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 unsigned int halt_poll_ns = 0;
  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 void kvm_kvfree(const void *addr)
 543 {
 544         if (is_vmalloc_addr(addr))
 545                 vfree(addr);
 546         else
 547                 kfree(addr);
 548 }
 549
 550 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 551 {
 552         if (!memslot->dirty_bitmap)
 553                 return;
 554
 555         kvm_kvfree(memslot->dirty_bitmap);
 556         memslot->dirty_bitmap = NULL;
 557 }
 558
 559 /*
 560  * Free any memory in @free but not in @dont.
 561  */
 562 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
 563                                   struct kvm_memory_slot *dont)
 564 {
 565         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 566                 kvm_destroy_dirty_bitmap(free);
 567
 568         kvm_arch_free_memslot(kvm, free, dont);
 569
 570         free->npages = 0;
 571 }
 572
 573 static void kvm_free_physmem(struct kvm *kvm)
 574 {
 575         struct kvm_memslots *slots = kvm->memslots;
 576         struct kvm_memory_slot *memslot;
 577
 578         kvm_for_each_memslot(memslot, slots)
 579                 kvm_free_physmem_slot(kvm, memslot, NULL);
 580
 581         kfree(kvm->memslots);
 582 }
 583
 584 static void kvm_destroy_devices(struct kvm *kvm)
 585 {
 586         struct list_head *node, *tmp;
 587
 588         list_for_each_safe(node, tmp, &kvm->devices) {
 589                 struct kvm_device *dev =
 590                         list_entry(node, struct kvm_device, vm_node);
 591
 592                 list_del(node);
 593                 dev->ops->destroy(dev);
 594         }
 595 }
 596
 597 static void kvm_destroy_vm(struct kvm *kvm)
 598 {
 599         int i;
 600         struct mm_struct *mm = kvm->mm;
 601
 602         kvm_arch_sync_events(kvm);
 603         spin_lock(&kvm_lock);
 604         list_del(&kvm->vm_list);
 605         spin_unlock(&kvm_lock);
 606         kvm_free_irq_routing(kvm);
 607         for (i = 0; i < KVM_NR_BUSES; i++)
 608                 kvm_io_bus_destroy(kvm->buses[i]);
 609         kvm_coalesced_mmio_free(kvm);
 610 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 611         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 612 #else
 613         kvm_arch_flush_shadow_all(kvm);
 614 #endif
 615         kvm_arch_destroy_vm(kvm);
 616         kvm_destroy_devices(kvm);
 617         kvm_free_physmem(kvm);
 618         cleanup_srcu_struct(&kvm->irq_srcu);
 619         cleanup_srcu_struct(&kvm->srcu);
 620         kvm_arch_free_vm(kvm);
 621         hardware_disable_all();
 622         mmdrop(mm);
 623 }
 624
 625 void kvm_get_kvm(struct kvm *kvm)
 626 {
 627         atomic_inc(&kvm->users_count);
 628 }
 629 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 630
 631 void kvm_put_kvm(struct kvm *kvm)
 632 {
 633         if (atomic_dec_and_test(&kvm->users_count))
 634                 kvm_destroy_vm(kvm);
 635 }
 636 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 637
 638
 639 static int kvm_vm_release(struct inode *inode, struct file *filp)
 640 {
 641         struct kvm *kvm = filp->private_data;
 642
 643         kvm_irqfd_release(kvm);
 644
 645         kvm_put_kvm(kvm);
 646         return 0;
 647 }
 648
 649 /*
 650  * Allocation size is twice as large as the actual dirty bitmap size.
 651  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 652  */
 653 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 654 {
 655         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 656
 657         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 658         if (!memslot->dirty_bitmap)
 659                 return -ENOMEM;
 660
 661         return 0;
 662 }
 663
 664 /*
 665  * Insert memslot and re-sort memslots based on their GFN,
 666  * so binary search could be used to lookup GFN.
 667  * Sorting algorithm takes advantage of having initially
 668  * sorted array and known changed memslot position.
 669  */
 670 static void update_memslots(struct kvm_memslots *slots,
 671                             struct kvm_memory_slot *new)
 672 {
 673         int id = new->id;
 674         int i = slots->id_to_index[id];
 675         struct kvm_memory_slot *mslots = slots->memslots;
 676
 677         WARN_ON(mslots[i].id != id);
 678         if (!new->npages) {
 679                 WARN_ON(!mslots[i].npages);
 680                 new->base_gfn = 0;
 681                 new->flags = 0;
 682                 if (mslots[i].npages)
 683                         slots->used_slots--;
 684         } else {
 685                 if (!mslots[i].npages)
 686                         slots->used_slots++;
 687         }
 688
 689         while (i < KVM_MEM_SLOTS_NUM - 1 &&
 690                new->base_gfn <= mslots[i + 1].base_gfn) {
 691                 if (!mslots[i + 1].npages)
 692                         break;
 693                 mslots[i] = mslots[i + 1];
 694                 slots->id_to_index[mslots[i].id] = i;
 695                 i++;
 696         }
 697
 698         /*
 699          * The ">=" is needed when creating a slot with base_gfn == 0,
 700          * so that it moves before all those with base_gfn == npages == 0.
 701          *
 702          * On the other hand, if new->npages is zero, the above loop has
 703          * already left i pointing to the beginning of the empty part of
 704          * mslots, and the ">=" would move the hole backwards in this
 705          * case---which is wrong.  So skip the loop when deleting a slot.
 706          */
 707         if (new->npages) {
 708                 while (i > 0 &&
 709                        new->base_gfn >= mslots[i - 1].base_gfn) {
 710                         mslots[i] = mslots[i - 1];
 711                         slots->id_to_index[mslots[i].id] = i;
 712                         i--;
 713                 }
 714         } else
 715                 WARN_ON_ONCE(i != slots->used_slots);
 716
 717         mslots[i] = *new;
 718         slots->id_to_index[mslots[i].id] = i;
 719 }
 720
 721 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 722 {
 723         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 724
 725 #ifdef __KVM_HAVE_READONLY_MEM
 726         valid_flags |= KVM_MEM_READONLY;
 727 #endif
 728
 729         if (mem->flags & ~valid_flags)
 730                 return -EINVAL;
 731
 732         return 0;
 733 }
 734
 735 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 736                 struct kvm_memslots *slots)
 737 {
 738         struct kvm_memslots *old_memslots = kvm->memslots;
 739
 740         /*
 741          * Set the low bit in the generation, which disables SPTE caching
 742          * until the end of synchronize_srcu_expedited.
 743          */
 744         WARN_ON(old_memslots->generation & 1);
 745         slots->generation = old_memslots->generation + 1;
 746
 747         rcu_assign_pointer(kvm->memslots, slots);
 748         synchronize_srcu_expedited(&kvm->srcu);
 749
 750         /*
 751          * Increment the new memslot generation a second time. This prevents
 752          * vm exits that race with memslot updates from caching a memslot
 753          * generation that will (potentially) be valid forever.
 754          */
 755         slots->generation++;
 756
 757         kvm_arch_memslots_updated(kvm);
 758
 759         return old_memslots;
 760 }
 761
 762 /*
 763  * Allocate some memory and give it an address in the guest physical address
 764  * space.
 765  *
 766  * Discontiguous memory is allowed, mostly for framebuffers.
 767  *
 768  * Must be called holding kvm->slots_lock for write.
 769  */
 770 int __kvm_set_memory_region(struct kvm *kvm,
 771                             struct kvm_userspace_memory_region *mem)
 772 {
 773         int r;
 774         gfn_t base_gfn;
 775         unsigned long npages;
 776         struct kvm_memory_slot *slot;
 777         struct kvm_memory_slot old, new;
 778         struct kvm_memslots *slots = NULL, *old_memslots;
 779         enum kvm_mr_change change;
 780
 781         r = check_memory_region_flags(mem);
 782         if (r)
 783                 goto out;
 784
 785         r = -EINVAL;
 786         /* General sanity checks */
 787         if (mem->memory_size & (PAGE_SIZE - 1))
 788                 goto out;
 789         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 790                 goto out;
 791         /* We can read the guest memory with __xxx_user() later on. */
 792         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 793             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 794              !access_ok(VERIFY_WRITE,
 795                         (void __user *)(unsigned long)mem->userspace_addr,
 796                         mem->memory_size)))
 797                 goto out;
 798         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 799                 goto out;
 800         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 801                 goto out;
 802
 803         slot = id_to_memslot(kvm->memslots, mem->slot);
 804         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 805         npages = mem->memory_size >> PAGE_SHIFT;
 806
 807         if (npages > KVM_MEM_MAX_NR_PAGES)
 808                 goto out;
 809
 810         if (!npages)
 811                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 812
 813         new = old = *slot;
 814
 815         new.id = mem->slot;
 816         new.base_gfn = base_gfn;
 817         new.npages = npages;
 818         new.flags = mem->flags;
 819
 820         if (npages) {
 821                 if (!old.npages)
 822                         change = KVM_MR_CREATE;
 823                 else { /* Modify an existing slot. */
 824                         if ((mem->userspace_addr != old.userspace_addr) ||
 825                             (npages != old.npages) ||
 826                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 827                                 goto out;
 828
 829                         if (base_gfn != old.base_gfn)
 830                                 change = KVM_MR_MOVE;
 831                         else if (new.flags != old.flags)
 832                                 change = KVM_MR_FLAGS_ONLY;
 833                         else { /* Nothing to change. */
 834                                 r = 0;
 835                                 goto out;
 836                         }
 837                 }
 838         } else if (old.npages) {
 839                 change = KVM_MR_DELETE;
 840         } else /* Modify a non-existent slot: disallowed. */
 841                 goto out;
 842
 843         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 844                 /* Check for overlaps */
 845                 r = -EEXIST;
 846                 kvm_for_each_memslot(slot, kvm->memslots) {
 847                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 848                             (slot->id == mem->slot))
 849                                 continue;
 850                         if (!((base_gfn + npages <= slot->base_gfn) ||
 851                               (base_gfn >= slot->base_gfn + slot->npages)))
 852                                 goto out;
 853                 }
 854         }
 855
 856         /* Free page dirty bitmap if unneeded */
 857         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 858                 new.dirty_bitmap = NULL;
 859
 860         r = -ENOMEM;
 861         if (change == KVM_MR_CREATE) {
 862                 new.userspace_addr = mem->userspace_addr;
 863
 864                 if (kvm_arch_create_memslot(kvm, &new, npages))
 865                         goto out_free;
 866         }
 867
 868         /* Allocate page dirty bitmap if needed */
 869         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 870                 if (kvm_create_dirty_bitmap(&new) < 0)
 871                         goto out_free;
 872         }
 873
 874         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 875                         GFP_KERNEL);
 876         if (!slots)
 877                 goto out_free;
 878
 879         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 880                 slot = id_to_memslot(slots, mem->slot);
 881                 slot->flags |= KVM_MEMSLOT_INVALID;
 882
 883                 old_memslots = install_new_memslots(kvm, slots);
 884
 885                 /* slot was deleted or moved, clear iommu mapping */
 886                 kvm_iommu_unmap_pages(kvm, &old);
 887                 /* From this point no new shadow pages pointing to a deleted,
 888                  * or moved, memslot will be created.
 889                  *
 890                  * validation of sp->gfn happens in:
 891                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 892                  *      - kvm_is_visible_gfn (mmu_check_roots)
 893                  */
 894                 kvm_arch_flush_shadow_memslot(kvm, slot);
 895
 896                 /*
 897                  * We can re-use the old_memslots from above, the only difference
 898                  * from the currently installed memslots is the invalid flag.  This
 899                  * will get overwritten by update_memslots anyway.
 900                  */
 901                 slots = old_memslots;
 902         }
 903
 904         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 905         if (r)
 906                 goto out_slots;
 907
 908         /* actual memory is freed via old in kvm_free_physmem_slot below */
 909         if (change == KVM_MR_DELETE) {
 910                 new.dirty_bitmap = NULL;
 911                 memset(&new.arch, 0, sizeof(new.arch));
 912         }
 913
 914         update_memslots(slots, &new);
 915         old_memslots = install_new_memslots(kvm, slots);
 916
 917         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 918
 919         kvm_free_physmem_slot(kvm, &old, &new);
 920         kfree(old_memslots);
 921
 922         /*
 923          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 924          * un-mapped and re-mapped if their base changes.  Since base change
 925          * unmapping is handled above with slot deletion, mapping alone is
 926          * needed here.  Anything else the iommu might care about for existing
 927          * slots (size changes, userspace addr changes and read-only flag
 928          * changes) is disallowed above, so any other attribute changes getting
 929          * here can be skipped.
 930          */
 931         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 932                 r = kvm_iommu_map_pages(kvm, &new);
 933                 return r;
 934         }
 935
 936         return 0;
 937
 938 out_slots:
 939         kfree(slots);
 940 out_free:
 941         kvm_free_physmem_slot(kvm, &new, &old);
 942 out:
 943         return r;
 944 }
 945 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 946
 947 int kvm_set_memory_region(struct kvm *kvm,
 948                           struct kvm_userspace_memory_region *mem)
 949 {
 950         int r;
 951
 952         mutex_lock(&kvm->slots_lock);
 953         r = __kvm_set_memory_region(kvm, mem);
 954         mutex_unlock(&kvm->slots_lock);
 955         return r;
 956 }
 957 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 958
 959 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 960                                           struct kvm_userspace_memory_region *mem)
 961 {
 962         if (mem->slot >= KVM_USER_MEM_SLOTS)
 963                 return -EINVAL;
 964         return kvm_set_memory_region(kvm, mem);
 965 }
 966
 967 int kvm_get_dirty_log(struct kvm *kvm,
 968                         struct kvm_dirty_log *log, int *is_dirty)
 969 {
 970         struct kvm_memory_slot *memslot;
 971         int r, i;
 972         unsigned long n;
 973         unsigned long any = 0;
 974
 975         r = -EINVAL;
 976         if (log->slot >= KVM_USER_MEM_SLOTS)
 977                 goto out;
 978
 979         memslot = id_to_memslot(kvm->memslots, log->slot);
 980         r = -ENOENT;
 981         if (!memslot->dirty_bitmap)
 982                 goto out;
 983
 984         n = kvm_dirty_bitmap_bytes(memslot);
 985
 986         for (i = 0; !any && i < n/sizeof(long); ++i)
 987                 any = memslot->dirty_bitmap[i];
 988
 989         r = -EFAULT;
 990         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 991                 goto out;
 992
 993         if (any)
 994                 *is_dirty = 1;
 995
 996         r = 0;
 997 out:
 998         return r;
 999 }
1000 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1001
1002 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1003 /**
1004  * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1005  *      are dirty write protect them for next write.
1006  * @kvm:        pointer to kvm instance
1007  * @log:        slot id and address to which we copy the log
1008  * @is_dirty:   flag set if any page is dirty
1009  *
1010  * We need to keep it in mind that VCPU threads can write to the bitmap
1011  * concurrently. So, to avoid losing track of dirty pages we keep the
1012  * following order:
1013  *
1014  *    1. Take a snapshot of the bit and clear it if needed.
1015  *    2. Write protect the corresponding page.
1016  *    3. Copy the snapshot to the userspace.
1017  *    4. Upon return caller flushes TLB's if needed.
1018  *
1019  * Between 2 and 4, the guest may write to the page using the remaining TLB
1020  * entry.  This is not a problem because the page is reported dirty using
1021  * the snapshot taken before and step 4 ensures that writes done after
1022  * exiting to userspace will be logged for the next call.
1023  *
1024  */
1025 int kvm_get_dirty_log_protect(struct kvm *kvm,
1026                         struct kvm_dirty_log *log, bool *is_dirty)
1027 {
1028         struct kvm_memory_slot *memslot;
1029         int r, i;
1030         unsigned long n;
1031         unsigned long *dirty_bitmap;
1032         unsigned long *dirty_bitmap_buffer;
1033
1034         r = -EINVAL;
1035         if (log->slot >= KVM_USER_MEM_SLOTS)
1036                 goto out;
1037
1038         memslot = id_to_memslot(kvm->memslots, log->slot);
1039
1040         dirty_bitmap = memslot->dirty_bitmap;
1041         r = -ENOENT;
1042         if (!dirty_bitmap)
1043                 goto out;
1044
1045         n = kvm_dirty_bitmap_bytes(memslot);
1046
1047         dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1048         memset(dirty_bitmap_buffer, 0, n);
1049
1050         spin_lock(&kvm->mmu_lock);
1051         *is_dirty = false;
1052         for (i = 0; i < n / sizeof(long); i++) {
1053                 unsigned long mask;
1054                 gfn_t offset;
1055
1056                 if (!dirty_bitmap[i])
1057                         continue;
1058
1059                 *is_dirty = true;
1060
1061                 mask = xchg(&dirty_bitmap[i], 0);
1062                 dirty_bitmap_buffer[i] = mask;
1063
1064                 offset = i * BITS_PER_LONG;
1065                 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, offset,
1066                                                                 mask);
1067         }
1068
1069         spin_unlock(&kvm->mmu_lock);
1070
1071         r = -EFAULT;
1072         if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1073                 goto out;
1074
1075         r = 0;
1076 out:
1077         return r;
1078 }
1079 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1080 #endif
1081
1082 bool kvm_largepages_enabled(void)
1083 {
1084         return largepages_enabled;
1085 }
1086
1087 void kvm_disable_largepages(void)
1088 {
1089         largepages_enabled = false;
1090 }
1091 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1092
1093 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1094 {
1095         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1096 }
1097 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1098
1099 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1100 {
1101         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1102
1103         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1104               memslot->flags & KVM_MEMSLOT_INVALID)
1105                 return 0;
1106
1107         return 1;
1108 }
1109 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1110
1111 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1112 {
1113         struct vm_area_struct *vma;
1114         unsigned long addr, size;
1115
1116         size = PAGE_SIZE;
1117
1118         addr = gfn_to_hva(kvm, gfn);
1119         if (kvm_is_error_hva(addr))
1120                 return PAGE_SIZE;
1121
1122         down_read(&current->mm->mmap_sem);
1123         vma = find_vma(current->mm, addr);
1124         if (!vma)
1125                 goto out;
1126
1127         size = vma_kernel_pagesize(vma);
1128
1129 out:
1130         up_read(&current->mm->mmap_sem);
1131
1132         return size;
1133 }
1134
1135 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1136 {
1137         return slot->flags & KVM_MEM_READONLY;
1138 }
1139
1140 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1141                                        gfn_t *nr_pages, bool write)
1142 {
1143         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1144                 return KVM_HVA_ERR_BAD;
1145
1146         if (memslot_is_readonly(slot) && write)
1147                 return KVM_HVA_ERR_RO_BAD;
1148
1149         if (nr_pages)
1150                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1151
1152         return __gfn_to_hva_memslot(slot, gfn);
1153 }
1154
1155 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1156                                      gfn_t *nr_pages)
1157 {
1158         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1159 }
1160
1161 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1162                                         gfn_t gfn)
1163 {
1164         return gfn_to_hva_many(slot, gfn, NULL);
1165 }
1166 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1167
1168 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1169 {
1170         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1171 }
1172 EXPORT_SYMBOL_GPL(gfn_to_hva);
1173
1174 /*
1175  * If writable is set to false, the hva returned by this function is only
1176  * allowed to be read.
1177  */
1178 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1179                                       gfn_t gfn, bool *writable)
1180 {
1181         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1182
1183         if (!kvm_is_error_hva(hva) && writable)
1184                 *writable = !memslot_is_readonly(slot);
1185
1186         return hva;
1187 }
1188
1189 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1190 {
1191         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1192
1193         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1194 }
1195
1196 static int kvm_read_hva(void *data, void __user *hva, int len)
1197 {
1198         return __copy_from_user(data, hva, len);
1199 }
1200
1201 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1202 {
1203         return __copy_from_user_inatomic(data, hva, len);
1204 }
1205
1206 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1207         unsigned long start, int write, struct page **page)
1208 {
1209         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1210
1211         if (write)
1212                 flags |= FOLL_WRITE;
1213
1214         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1215 }
1216
1217 static inline int check_user_page_hwpoison(unsigned long addr)
1218 {
1219         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1220
1221         rc = __get_user_pages(current, current->mm, addr, 1,
1222                               flags, NULL, NULL, NULL);
1223         return rc == -EHWPOISON;
1224 }
1225
1226 /*
1227  * The atomic path to get the writable pfn which will be stored in @pfn,
1228  * true indicates success, otherwise false is returned.
1229  */
1230 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1231                             bool write_fault, bool *writable, pfn_t *pfn)
1232 {
1233         struct page *page[1];
1234         int npages;
1235
1236         if (!(async || atomic))
1237                 return false;
1238
1239         /*
1240          * Fast pin a writable pfn only if it is a write fault request
1241          * or the caller allows to map a writable pfn for a read fault
1242          * request.
1243          */
1244         if (!(write_fault || writable))
1245                 return false;
1246
1247         npages = __get_user_pages_fast(addr, 1, 1, page);
1248         if (npages == 1) {
1249                 *pfn = page_to_pfn(page[0]);
1250
1251                 if (writable)
1252                         *writable = true;
1253                 return true;
1254         }
1255
1256         return false;
1257 }
1258
1259 /*
1260  * The slow path to get the pfn of the specified host virtual address,
1261  * 1 indicates success, -errno is returned if error is detected.
1262  */
1263 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1264                            bool *writable, pfn_t *pfn)
1265 {
1266         struct page *page[1];
1267         int npages = 0;
1268
1269         might_sleep();
1270
1271         if (writable)
1272                 *writable = write_fault;
1273
1274         if (async) {
1275                 down_read(&current->mm->mmap_sem);
1276                 npages = get_user_page_nowait(current, current->mm,
1277                                               addr, write_fault, page);
1278                 up_read(&current->mm->mmap_sem);
1279         } else
1280                 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1281                                                    write_fault, 0, page,
1282                                                    FOLL_TOUCH|FOLL_HWPOISON);
1283         if (npages != 1)
1284                 return npages;
1285
1286         /* map read fault as writable if possible */
1287         if (unlikely(!write_fault) && writable) {
1288                 struct page *wpage[1];
1289
1290                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1291                 if (npages == 1) {
1292                         *writable = true;
1293                         put_page(page[0]);
1294                         page[0] = wpage[0];
1295                 }
1296
1297                 npages = 1;
1298         }
1299         *pfn = page_to_pfn(page[0]);
1300         return npages;
1301 }
1302
1303 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1304 {
1305         if (unlikely(!(vma->vm_flags & VM_READ)))
1306                 return false;
1307
1308         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1309                 return false;
1310
1311         return true;
1312 }
1313
1314 /*
1315  * Pin guest page in memory and return its pfn.
1316  * @addr: host virtual address which maps memory to the guest
1317  * @atomic: whether this function can sleep
1318  * @async: whether this function need to wait IO complete if the
1319  *         host page is not in the memory
1320  * @write_fault: whether we should get a writable host page
1321  * @writable: whether it allows to map a writable host page for !@write_fault
1322  *
1323  * The function will map a writable host page for these two cases:
1324  * 1): @write_fault = true
1325  * 2): @write_fault = false && @writable, @writable will tell the caller
1326  *     whether the mapping is writable.
1327  */
1328 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1329                         bool write_fault, bool *writable)
1330 {
1331         struct vm_area_struct *vma;
1332         pfn_t pfn = 0;
1333         int npages;
1334
1335         /* we can do it either atomically or asynchronously, not both */
1336         BUG_ON(atomic && async);
1337
1338         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1339                 return pfn;
1340
1341         if (atomic)
1342                 return KVM_PFN_ERR_FAULT;
1343
1344         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1345         if (npages == 1)
1346                 return pfn;
1347
1348         down_read(&current->mm->mmap_sem);
1349         if (npages == -EHWPOISON ||
1350               (!async && check_user_page_hwpoison(addr))) {
1351                 pfn = KVM_PFN_ERR_HWPOISON;
1352                 goto exit;
1353         }
1354
1355         vma = find_vma_intersection(current->mm, addr, addr + 1);
1356
1357         if (vma == NULL)
1358                 pfn = KVM_PFN_ERR_FAULT;
1359         else if ((vma->vm_flags & VM_PFNMAP)) {
1360                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1361                         vma->vm_pgoff;
1362                 BUG_ON(!kvm_is_reserved_pfn(pfn));
1363         } else {
1364                 if (async && vma_is_valid(vma, write_fault))
1365                         *async = true;
1366                 pfn = KVM_PFN_ERR_FAULT;
1367         }
1368 exit:
1369         up_read(&current->mm->mmap_sem);
1370         return pfn;
1371 }
1372
1373 static pfn_t
1374 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1375                      bool *async, bool write_fault, bool *writable)
1376 {
1377         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1378
1379         if (addr == KVM_HVA_ERR_RO_BAD)
1380                 return KVM_PFN_ERR_RO_FAULT;
1381
1382         if (kvm_is_error_hva(addr))
1383                 return KVM_PFN_NOSLOT;
1384
1385         /* Do not map writable pfn in the readonly memslot. */
1386         if (writable && memslot_is_readonly(slot)) {
1387                 *writable = false;
1388                 writable = NULL;
1389         }
1390
1391         return hva_to_pfn(addr, atomic, async, write_fault,
1392                           writable);
1393 }
1394
1395 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1396                           bool write_fault, bool *writable)
1397 {
1398         struct kvm_memory_slot *slot;
1399
1400         if (async)
1401                 *async = false;
1402
1403         slot = gfn_to_memslot(kvm, gfn);
1404
1405         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1406                                     writable);
1407 }
1408
1409 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1410 {
1411         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1412 }
1413 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1414
1415 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1416                        bool write_fault, bool *writable)
1417 {
1418         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1419 }
1420 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1421
1422 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1423 {
1424         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1425 }
1426 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1427
1428 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1429                       bool *writable)
1430 {
1431         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1432 }
1433 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1434
1435 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1436 {
1437         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1438 }
1439
1440 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1441 {
1442         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1443 }
1444 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1445
1446 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1447                                                                   int nr_pages)
1448 {
1449         unsigned long addr;
1450         gfn_t entry;
1451
1452         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1453         if (kvm_is_error_hva(addr))
1454                 return -1;
1455
1456         if (entry < nr_pages)
1457                 return 0;
1458
1459         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1460 }
1461 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1462
1463 static struct page *kvm_pfn_to_page(pfn_t pfn)
1464 {
1465         if (is_error_noslot_pfn(pfn))
1466                 return KVM_ERR_PTR_BAD_PAGE;
1467
1468         if (kvm_is_reserved_pfn(pfn)) {
1469                 WARN_ON(1);
1470                 return KVM_ERR_PTR_BAD_PAGE;
1471         }
1472
1473         return pfn_to_page(pfn);
1474 }
1475
1476 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1477 {
1478         pfn_t pfn;
1479
1480         pfn = gfn_to_pfn(kvm, gfn);
1481
1482         return kvm_pfn_to_page(pfn);
1483 }
1484
1485 EXPORT_SYMBOL_GPL(gfn_to_page);
1486
1487 void kvm_release_page_clean(struct page *page)
1488 {
1489         WARN_ON(is_error_page(page));
1490
1491         kvm_release_pfn_clean(page_to_pfn(page));
1492 }
1493 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1494
1495 void kvm_release_pfn_clean(pfn_t pfn)
1496 {
1497         if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1498                 put_page(pfn_to_page(pfn));
1499 }
1500 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1501
1502 void kvm_release_page_dirty(struct page *page)
1503 {
1504         WARN_ON(is_error_page(page));
1505
1506         kvm_release_pfn_dirty(page_to_pfn(page));
1507 }
1508 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1509
1510 static void kvm_release_pfn_dirty(pfn_t pfn)
1511 {
1512         kvm_set_pfn_dirty(pfn);
1513         kvm_release_pfn_clean(pfn);
1514 }
1515
1516 void kvm_set_pfn_dirty(pfn_t pfn)
1517 {
1518         if (!kvm_is_reserved_pfn(pfn)) {
1519                 struct page *page = pfn_to_page(pfn);
1520                 if (!PageReserved(page))
1521                         SetPageDirty(page);
1522         }
1523 }
1524 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1525
1526 void kvm_set_pfn_accessed(pfn_t pfn)
1527 {
1528         if (!kvm_is_reserved_pfn(pfn))
1529                 mark_page_accessed(pfn_to_page(pfn));
1530 }
1531 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1532
1533 void kvm_get_pfn(pfn_t pfn)
1534 {
1535         if (!kvm_is_reserved_pfn(pfn))
1536                 get_page(pfn_to_page(pfn));
1537 }
1538 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1539
1540 static int next_segment(unsigned long len, int offset)
1541 {
1542         if (len > PAGE_SIZE - offset)
1543                 return PAGE_SIZE - offset;
1544         else
1545                 return len;
1546 }
1547
1548 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1549                         int len)
1550 {
1551         int r;
1552         unsigned long addr;
1553
1554         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1555         if (kvm_is_error_hva(addr))
1556                 return -EFAULT;
1557         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1558         if (r)
1559                 return -EFAULT;
1560         return 0;
1561 }
1562 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1563
1564 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1565 {
1566         gfn_t gfn = gpa >> PAGE_SHIFT;
1567         int seg;
1568         int offset = offset_in_page(gpa);
1569         int ret;
1570
1571         while ((seg = next_segment(len, offset)) != 0) {
1572                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1573                 if (ret < 0)
1574                         return ret;
1575                 offset = 0;
1576                 len -= seg;
1577                 data += seg;
1578                 ++gfn;
1579         }
1580         return 0;
1581 }
1582 EXPORT_SYMBOL_GPL(kvm_read_guest);
1583
1584 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1585                           unsigned long len)
1586 {
1587         int r;
1588         unsigned long addr;
1589         gfn_t gfn = gpa >> PAGE_SHIFT;
1590         int offset = offset_in_page(gpa);
1591
1592         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1593         if (kvm_is_error_hva(addr))
1594                 return -EFAULT;
1595         pagefault_disable();
1596         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1597         pagefault_enable();
1598         if (r)
1599                 return -EFAULT;
1600         return 0;
1601 }
1602 EXPORT_SYMBOL(kvm_read_guest_atomic);
1603
1604 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1605                          int offset, int len)
1606 {
1607         int r;
1608         unsigned long addr;
1609
1610         addr = gfn_to_hva(kvm, gfn);
1611         if (kvm_is_error_hva(addr))
1612                 return -EFAULT;
1613         r = __copy_to_user((void __user *)addr + offset, data, len);
1614         if (r)
1615                 return -EFAULT;
1616         mark_page_dirty(kvm, gfn);
1617         return 0;
1618 }
1619 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1620
1621 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1622                     unsigned long len)
1623 {
1624         gfn_t gfn = gpa >> PAGE_SHIFT;
1625         int seg;
1626         int offset = offset_in_page(gpa);
1627         int ret;
1628
1629         while ((seg = next_segment(len, offset)) != 0) {
1630                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1631                 if (ret < 0)
1632                         return ret;
1633                 offset = 0;
1634                 len -= seg;
1635                 data += seg;
1636                 ++gfn;
1637         }
1638         return 0;
1639 }
1640 EXPORT_SYMBOL_GPL(kvm_write_guest);
1641
1642 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1643                               gpa_t gpa, unsigned long len)
1644 {
1645         struct kvm_memslots *slots = kvm_memslots(kvm);
1646         int offset = offset_in_page(gpa);
1647         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1648         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1649         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1650         gfn_t nr_pages_avail;
1651
1652         ghc->gpa = gpa;
1653         ghc->generation = slots->generation;
1654         ghc->len = len;
1655         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1656         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1657         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1658                 ghc->hva += offset;
1659         } else {
1660                 /*
1661                  * If the requested region crosses two memslots, we still
1662                  * verify that the entire region is valid here.
1663                  */
1664                 while (start_gfn <= end_gfn) {
1665                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1666                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1667                                                    &nr_pages_avail);
1668                         if (kvm_is_error_hva(ghc->hva))
1669                                 return -EFAULT;
1670                         start_gfn += nr_pages_avail;
1671                 }
1672                 /* Use the slow path for cross page reads and writes. */
1673                 ghc->memslot = NULL;
1674         }
1675         return 0;
1676 }
1677 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1678
1679 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1680                            void *data, unsigned long len)
1681 {
1682         struct kvm_memslots *slots = kvm_memslots(kvm);
1683         int r;
1684
1685         BUG_ON(len > ghc->len);
1686
1687         if (slots->generation != ghc->generation)
1688                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1689
1690         if (unlikely(!ghc->memslot))
1691                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1692
1693         if (kvm_is_error_hva(ghc->hva))
1694                 return -EFAULT;
1695
1696         r = __copy_to_user((void __user *)ghc->hva, data, len);
1697         if (r)
1698                 return -EFAULT;
1699         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1700
1701         return 0;
1702 }
1703 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1704
1705 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1706                            void *data, unsigned long len)
1707 {
1708         struct kvm_memslots *slots = kvm_memslots(kvm);
1709         int r;
1710
1711         BUG_ON(len > ghc->len);
1712
1713         if (slots->generation != ghc->generation)
1714                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1715
1716         if (unlikely(!ghc->memslot))
1717                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1718
1719         if (kvm_is_error_hva(ghc->hva))
1720                 return -EFAULT;
1721
1722         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1723         if (r)
1724                 return -EFAULT;
1725
1726         return 0;
1727 }
1728 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1729
1730 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1731 {
1732         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1733
1734         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1735 }
1736 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1737
1738 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1739 {
1740         gfn_t gfn = gpa >> PAGE_SHIFT;
1741         int seg;
1742         int offset = offset_in_page(gpa);
1743         int ret;
1744
1745         while ((seg = next_segment(len, offset)) != 0) {
1746                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1747                 if (ret < 0)
1748                         return ret;
1749                 offset = 0;
1750                 len -= seg;
1751                 ++gfn;
1752         }
1753         return 0;
1754 }
1755 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1756
1757 static void mark_page_dirty_in_slot(struct kvm *kvm,
1758                                     struct kvm_memory_slot *memslot,
1759                                     gfn_t gfn)
1760 {
1761         if (memslot && memslot->dirty_bitmap) {
1762                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1763
1764                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1765         }
1766 }
1767
1768 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1769 {
1770         struct kvm_memory_slot *memslot;
1771
1772         memslot = gfn_to_memslot(kvm, gfn);
1773         mark_page_dirty_in_slot(kvm, memslot, gfn);
1774 }
1775 EXPORT_SYMBOL_GPL(mark_page_dirty);
1776
1777 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1778 {
1779         if (kvm_arch_vcpu_runnable(vcpu)) {
1780                 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1781                 return -EINTR;
1782         }
1783         if (kvm_cpu_has_pending_timer(vcpu))
1784                 return -EINTR;
1785         if (signal_pending(current))
1786                 return -EINTR;
1787
1788         return 0;
1789 }
1790
1791 /*
1792  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1793  */
1794 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1795 {
1796         ktime_t start, cur;
1797         DEFINE_WAIT(wait);
1798         bool waited = false;
1799
1800         start = cur = ktime_get();
1801         if (halt_poll_ns) {
1802                 ktime_t stop = ktime_add_ns(ktime_get(), halt_poll_ns);
1803                 do {
1804                         /*
1805                          * This sets KVM_REQ_UNHALT if an interrupt
1806                          * arrives.
1807                          */
1808                         if (kvm_vcpu_check_block(vcpu) < 0) {
1809                                 ++vcpu->stat.halt_successful_poll;
1810                                 goto out;
1811                         }
1812                         cur = ktime_get();
1813                 } while (single_task_running() && ktime_before(cur, stop));
1814         }
1815
1816         for (;;) {
1817                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1818
1819                 if (kvm_vcpu_check_block(vcpu) < 0)
1820                         break;
1821
1822                 waited = true;
1823                 schedule();
1824         }
1825
1826         finish_wait(&vcpu->wq, &wait);
1827         cur = ktime_get();
1828
1829 out:
1830         trace_kvm_vcpu_wakeup(ktime_to_ns(cur) - ktime_to_ns(start), waited);
1831 }
1832 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1833
1834 #ifndef CONFIG_S390
1835 /*
1836  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1837  */
1838 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1839 {
1840         int me;
1841         int cpu = vcpu->cpu;
1842         wait_queue_head_t *wqp;
1843
1844         wqp = kvm_arch_vcpu_wq(vcpu);
1845         if (waitqueue_active(wqp)) {
1846                 wake_up_interruptible(wqp);
1847                 ++vcpu->stat.halt_wakeup;
1848         }
1849
1850         me = get_cpu();
1851         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1852                 if (kvm_arch_vcpu_should_kick(vcpu))
1853                         smp_send_reschedule(cpu);
1854         put_cpu();
1855 }
1856 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1857 #endif /* !CONFIG_S390 */
1858
1859 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1860 {
1861         struct pid *pid;
1862         struct task_struct *task = NULL;
1863         int ret = 0;
1864
1865         rcu_read_lock();
1866         pid = rcu_dereference(target->pid);
1867         if (pid)
1868                 task = get_pid_task(pid, PIDTYPE_PID);
1869         rcu_read_unlock();
1870         if (!task)
1871                 return ret;
1872         ret = yield_to(task, 1);
1873         put_task_struct(task);
1874
1875         return ret;
1876 }
1877 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1878
1879 /*
1880  * Helper that checks whether a VCPU is eligible for directed yield.
1881  * Most eligible candidate to yield is decided by following heuristics:
1882  *
1883  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1884  *  (preempted lock holder), indicated by @in_spin_loop.
1885  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1886  *
1887  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1888  *  chance last time (mostly it has become eligible now since we have probably
1889  *  yielded to lockholder in last iteration. This is done by toggling
1890  *  @dy_eligible each time a VCPU checked for eligibility.)
1891  *
1892  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1893  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1894  *  burning. Giving priority for a potential lock-holder increases lock
1895  *  progress.
1896  *
1897  *  Since algorithm is based on heuristics, accessing another VCPU data without
1898  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1899  *  and continue with next VCPU and so on.
1900  */
1901 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1902 {
1903 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1904         bool eligible;
1905
1906         eligible = !vcpu->spin_loop.in_spin_loop ||
1907                     vcpu->spin_loop.dy_eligible;
1908
1909         if (vcpu->spin_loop.in_spin_loop)
1910                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1911
1912         return eligible;
1913 #else
1914         return true;
1915 #endif
1916 }
1917
1918 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1919 {
1920         struct kvm *kvm = me->kvm;
1921         struct kvm_vcpu *vcpu;
1922         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1923         int yielded = 0;
1924         int try = 3;
1925         int pass;
1926         int i;
1927
1928         kvm_vcpu_set_in_spin_loop(me, true);
1929         /*
1930          * We boost the priority of a VCPU that is runnable but not
1931          * currently running, because it got preempted by something
1932          * else and called schedule in __vcpu_run.  Hopefully that
1933          * VCPU is holding the lock that we need and will release it.
1934          * We approximate round-robin by starting at the last boosted VCPU.
1935          */
1936         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1937                 kvm_for_each_vcpu(i, vcpu, kvm) {
1938                         if (!pass && i <= last_boosted_vcpu) {
1939                                 i = last_boosted_vcpu;
1940                                 continue;
1941                         } else if (pass && i > last_boosted_vcpu)
1942                                 break;
1943                         if (!ACCESS_ONCE(vcpu->preempted))
1944                                 continue;
1945                         if (vcpu == me)
1946                                 continue;
1947                         if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1948                                 continue;
1949                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1950                                 continue;
1951
1952                         yielded = kvm_vcpu_yield_to(vcpu);
1953                         if (yielded > 0) {
1954                                 kvm->last_boosted_vcpu = i;
1955                                 break;
1956                         } else if (yielded < 0) {
1957                                 try--;
1958                                 if (!try)
1959                                         break;
1960                         }
1961                 }
1962         }
1963         kvm_vcpu_set_in_spin_loop(me, false);
1964
1965         /* Ensure vcpu is not eligible during next spinloop */
1966         kvm_vcpu_set_dy_eligible(me, false);
1967 }
1968 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1969
1970 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1971 {
1972         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1973         struct page *page;
1974
1975         if (vmf->pgoff == 0)
1976                 page = virt_to_page(vcpu->run);
1977 #ifdef CONFIG_X86
1978         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1979                 page = virt_to_page(vcpu->arch.pio_data);
1980 #endif
1981 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1982         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1983                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1984 #endif
1985         else
1986                 return kvm_arch_vcpu_fault(vcpu, vmf);
1987         get_page(page);
1988         vmf->page = page;
1989         return 0;
1990 }
1991
1992 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1993         .fault = kvm_vcpu_fault,
1994 };
1995
1996 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1997 {
1998         vma->vm_ops = &kvm_vcpu_vm_ops;
1999         return 0;
2000 }
2001
2002 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2003 {
2004         struct kvm_vcpu *vcpu = filp->private_data;
2005
2006         kvm_put_kvm(vcpu->kvm);
2007         return 0;
2008 }
2009
2010 static struct file_operations kvm_vcpu_fops = {
2011         .release        = kvm_vcpu_release,
2012         .unlocked_ioctl = kvm_vcpu_ioctl,
2013 #ifdef CONFIG_KVM_COMPAT
2014         .compat_ioctl   = kvm_vcpu_compat_ioctl,
2015 #endif
2016         .mmap           = kvm_vcpu_mmap,
2017         .llseek         = noop_llseek,
2018 };
2019
2020 /*
2021  * Allocates an inode for the vcpu.
2022  */
2023 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2024 {
2025         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2026 }
2027
2028 /*
2029  * Creates some virtual cpus.  Good luck creating more than one.
2030  */
2031 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2032 {
2033         int r;
2034         struct kvm_vcpu *vcpu, *v;
2035
2036         if (id >= KVM_MAX_VCPUS)
2037                 return -EINVAL;
2038
2039         vcpu = kvm_arch_vcpu_create(kvm, id);
2040         if (IS_ERR(vcpu))
2041                 return PTR_ERR(vcpu);
2042
2043         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2044
2045         r = kvm_arch_vcpu_setup(vcpu);
2046         if (r)
2047                 goto vcpu_destroy;
2048
2049         mutex_lock(&kvm->lock);
2050         if (!kvm_vcpu_compatible(vcpu)) {
2051                 r = -EINVAL;
2052                 goto unlock_vcpu_destroy;
2053         }
2054         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2055                 r = -EINVAL;
2056                 goto unlock_vcpu_destroy;
2057         }
2058
2059         kvm_for_each_vcpu(r, v, kvm)
2060                 if (v->vcpu_id == id) {
2061                         r = -EEXIST;
2062                         goto unlock_vcpu_destroy;
2063                 }
2064
2065         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2066
2067         /* Now it's all set up, let userspace reach it */
2068         kvm_get_kvm(kvm);
2069         r = create_vcpu_fd(vcpu);
2070         if (r < 0) {
2071                 kvm_put_kvm(kvm);
2072                 goto unlock_vcpu_destroy;
2073         }
2074
2075         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2076         smp_wmb();
2077         atomic_inc(&kvm->online_vcpus);
2078
2079         mutex_unlock(&kvm->lock);
2080         kvm_arch_vcpu_postcreate(vcpu);
2081         return r;
2082
2083 unlock_vcpu_destroy:
2084         mutex_unlock(&kvm->lock);
2085 vcpu_destroy:
2086         kvm_arch_vcpu_destroy(vcpu);
2087         return r;
2088 }
2089
2090 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2091 {
2092         if (sigset) {
2093                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2094                 vcpu->sigset_active = 1;
2095                 vcpu->sigset = *sigset;
2096         } else
2097                 vcpu->sigset_active = 0;
2098         return 0;
2099 }
2100
2101 static long kvm_vcpu_ioctl(struct file *filp,
2102                            unsigned int ioctl, unsigned long arg)
2103 {
2104         struct kvm_vcpu *vcpu = filp->private_data;
2105         void __user *argp = (void __user *)arg;
2106         int r;
2107         struct kvm_fpu *fpu = NULL;
2108         struct kvm_sregs *kvm_sregs = NULL;
2109
2110         if (vcpu->kvm->mm != current->mm)
2111                 return -EIO;
2112
2113         if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2114                 return -EINVAL;
2115
2116 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2117         /*
2118          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2119          * so vcpu_load() would break it.
2120          */
2121         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
2122                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2123 #endif
2124
2125
2126         r = vcpu_load(vcpu);
2127         if (r)
2128                 return r;
2129         switch (ioctl) {
2130         case KVM_RUN:
2131                 r = -EINVAL;
2132                 if (arg)
2133                         goto out;
2134                 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2135                         /* The thread running this VCPU changed. */
2136                         struct pid *oldpid = vcpu->pid;
2137                         struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2138                         rcu_assign_pointer(vcpu->pid, newpid);
2139                         if (oldpid)
2140                                 synchronize_rcu();
2141                         put_pid(oldpid);
2142                 }
2143                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2144                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2145                 break;
2146         case KVM_GET_REGS: {
2147                 struct kvm_regs *kvm_regs;
2148
2149                 r = -ENOMEM;
2150                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2151                 if (!kvm_regs)
2152                         goto out;
2153                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2154                 if (r)
2155                         goto out_free1;
2156                 r = -EFAULT;
2157                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2158                         goto out_free1;
2159                 r = 0;
2160 out_free1:
2161                 kfree(kvm_regs);
2162                 break;
2163         }
2164         case KVM_SET_REGS: {
2165                 struct kvm_regs *kvm_regs;
2166
2167                 r = -ENOMEM;
2168                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2169                 if (IS_ERR(kvm_regs)) {
2170                         r = PTR_ERR(kvm_regs);
2171                         goto out;
2172                 }
2173                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2174                 kfree(kvm_regs);
2175                 break;
2176         }
2177         case KVM_GET_SREGS: {
2178                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2179                 r = -ENOMEM;
2180                 if (!kvm_sregs)
2181                         goto out;
2182                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2183                 if (r)
2184                         goto out;
2185                 r = -EFAULT;
2186                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2187                         goto out;
2188                 r = 0;
2189                 break;
2190         }
2191         case KVM_SET_SREGS: {
2192                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2193                 if (IS_ERR(kvm_sregs)) {
2194                         r = PTR_ERR(kvm_sregs);
2195                         kvm_sregs = NULL;
2196                         goto out;
2197                 }
2198                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2199                 break;
2200         }
2201         case KVM_GET_MP_STATE: {
2202                 struct kvm_mp_state mp_state;
2203
2204                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2205                 if (r)
2206                         goto out;
2207                 r = -EFAULT;
2208                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2209                         goto out;
2210                 r = 0;
2211                 break;
2212         }
2213         case KVM_SET_MP_STATE: {
2214                 struct kvm_mp_state mp_state;
2215
2216                 r = -EFAULT;
2217                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2218                         goto out;
2219                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2220                 break;
2221         }
2222         case KVM_TRANSLATE: {
2223                 struct kvm_translation tr;
2224
2225                 r = -EFAULT;
2226                 if (copy_from_user(&tr, argp, sizeof tr))
2227                         goto out;
2228                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2229                 if (r)
2230                         goto out;
2231                 r = -EFAULT;
2232                 if (copy_to_user(argp, &tr, sizeof tr))
2233                         goto out;
2234                 r = 0;
2235                 break;
2236         }
2237         case KVM_SET_GUEST_DEBUG: {
2238                 struct kvm_guest_debug dbg;
2239
2240                 r = -EFAULT;
2241                 if (copy_from_user(&dbg, argp, sizeof dbg))
2242                         goto out;
2243                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2244                 break;
2245         }
2246         case KVM_SET_SIGNAL_MASK: {
2247                 struct kvm_signal_mask __user *sigmask_arg = argp;
2248                 struct kvm_signal_mask kvm_sigmask;
2249                 sigset_t sigset, *p;
2250
2251                 p = NULL;
2252                 if (argp) {
2253                         r = -EFAULT;
2254                         if (copy_from_user(&kvm_sigmask, argp,
2255                                            sizeof kvm_sigmask))
2256                                 goto out;
2257                         r = -EINVAL;
2258                         if (kvm_sigmask.len != sizeof sigset)
2259                                 goto out;
2260                         r = -EFAULT;
2261                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2262                                            sizeof sigset))
2263                                 goto out;
2264                         p = &sigset;
2265                 }
2266                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2267                 break;
2268         }
2269         case KVM_GET_FPU: {
2270                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2271                 r = -ENOMEM;
2272                 if (!fpu)
2273                         goto out;
2274                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2275                 if (r)
2276                         goto out;
2277                 r = -EFAULT;
2278                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2279                         goto out;
2280                 r = 0;
2281                 break;
2282         }
2283         case KVM_SET_FPU: {
2284                 fpu = memdup_user(argp, sizeof(*fpu));
2285                 if (IS_ERR(fpu)) {
2286                         r = PTR_ERR(fpu);
2287                         fpu = NULL;
2288                         goto out;
2289                 }
2290                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2291                 break;
2292         }
2293         default:
2294                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2295         }
2296 out:
2297         vcpu_put(vcpu);
2298         kfree(fpu);
2299         kfree(kvm_sregs);
2300         return r;
2301 }
2302
2303 #ifdef CONFIG_KVM_COMPAT
2304 static long kvm_vcpu_compat_ioctl(struct file *filp,
2305                                   unsigned int ioctl, unsigned long arg)
2306 {
2307         struct kvm_vcpu *vcpu = filp->private_data;
2308         void __user *argp = compat_ptr(arg);
2309         int r;
2310
2311         if (vcpu->kvm->mm != current->mm)
2312                 return -EIO;
2313
2314         switch (ioctl) {
2315         case KVM_SET_SIGNAL_MASK: {
2316                 struct kvm_signal_mask __user *sigmask_arg = argp;
2317                 struct kvm_signal_mask kvm_sigmask;
2318                 compat_sigset_t csigset;
2319                 sigset_t sigset;
2320
2321                 if (argp) {
2322                         r = -EFAULT;
2323                         if (copy_from_user(&kvm_sigmask, argp,
2324                                            sizeof kvm_sigmask))
2325                                 goto out;
2326                         r = -EINVAL;
2327                         if (kvm_sigmask.len != sizeof csigset)
2328                                 goto out;
2329                         r = -EFAULT;
2330                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2331                                            sizeof csigset))
2332                                 goto out;
2333                         sigset_from_compat(&sigset, &csigset);
2334                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2335                 } else
2336                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2337                 break;
2338         }
2339         default:
2340                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2341         }
2342
2343 out:
2344         return r;
2345 }
2346 #endif
2347
2348 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2349                                  int (*accessor)(struct kvm_device *dev,
2350                                                  struct kvm_device_attr *attr),
2351                                  unsigned long arg)
2352 {
2353         struct kvm_device_attr attr;
2354
2355         if (!accessor)
2356                 return -EPERM;
2357
2358         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2359                 return -EFAULT;
2360
2361         return accessor(dev, &attr);
2362 }
2363
2364 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2365                              unsigned long arg)
2366 {
2367         struct kvm_device *dev = filp->private_data;
2368
2369         switch (ioctl) {
2370         case KVM_SET_DEVICE_ATTR:
2371                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2372         case KVM_GET_DEVICE_ATTR:
2373                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2374         case KVM_HAS_DEVICE_ATTR:
2375                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2376         default:
2377                 if (dev->ops->ioctl)
2378                         return dev->ops->ioctl(dev, ioctl, arg);
2379
2380                 return -ENOTTY;
2381         }
2382 }
2383
2384 static int kvm_device_release(struct inode *inode, struct file *filp)
2385 {
2386         struct kvm_device *dev = filp->private_data;
2387         struct kvm *kvm = dev->kvm;
2388
2389         kvm_put_kvm(kvm);
2390         return 0;
2391 }
2392
2393 static const struct file_operations kvm_device_fops = {
2394         .unlocked_ioctl = kvm_device_ioctl,
2395 #ifdef CONFIG_KVM_COMPAT
2396         .compat_ioctl = kvm_device_ioctl,
2397 #endif
2398         .release = kvm_device_release,
2399 };
2400
2401 struct kvm_device *kvm_device_from_filp(struct file *filp)
2402 {
2403         if (filp->f_op != &kvm_device_fops)
2404                 return NULL;
2405
2406         return filp->private_data;
2407 }
2408
2409 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2410 #ifdef CONFIG_KVM_MPIC
2411         [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
2412         [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
2413 #endif
2414
2415 #ifdef CONFIG_KVM_XICS
2416         [KVM_DEV_TYPE_XICS]             = &kvm_xics_ops,
2417 #endif
2418 };
2419
2420 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2421 {
2422         if (type >= ARRAY_SIZE(kvm_device_ops_table))
2423                 return -ENOSPC;
2424
2425         if (kvm_device_ops_table[type] != NULL)
2426                 return -EEXIST;
2427
2428         kvm_device_ops_table[type] = ops;
2429         return 0;
2430 }
2431
2432 void kvm_unregister_device_ops(u32 type)
2433 {
2434         if (kvm_device_ops_table[type] != NULL)
2435                 kvm_device_ops_table[type] = NULL;
2436 }
2437
2438 static int kvm_ioctl_create_device(struct kvm *kvm,
2439                                    struct kvm_create_device *cd)
2440 {
2441         struct kvm_device_ops *ops = NULL;
2442         struct kvm_device *dev;
2443         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2444         int ret;
2445
2446         if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2447                 return -ENODEV;
2448
2449         ops = kvm_device_ops_table[cd->type];
2450         if (ops == NULL)
2451                 return -ENODEV;
2452
2453         if (test)
2454                 return 0;
2455
2456         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2457         if (!dev)
2458                 return -ENOMEM;
2459
2460         dev->ops = ops;
2461         dev->kvm = kvm;
2462
2463         ret = ops->create(dev, cd->type);
2464         if (ret < 0) {
2465                 kfree(dev);
2466                 return ret;
2467         }
2468
2469         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2470         if (ret < 0) {
2471                 ops->destroy(dev);
2472                 return ret;
2473         }
2474
2475         list_add(&dev->vm_node, &kvm->devices);
2476         kvm_get_kvm(kvm);
2477         cd->fd = ret;
2478         return 0;
2479 }
2480
2481 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2482 {
2483         switch (arg) {
2484         case KVM_CAP_USER_MEMORY:
2485         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2486         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2487 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2488         case KVM_CAP_SET_BOOT_CPU_ID:
2489 #endif
2490         case KVM_CAP_INTERNAL_ERROR_DATA:
2491 #ifdef CONFIG_HAVE_KVM_MSI
2492         case KVM_CAP_SIGNAL_MSI:
2493 #endif
2494 #ifdef CONFIG_HAVE_KVM_IRQFD
2495         case KVM_CAP_IRQFD_RESAMPLE:
2496 #endif
2497         case KVM_CAP_CHECK_EXTENSION_VM:
2498                 return 1;
2499 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2500         case KVM_CAP_IRQ_ROUTING:
2501                 return KVM_MAX_IRQ_ROUTES;
2502 #endif
2503         default:
2504                 break;
2505         }
2506         return kvm_vm_ioctl_check_extension(kvm, arg);
2507 }
2508
2509 static long kvm_vm_ioctl(struct file *filp,
2510                            unsigned int ioctl, unsigned long arg)
2511 {
2512         struct kvm *kvm = filp->private_data;
2513         void __user *argp = (void __user *)arg;
2514         int r;
2515
2516         if (kvm->mm != current->mm)
2517                 return -EIO;
2518         switch (ioctl) {
2519         case KVM_CREATE_VCPU:
2520                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2521                 break;
2522         case KVM_SET_USER_MEMORY_REGION: {
2523                 struct kvm_userspace_memory_region kvm_userspace_mem;
2524
2525                 r = -EFAULT;
2526                 if (copy_from_user(&kvm_userspace_mem, argp,
2527                                                 sizeof kvm_userspace_mem))
2528                         goto out;
2529
2530                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2531                 break;
2532         }
2533         case KVM_GET_DIRTY_LOG: {
2534                 struct kvm_dirty_log log;
2535
2536                 r = -EFAULT;
2537                 if (copy_from_user(&log, argp, sizeof log))
2538                         goto out;
2539                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2540                 break;
2541         }
2542 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2543         case KVM_REGISTER_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_register_coalesced_mmio(kvm, &zone);
2549                 break;
2550         }
2551         case KVM_UNREGISTER_COALESCED_MMIO: {
2552                 struct kvm_coalesced_mmio_zone zone;
2553                 r = -EFAULT;
2554                 if (copy_from_user(&zone, argp, sizeof zone))
2555                         goto out;
2556                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2557                 break;
2558         }
2559 #endif
2560         case KVM_IRQFD: {
2561                 struct kvm_irqfd data;
2562
2563                 r = -EFAULT;
2564                 if (copy_from_user(&data, argp, sizeof data))
2565                         goto out;
2566                 r = kvm_irqfd(kvm, &data);
2567                 break;
2568         }
2569         case KVM_IOEVENTFD: {
2570                 struct kvm_ioeventfd data;
2571
2572                 r = -EFAULT;
2573                 if (copy_from_user(&data, argp, sizeof data))
2574                         goto out;
2575                 r = kvm_ioeventfd(kvm, &data);
2576                 break;
2577         }
2578 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2579         case KVM_SET_BOOT_CPU_ID:
2580                 r = 0;
2581                 mutex_lock(&kvm->lock);
2582                 if (atomic_read(&kvm->online_vcpus) != 0)
2583                         r = -EBUSY;
2584                 else
2585                         kvm->bsp_vcpu_id = arg;
2586                 mutex_unlock(&kvm->lock);
2587                 break;
2588 #endif
2589 #ifdef CONFIG_HAVE_KVM_MSI
2590         case KVM_SIGNAL_MSI: {
2591                 struct kvm_msi msi;
2592
2593                 r = -EFAULT;
2594                 if (copy_from_user(&msi, argp, sizeof msi))
2595                         goto out;
2596                 r = kvm_send_userspace_msi(kvm, &msi);
2597                 break;
2598         }
2599 #endif
2600 #ifdef __KVM_HAVE_IRQ_LINE
2601         case KVM_IRQ_LINE_STATUS:
2602         case KVM_IRQ_LINE: {
2603                 struct kvm_irq_level irq_event;
2604
2605                 r = -EFAULT;
2606                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2607                         goto out;
2608
2609                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2610                                         ioctl == KVM_IRQ_LINE_STATUS);
2611                 if (r)
2612                         goto out;
2613
2614                 r = -EFAULT;
2615                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2616                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2617                                 goto out;
2618                 }
2619
2620                 r = 0;
2621                 break;
2622         }
2623 #endif
2624 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2625         case KVM_SET_GSI_ROUTING: {
2626                 struct kvm_irq_routing routing;
2627                 struct kvm_irq_routing __user *urouting;
2628                 struct kvm_irq_routing_entry *entries;
2629
2630                 r = -EFAULT;
2631                 if (copy_from_user(&routing, argp, sizeof(routing)))
2632                         goto out;
2633                 r = -EINVAL;
2634                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2635                         goto out;
2636                 if (routing.flags)
2637                         goto out;
2638                 r = -ENOMEM;
2639                 entries = vmalloc(routing.nr * sizeof(*entries));
2640                 if (!entries)
2641                         goto out;
2642                 r = -EFAULT;
2643                 urouting = argp;
2644                 if (copy_from_user(entries, urouting->entries,
2645                                    routing.nr * sizeof(*entries)))
2646                         goto out_free_irq_routing;
2647                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2648                                         routing.flags);
2649         out_free_irq_routing:
2650                 vfree(entries);
2651                 break;
2652         }
2653 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2654         case KVM_CREATE_DEVICE: {
2655                 struct kvm_create_device cd;
2656
2657                 r = -EFAULT;
2658                 if (copy_from_user(&cd, argp, sizeof(cd)))
2659                         goto out;
2660
2661                 r = kvm_ioctl_create_device(kvm, &cd);
2662                 if (r)
2663                         goto out;
2664
2665                 r = -EFAULT;
2666                 if (copy_to_user(argp, &cd, sizeof(cd)))
2667                         goto out;
2668
2669                 r = 0;
2670                 break;
2671         }
2672         case KVM_CHECK_EXTENSION:
2673                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2674                 break;
2675         default:
2676                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2677         }
2678 out:
2679         return r;
2680 }
2681
2682 #ifdef CONFIG_KVM_COMPAT
2683 struct compat_kvm_dirty_log {
2684         __u32 slot;
2685         __u32 padding1;
2686         union {
2687                 compat_uptr_t dirty_bitmap; /* one bit per page */
2688                 __u64 padding2;
2689         };
2690 };
2691
2692 static long kvm_vm_compat_ioctl(struct file *filp,
2693                            unsigned int ioctl, unsigned long arg)
2694 {
2695         struct kvm *kvm = filp->private_data;
2696         int r;
2697
2698         if (kvm->mm != current->mm)
2699                 return -EIO;
2700         switch (ioctl) {
2701         case KVM_GET_DIRTY_LOG: {
2702                 struct compat_kvm_dirty_log compat_log;
2703                 struct kvm_dirty_log log;
2704
2705                 r = -EFAULT;
2706                 if (copy_from_user(&compat_log, (void __user *)arg,
2707                                    sizeof(compat_log)))
2708                         goto out;
2709                 log.slot         = compat_log.slot;
2710                 log.padding1     = compat_log.padding1;
2711                 log.padding2     = compat_log.padding2;
2712                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2713
2714                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2715                 break;
2716         }
2717         default:
2718                 r = kvm_vm_ioctl(filp, ioctl, arg);
2719         }
2720
2721 out:
2722         return r;
2723 }
2724 #endif
2725
2726 static struct file_operations kvm_vm_fops = {
2727         .release        = kvm_vm_release,
2728         .unlocked_ioctl = kvm_vm_ioctl,
2729 #ifdef CONFIG_KVM_COMPAT
2730         .compat_ioctl   = kvm_vm_compat_ioctl,
2731 #endif
2732         .llseek         = noop_llseek,
2733 };
2734
2735 static int kvm_dev_ioctl_create_vm(unsigned long type)
2736 {
2737         int r;
2738         struct kvm *kvm;
2739
2740         kvm = kvm_create_vm(type);
2741         if (IS_ERR(kvm))
2742                 return PTR_ERR(kvm);
2743 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2744         r = kvm_coalesced_mmio_init(kvm);
2745         if (r < 0) {
2746                 kvm_put_kvm(kvm);
2747                 return r;
2748         }
2749 #endif
2750         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2751         if (r < 0)
2752                 kvm_put_kvm(kvm);
2753
2754         return r;
2755 }
2756
2757 static long kvm_dev_ioctl(struct file *filp,
2758                           unsigned int ioctl, unsigned long arg)
2759 {
2760         long r = -EINVAL;
2761
2762         switch (ioctl) {
2763         case KVM_GET_API_VERSION:
2764                 if (arg)
2765                         goto out;
2766                 r = KVM_API_VERSION;
2767                 break;
2768         case KVM_CREATE_VM:
2769                 r = kvm_dev_ioctl_create_vm(arg);
2770                 break;
2771         case KVM_CHECK_EXTENSION:
2772                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2773                 break;
2774         case KVM_GET_VCPU_MMAP_SIZE:
2775                 if (arg)
2776                         goto out;
2777                 r = PAGE_SIZE;     /* struct kvm_run */
2778 #ifdef CONFIG_X86
2779                 r += PAGE_SIZE;    /* pio data page */
2780 #endif
2781 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2782                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2783 #endif
2784                 break;
2785         case KVM_TRACE_ENABLE:
2786         case KVM_TRACE_PAUSE:
2787         case KVM_TRACE_DISABLE:
2788                 r = -EOPNOTSUPP;
2789                 break;
2790         default:
2791                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2792         }
2793 out:
2794         return r;
2795 }
2796
2797 static struct file_operations kvm_chardev_ops = {
2798         .unlocked_ioctl = kvm_dev_ioctl,
2799         .compat_ioctl   = kvm_dev_ioctl,
2800         .llseek         = noop_llseek,
2801 };
2802
2803 static struct miscdevice kvm_dev = {
2804         KVM_MINOR,
2805         "kvm",
2806         &kvm_chardev_ops,
2807 };
2808
2809 static void hardware_enable_nolock(void *junk)
2810 {
2811         int cpu = raw_smp_processor_id();
2812         int r;
2813
2814         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2815                 return;
2816
2817         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2818
2819         r = kvm_arch_hardware_enable();
2820
2821         if (r) {
2822                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2823                 atomic_inc(&hardware_enable_failed);
2824                 printk(KERN_INFO "kvm: enabling virtualization on "
2825                                  "CPU%d failed\n", cpu);
2826         }
2827 }
2828
2829 static void hardware_enable(void)
2830 {
2831         raw_spin_lock(&kvm_count_lock);
2832         if (kvm_usage_count)
2833                 hardware_enable_nolock(NULL);
2834         raw_spin_unlock(&kvm_count_lock);
2835 }
2836
2837 static void hardware_disable_nolock(void *junk)
2838 {
2839         int cpu = raw_smp_processor_id();
2840
2841         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2842                 return;
2843         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2844         kvm_arch_hardware_disable();
2845 }
2846
2847 static void hardware_disable(void)
2848 {
2849         raw_spin_lock(&kvm_count_lock);
2850         if (kvm_usage_count)
2851                 hardware_disable_nolock(NULL);
2852         raw_spin_unlock(&kvm_count_lock);
2853 }
2854
2855 static void hardware_disable_all_nolock(void)
2856 {
2857         BUG_ON(!kvm_usage_count);
2858
2859         kvm_usage_count--;
2860         if (!kvm_usage_count)
2861                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2862 }
2863
2864 static void hardware_disable_all(void)
2865 {
2866         raw_spin_lock(&kvm_count_lock);
2867         hardware_disable_all_nolock();
2868         raw_spin_unlock(&kvm_count_lock);
2869 }
2870
2871 static int hardware_enable_all(void)
2872 {
2873         int r = 0;
2874
2875         raw_spin_lock(&kvm_count_lock);
2876
2877         kvm_usage_count++;
2878         if (kvm_usage_count == 1) {
2879                 atomic_set(&hardware_enable_failed, 0);
2880                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2881
2882                 if (atomic_read(&hardware_enable_failed)) {
2883                         hardware_disable_all_nolock();
2884                         r = -EBUSY;
2885                 }
2886         }
2887
2888         raw_spin_unlock(&kvm_count_lock);
2889
2890         return r;
2891 }
2892
2893 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2894                            void *v)
2895 {
2896         int cpu = (long)v;
2897
2898         val &= ~CPU_TASKS_FROZEN;
2899         switch (val) {
2900         case CPU_DYING:
2901                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2902                        cpu);
2903                 hardware_disable();
2904                 break;
2905         case CPU_STARTING:
2906                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2907                        cpu);
2908                 hardware_enable();
2909                 break;
2910         }
2911         return NOTIFY_OK;
2912 }
2913
2914 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2915                       void *v)
2916 {
2917         /*
2918          * Some (well, at least mine) BIOSes hang on reboot if
2919          * in vmx root mode.
2920          *
2921          * And Intel TXT required VMX off for all cpu when system shutdown.
2922          */
2923         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2924         kvm_rebooting = true;
2925         on_each_cpu(hardware_disable_nolock, NULL, 1);
2926         return NOTIFY_OK;
2927 }
2928
2929 static struct notifier_block kvm_reboot_notifier = {
2930         .notifier_call = kvm_reboot,
2931         .priority = 0,
2932 };
2933
2934 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2935 {
2936         int i;
2937
2938         for (i = 0; i < bus->dev_count; i++) {
2939                 struct kvm_io_device *pos = bus->range[i].dev;
2940
2941                 kvm_iodevice_destructor(pos);
2942         }
2943         kfree(bus);
2944 }
2945
2946 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2947                                  const struct kvm_io_range *r2)
2948 {
2949         if (r1->addr < r2->addr)
2950                 return -1;
2951         if (r1->addr + r1->len > r2->addr + r2->len)
2952                 return 1;
2953         return 0;
2954 }
2955
2956 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2957 {
2958         return kvm_io_bus_cmp(p1, p2);
2959 }
2960
2961 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2962                           gpa_t addr, int len)
2963 {
2964         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2965                 .addr = addr,
2966                 .len = len,
2967                 .dev = dev,
2968         };
2969
2970         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2971                 kvm_io_bus_sort_cmp, NULL);
2972
2973         return 0;
2974 }
2975
2976 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2977                              gpa_t addr, int len)
2978 {
2979         struct kvm_io_range *range, key;
2980         int off;
2981
2982         key = (struct kvm_io_range) {
2983                 .addr = addr,
2984                 .len = len,
2985         };
2986
2987         range = bsearch(&key, bus->range, bus->dev_count,
2988                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2989         if (range == NULL)
2990                 return -ENOENT;
2991
2992         off = range - bus->range;
2993
2994         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2995                 off--;
2996
2997         return off;
2998 }
2999
3000 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
3001                               struct kvm_io_range *range, const void *val)
3002 {
3003         int idx;
3004
3005         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3006         if (idx < 0)
3007                 return -EOPNOTSUPP;
3008
3009         while (idx < bus->dev_count &&
3010                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3011                 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
3012                                         range->len, val))
3013                         return idx;
3014                 idx++;
3015         }
3016
3017         return -EOPNOTSUPP;
3018 }
3019
3020 /* kvm_io_bus_write - called under kvm->slots_lock */
3021 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3022                      int len, const void *val)
3023 {
3024         struct kvm_io_bus *bus;
3025         struct kvm_io_range range;
3026         int r;
3027
3028         range = (struct kvm_io_range) {
3029                 .addr = addr,
3030                 .len = len,
3031         };
3032
3033         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3034         r = __kvm_io_bus_write(bus, &range, val);
3035         return r < 0 ? r : 0;
3036 }
3037
3038 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3039 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3040                             int len, const void *val, long cookie)
3041 {
3042         struct kvm_io_bus *bus;
3043         struct kvm_io_range range;
3044
3045         range = (struct kvm_io_range) {
3046                 .addr = addr,
3047                 .len = len,
3048         };
3049
3050         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3051
3052         /* First try the device referenced by cookie. */
3053         if ((cookie >= 0) && (cookie < bus->dev_count) &&
3054             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3055                 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
3056                                         val))
3057                         return cookie;
3058
3059         /*
3060          * cookie contained garbage; fall back to search and return the
3061          * correct cookie value.
3062          */
3063         return __kvm_io_bus_write(bus, &range, val);
3064 }
3065
3066 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
3067                              void *val)
3068 {
3069         int idx;
3070
3071         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3072         if (idx < 0)
3073                 return -EOPNOTSUPP;
3074
3075         while (idx < bus->dev_count &&
3076                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3077                 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
3078                                        range->len, val))
3079                         return idx;
3080                 idx++;
3081         }
3082
3083         return -EOPNOTSUPP;
3084 }
3085 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3086
3087 /* kvm_io_bus_read - called under kvm->slots_lock */
3088 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3089                     int len, void *val)
3090 {
3091         struct kvm_io_bus *bus;
3092         struct kvm_io_range range;
3093         int r;
3094
3095         range = (struct kvm_io_range) {
3096                 .addr = addr,
3097                 .len = len,
3098         };
3099
3100         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3101         r = __kvm_io_bus_read(bus, &range, val);
3102         return r < 0 ? r : 0;
3103 }
3104
3105
3106 /* Caller must hold slots_lock. */
3107 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3108                             int len, struct kvm_io_device *dev)
3109 {
3110         struct kvm_io_bus *new_bus, *bus;
3111
3112         bus = kvm->buses[bus_idx];
3113         /* exclude ioeventfd which is limited by maximum fd */
3114         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3115                 return -ENOSPC;
3116
3117         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3118                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3119         if (!new_bus)
3120                 return -ENOMEM;
3121         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3122                sizeof(struct kvm_io_range)));
3123         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3124         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3125         synchronize_srcu_expedited(&kvm->srcu);
3126         kfree(bus);
3127
3128         return 0;
3129 }
3130
3131 /* Caller must hold slots_lock. */
3132 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3133                               struct kvm_io_device *dev)
3134 {
3135         int i, r;
3136         struct kvm_io_bus *new_bus, *bus;
3137
3138         bus = kvm->buses[bus_idx];
3139         r = -ENOENT;
3140         for (i = 0; i < bus->dev_count; i++)
3141                 if (bus->range[i].dev == dev) {
3142                         r = 0;
3143                         break;
3144                 }
3145
3146         if (r)
3147                 return r;
3148
3149         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3150                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3151         if (!new_bus)
3152                 return -ENOMEM;
3153
3154         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3155         new_bus->dev_count--;
3156         memcpy(new_bus->range + i, bus->range + i + 1,
3157                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3158
3159         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3160         synchronize_srcu_expedited(&kvm->srcu);
3161         kfree(bus);
3162         return r;
3163 }
3164
3165 static struct notifier_block kvm_cpu_notifier = {
3166         .notifier_call = kvm_cpu_hotplug,
3167 };
3168
3169 static int vm_stat_get(void *_offset, u64 *val)
3170 {
3171         unsigned offset = (long)_offset;
3172         struct kvm *kvm;
3173
3174         *val = 0;
3175         spin_lock(&kvm_lock);
3176         list_for_each_entry(kvm, &vm_list, vm_list)
3177                 *val += *(u32 *)((void *)kvm + offset);
3178         spin_unlock(&kvm_lock);
3179         return 0;
3180 }
3181
3182 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3183
3184 static int vcpu_stat_get(void *_offset, u64 *val)
3185 {
3186         unsigned offset = (long)_offset;
3187         struct kvm *kvm;
3188         struct kvm_vcpu *vcpu;
3189         int i;
3190
3191         *val = 0;
3192         spin_lock(&kvm_lock);
3193         list_for_each_entry(kvm, &vm_list, vm_list)
3194                 kvm_for_each_vcpu(i, vcpu, kvm)
3195                         *val += *(u32 *)((void *)vcpu + offset);
3196
3197         spin_unlock(&kvm_lock);
3198         return 0;
3199 }
3200
3201 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3202
3203 static const struct file_operations *stat_fops[] = {
3204         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3205         [KVM_STAT_VM]   = &vm_stat_fops,
3206 };
3207
3208 static int kvm_init_debug(void)
3209 {
3210         int r = -EEXIST;
3211         struct kvm_stats_debugfs_item *p;
3212
3213         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3214         if (kvm_debugfs_dir == NULL)
3215                 goto out;
3216
3217         for (p = debugfs_entries; p->name; ++p) {
3218                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3219                                                 (void *)(long)p->offset,
3220                                                 stat_fops[p->kind]);
3221                 if (p->dentry == NULL)
3222                         goto out_dir;
3223         }
3224
3225         return 0;
3226
3227 out_dir:
3228         debugfs_remove_recursive(kvm_debugfs_dir);
3229 out:
3230         return r;
3231 }
3232
3233 static void kvm_exit_debug(void)
3234 {
3235         struct kvm_stats_debugfs_item *p;
3236
3237         for (p = debugfs_entries; p->name; ++p)
3238                 debugfs_remove(p->dentry);
3239         debugfs_remove(kvm_debugfs_dir);
3240 }
3241
3242 static int kvm_suspend(void)
3243 {
3244         if (kvm_usage_count)
3245                 hardware_disable_nolock(NULL);
3246         return 0;
3247 }
3248
3249 static void kvm_resume(void)
3250 {
3251         if (kvm_usage_count) {
3252                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3253                 hardware_enable_nolock(NULL);
3254         }
3255 }
3256
3257 static struct syscore_ops kvm_syscore_ops = {
3258         .suspend = kvm_suspend,
3259         .resume = kvm_resume,
3260 };
3261
3262 static inline
3263 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3264 {
3265         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3266 }
3267
3268 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3269 {
3270         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3271         if (vcpu->preempted)
3272                 vcpu->preempted = false;
3273
3274         kvm_arch_sched_in(vcpu, cpu);
3275
3276         kvm_arch_vcpu_load(vcpu, cpu);
3277 }
3278
3279 static void kvm_sched_out(struct preempt_notifier *pn,
3280                           struct task_struct *next)
3281 {
3282         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3283
3284         if (current->state == TASK_RUNNING)
3285                 vcpu->preempted = true;
3286         kvm_arch_vcpu_put(vcpu);
3287 }
3288
3289 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3290                   struct module *module)
3291 {
3292         int r;
3293         int cpu;
3294
3295         r = kvm_arch_init(opaque);
3296         if (r)
3297                 goto out_fail;
3298
3299         /*
3300          * kvm_arch_init makes sure there's at most one caller
3301          * for architectures that support multiple implementations,
3302          * like intel and amd on x86.
3303          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3304          * conflicts in case kvm is already setup for another implementation.
3305          */
3306         r = kvm_irqfd_init();
3307         if (r)
3308                 goto out_irqfd;
3309
3310         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3311                 r = -ENOMEM;
3312                 goto out_free_0;
3313         }
3314
3315         r = kvm_arch_hardware_setup();
3316         if (r < 0)
3317                 goto out_free_0a;
3318
3319         for_each_online_cpu(cpu) {
3320                 smp_call_function_single(cpu,
3321                                 kvm_arch_check_processor_compat,
3322                                 &r, 1);
3323                 if (r < 0)
3324                         goto out_free_1;
3325         }
3326
3327         r = register_cpu_notifier(&kvm_cpu_notifier);
3328         if (r)
3329                 goto out_free_2;
3330         register_reboot_notifier(&kvm_reboot_notifier);
3331
3332         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3333         if (!vcpu_align)
3334                 vcpu_align = __alignof__(struct kvm_vcpu);
3335         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3336                                            0, NULL);
3337         if (!kvm_vcpu_cache) {
3338                 r = -ENOMEM;
3339                 goto out_free_3;
3340         }
3341
3342         r = kvm_async_pf_init();
3343         if (r)
3344                 goto out_free;
3345
3346         kvm_chardev_ops.owner = module;
3347         kvm_vm_fops.owner = module;
3348         kvm_vcpu_fops.owner = module;
3349
3350         r = misc_register(&kvm_dev);
3351         if (r) {
3352                 printk(KERN_ERR "kvm: misc device register failed\n");
3353                 goto out_unreg;
3354         }
3355
3356         register_syscore_ops(&kvm_syscore_ops);
3357
3358         kvm_preempt_ops.sched_in = kvm_sched_in;
3359         kvm_preempt_ops.sched_out = kvm_sched_out;
3360
3361         r = kvm_init_debug();
3362         if (r) {
3363                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3364                 goto out_undebugfs;
3365         }
3366
3367         r = kvm_vfio_ops_init();
3368         WARN_ON(r);
3369
3370         return 0;
3371
3372 out_undebugfs:
3373         unregister_syscore_ops(&kvm_syscore_ops);
3374         misc_deregister(&kvm_dev);
3375 out_unreg:
3376         kvm_async_pf_deinit();
3377 out_free:
3378         kmem_cache_destroy(kvm_vcpu_cache);
3379 out_free_3:
3380         unregister_reboot_notifier(&kvm_reboot_notifier);
3381         unregister_cpu_notifier(&kvm_cpu_notifier);
3382 out_free_2:
3383 out_free_1:
3384         kvm_arch_hardware_unsetup();
3385 out_free_0a:
3386         free_cpumask_var(cpus_hardware_enabled);
3387 out_free_0:
3388         kvm_irqfd_exit();
3389 out_irqfd:
3390         kvm_arch_exit();
3391 out_fail:
3392         return r;
3393 }
3394 EXPORT_SYMBOL_GPL(kvm_init);
3395
3396 void kvm_exit(void)
3397 {
3398         kvm_exit_debug();
3399         misc_deregister(&kvm_dev);
3400         kmem_cache_destroy(kvm_vcpu_cache);
3401         kvm_async_pf_deinit();
3402         unregister_syscore_ops(&kvm_syscore_ops);
3403         unregister_reboot_notifier(&kvm_reboot_notifier);
3404         unregister_cpu_notifier(&kvm_cpu_notifier);
3405         on_each_cpu(hardware_disable_nolock, NULL, 1);
3406         kvm_arch_hardware_unsetup();
3407         kvm_arch_exit();
3408         kvm_irqfd_exit();
3409         free_cpumask_var(cpus_hardware_enabled);
3410         kvm_vfio_ops_exit();
3411 }
3412 EXPORT_SYMBOL_GPL(kvm_exit);