1874                                 continue;
1875
1876                         yielded = kvm_vcpu_yield_to(vcpu);
1877                         if (yielded > 0) {
1878                                 kvm->last_boosted_vcpu = i;
1879                                 break;
1880                         } else if (yielded < 0) {
1881                                 try--;
1882                                 if (!try)
1883                                         break;
1884                         }
1885                 }
1886         }
1887         kvm_vcpu_set_in_spin_loop(me, false);
1888
1889         /* Ensure vcpu is not eligible during next spinloop */
1890         kvm_vcpu_set_dy_eligible(me, false);
1891 }
1892 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1893
1894 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1895 {
1896         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1897         struct page *page;
1898
1899         if (vmf->pgoff == 0)
1900                 page = virt_to_page(vcpu->run);
1901 #ifdef CONFIG_X86
1902         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1903                 page = virt_to_page(vcpu->arch.pio_data);
1904 #endif
1905 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1906         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1907                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1908 #endif
1909         else
1910                 return kvm_arch_vcpu_fault(vcpu, vmf);
1911         get_page(page);
1912         vmf->page = page;
1913         return 0;
1914 }
1915
1916 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1917         .fault = kvm_vcpu_fault,
1918 };
1919
1920 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1921 {
1922         vma->vm_ops = &kvm_vcpu_vm_ops;
1923         return 0;
1924 }
1925
1926 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1927 {
1928         struct kvm_vcpu *vcpu = filp->private_data;
1929
1930         kvm_put_kvm(vcpu->kvm);
1931         return 0;
1932 }
1933
1934 static struct file_operations kvm_vcpu_fops = {
1935         .release        = kvm_vcpu_release,
1936         .unlocked_ioctl = kvm_vcpu_ioctl,
1937 #ifdef CONFIG_COMPAT
1938         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1939 #endif
1940         .mmap           = kvm_vcpu_mmap,
1941         .llseek         = noop_llseek,
1942 };
1943
1944 /*
1945  * Allocates an inode for the vcpu.
1946  */
1947 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1948 {
1949         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1950 }
1951
1952 /*
1953  * Creates some virtual cpus.  Good luck creating more than one.
1954  */
1955 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1956 {
1957         int r;
1958         struct kvm_vcpu *vcpu, *v;
1959
1960         if (id >= KVM_MAX_VCPUS)
1961                 return -EINVAL;
1962
1963         vcpu = kvm_arch_vcpu_create(kvm, id);
1964         if (IS_ERR(vcpu))
1965                 return PTR_ERR(vcpu);
1966
1967         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1968
1969         r = kvm_arch_vcpu_setup(vcpu);
1970         if (r)
1971                 goto vcpu_destroy;
1972
1973         mutex_lock(&kvm->lock);
1974         if (!kvm_vcpu_compatible(vcpu)) {
1975                 r = -EINVAL;
1976                 goto unlock_vcpu_destroy;
1977         }
1978         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1979                 r = -EINVAL;
1980                 goto unlock_vcpu_destroy;
1981         }
1982
1983         kvm_for_each_vcpu(r, v, kvm)
1984                 if (v->vcpu_id == id) {
1985                         r = -EEXIST;
1986                         goto unlock_vcpu_destroy;
1987                 }
1988
1989         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1990
1991         /* Now it's all set up, let userspace reach it */
1992         kvm_get_kvm(kvm);
1993         r = create_vcpu_fd(vcpu);
1994         if (r < 0) {
1995                 kvm_put_kvm(kvm);
1996                 goto unlock_vcpu_destroy;
1997         }
1998
1999         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2000         smp_wmb();
2001         atomic_inc(&kvm->online_vcpus);
2002
2003         mutex_unlock(&kvm->lock);
2004         kvm_arch_vcpu_postcreate(vcpu);
2005         return r;
2006
2007 unlock_vcpu_destroy:
2008         mutex_unlock(&kvm->lock);
2009 vcpu_destroy:
2010         kvm_arch_vcpu_destroy(vcpu);
2011         return r;
2012 }
2013
2014 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2015 {
2016         if (sigset) {
2017                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2018                 vcpu->sigset_active = 1;
2019                 vcpu->sigset = *sigset;
2020         } else
2021                 vcpu->sigset_active = 0;
2022         return 0;
2023 }
2024
2025 static long kvm_vcpu_ioctl(struct file *filp,
2026                            unsigned int ioctl, unsigned long arg)
2027 {
2028         struct kvm_vcpu *vcpu = filp->private_data;
2029         void __user *argp = (void __user *)arg;
2030         int r;
2031         struct kvm_fpu *fpu = NULL;
2032         struct kvm_sregs *kvm_sregs = NULL;
2033
2034         if (vcpu->kvm->mm != current->mm)
2035                 return -EIO;
2036
2037         if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2038                 return -EINVAL;
2039
2040 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2041         /*
2042          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2043          * so vcpu_load() would break it.
2044          */
2045         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2046                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2047 #endif
2048
2049
2050         r = vcpu_load(vcpu);
2051         if (r)
2052                 return r;
2053         switch (ioctl) {
2054         case KVM_RUN:
2055                 r = -EINVAL;
2056                 if (arg)
2057                         goto out;
2058                 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2059                         /* The thread running this VCPU changed. */
2060                         struct pid *oldpid = vcpu->pid;
2061                         struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2062                         rcu_assign_pointer(vcpu->pid, newpid);
2063                         if (oldpid)
2064                                 synchronize_rcu();
2065                         put_pid(oldpid);
2066                 }
2067                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2068                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2069                 break;
2070         case KVM_GET_REGS: {
2071                 struct kvm_regs *kvm_regs;
2072
2073                 r = -ENOMEM;
2074                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2075                 if (!kvm_regs)
2076                         goto out;
2077                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2078                 if (r)
2079                         goto out_free1;
2080                 r = -EFAULT;
2081                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2082                         goto out_free1;
2083                 r = 0;
2084 out_free1:
2085                 kfree(kvm_regs);
2086                 break;
2087         }
2088         case KVM_SET_REGS: {
2089                 struct kvm_regs *kvm_regs;
2090
2091                 r = -ENOMEM;
2092                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2093                 if (IS_ERR(kvm_regs)) {
2094                         r = PTR_ERR(kvm_regs);
2095                         goto out;
2096                 }
2097                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2098                 kfree(kvm_regs);
2099                 break;
2100         }
2101         case KVM_GET_SREGS: {
2102                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2103                 r = -ENOMEM;
2104                 if (!kvm_sregs)
2105                         goto out;
2106                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2107                 if (r)
2108                         goto out;
2109                 r = -EFAULT;
2110                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2111                         goto out;
2112                 r = 0;
2113                 break;
2114         }
2115         case KVM_SET_SREGS: {
2116                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2117                 if (IS_ERR(kvm_sregs)) {
2118                         r = PTR_ERR(kvm_sregs);
2119                         kvm_sregs = NULL;
2120                         goto out;
2121                 }
2122                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2123                 break;
2124         }
2125         case KVM_GET_MP_STATE: {
2126                 struct kvm_mp_state mp_state;
2127
2128                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2129                 if (r)
2130                         goto out;
2131                 r = -EFAULT;
2132                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2133                         goto out;
2134                 r = 0;
2135                 break;
2136         }
2137         case KVM_SET_MP_STATE: {
2138                 struct kvm_mp_state mp_state;
2139
2140                 r = -EFAULT;
2141                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2142                         goto out;
2143                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2144                 break;
2145         }
2146         case KVM_TRANSLATE: {
2147                 struct kvm_translation tr;
2148
2149                 r = -EFAULT;
2150                 if (copy_from_user(&tr, argp, sizeof tr))
2151                         goto out;
2152                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2153                 if (r)
2154                         goto out;
2155                 r = -EFAULT;
2156                 if (copy_to_user(argp, &tr, sizeof tr))
2157                         goto out;
2158                 r = 0;
2159                 break;
2160         }
2161         case KVM_SET_GUEST_DEBUG: {
2162                 struct kvm_guest_debug dbg;
2163
2164                 r = -EFAULT;
2165                 if (copy_from_user(&dbg, argp, sizeof dbg))
2166                         goto out;
2167                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2168                 break;
2169         }
2170         case KVM_SET_SIGNAL_MASK: {
2171                 struct kvm_signal_mask __user *sigmask_arg = argp;
2172                 struct kvm_signal_mask kvm_sigmask;
2173                 sigset_t sigset, *p;
2174
2175                 p = NULL;
2176                 if (argp) {
2177                         r = -EFAULT;
2178                         if (copy_from_user(&kvm_sigmask, argp,
2179                                            sizeof kvm_sigmask))
2180                                 goto out;
2181                         r = -EINVAL;
2182                         if (kvm_sigmask.len != sizeof sigset)
2183                                 goto out;
2184                         r = -EFAULT;
2185                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2186                                            sizeof sigset))
2187                                 goto out;
2188                         p = &sigset;
2189                 }
2190                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2191                 break;
2192         }
2193         case KVM_GET_FPU: {
2194                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2195                 r = -ENOMEM;
2196                 if (!fpu)
2197                         goto out;
2198                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2199                 if (r)
2200                         goto out;
2201                 r = -EFAULT;
2202                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2203                         goto out;
2204                 r = 0;
2205                 break;
2206         }
2207         case KVM_SET_FPU: {
2208                 fpu = memdup_user(argp, sizeof(*fpu));
2209                 if (IS_ERR(fpu)) {
2210                         r = PTR_ERR(fpu);
2211                         fpu = NULL;
2212                         goto out;
2213                 }
2214                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2215                 break;
2216         }
2217         default:
2218                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2219         }
2220 out:
2221         vcpu_put(vcpu);
2222         kfree(fpu);
2223         kfree(kvm_sregs);
2224         return r;
2225 }
2226
2227 #ifdef CONFIG_COMPAT
2228 static long kvm_vcpu_compat_ioctl(struct file *filp,
2229                                   unsigned int ioctl, unsigned long arg)
2230 {
2231         struct kvm_vcpu *vcpu = filp->private_data;
2232         void __user *argp = compat_ptr(arg);
2233         int r;
2234
2235         if (vcpu->kvm->mm != current->mm)
2236                 return -EIO;
2237
2238         switch (ioctl) {
2239         case KVM_SET_SIGNAL_MASK: {
2240                 struct kvm_signal_mask __user *sigmask_arg = argp;
2241                 struct kvm_signal_mask kvm_sigmask;
2242                 compat_sigset_t csigset;
2243                 sigset_t sigset;
2244
2245                 if (argp) {
2246                         r = -EFAULT;
2247                         if (copy_from_user(&kvm_sigmask, argp,
2248                                            sizeof kvm_sigmask))
2249                                 goto out;
2250                         r = -EINVAL;
2251                         if (kvm_sigmask.len != sizeof csigset)
2252                                 goto out;
2253                         r = -EFAULT;
2254                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2255                                            sizeof csigset))
2256                                 goto out;
2257                         sigset_from_compat(&sigset, &csigset);
2258                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2259                 } else
2260                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2261                 break;
2262         }
2263         default:
2264                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2265         }
2266
2267 out:
2268         return r;
2269 }
2270 #endif
2271
2272 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2273                                  int (*accessor)(struct kvm_device *dev,
2274                                                  struct kvm_device_attr *attr),
2275                                  unsigned long arg)
2276 {
2277         struct kvm_device_attr attr;
2278
2279         if (!accessor)
2280                 return -EPERM;
2281
2282         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2283                 return -EFAULT;
2284
2285         return accessor(dev, &attr);
2286 }
2287
2288 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2289                              unsigned long arg)
2290 {
2291         struct kvm_device *dev = filp->private_data;
2292
2293         switch (ioctl) {
2294         case KVM_SET_DEVICE_ATTR:
2295                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2296         case KVM_GET_DEVICE_ATTR:
2297                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2298         case KVM_HAS_DEVICE_ATTR:
2299                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2300         default:
2301                 if (dev->ops->ioctl)
2302                         return dev->ops->ioctl(dev, ioctl, arg);
2303
2304                 return -ENOTTY;
2305         }
2306 }
2307
2308 static int kvm_device_release(struct inode *inode, struct file *filp)
2309 {
2310         struct kvm_device *dev = filp->private_data;
2311         struct kvm *kvm = dev->kvm;
2312
2313         kvm_put_kvm(kvm);
2314         return 0;
2315 }
2316
2317 static const struct file_operations kvm_device_fops = {
2318         .unlocked_ioctl = kvm_device_ioctl,
2319 #ifdef CONFIG_COMPAT
2320         .compat_ioctl = kvm_device_ioctl,
2321 #endif
2322         .release = kvm_device_release,
2323 };
2324
2325 struct kvm_device *kvm_device_from_filp(struct file *filp)
2326 {
2327         if (filp->f_op != &kvm_device_fops)
2328                 return NULL;
2329
2330         return filp->private_data;
2331 }
2332
2333 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2334 #ifdef CONFIG_KVM_MPIC
2335         [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
2336         [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
2337 #endif
2338
2339 #ifdef CONFIG_KVM_XICS
2340         [KVM_DEV_TYPE_XICS]             = &kvm_xics_ops,
2341 #endif
2342 };
2343
2344 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2345 {
2346         if (type >= ARRAY_SIZE(kvm_device_ops_table))
2347                 return -ENOSPC;
2348
2349         if (kvm_device_ops_table[type] != NULL)
2350                 return -EEXIST;
2351
2352         kvm_device_ops_table[type] = ops;
2353         return 0;
2354 }
2355
2356 void kvm_unregister_device_ops(u32 type)
2357 {
2358         if (kvm_device_ops_table[type] != NULL)
2359                 kvm_device_ops_table[type] = NULL;
2360 }
2361
2362 static int kvm_ioctl_create_device(struct kvm *kvm,
2363                                    struct kvm_create_device *cd)
2364 {
2365         struct kvm_device_ops *ops = NULL;
2366         struct kvm_device *dev;
2367         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2368         int ret;
2369
2370         if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2371                 return -ENODEV;
2372
2373         ops = kvm_device_ops_table[cd->type];
2374         if (ops == NULL)
2375                 return -ENODEV;
2376
2377         if (test)
2378                 return 0;
2379
2380         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2381         if (!dev)
2382                 return -ENOMEM;
2383
2384         dev->ops = ops;
2385         dev->kvm = kvm;
2386
2387         ret = ops->create(dev, cd->type);
2388         if (ret < 0) {
2389                 kfree(dev);
2390                 return ret;
2391         }
2392
2393         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2394         if (ret < 0) {
2395                 ops->destroy(dev);
2396                 return ret;
2397         }
2398
2399         list_add(&dev->vm_node, &kvm->devices);
2400         kvm_get_kvm(kvm);
2401         cd->fd = ret;
2402         return 0;
2403 }
2404
2405 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2406 {
2407         switch (arg) {
2408         case KVM_CAP_USER_MEMORY:
2409         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2410         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2411 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2412         case KVM_CAP_SET_BOOT_CPU_ID:
2413 #endif
2414         case KVM_CAP_INTERNAL_ERROR_DATA:
2415 #ifdef CONFIG_HAVE_KVM_MSI
2416         case KVM_CAP_SIGNAL_MSI:
2417 #endif
2418 #ifdef CONFIG_HAVE_KVM_IRQFD
2419         case KVM_CAP_IRQFD_RESAMPLE:
2420 #endif
2421         case KVM_CAP_CHECK_EXTENSION_VM:
2422                 return 1;
2423 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2424         case KVM_CAP_IRQ_ROUTING:
2425                 return KVM_MAX_IRQ_ROUTES;
2426 #endif
2427         default:
2428                 break;
2429         }
2430         return kvm_vm_ioctl_check_extension(kvm, arg);
2431 }
2432
2433 static long kvm_vm_ioctl(struct file *filp,
2434                            unsigned int ioctl, unsigned long arg)
2435 {
2436         struct kvm *kvm = filp->private_data;
2437         void __user *argp = (void __user *)arg;
2438         int r;
2439
2440         if (kvm->mm != current->mm)
2441                 return -EIO;
2442         switch (ioctl) {
2443         case KVM_CREATE_VCPU:
2444                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2445                 break;
2446         case KVM_SET_USER_MEMORY_REGION: {
2447                 struct kvm_userspace_memory_region kvm_userspace_mem;
2448
2449                 r = -EFAULT;
2450                 if (copy_from_user(&kvm_userspace_mem, argp,
2451                                                 sizeof kvm_userspace_mem))
2452                         goto out;
2453
2454                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2455                 break;
2456         }
2457         case KVM_GET_DIRTY_LOG: {
2458                 struct kvm_dirty_log log;
2459
2460                 r = -EFAULT;
2461                 if (copy_from_user(&log, argp, sizeof log))
2462                         goto out;
2463                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2464                 break;
2465         }
2466 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2467         case KVM_REGISTER_COALESCED_MMIO: {
2468                 struct kvm_coalesced_mmio_zone zone;
2469                 r = -EFAULT;
2470                 if (copy_from_user(&zone, argp, sizeof zone))
2471                         goto out;
2472                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2473                 break;
2474         }
2475         case KVM_UNREGISTER_COALESCED_MMIO: {
2476                 struct kvm_coalesced_mmio_zone zone;
2477                 r = -EFAULT;
2478                 if (copy_from_user(&zone, argp, sizeof zone))
2479                         goto out;
2480                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2481                 break;
2482         }
2483 #endif
2484         case KVM_IRQFD: {
2485                 struct kvm_irqfd data;
2486
2487                 r = -EFAULT;
2488                 if (copy_from_user(&data, argp, sizeof data))
2489                         goto out;
2490                 r = kvm_irqfd(kvm, &data);
2491                 break;
2492         }
2493         case KVM_IOEVENTFD: {
2494                 struct kvm_ioeventfd data;
2495
2496                 r = -EFAULT;
2497                 if (copy_from_user(&data, argp, sizeof data))
2498                         goto out;
2499                 r = kvm_ioeventfd(kvm, &data);
2500                 break;
2501         }
2502 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2503         case KVM_SET_BOOT_CPU_ID:
2504                 r = 0;
2505                 mutex_lock(&kvm->lock);
2506                 if (atomic_read(&kvm->online_vcpus) != 0)
2507                         r = -EBUSY;
2508                 else
2509                         kvm->bsp_vcpu_id = arg;
2510                 mutex_unlock(&kvm->lock);
2511                 break;
2512 #endif
2513 #ifdef CONFIG_HAVE_KVM_MSI
2514         case KVM_SIGNAL_MSI: {
2515                 struct kvm_msi msi;
2516
2517                 r = -EFAULT;
2518                 if (copy_from_user(&msi, argp, sizeof msi))
2519                         goto out;
2520                 r = kvm_send_userspace_msi(kvm, &msi);
2521                 break;
2522         }
2523 #endif
2524 #ifdef __KVM_HAVE_IRQ_LINE
2525         case KVM_IRQ_LINE_STATUS:
2526         case KVM_IRQ_LINE: {
2527                 struct kvm_irq_level irq_event;
2528
2529                 r = -EFAULT;
2530                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2531                         goto out;
2532
2533                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2534                                         ioctl == KVM_IRQ_LINE_STATUS);
2535                 if (r)
2536                         goto out;
2537
2538                 r = -EFAULT;
2539                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2540                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2541                                 goto out;
2542                 }
2543
2544                 r = 0;
2545                 break;
2546         }
2547 #endif
2548 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2549         case KVM_SET_GSI_ROUTING: {
2550                 struct kvm_irq_routing routing;
2551                 struct kvm_irq_routing __user *urouting;
2552                 struct kvm_irq_routing_entry *entries;
2553
2554                 r = -EFAULT;
2555                 if (copy_from_user(&routing, argp, sizeof(routing)))
2556                         goto out;
2557                 r = -EINVAL;
2558                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2559                         goto out;
2560                 if (routing.flags)
2561                         goto out;
2562                 r = -ENOMEM;
2563                 entries = vmalloc(routing.nr * sizeof(*entries));
2564                 if (!entries)
2565                         goto out;
2566                 r = -EFAULT;
2567                 urouting = argp;
2568                 if (copy_from_user(entries, urouting->entries,
2569                                    routing.nr * sizeof(*entries)))
2570                         goto out_free_irq_routing;
2571                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2572                                         routing.flags);
2573         out_free_irq_routing:
2574                 vfree(entries);
2575                 break;
2576         }
2577 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2578         case KVM_CREATE_DEVICE: {
2579                 struct kvm_create_device cd;
2580
2581                 r = -EFAULT;
2582                 if (copy_from_user(&cd, argp, sizeof(cd)))
2583                         goto out;
2584
2585                 r = kvm_ioctl_create_device(kvm, &cd);
2586                 if (r)
2587                         goto out;
2588
2589                 r = -EFAULT;
2590                 if (copy_to_user(argp, &cd, sizeof(cd)))
2591                         goto out;
2592
2593                 r = 0;
2594                 break;
2595         }
2596         case KVM_CHECK_EXTENSION:
2597                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2598                 break;
2599         default:
2600                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2601         }
2602 out:
2603         return r;
2604 }
2605
2606 #ifdef CONFIG_COMPAT
2607 struct compat_kvm_dirty_log {
2608         __u32 slot;
2609         __u32 padding1;
2610         union {
2611                 compat_uptr_t dirty_bitmap; /* one bit per page */
2612                 __u64 padding2;
2613         };
2614 };
2615
2616 static long kvm_vm_compat_ioctl(struct file *filp,
2617                            unsigned int ioctl, unsigned long arg)
2618 {
2619         struct kvm *kvm = filp->private_data;
2620         int r;
2621
2622         if (kvm->mm != current->mm)
2623                 return -EIO;
2624         switch (ioctl) {
2625         case KVM_GET_DIRTY_LOG: {
2626                 struct compat_kvm_dirty_log compat_log;
2627                 struct kvm_dirty_log log;
2628
2629                 r = -EFAULT;
2630                 if (copy_from_user(&compat_log, (void __user *)arg,
2631                                    sizeof(compat_log)))
2632                         goto out;
2633                 log.slot         = compat_log.slot;
2634                 log.padding1     = compat_log.padding1;
2635                 log.padding2     = compat_log.padding2;
2636                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2637
2638                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2639                 break;
2640         }
2641         default:
2642                 r = kvm_vm_ioctl(filp, ioctl, arg);
2643         }
2644
2645 out:
2646         return r;
2647 }
2648 #endif
2649
2650 static struct file_operations kvm_vm_fops = {
2651         .release        = kvm_vm_release,
2652         .unlocked_ioctl = kvm_vm_ioctl,
2653 #ifdef CONFIG_COMPAT
2654         .compat_ioctl   = kvm_vm_compat_ioctl,
2655 #endif
2656         .llseek         = noop_llseek,
2657 };
2658
2659 static int kvm_dev_ioctl_create_vm(unsigned long type)
2660 {
2661         int r;
2662         struct kvm *kvm;
2663
2664         kvm = kvm_create_vm(type);
2665         if (IS_ERR(kvm))
2666                 return PTR_ERR(kvm);
2667 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2668         r = kvm_coalesced_mmio_init(kvm);
2669         if (r < 0) {
2670                 kvm_put_kvm(kvm);
2671                 return r;
2672         }
2673 #endif
2674         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2675         if (r < 0)
2676                 kvm_put_kvm(kvm);
2677
2678         return r;
2679 }
2680
2681 static long kvm_dev_ioctl(struct file *filp,
2682                           unsigned int ioctl, unsigned long arg)
2683 {
2684         long r = -EINVAL;
2685
2686         switch (ioctl) {
2687         case KVM_GET_API_VERSION:
2688                 if (arg)
2689                         goto out;
2690                 r = KVM_API_VERSION;
2691                 break;
2692         case KVM_CREATE_VM:
2693                 r = kvm_dev_ioctl_create_vm(arg);
2694                 break;
2695         case KVM_CHECK_EXTENSION:
2696                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2697                 break;
2698         case KVM_GET_VCPU_MMAP_SIZE:
2699                 if (arg)
2700                         goto out;
2701                 r = PAGE_SIZE;     /* struct kvm_run */
2702 #ifdef CONFIG_X86
2703                 r += PAGE_SIZE;    /* pio data page */
2704 #endif
2705 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2706                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2707 #endif
2708                 break;
2709         case KVM_TRACE_ENABLE:
2710         case KVM_TRACE_PAUSE:
2711         case KVM_TRACE_DISABLE:
2712                 r = -EOPNOTSUPP;
2713                 break;
2714         default:
2715                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2716         }
2717 out:
2718         return r;
2719 }
2720
2721 static struct file_operations kvm_chardev_ops = {
2722         .unlocked_ioctl = kvm_dev_ioctl,
2723         .compat_ioctl   = kvm_dev_ioctl,
2724         .llseek         = noop_llseek,
2725 };
2726
2727 static struct miscdevice kvm_dev = {
2728         KVM_MINOR,
2729         "kvm",
2730         &kvm_chardev_ops,
2731 };
2732
2733 static void hardware_enable_nolock(void *junk)
2734 {
2735         int cpu = raw_smp_processor_id();
2736         int r;
2737
2738         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2739                 return;
2740
2741         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2742
2743         r = kvm_arch_hardware_enable();
2744
2745         if (r) {
2746                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2747                 atomic_inc(&hardware_enable_failed);
2748                 printk(KERN_INFO "kvm: enabling virtualization on "
2749                                  "CPU%d failed\n", cpu);
2750         }
2751 }
2752
2753 static void hardware_enable(void)
2754 {
2755         raw_spin_lock(&kvm_count_lock);
2756         if (kvm_usage_count)
2757                 hardware_enable_nolock(NULL);
2758         raw_spin_unlock(&kvm_count_lock);
2759 }
2760
2761 static void hardware_disable_nolock(void *junk)
2762 {
2763         int cpu = raw_smp_processor_id();
2764
2765         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2766                 return;
2767         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2768         kvm_arch_hardware_disable();
2769 }
2770
2771 static void hardware_disable(void)
2772 {
2773         raw_spin_lock(&kvm_count_lock);
2774         if (kvm_usage_count)
2775                 hardware_disable_nolock(NULL);
2776         raw_spin_unlock(&kvm_count_lock);
2777 }
2778
2779 static void hardware_disable_all_nolock(void)
2780 {
2781         BUG_ON(!kvm_usage_count);
2782
2783         kvm_usage_count--;
2784         if (!kvm_usage_count)
2785                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2786 }
2787
2788 static void hardware_disable_all(void)
2789 {
2790         raw_spin_lock(&kvm_count_lock);
2791         hardware_disable_all_nolock();
2792         raw_spin_unlock(&kvm_count_lock);
2793 }
2794
2795 static int hardware_enable_all(void)
2796 {
2797         int r = 0;
2798
2799         raw_spin_lock(&kvm_count_lock);
2800
2801         kvm_usage_count++;
2802         if (kvm_usage_count == 1) {
2803                 atomic_set(&hardware_enable_failed, 0);
2804                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2805
2806                 if (atomic_read(&hardware_enable_failed)) {
2807                         hardware_disable_all_nolock();
2808                         r = -EBUSY;
2809                 }
2810         }
2811
2812         raw_spin_unlock(&kvm_count_lock);
2813
2814         return r;
2815 }
2816
2817 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2818                            void *v)
2819 {
2820         int cpu = (long)v;
2821
2822         val &= ~CPU_TASKS_FROZEN;
2823         switch (val) {
2824         case CPU_DYING:
2825                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2826                        cpu);
2827                 hardware_disable();
2828                 break;
2829         case CPU_STARTING:
2830                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2831                        cpu);
2832                 hardware_enable();
2833                 break;
2834         }
2835         return NOTIFY_OK;
2836 }
2837
2838 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2839                       void *v)
2840 {
2841         /*
2842          * Some (well, at least mine) BIOSes hang on reboot if
2843          * in vmx root mode.
2844          *
2845          * And Intel TXT required VMX off for all cpu when system shutdown.
2846          */
2847         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2848         kvm_rebooting = true;
2849         on_each_cpu(hardware_disable_nolock, NULL, 1);
2850         return NOTIFY_OK;
2851 }
2852
2853 static struct notifier_block kvm_reboot_notifier = {
2854         .notifier_call = kvm_reboot,
2855         .priority = 0,
2856 };
2857
2858 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2859 {
2860         int i;
2861
2862         for (i = 0; i < bus->dev_count; i++) {
2863                 struct kvm_io_device *pos = bus->range[i].dev;
2864
2865                 kvm_iodevice_destructor(pos);
2866         }
2867         kfree(bus);
2868 }
2869
2870 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2871                                  const struct kvm_io_range *r2)
2872 {
2873         if (r1->addr < r2->addr)
2874                 return -1;
2875         if (r1->addr + r1->len > r2->addr + r2->len)
2876                 return 1;
2877         return 0;
2878 }
2879
2880 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2881 {
2882         return kvm_io_bus_cmp(p1, p2);
2883 }
2884
2885 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2886                           gpa_t addr, int len)
2887 {
2888         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2889                 .addr = addr,
2890                 .len = len,
2891                 .dev = dev,
2892         };
2893
2894         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2895                 kvm_io_bus_sort_cmp, NULL);
2896
2897         return 0;
2898 }
2899
2900 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2901                              gpa_t addr, int len)
2902 {
2903         struct kvm_io_range *range, key;
2904         int off;
2905
2906         key = (struct kvm_io_range) {
2907                 .addr = addr,
2908                 .len = len,
2909         };
2910
2911         range = bsearch(&key, bus->range, bus->dev_count,
2912                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2913         if (range == NULL)
2914                 return -ENOENT;
2915
2916         off = range - bus->range;
2917
2918         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2919                 off--;
2920
2921         return off;
2922 }
2923
2924 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2925                               struct kvm_io_range *range, const void *val)
2926 {
2927         int idx;
2928
2929         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2930         if (idx < 0)
2931                 return -EOPNOTSUPP;
2932
2933         while (idx < bus->dev_count &&
2934                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2935                 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2936                                         range->len, val))
2937                         return idx;
2938                 idx++;
2939         }
2940
2941         return -EOPNOTSUPP;
2942 }
2943
2944 /* kvm_io_bus_write - called under kvm->slots_lock */
2945 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2946                      int len, const void *val)
2947 {
2948         struct kvm_io_bus *bus;
2949         struct kvm_io_range range;
2950         int r;
2951
2952         range = (struct kvm_io_range) {
2953                 .addr = addr,
2954                 .len = len,
2955         };
2956
2957         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2958         r = __kvm_io_bus_write(bus, &range, val);
2959         return r < 0 ? r : 0;
2960 }
2961
2962 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2963 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2964                             int len, const void *val, long cookie)
2965 {
2966         struct kvm_io_bus *bus;
2967         struct kvm_io_range range;
2968
2969         range = (struct kvm_io_range) {
2970                 .addr = addr,
2971                 .len = len,
2972         };
2973
2974         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2975
2976         /* First try the device referenced by cookie. */
2977         if ((cookie >= 0) && (cookie < bus->dev_count) &&
2978             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2979                 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2980                                         val))
2981                         return cookie;
2982
2983         /*
2984          * cookie contained garbage; fall back to search and return the
2985          * correct cookie value.
2986          */
2987         return __kvm_io_bus_write(bus, &range, val);
2988 }
2989
2990 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2991                              void *val)
2992 {
2993         int idx;
2994
2995         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2996         if (idx < 0)
2997                 return -EOPNOTSUPP;
2998
2999         while (idx < bus->dev_count &&
3000                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3001                 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
3002                                        range->len, val))
3003                         return idx;
3004                 idx++;
3005         }
3006
3007         return -EOPNOTSUPP;
3008 }
3009 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3010
3011 /* kvm_io_bus_read - called under kvm->slots_lock */
3012 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3013                     int len, void *val)
3014 {
3015         struct kvm_io_bus *bus;
3016         struct kvm_io_range range;
3017         int r;
3018
3019         range = (struct kvm_io_range) {
3020                 .addr = addr,
3021                 .len = len,
3022         };
3023
3024         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3025         r = __kvm_io_bus_read(bus, &range, val);
3026         return r < 0 ? r : 0;
3027 }
3028
3029
3030 /* Caller must hold slots_lock. */
3031 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3032                             int len, struct kvm_io_device *dev)
3033 {
3034         struct kvm_io_bus *new_bus, *bus;
3035
3036         bus = kvm->buses[bus_idx];
3037         /* exclude ioeventfd which is limited by maximum fd */
3038         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3039                 return -ENOSPC;
3040
3041         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3042                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3043         if (!new_bus)
3044                 return -ENOMEM;
3045         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3046                sizeof(struct kvm_io_range)));
3047         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3048         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3049         synchronize_srcu_expedited(&kvm->srcu);
3050         kfree(bus);
3051
3052         return 0;
3053 }
3054
3055 /* Caller must hold slots_lock. */
3056 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3057                               struct kvm_io_device *dev)
3058 {
3059         int i, r;
3060         struct kvm_io_bus *new_bus, *bus;
3061
3062         bus = kvm->buses[bus_idx];
3063         r = -ENOENT;
3064         for (i = 0; i < bus->dev_count; i++)
3065                 if (bus->range[i].dev == dev) {
3066                         r = 0;
3067                         break;
3068                 }
3069
3070         if (r)
3071                 return r;
3072
3073         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3074                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3075         if (!new_bus)
3076                 return -ENOMEM;
3077
3078         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3079         new_bus->dev_count--;
3080         memcpy(new_bus->range + i, bus->range + i + 1,
3081                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3082
3083         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3084         synchronize_srcu_expedited(&kvm->srcu);
3085         kfree(bus);
3086         return r;
3087 }
3088
3089 static struct notifier_block kvm_cpu_notifier = {
3090         .notifier_call = kvm_cpu_hotplug,
3091 };
3092
3093 static int vm_stat_get(void *_offset, u64 *val)
3094 {
3095         unsigned offset = (long)_offset;
3096         struct kvm *kvm;
3097
3098         *val = 0;
3099         spin_lock(&kvm_lock);
3100         list_for_each_entry(kvm, &vm_list, vm_list)
3101                 *val += *(u32 *)((void *)kvm + offset);
3102         spin_unlock(&kvm_lock);
3103         return 0;
3104 }
3105
3106 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3107
3108 static int vcpu_stat_get(void *_offset, u64 *val)
3109 {
3110         unsigned offset = (long)_offset;
3111         struct kvm *kvm;
3112         struct kvm_vcpu *vcpu;
3113         int i;
3114
3115         *val = 0;
3116         spin_lock(&kvm_lock);
3117         list_for_each_entry(kvm, &vm_list, vm_list)
3118                 kvm_for_each_vcpu(i, vcpu, kvm)
3119                         *val += *(u32 *)((void *)vcpu + offset);
3120
3121         spin_unlock(&kvm_lock);
3122         return 0;
3123 }
3124
3125 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3126
3127 static const struct file_operations *stat_fops[] = {
3128         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3129         [KVM_STAT_VM]   = &vm_stat_fops,
3130 };
3131
3132 static int kvm_init_debug(void)
3133 {
3134         int r = -EEXIST;
3135         struct kvm_stats_debugfs_item *p;
3136
3137         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3138         if (kvm_debugfs_dir == NULL)
3139                 goto out;
3140
3141         for (p = debugfs_entries; p->name; ++p) {
3142                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3143                                                 (void *)(long)p->offset,
3144                                                 stat_fops[p->kind]);
3145                 if (p->dentry == NULL)
3146                         goto out_dir;
3147         }
3148
3149         return 0;
3150
3151 out_dir:
3152         debugfs_remove_recursive(kvm_debugfs_dir);
3153 out:
3154         return r;
3155 }
3156
3157 static void kvm_exit_debug(void)
3158 {
3159         struct kvm_stats_debugfs_item *p;
3160
3161         for (p = debugfs_entries; p->name; ++p)
3162                 debugfs_remove(p->dentry);
3163         debugfs_remove(kvm_debugfs_dir);
3164 }
3165
3166 static int kvm_suspend(void)
3167 {
3168         if (kvm_usage_count)
3169                 hardware_disable_nolock(NULL);
3170         return 0;
3171 }
3172
3173 static void kvm_resume(void)
3174 {
3175         if (kvm_usage_count) {
3176                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3177                 hardware_enable_nolock(NULL);
3178         }
3179 }
3180
3181 static struct syscore_ops kvm_syscore_ops = {
3182         .suspend = kvm_suspend,
3183         .resume = kvm_resume,
3184 };
3185
3186 static inline
3187 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3188 {
3189         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3190 }
3191
3192 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3193 {
3194         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3195         if (vcpu->preempted)
3196                 vcpu->preempted = false;
3197
3198         kvm_arch_sched_in(vcpu, cpu);
3199
3200         kvm_arch_vcpu_load(vcpu, cpu);
3201 }
3202
3203 static void kvm_sched_out(struct preempt_notifier *pn,
3204                           struct task_struct *next)
3205 {
3206         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3207
3208         if (current->state == TASK_RUNNING)
3209                 vcpu->preempted = true;
3210         kvm_arch_vcpu_put(vcpu);
3211 }
3212
3213 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3214                   struct module *module)
3215 {
3216         int r;
3217         int cpu;
3218
3219         r = kvm_arch_init(opaque);
3220         if (r)
3221                 goto out_fail;
3222
3223         /*
3224          * kvm_arch_init makes sure there's at most one caller
3225          * for architectures that support multiple implementations,
3226          * like intel and amd on x86.
3227          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3228          * conflicts in case kvm is already setup for another implementation.
3229          */
3230         r = kvm_irqfd_init();
3231         if (r)
3232                 goto out_irqfd;
3233
3234         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3235                 r = -ENOMEM;
3236                 goto out_free_0;
3237         }
3238
3239         r = kvm_arch_hardware_setup();
3240         if (r < 0)
3241                 goto out_free_0a;
3242
3243         for_each_online_cpu(cpu) {
3244                 smp_call_function_single(cpu,
3245                                 kvm_arch_check_processor_compat,
3246                                 &r, 1);
3247                 if (r < 0)
3248                         goto out_free_1;
3249         }
3250
3251         r = register_cpu_notifier(&kvm_cpu_notifier);
3252         if (r)
3253                 goto out_free_2;
3254         register_reboot_notifier(&kvm_reboot_notifier);
3255
3256         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3257         if (!vcpu_align)
3258                 vcpu_align = __alignof__(struct kvm_vcpu);
3259         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3260                                            0, NULL);
3261         if (!kvm_vcpu_cache) {
3262                 r = -ENOMEM;
3263                 goto out_free_3;
3264         }
3265
3266         r = kvm_async_pf_init();
3267         if (r)
3268                 goto out_free;
3269
3270         kvm_chardev_ops.owner = module;
3271         kvm_vm_fops.owner = module;
3272         kvm_vcpu_fops.owner = module;
3273
3274         r = misc_register(&kvm_dev);
3275         if (r) {
3276                 printk(KERN_ERR "kvm: misc device register failed\n");
3277                 goto out_unreg;
3278         }
3279
3280         register_syscore_ops(&kvm_syscore_ops);
3281
3282         kvm_preempt_ops.sched_in = kvm_sched_in;
3283         kvm_preempt_ops.sched_out = kvm_sched_out;
3284
3285         r = kvm_init_debug();
3286         if (r) {
3287                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3288                 goto out_undebugfs;
3289         }
3290
3291         r = kvm_vfio_ops_init();
3292         WARN_ON(r);
3293
3294         return 0;
3295
3296 out_undebugfs:
3297         unregister_syscore_ops(&kvm_syscore_ops);
3298         misc_deregister(&kvm_dev);
3299 out_unreg:
3300         kvm_async_pf_deinit();
3301 out_free:
3302         kmem_cache_destroy(kvm_vcpu_cache);
3303 out_free_3:
3304         unregister_reboot_notifier(&kvm_reboot_notifier);
3305         unregister_cpu_notifier(&kvm_cpu_notifier);
3306 out_free_2:
3307 out_free_1:
3308         kvm_arch_hardware_unsetup();
3309 out_free_0a:
3310         free_cpumask_var(cpus_hardware_enabled);
3311 out_free_0:
3312         kvm_irqfd_exit();
3313 out_irqfd:
3314         kvm_arch_exit();
3315 out_fail:
3316         return r;
3317 }
3318 EXPORT_SYMBOL_GPL(kvm_init);
3319
3320 void kvm_exit(void)
3321 {
3322         kvm_exit_debug();
3323         misc_deregister(&kvm_dev);
3324         kmem_cache_destroy(kvm_vcpu_cache);
3325         kvm_async_pf_deinit();
3326         unregister_syscore_ops(&kvm_syscore_ops);
3327         unregister_reboot_notifier(&kvm_reboot_notifier);
3328         unregister_cpu_notifier(&kvm_cpu_notifier);
3329         on_each_cpu(hardware_disable_nolock, NULL, 1);
3330         kvm_arch_hardware_unsetup();
3331         kvm_arch_exit();
3332         kvm_irqfd_exit();
3333         free_cpumask_var(cpus_hardware_enabled);
3334         kvm_vfio_ops_exit();
3335 }
3336 EXPORT_SYMBOL_GPL(kvm_exit);