virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/syscore_ops.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/ioctl.h>
  56 #include <asm/uaccess.h>
  57 #include <asm/pgtable.h>
  58
  59 #include "coalesced_mmio.h"
  60 #include "async_pf.h"
  61 #include "vfio.h"
  62
  63 #define CREATE_TRACE_POINTS
  64 #include <trace/events/kvm.h>
  65
  66 MODULE_AUTHOR("Qumranet");
  67 MODULE_LICENSE("GPL");
  68
  69 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 = 0;
  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
  93 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  94                            unsigned long arg);
  95 #ifdef CONFIG_KVM_COMPAT
  96 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  97                                   unsigned long arg);
  98 #endif
  99 static int hardware_enable_all(void);
 100 static void hardware_disable_all(void);
 101
 102 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
 103
 104 static void kvm_release_pfn_dirty(pfn_t pfn);
 105 static void mark_page_dirty_in_slot(struct kvm *kvm,
 106                                     struct kvm_memory_slot *memslot, gfn_t gfn);
 107
 108 __visible bool kvm_rebooting;
 109 EXPORT_SYMBOL_GPL(kvm_rebooting);
 110
 111 static bool largepages_enabled = true;
 112
 113 bool kvm_is_reserved_pfn(pfn_t pfn)
 114 {
 115         if (pfn_valid(pfn))
 116                 return PageReserved(pfn_to_page(pfn));
 117
 118         return true;
 119 }
 120
 121 /*
 122  * Switches to specified vcpu, until a matching vcpu_put()
 123  */
 124 int vcpu_load(struct kvm_vcpu *vcpu)
 125 {
 126         int cpu;
 127
 128         if (mutex_lock_killable(&vcpu->mutex))
 129                 return -EINTR;
 130         cpu = get_cpu();
 131         preempt_notifier_register(&vcpu->preempt_notifier);
 132         kvm_arch_vcpu_load(vcpu, cpu);
 133         put_cpu();
 134         return 0;
 135 }
 136
 137 void vcpu_put(struct kvm_vcpu *vcpu)
 138 {
 139         preempt_disable();
 140         kvm_arch_vcpu_put(vcpu);
 141         preempt_notifier_unregister(&vcpu->preempt_notifier);
 142         preempt_enable();
 143         mutex_unlock(&vcpu->mutex);
 144 }
 145
 146 static void ack_flush(void *_completed)
 147 {
 148 }
 149
 150 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
 151 {
 152         int i, cpu, me;
 153         cpumask_var_t cpus;
 154         bool called = true;
 155         struct kvm_vcpu *vcpu;
 156
 157         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 158
 159         me = get_cpu();
 160         kvm_for_each_vcpu(i, vcpu, kvm) {
 161                 kvm_make_request(req, vcpu);
 162                 cpu = vcpu->cpu;
 163
 164                 /* Set ->requests bit before we read ->mode */
 165                 smp_mb();
 166
 167                 if (cpus != NULL && cpu != -1 && cpu != me &&
 168                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 169                         cpumask_set_cpu(cpu, cpus);
 170         }
 171         if (unlikely(cpus == NULL))
 172                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 173         else if (!cpumask_empty(cpus))
 174                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 175         else
 176                 called = false;
 177         put_cpu();
 178         free_cpumask_var(cpus);
 179         return called;
 180 }
 181
 182 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
 183 void kvm_flush_remote_tlbs(struct kvm *kvm)
 184 {
 185         long dirty_count = kvm->tlbs_dirty;
 186
 187         smp_mb();
 188         if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 189                 ++kvm->stat.remote_tlb_flush;
 190         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 191 }
 192 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 193 #endif
 194
 195 void kvm_reload_remote_mmus(struct kvm *kvm)
 196 {
 197         kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 198 }
 199
 200 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 201 {
 202         kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 203 }
 204
 205 void kvm_make_scan_ioapic_request(struct kvm *kvm)
 206 {
 207         kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
 208 }
 209
 210 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 211 {
 212         struct page *page;
 213         int r;
 214
 215         mutex_init(&vcpu->mutex);
 216         vcpu->cpu = -1;
 217         vcpu->kvm = kvm;
 218         vcpu->vcpu_id = id;
 219         vcpu->pid = NULL;
 220         init_waitqueue_head(&vcpu->wq);
 221         kvm_async_pf_vcpu_init(vcpu);
 222
 223         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 224         if (!page) {
 225                 r = -ENOMEM;
 226                 goto fail;
 227         }
 228         vcpu->run = page_address(page);
 229
 230         kvm_vcpu_set_in_spin_loop(vcpu, false);
 231         kvm_vcpu_set_dy_eligible(vcpu, false);
 232         vcpu->preempted = false;
 233
 234         r = kvm_arch_vcpu_init(vcpu);
 235         if (r < 0)
 236                 goto fail_free_run;
 237         return 0;
 238
 239 fail_free_run:
 240         free_page((unsigned long)vcpu->run);
 241 fail:
 242         return r;
 243 }
 244 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 245
 246 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 247 {
 248         put_pid(vcpu->pid);
 249         kvm_arch_vcpu_uninit(vcpu);
 250         free_page((unsigned long)vcpu->run);
 251 }
 252 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 253
 254 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 255 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 256 {
 257         return container_of(mn, struct kvm, mmu_notifier);
 258 }
 259
 260 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 261                                              struct mm_struct *mm,
 262                                              unsigned long address)
 263 {
 264         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 265         int need_tlb_flush, idx;
 266
 267         /*
 268          * When ->invalidate_page runs, the linux pte has been zapped
 269          * already but the page is still allocated until
 270          * ->invalidate_page returns. So if we increase the sequence
 271          * here the kvm page fault will notice if the spte can't be
 272          * established because the page is going to be freed. If
 273          * instead the kvm page fault establishes the spte before
 274          * ->invalidate_page runs, kvm_unmap_hva will release it
 275          * before returning.
 276          *
 277          * The sequence increase only need to be seen at spin_unlock
 278          * time, and not at spin_lock time.
 279          *
 280          * Increasing the sequence after the spin_unlock would be
 281          * unsafe because the kvm page fault could then establish the
 282          * pte after kvm_unmap_hva returned, without noticing the page
 283          * is going to be freed.
 284          */
 285         idx = srcu_read_lock(&kvm->srcu);
 286         spin_lock(&kvm->mmu_lock);
 287
 288         kvm->mmu_notifier_seq++;
 289         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 290         /* we've to flush the tlb before the pages can be freed */
 291         if (need_tlb_flush)
 292                 kvm_flush_remote_tlbs(kvm);
 293
 294         spin_unlock(&kvm->mmu_lock);
 295
 296         kvm_arch_mmu_notifier_invalidate_page(kvm, address);
 297
 298         srcu_read_unlock(&kvm->srcu, idx);
 299 }
 300
 301 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 302                                         struct mm_struct *mm,
 303                                         unsigned long address,
 304                                         pte_t pte)
 305 {
 306         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 307         int idx;
 308
 309         idx = srcu_read_lock(&kvm->srcu);
 310         spin_lock(&kvm->mmu_lock);
 311         kvm->mmu_notifier_seq++;
 312         kvm_set_spte_hva(kvm, address, pte);
 313         spin_unlock(&kvm->mmu_lock);
 314         srcu_read_unlock(&kvm->srcu, idx);
 315 }
 316
 317 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 318                                                     struct mm_struct *mm,
 319                                                     unsigned long start,
 320                                                     unsigned long end)
 321 {
 322         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 323         int need_tlb_flush = 0, idx;
 324
 325         idx = srcu_read_lock(&kvm->srcu);
 326         spin_lock(&kvm->mmu_lock);
 327         /*
 328          * The count increase must become visible at unlock time as no
 329          * spte can be established without taking the mmu_lock and
 330          * count is also read inside the mmu_lock critical section.
 331          */
 332         kvm->mmu_notifier_count++;
 333         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 334         need_tlb_flush |= kvm->tlbs_dirty;
 335         /* we've to flush the tlb before the pages can be freed */
 336         if (need_tlb_flush)
 337                 kvm_flush_remote_tlbs(kvm);
 338
 339         spin_unlock(&kvm->mmu_lock);
 340         srcu_read_unlock(&kvm->srcu, idx);
 341 }
 342
 343 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 344                                                   struct mm_struct *mm,
 345                                                   unsigned long start,
 346                                                   unsigned long end)
 347 {
 348         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 349
 350         spin_lock(&kvm->mmu_lock);
 351         /*
 352          * This sequence increase will notify the kvm page fault that
 353          * the page that is going to be mapped in the spte could have
 354          * been freed.
 355          */
 356         kvm->mmu_notifier_seq++;
 357         smp_wmb();
 358         /*
 359          * The above sequence increase must be visible before the
 360          * below count decrease, which is ensured by the smp_wmb above
 361          * in conjunction with the smp_rmb in mmu_notifier_retry().
 362          */
 363         kvm->mmu_notifier_count--;
 364         spin_unlock(&kvm->mmu_lock);
 365
 366         BUG_ON(kvm->mmu_notifier_count < 0);
 367 }
 368
 369 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 370                                               struct mm_struct *mm,
 371                                               unsigned long start,
 372                                               unsigned long end)
 373 {
 374         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 375         int young, idx;
 376
 377         idx = srcu_read_lock(&kvm->srcu);
 378         spin_lock(&kvm->mmu_lock);
 379
 380         young = kvm_age_hva(kvm, start, end);
 381         if (young)
 382                 kvm_flush_remote_tlbs(kvm);
 383
 384         spin_unlock(&kvm->mmu_lock);
 385         srcu_read_unlock(&kvm->srcu, idx);
 386
 387         return young;
 388 }
 389
 390 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 391                                        struct mm_struct *mm,
 392                                        unsigned long address)
 393 {
 394         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 395         int young, idx;
 396
 397         idx = srcu_read_lock(&kvm->srcu);
 398         spin_lock(&kvm->mmu_lock);
 399         young = kvm_test_age_hva(kvm, address);
 400         spin_unlock(&kvm->mmu_lock);
 401         srcu_read_unlock(&kvm->srcu, idx);
 402
 403         return young;
 404 }
 405
 406 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 407                                      struct mm_struct *mm)
 408 {
 409         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 410         int idx;
 411
 412         idx = srcu_read_lock(&kvm->srcu);
 413         kvm_arch_flush_shadow_all(kvm);
 414         srcu_read_unlock(&kvm->srcu, idx);
 415 }
 416
 417 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 418         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 419         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 420         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 421         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 422         .test_young             = kvm_mmu_notifier_test_young,
 423         .change_pte             = kvm_mmu_notifier_change_pte,
 424         .release                = kvm_mmu_notifier_release,
 425 };
 426
 427 static int kvm_init_mmu_notifier(struct kvm *kvm)
 428 {
 429         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 430         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 431 }
 432
 433 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 434
 435 static int kvm_init_mmu_notifier(struct kvm *kvm)
 436 {
 437         return 0;
 438 }
 439
 440 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 441
 442 static void kvm_init_memslots_id(struct kvm *kvm)
 443 {
 444         int i;
 445         struct kvm_memslots *slots = kvm->memslots;
 446
 447         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 448                 slots->id_to_index[i] = slots->memslots[i].id = i;
 449 }
 450
 451 static struct kvm *kvm_create_vm(unsigned long type)
 452 {
 453         int r, i;
 454         struct kvm *kvm = kvm_arch_alloc_vm();
 455
 456         if (!kvm)
 457                 return ERR_PTR(-ENOMEM);
 458
 459         r = kvm_arch_init_vm(kvm, type);
 460         if (r)
 461                 goto out_err_no_disable;
 462
 463         r = hardware_enable_all();
 464         if (r)
 465                 goto out_err_no_disable;
 466
 467 #ifdef CONFIG_HAVE_KVM_IRQFD
 468         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 469 #endif
 470
 471         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 472
 473         r = -ENOMEM;
 474         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 475         if (!kvm->memslots)
 476                 goto out_err_no_srcu;
 477
 478         /*
 479          * Init kvm generation close to the maximum to easily test the
 480          * code of handling generation number wrap-around.
 481          */
 482         kvm->memslots->generation = -150;
 483
 484         kvm_init_memslots_id(kvm);
 485         if (init_srcu_struct(&kvm->srcu))
 486                 goto out_err_no_srcu;
 487         if (init_srcu_struct(&kvm->irq_srcu))
 488                 goto out_err_no_irq_srcu;
 489         for (i = 0; i < KVM_NR_BUSES; i++) {
 490                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 491                                         GFP_KERNEL);
 492                 if (!kvm->buses[i])
 493                         goto out_err;
 494         }
 495
 496         spin_lock_init(&kvm->mmu_lock);
 497         kvm->mm = current->mm;
 498         atomic_inc(&kvm->mm->mm_count);
 499         kvm_eventfd_init(kvm);
 500         mutex_init(&kvm->lock);
 501         mutex_init(&kvm->irq_lock);
 502         mutex_init(&kvm->slots_lock);
 503         atomic_set(&kvm->users_count, 1);
 504         INIT_LIST_HEAD(&kvm->devices);
 505
 506         r = kvm_init_mmu_notifier(kvm);
 507         if (r)
 508                 goto out_err;
 509
 510         spin_lock(&kvm_lock);
 511         list_add(&kvm->vm_list, &vm_list);
 512         spin_unlock(&kvm_lock);
 513
 514         return kvm;
 515
 516 out_err:
 517         cleanup_srcu_struct(&kvm->irq_srcu);
 518 out_err_no_irq_srcu:
 519         cleanup_srcu_struct(&kvm->srcu);
 520 out_err_no_srcu:
 521         hardware_disable_all();
 522 out_err_no_disable:
 523         for (i = 0; i < KVM_NR_BUSES; i++)
 524                 kfree(kvm->buses[i]);
 525         kfree(kvm->memslots);
 526         kvm_arch_free_vm(kvm);
 527         return ERR_PTR(r);
 528 }
 529
 530 /*
 531  * Avoid using vmalloc for a small buffer.
 532  * Should not be used when the size is statically known.
 533  */
 534 void *kvm_kvzalloc(unsigned long size)
 535 {
 536         if (size > PAGE_SIZE)
 537                 return vzalloc(size);
 538         else
 539                 return kzalloc(size, GFP_KERNEL);
 540 }
 541
 542 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 543 {
 544         if (!memslot->dirty_bitmap)
 545                 return;
 546
 547         kvfree(memslot->dirty_bitmap);
 548         memslot->dirty_bitmap = NULL;
 549 }
 550
 551 /*
 552  * Free any memory in @free but not in @dont.
 553  */
 554 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
 555                                   struct kvm_memory_slot *dont)
 556 {
 557         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 558                 kvm_destroy_dirty_bitmap(free);
 559
 560         kvm_arch_free_memslot(kvm, free, dont);
 561
 562         free->npages = 0;
 563 }
 564
 565 static void kvm_free_physmem(struct kvm *kvm)
 566 {
 567         struct kvm_memslots *slots = kvm->memslots;
 568         struct kvm_memory_slot *memslot;
 569
 570         kvm_for_each_memslot(memslot, slots)
 571                 kvm_free_physmem_slot(kvm, memslot, NULL);
 572
 573         kfree(kvm->memslots);
 574 }
 575
 576 static void kvm_destroy_devices(struct kvm *kvm)
 577 {
 578         struct list_head *node, *tmp;
 579
 580         list_for_each_safe(node, tmp, &kvm->devices) {
 581                 struct kvm_device *dev =
 582                         list_entry(node, struct kvm_device, vm_node);
 583
 584                 list_del(node);
 585                 dev->ops->destroy(dev);
 586         }
 587 }
 588
 589 static void kvm_destroy_vm(struct kvm *kvm)
 590 {
 591         int i;
 592         struct mm_struct *mm = kvm->mm;
 593
 594         kvm_arch_sync_events(kvm);
 595         spin_lock(&kvm_lock);
 596         list_del(&kvm->vm_list);
 597         spin_unlock(&kvm_lock);
 598         kvm_free_irq_routing(kvm);
 599         for (i = 0; i < KVM_NR_BUSES; i++)
 600                 kvm_io_bus_destroy(kvm->buses[i]);
 601         kvm_coalesced_mmio_free(kvm);
 602 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 603         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 604 #else
 605         kvm_arch_flush_shadow_all(kvm);
 606 #endif
 607         kvm_arch_destroy_vm(kvm);
 608         kvm_destroy_devices(kvm);
 609         kvm_free_physmem(kvm);
 610         cleanup_srcu_struct(&kvm->irq_srcu);
 611         cleanup_srcu_struct(&kvm->srcu);
 612         kvm_arch_free_vm(kvm);
 613         hardware_disable_all();
 614         mmdrop(mm);
 615 }
 616
 617 void kvm_get_kvm(struct kvm *kvm)
 618 {
 619         atomic_inc(&kvm->users_count);
 620 }
 621 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 622
 623 void kvm_put_kvm(struct kvm *kvm)
 624 {
 625         if (atomic_dec_and_test(&kvm->users_count))
 626                 kvm_destroy_vm(kvm);
 627 }
 628 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 629
 630
 631 static int kvm_vm_release(struct inode *inode, struct file *filp)
 632 {
 633         struct kvm *kvm = filp->private_data;
 634
 635         kvm_irqfd_release(kvm);
 636
 637         kvm_put_kvm(kvm);
 638         return 0;
 639 }
 640
 641 /*
 642  * Allocation size is twice as large as the actual dirty bitmap size.
 643  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 644  */
 645 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 646 {
 647         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 648
 649         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 650         if (!memslot->dirty_bitmap)
 651                 return -ENOMEM;
 652
 653         return 0;
 654 }
 655
 656 /*
 657  * Insert memslot and re-sort memslots based on their GFN,
 658  * so binary search could be used to lookup GFN.
 659  * Sorting algorithm takes advantage of having initially
 660  * sorted array and known changed memslot position.
 661  */
 662 static void update_memslots(struct kvm_memslots *slots,
 663                             struct kvm_memory_slot *new)
 664 {
 665         int id = new->id;
 666         int i = slots->id_to_index[id];
 667         struct kvm_memory_slot *mslots = slots->memslots;
 668
 669         WARN_ON(mslots[i].id != id);
 670         if (!new->npages) {
 671                 WARN_ON(!mslots[i].npages);
 672                 new->base_gfn = 0;
 673                 new->flags = 0;
 674                 if (mslots[i].npages)
 675                         slots->used_slots--;
 676         } else {
 677                 if (!mslots[i].npages)
 678                         slots->used_slots++;
 679         }
 680
 681         while (i < KVM_MEM_SLOTS_NUM - 1 &&
 682                new->base_gfn <= mslots[i + 1].base_gfn) {
 683                 if (!mslots[i + 1].npages)
 684                         break;
 685                 mslots[i] = mslots[i + 1];
 686                 slots->id_to_index[mslots[i].id] = i;
 687                 i++;
 688         }
 689
 690         /*
 691          * The ">=" is needed when creating a slot with base_gfn == 0,
 692          * so that it moves before all those with base_gfn == npages == 0.
 693          *
 694          * On the other hand, if new->npages is zero, the above loop has
 695          * already left i pointing to the beginning of the empty part of
 696          * mslots, and the ">=" would move the hole backwards in this
 697          * case---which is wrong.  So skip the loop when deleting a slot.
 698          */
 699         if (new->npages) {
 700                 while (i > 0 &&
 701                        new->base_gfn >= mslots[i - 1].base_gfn) {
 702                         mslots[i] = mslots[i - 1];
 703                         slots->id_to_index[mslots[i].id] = i;
 704                         i--;
 705                 }
 706         } else
 707                 WARN_ON_ONCE(i != slots->used_slots);
 708
 709         mslots[i] = *new;
 710         slots->id_to_index[mslots[i].id] = i;
 711 }
 712
 713 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 714 {
 715         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 716
 717 #ifdef __KVM_HAVE_READONLY_MEM
 718         valid_flags |= KVM_MEM_READONLY;
 719 #endif
 720
 721         if (mem->flags & ~valid_flags)
 722                 return -EINVAL;
 723
 724         return 0;
 725 }
 726
 727 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 728                 struct kvm_memslots *slots)
 729 {
 730         struct kvm_memslots *old_memslots = kvm->memslots;
 731
 732         /*
 733          * Set the low bit in the generation, which disables SPTE caching
 734          * until the end of synchronize_srcu_expedited.
 735          */
 736         WARN_ON(old_memslots->generation & 1);
 737         slots->generation = old_memslots->generation + 1;
 738
 739         rcu_assign_pointer(kvm->memslots, slots);
 740         synchronize_srcu_expedited(&kvm->srcu);
 741
 742         /*
 743          * Increment the new memslot generation a second time. This prevents
 744          * vm exits that race with memslot updates from caching a memslot
 745          * generation that will (potentially) be valid forever.
 746          */
 747         slots->generation++;
 748
 749         kvm_arch_memslots_updated(kvm);
 750
 751         return old_memslots;
 752 }
 753
 754 /*
 755  * Allocate some memory and give it an address in the guest physical address
 756  * space.
 757  *
 758  * Discontiguous memory is allowed, mostly for framebuffers.
 759  *
 760  * Must be called holding kvm->slots_lock for write.
 761  */
 762 int __kvm_set_memory_region(struct kvm *kvm,
 763                             struct kvm_userspace_memory_region *mem)
 764 {
 765         int r;
 766         gfn_t base_gfn;
 767         unsigned long npages;
 768         struct kvm_memory_slot *slot;
 769         struct kvm_memory_slot old, new;
 770         struct kvm_memslots *slots = NULL, *old_memslots;
 771         enum kvm_mr_change change;
 772
 773         r = check_memory_region_flags(mem);
 774         if (r)
 775                 goto out;
 776
 777         r = -EINVAL;
 778         /* General sanity checks */
 779         if (mem->memory_size & (PAGE_SIZE - 1))
 780                 goto out;
 781         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 782                 goto out;
 783         /* We can read the guest memory with __xxx_user() later on. */
 784         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 785             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 786              !access_ok(VERIFY_WRITE,
 787                         (void __user *)(unsigned long)mem->userspace_addr,
 788                         mem->memory_size)))
 789                 goto out;
 790         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 791                 goto out;
 792         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 793                 goto out;
 794
 795         slot = id_to_memslot(kvm->memslots, mem->slot);
 796         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 797         npages = mem->memory_size >> PAGE_SHIFT;
 798
 799         if (npages > KVM_MEM_MAX_NR_PAGES)
 800                 goto out;
 801
 802         if (!npages)
 803                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 804
 805         new = old = *slot;
 806
 807         new.id = mem->slot;
 808         new.base_gfn = base_gfn;
 809         new.npages = npages;
 810         new.flags = mem->flags;
 811
 812         if (npages) {
 813                 if (!old.npages)
 814                         change = KVM_MR_CREATE;
 815                 else { /* Modify an existing slot. */
 816                         if ((mem->userspace_addr != old.userspace_addr) ||
 817                             (npages != old.npages) ||
 818                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 819                                 goto out;
 820
 821                         if (base_gfn != old.base_gfn)
 822                                 change = KVM_MR_MOVE;
 823                         else if (new.flags != old.flags)
 824                                 change = KVM_MR_FLAGS_ONLY;
 825                         else { /* Nothing to change. */
 826                                 r = 0;
 827                                 goto out;
 828                         }
 829                 }
 830         } else if (old.npages) {
 831                 change = KVM_MR_DELETE;
 832         } else /* Modify a non-existent slot: disallowed. */
 833                 goto out;
 834
 835         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 836                 /* Check for overlaps */
 837                 r = -EEXIST;
 838                 kvm_for_each_memslot(slot, kvm->memslots) {
 839                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 840                             (slot->id == mem->slot))
 841                                 continue;
 842                         if (!((base_gfn + npages <= slot->base_gfn) ||
 843                               (base_gfn >= slot->base_gfn + slot->npages)))
 844                                 goto out;
 845                 }
 846         }
 847
 848         /* Free page dirty bitmap if unneeded */
 849         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 850                 new.dirty_bitmap = NULL;
 851
 852         r = -ENOMEM;
 853         if (change == KVM_MR_CREATE) {
 854                 new.userspace_addr = mem->userspace_addr;
 855
 856                 if (kvm_arch_create_memslot(kvm, &new, npages))
 857                         goto out_free;
 858         }
 859
 860         /* Allocate page dirty bitmap if needed */
 861         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 862                 if (kvm_create_dirty_bitmap(&new) < 0)
 863                         goto out_free;
 864         }
 865
 866         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 867                         GFP_KERNEL);
 868         if (!slots)
 869                 goto out_free;
 870
 871         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 872                 slot = id_to_memslot(slots, mem->slot);
 873                 slot->flags |= KVM_MEMSLOT_INVALID;
 874
 875                 old_memslots = install_new_memslots(kvm, slots);
 876
 877                 /* slot was deleted or moved, clear iommu mapping */
 878                 kvm_iommu_unmap_pages(kvm, &old);
 879                 /* From this point no new shadow pages pointing to a deleted,
 880                  * or moved, memslot will be created.
 881                  *
 882                  * validation of sp->gfn happens in:
 883                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 884                  *      - kvm_is_visible_gfn (mmu_check_roots)
 885                  */
 886                 kvm_arch_flush_shadow_memslot(kvm, slot);
 887
 888                 /*
 889                  * We can re-use the old_memslots from above, the only difference
 890                  * from the currently installed memslots is the invalid flag.  This
 891                  * will get overwritten by update_memslots anyway.
 892                  */
 893                 slots = old_memslots;
 894         }
 895
 896         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 897         if (r)
 898                 goto out_slots;
 899
 900         /* actual memory is freed via old in kvm_free_physmem_slot below */
 901         if (change == KVM_MR_DELETE) {
 902                 new.dirty_bitmap = NULL;
 903                 memset(&new.arch, 0, sizeof(new.arch));
 904         }
 905
 906         update_memslots(slots, &new);
 907         old_memslots = install_new_memslots(kvm, slots);
 908
 909         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 910
 911         kvm_free_physmem_slot(kvm, &old, &new);
 912         kfree(old_memslots);
 913
 914         /*
 915          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 916          * un-mapped and re-mapped if their base changes.  Since base change
 917          * unmapping is handled above with slot deletion, mapping alone is
 918          * needed here.  Anything else the iommu might care about for existing
 919          * slots (size changes, userspace addr changes and read-only flag
 920          * changes) is disallowed above, so any other attribute changes getting
 921          * here can be skipped.
 922          */
 923         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 924                 r = kvm_iommu_map_pages(kvm, &new);
 925                 return r;
 926         }
 927
 928         return 0;
 929
 930 out_slots:
 931         kfree(slots);
 932 out_free:
 933         kvm_free_physmem_slot(kvm, &new, &old);
 934 out:
 935         return r;
 936 }
 937 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 938
 939 int kvm_set_memory_region(struct kvm *kvm,
 940                           struct kvm_userspace_memory_region *mem)
 941 {
 942         int r;
 943
 944         mutex_lock(&kvm->slots_lock);
 945         r = __kvm_set_memory_region(kvm, mem);
 946         mutex_unlock(&kvm->slots_lock);
 947         return r;
 948 }
 949 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 950
 951 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 952                                           struct kvm_userspace_memory_region *mem)
 953 {
 954         if (mem->slot >= KVM_USER_MEM_SLOTS)
 955                 return -EINVAL;
 956         return kvm_set_memory_region(kvm, mem);
 957 }
 958
 959 int kvm_get_dirty_log(struct kvm *kvm,
 960                         struct kvm_dirty_log *log, int *is_dirty)
 961 {
 962         struct kvm_memory_slot *memslot;
 963         int r, i;
 964         unsigned long n;
 965         unsigned long any = 0;
 966
 967         r = -EINVAL;
 968         if (log->slot >= KVM_USER_MEM_SLOTS)
 969                 goto out;
 970
 971         memslot = id_to_memslot(kvm->memslots, log->slot);
 972         r = -ENOENT;
 973         if (!memslot->dirty_bitmap)
 974                 goto out;
 975
 976         n = kvm_dirty_bitmap_bytes(memslot);
 977
 978         for (i = 0; !any && i < n/sizeof(long); ++i)
 979                 any = memslot->dirty_bitmap[i];
 980
 981         r = -EFAULT;
 982         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 983                 goto out;
 984
 985         if (any)
 986                 *is_dirty = 1;
 987
 988         r = 0;
 989 out:
 990         return r;
 991 }
 992 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
 993
 994 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
 995 /**
 996  * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
 997  *      are dirty write protect them for next write.
 998  * @kvm:        pointer to kvm instance
 999  * @log:        slot id and address to which we copy the log
1000  * @is_dirty:   flag set if any page is dirty
1001  *
1002  * We need to keep it in mind that VCPU threads can write to the bitmap
1003  * concurrently. So, to avoid losing track of dirty pages we keep the
1004  * following order:
1005  *
1006  *    1. Take a snapshot of the bit and clear it if needed.
1007  *    2. Write protect the corresponding page.
1008  *    3. Copy the snapshot to the userspace.
1009  *    4. Upon return caller flushes TLB's if needed.
1010  *
1011  * Between 2 and 4, the guest may write to the page using the remaining TLB
1012  * entry.  This is not a problem because the page is reported dirty using
1013  * the snapshot taken before and step 4 ensures that writes done after
1014  * exiting to userspace will be logged for the next call.
1015  *
1016  */
1017 int kvm_get_dirty_log_protect(struct kvm *kvm,
1018                         struct kvm_dirty_log *log, bool *is_dirty)
1019 {
1020         struct kvm_memory_slot *memslot;
1021         int r, i;
1022         unsigned long n;
1023         unsigned long *dirty_bitmap;
1024         unsigned long *dirty_bitmap_buffer;
1025
1026         r = -EINVAL;
1027         if (log->slot >= KVM_USER_MEM_SLOTS)
1028                 goto out;
1029
1030         memslot = id_to_memslot(kvm->memslots, log->slot);
1031
1032         dirty_bitmap = memslot->dirty_bitmap;
1033         r = -ENOENT;
1034         if (!dirty_bitmap)
1035                 goto out;
1036
1037         n = kvm_dirty_bitmap_bytes(memslot);
1038
1039         dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1040         memset(dirty_bitmap_buffer, 0, n);
1041
1042         spin_lock(&kvm->mmu_lock);
1043         *is_dirty = false;
1044         for (i = 0; i < n / sizeof(long); i++) {
1045                 unsigned long mask;
1046                 gfn_t offset;
1047
1048                 if (!dirty_bitmap[i])
1049                         continue;
1050
1051                 *is_dirty = true;
1052
1053                 mask = xchg(&dirty_bitmap[i], 0);
1054                 dirty_bitmap_buffer[i] = mask;
1055
1056                 offset = i * BITS_PER_LONG;
1057                 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, offset,
1058                                                                 mask);
1059         }
1060
1061         spin_unlock(&kvm->mmu_lock);
1062
1063         r = -EFAULT;
1064         if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1065                 goto out;
1066
1067         r = 0;
1068 out:
1069         return r;
1070 }
1071 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1072 #endif
1073
1074 bool kvm_largepages_enabled(void)
1075 {
1076         return largepages_enabled;
1077 }
1078
1079 void kvm_disable_largepages(void)
1080 {
1081         largepages_enabled = false;
1082 }
1083 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1084
1085 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1086 {
1087         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1088 }
1089 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1090
1091 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1092 {
1093         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1094
1095         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1096               memslot->flags & KVM_MEMSLOT_INVALID)
1097                 return 0;
1098
1099         return 1;
1100 }
1101 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1102
1103 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1104 {
1105         struct vm_area_struct *vma;
1106         unsigned long addr, size;
1107
1108         size = PAGE_SIZE;
1109
1110         addr = gfn_to_hva(kvm, gfn);
1111         if (kvm_is_error_hva(addr))
1112                 return PAGE_SIZE;
1113
1114         down_read(&current->mm->mmap_sem);
1115         vma = find_vma(current->mm, addr);
1116         if (!vma)
1117                 goto out;
1118
1119         size = vma_kernel_pagesize(vma);
1120
1121 out:
1122         up_read(&current->mm->mmap_sem);
1123
1124         return size;
1125 }
1126
1127 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1128 {
1129         return slot->flags & KVM_MEM_READONLY;
1130 }
1131
1132 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1133                                        gfn_t *nr_pages, bool write)
1134 {
1135         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1136                 return KVM_HVA_ERR_BAD;
1137
1138         if (memslot_is_readonly(slot) && write)
1139                 return KVM_HVA_ERR_RO_BAD;
1140
1141         if (nr_pages)
1142                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1143
1144         return __gfn_to_hva_memslot(slot, gfn);
1145 }
1146
1147 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1148                                      gfn_t *nr_pages)
1149 {
1150         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1151 }
1152
1153 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1154                                         gfn_t gfn)
1155 {
1156         return gfn_to_hva_many(slot, gfn, NULL);
1157 }
1158 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1159
1160 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1161 {
1162         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1163 }
1164 EXPORT_SYMBOL_GPL(gfn_to_hva);
1165
1166 /*
1167  * If writable is set to false, the hva returned by this function is only
1168  * allowed to be read.
1169  */
1170 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1171                                       gfn_t gfn, bool *writable)
1172 {
1173         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1174
1175         if (!kvm_is_error_hva(hva) && writable)
1176                 *writable = !memslot_is_readonly(slot);
1177
1178         return hva;
1179 }
1180
1181 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1182 {
1183         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1184
1185         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1186 }
1187
1188 static int kvm_read_hva(void *data, void __user *hva, int len)
1189 {
1190         return __copy_from_user(data, hva, len);
1191 }
1192
1193 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1194 {
1195         return __copy_from_user_inatomic(data, hva, len);
1196 }
1197
1198 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1199         unsigned long start, int write, struct page **page)
1200 {
1201         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1202
1203         if (write)
1204                 flags |= FOLL_WRITE;
1205
1206         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1207 }
1208
1209 static inline int check_user_page_hwpoison(unsigned long addr)
1210 {
1211         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1212
1213         rc = __get_user_pages(current, current->mm, addr, 1,
1214                               flags, NULL, NULL, NULL);
1215         return rc == -EHWPOISON;
1216 }
1217
1218 /*
1219  * The atomic path to get the writable pfn which will be stored in @pfn,
1220  * true indicates success, otherwise false is returned.
1221  */
1222 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1223                             bool write_fault, bool *writable, pfn_t *pfn)
1224 {
1225         struct page *page[1];
1226         int npages;
1227
1228         if (!(async || atomic))
1229                 return false;
1230
1231         /*
1232          * Fast pin a writable pfn only if it is a write fault request
1233          * or the caller allows to map a writable pfn for a read fault
1234          * request.
1235          */
1236         if (!(write_fault || writable))
1237                 return false;
1238
1239         npages = __get_user_pages_fast(addr, 1, 1, page);
1240         if (npages == 1) {
1241                 *pfn = page_to_pfn(page[0]);
1242
1243                 if (writable)
1244                         *writable = true;
1245                 return true;
1246         }
1247
1248         return false;
1249 }
1250
1251 /*
1252  * The slow path to get the pfn of the specified host virtual address,
1253  * 1 indicates success, -errno is returned if error is detected.
1254  */
1255 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1256                            bool *writable, pfn_t *pfn)
1257 {
1258         struct page *page[1];
1259         int npages = 0;
1260
1261         might_sleep();
1262
1263         if (writable)
1264                 *writable = write_fault;
1265
1266         if (async) {
1267                 down_read(&current->mm->mmap_sem);
1268                 npages = get_user_page_nowait(current, current->mm,
1269                                               addr, write_fault, page);
1270                 up_read(&current->mm->mmap_sem);
1271         } else
1272                 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1273                                                    write_fault, 0, page,
1274                                                    FOLL_TOUCH|FOLL_HWPOISON);
1275         if (npages != 1)
1276                 return npages;
1277
1278         /* map read fault as writable if possible */
1279         if (unlikely(!write_fault) && writable) {
1280                 struct page *wpage[1];
1281
1282                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1283                 if (npages == 1) {
1284                         *writable = true;
1285                         put_page(page[0]);
1286                         page[0] = wpage[0];
1287                 }
1288
1289                 npages = 1;
1290         }
1291         *pfn = page_to_pfn(page[0]);
1292         return npages;
1293 }
1294
1295 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1296 {
1297         if (unlikely(!(vma->vm_flags & VM_READ)))
1298                 return false;
1299
1300         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1301                 return false;
1302
1303         return true;
1304 }
1305
1306 /*
1307  * Pin guest page in memory and return its pfn.
1308  * @addr: host virtual address which maps memory to the guest
1309  * @atomic: whether this function can sleep
1310  * @async: whether this function need to wait IO complete if the
1311  *         host page is not in the memory
1312  * @write_fault: whether we should get a writable host page
1313  * @writable: whether it allows to map a writable host page for !@write_fault
1314  *
1315  * The function will map a writable host page for these two cases:
1316  * 1): @write_fault = true
1317  * 2): @write_fault = false && @writable, @writable will tell the caller
1318  *     whether the mapping is writable.
1319  */
1320 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1321                         bool write_fault, bool *writable)
1322 {
1323         struct vm_area_struct *vma;
1324         pfn_t pfn = 0;
1325         int npages;
1326
1327         /* we can do it either atomically or asynchronously, not both */
1328         BUG_ON(atomic && async);
1329
1330         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1331                 return pfn;
1332
1333         if (atomic)
1334                 return KVM_PFN_ERR_FAULT;
1335
1336         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1337         if (npages == 1)
1338                 return pfn;
1339
1340         down_read(&current->mm->mmap_sem);
1341         if (npages == -EHWPOISON ||
1342               (!async && check_user_page_hwpoison(addr))) {
1343                 pfn = KVM_PFN_ERR_HWPOISON;
1344                 goto exit;
1345         }
1346
1347         vma = find_vma_intersection(current->mm, addr, addr + 1);
1348
1349         if (vma == NULL)
1350                 pfn = KVM_PFN_ERR_FAULT;
1351         else if ((vma->vm_flags & VM_PFNMAP)) {
1352                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1353                         vma->vm_pgoff;
1354                 BUG_ON(!kvm_is_reserved_pfn(pfn));
1355         } else {
1356                 if (async && vma_is_valid(vma, write_fault))
1357                         *async = true;
1358                 pfn = KVM_PFN_ERR_FAULT;
1359         }
1360 exit:
1361         up_read(&current->mm->mmap_sem);
1362         return pfn;
1363 }
1364
1365 static pfn_t
1366 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1367                      bool *async, bool write_fault, bool *writable)
1368 {
1369         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1370
1371         if (addr == KVM_HVA_ERR_RO_BAD)
1372                 return KVM_PFN_ERR_RO_FAULT;
1373
1374         if (kvm_is_error_hva(addr))
1375                 return KVM_PFN_NOSLOT;
1376
1377         /* Do not map writable pfn in the readonly memslot. */
1378         if (writable && memslot_is_readonly(slot)) {
1379                 *writable = false;
1380                 writable = NULL;
1381         }
1382
1383         return hva_to_pfn(addr, atomic, async, write_fault,
1384                           writable);
1385 }
1386
1387 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1388                           bool write_fault, bool *writable)
1389 {
1390         struct kvm_memory_slot *slot;
1391
1392         if (async)
1393                 *async = false;
1394
1395         slot = gfn_to_memslot(kvm, gfn);
1396
1397         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1398                                     writable);
1399 }
1400
1401 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1402 {
1403         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1404 }
1405 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1406
1407 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1408                        bool write_fault, bool *writable)
1409 {
1410         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1411 }
1412 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1413
1414 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1415 {
1416         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1417 }
1418 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1419
1420 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1421                       bool *writable)
1422 {
1423         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1424 }
1425 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1426
1427 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1428 {
1429         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1430 }
1431
1432 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1433 {
1434         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1435 }
1436 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1437
1438 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1439                                                                   int nr_pages)
1440 {
1441         unsigned long addr;
1442         gfn_t entry;
1443
1444         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1445         if (kvm_is_error_hva(addr))
1446                 return -1;
1447
1448         if (entry < nr_pages)
1449                 return 0;
1450
1451         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1452 }
1453 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1454
1455 static struct page *kvm_pfn_to_page(pfn_t pfn)
1456 {
1457         if (is_error_noslot_pfn(pfn))
1458                 return KVM_ERR_PTR_BAD_PAGE;
1459
1460         if (kvm_is_reserved_pfn(pfn)) {
1461                 WARN_ON(1);
1462                 return KVM_ERR_PTR_BAD_PAGE;
1463         }
1464
1465         return pfn_to_page(pfn);
1466 }
1467
1468 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1469 {
1470         pfn_t pfn;
1471
1472         pfn = gfn_to_pfn(kvm, gfn);
1473
1474         return kvm_pfn_to_page(pfn);
1475 }
1476
1477 EXPORT_SYMBOL_GPL(gfn_to_page);
1478
1479 void kvm_release_page_clean(struct page *page)
1480 {
1481         WARN_ON(is_error_page(page));
1482
1483         kvm_release_pfn_clean(page_to_pfn(page));
1484 }
1485 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1486
1487 void kvm_release_pfn_clean(pfn_t pfn)
1488 {
1489         if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1490                 put_page(pfn_to_page(pfn));
1491 }
1492 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1493
1494 void kvm_release_page_dirty(struct page *page)
1495 {
1496         WARN_ON(is_error_page(page));
1497
1498         kvm_release_pfn_dirty(page_to_pfn(page));
1499 }
1500 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1501
1502 static void kvm_release_pfn_dirty(pfn_t pfn)
1503 {
1504         kvm_set_pfn_dirty(pfn);
1505         kvm_release_pfn_clean(pfn);
1506 }
1507
1508 void kvm_set_pfn_dirty(pfn_t pfn)
1509 {
1510         if (!kvm_is_reserved_pfn(pfn)) {
1511                 struct page *page = pfn_to_page(pfn);
1512                 if (!PageReserved(page))
1513                         SetPageDirty(page);
1514         }
1515 }
1516 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1517
1518 void kvm_set_pfn_accessed(pfn_t pfn)
1519 {
1520         if (!kvm_is_reserved_pfn(pfn))
1521                 mark_page_accessed(pfn_to_page(pfn));
1522 }
1523 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1524
1525 void kvm_get_pfn(pfn_t pfn)
1526 {
1527         if (!kvm_is_reserved_pfn(pfn))
1528                 get_page(pfn_to_page(pfn));
1529 }
1530 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1531
1532 static int next_segment(unsigned long len, int offset)
1533 {
1534         if (len > PAGE_SIZE - offset)
1535                 return PAGE_SIZE - offset;
1536         else
1537                 return len;
1538 }
1539
1540 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1541                         int len)
1542 {
1543         int r;
1544         unsigned long addr;
1545
1546         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1547         if (kvm_is_error_hva(addr))
1548                 return -EFAULT;
1549         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1550         if (r)
1551                 return -EFAULT;
1552         return 0;
1553 }
1554 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1555
1556 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1557 {
1558         gfn_t gfn = gpa >> PAGE_SHIFT;
1559         int seg;
1560         int offset = offset_in_page(gpa);
1561         int ret;
1562
1563         while ((seg = next_segment(len, offset)) != 0) {
1564                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1565                 if (ret < 0)
1566                         return ret;
1567                 offset = 0;
1568                 len -= seg;
1569                 data += seg;
1570                 ++gfn;
1571         }
1572         return 0;
1573 }
1574 EXPORT_SYMBOL_GPL(kvm_read_guest);
1575
1576 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1577                           unsigned long len)
1578 {
1579         int r;
1580         unsigned long addr;
1581         gfn_t gfn = gpa >> PAGE_SHIFT;
1582         int offset = offset_in_page(gpa);
1583
1584         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1585         if (kvm_is_error_hva(addr))
1586                 return -EFAULT;
1587         pagefault_disable();
1588         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1589         pagefault_enable();
1590         if (r)
1591                 return -EFAULT;
1592         return 0;
1593 }
1594 EXPORT_SYMBOL(kvm_read_guest_atomic);
1595
1596 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1597                          int offset, int len)
1598 {
1599         int r;
1600         unsigned long addr;
1601
1602         addr = gfn_to_hva(kvm, gfn);
1603         if (kvm_is_error_hva(addr))
1604                 return -EFAULT;
1605         r = __copy_to_user((void __user *)addr + offset, data, len);
1606         if (r)
1607                 return -EFAULT;
1608         mark_page_dirty(kvm, gfn);
1609         return 0;
1610 }
1611 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1612
1613 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1614                     unsigned long len)
1615 {
1616         gfn_t gfn = gpa >> PAGE_SHIFT;
1617         int seg;
1618         int offset = offset_in_page(gpa);
1619         int ret;
1620
1621         while ((seg = next_segment(len, offset)) != 0) {
1622                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1623                 if (ret < 0)
1624                         return ret;
1625                 offset = 0;
1626                 len -= seg;
1627                 data += seg;
1628                 ++gfn;
1629         }
1630         return 0;
1631 }
1632 EXPORT_SYMBOL_GPL(kvm_write_guest);
1633
1634 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1635                               gpa_t gpa, unsigned long len)
1636 {
1637         struct kvm_memslots *slots = kvm_memslots(kvm);
1638         int offset = offset_in_page(gpa);
1639         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1640         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1641         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1642         gfn_t nr_pages_avail;
1643
1644         ghc->gpa = gpa;
1645         ghc->generation = slots->generation;
1646         ghc->len = len;
1647         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1648         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1649         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1650                 ghc->hva += offset;
1651         } else {
1652                 /*
1653                  * If the requested region crosses two memslots, we still
1654                  * verify that the entire region is valid here.
1655                  */
1656                 while (start_gfn <= end_gfn) {
1657                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1658                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1659                                                    &nr_pages_avail);
1660                         if (kvm_is_error_hva(ghc->hva))
1661                                 return -EFAULT;
1662                         start_gfn += nr_pages_avail;
1663                 }
1664                 /* Use the slow path for cross page reads and writes. */
1665                 ghc->memslot = NULL;
1666         }
1667         return 0;
1668 }
1669 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1670
1671 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1672                            void *data, unsigned long len)
1673 {
1674         struct kvm_memslots *slots = kvm_memslots(kvm);
1675         int r;
1676
1677         BUG_ON(len > ghc->len);
1678
1679         if (slots->generation != ghc->generation)
1680                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1681
1682         if (unlikely(!ghc->memslot))
1683                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1684
1685         if (kvm_is_error_hva(ghc->hva))
1686                 return -EFAULT;
1687
1688         r = __copy_to_user((void __user *)ghc->hva, data, len);
1689         if (r)
1690                 return -EFAULT;
1691         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1692
1693         return 0;
1694 }
1695 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1696
1697 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1698                            void *data, unsigned long len)
1699 {
1700         struct kvm_memslots *slots = kvm_memslots(kvm);
1701         int r;
1702
1703         BUG_ON(len > ghc->len);
1704
1705         if (slots->generation != ghc->generation)
1706                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1707
1708         if (unlikely(!ghc->memslot))
1709                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1710
1711         if (kvm_is_error_hva(ghc->hva))
1712                 return -EFAULT;
1713
1714         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1715         if (r)
1716                 return -EFAULT;
1717
1718         return 0;
1719 }
1720 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1721
1722 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1723 {
1724         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1725
1726         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1727 }
1728 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1729
1730 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1731 {
1732         gfn_t gfn = gpa >> PAGE_SHIFT;
1733         int seg;
1734         int offset = offset_in_page(gpa);
1735         int ret;
1736
1737         while ((seg = next_segment(len, offset)) != 0) {
1738                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1739                 if (ret < 0)
1740                         return ret;
1741                 offset = 0;
1742                 len -= seg;
1743                 ++gfn;
1744         }
1745         return 0;
1746 }
1747 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1748
1749 static void mark_page_dirty_in_slot(struct kvm *kvm,
1750                                     struct kvm_memory_slot *memslot,
1751                                     gfn_t gfn)
1752 {
1753         if (memslot && memslot->dirty_bitmap) {
1754                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1755
1756                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1757         }
1758 }
1759
1760 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1761 {
1762         struct kvm_memory_slot *memslot;
1763
1764         memslot = gfn_to_memslot(kvm, gfn);
1765         mark_page_dirty_in_slot(kvm, memslot, gfn);
1766 }
1767 EXPORT_SYMBOL_GPL(mark_page_dirty);
1768
1769 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1770 {
1771         if (kvm_arch_vcpu_runnable(vcpu)) {
1772                 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1773                 return -EINTR;
1774         }
1775         if (kvm_cpu_has_pending_timer(vcpu))
1776                 return -EINTR;
1777         if (signal_pending(current))
1778                 return -EINTR;
1779
1780         return 0;
1781 }
1782
1783 /*
1784  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1785  */
1786 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1787 {
1788         ktime_t start, cur;
1789         DEFINE_WAIT(wait);
1790         bool waited = false;
1791
1792         start = cur = ktime_get();
1793         if (halt_poll_ns) {
1794                 ktime_t stop = ktime_add_ns(ktime_get(), halt_poll_ns);
1795                 do {
1796                         /*
1797                          * This sets KVM_REQ_UNHALT if an interrupt
1798                          * arrives.
1799                          */
1800                         if (kvm_vcpu_check_block(vcpu) < 0) {
1801                                 ++vcpu->stat.halt_successful_poll;
1802                                 goto out;
1803                         }
1804                         cur = ktime_get();
1805                 } while (single_task_running() && ktime_before(cur, stop));
1806         }
1807
1808         for (;;) {
1809                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1810
1811                 if (kvm_vcpu_check_block(vcpu) < 0)
1812                         break;
1813
1814                 waited = true;
1815                 schedule();
1816         }
1817
1818         finish_wait(&vcpu->wq, &wait);
1819         cur = ktime_get();
1820
1821 out:
1822         trace_kvm_vcpu_wakeup(ktime_to_ns(cur) - ktime_to_ns(start), waited);
1823 }
1824 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1825
1826 #ifndef CONFIG_S390
1827 /*
1828  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1829  */
1830 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1831 {
1832         int me;
1833         int cpu = vcpu->cpu;
1834         wait_queue_head_t *wqp;
1835
1836         wqp = kvm_arch_vcpu_wq(vcpu);
1837         if (waitqueue_active(wqp)) {
1838                 wake_up_interruptible(wqp);
1839                 ++vcpu->stat.halt_wakeup;
1840         }
1841
1842         me = get_cpu();
1843         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1844                 if (kvm_arch_vcpu_should_kick(vcpu))
1845                         smp_send_reschedule(cpu);
1846         put_cpu();
1847 }
1848 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1849 #endif /* !CONFIG_S390 */
1850
1851 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1852 {
1853         struct pid *pid;
1854         struct task_struct *task = NULL;
1855         int ret = 0;
1856
1857         rcu_read_lock();
1858         pid = rcu_dereference(target->pid);
1859         if (pid)
1860                 task = get_pid_task(pid, PIDTYPE_PID);
1861         rcu_read_unlock();
1862         if (!task)
1863                 return ret;
1864         ret = yield_to(task, 1);
1865         put_task_struct(task);
1866
1867         return ret;
1868 }
1869 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1870
1871 /*
1872  * Helper that checks whether a VCPU is eligible for directed yield.
1873  * Most eligible candidate to yield is decided by following heuristics:
1874  *
1875  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1876  *  (preempted lock holder), indicated by @in_spin_loop.
1877  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1878  *
1879  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1880  *  chance last time (mostly it has become eligible now since we have probably
1881  *  yielded to lockholder in last iteration. This is done by toggling
1882  *  @dy_eligible each time a VCPU checked for eligibility.)
1883  *
1884  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1885  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1886  *  burning. Giving priority for a potential lock-holder increases lock
1887  *  progress.
1888  *
1889  *  Since algorithm is based on heuristics, accessing another VCPU data without
1890  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1891  *  and continue with next VCPU and so on.
1892  */
1893 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1894 {
1895 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1896         bool eligible;
1897
1898         eligible = !vcpu->spin_loop.in_spin_loop ||
1899                     vcpu->spin_loop.dy_eligible;
1900
1901         if (vcpu->spin_loop.in_spin_loop)
1902                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1903
1904         return eligible;
1905 #else
1906         return true;
1907 #endif
1908 }
1909
1910 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1911 {
1912         struct kvm *kvm = me->kvm;
1913         struct kvm_vcpu *vcpu;
1914         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1915         int yielded = 0;
1916         int try = 3;
1917         int pass;
1918         int i;
1919
1920         kvm_vcpu_set_in_spin_loop(me, true);
1921         /*
1922          * We boost the priority of a VCPU that is runnable but not
1923          * currently running, because it got preempted by something
1924          * else and called schedule in __vcpu_run.  Hopefully that
1925          * VCPU is holding the lock that we need and will release it.
1926          * We approximate round-robin by starting at the last boosted VCPU.
1927          */
1928         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1929                 kvm_for_each_vcpu(i, vcpu, kvm) {
1930                         if (!pass && i <= last_boosted_vcpu) {
1931                                 i = last_boosted_vcpu;
1932                                 continue;
1933                         } else if (pass && i > last_boosted_vcpu)
1934                                 break;
1935                         if (!ACCESS_ONCE(vcpu->preempted))
1936                                 continue;
1937                         if (vcpu == me)
1938                                 continue;
1939                         if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1940                                 continue;
1941                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1942                                 continue;
1943
1944                         yielded = kvm_vcpu_yield_to(vcpu);
1945                         if (yielded > 0) {
1946                                 kvm->last_boosted_vcpu = i;
1947                                 break;
1948                         } else if (yielded < 0) {
1949                                 try--;
1950                                 if (!try)
1951                                         break;
1952                         }
1953                 }
1954         }
1955         kvm_vcpu_set_in_spin_loop(me, false);
1956
1957         /* Ensure vcpu is not eligible during next spinloop */
1958         kvm_vcpu_set_dy_eligible(me, false);
1959 }
1960 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1961
1962 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1963 {
1964         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1965         struct page *page;
1966
1967         if (vmf->pgoff == 0)
1968                 page = virt_to_page(vcpu->run);
1969 #ifdef CONFIG_X86
1970         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1971                 page = virt_to_page(vcpu->arch.pio_data);
1972 #endif
1973 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1974         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1975                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1976 #endif
1977         else
1978                 return kvm_arch_vcpu_fault(vcpu, vmf);
1979         get_page(page);
1980         vmf->page = page;
1981         return 0;
1982 }
1983
1984 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1985         .fault = kvm_vcpu_fault,
1986 };
1987
1988 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1989 {
1990         vma->vm_ops = &kvm_vcpu_vm_ops;
1991         return 0;
1992 }
1993
1994 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1995 {
1996         struct kvm_vcpu *vcpu = filp->private_data;
1997
1998         kvm_put_kvm(vcpu->kvm);
1999         return 0;
2000 }
2001
2002 static struct file_operations kvm_vcpu_fops = {
2003         .release        = kvm_vcpu_release,
2004         .unlocked_ioctl = kvm_vcpu_ioctl,
2005 #ifdef CONFIG_KVM_COMPAT
2006         .compat_ioctl   = kvm_vcpu_compat_ioctl,
2007 #endif
2008         .mmap           = kvm_vcpu_mmap,
2009         .llseek         = noop_llseek,
2010 };
2011
2012 /*
2013  * Allocates an inode for the vcpu.
2014  */
2015 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2016 {
2017         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2018 }
2019
2020 /*
2021  * Creates some virtual cpus.  Good luck creating more than one.
2022  */
2023 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2024 {
2025         int r;
2026         struct kvm_vcpu *vcpu, *v;
2027
2028         if (id >= KVM_MAX_VCPUS)
2029                 return -EINVAL;
2030
2031         vcpu = kvm_arch_vcpu_create(kvm, id);
2032         if (IS_ERR(vcpu))
2033                 return PTR_ERR(vcpu);
2034
2035         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2036
2037         r = kvm_arch_vcpu_setup(vcpu);
2038         if (r)
2039                 goto vcpu_destroy;
2040
2041         mutex_lock(&kvm->lock);
2042         if (!kvm_vcpu_compatible(vcpu)) {
2043                 r = -EINVAL;
2044                 goto unlock_vcpu_destroy;
2045         }
2046         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2047                 r = -EINVAL;
2048                 goto unlock_vcpu_destroy;
2049         }
2050
2051         kvm_for_each_vcpu(r, v, kvm)
2052                 if (v->vcpu_id == id) {
2053                         r = -EEXIST;
2054                         goto unlock_vcpu_destroy;
2055                 }
2056
2057         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2058
2059         /* Now it's all set up, let userspace reach it */
2060         kvm_get_kvm(kvm);
2061         r = create_vcpu_fd(vcpu);
2062         if (r < 0) {
2063                 kvm_put_kvm(kvm);
2064                 goto unlock_vcpu_destroy;
2065         }
2066
2067         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2068         smp_wmb();
2069         atomic_inc(&kvm->online_vcpus);
2070
2071         mutex_unlock(&kvm->lock);
2072         kvm_arch_vcpu_postcreate(vcpu);
2073         return r;
2074
2075 unlock_vcpu_destroy:
2076         mutex_unlock(&kvm->lock);
2077 vcpu_destroy:
2078         kvm_arch_vcpu_destroy(vcpu);
2079         return r;
2080 }
2081
2082 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2083 {
2084         if (sigset) {
2085                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2086                 vcpu->sigset_active = 1;
2087                 vcpu->sigset = *sigset;
2088         } else
2089                 vcpu->sigset_active = 0;
2090         return 0;
2091 }
2092
2093 static long kvm_vcpu_ioctl(struct file *filp,
2094                            unsigned int ioctl, unsigned long arg)
2095 {
2096         struct kvm_vcpu *vcpu = filp->private_data;
2097         void __user *argp = (void __user *)arg;
2098         int r;
2099         struct kvm_fpu *fpu = NULL;
2100         struct kvm_sregs *kvm_sregs = NULL;
2101
2102         if (vcpu->kvm->mm != current->mm)
2103                 return -EIO;
2104
2105         if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2106                 return -EINVAL;
2107
2108 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2109         /*
2110          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2111          * so vcpu_load() would break it.
2112          */
2113         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2114                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2115 #endif
2116
2117
2118         r = vcpu_load(vcpu);
2119         if (r)
2120                 return r;
2121         switch (ioctl) {
2122         case KVM_RUN:
2123                 r = -EINVAL;
2124                 if (arg)
2125                         goto out;
2126                 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2127                         /* The thread running this VCPU changed. */
2128                         struct pid *oldpid = vcpu->pid;
2129                         struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2130                         rcu_assign_pointer(vcpu->pid, newpid);
2131                         if (oldpid)
2132                                 synchronize_rcu();
2133                         put_pid(oldpid);
2134                 }
2135                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2136                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2137                 break;
2138         case KVM_GET_REGS: {
2139                 struct kvm_regs *kvm_regs;
2140
2141                 r = -ENOMEM;
2142                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2143                 if (!kvm_regs)
2144                         goto out;
2145                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2146                 if (r)
2147                         goto out_free1;
2148                 r = -EFAULT;
2149                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2150                         goto out_free1;
2151                 r = 0;
2152 out_free1:
2153                 kfree(kvm_regs);
2154                 break;
2155         }
2156         case KVM_SET_REGS: {
2157                 struct kvm_regs *kvm_regs;
2158
2159                 r = -ENOMEM;
2160                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2161                 if (IS_ERR(kvm_regs)) {
2162                         r = PTR_ERR(kvm_regs);
2163                         goto out;
2164                 }
2165                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2166                 kfree(kvm_regs);
2167                 break;
2168         }
2169         case KVM_GET_SREGS: {
2170                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2171                 r = -ENOMEM;
2172                 if (!kvm_sregs)
2173                         goto out;
2174                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2175                 if (r)
2176                         goto out;
2177                 r = -EFAULT;
2178                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2179                         goto out;
2180                 r = 0;
2181                 break;
2182         }
2183         case KVM_SET_SREGS: {
2184                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2185                 if (IS_ERR(kvm_sregs)) {
2186                         r = PTR_ERR(kvm_sregs);
2187                         kvm_sregs = NULL;
2188                         goto out;
2189                 }
2190                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2191                 break;
2192         }
2193         case KVM_GET_MP_STATE: {
2194                 struct kvm_mp_state mp_state;
2195
2196                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2197                 if (r)
2198                         goto out;
2199                 r = -EFAULT;
2200                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2201                         goto out;
2202                 r = 0;
2203                 break;
2204         }
2205         case KVM_SET_MP_STATE: {
2206                 struct kvm_mp_state mp_state;
2207
2208                 r = -EFAULT;
2209                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2210                         goto out;
2211                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2212                 break;
2213         }
2214         case KVM_TRANSLATE: {
2215                 struct kvm_translation tr;
2216
2217                 r = -EFAULT;
2218                 if (copy_from_user(&tr, argp, sizeof tr))
2219                         goto out;
2220                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2221                 if (r)
2222                         goto out;
2223                 r = -EFAULT;
2224                 if (copy_to_user(argp, &tr, sizeof tr))
2225                         goto out;
2226                 r = 0;
2227                 break;
2228         }
2229         case KVM_SET_GUEST_DEBUG: {
2230                 struct kvm_guest_debug dbg;
2231
2232                 r = -EFAULT;
2233                 if (copy_from_user(&dbg, argp, sizeof dbg))
2234                         goto out;
2235                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2236                 break;
2237         }
2238         case KVM_SET_SIGNAL_MASK: {
2239                 struct kvm_signal_mask __user *sigmask_arg = argp;
2240                 struct kvm_signal_mask kvm_sigmask;
2241                 sigset_t sigset, *p;
2242
2243                 p = NULL;
2244                 if (argp) {
2245                         r = -EFAULT;
2246                         if (copy_from_user(&kvm_sigmask, argp,
2247                                            sizeof kvm_sigmask))
2248                                 goto out;
2249                         r = -EINVAL;
2250                         if (kvm_sigmask.len != sizeof sigset)
2251                                 goto out;
2252                         r = -EFAULT;
2253                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2254                                            sizeof sigset))
2255                                 goto out;
2256                         p = &sigset;
2257                 }
2258                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2259                 break;
2260         }
2261         case KVM_GET_FPU: {
2262                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2263                 r = -ENOMEM;
2264                 if (!fpu)
2265                         goto out;
2266                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2267                 if (r)
2268                         goto out;
2269                 r = -EFAULT;
2270                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2271                         goto out;
2272                 r = 0;
2273                 break;
2274         }
2275         case KVM_SET_FPU: {
2276                 fpu = memdup_user(argp, sizeof(*fpu));
2277                 if (IS_ERR(fpu)) {
2278                         r = PTR_ERR(fpu);
2279                         fpu = NULL;
2280                         goto out;
2281                 }
2282                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2283                 break;
2284         }
2285         default:
2286                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2287         }
2288 out:
2289         vcpu_put(vcpu);
2290         kfree(fpu);
2291         kfree(kvm_sregs);
2292         return r;
2293 }
2294
2295 #ifdef CONFIG_KVM_COMPAT
2296 static long kvm_vcpu_compat_ioctl(struct file *filp,
2297                                   unsigned int ioctl, unsigned long arg)
2298 {
2299         struct kvm_vcpu *vcpu = filp->private_data;
2300         void __user *argp = compat_ptr(arg);
2301         int r;
2302
2303         if (vcpu->kvm->mm != current->mm)
2304                 return -EIO;
2305
2306         switch (ioctl) {
2307         case KVM_SET_SIGNAL_MASK: {
2308                 struct kvm_signal_mask __user *sigmask_arg = argp;
2309                 struct kvm_signal_mask kvm_sigmask;
2310                 compat_sigset_t csigset;
2311                 sigset_t sigset;
2312
2313                 if (argp) {
2314                         r = -EFAULT;
2315                         if (copy_from_user(&kvm_sigmask, argp,
2316                                            sizeof kvm_sigmask))
2317                                 goto out;
2318                         r = -EINVAL;
2319                         if (kvm_sigmask.len != sizeof csigset)
2320                                 goto out;
2321                         r = -EFAULT;
2322                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2323                                            sizeof csigset))
2324                                 goto out;
2325                         sigset_from_compat(&sigset, &csigset);
2326                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2327                 } else
2328                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2329                 break;
2330         }
2331         default:
2332                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2333         }
2334
2335 out:
2336         return r;
2337 }
2338 #endif
2339
2340 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2341                                  int (*accessor)(struct kvm_device *dev,
2342                                                  struct kvm_device_attr *attr),
2343                                  unsigned long arg)
2344 {
2345         struct kvm_device_attr attr;
2346
2347         if (!accessor)
2348                 return -EPERM;
2349
2350         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2351                 return -EFAULT;
2352
2353         return accessor(dev, &attr);
2354 }
2355
2356 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2357                              unsigned long arg)
2358 {
2359         struct kvm_device *dev = filp->private_data;
2360
2361         switch (ioctl) {
2362         case KVM_SET_DEVICE_ATTR:
2363                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2364         case KVM_GET_DEVICE_ATTR:
2365                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2366         case KVM_HAS_DEVICE_ATTR:
2367                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2368         default:
2369                 if (dev->ops->ioctl)
2370                         return dev->ops->ioctl(dev, ioctl, arg);
2371
2372                 return -ENOTTY;
2373         }
2374 }
2375
2376 static int kvm_device_release(struct inode *inode, struct file *filp)
2377 {
2378         struct kvm_device *dev = filp->private_data;
2379         struct kvm *kvm = dev->kvm;
2380
2381         kvm_put_kvm(kvm);
2382         return 0;
2383 }
2384
2385 static const struct file_operations kvm_device_fops = {
2386         .unlocked_ioctl = kvm_device_ioctl,
2387 #ifdef CONFIG_KVM_COMPAT
2388         .compat_ioctl = kvm_device_ioctl,
2389 #endif
2390         .release = kvm_device_release,
2391 };
2392
2393 struct kvm_device *kvm_device_from_filp(struct file *filp)
2394 {
2395         if (filp->f_op != &kvm_device_fops)
2396                 return NULL;
2397
2398         return filp->private_data;
2399 }
2400
2401 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2402 #ifdef CONFIG_KVM_MPIC
2403         [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
2404         [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
2405 #endif
2406
2407 #ifdef CONFIG_KVM_XICS
2408         [KVM_DEV_TYPE_XICS]             = &kvm_xics_ops,
2409 #endif
2410 };
2411
2412 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2413 {
2414         if (type >= ARRAY_SIZE(kvm_device_ops_table))
2415                 return -ENOSPC;
2416
2417         if (kvm_device_ops_table[type] != NULL)
2418                 return -EEXIST;
2419
2420         kvm_device_ops_table[type] = ops;
2421         return 0;
2422 }
2423
2424 void kvm_unregister_device_ops(u32 type)
2425 {
2426         if (kvm_device_ops_table[type] != NULL)
2427                 kvm_device_ops_table[type] = NULL;
2428 }
2429
2430 static int kvm_ioctl_create_device(struct kvm *kvm,
2431                                    struct kvm_create_device *cd)
2432 {
2433         struct kvm_device_ops *ops = NULL;
2434         struct kvm_device *dev;
2435         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2436         int ret;
2437
2438         if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2439                 return -ENODEV;
2440
2441         ops = kvm_device_ops_table[cd->type];
2442         if (ops == NULL)
2443                 return -ENODEV;
2444
2445         if (test)
2446                 return 0;
2447
2448         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2449         if (!dev)
2450                 return -ENOMEM;
2451
2452         dev->ops = ops;
2453         dev->kvm = kvm;
2454
2455         ret = ops->create(dev, cd->type);
2456         if (ret < 0) {
2457                 kfree(dev);
2458                 return ret;
2459         }
2460
2461         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2462         if (ret < 0) {
2463                 ops->destroy(dev);
2464                 return ret;
2465         }
2466
2467         list_add(&dev->vm_node, &kvm->devices);
2468         kvm_get_kvm(kvm);
2469         cd->fd = ret;
2470         return 0;
2471 }
2472
2473 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2474 {
2475         switch (arg) {
2476         case KVM_CAP_USER_MEMORY:
2477         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2478         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2479 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2480         case KVM_CAP_SET_BOOT_CPU_ID:
2481 #endif
2482         case KVM_CAP_INTERNAL_ERROR_DATA:
2483 #ifdef CONFIG_HAVE_KVM_MSI
2484         case KVM_CAP_SIGNAL_MSI:
2485 #endif
2486 #ifdef CONFIG_HAVE_KVM_IRQFD
2487         case KVM_CAP_IRQFD_RESAMPLE:
2488 #endif
2489         case KVM_CAP_CHECK_EXTENSION_VM:
2490                 return 1;
2491 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2492         case KVM_CAP_IRQ_ROUTING:
2493                 return KVM_MAX_IRQ_ROUTES;
2494 #endif
2495         default:
2496                 break;
2497         }
2498         return kvm_vm_ioctl_check_extension(kvm, arg);
2499 }
2500
2501 static long kvm_vm_ioctl(struct file *filp,
2502                            unsigned int ioctl, unsigned long arg)
2503 {
2504         struct kvm *kvm = filp->private_data;
2505         void __user *argp = (void __user *)arg;
2506         int r;
2507
2508         if (kvm->mm != current->mm)
2509                 return -EIO;
2510         switch (ioctl) {
2511         case KVM_CREATE_VCPU:
2512                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2513                 break;
2514         case KVM_SET_USER_MEMORY_REGION: {
2515                 struct kvm_userspace_memory_region kvm_userspace_mem;
2516
2517                 r = -EFAULT;
2518                 if (copy_from_user(&kvm_userspace_mem, argp,
2519                                                 sizeof kvm_userspace_mem))
2520                         goto out;
2521
2522                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2523                 break;
2524         }
2525         case KVM_GET_DIRTY_LOG: {
2526                 struct kvm_dirty_log log;
2527
2528                 r = -EFAULT;
2529                 if (copy_from_user(&log, argp, sizeof log))
2530                         goto out;
2531                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2532                 break;
2533         }
2534 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2535         case KVM_REGISTER_COALESCED_MMIO: {
2536                 struct kvm_coalesced_mmio_zone zone;
2537                 r = -EFAULT;
2538                 if (copy_from_user(&zone, argp, sizeof zone))
2539                         goto out;
2540                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2541                 break;
2542         }
2543         case KVM_UNREGISTER_COALESCED_MMIO: {
2544                 struct kvm_coalesced_mmio_zone zone;
2545                 r = -EFAULT;
2546                 if (copy_from_user(&zone, argp, sizeof zone))
2547                         goto out;
2548                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2549                 break;
2550         }
2551 #endif
2552         case KVM_IRQFD: {
2553                 struct kvm_irqfd data;
2554
2555                 r = -EFAULT;
2556                 if (copy_from_user(&data, argp, sizeof data))
2557                         goto out;
2558                 r = kvm_irqfd(kvm, &data);
2559                 break;
2560         }
2561         case KVM_IOEVENTFD: {
2562                 struct kvm_ioeventfd data;
2563
2564                 r = -EFAULT;
2565                 if (copy_from_user(&data, argp, sizeof data))
2566                         goto out;
2567                 r = kvm_ioeventfd(kvm, &data);
2568                 break;
2569         }
2570 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2571         case KVM_SET_BOOT_CPU_ID:
2572                 r = 0;
2573                 mutex_lock(&kvm->lock);
2574                 if (atomic_read(&kvm->online_vcpus) != 0)
2575                         r = -EBUSY;
2576                 else
2577                         kvm->bsp_vcpu_id = arg;
2578                 mutex_unlock(&kvm->lock);
2579                 break;
2580 #endif
2581 #ifdef CONFIG_HAVE_KVM_MSI
2582         case KVM_SIGNAL_MSI: {
2583                 struct kvm_msi msi;
2584
2585                 r = -EFAULT;
2586                 if (copy_from_user(&msi, argp, sizeof msi))
2587                         goto out;
2588                 r = kvm_send_userspace_msi(kvm, &msi);
2589                 break;
2590         }
2591 #endif
2592 #ifdef __KVM_HAVE_IRQ_LINE
2593         case KVM_IRQ_LINE_STATUS:
2594         case KVM_IRQ_LINE: {
2595                 struct kvm_irq_level irq_event;
2596
2597                 r = -EFAULT;
2598                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2599                         goto out;
2600
2601                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2602                                         ioctl == KVM_IRQ_LINE_STATUS);
2603                 if (r)
2604                         goto out;
2605
2606                 r = -EFAULT;
2607                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2608                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2609                                 goto out;
2610                 }
2611
2612                 r = 0;
2613                 break;
2614         }
2615 #endif
2616 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2617         case KVM_SET_GSI_ROUTING: {
2618                 struct kvm_irq_routing routing;
2619                 struct kvm_irq_routing __user *urouting;
2620                 struct kvm_irq_routing_entry *entries;
2621
2622                 r = -EFAULT;
2623                 if (copy_from_user(&routing, argp, sizeof(routing)))
2624                         goto out;
2625                 r = -EINVAL;
2626                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2627                         goto out;
2628                 if (routing.flags)
2629                         goto out;
2630                 r = -ENOMEM;
2631                 entries = vmalloc(routing.nr * sizeof(*entries));
2632                 if (!entries)
2633                         goto out;
2634                 r = -EFAULT;
2635                 urouting = argp;
2636                 if (copy_from_user(entries, urouting->entries,
2637                                    routing.nr * sizeof(*entries)))
2638                         goto out_free_irq_routing;
2639                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2640                                         routing.flags);
2641         out_free_irq_routing:
2642                 vfree(entries);
2643                 break;
2644         }
2645 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2646         case KVM_CREATE_DEVICE: {
2647                 struct kvm_create_device cd;
2648
2649                 r = -EFAULT;
2650                 if (copy_from_user(&cd, argp, sizeof(cd)))
2651                         goto out;
2652
2653                 r = kvm_ioctl_create_device(kvm, &cd);
2654                 if (r)
2655                         goto out;
2656
2657                 r = -EFAULT;
2658                 if (copy_to_user(argp, &cd, sizeof(cd)))
2659                         goto out;
2660
2661                 r = 0;
2662                 break;
2663         }
2664         case KVM_CHECK_EXTENSION:
2665                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2666                 break;
2667         default:
2668                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2669         }
2670 out:
2671         return r;
2672 }
2673
2674 #ifdef CONFIG_KVM_COMPAT
2675 struct compat_kvm_dirty_log {
2676         __u32 slot;
2677         __u32 padding1;
2678         union {
2679                 compat_uptr_t dirty_bitmap; /* one bit per page */
2680                 __u64 padding2;
2681         };
2682 };
2683
2684 static long kvm_vm_compat_ioctl(struct file *filp,
2685                            unsigned int ioctl, unsigned long arg)
2686 {
2687         struct kvm *kvm = filp->private_data;
2688         int r;
2689
2690         if (kvm->mm != current->mm)
2691                 return -EIO;
2692         switch (ioctl) {
2693         case KVM_GET_DIRTY_LOG: {
2694                 struct compat_kvm_dirty_log compat_log;
2695                 struct kvm_dirty_log log;
2696
2697                 r = -EFAULT;
2698                 if (copy_from_user(&compat_log, (void __user *)arg,
2699                                    sizeof(compat_log)))
2700                         goto out;
2701                 log.slot         = compat_log.slot;
2702                 log.padding1     = compat_log.padding1;
2703                 log.padding2     = compat_log.padding2;
2704                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2705
2706                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2707                 break;
2708         }
2709         default:
2710                 r = kvm_vm_ioctl(filp, ioctl, arg);
2711         }
2712
2713 out:
2714         return r;
2715 }
2716 #endif
2717
2718 static struct file_operations kvm_vm_fops = {
2719         .release        = kvm_vm_release,
2720         .unlocked_ioctl = kvm_vm_ioctl,
2721 #ifdef CONFIG_KVM_COMPAT
2722         .compat_ioctl   = kvm_vm_compat_ioctl,
2723 #endif
2724         .llseek         = noop_llseek,
2725 };
2726
2727 static int kvm_dev_ioctl_create_vm(unsigned long type)
2728 {
2729         int r;
2730         struct kvm *kvm;
2731
2732         kvm = kvm_create_vm(type);
2733         if (IS_ERR(kvm))
2734                 return PTR_ERR(kvm);
2735 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2736         r = kvm_coalesced_mmio_init(kvm);
2737         if (r < 0) {
2738                 kvm_put_kvm(kvm);
2739                 return r;
2740         }
2741 #endif
2742         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2743         if (r < 0)
2744                 kvm_put_kvm(kvm);
2745
2746         return r;
2747 }
2748
2749 static long kvm_dev_ioctl(struct file *filp,
2750                           unsigned int ioctl, unsigned long arg)
2751 {
2752         long r = -EINVAL;
2753
2754         switch (ioctl) {
2755         case KVM_GET_API_VERSION:
2756                 if (arg)
2757                         goto out;
2758                 r = KVM_API_VERSION;
2759                 break;
2760         case KVM_CREATE_VM:
2761                 r = kvm_dev_ioctl_create_vm(arg);
2762                 break;
2763         case KVM_CHECK_EXTENSION:
2764                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2765                 break;
2766         case KVM_GET_VCPU_MMAP_SIZE:
2767                 if (arg)
2768                         goto out;
2769                 r = PAGE_SIZE;     /* struct kvm_run */
2770 #ifdef CONFIG_X86
2771                 r += PAGE_SIZE;    /* pio data page */
2772 #endif
2773 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2774                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2775 #endif
2776                 break;
2777         case KVM_TRACE_ENABLE:
2778         case KVM_TRACE_PAUSE:
2779         case KVM_TRACE_DISABLE:
2780                 r = -EOPNOTSUPP;
2781                 break;
2782         default:
2783                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2784         }
2785 out:
2786         return r;
2787 }
2788
2789 static struct file_operations kvm_chardev_ops = {
2790         .unlocked_ioctl = kvm_dev_ioctl,
2791         .compat_ioctl   = kvm_dev_ioctl,
2792         .llseek         = noop_llseek,
2793 };
2794
2795 static struct miscdevice kvm_dev = {
2796         KVM_MINOR,
2797         "kvm",
2798         &kvm_chardev_ops,
2799 };
2800
2801 static void hardware_enable_nolock(void *junk)
2802 {
2803         int cpu = raw_smp_processor_id();
2804         int r;
2805
2806         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2807                 return;
2808
2809         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2810
2811         r = kvm_arch_hardware_enable();
2812
2813         if (r) {
2814                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2815                 atomic_inc(&hardware_enable_failed);
2816                 printk(KERN_INFO "kvm: enabling virtualization on "
2817                                  "CPU%d failed\n", cpu);
2818         }
2819 }
2820
2821 static void hardware_enable(void)
2822 {
2823         raw_spin_lock(&kvm_count_lock);
2824         if (kvm_usage_count)
2825                 hardware_enable_nolock(NULL);
2826         raw_spin_unlock(&kvm_count_lock);
2827 }
2828
2829 static void hardware_disable_nolock(void *junk)
2830 {
2831         int cpu = raw_smp_processor_id();
2832
2833         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2834                 return;
2835         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2836         kvm_arch_hardware_disable();
2837 }
2838
2839 static void hardware_disable(void)
2840 {
2841         raw_spin_lock(&kvm_count_lock);
2842         if (kvm_usage_count)
2843                 hardware_disable_nolock(NULL);
2844         raw_spin_unlock(&kvm_count_lock);
2845 }
2846
2847 static void hardware_disable_all_nolock(void)
2848 {
2849         BUG_ON(!kvm_usage_count);
2850
2851         kvm_usage_count--;
2852         if (!kvm_usage_count)
2853                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2854 }
2855
2856 static void hardware_disable_all(void)
2857 {
2858         raw_spin_lock(&kvm_count_lock);
2859         hardware_disable_all_nolock();
2860         raw_spin_unlock(&kvm_count_lock);
2861 }
2862
2863 static int hardware_enable_all(void)
2864 {
2865         int r = 0;
2866
2867         raw_spin_lock(&kvm_count_lock);
2868
2869         kvm_usage_count++;
2870         if (kvm_usage_count == 1) {
2871                 atomic_set(&hardware_enable_failed, 0);
2872                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2873
2874                 if (atomic_read(&hardware_enable_failed)) {
2875                         hardware_disable_all_nolock();
2876                         r = -EBUSY;
2877                 }
2878         }
2879
2880         raw_spin_unlock(&kvm_count_lock);
2881
2882         return r;
2883 }
2884
2885 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2886                            void *v)
2887 {
2888         int cpu = (long)v;
2889
2890         val &= ~CPU_TASKS_FROZEN;
2891         switch (val) {
2892         case CPU_DYING:
2893                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2894                        cpu);
2895                 hardware_disable();
2896                 break;
2897         case CPU_STARTING:
2898                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2899                        cpu);
2900                 hardware_enable();
2901                 break;
2902         }
2903         return NOTIFY_OK;
2904 }
2905
2906 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2907                       void *v)
2908 {
2909         /*
2910          * Some (well, at least mine) BIOSes hang on reboot if
2911          * in vmx root mode.
2912          *
2913          * And Intel TXT required VMX off for all cpu when system shutdown.
2914          */
2915         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2916         kvm_rebooting = true;
2917         on_each_cpu(hardware_disable_nolock, NULL, 1);
2918         return NOTIFY_OK;
2919 }
2920
2921 static struct notifier_block kvm_reboot_notifier = {
2922         .notifier_call = kvm_reboot,
2923         .priority = 0,
2924 };
2925
2926 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2927 {
2928         int i;
2929
2930         for (i = 0; i < bus->dev_count; i++) {
2931                 struct kvm_io_device *pos = bus->range[i].dev;
2932
2933                 kvm_iodevice_destructor(pos);
2934         }
2935         kfree(bus);
2936 }
2937
2938 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2939                                  const struct kvm_io_range *r2)
2940 {
2941         if (r1->addr < r2->addr)
2942                 return -1;
2943         if (r1->addr + r1->len > r2->addr + r2->len)
2944                 return 1;
2945         return 0;
2946 }
2947
2948 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2949 {
2950         return kvm_io_bus_cmp(p1, p2);
2951 }
2952
2953 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2954                           gpa_t addr, int len)
2955 {
2956         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2957                 .addr = addr,
2958                 .len = len,
2959                 .dev = dev,
2960         };
2961
2962         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2963                 kvm_io_bus_sort_cmp, NULL);
2964
2965         return 0;
2966 }
2967
2968 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2969                              gpa_t addr, int len)
2970 {
2971         struct kvm_io_range *range, key;
2972         int off;
2973
2974         key = (struct kvm_io_range) {
2975                 .addr = addr,
2976                 .len = len,
2977         };
2978
2979         range = bsearch(&key, bus->range, bus->dev_count,
2980                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2981         if (range == NULL)
2982                 return -ENOENT;
2983
2984         off = range - bus->range;
2985
2986         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2987                 off--;
2988
2989         return off;
2990 }
2991
2992 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2993                               struct kvm_io_range *range, const void *val)
2994 {
2995         int idx;
2996
2997         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2998         if (idx < 0)
2999                 return -EOPNOTSUPP;
3000
3001         while (idx < bus->dev_count &&
3002                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3003                 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
3004                                         range->len, val))
3005                         return idx;
3006                 idx++;
3007         }
3008
3009         return -EOPNOTSUPP;
3010 }
3011
3012 /* kvm_io_bus_write - called under kvm->slots_lock */
3013 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3014                      int len, const void *val)
3015 {
3016         struct kvm_io_bus *bus;
3017         struct kvm_io_range range;
3018         int r;
3019
3020         range = (struct kvm_io_range) {
3021                 .addr = addr,
3022                 .len = len,
3023         };
3024
3025         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3026         r = __kvm_io_bus_write(bus, &range, val);
3027         return r < 0 ? r : 0;
3028 }
3029
3030 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3031 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3032                             int len, const void *val, long cookie)
3033 {
3034         struct kvm_io_bus *bus;
3035         struct kvm_io_range range;
3036
3037         range = (struct kvm_io_range) {
3038                 .addr = addr,
3039                 .len = len,
3040         };
3041
3042         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3043
3044         /* First try the device referenced by cookie. */
3045         if ((cookie >= 0) && (cookie < bus->dev_count) &&
3046             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3047                 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
3048                                         val))
3049                         return cookie;
3050
3051         /*
3052          * cookie contained garbage; fall back to search and return the
3053          * correct cookie value.
3054          */
3055         return __kvm_io_bus_write(bus, &range, val);
3056 }
3057
3058 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
3059                              void *val)
3060 {
3061         int idx;
3062
3063         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3064         if (idx < 0)
3065                 return -EOPNOTSUPP;
3066
3067         while (idx < bus->dev_count &&
3068                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3069                 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
3070                                        range->len, val))
3071                         return idx;
3072                 idx++;
3073         }
3074
3075         return -EOPNOTSUPP;
3076 }
3077 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3078
3079 /* kvm_io_bus_read - called under kvm->slots_lock */
3080 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3081                     int len, void *val)
3082 {
3083         struct kvm_io_bus *bus;
3084         struct kvm_io_range range;
3085         int r;
3086
3087         range = (struct kvm_io_range) {
3088                 .addr = addr,
3089                 .len = len,
3090         };
3091
3092         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3093         r = __kvm_io_bus_read(bus, &range, val);
3094         return r < 0 ? r : 0;
3095 }
3096
3097
3098 /* Caller must hold slots_lock. */
3099 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3100                             int len, struct kvm_io_device *dev)
3101 {
3102         struct kvm_io_bus *new_bus, *bus;
3103
3104         bus = kvm->buses[bus_idx];
3105         /* exclude ioeventfd which is limited by maximum fd */
3106         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3107                 return -ENOSPC;
3108
3109         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3110                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3111         if (!new_bus)
3112                 return -ENOMEM;
3113         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3114                sizeof(struct kvm_io_range)));
3115         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3116         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3117         synchronize_srcu_expedited(&kvm->srcu);
3118         kfree(bus);
3119
3120         return 0;
3121 }
3122
3123 /* Caller must hold slots_lock. */
3124 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3125                               struct kvm_io_device *dev)
3126 {
3127         int i, r;
3128         struct kvm_io_bus *new_bus, *bus;
3129
3130         bus = kvm->buses[bus_idx];
3131         r = -ENOENT;
3132         for (i = 0; i < bus->dev_count; i++)
3133                 if (bus->range[i].dev == dev) {
3134                         r = 0;
3135                         break;
3136                 }
3137
3138         if (r)
3139                 return r;
3140
3141         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3142                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3143         if (!new_bus)
3144                 return -ENOMEM;
3145
3146         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3147         new_bus->dev_count--;
3148         memcpy(new_bus->range + i, bus->range + i + 1,
3149                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3150
3151         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3152         synchronize_srcu_expedited(&kvm->srcu);
3153         kfree(bus);
3154         return r;
3155 }
3156
3157 static struct notifier_block kvm_cpu_notifier = {
3158         .notifier_call = kvm_cpu_hotplug,
3159 };
3160
3161 static int vm_stat_get(void *_offset, u64 *val)
3162 {
3163         unsigned offset = (long)_offset;
3164         struct kvm *kvm;
3165
3166         *val = 0;
3167         spin_lock(&kvm_lock);
3168         list_for_each_entry(kvm, &vm_list, vm_list)
3169                 *val += *(u32 *)((void *)kvm + offset);
3170         spin_unlock(&kvm_lock);
3171         return 0;
3172 }
3173
3174 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3175
3176 static int vcpu_stat_get(void *_offset, u64 *val)
3177 {
3178         unsigned offset = (long)_offset;
3179         struct kvm *kvm;
3180         struct kvm_vcpu *vcpu;
3181         int i;
3182
3183         *val = 0;
3184         spin_lock(&kvm_lock);
3185         list_for_each_entry(kvm, &vm_list, vm_list)
3186                 kvm_for_each_vcpu(i, vcpu, kvm)
3187                         *val += *(u32 *)((void *)vcpu + offset);
3188
3189         spin_unlock(&kvm_lock);
3190         return 0;
3191 }
3192
3193 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3194
3195 static const struct file_operations *stat_fops[] = {
3196         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3197         [KVM_STAT_VM]   = &vm_stat_fops,
3198 };
3199
3200 static int kvm_init_debug(void)
3201 {
3202         int r = -EEXIST;
3203         struct kvm_stats_debugfs_item *p;
3204
3205         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3206         if (kvm_debugfs_dir == NULL)
3207                 goto out;
3208
3209         for (p = debugfs_entries; p->name; ++p) {
3210                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3211                                                 (void *)(long)p->offset,
3212                                                 stat_fops[p->kind]);
3213                 if (p->dentry == NULL)
3214                         goto out_dir;
3215         }
3216
3217         return 0;
3218
3219 out_dir:
3220         debugfs_remove_recursive(kvm_debugfs_dir);
3221 out:
3222         return r;
3223 }
3224
3225 static void kvm_exit_debug(void)
3226 {
3227         struct kvm_stats_debugfs_item *p;
3228
3229         for (p = debugfs_entries; p->name; ++p)
3230                 debugfs_remove(p->dentry);
3231         debugfs_remove(kvm_debugfs_dir);
3232 }
3233
3234 static int kvm_suspend(void)
3235 {
3236         if (kvm_usage_count)
3237                 hardware_disable_nolock(NULL);
3238         return 0;
3239 }
3240
3241 static void kvm_resume(void)
3242 {
3243         if (kvm_usage_count) {
3244                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3245                 hardware_enable_nolock(NULL);
3246         }
3247 }
3248
3249 static struct syscore_ops kvm_syscore_ops = {
3250         .suspend = kvm_suspend,
3251         .resume = kvm_resume,
3252 };
3253
3254 static inline
3255 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3256 {
3257         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3258 }
3259
3260 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3261 {
3262         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3263         if (vcpu->preempted)
3264                 vcpu->preempted = false;
3265
3266         kvm_arch_sched_in(vcpu, cpu);
3267
3268         kvm_arch_vcpu_load(vcpu, cpu);
3269 }
3270
3271 static void kvm_sched_out(struct preempt_notifier *pn,
3272                           struct task_struct *next)
3273 {
3274         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3275
3276         if (current->state == TASK_RUNNING)
3277                 vcpu->preempted = true;
3278         kvm_arch_vcpu_put(vcpu);
3279 }
3280
3281 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3282                   struct module *module)
3283 {
3284         int r;
3285         int cpu;
3286
3287         r = kvm_arch_init(opaque);
3288         if (r)
3289                 goto out_fail;
3290
3291         /*
3292          * kvm_arch_init makes sure there's at most one caller
3293          * for architectures that support multiple implementations,
3294          * like intel and amd on x86.
3295          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3296          * conflicts in case kvm is already setup for another implementation.
3297          */
3298         r = kvm_irqfd_init();
3299         if (r)
3300                 goto out_irqfd;
3301
3302         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3303                 r = -ENOMEM;
3304                 goto out_free_0;
3305         }
3306
3307         r = kvm_arch_hardware_setup();
3308         if (r < 0)
3309                 goto out_free_0a;
3310
3311         for_each_online_cpu(cpu) {
3312                 smp_call_function_single(cpu,
3313                                 kvm_arch_check_processor_compat,
3314                                 &r, 1);
3315                 if (r < 0)
3316                         goto out_free_1;
3317         }
3318
3319         r = register_cpu_notifier(&kvm_cpu_notifier);
3320         if (r)
3321                 goto out_free_2;
3322         register_reboot_notifier(&kvm_reboot_notifier);
3323
3324         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3325         if (!vcpu_align)
3326                 vcpu_align = __alignof__(struct kvm_vcpu);
3327         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3328                                            0, NULL);
3329         if (!kvm_vcpu_cache) {
3330                 r = -ENOMEM;
3331                 goto out_free_3;
3332         }
3333
3334         r = kvm_async_pf_init();
3335         if (r)
3336                 goto out_free;
3337
3338         kvm_chardev_ops.owner = module;
3339         kvm_vm_fops.owner = module;
3340         kvm_vcpu_fops.owner = module;
3341
3342         r = misc_register(&kvm_dev);
3343         if (r) {
3344                 printk(KERN_ERR "kvm: misc device register failed\n");
3345                 goto out_unreg;
3346         }
3347
3348         register_syscore_ops(&kvm_syscore_ops);
3349
3350         kvm_preempt_ops.sched_in = kvm_sched_in;
3351         kvm_preempt_ops.sched_out = kvm_sched_out;
3352
3353         r = kvm_init_debug();
3354         if (r) {
3355                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3356                 goto out_undebugfs;
3357         }
3358
3359         r = kvm_vfio_ops_init();
3360         WARN_ON(r);
3361
3362         return 0;
3363
3364 out_undebugfs:
3365         unregister_syscore_ops(&kvm_syscore_ops);
3366         misc_deregister(&kvm_dev);
3367 out_unreg:
3368         kvm_async_pf_deinit();
3369 out_free:
3370         kmem_cache_destroy(kvm_vcpu_cache);
3371 out_free_3:
3372         unregister_reboot_notifier(&kvm_reboot_notifier);
3373         unregister_cpu_notifier(&kvm_cpu_notifier);
3374 out_free_2:
3375 out_free_1:
3376         kvm_arch_hardware_unsetup();
3377 out_free_0a:
3378         free_cpumask_var(cpus_hardware_enabled);
3379 out_free_0:
3380         kvm_irqfd_exit();
3381 out_irqfd:
3382         kvm_arch_exit();
3383 out_fail:
3384         return r;
3385 }
3386 EXPORT_SYMBOL_GPL(kvm_init);
3387
3388 void kvm_exit(void)
3389 {
3390         kvm_exit_debug();
3391         misc_deregister(&kvm_dev);
3392         kmem_cache_destroy(kvm_vcpu_cache);
3393         kvm_async_pf_deinit();
3394         unregister_syscore_ops(&kvm_syscore_ops);
3395         unregister_reboot_notifier(&kvm_reboot_notifier);
3396         unregister_cpu_notifier(&kvm_cpu_notifier);
3397         on_each_cpu(hardware_disable_nolock, NULL, 1);
3398         kvm_arch_hardware_unsetup();
3399         kvm_arch_exit();
3400         kvm_irqfd_exit();
3401         free_cpumask_var(cpus_hardware_enabled);
3402         kvm_vfio_ops_exit();
3403 }
3404 EXPORT_SYMBOL_GPL(kvm_exit);