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