kernel/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 <kvm/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 static unsigned int halt_poll_ns;
  70 module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
  71
  72 /*
  73  * Ordering of locks:
  74  *
  75  *      kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  76  */
  77
  78 DEFINE_SPINLOCK(kvm_lock);
  79 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
  80 LIST_HEAD(vm_list);
  81
  82 static cpumask_var_t cpus_hardware_enabled;
  83 static int kvm_usage_count;
  84 static atomic_t hardware_enable_failed;
  85
  86 struct kmem_cache *kvm_vcpu_cache;
  87 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  88
  89 static __read_mostly struct preempt_ops kvm_preempt_ops;
  90
  91 struct dentry *kvm_debugfs_dir;
  92 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
  93
  94 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  95                            unsigned long arg);
  96 #ifdef CONFIG_KVM_COMPAT
  97 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  98                                   unsigned long arg);
  99 #endif
 100 static int hardware_enable_all(void);
 101 static void hardware_disable_all(void);
 102
 103 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
 104
 105 static void kvm_release_pfn_dirty(pfn_t pfn);
 106 static void mark_page_dirty_in_slot(struct kvm *kvm,
 107                                     struct kvm_memory_slot *memslot, gfn_t gfn);
 108
 109 __visible bool kvm_rebooting;
 110 EXPORT_SYMBOL_GPL(kvm_rebooting);
 111
 112 static bool largepages_enabled = true;
 113
 114 bool kvm_is_reserved_pfn(pfn_t pfn)
 115 {
 116         if (pfn_valid(pfn))
 117                 return PageReserved(pfn_to_page(pfn));
 118
 119         return true;
 120 }
 121
 122 /*
 123  * Switches to specified vcpu, until a matching vcpu_put()
 124  */
 125 int vcpu_load(struct kvm_vcpu *vcpu)
 126 {
 127         int cpu;
 128
 129         if (mutex_lock_killable(&vcpu->mutex))
 130                 return -EINTR;
 131         cpu = get_cpu();
 132         preempt_notifier_register(&vcpu->preempt_notifier);
 133         kvm_arch_vcpu_load(vcpu, cpu);
 134         put_cpu();
 135         return 0;
 136 }
 137
 138 void vcpu_put(struct kvm_vcpu *vcpu)
 139 {
 140         preempt_disable();
 141         kvm_arch_vcpu_put(vcpu);
 142         preempt_notifier_unregister(&vcpu->preempt_notifier);
 143         preempt_enable();
 144         mutex_unlock(&vcpu->mutex);
 145 }
 146
 147 static void ack_flush(void *_completed)
 148 {
 149 }
 150
 151 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
 152 {
 153         int i, cpu, me;
 154         cpumask_var_t cpus;
 155         bool called = true;
 156         struct kvm_vcpu *vcpu;
 157
 158         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 159
 160         me = get_cpu();
 161         kvm_for_each_vcpu(i, vcpu, kvm) {
 162                 kvm_make_request(req, vcpu);
 163                 cpu = vcpu->cpu;
 164
 165                 /* Set ->requests bit before we read ->mode */
 166                 smp_mb();
 167
 168                 if (cpus != NULL && cpu != -1 && cpu != me &&
 169                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 170                         cpumask_set_cpu(cpu, cpus);
 171         }
 172         if (unlikely(cpus == NULL))
 173                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 174         else if (!cpumask_empty(cpus))
 175                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 176         else
 177                 called = false;
 178         put_cpu();
 179         free_cpumask_var(cpus);
 180         return called;
 181 }
 182
 183 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
 184 void kvm_flush_remote_tlbs(struct kvm *kvm)
 185 {
 186         long dirty_count = kvm->tlbs_dirty;
 187
 188         smp_mb();
 189         if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 190                 ++kvm->stat.remote_tlb_flush;
 191         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 192 }
 193 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 194 #endif
 195
 196 void kvm_reload_remote_mmus(struct kvm *kvm)
 197 {
 198         kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 199 }
 200
 201 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 202 {
 203         kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 204 }
 205
 206 void kvm_make_scan_ioapic_request(struct kvm *kvm)
 207 {
 208         kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
 209 }
 210
 211 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 212 {
 213         struct page *page;
 214         int r;
 215
 216         mutex_init(&vcpu->mutex);
 217         vcpu->cpu = -1;
 218         vcpu->kvm = kvm;
 219         vcpu->vcpu_id = id;
 220         vcpu->pid = NULL;
 221         init_swait_head(&vcpu->wq);
 222         kvm_async_pf_vcpu_init(vcpu);
 223
 224         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 225         if (!page) {
 226                 r = -ENOMEM;
 227                 goto fail;
 228         }
 229         vcpu->run = page_address(page);
 230
 231         kvm_vcpu_set_in_spin_loop(vcpu, false);
 232         kvm_vcpu_set_dy_eligible(vcpu, false);
 233         vcpu->preempted = false;
 234
 235         r = kvm_arch_vcpu_init(vcpu);
 236         if (r < 0)
 237                 goto fail_free_run;
 238         return 0;
 239
 240 fail_free_run:
 241         free_page((unsigned long)vcpu->run);
 242 fail:
 243         return r;
 244 }
 245 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 246
 247 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 248 {
 249         put_pid(vcpu->pid);
 250         kvm_arch_vcpu_uninit(vcpu);
 251         free_page((unsigned long)vcpu->run);
 252 }
 253 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 254
 255 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 256 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 257 {
 258         return container_of(mn, struct kvm, mmu_notifier);
 259 }
 260
 261 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 262                                              struct mm_struct *mm,
 263                                              unsigned long address)
 264 {
 265         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 266         int need_tlb_flush, idx;
 267
 268         /*
 269          * When ->invalidate_page runs, the linux pte has been zapped
 270          * already but the page is still allocated until
 271          * ->invalidate_page returns. So if we increase the sequence
 272          * here the kvm page fault will notice if the spte can't be
 273          * established because the page is going to be freed. If
 274          * instead the kvm page fault establishes the spte before
 275          * ->invalidate_page runs, kvm_unmap_hva will release it
 276          * before returning.
 277          *
 278          * The sequence increase only need to be seen at spin_unlock
 279          * time, and not at spin_lock time.
 280          *
 281          * Increasing the sequence after the spin_unlock would be
 282          * unsafe because the kvm page fault could then establish the
 283          * pte after kvm_unmap_hva returned, without noticing the page
 284          * is going to be freed.
 285          */
 286         idx = srcu_read_lock(&kvm->srcu);
 287         spin_lock(&kvm->mmu_lock);
 288
 289         kvm->mmu_notifier_seq++;
 290         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 291         /* we've to flush the tlb before the pages can be freed */
 292         if (need_tlb_flush)
 293                 kvm_flush_remote_tlbs(kvm);
 294
 295         spin_unlock(&kvm->mmu_lock);
 296
 297         kvm_arch_mmu_notifier_invalidate_page(kvm, address);
 298
 299         srcu_read_unlock(&kvm->srcu, idx);
 300 }
 301
 302 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 303                                         struct mm_struct *mm,
 304                                         unsigned long address,
 305                                         pte_t pte)
 306 {
 307         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 308         int idx;
 309
 310         idx = srcu_read_lock(&kvm->srcu);
 311         spin_lock(&kvm->mmu_lock);
 312         kvm->mmu_notifier_seq++;
 313         kvm_set_spte_hva(kvm, address, pte);
 314         spin_unlock(&kvm->mmu_lock);
 315         srcu_read_unlock(&kvm->srcu, idx);
 316 }
 317
 318 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 319                                                     struct mm_struct *mm,
 320                                                     unsigned long start,
 321                                                     unsigned long end)
 322 {
 323         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 324         int need_tlb_flush = 0, idx;
 325
 326         idx = srcu_read_lock(&kvm->srcu);
 327         spin_lock(&kvm->mmu_lock);
 328         /*
 329          * The count increase must become visible at unlock time as no
 330          * spte can be established without taking the mmu_lock and
 331          * count is also read inside the mmu_lock critical section.
 332          */
 333         kvm->mmu_notifier_count++;
 334         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 335         need_tlb_flush |= kvm->tlbs_dirty;
 336         /* we've to flush the tlb before the pages can be freed */
 337         if (need_tlb_flush)
 338                 kvm_flush_remote_tlbs(kvm);
 339
 340         spin_unlock(&kvm->mmu_lock);
 341         srcu_read_unlock(&kvm->srcu, idx);
 342 }
 343
 344 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 345                                                   struct mm_struct *mm,
 346                                                   unsigned long start,
 347                                                   unsigned long end)
 348 {
 349         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 350
 351         spin_lock(&kvm->mmu_lock);
 352         /*
 353          * This sequence increase will notify the kvm page fault that
 354          * the page that is going to be mapped in the spte could have
 355          * been freed.
 356          */
 357         kvm->mmu_notifier_seq++;
 358         smp_wmb();
 359         /*
 360          * The above sequence increase must be visible before the
 361          * below count decrease, which is ensured by the smp_wmb above
 362          * in conjunction with the smp_rmb in mmu_notifier_retry().
 363          */
 364         kvm->mmu_notifier_count--;
 365         spin_unlock(&kvm->mmu_lock);
 366
 367         BUG_ON(kvm->mmu_notifier_count < 0);
 368 }
 369
 370 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 371                                               struct mm_struct *mm,
 372                                               unsigned long start,
 373                                               unsigned long end)
 374 {
 375         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 376         int young, idx;
 377
 378         idx = srcu_read_lock(&kvm->srcu);
 379         spin_lock(&kvm->mmu_lock);
 380
 381         young = kvm_age_hva(kvm, start, end);
 382         if (young)
 383                 kvm_flush_remote_tlbs(kvm);
 384
 385         spin_unlock(&kvm->mmu_lock);
 386         srcu_read_unlock(&kvm->srcu, idx);
 387
 388         return young;
 389 }
 390
 391 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 392                                        struct mm_struct *mm,
 393                                        unsigned long address)
 394 {
 395         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 396         int young, idx;
 397
 398         idx = srcu_read_lock(&kvm->srcu);
 399         spin_lock(&kvm->mmu_lock);
 400         young = kvm_test_age_hva(kvm, address);
 401         spin_unlock(&kvm->mmu_lock);
 402         srcu_read_unlock(&kvm->srcu, idx);
 403
 404         return young;
 405 }
 406
 407 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 408                                      struct mm_struct *mm)
 409 {
 410         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 411         int idx;
 412
 413         idx = srcu_read_lock(&kvm->srcu);
 414         kvm_arch_flush_shadow_all(kvm);
 415         srcu_read_unlock(&kvm->srcu, idx);
 416 }
 417
 418 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 419         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 420         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 421         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 422         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 423         .test_young             = kvm_mmu_notifier_test_young,
 424         .change_pte             = kvm_mmu_notifier_change_pte,
 425         .release                = kvm_mmu_notifier_release,
 426 };
 427
 428 static int kvm_init_mmu_notifier(struct kvm *kvm)
 429 {
 430         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 431         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 432 }
 433
 434 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 435
 436 static int kvm_init_mmu_notifier(struct kvm *kvm)
 437 {
 438         return 0;
 439 }
 440
 441 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 442
 443 static void kvm_init_memslots_id(struct kvm *kvm)
 444 {
 445         int i;
 446         struct kvm_memslots *slots = kvm->memslots;
 447
 448         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 449                 slots->id_to_index[i] = slots->memslots[i].id = i;
 450 }
 451
 452 static struct kvm *kvm_create_vm(unsigned long type)
 453 {
 454         int r, i;
 455         struct kvm *kvm = kvm_arch_alloc_vm();
 456
 457         if (!kvm)
 458                 return ERR_PTR(-ENOMEM);
 459
 460         r = kvm_arch_init_vm(kvm, type);
 461         if (r)
 462                 goto out_err_no_disable;
 463
 464         r = hardware_enable_all();
 465         if (r)
 466                 goto out_err_no_disable;
 467
 468 #ifdef CONFIG_HAVE_KVM_IRQFD
 469         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 470 #endif
 471
 472         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 473
 474         r = -ENOMEM;
 475         kvm->memslots = kvm_kvzalloc(sizeof(struct kvm_memslots));
 476         if (!kvm->memslots)
 477                 goto out_err_no_srcu;
 478
 479         /*
 480          * Init kvm generation close to the maximum to easily test the
 481          * code of handling generation number wrap-around.
 482          */
 483         kvm->memslots->generation = -150;
 484
 485         kvm_init_memslots_id(kvm);
 486         if (init_srcu_struct(&kvm->srcu))
 487                 goto out_err_no_srcu;
 488         if (init_srcu_struct(&kvm->irq_srcu))
 489                 goto out_err_no_irq_srcu;
 490         for (i = 0; i < KVM_NR_BUSES; i++) {
 491                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 492                                         GFP_KERNEL);
 493                 if (!kvm->buses[i])
 494                         goto out_err;
 495         }
 496
 497         spin_lock_init(&kvm->mmu_lock);
 498         kvm->mm = current->mm;
 499         atomic_inc(&kvm->mm->mm_count);
 500         kvm_eventfd_init(kvm);
 501         mutex_init(&kvm->lock);
 502         mutex_init(&kvm->irq_lock);
 503         mutex_init(&kvm->slots_lock);
 504         atomic_set(&kvm->users_count, 1);
 505         INIT_LIST_HEAD(&kvm->devices);
 506
 507         r = kvm_init_mmu_notifier(kvm);
 508         if (r)
 509                 goto out_err;
 510
 511         spin_lock(&kvm_lock);
 512         list_add(&kvm->vm_list, &vm_list);
 513         spin_unlock(&kvm_lock);
 514
 515         return kvm;
 516
 517 out_err:
 518         cleanup_srcu_struct(&kvm->irq_srcu);
 519 out_err_no_irq_srcu:
 520         cleanup_srcu_struct(&kvm->srcu);
 521 out_err_no_srcu:
 522         hardware_disable_all();
 523 out_err_no_disable:
 524         for (i = 0; i < KVM_NR_BUSES; i++)
 525                 kfree(kvm->buses[i]);
 526         kvfree(kvm->memslots);
 527         kvm_arch_free_vm(kvm);
 528         return ERR_PTR(r);
 529 }
 530
 531 /*
 532  * Avoid using vmalloc for a small buffer.
 533  * Should not be used when the size is statically known.
 534  */
 535 void *kvm_kvzalloc(unsigned long size)
 536 {
 537         if (size > PAGE_SIZE)
 538                 return vzalloc(size);
 539         else
 540                 return kzalloc(size, GFP_KERNEL);
 541 }
 542
 543 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 544 {
 545         if (!memslot->dirty_bitmap)
 546                 return;
 547
 548         kvfree(memslot->dirty_bitmap);
 549         memslot->dirty_bitmap = NULL;
 550 }
 551
 552 /*
 553  * Free any memory in @free but not in @dont.
 554  */
 555 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
 556                                   struct kvm_memory_slot *dont)
 557 {
 558         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 559                 kvm_destroy_dirty_bitmap(free);
 560
 561         kvm_arch_free_memslot(kvm, free, dont);
 562
 563         free->npages = 0;
 564 }
 565
 566 static void kvm_free_physmem(struct kvm *kvm)
 567 {
 568         struct kvm_memslots *slots = kvm->memslots;
 569         struct kvm_memory_slot *memslot;
 570
 571         kvm_for_each_memslot(memslot, slots)
 572                 kvm_free_physmem_slot(kvm, memslot, NULL);
 573
 574         kvfree(kvm->memslots);
 575 }
 576
 577 static void kvm_destroy_devices(struct kvm *kvm)
 578 {
 579         struct list_head *node, *tmp;
 580
 581         list_for_each_safe(node, tmp, &kvm->devices) {
 582                 struct kvm_device *dev =
 583                         list_entry(node, struct kvm_device, vm_node);
 584
 585                 list_del(node);
 586                 dev->ops->destroy(dev);
 587         }
 588 }
 589
 590 static void kvm_destroy_vm(struct kvm *kvm)
 591 {
 592         int i;
 593         struct mm_struct *mm = kvm->mm;
 594
 595         kvm_arch_sync_events(kvm);
 596         spin_lock(&kvm_lock);
 597         list_del(&kvm->vm_list);
 598         spin_unlock(&kvm_lock);
 599         kvm_free_irq_routing(kvm);
 600         for (i = 0; i < KVM_NR_BUSES; i++)
 601                 kvm_io_bus_destroy(kvm->buses[i]);
 602         kvm_coalesced_mmio_free(kvm);
 603 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 604         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 605 #else
 606         kvm_arch_flush_shadow_all(kvm);
 607 #endif
 608         kvm_arch_destroy_vm(kvm);
 609         kvm_destroy_devices(kvm);
 610         kvm_free_physmem(kvm);
 611         cleanup_srcu_struct(&kvm->irq_srcu);
 612         cleanup_srcu_struct(&kvm->srcu);
 613         kvm_arch_free_vm(kvm);
 614         hardware_disable_all();
 615         mmdrop(mm);
 616 }
 617
 618 void kvm_get_kvm(struct kvm *kvm)
 619 {
 620         atomic_inc(&kvm->users_count);
 621 }
 622 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 623
 624 void kvm_put_kvm(struct kvm *kvm)
 625 {
 626         if (atomic_dec_and_test(&kvm->users_count))
 627                 kvm_destroy_vm(kvm);
 628 }
 629 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 630
 631
 632 static int kvm_vm_release(struct inode *inode, struct file *filp)
 633 {
 634         struct kvm *kvm = filp->private_data;
 635
 636         kvm_irqfd_release(kvm);
 637
 638         kvm_put_kvm(kvm);
 639         return 0;
 640 }
 641
 642 /*
 643  * Allocation size is twice as large as the actual dirty bitmap size.
 644  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 645  */
 646 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 647 {
 648         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 649
 650         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 651         if (!memslot->dirty_bitmap)
 652                 return -ENOMEM;
 653
 654         return 0;
 655 }
 656
 657 /*
 658  * Insert memslot and re-sort memslots based on their GFN,
 659  * so binary search could be used to lookup GFN.
 660  * Sorting algorithm takes advantage of having initially
 661  * sorted array and known changed memslot position.
 662  */
 663 static void update_memslots(struct kvm_memslots *slots,
 664                             struct kvm_memory_slot *new)
 665 {
 666         int id = new->id;
 667         int i = slots->id_to_index[id];
 668         struct kvm_memory_slot *mslots = slots->memslots;
 669
 670         WARN_ON(mslots[i].id != id);
 671         if (!new->npages) {
 672                 WARN_ON(!mslots[i].npages);
 673                 new->base_gfn = 0;
 674                 new->flags = 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
 691         /*
 692          * The ">=" is needed when creating a slot with base_gfn == 0,
 693          * so that it moves before all those with base_gfn == npages == 0.
 694          *
 695          * On the other hand, if new->npages is zero, the above loop has
 696          * already left i pointing to the beginning of the empty part of
 697          * mslots, and the ">=" would move the hole backwards in this
 698          * case---which is wrong.  So skip the loop when deleting a slot.
 699          */
 700         if (new->npages) {
 701                 while (i > 0 &&
 702                        new->base_gfn >= mslots[i - 1].base_gfn) {
 703                         mslots[i] = mslots[i - 1];
 704                         slots->id_to_index[mslots[i].id] = i;
 705                         i--;
 706                 }
 707         } else
 708                 WARN_ON_ONCE(i != slots->used_slots);
 709
 710         mslots[i] = *new;
 711         slots->id_to_index[mslots[i].id] = i;
 712 }
 713
 714 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 715 {
 716         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 717
 718 #ifdef __KVM_HAVE_READONLY_MEM
 719         valid_flags |= KVM_MEM_READONLY;
 720 #endif
 721
 722         if (mem->flags & ~valid_flags)
 723                 return -EINVAL;
 724
 725         return 0;
 726 }
 727
 728 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 729                 struct kvm_memslots *slots)
 730 {
 731         struct kvm_memslots *old_memslots = kvm->memslots;
 732
 733         /*
 734          * Set the low bit in the generation, which disables SPTE caching
 735          * until the end of synchronize_srcu_expedited.
 736          */
 737         WARN_ON(old_memslots->generation & 1);
 738         slots->generation = old_memslots->generation + 1;
 739
 740         rcu_assign_pointer(kvm->memslots, slots);
 741         synchronize_srcu_expedited(&kvm->srcu);
 742
 743         /*
 744          * Increment the new memslot generation a second time. This prevents
 745          * vm exits that race with memslot updates from caching a memslot
 746          * generation that will (potentially) be valid forever.
 747          */
 748         slots->generation++;
 749
 750         kvm_arch_memslots_updated(kvm);
 751
 752         return old_memslots;
 753 }
 754
 755 /*
 756  * Allocate some memory and give it an address in the guest physical address
 757  * space.
 758  *
 759  * Discontiguous memory is allowed, mostly for framebuffers.
 760  *
 761  * Must be called holding kvm->slots_lock for write.
 762  */
 763 int __kvm_set_memory_region(struct kvm *kvm,
 764                             struct kvm_userspace_memory_region *mem)
 765 {
 766         int r;
 767         gfn_t base_gfn;
 768         unsigned long npages;
 769         struct kvm_memory_slot *slot;
 770         struct kvm_memory_slot old, new;
 771         struct kvm_memslots *slots = NULL, *old_memslots;
 772         enum kvm_mr_change change;
 773
 774         r = check_memory_region_flags(mem);
 775         if (r)
 776                 goto out;
 777
 778         r = -EINVAL;
 779         /* General sanity checks */
 780         if (mem->memory_size & (PAGE_SIZE - 1))
 781                 goto out;
 782         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 783                 goto out;
 784         /* We can read the guest memory with __xxx_user() later on. */
 785         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 786             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 787              !access_ok(VERIFY_WRITE,
 788                         (void __user *)(unsigned long)mem->userspace_addr,
 789                         mem->memory_size)))
 790                 goto out;
 791         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 792                 goto out;
 793         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 794                 goto out;
 795
 796         slot = id_to_memslot(kvm->memslots, mem->slot);
 797         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 798         npages = mem->memory_size >> PAGE_SHIFT;
 799
 800         if (npages > KVM_MEM_MAX_NR_PAGES)
 801                 goto out;
 802
 803         if (!npages)
 804                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 805
 806         new = old = *slot;
 807
 808         new.id = mem->slot;
 809         new.base_gfn = base_gfn;
 810         new.npages = npages;
 811         new.flags = mem->flags;
 812
 813         if (npages) {
 814                 if (!old.npages)
 815                         change = KVM_MR_CREATE;
 816                 else { /* Modify an existing slot. */
 817                         if ((mem->userspace_addr != old.userspace_addr) ||
 818                             (npages != old.npages) ||
 819                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 820                                 goto out;
 821
 822                         if (base_gfn != old.base_gfn)
 823                                 change = KVM_MR_MOVE;
 824                         else if (new.flags != old.flags)
 825                                 change = KVM_MR_FLAGS_ONLY;
 826                         else { /* Nothing to change. */
 827                                 r = 0;
 828                                 goto out;
 829                         }
 830                 }
 831         } else if (old.npages) {
 832                 change = KVM_MR_DELETE;
 833         } else /* Modify a non-existent slot: disallowed. */
 834                 goto out;
 835
 836         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 837                 /* Check for overlaps */
 838                 r = -EEXIST;
 839                 kvm_for_each_memslot(slot, kvm->memslots) {
 840                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 841                             (slot->id == mem->slot))
 842                                 continue;
 843                         if (!((base_gfn + npages <= slot->base_gfn) ||
 844                               (base_gfn >= slot->base_gfn + slot->npages)))
 845                                 goto out;
 846                 }
 847         }
 848
 849         /* Free page dirty bitmap if unneeded */
 850         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 851                 new.dirty_bitmap = NULL;
 852
 853         r = -ENOMEM;
 854         if (change == KVM_MR_CREATE) {
 855                 new.userspace_addr = mem->userspace_addr;
 856
 857                 if (kvm_arch_create_memslot(kvm, &new, npages))
 858                         goto out_free;
 859         }
 860
 861         /* Allocate page dirty bitmap if needed */
 862         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 863                 if (kvm_create_dirty_bitmap(&new) < 0)
 864                         goto out_free;
 865         }
 866
 867         slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
 868         if (!slots)
 869                 goto out_free;
 870         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 871
 872         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 873                 slot = id_to_memslot(slots, mem->slot);
 874                 slot->flags |= KVM_MEMSLOT_INVALID;
 875
 876                 old_memslots = install_new_memslots(kvm, slots);
 877
 878                 /* slot was deleted or moved, clear iommu mapping */
 879                 kvm_iommu_unmap_pages(kvm, &old);
 880                 /* From this point no new shadow pages pointing to a deleted,
 881                  * or moved, memslot will be created.
 882                  *
 883                  * validation of sp->gfn happens in:
 884                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 885                  *      - kvm_is_visible_gfn (mmu_check_roots)
 886                  */
 887                 kvm_arch_flush_shadow_memslot(kvm, slot);
 888
 889                 /*
 890                  * We can re-use the old_memslots from above, the only difference
 891                  * from the currently installed memslots is the invalid flag.  This
 892                  * will get overwritten by update_memslots anyway.
 893                  */
 894                 slots = old_memslots;
 895         }
 896
 897         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 898         if (r)
 899                 goto out_slots;
 900
 901         /* actual memory is freed via old in kvm_free_physmem_slot below */
 902         if (change == KVM_MR_DELETE) {
 903                 new.dirty_bitmap = NULL;
 904                 memset(&new.arch, 0, sizeof(new.arch));
 905         }
 906
 907         update_memslots(slots, &new);
 908         old_memslots = install_new_memslots(kvm, slots);
 909
 910         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 911
 912         kvm_free_physmem_slot(kvm, &old, &new);
 913         kvfree(old_memslots);
 914
 915         /*
 916          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 917          * un-mapped and re-mapped if their base changes.  Since base change
 918          * unmapping is handled above with slot deletion, mapping alone is
 919          * needed here.  Anything else the iommu might care about for existing
 920          * slots (size changes, userspace addr changes and read-only flag
 921          * changes) is disallowed above, so any other attribute changes getting
 922          * here can be skipped.
 923          */
 924         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 925                 r = kvm_iommu_map_pages(kvm, &new);
 926                 return r;
 927         }
 928
 929         return 0;
 930
 931 out_slots:
 932         kvfree(slots);
 933 out_free:
 934         kvm_free_physmem_slot(kvm, &new, &old);
 935 out:
 936         return r;
 937 }
 938 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 939
 940 int kvm_set_memory_region(struct kvm *kvm,
 941                           struct kvm_userspace_memory_region *mem)
 942 {
 943         int r;
 944
 945         mutex_lock(&kvm->slots_lock);
 946         r = __kvm_set_memory_region(kvm, mem);
 947         mutex_unlock(&kvm->slots_lock);
 948         return r;
 949 }
 950 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 951
 952 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 953                                           struct kvm_userspace_memory_region *mem)
 954 {
 955         if (mem->slot >= KVM_USER_MEM_SLOTS)
 956                 return -EINVAL;
 957         return kvm_set_memory_region(kvm, mem);
 958 }
 959
 960 int kvm_get_dirty_log(struct kvm *kvm,
 961                         struct kvm_dirty_log *log, int *is_dirty)
 962 {
 963         struct kvm_memory_slot *memslot;
 964         int r, i;
 965         unsigned long n;
 966         unsigned long any = 0;
 967
 968         r = -EINVAL;
 969         if (log->slot >= KVM_USER_MEM_SLOTS)
 970                 goto out;
 971
 972         memslot = id_to_memslot(kvm->memslots, log->slot);
 973         r = -ENOENT;
 974         if (!memslot->dirty_bitmap)
 975                 goto out;
 976
 977         n = kvm_dirty_bitmap_bytes(memslot);
 978
 979         for (i = 0; !any && i < n/sizeof(long); ++i)
 980                 any = memslot->dirty_bitmap[i];
 981
 982         r = -EFAULT;
 983         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 984                 goto out;
 985
 986         if (any)
 987                 *is_dirty = 1;
 988
 989         r = 0;
 990 out:
 991         return r;
 992 }
 993 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
 994
 995 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
 996 /**
 997  * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
 998  *      are dirty write protect them for next write.
 999  * @kvm:        pointer to kvm instance
1000  * @log:        slot id and address to which we copy the log
1001  * @is_dirty:   flag set if any page is dirty
1002  *
1003  * We need to keep it in mind that VCPU threads can write to the bitmap
1004  * concurrently. So, to avoid losing track of dirty pages we keep the
1005  * following order:
1006  *
1007  *    1. Take a snapshot of the bit and clear it if needed.
1008  *    2. Write protect the corresponding page.
1009  *    3. Copy the snapshot to the userspace.
1010  *    4. Upon return caller flushes TLB's if needed.
1011  *
1012  * Between 2 and 4, the guest may write to the page using the remaining TLB
1013  * entry.  This is not a problem because the page is reported dirty using
1014  * the snapshot taken before and step 4 ensures that writes done after
1015  * exiting to userspace will be logged for the next call.
1016  *
1017  */
1018 int kvm_get_dirty_log_protect(struct kvm *kvm,
1019                         struct kvm_dirty_log *log, bool *is_dirty)
1020 {
1021         struct kvm_memory_slot *memslot;
1022         int r, i;
1023         unsigned long n;
1024         unsigned long *dirty_bitmap;
1025         unsigned long *dirty_bitmap_buffer;
1026
1027         r = -EINVAL;
1028         if (log->slot >= KVM_USER_MEM_SLOTS)
1029                 goto out;
1030
1031         memslot = id_to_memslot(kvm->memslots, log->slot);
1032
1033         dirty_bitmap = memslot->dirty_bitmap;
1034         r = -ENOENT;
1035         if (!dirty_bitmap)
1036                 goto out;
1037
1038         n = kvm_dirty_bitmap_bytes(memslot);
1039
1040         dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1041         memset(dirty_bitmap_buffer, 0, n);
1042
1043         spin_lock(&kvm->mmu_lock);
1044         *is_dirty = false;
1045         for (i = 0; i < n / sizeof(long); i++) {
1046                 unsigned long mask;
1047                 gfn_t offset;
1048
1049                 if (!dirty_bitmap[i])
1050                         continue;
1051
1052                 *is_dirty = true;
1053
1054                 mask = xchg(&dirty_bitmap[i], 0);
1055                 dirty_bitmap_buffer[i] = mask;
1056
1057                 if (mask) {
1058                         offset = i * BITS_PER_LONG;
1059                         kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1060                                                                 offset, mask);
1061                 }
1062         }
1063
1064         spin_unlock(&kvm->mmu_lock);
1065
1066         r = -EFAULT;
1067         if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1068                 goto out;
1069
1070         r = 0;
1071 out:
1072         return r;
1073 }
1074 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1075 #endif
1076
1077 bool kvm_largepages_enabled(void)
1078 {
1079         return largepages_enabled;
1080 }
1081
1082 void kvm_disable_largepages(void)
1083 {
1084         largepages_enabled = false;
1085 }
1086 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1087
1088 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1089 {
1090         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1091 }
1092 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1093
1094 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1095 {
1096         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1097
1098         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1099               memslot->flags & KVM_MEMSLOT_INVALID)
1100                 return 0;
1101
1102         return 1;
1103 }
1104 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1105
1106 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1107 {
1108         struct vm_area_struct *vma;
1109         unsigned long addr, size;
1110
1111         size = PAGE_SIZE;
1112
1113         addr = gfn_to_hva(kvm, gfn);
1114         if (kvm_is_error_hva(addr))
1115                 return PAGE_SIZE;
1116
1117         down_read(&current->mm->mmap_sem);
1118         vma = find_vma(current->mm, addr);
1119         if (!vma)
1120                 goto out;
1121
1122         size = vma_kernel_pagesize(vma);
1123
1124 out:
1125         up_read(&current->mm->mmap_sem);
1126
1127         return size;
1128 }
1129
1130 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1131 {
1132         return slot->flags & KVM_MEM_READONLY;
1133 }
1134
1135 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1136                                        gfn_t *nr_pages, bool write)
1137 {
1138         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1139                 return KVM_HVA_ERR_BAD;
1140
1141         if (memslot_is_readonly(slot) && write)
1142                 return KVM_HVA_ERR_RO_BAD;
1143
1144         if (nr_pages)
1145                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1146
1147         return __gfn_to_hva_memslot(slot, gfn);
1148 }
1149
1150 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1151                                      gfn_t *nr_pages)
1152 {
1153         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1154 }
1155
1156 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1157                                         gfn_t gfn)
1158 {
1159         return gfn_to_hva_many(slot, gfn, NULL);
1160 }
1161 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1162
1163 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1164 {
1165         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1166 }
1167 EXPORT_SYMBOL_GPL(gfn_to_hva);
1168
1169 /*
1170  * If writable is set to false, the hva returned by this function is only
1171  * allowed to be read.
1172  */
1173 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1174                                       gfn_t gfn, bool *writable)
1175 {
1176         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1177
1178         if (!kvm_is_error_hva(hva) && writable)
1179                 *writable = !memslot_is_readonly(slot);
1180
1181         return hva;
1182 }
1183
1184 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1185 {
1186         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1187
1188         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1189 }
1190
1191 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1192         unsigned long start, int write, struct page **page)
1193 {
1194         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1195
1196         if (write)
1197                 flags |= FOLL_WRITE;
1198
1199         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1200 }
1201
1202 static inline int check_user_page_hwpoison(unsigned long addr)
1203 {
1204         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1205
1206         rc = __get_user_pages(current, current->mm, addr, 1,
1207                               flags, NULL, NULL, NULL);
1208         return rc == -EHWPOISON;
1209 }
1210
1211 /*
1212  * The atomic path to get the writable pfn which will be stored in @pfn,
1213  * true indicates success, otherwise false is returned.
1214  */
1215 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1216                             bool write_fault, bool *writable, pfn_t *pfn)
1217 {
1218         struct page *page[1];
1219         int npages;
1220
1221         if (!(async || atomic))
1222                 return false;
1223
1224         /*
1225          * Fast pin a writable pfn only if it is a write fault request
1226          * or the caller allows to map a writable pfn for a read fault
1227          * request.
1228          */
1229         if (!(write_fault || writable))
1230                 return false;
1231
1232         npages = __get_user_pages_fast(addr, 1, 1, page);
1233         if (npages == 1) {
1234                 *pfn = page_to_pfn(page[0]);
1235
1236                 if (writable)
1237                         *writable = true;
1238                 return true;
1239         }
1240
1241         return false;
1242 }
1243
1244 /*
1245  * The slow path to get the pfn of the specified host virtual address,
1246  * 1 indicates success, -errno is returned if error is detected.
1247  */
1248 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1249                            bool *writable, pfn_t *pfn)
1250 {
1251         struct page *page[1];
1252         int npages = 0;
1253
1254         might_sleep();
1255
1256         if (writable)
1257                 *writable = write_fault;
1258
1259         if (async) {
1260                 down_read(&current->mm->mmap_sem);
1261                 npages = get_user_page_nowait(current, current->mm,
1262                                               addr, write_fault, page);
1263                 up_read(&current->mm->mmap_sem);
1264         } else
1265                 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1266                                                    write_fault, 0, page,
1267                                                    FOLL_TOUCH|FOLL_HWPOISON);
1268         if (npages != 1)
1269                 return npages;
1270
1271         /* map read fault as writable if possible */
1272         if (unlikely(!write_fault) && writable) {
1273                 struct page *wpage[1];
1274
1275                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1276                 if (npages == 1) {
1277                         *writable = true;
1278                         put_page(page[0]);
1279                         page[0] = wpage[0];
1280                 }
1281
1282                 npages = 1;
1283         }
1284         *pfn = page_to_pfn(page[0]);
1285         return npages;
1286 }
1287
1288 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1289 {
1290         if (unlikely(!(vma->vm_flags & VM_READ)))
1291                 return false;
1292
1293         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1294                 return false;
1295
1296         return true;
1297 }
1298
1299 /*
1300  * Pin guest page in memory and return its pfn.
1301  * @addr: host virtual address which maps memory to the guest
1302  * @atomic: whether this function can sleep
1303  * @async: whether this function need to wait IO complete if the
1304  *         host page is not in the memory
1305  * @write_fault: whether we should get a writable host page
1306  * @writable: whether it allows to map a writable host page for !@write_fault
1307  *
1308  * The function will map a writable host page for these two cases:
1309  * 1): @write_fault = true
1310  * 2): @write_fault = false && @writable, @writable will tell the caller
1311  *     whether the mapping is writable.
1312  */
1313 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1314                         bool write_fault, bool *writable)
1315 {
1316         struct vm_area_struct *vma;
1317         pfn_t pfn = 0;
1318         int npages;
1319
1320         /* we can do it either atomically or asynchronously, not both */
1321         BUG_ON(atomic && async);
1322
1323         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1324                 return pfn;
1325
1326         if (atomic)
1327                 return KVM_PFN_ERR_FAULT;
1328
1329         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1330         if (npages == 1)
1331                 return pfn;
1332
1333         down_read(&current->mm->mmap_sem);
1334         if (npages == -EHWPOISON ||
1335               (!async && check_user_page_hwpoison(addr))) {
1336                 pfn = KVM_PFN_ERR_HWPOISON;
1337                 goto exit;
1338         }
1339
1340         vma = find_vma_intersection(current->mm, addr, addr + 1);
1341
1342         if (vma == NULL)
1343                 pfn = KVM_PFN_ERR_FAULT;
1344         else if ((vma->vm_flags & VM_PFNMAP)) {
1345                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1346                         vma->vm_pgoff;
1347                 BUG_ON(!kvm_is_reserved_pfn(pfn));
1348         } else {
1349                 if (async && vma_is_valid(vma, write_fault))
1350                         *async = true;
1351                 pfn = KVM_PFN_ERR_FAULT;
1352         }
1353 exit:
1354         up_read(&current->mm->mmap_sem);
1355         return pfn;
1356 }
1357
1358 static pfn_t
1359 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1360                      bool *async, bool write_fault, bool *writable)
1361 {
1362         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1363
1364         if (addr == KVM_HVA_ERR_RO_BAD)
1365                 return KVM_PFN_ERR_RO_FAULT;
1366
1367         if (kvm_is_error_hva(addr))
1368                 return KVM_PFN_NOSLOT;
1369
1370         /* Do not map writable pfn in the readonly memslot. */
1371         if (writable && memslot_is_readonly(slot)) {
1372                 *writable = false;
1373                 writable = NULL;
1374         }
1375
1376         return hva_to_pfn(addr, atomic, async, write_fault,
1377                           writable);
1378 }
1379
1380 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1381                           bool write_fault, bool *writable)
1382 {
1383         struct kvm_memory_slot *slot;
1384
1385         if (async)
1386                 *async = false;
1387
1388         slot = gfn_to_memslot(kvm, gfn);
1389
1390         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1391                                     writable);
1392 }
1393
1394 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1395 {
1396         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1397 }
1398 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1399
1400 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1401                        bool write_fault, bool *writable)
1402 {
1403         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1404 }
1405 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1406
1407 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1408 {
1409         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1410 }
1411 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1412
1413 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1414                       bool *writable)
1415 {
1416         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1417 }
1418 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1419
1420 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1421 {
1422         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1423 }
1424
1425 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1426 {
1427         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1428 }
1429 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1430
1431 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1432                                                                   int nr_pages)
1433 {
1434         unsigned long addr;
1435         gfn_t entry;
1436
1437         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1438         if (kvm_is_error_hva(addr))
1439                 return -1;
1440
1441         if (entry < nr_pages)
1442                 return 0;
1443
1444         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1445 }
1446 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1447
1448 static struct page *kvm_pfn_to_page(pfn_t pfn)
1449 {
1450         if (is_error_noslot_pfn(pfn))
1451                 return KVM_ERR_PTR_BAD_PAGE;
1452
1453         if (kvm_is_reserved_pfn(pfn)) {
1454                 WARN_ON(1);
1455                 return KVM_ERR_PTR_BAD_PAGE;
1456         }
1457
1458         return pfn_to_page(pfn);
1459 }
1460
1461 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1462 {
1463         pfn_t pfn;
1464
1465         pfn = gfn_to_pfn(kvm, gfn);
1466
1467         return kvm_pfn_to_page(pfn);
1468 }
1469 EXPORT_SYMBOL_GPL(gfn_to_page);
1470
1471 void kvm_release_page_clean(struct page *page)
1472 {
1473         WARN_ON(is_error_page(page));
1474
1475         kvm_release_pfn_clean(page_to_pfn(page));
1476 }
1477 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1478
1479 void kvm_release_pfn_clean(pfn_t pfn)
1480 {
1481         if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1482                 put_page(pfn_to_page(pfn));
1483 }
1484 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1485
1486 void kvm_release_page_dirty(struct page *page)
1487 {
1488         WARN_ON(is_error_page(page));
1489
1490         kvm_release_pfn_dirty(page_to_pfn(page));
1491 }
1492 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1493
1494 static void kvm_release_pfn_dirty(pfn_t pfn)
1495 {
1496         kvm_set_pfn_dirty(pfn);
1497         kvm_release_pfn_clean(pfn);
1498 }
1499
1500 void kvm_set_pfn_dirty(pfn_t pfn)
1501 {
1502         if (!kvm_is_reserved_pfn(pfn)) {
1503                 struct page *page = pfn_to_page(pfn);
1504
1505                 if (!PageReserved(page))
1506                         SetPageDirty(page);
1507         }
1508 }
1509 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1510
1511 void kvm_set_pfn_accessed(pfn_t pfn)
1512 {
1513         if (!kvm_is_reserved_pfn(pfn))
1514                 mark_page_accessed(pfn_to_page(pfn));
1515 }
1516 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1517
1518 void kvm_get_pfn(pfn_t pfn)
1519 {
1520         if (!kvm_is_reserved_pfn(pfn))
1521                 get_page(pfn_to_page(pfn));
1522 }
1523 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1524
1525 static int next_segment(unsigned long len, int offset)
1526 {
1527         if (len > PAGE_SIZE - offset)
1528                 return PAGE_SIZE - offset;
1529         else
1530                 return len;
1531 }
1532
1533 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1534                         int len)
1535 {
1536         int r;
1537         unsigned long addr;
1538
1539         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1540         if (kvm_is_error_hva(addr))
1541                 return -EFAULT;
1542         r = __copy_from_user(data, (void __user *)addr + offset, len);
1543         if (r)
1544                 return -EFAULT;
1545         return 0;
1546 }
1547 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1548
1549 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1550 {
1551         gfn_t gfn = gpa >> PAGE_SHIFT;
1552         int seg;
1553         int offset = offset_in_page(gpa);
1554         int ret;
1555
1556         while ((seg = next_segment(len, offset)) != 0) {
1557                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1558                 if (ret < 0)
1559                         return ret;
1560                 offset = 0;
1561                 len -= seg;
1562                 data += seg;
1563                 ++gfn;
1564         }
1565         return 0;
1566 }
1567 EXPORT_SYMBOL_GPL(kvm_read_guest);
1568
1569 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1570                           unsigned long len)
1571 {
1572         int r;
1573         unsigned long addr;
1574         gfn_t gfn = gpa >> PAGE_SHIFT;
1575         int offset = offset_in_page(gpa);
1576
1577         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1578         if (kvm_is_error_hva(addr))
1579                 return -EFAULT;
1580         pagefault_disable();
1581         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1582         pagefault_enable();
1583         if (r)
1584                 return -EFAULT;
1585         return 0;
1586 }
1587 EXPORT_SYMBOL(kvm_read_guest_atomic);
1588
1589 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1590                          int offset, int len)
1591 {
1592         int r;
1593         unsigned long addr;
1594
1595         addr = gfn_to_hva(kvm, gfn);
1596         if (kvm_is_error_hva(addr))
1597                 return -EFAULT;
1598         r = __copy_to_user((void __user *)addr + offset, data, len);
1599         if (r)
1600                 return -EFAULT;
1601         mark_page_dirty(kvm, gfn);
1602         return 0;
1603 }
1604 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1605
1606 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1607                     unsigned long len)
1608 {
1609         gfn_t gfn = gpa >> PAGE_SHIFT;
1610         int seg;
1611         int offset = offset_in_page(gpa);
1612         int ret;
1613
1614         while ((seg = next_segment(len, offset)) != 0) {
1615                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1616                 if (ret < 0)
1617                         return ret;
1618                 offset = 0;
1619                 len -= seg;
1620                 data += seg;
1621                 ++gfn;
1622         }
1623         return 0;
1624 }
1625 EXPORT_SYMBOL_GPL(kvm_write_guest);
1626
1627 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1628                               gpa_t gpa, unsigned long len)
1629 {
1630         struct kvm_memslots *slots = kvm_memslots(kvm);
1631         int offset = offset_in_page(gpa);
1632         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1633         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1634         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1635         gfn_t nr_pages_avail;
1636
1637         ghc->gpa = gpa;
1638         ghc->generation = slots->generation;
1639         ghc->len = len;
1640         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1641         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1642         if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1643                 ghc->hva += offset;
1644         } else {
1645                 /*
1646                  * If the requested region crosses two memslots, we still
1647                  * verify that the entire region is valid here.
1648                  */
1649                 while (start_gfn <= end_gfn) {
1650                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1651                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1652                                                    &nr_pages_avail);
1653                         if (kvm_is_error_hva(ghc->hva))
1654                                 return -EFAULT;
1655                         start_gfn += nr_pages_avail;
1656                 }
1657                 /* Use the slow path for cross page reads and writes. */
1658                 ghc->memslot = NULL;
1659         }
1660         return 0;
1661 }
1662 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1663
1664 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1665                            void *data, unsigned long len)
1666 {
1667         struct kvm_memslots *slots = kvm_memslots(kvm);
1668         int r;
1669
1670         BUG_ON(len > ghc->len);
1671
1672         if (slots->generation != ghc->generation)
1673                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1674
1675         if (unlikely(!ghc->memslot))
1676                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1677
1678         if (kvm_is_error_hva(ghc->hva))
1679                 return -EFAULT;
1680
1681         r = __copy_to_user((void __user *)ghc->hva, data, len);
1682         if (r)
1683                 return -EFAULT;
1684         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1685
1686         return 0;
1687 }
1688 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1689
1690 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1691                            void *data, unsigned long len)
1692 {
1693         struct kvm_memslots *slots = kvm_memslots(kvm);
1694         int r;
1695
1696         BUG_ON(len > ghc->len);
1697
1698         if (slots->generation != ghc->generation)
1699                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1700
1701         if (unlikely(!ghc->memslot))
1702                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1703
1704         if (kvm_is_error_hva(ghc->hva))
1705                 return -EFAULT;
1706
1707         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1708         if (r)
1709                 return -EFAULT;
1710
1711         return 0;
1712 }
1713 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1714
1715 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1716 {
1717         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1718
1719         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1720 }
1721 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1722
1723 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1724 {
1725         gfn_t gfn = gpa >> PAGE_SHIFT;
1726         int seg;
1727         int offset = offset_in_page(gpa);
1728         int ret;
1729
1730         while ((seg = next_segment(len, offset)) != 0) {
1731                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1732                 if (ret < 0)
1733                         return ret;
1734                 offset = 0;
1735                 len -= seg;
1736                 ++gfn;
1737         }
1738         return 0;
1739 }
1740 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1741
1742 static void mark_page_dirty_in_slot(struct kvm *kvm,
1743                                     struct kvm_memory_slot *memslot,
1744                                     gfn_t gfn)
1745 {
1746         if (memslot && memslot->dirty_bitmap) {
1747                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1748
1749                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1750         }
1751 }
1752
1753 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1754 {
1755         struct kvm_memory_slot *memslot;
1756
1757         memslot = gfn_to_memslot(kvm, gfn);
1758         mark_page_dirty_in_slot(kvm, memslot, gfn);
1759 }
1760 EXPORT_SYMBOL_GPL(mark_page_dirty);
1761
1762 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1763 {
1764         if (kvm_arch_vcpu_runnable(vcpu)) {
1765                 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1766                 return -EINTR;
1767         }
1768         if (kvm_cpu_has_pending_timer(vcpu))
1769                 return -EINTR;
1770         if (signal_pending(current))
1771                 return -EINTR;
1772
1773         return 0;
1774 }
1775
1776 /*
1777  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1778  */
1779 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1780 {
1781         ktime_t start, cur;
1782         DEFINE_SWAITER(wait);
1783         bool waited = false;
1784
1785         start = cur = ktime_get();
1786         if (halt_poll_ns) {
1787                 ktime_t stop = ktime_add_ns(ktime_get(), halt_poll_ns);
1788
1789                 do {
1790                         /*
1791                          * This sets KVM_REQ_UNHALT if an interrupt
1792                          * arrives.
1793                          */
1794                         if (kvm_vcpu_check_block(vcpu) < 0) {
1795                                 ++vcpu->stat.halt_successful_poll;
1796                                 goto out;
1797                         }
1798                         cur = ktime_get();
1799                 } while (single_task_running() && ktime_before(cur, stop));
1800         }
1801
1802         for (;;) {
1803                 swait_prepare(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1804
1805                 if (kvm_vcpu_check_block(vcpu) < 0)
1806                         break;
1807
1808                 waited = true;
1809                 schedule();
1810         }
1811
1812         swait_finish(&vcpu->wq, &wait);
1813         cur = ktime_get();
1814
1815 out:
1816         trace_kvm_vcpu_wakeup(ktime_to_ns(cur) - ktime_to_ns(start), waited);
1817 }
1818 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1819
1820 #ifndef CONFIG_S390
1821 /*
1822  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1823  */
1824 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1825 {
1826         int me;
1827         int cpu = vcpu->cpu;
1828         struct swait_head *wqp;
1829
1830         wqp = kvm_arch_vcpu_wq(vcpu);
1831         if (swaitqueue_active(wqp)) {
1832                 swait_wake_interruptible(wqp);
1833                 ++vcpu->stat.halt_wakeup;
1834         }
1835
1836         me = get_cpu();
1837         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1838                 if (kvm_arch_vcpu_should_kick(vcpu))
1839                         smp_send_reschedule(cpu);
1840         put_cpu();
1841 }
1842 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1843 #endif /* !CONFIG_S390 */
1844
1845 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1846 {
1847         struct pid *pid;
1848         struct task_struct *task = NULL;
1849         int ret = 0;
1850
1851         rcu_read_lock();
1852         pid = rcu_dereference(target->pid);
1853         if (pid)
1854                 task = get_pid_task(pid, PIDTYPE_PID);
1855         rcu_read_unlock();
1856         if (!task)
1857                 return ret;
1858         ret = yield_to(task, 1);
1859         put_task_struct(task);
1860
1861         return ret;
1862 }
1863 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1864
1865 /*
1866  * Helper that checks whether a VCPU is eligible for directed yield.
1867  * Most eligible candidate to yield is decided by following heuristics:
1868  *
1869  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1870  *  (preempted lock holder), indicated by @in_spin_loop.
1871  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1872  *
1873  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1874  *  chance last time (mostly it has become eligible now since we have probably
1875  *  yielded to lockholder in last iteration. This is done by toggling
1876  *  @dy_eligible each time a VCPU checked for eligibility.)
1877  *
1878  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1879  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1880  *  burning. Giving priority for a potential lock-holder increases lock
1881  *  progress.
1882  *
1883  *  Since algorithm is based on heuristics, accessing another VCPU data without
1884  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1885  *  and continue with next VCPU and so on.
1886  */
1887 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1888 {
1889 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1890         bool eligible;
1891
1892         eligible = !vcpu->spin_loop.in_spin_loop ||
1893                     vcpu->spin_loop.dy_eligible;
1894
1895         if (vcpu->spin_loop.in_spin_loop)
1896                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1897
1898         return eligible;
1899 #else
1900         return true;
1901 #endif
1902 }
1903
1904 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1905 {
1906         struct kvm *kvm = me->kvm;
1907         struct kvm_vcpu *vcpu;
1908         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1909         int yielded = 0;
1910         int try = 3;
1911         int pass;
1912         int i;
1913
1914         kvm_vcpu_set_in_spin_loop(me, true);
1915         /*
1916          * We boost the priority of a VCPU that is runnable but not
1917          * currently running, because it got preempted by something
1918          * else and called schedule in __vcpu_run.  Hopefully that
1919          * VCPU is holding the lock that we need and will release it.
1920          * We approximate round-robin by starting at the last boosted VCPU.
1921          */
1922         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1923                 kvm_for_each_vcpu(i, vcpu, kvm) {
1924                         if (!pass && i <= last_boosted_vcpu) {
1925                                 i = last_boosted_vcpu;
1926                                 continue;
1927                         } else if (pass && i > last_boosted_vcpu)
1928                                 break;
1929                         if (!ACCESS_ONCE(vcpu->preempted))
1930                                 continue;
1931                         if (vcpu == me)
1932                                 continue;
1933                         if (swaitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1934                                 continue;
1935                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1936                                 continue;
1937
1938                         yielded = kvm_vcpu_yield_to(vcpu);
1939                         if (yielded > 0) {
1940                                 kvm->last_boosted_vcpu = i;
1941                                 break;
1942                         } else if (yielded < 0) {
1943                                 try--;
1944                                 if (!try)
1945                                         break;
1946                         }
1947                 }
1948         }
1949         kvm_vcpu_set_in_spin_loop(me, false);
1950
1951         /* Ensure vcpu is not eligible during next spinloop */
1952         kvm_vcpu_set_dy_eligible(me, false);
1953 }
1954 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1955
1956 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1957 {
1958         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1959         struct page *page;
1960
1961         if (vmf->pgoff == 0)
1962                 page = virt_to_page(vcpu->run);
1963 #ifdef CONFIG_X86
1964         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1965                 page = virt_to_page(vcpu->arch.pio_data);
1966 #endif
1967 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1968         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1969                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1970 #endif
1971         else
1972                 return kvm_arch_vcpu_fault(vcpu, vmf);
1973         get_page(page);
1974         vmf->page = page;
1975         return 0;
1976 }
1977
1978 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1979         .fault = kvm_vcpu_fault,
1980 };
1981
1982 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1983 {
1984         vma->vm_ops = &kvm_vcpu_vm_ops;
1985         return 0;
1986 }
1987
1988 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1989 {
1990         struct kvm_vcpu *vcpu = filp->private_data;
1991
1992         kvm_put_kvm(vcpu->kvm);
1993         return 0;
1994 }
1995
1996 static struct file_operations kvm_vcpu_fops = {
1997         .release        = kvm_vcpu_release,
1998         .unlocked_ioctl = kvm_vcpu_ioctl,
1999 #ifdef CONFIG_KVM_COMPAT
2000         .compat_ioctl   = kvm_vcpu_compat_ioctl,
2001 #endif
2002         .mmap           = kvm_vcpu_mmap,
2003         .llseek         = noop_llseek,
2004 };
2005
2006 /*
2007  * Allocates an inode for the vcpu.
2008  */
2009 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2010 {
2011         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2012 }
2013
2014 /*
2015  * Creates some virtual cpus.  Good luck creating more than one.
2016  */
2017 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2018 {
2019         int r;
2020         struct kvm_vcpu *vcpu, *v;
2021
2022         if (id >= KVM_MAX_VCPUS)
2023                 return -EINVAL;
2024
2025         vcpu = kvm_arch_vcpu_create(kvm, id);
2026         if (IS_ERR(vcpu))
2027                 return PTR_ERR(vcpu);
2028
2029         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2030
2031         r = kvm_arch_vcpu_setup(vcpu);
2032         if (r)
2033                 goto vcpu_destroy;
2034
2035         mutex_lock(&kvm->lock);
2036         if (!kvm_vcpu_compatible(vcpu)) {
2037                 r = -EINVAL;
2038                 goto unlock_vcpu_destroy;
2039         }
2040         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2041                 r = -EINVAL;
2042                 goto unlock_vcpu_destroy;
2043         }
2044
2045         kvm_for_each_vcpu(r, v, kvm)
2046                 if (v->vcpu_id == id) {
2047                         r = -EEXIST;
2048                         goto unlock_vcpu_destroy;
2049                 }
2050
2051         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2052
2053         /* Now it's all set up, let userspace reach it */
2054         kvm_get_kvm(kvm);
2055         r = create_vcpu_fd(vcpu);
2056         if (r < 0) {
2057                 kvm_put_kvm(kvm);
2058                 goto unlock_vcpu_destroy;
2059         }
2060
2061         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2062         smp_wmb();
2063         atomic_inc(&kvm->online_vcpus);
2064
2065         mutex_unlock(&kvm->lock);
2066         kvm_arch_vcpu_postcreate(vcpu);
2067         return r;
2068
2069 unlock_vcpu_destroy:
2070         mutex_unlock(&kvm->lock);
2071 vcpu_destroy:
2072         kvm_arch_vcpu_destroy(vcpu);
2073         return r;
2074 }
2075
2076 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2077 {
2078         if (sigset) {
2079                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2080                 vcpu->sigset_active = 1;
2081                 vcpu->sigset = *sigset;
2082         } else
2083                 vcpu->sigset_active = 0;
2084         return 0;
2085 }
2086
2087 static long kvm_vcpu_ioctl(struct file *filp,
2088                            unsigned int ioctl, unsigned long arg)
2089 {
2090         struct kvm_vcpu *vcpu = filp->private_data;
2091         void __user *argp = (void __user *)arg;
2092         int r;
2093         struct kvm_fpu *fpu = NULL;
2094         struct kvm_sregs *kvm_sregs = NULL;
2095
2096         if (vcpu->kvm->mm != current->mm)
2097                 return -EIO;
2098
2099         if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2100                 return -EINVAL;
2101
2102 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2103         /*
2104          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2105          * so vcpu_load() would break it.
2106          */
2107         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2108                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2109 #endif
2110
2111
2112         r = vcpu_load(vcpu);
2113         if (r)
2114                 return r;
2115         switch (ioctl) {
2116         case KVM_RUN:
2117                 r = -EINVAL;
2118                 if (arg)
2119                         goto out;
2120                 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2121                         /* The thread running this VCPU changed. */
2122                         struct pid *oldpid = vcpu->pid;
2123                         struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2124
2125                         rcu_assign_pointer(vcpu->pid, newpid);
2126                         if (oldpid)
2127                                 synchronize_rcu();
2128                         put_pid(oldpid);
2129                 }
2130                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2131                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2132                 break;
2133         case KVM_GET_REGS: {
2134                 struct kvm_regs *kvm_regs;
2135
2136                 r = -ENOMEM;
2137                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2138                 if (!kvm_regs)
2139                         goto out;
2140                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2141                 if (r)
2142                         goto out_free1;
2143                 r = -EFAULT;
2144                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2145                         goto out_free1;
2146                 r = 0;
2147 out_free1:
2148                 kfree(kvm_regs);
2149                 break;
2150         }
2151         case KVM_SET_REGS: {
2152                 struct kvm_regs *kvm_regs;
2153
2154                 r = -ENOMEM;
2155                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2156                 if (IS_ERR(kvm_regs)) {
2157                         r = PTR_ERR(kvm_regs);
2158                         goto out;
2159                 }
2160                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2161                 kfree(kvm_regs);
2162                 break;
2163         }
2164         case KVM_GET_SREGS: {
2165                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2166                 r = -ENOMEM;
2167                 if (!kvm_sregs)
2168                         goto out;
2169                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2170                 if (r)
2171                         goto out;
2172                 r = -EFAULT;
2173                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2174                         goto out;
2175                 r = 0;
2176                 break;
2177         }
2178         case KVM_SET_SREGS: {
2179                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2180                 if (IS_ERR(kvm_sregs)) {
2181                         r = PTR_ERR(kvm_sregs);
2182                         kvm_sregs = NULL;
2183                         goto out;
2184                 }
2185                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2186                 break;
2187         }
2188         case KVM_GET_MP_STATE: {
2189                 struct kvm_mp_state mp_state;
2190
2191                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2192                 if (r)
2193                         goto out;
2194                 r = -EFAULT;
2195                 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2196                         goto out;
2197                 r = 0;
2198                 break;
2199         }
2200         case KVM_SET_MP_STATE: {
2201                 struct kvm_mp_state mp_state;
2202
2203                 r = -EFAULT;
2204                 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2205                         goto out;
2206                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2207                 break;
2208         }
2209         case KVM_TRANSLATE: {
2210                 struct kvm_translation tr;
2211
2212                 r = -EFAULT;
2213                 if (copy_from_user(&tr, argp, sizeof(tr)))
2214                         goto out;
2215                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2216                 if (r)
2217                         goto out;
2218                 r = -EFAULT;
2219                 if (copy_to_user(argp, &tr, sizeof(tr)))
2220                         goto out;
2221                 r = 0;
2222                 break;
2223         }
2224         case KVM_SET_GUEST_DEBUG: {
2225                 struct kvm_guest_debug dbg;
2226
2227                 r = -EFAULT;
2228                 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2229                         goto out;
2230                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2231                 break;
2232         }
2233         case KVM_SET_SIGNAL_MASK: {
2234                 struct kvm_signal_mask __user *sigmask_arg = argp;
2235                 struct kvm_signal_mask kvm_sigmask;
2236                 sigset_t sigset, *p;
2237
2238                 p = NULL;
2239                 if (argp) {
2240                         r = -EFAULT;
2241                         if (copy_from_user(&kvm_sigmask, argp,
2242                                            sizeof(kvm_sigmask)))
2243                                 goto out;
2244                         r = -EINVAL;
2245                         if (kvm_sigmask.len != sizeof(sigset))
2246                                 goto out;
2247                         r = -EFAULT;
2248                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2249                                            sizeof(sigset)))
2250                                 goto out;
2251                         p = &sigset;
2252                 }
2253                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2254                 break;
2255         }
2256         case KVM_GET_FPU: {
2257                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2258                 r = -ENOMEM;
2259                 if (!fpu)
2260                         goto out;
2261                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2262                 if (r)
2263                         goto out;
2264                 r = -EFAULT;
2265                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2266                         goto out;
2267                 r = 0;
2268                 break;
2269         }
2270         case KVM_SET_FPU: {
2271                 fpu = memdup_user(argp, sizeof(*fpu));
2272                 if (IS_ERR(fpu)) {
2273                         r = PTR_ERR(fpu);
2274                         fpu = NULL;
2275                         goto out;
2276                 }
2277                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2278                 break;
2279         }
2280         default:
2281                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2282         }
2283 out:
2284         vcpu_put(vcpu);
2285         kfree(fpu);
2286         kfree(kvm_sregs);
2287         return r;
2288 }
2289
2290 #ifdef CONFIG_KVM_COMPAT
2291 static long kvm_vcpu_compat_ioctl(struct file *filp,
2292                                   unsigned int ioctl, unsigned long arg)
2293 {
2294         struct kvm_vcpu *vcpu = filp->private_data;
2295         void __user *argp = compat_ptr(arg);
2296         int r;
2297
2298         if (vcpu->kvm->mm != current->mm)
2299                 return -EIO;
2300
2301         switch (ioctl) {
2302         case KVM_SET_SIGNAL_MASK: {
2303                 struct kvm_signal_mask __user *sigmask_arg = argp;
2304                 struct kvm_signal_mask kvm_sigmask;
2305                 compat_sigset_t csigset;
2306                 sigset_t sigset;
2307
2308                 if (argp) {
2309                         r = -EFAULT;
2310                         if (copy_from_user(&kvm_sigmask, argp,
2311                                            sizeof(kvm_sigmask)))
2312                                 goto out;
2313                         r = -EINVAL;
2314                         if (kvm_sigmask.len != sizeof(csigset))
2315                                 goto out;
2316                         r = -EFAULT;
2317                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2318                                            sizeof(csigset)))
2319                                 goto out;
2320                         sigset_from_compat(&sigset, &csigset);
2321                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2322                 } else
2323                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2324                 break;
2325         }
2326         default:
2327                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2328         }
2329
2330 out:
2331         return r;
2332 }
2333 #endif
2334
2335 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2336                                  int (*accessor)(struct kvm_device *dev,
2337                                                  struct kvm_device_attr *attr),
2338                                  unsigned long arg)
2339 {
2340         struct kvm_device_attr attr;
2341
2342         if (!accessor)
2343                 return -EPERM;
2344
2345         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2346                 return -EFAULT;
2347
2348         return accessor(dev, &attr);
2349 }
2350
2351 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2352                              unsigned long arg)
2353 {
2354         struct kvm_device *dev = filp->private_data;
2355
2356         switch (ioctl) {
2357         case KVM_SET_DEVICE_ATTR:
2358                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2359         case KVM_GET_DEVICE_ATTR:
2360                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2361         case KVM_HAS_DEVICE_ATTR:
2362                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2363         default:
2364                 if (dev->ops->ioctl)
2365                         return dev->ops->ioctl(dev, ioctl, arg);
2366
2367                 return -ENOTTY;
2368         }
2369 }
2370
2371 static int kvm_device_release(struct inode *inode, struct file *filp)
2372 {
2373         struct kvm_device *dev = filp->private_data;
2374         struct kvm *kvm = dev->kvm;
2375
2376         kvm_put_kvm(kvm);
2377         return 0;
2378 }
2379
2380 static const struct file_operations kvm_device_fops = {
2381         .unlocked_ioctl = kvm_device_ioctl,
2382 #ifdef CONFIG_KVM_COMPAT
2383         .compat_ioctl = kvm_device_ioctl,
2384 #endif
2385         .release = kvm_device_release,
2386 };
2387
2388 struct kvm_device *kvm_device_from_filp(struct file *filp)
2389 {
2390         if (filp->f_op != &kvm_device_fops)
2391                 return NULL;
2392
2393         return filp->private_data;
2394 }
2395
2396 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2397 #ifdef CONFIG_KVM_MPIC
2398         [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
2399         [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
2400 #endif
2401
2402 #ifdef CONFIG_KVM_XICS
2403         [KVM_DEV_TYPE_XICS]             = &kvm_xics_ops,
2404 #endif
2405 };
2406
2407 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2408 {
2409         if (type >= ARRAY_SIZE(kvm_device_ops_table))
2410                 return -ENOSPC;
2411
2412         if (kvm_device_ops_table[type] != NULL)
2413                 return -EEXIST;
2414
2415         kvm_device_ops_table[type] = ops;
2416         return 0;
2417 }
2418
2419 void kvm_unregister_device_ops(u32 type)
2420 {
2421         if (kvm_device_ops_table[type] != NULL)
2422                 kvm_device_ops_table[type] = NULL;
2423 }
2424
2425 static int kvm_ioctl_create_device(struct kvm *kvm,
2426                                    struct kvm_create_device *cd)
2427 {
2428         struct kvm_device_ops *ops = NULL;
2429         struct kvm_device *dev;
2430         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2431         int ret;
2432
2433         if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2434                 return -ENODEV;
2435
2436         ops = kvm_device_ops_table[cd->type];
2437         if (ops == NULL)
2438                 return -ENODEV;
2439
2440         if (test)
2441                 return 0;
2442
2443         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2444         if (!dev)
2445                 return -ENOMEM;
2446
2447         dev->ops = ops;
2448         dev->kvm = kvm;
2449
2450         ret = ops->create(dev, cd->type);
2451         if (ret < 0) {
2452                 kfree(dev);
2453                 return ret;
2454         }
2455
2456         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2457         if (ret < 0) {
2458                 ops->destroy(dev);
2459                 return ret;
2460         }
2461
2462         list_add(&dev->vm_node, &kvm->devices);
2463         kvm_get_kvm(kvm);
2464         cd->fd = ret;
2465         return 0;
2466 }
2467
2468 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2469 {
2470         switch (arg) {
2471         case KVM_CAP_USER_MEMORY:
2472         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2473         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2474 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2475         case KVM_CAP_SET_BOOT_CPU_ID:
2476 #endif
2477         case KVM_CAP_INTERNAL_ERROR_DATA:
2478 #ifdef CONFIG_HAVE_KVM_MSI
2479         case KVM_CAP_SIGNAL_MSI:
2480 #endif
2481 #ifdef CONFIG_HAVE_KVM_IRQFD
2482         case KVM_CAP_IRQFD:
2483         case KVM_CAP_IRQFD_RESAMPLE:
2484 #endif
2485         case KVM_CAP_CHECK_EXTENSION_VM:
2486                 return 1;
2487 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2488         case KVM_CAP_IRQ_ROUTING:
2489                 return KVM_MAX_IRQ_ROUTES;
2490 #endif
2491         default:
2492                 break;
2493         }
2494         return kvm_vm_ioctl_check_extension(kvm, arg);
2495 }
2496
2497 static long kvm_vm_ioctl(struct file *filp,
2498                            unsigned int ioctl, unsigned long arg)
2499 {
2500         struct kvm *kvm = filp->private_data;
2501         void __user *argp = (void __user *)arg;
2502         int r;
2503
2504         if (kvm->mm != current->mm)
2505                 return -EIO;
2506         switch (ioctl) {
2507         case KVM_CREATE_VCPU:
2508                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2509                 break;
2510         case KVM_SET_USER_MEMORY_REGION: {
2511                 struct kvm_userspace_memory_region kvm_userspace_mem;
2512
2513                 r = -EFAULT;
2514                 if (copy_from_user(&kvm_userspace_mem, argp,
2515                                                 sizeof(kvm_userspace_mem)))
2516                         goto out;
2517
2518                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2519                 break;
2520         }
2521         case KVM_GET_DIRTY_LOG: {
2522                 struct kvm_dirty_log log;
2523
2524                 r = -EFAULT;
2525                 if (copy_from_user(&log, argp, sizeof(log)))
2526                         goto out;
2527                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2528                 break;
2529         }
2530 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2531         case KVM_REGISTER_COALESCED_MMIO: {
2532                 struct kvm_coalesced_mmio_zone zone;
2533
2534                 r = -EFAULT;
2535                 if (copy_from_user(&zone, argp, sizeof(zone)))
2536                         goto out;
2537                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2538                 break;
2539         }
2540         case KVM_UNREGISTER_COALESCED_MMIO: {
2541                 struct kvm_coalesced_mmio_zone zone;
2542
2543                 r = -EFAULT;
2544                 if (copy_from_user(&zone, argp, sizeof(zone)))
2545                         goto out;
2546                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2547                 break;
2548         }
2549 #endif
2550         case KVM_IRQFD: {
2551                 struct kvm_irqfd data;
2552
2553                 r = -EFAULT;
2554                 if (copy_from_user(&data, argp, sizeof(data)))
2555                         goto out;
2556                 r = kvm_irqfd(kvm, &data);
2557                 break;
2558         }
2559         case KVM_IOEVENTFD: {
2560                 struct kvm_ioeventfd data;
2561
2562                 r = -EFAULT;
2563                 if (copy_from_user(&data, argp, sizeof(data)))
2564                         goto out;
2565                 r = kvm_ioeventfd(kvm, &data);
2566                 break;
2567         }
2568 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2569         case KVM_SET_BOOT_CPU_ID:
2570                 r = 0;
2571                 mutex_lock(&kvm->lock);
2572                 if (atomic_read(&kvm->online_vcpus) != 0)
2573                         r = -EBUSY;
2574                 else
2575                         kvm->bsp_vcpu_id = arg;
2576                 mutex_unlock(&kvm->lock);
2577                 break;
2578 #endif
2579 #ifdef CONFIG_HAVE_KVM_MSI
2580         case KVM_SIGNAL_MSI: {
2581                 struct kvm_msi msi;
2582
2583                 r = -EFAULT;
2584                 if (copy_from_user(&msi, argp, sizeof(msi)))
2585                         goto out;
2586                 r = kvm_send_userspace_msi(kvm, &msi);
2587                 break;
2588         }
2589 #endif
2590 #ifdef __KVM_HAVE_IRQ_LINE
2591         case KVM_IRQ_LINE_STATUS:
2592         case KVM_IRQ_LINE: {
2593                 struct kvm_irq_level irq_event;
2594
2595                 r = -EFAULT;
2596                 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2597                         goto out;
2598
2599                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2600                                         ioctl == KVM_IRQ_LINE_STATUS);
2601                 if (r)
2602                         goto out;
2603
2604                 r = -EFAULT;
2605                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2606                         if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2607                                 goto out;
2608                 }
2609
2610                 r = 0;
2611                 break;
2612         }
2613 #endif
2614 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2615         case KVM_SET_GSI_ROUTING: {
2616                 struct kvm_irq_routing routing;
2617                 struct kvm_irq_routing __user *urouting;
2618                 struct kvm_irq_routing_entry *entries;
2619
2620                 r = -EFAULT;
2621                 if (copy_from_user(&routing, argp, sizeof(routing)))
2622                         goto out;
2623                 r = -EINVAL;
2624                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2625                         goto out;
2626                 if (routing.flags)
2627                         goto out;
2628                 r = -ENOMEM;
2629                 entries = vmalloc(routing.nr * sizeof(*entries));
2630                 if (!entries)
2631                         goto out;
2632                 r = -EFAULT;
2633                 urouting = argp;
2634                 if (copy_from_user(entries, urouting->entries,
2635                                    routing.nr * sizeof(*entries)))
2636                         goto out_free_irq_routing;
2637                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2638                                         routing.flags);
2639 out_free_irq_routing:
2640                 vfree(entries);
2641                 break;
2642         }
2643 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2644         case KVM_CREATE_DEVICE: {
2645                 struct kvm_create_device cd;
2646
2647                 r = -EFAULT;
2648                 if (copy_from_user(&cd, argp, sizeof(cd)))
2649                         goto out;
2650
2651                 r = kvm_ioctl_create_device(kvm, &cd);
2652                 if (r)
2653                         goto out;
2654
2655                 r = -EFAULT;
2656                 if (copy_to_user(argp, &cd, sizeof(cd)))
2657                         goto out;
2658
2659                 r = 0;
2660                 break;
2661         }
2662         case KVM_CHECK_EXTENSION:
2663                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2664                 break;
2665         default:
2666                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2667         }
2668 out:
2669         return r;
2670 }
2671
2672 #ifdef CONFIG_KVM_COMPAT
2673 struct compat_kvm_dirty_log {
2674         __u32 slot;
2675         __u32 padding1;
2676         union {
2677                 compat_uptr_t dirty_bitmap; /* one bit per page */
2678                 __u64 padding2;
2679         };
2680 };
2681
2682 static long kvm_vm_compat_ioctl(struct file *filp,
2683                            unsigned int ioctl, unsigned long arg)
2684 {
2685         struct kvm *kvm = filp->private_data;
2686         int r;
2687
2688         if (kvm->mm != current->mm)
2689                 return -EIO;
2690         switch (ioctl) {
2691         case KVM_GET_DIRTY_LOG: {
2692                 struct compat_kvm_dirty_log compat_log;
2693                 struct kvm_dirty_log log;
2694
2695                 r = -EFAULT;
2696                 if (copy_from_user(&compat_log, (void __user *)arg,
2697                                    sizeof(compat_log)))
2698                         goto out;
2699                 log.slot         = compat_log.slot;
2700                 log.padding1     = compat_log.padding1;
2701                 log.padding2     = compat_log.padding2;
2702                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2703
2704                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2705                 break;
2706         }
2707         default:
2708                 r = kvm_vm_ioctl(filp, ioctl, arg);
2709         }
2710
2711 out:
2712         return r;
2713 }
2714 #endif
2715
2716 static struct file_operations kvm_vm_fops = {
2717         .release        = kvm_vm_release,
2718         .unlocked_ioctl = kvm_vm_ioctl,
2719 #ifdef CONFIG_KVM_COMPAT
2720         .compat_ioctl   = kvm_vm_compat_ioctl,
2721 #endif
2722         .llseek         = noop_llseek,
2723 };
2724
2725 static int kvm_dev_ioctl_create_vm(unsigned long type)
2726 {
2727         int r;
2728         struct kvm *kvm;
2729
2730         kvm = kvm_create_vm(type);
2731         if (IS_ERR(kvm))
2732                 return PTR_ERR(kvm);
2733 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2734         r = kvm_coalesced_mmio_init(kvm);
2735         if (r < 0) {
2736                 kvm_put_kvm(kvm);
2737                 return r;
2738         }
2739 #endif
2740         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2741         if (r < 0)
2742                 kvm_put_kvm(kvm);
2743
2744         return r;
2745 }
2746
2747 static long kvm_dev_ioctl(struct file *filp,
2748                           unsigned int ioctl, unsigned long arg)
2749 {
2750         long r = -EINVAL;
2751
2752         switch (ioctl) {
2753         case KVM_GET_API_VERSION:
2754                 if (arg)
2755                         goto out;
2756                 r = KVM_API_VERSION;
2757                 break;
2758         case KVM_CREATE_VM:
2759                 r = kvm_dev_ioctl_create_vm(arg);
2760                 break;
2761         case KVM_CHECK_EXTENSION:
2762                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2763                 break;
2764         case KVM_GET_VCPU_MMAP_SIZE:
2765                 if (arg)
2766                         goto out;
2767                 r = PAGE_SIZE;     /* struct kvm_run */
2768 #ifdef CONFIG_X86
2769                 r += PAGE_SIZE;    /* pio data page */
2770 #endif
2771 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2772                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2773 #endif
2774                 break;
2775         case KVM_TRACE_ENABLE:
2776         case KVM_TRACE_PAUSE:
2777         case KVM_TRACE_DISABLE:
2778                 r = -EOPNOTSUPP;
2779                 break;
2780         default:
2781                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2782         }
2783 out:
2784         return r;
2785 }
2786
2787 static struct file_operations kvm_chardev_ops = {
2788         .unlocked_ioctl = kvm_dev_ioctl,
2789         .compat_ioctl   = kvm_dev_ioctl,
2790         .llseek         = noop_llseek,
2791 };
2792
2793 static struct miscdevice kvm_dev = {
2794         KVM_MINOR,
2795         "kvm",
2796         &kvm_chardev_ops,
2797 };
2798
2799 static void hardware_enable_nolock(void *junk)
2800 {
2801         int cpu = raw_smp_processor_id();
2802         int r;
2803
2804         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2805                 return;
2806
2807         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2808
2809         r = kvm_arch_hardware_enable();
2810
2811         if (r) {
2812                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2813                 atomic_inc(&hardware_enable_failed);
2814                 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
2815         }
2816 }
2817
2818 static void hardware_enable(void)
2819 {
2820         raw_spin_lock(&kvm_count_lock);
2821         if (kvm_usage_count)
2822                 hardware_enable_nolock(NULL);
2823         raw_spin_unlock(&kvm_count_lock);
2824 }
2825
2826 static void hardware_disable_nolock(void *junk)
2827 {
2828         int cpu = raw_smp_processor_id();
2829
2830         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2831                 return;
2832         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2833         kvm_arch_hardware_disable();
2834 }
2835
2836 static void hardware_disable(void)
2837 {
2838         raw_spin_lock(&kvm_count_lock);
2839         if (kvm_usage_count)
2840                 hardware_disable_nolock(NULL);
2841         raw_spin_unlock(&kvm_count_lock);
2842 }
2843
2844 static void hardware_disable_all_nolock(void)
2845 {
2846         BUG_ON(!kvm_usage_count);
2847
2848         kvm_usage_count--;
2849         if (!kvm_usage_count)
2850                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2851 }
2852
2853 static void hardware_disable_all(void)
2854 {
2855         raw_spin_lock(&kvm_count_lock);
2856         hardware_disable_all_nolock();
2857         raw_spin_unlock(&kvm_count_lock);
2858 }
2859
2860 static int hardware_enable_all(void)
2861 {
2862         int r = 0;
2863
2864         raw_spin_lock(&kvm_count_lock);
2865
2866         kvm_usage_count++;
2867         if (kvm_usage_count == 1) {
2868                 atomic_set(&hardware_enable_failed, 0);
2869                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2870
2871                 if (atomic_read(&hardware_enable_failed)) {
2872                         hardware_disable_all_nolock();
2873                         r = -EBUSY;
2874                 }
2875         }
2876
2877         raw_spin_unlock(&kvm_count_lock);
2878
2879         return r;
2880 }
2881
2882 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2883                            void *v)
2884 {
2885         int cpu = (long)v;
2886
2887         val &= ~CPU_TASKS_FROZEN;
2888         switch (val) {
2889         case CPU_DYING:
2890                 pr_info("kvm: disabling virtualization on CPU%d\n",
2891                        cpu);
2892                 hardware_disable();
2893                 break;
2894         case CPU_STARTING:
2895                 pr_info("kvm: enabling virtualization on CPU%d\n",
2896                        cpu);
2897                 hardware_enable();
2898                 break;
2899         }
2900         return NOTIFY_OK;
2901 }
2902
2903 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2904                       void *v)
2905 {
2906         /*
2907          * Some (well, at least mine) BIOSes hang on reboot if
2908          * in vmx root mode.
2909          *
2910          * And Intel TXT required VMX off for all cpu when system shutdown.
2911          */
2912         pr_info("kvm: exiting hardware virtualization\n");
2913         kvm_rebooting = true;
2914         on_each_cpu(hardware_disable_nolock, NULL, 1);
2915         return NOTIFY_OK;
2916 }
2917
2918 static struct notifier_block kvm_reboot_notifier = {
2919         .notifier_call = kvm_reboot,
2920         .priority = 0,
2921 };
2922
2923 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2924 {
2925         int i;
2926
2927         for (i = 0; i < bus->dev_count; i++) {
2928                 struct kvm_io_device *pos = bus->range[i].dev;
2929
2930                 kvm_iodevice_destructor(pos);
2931         }
2932         kfree(bus);
2933 }
2934
2935 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2936                                  const struct kvm_io_range *r2)
2937 {
2938         if (r1->addr < r2->addr)
2939                 return -1;
2940         if (r1->addr + r1->len > r2->addr + r2->len)
2941                 return 1;
2942         return 0;
2943 }
2944
2945 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2946 {
2947         return kvm_io_bus_cmp(p1, p2);
2948 }
2949
2950 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2951                           gpa_t addr, int len)
2952 {
2953         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2954                 .addr = addr,
2955                 .len = len,
2956                 .dev = dev,
2957         };
2958
2959         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2960                 kvm_io_bus_sort_cmp, NULL);
2961
2962         return 0;
2963 }
2964
2965 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2966                              gpa_t addr, int len)
2967 {
2968         struct kvm_io_range *range, key;
2969         int off;
2970
2971         key = (struct kvm_io_range) {
2972                 .addr = addr,
2973                 .len = len,
2974         };
2975
2976         range = bsearch(&key, bus->range, bus->dev_count,
2977                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2978         if (range == NULL)
2979                 return -ENOENT;
2980
2981         off = range - bus->range;
2982
2983         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2984                 off--;
2985
2986         return off;
2987 }
2988
2989 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
2990                               struct kvm_io_range *range, const void *val)
2991 {
2992         int idx;
2993
2994         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2995         if (idx < 0)
2996                 return -EOPNOTSUPP;
2997
2998         while (idx < bus->dev_count &&
2999                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3000                 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3001                                         range->len, val))
3002                         return idx;
3003                 idx++;
3004         }
3005
3006         return -EOPNOTSUPP;
3007 }
3008
3009 /* kvm_io_bus_write - called under kvm->slots_lock */
3010 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3011                      int len, const void *val)
3012 {
3013         struct kvm_io_bus *bus;
3014         struct kvm_io_range range;
3015         int r;
3016
3017         range = (struct kvm_io_range) {
3018                 .addr = addr,
3019                 .len = len,
3020         };
3021
3022         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3023         r = __kvm_io_bus_write(vcpu, bus, &range, val);
3024         return r < 0 ? r : 0;
3025 }
3026
3027 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3028 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3029                             gpa_t addr, int len, const void *val, long cookie)
3030 {
3031         struct kvm_io_bus *bus;
3032         struct kvm_io_range range;
3033
3034         range = (struct kvm_io_range) {
3035                 .addr = addr,
3036                 .len = len,
3037         };
3038
3039         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3040
3041         /* First try the device referenced by cookie. */
3042         if ((cookie >= 0) && (cookie < bus->dev_count) &&
3043             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3044                 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3045                                         val))
3046                         return cookie;
3047
3048         /*
3049          * cookie contained garbage; fall back to search and return the
3050          * correct cookie value.
3051          */
3052         return __kvm_io_bus_write(vcpu, bus, &range, val);
3053 }
3054
3055 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3056                              struct kvm_io_range *range, void *val)
3057 {
3058         int idx;
3059
3060         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3061         if (idx < 0)
3062                 return -EOPNOTSUPP;
3063
3064         while (idx < bus->dev_count &&
3065                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3066                 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3067                                        range->len, val))
3068                         return idx;
3069                 idx++;
3070         }
3071
3072         return -EOPNOTSUPP;
3073 }
3074 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3075
3076 /* kvm_io_bus_read - called under kvm->slots_lock */
3077 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3078                     int len, void *val)
3079 {
3080         struct kvm_io_bus *bus;
3081         struct kvm_io_range range;
3082         int r;
3083
3084         range = (struct kvm_io_range) {
3085                 .addr = addr,
3086                 .len = len,
3087         };
3088
3089         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3090         r = __kvm_io_bus_read(vcpu, bus, &range, val);
3091         return r < 0 ? r : 0;
3092 }
3093
3094
3095 /* Caller must hold slots_lock. */
3096 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3097                             int len, struct kvm_io_device *dev)
3098 {
3099         struct kvm_io_bus *new_bus, *bus;
3100
3101         bus = kvm->buses[bus_idx];
3102         /* exclude ioeventfd which is limited by maximum fd */
3103         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3104                 return -ENOSPC;
3105
3106         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3107                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3108         if (!new_bus)
3109                 return -ENOMEM;
3110         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3111                sizeof(struct kvm_io_range)));
3112         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3113         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3114         synchronize_srcu_expedited(&kvm->srcu);
3115         kfree(bus);
3116
3117         return 0;
3118 }
3119
3120 /* Caller must hold slots_lock. */
3121 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3122                               struct kvm_io_device *dev)
3123 {
3124         int i, r;
3125         struct kvm_io_bus *new_bus, *bus;
3126
3127         bus = kvm->buses[bus_idx];
3128         r = -ENOENT;
3129         for (i = 0; i < bus->dev_count; i++)
3130                 if (bus->range[i].dev == dev) {
3131                         r = 0;
3132                         break;
3133                 }
3134
3135         if (r)
3136                 return r;
3137
3138         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3139                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3140         if (!new_bus)
3141                 return -ENOMEM;
3142
3143         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3144         new_bus->dev_count--;
3145         memcpy(new_bus->range + i, bus->range + i + 1,
3146                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3147
3148         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3149         synchronize_srcu_expedited(&kvm->srcu);
3150         kfree(bus);
3151         return r;
3152 }
3153
3154 static struct notifier_block kvm_cpu_notifier = {
3155         .notifier_call = kvm_cpu_hotplug,
3156 };
3157
3158 static int vm_stat_get(void *_offset, u64 *val)
3159 {
3160         unsigned offset = (long)_offset;
3161         struct kvm *kvm;
3162
3163         *val = 0;
3164         spin_lock(&kvm_lock);
3165         list_for_each_entry(kvm, &vm_list, vm_list)
3166                 *val += *(u32 *)((void *)kvm + offset);
3167         spin_unlock(&kvm_lock);
3168         return 0;
3169 }
3170
3171 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3172
3173 static int vcpu_stat_get(void *_offset, u64 *val)
3174 {
3175         unsigned offset = (long)_offset;
3176         struct kvm *kvm;
3177         struct kvm_vcpu *vcpu;
3178         int i;
3179
3180         *val = 0;
3181         spin_lock(&kvm_lock);
3182         list_for_each_entry(kvm, &vm_list, vm_list)
3183                 kvm_for_each_vcpu(i, vcpu, kvm)
3184                         *val += *(u32 *)((void *)vcpu + offset);
3185
3186         spin_unlock(&kvm_lock);
3187         return 0;
3188 }
3189
3190 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3191
3192 static const struct file_operations *stat_fops[] = {
3193         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3194         [KVM_STAT_VM]   = &vm_stat_fops,
3195 };
3196
3197 static int kvm_init_debug(void)
3198 {
3199         int r = -EEXIST;
3200         struct kvm_stats_debugfs_item *p;
3201
3202         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3203         if (kvm_debugfs_dir == NULL)
3204                 goto out;
3205
3206         for (p = debugfs_entries; p->name; ++p) {
3207                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3208                                                 (void *)(long)p->offset,
3209                                                 stat_fops[p->kind]);
3210                 if (p->dentry == NULL)
3211                         goto out_dir;
3212         }
3213
3214         return 0;
3215
3216 out_dir:
3217         debugfs_remove_recursive(kvm_debugfs_dir);
3218 out:
3219         return r;
3220 }
3221
3222 static void kvm_exit_debug(void)
3223 {
3224         struct kvm_stats_debugfs_item *p;
3225
3226         for (p = debugfs_entries; p->name; ++p)
3227                 debugfs_remove(p->dentry);
3228         debugfs_remove(kvm_debugfs_dir);
3229 }
3230
3231 static int kvm_suspend(void)
3232 {
3233         if (kvm_usage_count)
3234                 hardware_disable_nolock(NULL);
3235         return 0;
3236 }
3237
3238 static void kvm_resume(void)
3239 {
3240         if (kvm_usage_count) {
3241                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3242                 hardware_enable_nolock(NULL);
3243         }
3244 }
3245
3246 static struct syscore_ops kvm_syscore_ops = {
3247         .suspend = kvm_suspend,
3248         .resume = kvm_resume,
3249 };
3250
3251 static inline
3252 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3253 {
3254         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3255 }
3256
3257 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3258 {
3259         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3260
3261         if (vcpu->preempted)
3262                 vcpu->preempted = false;
3263
3264         kvm_arch_sched_in(vcpu, cpu);
3265
3266         kvm_arch_vcpu_load(vcpu, cpu);
3267 }
3268
3269 static void kvm_sched_out(struct preempt_notifier *pn,
3270                           struct task_struct *next)
3271 {
3272         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3273
3274         if (current->state == TASK_RUNNING)
3275                 vcpu->preempted = true;
3276         kvm_arch_vcpu_put(vcpu);
3277 }
3278
3279 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3280                   struct module *module)
3281 {
3282         int r;
3283         int cpu;
3284
3285         r = kvm_arch_init(opaque);
3286         if (r)
3287                 goto out_fail;
3288
3289         /*
3290          * kvm_arch_init makes sure there's at most one caller
3291          * for architectures that support multiple implementations,
3292          * like intel and amd on x86.
3293          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3294          * conflicts in case kvm is already setup for another implementation.
3295          */
3296         r = kvm_irqfd_init();
3297         if (r)
3298                 goto out_irqfd;
3299
3300         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3301                 r = -ENOMEM;
3302                 goto out_free_0;
3303         }
3304
3305         r = kvm_arch_hardware_setup();
3306         if (r < 0)
3307                 goto out_free_0a;
3308
3309         for_each_online_cpu(cpu) {
3310                 smp_call_function_single(cpu,
3311                                 kvm_arch_check_processor_compat,
3312                                 &r, 1);
3313                 if (r < 0)
3314                         goto out_free_1;
3315         }
3316
3317         r = register_cpu_notifier(&kvm_cpu_notifier);
3318         if (r)
3319                 goto out_free_2;
3320         register_reboot_notifier(&kvm_reboot_notifier);
3321
3322         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3323         if (!vcpu_align)
3324                 vcpu_align = __alignof__(struct kvm_vcpu);
3325         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3326                                            0, NULL);
3327         if (!kvm_vcpu_cache) {
3328                 r = -ENOMEM;
3329                 goto out_free_3;
3330         }
3331
3332         r = kvm_async_pf_init();
3333         if (r)
3334                 goto out_free;
3335
3336         kvm_chardev_ops.owner = module;
3337         kvm_vm_fops.owner = module;
3338         kvm_vcpu_fops.owner = module;
3339
3340         r = misc_register(&kvm_dev);
3341         if (r) {
3342                 pr_err("kvm: misc device register failed\n");
3343                 goto out_unreg;
3344         }
3345
3346         register_syscore_ops(&kvm_syscore_ops);
3347
3348         kvm_preempt_ops.sched_in = kvm_sched_in;
3349         kvm_preempt_ops.sched_out = kvm_sched_out;
3350
3351         r = kvm_init_debug();
3352         if (r) {
3353                 pr_err("kvm: create debugfs files failed\n");
3354                 goto out_undebugfs;
3355         }
3356
3357         r = kvm_vfio_ops_init();
3358         WARN_ON(r);
3359
3360         return 0;
3361
3362 out_undebugfs:
3363         unregister_syscore_ops(&kvm_syscore_ops);
3364         misc_deregister(&kvm_dev);
3365 out_unreg:
3366         kvm_async_pf_deinit();
3367 out_free:
3368         kmem_cache_destroy(kvm_vcpu_cache);
3369 out_free_3:
3370         unregister_reboot_notifier(&kvm_reboot_notifier);
3371         unregister_cpu_notifier(&kvm_cpu_notifier);
3372 out_free_2:
3373 out_free_1:
3374         kvm_arch_hardware_unsetup();
3375 out_free_0a:
3376         free_cpumask_var(cpus_hardware_enabled);
3377 out_free_0:
3378         kvm_irqfd_exit();
3379 out_irqfd:
3380         kvm_arch_exit();
3381 out_fail:
3382         return r;
3383 }
3384 EXPORT_SYMBOL_GPL(kvm_init);
3385
3386 void kvm_exit(void)
3387 {
3388         kvm_exit_debug();
3389         misc_deregister(&kvm_dev);
3390         kmem_cache_destroy(kvm_vcpu_cache);
3391         kvm_async_pf_deinit();
3392         unregister_syscore_ops(&kvm_syscore_ops);
3393         unregister_reboot_notifier(&kvm_reboot_notifier);
3394         unregister_cpu_notifier(&kvm_cpu_notifier);
3395         on_each_cpu(hardware_disable_nolock, NULL, 1);
3396         kvm_arch_hardware_unsetup();
3397         kvm_arch_exit();
3398         kvm_irqfd_exit();
3399         free_cpumask_var(cpus_hardware_enabled);
3400         kvm_vfio_ops_exit();
3401 }
3402 EXPORT_SYMBOL_GPL(kvm_exit);