virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include <kvm/iodev.h>
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/syscore_ops.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/ioctl.h>
  56 #include <asm/uaccess.h>
  57 #include <asm/pgtable.h>
  58
  59 #include "coalesced_mmio.h"
  60 #include "async_pf.h"
  61 #include "vfio.h"
  62
  63 #define CREATE_TRACE_POINTS
  64 #include <trace/events/kvm.h>
  65
  66 MODULE_AUTHOR("Qumranet");
  67 MODULE_LICENSE("GPL");
  68
  69 static unsigned int halt_poll_ns;
  70 module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
  71
  72 /*
  73  * Ordering of locks:
  74  *
  75  *      kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  76  */
  77
  78 DEFINE_SPINLOCK(kvm_lock);
  79 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
  80 LIST_HEAD(vm_list);
  81
  82 static cpumask_var_t cpus_hardware_enabled;
  83 static int kvm_usage_count;
  84 static atomic_t hardware_enable_failed;
  85
  86 struct kmem_cache *kvm_vcpu_cache;
  87 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  88
  89 static __read_mostly struct preempt_ops kvm_preempt_ops;
  90
  91 struct dentry *kvm_debugfs_dir;
  92 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
  93
  94 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  95                            unsigned long arg);
  96 #ifdef CONFIG_KVM_COMPAT
  97 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  98                                   unsigned long arg);
  99 #endif
 100 static int hardware_enable_all(void);
 101 static void hardware_disable_all(void);
 102
 103 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
 104
 105 static void kvm_release_pfn_dirty(pfn_t pfn);
 106 static void mark_page_dirty_in_slot(struct kvm *kvm,
 107                                     struct kvm_memory_slot *memslot, gfn_t gfn);
 108
 109 __visible bool kvm_rebooting;
 110 EXPORT_SYMBOL_GPL(kvm_rebooting);
 111
 112 static bool largepages_enabled = true;
 113
 114 bool kvm_is_reserved_pfn(pfn_t pfn)
 115 {
 116         if (pfn_valid(pfn))
 117                 return PageReserved(pfn_to_page(pfn));
 118
 119         return true;
 120 }
 121
 122 /*
 123  * Switches to specified vcpu, until a matching vcpu_put()
 124  */
 125 int vcpu_load(struct kvm_vcpu *vcpu)
 126 {
 127         int cpu;
 128
 129         if (mutex_lock_killable(&vcpu->mutex))
 130                 return -EINTR;
 131         cpu = get_cpu();
 132         preempt_notifier_register(&vcpu->preempt_notifier);
 133         kvm_arch_vcpu_load(vcpu, cpu);
 134         put_cpu();
 135         return 0;
 136 }
 137
 138 void vcpu_put(struct kvm_vcpu *vcpu)
 139 {
 140         preempt_disable();
 141         kvm_arch_vcpu_put(vcpu);
 142         preempt_notifier_unregister(&vcpu->preempt_notifier);
 143         preempt_enable();
 144         mutex_unlock(&vcpu->mutex);
 145 }
 146
 147 static void ack_flush(void *_completed)
 148 {
 149 }
 150
 151 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
 152 {
 153         int i, cpu, me;
 154         cpumask_var_t cpus;
 155         bool called = true;
 156         struct kvm_vcpu *vcpu;
 157
 158         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 159
 160         me = get_cpu();
 161         kvm_for_each_vcpu(i, vcpu, kvm) {
 162                 kvm_make_request(req, vcpu);
 163                 cpu = vcpu->cpu;
 164
 165                 /* Set ->requests bit before we read ->mode */
 166                 smp_mb();
 167
 168                 if (cpus != NULL && cpu != -1 && cpu != me &&
 169                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 170                         cpumask_set_cpu(cpu, cpus);
 171         }
 172         if (unlikely(cpus == NULL))
 173                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 174         else if (!cpumask_empty(cpus))
 175                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 176         else
 177                 called = false;
 178         put_cpu();
 179         free_cpumask_var(cpus);
 180         return called;
 181 }
 182
 183 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
 184 void kvm_flush_remote_tlbs(struct kvm *kvm)
 185 {
 186         long dirty_count = kvm->tlbs_dirty;
 187
 188         smp_mb();
 189         if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 190                 ++kvm->stat.remote_tlb_flush;
 191         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 192 }
 193 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 194 #endif
 195
 196 void kvm_reload_remote_mmus(struct kvm *kvm)
 197 {
 198         kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 199 }
 200
 201 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 202 {
 203         kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 204 }
 205
 206 void kvm_make_scan_ioapic_request(struct kvm *kvm)
 207 {
 208         kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
 209 }
 210
 211 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 212 {
 213         struct page *page;
 214         int r;
 215
 216         mutex_init(&vcpu->mutex);
 217         vcpu->cpu = -1;
 218         vcpu->kvm = kvm;
 219         vcpu->vcpu_id = id;
 220         vcpu->pid = NULL;
 221         init_waitqueue_head(&vcpu->wq);
 222         kvm_async_pf_vcpu_init(vcpu);
 223
 224         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 225         if (!page) {
 226                 r = -ENOMEM;
 227                 goto fail;
 228         }
 229         vcpu->run = page_address(page);
 230
 231         kvm_vcpu_set_in_spin_loop(vcpu, false);
 232         kvm_vcpu_set_dy_eligible(vcpu, false);
 233         vcpu->preempted = false;
 234
 235         r = kvm_arch_vcpu_init(vcpu);
 236         if (r < 0)
 237                 goto fail_free_run;
 238         return 0;
 239
 240 fail_free_run:
 241         free_page((unsigned long)vcpu->run);
 242 fail:
 243         return r;
 244 }
 245 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 246
 247 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 248 {
 249         put_pid(vcpu->pid);
 250         kvm_arch_vcpu_uninit(vcpu);
 251         free_page((unsigned long)vcpu->run);
 252 }
 253 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 254
 255 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 256 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 257 {
 258         return container_of(mn, struct kvm, mmu_notifier);
 259 }
 260
 261 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 262                                              struct mm_struct *mm,
 263                                              unsigned long address)
 264 {
 265         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 266         int need_tlb_flush, idx;
 267
 268         /*
 269          * When ->invalidate_page runs, the linux pte has been zapped
 270          * already but the page is still allocated until
 271          * ->invalidate_page returns. So if we increase the sequence
 272          * here the kvm page fault will notice if the spte can't be
 273          * established because the page is going to be freed. If
 274          * instead the kvm page fault establishes the spte before
 275          * ->invalidate_page runs, kvm_unmap_hva will release it
 276          * before returning.
 277          *
 278          * The sequence increase only need to be seen at spin_unlock
 279          * time, and not at spin_lock time.
 280          *
 281          * Increasing the sequence after the spin_unlock would be
 282          * unsafe because the kvm page fault could then establish the
 283          * pte after kvm_unmap_hva returned, without noticing the page
 284          * is going to be freed.
 285          */
 286         idx = srcu_read_lock(&kvm->srcu);
 287         spin_lock(&kvm->mmu_lock);
 288
 289         kvm->mmu_notifier_seq++;
 290         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 291         /* we've to flush the tlb before the pages can be freed */
 292         if (need_tlb_flush)
 293                 kvm_flush_remote_tlbs(kvm);
 294
 295         spin_unlock(&kvm->mmu_lock);
 296
 297         kvm_arch_mmu_notifier_invalidate_page(kvm, address);
 298
 299         srcu_read_unlock(&kvm->srcu, idx);
 300 }
 301
 302 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 303                                         struct mm_struct *mm,
 304                                         unsigned long address,
 305                                         pte_t pte)
 306 {
 307         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 308         int idx;
 309
 310         idx = srcu_read_lock(&kvm->srcu);
 311         spin_lock(&kvm->mmu_lock);
 312         kvm->mmu_notifier_seq++;
 313         kvm_set_spte_hva(kvm, address, pte);
 314         spin_unlock(&kvm->mmu_lock);
 315         srcu_read_unlock(&kvm->srcu, idx);
 316 }
 317
 318 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 319                                                     struct mm_struct *mm,
 320                                                     unsigned long start,
 321                                                     unsigned long end)
 322 {
 323         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 324         int need_tlb_flush = 0, idx;
 325
 326         idx = srcu_read_lock(&kvm->srcu);
 327         spin_lock(&kvm->mmu_lock);
 328         /*
 329          * The count increase must become visible at unlock time as no
 330          * spte can be established without taking the mmu_lock and
 331          * count is also read inside the mmu_lock critical section.
 332          */
 333         kvm->mmu_notifier_count++;
 334         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 335         need_tlb_flush |= kvm->tlbs_dirty;
 336         /* we've to flush the tlb before the pages can be freed */
 337         if (need_tlb_flush)
 338                 kvm_flush_remote_tlbs(kvm);
 339
 340         spin_unlock(&kvm->mmu_lock);
 341         srcu_read_unlock(&kvm->srcu, idx);
 342 }
 343
 344 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 345                                                   struct mm_struct *mm,
 346                                                   unsigned long start,
 347                                                   unsigned long end)
 348 {
 349         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 350
 351         spin_lock(&kvm->mmu_lock);
 352         /*
 353          * This sequence increase will notify the kvm page fault that
 354          * the page that is going to be mapped in the spte could have
 355          * been freed.
 356          */
 357         kvm->mmu_notifier_seq++;
 358         smp_wmb();
 359         /*
 360          * The above sequence increase must be visible before the
 361          * below count decrease, which is ensured by the smp_wmb above
 362          * in conjunction with the smp_rmb in mmu_notifier_retry().
 363          */
 364         kvm->mmu_notifier_count--;
 365         spin_unlock(&kvm->mmu_lock);
 366
 367         BUG_ON(kvm->mmu_notifier_count < 0);
 368 }
 369
 370 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 371                                               struct mm_struct *mm,
 372                                               unsigned long start,
 373                                               unsigned long end)
 374 {
 375         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 376         int young, idx;
 377
 378         idx = srcu_read_lock(&kvm->srcu);
 379         spin_lock(&kvm->mmu_lock);
 380
 381         young = kvm_age_hva(kvm, start, end);
 382         if (young)
 383                 kvm_flush_remote_tlbs(kvm);
 384
 385         spin_unlock(&kvm->mmu_lock);
 386         srcu_read_unlock(&kvm->srcu, idx);
 387
 388         return young;
 389 }
 390
 391 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 392                                        struct mm_struct *mm,
 393                                        unsigned long address)
 394 {
 395         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 396         int young, idx;
 397
 398         idx = srcu_read_lock(&kvm->srcu);
 399         spin_lock(&kvm->mmu_lock);
 400         young = kvm_test_age_hva(kvm, address);
 401         spin_unlock(&kvm->mmu_lock);
 402         srcu_read_unlock(&kvm->srcu, idx);
 403
 404         return young;
 405 }
 406
 407 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 408                                      struct mm_struct *mm)
 409 {
 410         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 411         int idx;
 412
 413         idx = srcu_read_lock(&kvm->srcu);
 414         kvm_arch_flush_shadow_all(kvm);
 415         srcu_read_unlock(&kvm->srcu, idx);
 416 }
 417
 418 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 419         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 420         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 421         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 422         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 423         .test_young             = kvm_mmu_notifier_test_young,
 424         .change_pte             = kvm_mmu_notifier_change_pte,
 425         .release                = kvm_mmu_notifier_release,
 426 };
 427
 428 static int kvm_init_mmu_notifier(struct kvm *kvm)
 429 {
 430         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 431         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 432 }
 433
 434 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 435
 436 static int kvm_init_mmu_notifier(struct kvm *kvm)
 437 {
 438         return 0;
 439 }
 440
 441 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 442
 443 static struct kvm_memslots *kvm_alloc_memslots(void)
 444 {
 445         int i;
 446         struct kvm_memslots *slots;
 447
 448         slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
 449         if (!slots)
 450                 return NULL;
 451
 452         /*
 453          * Init kvm generation close to the maximum to easily test the
 454          * code of handling generation number wrap-around.
 455          */
 456         slots->generation = -150;
 457         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 458                 slots->id_to_index[i] = slots->memslots[i].id = i;
 459
 460         return slots;
 461 }
 462
 463 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 464 {
 465         if (!memslot->dirty_bitmap)
 466                 return;
 467
 468         kvfree(memslot->dirty_bitmap);
 469         memslot->dirty_bitmap = NULL;
 470 }
 471
 472 /*
 473  * Free any memory in @free but not in @dont.
 474  */
 475 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 476                               struct kvm_memory_slot *dont)
 477 {
 478         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 479                 kvm_destroy_dirty_bitmap(free);
 480
 481         kvm_arch_free_memslot(kvm, free, dont);
 482
 483         free->npages = 0;
 484 }
 485
 486 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
 487 {
 488         struct kvm_memory_slot *memslot;
 489
 490         if (!slots)
 491                 return;
 492
 493         kvm_for_each_memslot(memslot, slots)
 494                 kvm_free_memslot(kvm, memslot, NULL);
 495
 496         kvfree(slots);
 497 }
 498
 499 static struct kvm *kvm_create_vm(unsigned long type)
 500 {
 501         int r, i;
 502         struct kvm *kvm = kvm_arch_alloc_vm();
 503
 504         if (!kvm)
 505                 return ERR_PTR(-ENOMEM);
 506
 507         r = kvm_arch_init_vm(kvm, type);
 508         if (r)
 509                 goto out_err_no_disable;
 510
 511         r = hardware_enable_all();
 512         if (r)
 513                 goto out_err_no_disable;
 514
 515 #ifdef CONFIG_HAVE_KVM_IRQFD
 516         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 517 #endif
 518
 519         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 520
 521         r = -ENOMEM;
 522         kvm->memslots = kvm_alloc_memslots();
 523         if (!kvm->memslots)
 524                 goto out_err_no_srcu;
 525
 526         if (init_srcu_struct(&kvm->srcu))
 527                 goto out_err_no_srcu;
 528         if (init_srcu_struct(&kvm->irq_srcu))
 529                 goto out_err_no_irq_srcu;
 530         for (i = 0; i < KVM_NR_BUSES; i++) {
 531                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 532                                         GFP_KERNEL);
 533                 if (!kvm->buses[i])
 534                         goto out_err;
 535         }
 536
 537         spin_lock_init(&kvm->mmu_lock);
 538         kvm->mm = current->mm;
 539         atomic_inc(&kvm->mm->mm_count);
 540         kvm_eventfd_init(kvm);
 541         mutex_init(&kvm->lock);
 542         mutex_init(&kvm->irq_lock);
 543         mutex_init(&kvm->slots_lock);
 544         atomic_set(&kvm->users_count, 1);
 545         INIT_LIST_HEAD(&kvm->devices);
 546
 547         r = kvm_init_mmu_notifier(kvm);
 548         if (r)
 549                 goto out_err;
 550
 551         spin_lock(&kvm_lock);
 552         list_add(&kvm->vm_list, &vm_list);
 553         spin_unlock(&kvm_lock);
 554
 555         return kvm;
 556
 557 out_err:
 558         cleanup_srcu_struct(&kvm->irq_srcu);
 559 out_err_no_irq_srcu:
 560         cleanup_srcu_struct(&kvm->srcu);
 561 out_err_no_srcu:
 562         hardware_disable_all();
 563 out_err_no_disable:
 564         for (i = 0; i < KVM_NR_BUSES; i++)
 565                 kfree(kvm->buses[i]);
 566         kvm_free_memslots(kvm, kvm->memslots);
 567         kvm_arch_free_vm(kvm);
 568         return ERR_PTR(r);
 569 }
 570
 571 /*
 572  * Avoid using vmalloc for a small buffer.
 573  * Should not be used when the size is statically known.
 574  */
 575 void *kvm_kvzalloc(unsigned long size)
 576 {
 577         if (size > PAGE_SIZE)
 578                 return vzalloc(size);
 579         else
 580                 return kzalloc(size, GFP_KERNEL);
 581 }
 582
 583 static void kvm_destroy_devices(struct kvm *kvm)
 584 {
 585         struct list_head *node, *tmp;
 586
 587         list_for_each_safe(node, tmp, &kvm->devices) {
 588                 struct kvm_device *dev =
 589                         list_entry(node, struct kvm_device, vm_node);
 590
 591                 list_del(node);
 592                 dev->ops->destroy(dev);
 593         }
 594 }
 595
 596 static void kvm_destroy_vm(struct kvm *kvm)
 597 {
 598         int i;
 599         struct mm_struct *mm = kvm->mm;
 600
 601         kvm_arch_sync_events(kvm);
 602         spin_lock(&kvm_lock);
 603         list_del(&kvm->vm_list);
 604         spin_unlock(&kvm_lock);
 605         kvm_free_irq_routing(kvm);
 606         for (i = 0; i < KVM_NR_BUSES; i++)
 607                 kvm_io_bus_destroy(kvm->buses[i]);
 608         kvm_coalesced_mmio_free(kvm);
 609 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 610         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 611 #else
 612         kvm_arch_flush_shadow_all(kvm);
 613 #endif
 614         kvm_arch_destroy_vm(kvm);
 615         kvm_destroy_devices(kvm);
 616         kvm_free_memslots(kvm, kvm->memslots);
 617         cleanup_srcu_struct(&kvm->irq_srcu);
 618         cleanup_srcu_struct(&kvm->srcu);
 619         kvm_arch_free_vm(kvm);
 620         hardware_disable_all();
 621         mmdrop(mm);
 622 }
 623
 624 void kvm_get_kvm(struct kvm *kvm)
 625 {
 626         atomic_inc(&kvm->users_count);
 627 }
 628 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 629
 630 void kvm_put_kvm(struct kvm *kvm)
 631 {
 632         if (atomic_dec_and_test(&kvm->users_count))
 633                 kvm_destroy_vm(kvm);
 634 }
 635 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 636
 637
 638 static int kvm_vm_release(struct inode *inode, struct file *filp)
 639 {
 640         struct kvm *kvm = filp->private_data;
 641
 642         kvm_irqfd_release(kvm);
 643
 644         kvm_put_kvm(kvm);
 645         return 0;
 646 }
 647
 648 /*
 649  * Allocation size is twice as large as the actual dirty bitmap size.
 650  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 651  */
 652 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 653 {
 654         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 655
 656         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 657         if (!memslot->dirty_bitmap)
 658                 return -ENOMEM;
 659
 660         return 0;
 661 }
 662
 663 /*
 664  * Insert memslot and re-sort memslots based on their GFN,
 665  * so binary search could be used to lookup GFN.
 666  * Sorting algorithm takes advantage of having initially
 667  * sorted array and known changed memslot position.
 668  */
 669 static void update_memslots(struct kvm_memslots *slots,
 670                             struct kvm_memory_slot *new)
 671 {
 672         int id = new->id;
 673         int i = slots->id_to_index[id];
 674         struct kvm_memory_slot *mslots = slots->memslots;
 675
 676         WARN_ON(mslots[i].id != id);
 677         if (!new->npages) {
 678                 WARN_ON(!mslots[i].npages);
 679                 if (mslots[i].npages)
 680                         slots->used_slots--;
 681         } else {
 682                 if (!mslots[i].npages)
 683                         slots->used_slots++;
 684         }
 685
 686         while (i < KVM_MEM_SLOTS_NUM - 1 &&
 687                new->base_gfn <= mslots[i + 1].base_gfn) {
 688                 if (!mslots[i + 1].npages)
 689                         break;
 690                 mslots[i] = mslots[i + 1];
 691                 slots->id_to_index[mslots[i].id] = i;
 692                 i++;
 693         }
 694
 695         /*
 696          * The ">=" is needed when creating a slot with base_gfn == 0,
 697          * so that it moves before all those with base_gfn == npages == 0.
 698          *
 699          * On the other hand, if new->npages is zero, the above loop has
 700          * already left i pointing to the beginning of the empty part of
 701          * mslots, and the ">=" would move the hole backwards in this
 702          * case---which is wrong.  So skip the loop when deleting a slot.
 703          */
 704         if (new->npages) {
 705                 while (i > 0 &&
 706                        new->base_gfn >= mslots[i - 1].base_gfn) {
 707                         mslots[i] = mslots[i - 1];
 708                         slots->id_to_index[mslots[i].id] = i;
 709                         i--;
 710                 }
 711         } else
 712                 WARN_ON_ONCE(i != slots->used_slots);
 713
 714         mslots[i] = *new;
 715         slots->id_to_index[mslots[i].id] = i;
 716 }
 717
 718 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
 719 {
 720         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 721
 722 #ifdef __KVM_HAVE_READONLY_MEM
 723         valid_flags |= KVM_MEM_READONLY;
 724 #endif
 725
 726         if (mem->flags & ~valid_flags)
 727                 return -EINVAL;
 728
 729         return 0;
 730 }
 731
 732 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 733                 struct kvm_memslots *slots)
 734 {
 735         struct kvm_memslots *old_memslots = kvm_memslots(kvm);
 736
 737         /*
 738          * Set the low bit in the generation, which disables SPTE caching
 739          * until the end of synchronize_srcu_expedited.
 740          */
 741         WARN_ON(old_memslots->generation & 1);
 742         slots->generation = old_memslots->generation + 1;
 743
 744         rcu_assign_pointer(kvm->memslots, slots);
 745         synchronize_srcu_expedited(&kvm->srcu);
 746
 747         /*
 748          * Increment the new memslot generation a second time. This prevents
 749          * vm exits that race with memslot updates from caching a memslot
 750          * generation that will (potentially) be valid forever.
 751          */
 752         slots->generation++;
 753
 754         kvm_arch_memslots_updated(kvm);
 755
 756         return old_memslots;
 757 }
 758
 759 /*
 760  * Allocate some memory and give it an address in the guest physical address
 761  * space.
 762  *
 763  * Discontiguous memory is allowed, mostly for framebuffers.
 764  *
 765  * Must be called holding kvm->slots_lock for write.
 766  */
 767 int __kvm_set_memory_region(struct kvm *kvm,
 768                             const struct kvm_userspace_memory_region *mem)
 769 {
 770         int r;
 771         gfn_t base_gfn;
 772         unsigned long npages;
 773         struct kvm_memory_slot *slot;
 774         struct kvm_memory_slot old, new;
 775         struct kvm_memslots *slots = NULL, *old_memslots;
 776         enum kvm_mr_change change;
 777
 778         r = check_memory_region_flags(mem);
 779         if (r)
 780                 goto out;
 781
 782         r = -EINVAL;
 783         /* General sanity checks */
 784         if (mem->memory_size & (PAGE_SIZE - 1))
 785                 goto out;
 786         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 787                 goto out;
 788         /* We can read the guest memory with __xxx_user() later on. */
 789         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 790             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 791              !access_ok(VERIFY_WRITE,
 792                         (void __user *)(unsigned long)mem->userspace_addr,
 793                         mem->memory_size)))
 794                 goto out;
 795         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 796                 goto out;
 797         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 798                 goto out;
 799
 800         slot = id_to_memslot(kvm_memslots(kvm), mem->slot);
 801         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 802         npages = mem->memory_size >> PAGE_SHIFT;
 803
 804         if (npages > KVM_MEM_MAX_NR_PAGES)
 805                 goto out;
 806
 807         new = old = *slot;
 808
 809         new.id = mem->slot;
 810         new.base_gfn = base_gfn;
 811         new.npages = npages;
 812         new.flags = mem->flags;
 813
 814         if (npages) {
 815                 if (!old.npages)
 816                         change = KVM_MR_CREATE;
 817                 else { /* Modify an existing slot. */
 818                         if ((mem->userspace_addr != old.userspace_addr) ||
 819                             (npages != old.npages) ||
 820                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 821                                 goto out;
 822
 823                         if (base_gfn != old.base_gfn)
 824                                 change = KVM_MR_MOVE;
 825                         else if (new.flags != old.flags)
 826                                 change = KVM_MR_FLAGS_ONLY;
 827                         else { /* Nothing to change. */
 828                                 r = 0;
 829                                 goto out;
 830                         }
 831                 }
 832         } else {
 833                 if (!old.npages)
 834                         goto out;
 835
 836                 change = KVM_MR_DELETE;
 837                 new.base_gfn = 0;
 838                 new.flags = 0;
 839         }
 840
 841         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 842                 /* Check for overlaps */
 843                 r = -EEXIST;
 844                 kvm_for_each_memslot(slot, kvm_memslots(kvm)) {
 845                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 846                             (slot->id == mem->slot))
 847                                 continue;
 848                         if (!((base_gfn + npages <= slot->base_gfn) ||
 849                               (base_gfn >= slot->base_gfn + slot->npages)))
 850                                 goto out;
 851                 }
 852         }
 853
 854         /* Free page dirty bitmap if unneeded */
 855         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 856                 new.dirty_bitmap = NULL;
 857
 858         r = -ENOMEM;
 859         if (change == KVM_MR_CREATE) {
 860                 new.userspace_addr = mem->userspace_addr;
 861
 862                 if (kvm_arch_create_memslot(kvm, &new, npages))
 863                         goto out_free;
 864         }
 865
 866         /* Allocate page dirty bitmap if needed */
 867         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 868                 if (kvm_create_dirty_bitmap(&new) < 0)
 869                         goto out_free;
 870         }
 871
 872         slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
 873         if (!slots)
 874                 goto out_free;
 875         memcpy(slots, kvm_memslots(kvm), sizeof(struct kvm_memslots));
 876
 877         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 878                 slot = id_to_memslot(slots, mem->slot);
 879                 slot->flags |= KVM_MEMSLOT_INVALID;
 880
 881                 old_memslots = install_new_memslots(kvm, slots);
 882
 883                 /* slot was deleted or moved, clear iommu mapping */
 884                 kvm_iommu_unmap_pages(kvm, &old);
 885                 /* From this point no new shadow pages pointing to a deleted,
 886                  * or moved, memslot will be created.
 887                  *
 888                  * validation of sp->gfn happens in:
 889                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 890                  *      - kvm_is_visible_gfn (mmu_check_roots)
 891                  */
 892                 kvm_arch_flush_shadow_memslot(kvm, slot);
 893
 894                 /*
 895                  * We can re-use the old_memslots from above, the only difference
 896                  * from the currently installed memslots is the invalid flag.  This
 897                  * will get overwritten by update_memslots anyway.
 898                  */
 899                 slots = old_memslots;
 900         }
 901
 902         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 903         if (r)
 904                 goto out_slots;
 905
 906         /* actual memory is freed via old in kvm_free_memslot below */
 907         if (change == KVM_MR_DELETE) {
 908                 new.dirty_bitmap = NULL;
 909                 memset(&new.arch, 0, sizeof(new.arch));
 910         }
 911
 912         update_memslots(slots, &new);
 913         old_memslots = install_new_memslots(kvm, slots);
 914
 915         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 916
 917         kvm_free_memslot(kvm, &old, &new);
 918         kvfree(old_memslots);
 919
 920         /*
 921          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 922          * un-mapped and re-mapped if their base changes.  Since base change
 923          * unmapping is handled above with slot deletion, mapping alone is
 924          * needed here.  Anything else the iommu might care about for existing
 925          * slots (size changes, userspace addr changes and read-only flag
 926          * changes) is disallowed above, so any other attribute changes getting
 927          * here can be skipped.
 928          */
 929         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 930                 r = kvm_iommu_map_pages(kvm, &new);
 931                 return r;
 932         }
 933
 934         return 0;
 935
 936 out_slots:
 937         kvfree(slots);
 938 out_free:
 939         kvm_free_memslot(kvm, &new, &old);
 940 out:
 941         return r;
 942 }
 943 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 944
 945 int kvm_set_memory_region(struct kvm *kvm,
 946                           const struct kvm_userspace_memory_region *mem)
 947 {
 948         int r;
 949
 950         mutex_lock(&kvm->slots_lock);
 951         r = __kvm_set_memory_region(kvm, mem);
 952         mutex_unlock(&kvm->slots_lock);
 953         return r;
 954 }
 955 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 956
 957 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 958                                           struct kvm_userspace_memory_region *mem)
 959 {
 960         if (mem->slot >= KVM_USER_MEM_SLOTS)
 961                 return -EINVAL;
 962
 963         return kvm_set_memory_region(kvm, mem);
 964 }
 965
 966 int kvm_get_dirty_log(struct kvm *kvm,
 967                         struct kvm_dirty_log *log, int *is_dirty)
 968 {
 969         struct kvm_memslots *slots;
 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         slots = kvm_memslots(kvm);
 980         memslot = id_to_memslot(slots, log->slot);
 981         r = -ENOENT;
 982         if (!memslot->dirty_bitmap)
 983                 goto out;
 984
 985         n = kvm_dirty_bitmap_bytes(memslot);
 986
 987         for (i = 0; !any && i < n/sizeof(long); ++i)
 988                 any = memslot->dirty_bitmap[i];
 989
 990         r = -EFAULT;
 991         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 992                 goto out;
 993
 994         if (any)
 995                 *is_dirty = 1;
 996
 997         r = 0;
 998 out:
 999         return r;
1000 }
1001 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1002
1003 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1004 /**
1005  * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1006  *      are dirty write protect them for next write.
1007  * @kvm:        pointer to kvm instance
1008  * @log:        slot id and address to which we copy the log
1009  * @is_dirty:   flag set if any page is dirty
1010  *
1011  * We need to keep it in mind that VCPU threads can write to the bitmap
1012  * concurrently. So, to avoid losing track of dirty pages we keep the
1013  * following order:
1014  *
1015  *    1. Take a snapshot of the bit and clear it if needed.
1016  *    2. Write protect the corresponding page.
1017  *    3. Copy the snapshot to the userspace.
1018  *    4. Upon return caller flushes TLB's if needed.
1019  *
1020  * Between 2 and 4, the guest may write to the page using the remaining TLB
1021  * entry.  This is not a problem because the page is reported dirty using
1022  * the snapshot taken before and step 4 ensures that writes done after
1023  * exiting to userspace will be logged for the next call.
1024  *
1025  */
1026 int kvm_get_dirty_log_protect(struct kvm *kvm,
1027                         struct kvm_dirty_log *log, bool *is_dirty)
1028 {
1029         struct kvm_memslots *slots;
1030         struct kvm_memory_slot *memslot;
1031         int r, i;
1032         unsigned long n;
1033         unsigned long *dirty_bitmap;
1034         unsigned long *dirty_bitmap_buffer;
1035
1036         r = -EINVAL;
1037         if (log->slot >= KVM_USER_MEM_SLOTS)
1038                 goto out;
1039
1040         slots = kvm_memslots(kvm);
1041         memslot = id_to_memslot(slots, log->slot);
1042
1043         dirty_bitmap = memslot->dirty_bitmap;
1044         r = -ENOENT;
1045         if (!dirty_bitmap)
1046                 goto out;
1047
1048         n = kvm_dirty_bitmap_bytes(memslot);
1049
1050         dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1051         memset(dirty_bitmap_buffer, 0, n);
1052
1053         spin_lock(&kvm->mmu_lock);
1054         *is_dirty = false;
1055         for (i = 0; i < n / sizeof(long); i++) {
1056                 unsigned long mask;
1057                 gfn_t offset;
1058
1059                 if (!dirty_bitmap[i])
1060                         continue;
1061
1062                 *is_dirty = true;
1063
1064                 mask = xchg(&dirty_bitmap[i], 0);
1065                 dirty_bitmap_buffer[i] = mask;
1066
1067                 if (mask) {
1068                         offset = i * BITS_PER_LONG;
1069                         kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1070                                                                 offset, mask);
1071                 }
1072         }
1073
1074         spin_unlock(&kvm->mmu_lock);
1075
1076         r = -EFAULT;
1077         if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1078                 goto out;
1079
1080         r = 0;
1081 out:
1082         return r;
1083 }
1084 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1085 #endif
1086
1087 bool kvm_largepages_enabled(void)
1088 {
1089         return largepages_enabled;
1090 }
1091
1092 void kvm_disable_largepages(void)
1093 {
1094         largepages_enabled = false;
1095 }
1096 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1097
1098 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1099 {
1100         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1101 }
1102 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1103
1104 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1105 {
1106         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1107
1108         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1109               memslot->flags & KVM_MEMSLOT_INVALID)
1110                 return 0;
1111
1112         return 1;
1113 }
1114 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1115
1116 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1117 {
1118         struct vm_area_struct *vma;
1119         unsigned long addr, size;
1120
1121         size = PAGE_SIZE;
1122
1123         addr = gfn_to_hva(kvm, gfn);
1124         if (kvm_is_error_hva(addr))
1125                 return PAGE_SIZE;
1126
1127         down_read(&current->mm->mmap_sem);
1128         vma = find_vma(current->mm, addr);
1129         if (!vma)
1130                 goto out;
1131
1132         size = vma_kernel_pagesize(vma);
1133
1134 out:
1135         up_read(&current->mm->mmap_sem);
1136
1137         return size;
1138 }
1139
1140 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1141 {
1142         return slot->flags & KVM_MEM_READONLY;
1143 }
1144
1145 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1146                                        gfn_t *nr_pages, bool write)
1147 {
1148         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1149                 return KVM_HVA_ERR_BAD;
1150
1151         if (memslot_is_readonly(slot) && write)
1152                 return KVM_HVA_ERR_RO_BAD;
1153
1154         if (nr_pages)
1155                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1156
1157         return __gfn_to_hva_memslot(slot, gfn);
1158 }
1159
1160 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1161                                      gfn_t *nr_pages)
1162 {
1163         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1164 }
1165
1166 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1167                                         gfn_t gfn)
1168 {
1169         return gfn_to_hva_many(slot, gfn, NULL);
1170 }
1171 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1172
1173 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1174 {
1175         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1176 }
1177 EXPORT_SYMBOL_GPL(gfn_to_hva);
1178
1179 /*
1180  * If writable is set to false, the hva returned by this function is only
1181  * allowed to be read.
1182  */
1183 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1184                                       gfn_t gfn, bool *writable)
1185 {
1186         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1187
1188         if (!kvm_is_error_hva(hva) && writable)
1189                 *writable = !memslot_is_readonly(slot);
1190
1191         return hva;
1192 }
1193
1194 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1195 {
1196         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1197
1198         return gfn_to_hva_memslot_prot(slot, gfn, writable);
1199 }
1200
1201 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1202         unsigned long start, int write, struct page **page)
1203 {
1204         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1205
1206         if (write)
1207                 flags |= FOLL_WRITE;
1208
1209         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1210 }
1211
1212 static inline int check_user_page_hwpoison(unsigned long addr)
1213 {
1214         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1215
1216         rc = __get_user_pages(current, current->mm, addr, 1,
1217                               flags, NULL, NULL, NULL);
1218         return rc == -EHWPOISON;
1219 }
1220
1221 /*
1222  * The atomic path to get the writable pfn which will be stored in @pfn,
1223  * true indicates success, otherwise false is returned.
1224  */
1225 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1226                             bool write_fault, bool *writable, pfn_t *pfn)
1227 {
1228         struct page *page[1];
1229         int npages;
1230
1231         if (!(async || atomic))
1232                 return false;
1233
1234         /*
1235          * Fast pin a writable pfn only if it is a write fault request
1236          * or the caller allows to map a writable pfn for a read fault
1237          * request.
1238          */
1239         if (!(write_fault || writable))
1240                 return false;
1241
1242         npages = __get_user_pages_fast(addr, 1, 1, page);
1243         if (npages == 1) {
1244                 *pfn = page_to_pfn(page[0]);
1245
1246                 if (writable)
1247                         *writable = true;
1248                 return true;
1249         }
1250
1251         return false;
1252 }
1253
1254 /*
1255  * The slow path to get the pfn of the specified host virtual address,
1256  * 1 indicates success, -errno is returned if error is detected.
1257  */
1258 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1259                            bool *writable, pfn_t *pfn)
1260 {
1261         struct page *page[1];
1262         int npages = 0;
1263
1264         might_sleep();
1265
1266         if (writable)
1267                 *writable = write_fault;
1268
1269         if (async) {
1270                 down_read(&current->mm->mmap_sem);
1271                 npages = get_user_page_nowait(current, current->mm,
1272                                               addr, write_fault, page);
1273                 up_read(&current->mm->mmap_sem);
1274         } else
1275                 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1276                                                    write_fault, 0, page,
1277                                                    FOLL_TOUCH|FOLL_HWPOISON);
1278         if (npages != 1)
1279                 return npages;
1280
1281         /* map read fault as writable if possible */
1282         if (unlikely(!write_fault) && writable) {
1283                 struct page *wpage[1];
1284
1285                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1286                 if (npages == 1) {
1287                         *writable = true;
1288                         put_page(page[0]);
1289                         page[0] = wpage[0];
1290                 }
1291
1292                 npages = 1;
1293         }
1294         *pfn = page_to_pfn(page[0]);
1295         return npages;
1296 }
1297
1298 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1299 {
1300         if (unlikely(!(vma->vm_flags & VM_READ)))
1301                 return false;
1302
1303         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1304                 return false;
1305
1306         return true;
1307 }
1308
1309 /*
1310  * Pin guest page in memory and return its pfn.
1311  * @addr: host virtual address which maps memory to the guest
1312  * @atomic: whether this function can sleep
1313  * @async: whether this function need to wait IO complete if the
1314  *         host page is not in the memory
1315  * @write_fault: whether we should get a writable host page
1316  * @writable: whether it allows to map a writable host page for !@write_fault
1317  *
1318  * The function will map a writable host page for these two cases:
1319  * 1): @write_fault = true
1320  * 2): @write_fault = false && @writable, @writable will tell the caller
1321  *     whether the mapping is writable.
1322  */
1323 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1324                         bool write_fault, bool *writable)
1325 {
1326         struct vm_area_struct *vma;
1327         pfn_t pfn = 0;
1328         int npages;
1329
1330         /* we can do it either atomically or asynchronously, not both */
1331         BUG_ON(atomic && async);
1332
1333         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1334                 return pfn;
1335
1336         if (atomic)
1337                 return KVM_PFN_ERR_FAULT;
1338
1339         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1340         if (npages == 1)
1341                 return pfn;
1342
1343         down_read(&current->mm->mmap_sem);
1344         if (npages == -EHWPOISON ||
1345               (!async && check_user_page_hwpoison(addr))) {
1346                 pfn = KVM_PFN_ERR_HWPOISON;
1347                 goto exit;
1348         }
1349
1350         vma = find_vma_intersection(current->mm, addr, addr + 1);
1351
1352         if (vma == NULL)
1353                 pfn = KVM_PFN_ERR_FAULT;
1354         else if ((vma->vm_flags & VM_PFNMAP)) {
1355                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1356                         vma->vm_pgoff;
1357                 BUG_ON(!kvm_is_reserved_pfn(pfn));
1358         } else {
1359                 if (async && vma_is_valid(vma, write_fault))
1360                         *async = true;
1361                 pfn = KVM_PFN_ERR_FAULT;
1362         }
1363 exit:
1364         up_read(&current->mm->mmap_sem);
1365         return pfn;
1366 }
1367
1368 pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1369                            bool *async, bool write_fault, bool *writable)
1370 {
1371         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1372
1373         if (addr == KVM_HVA_ERR_RO_BAD)
1374                 return KVM_PFN_ERR_RO_FAULT;
1375
1376         if (kvm_is_error_hva(addr))
1377                 return KVM_PFN_NOSLOT;
1378
1379         /* Do not map writable pfn in the readonly memslot. */
1380         if (writable && memslot_is_readonly(slot)) {
1381                 *writable = false;
1382                 writable = NULL;
1383         }
1384
1385         return hva_to_pfn(addr, atomic, async, write_fault,
1386                           writable);
1387 }
1388 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1389
1390 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic,
1391                           bool write_fault, bool *writable)
1392 {
1393         struct kvm_memory_slot *slot;
1394
1395         slot = gfn_to_memslot(kvm, gfn);
1396
1397         return __gfn_to_pfn_memslot(slot, gfn, atomic, NULL, write_fault,
1398                                     writable);
1399 }
1400
1401 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1402 {
1403         return __gfn_to_pfn(kvm, gfn, true, true, NULL);
1404 }
1405 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1406
1407 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1408 {
1409         return __gfn_to_pfn(kvm, gfn, false, true, NULL);
1410 }
1411 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1412
1413 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1414                       bool *writable)
1415 {
1416         return __gfn_to_pfn(kvm, gfn, false, write_fault, writable);
1417 }
1418 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1419
1420 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1421 {
1422         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1423 }
1424
1425 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1426 {
1427         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1428 }
1429 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1430
1431 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1432                                                                   int nr_pages)
1433 {
1434         unsigned long addr;
1435         gfn_t entry;
1436
1437         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1438         if (kvm_is_error_hva(addr))
1439                 return -1;
1440
1441         if (entry < nr_pages)
1442                 return 0;
1443
1444         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1445 }
1446 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1447
1448 static struct page *kvm_pfn_to_page(pfn_t pfn)
1449 {
1450         if (is_error_noslot_pfn(pfn))
1451                 return KVM_ERR_PTR_BAD_PAGE;
1452
1453         if (kvm_is_reserved_pfn(pfn)) {
1454                 WARN_ON(1);
1455                 return KVM_ERR_PTR_BAD_PAGE;
1456         }
1457
1458         return pfn_to_page(pfn);
1459 }
1460
1461 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1462 {
1463         pfn_t pfn;
1464
1465         pfn = gfn_to_pfn(kvm, gfn);
1466
1467         return kvm_pfn_to_page(pfn);
1468 }
1469 EXPORT_SYMBOL_GPL(gfn_to_page);
1470
1471 void kvm_release_page_clean(struct page *page)
1472 {
1473         WARN_ON(is_error_page(page));
1474
1475         kvm_release_pfn_clean(page_to_pfn(page));
1476 }
1477 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1478
1479 void kvm_release_pfn_clean(pfn_t pfn)
1480 {
1481         if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1482                 put_page(pfn_to_page(pfn));
1483 }
1484 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1485
1486 void kvm_release_page_dirty(struct page *page)
1487 {
1488         WARN_ON(is_error_page(page));
1489
1490         kvm_release_pfn_dirty(page_to_pfn(page));
1491 }
1492 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1493
1494 static void kvm_release_pfn_dirty(pfn_t pfn)
1495 {
1496         kvm_set_pfn_dirty(pfn);
1497         kvm_release_pfn_clean(pfn);
1498 }
1499
1500 void kvm_set_pfn_dirty(pfn_t pfn)
1501 {
1502         if (!kvm_is_reserved_pfn(pfn)) {
1503                 struct page *page = pfn_to_page(pfn);
1504
1505                 if (!PageReserved(page))
1506                         SetPageDirty(page);
1507         }
1508 }
1509 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1510
1511 void kvm_set_pfn_accessed(pfn_t pfn)
1512 {
1513         if (!kvm_is_reserved_pfn(pfn))
1514                 mark_page_accessed(pfn_to_page(pfn));
1515 }
1516 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1517
1518 void kvm_get_pfn(pfn_t pfn)
1519 {
1520         if (!kvm_is_reserved_pfn(pfn))
1521                 get_page(pfn_to_page(pfn));
1522 }
1523 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1524
1525 static int next_segment(unsigned long len, int offset)
1526 {
1527         if (len > PAGE_SIZE - offset)
1528                 return PAGE_SIZE - offset;
1529         else
1530                 return len;
1531 }
1532
1533 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1534                         int len)
1535 {
1536         int r;
1537         unsigned long addr;
1538
1539         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1540         if (kvm_is_error_hva(addr))
1541                 return -EFAULT;
1542         r = __copy_from_user(data, (void __user *)addr + offset, len);
1543         if (r)
1544                 return -EFAULT;
1545         return 0;
1546 }
1547 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1548
1549 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1550 {
1551         gfn_t gfn = gpa >> PAGE_SHIFT;
1552         int seg;
1553         int offset = offset_in_page(gpa);
1554         int ret;
1555
1556         while ((seg = next_segment(len, offset)) != 0) {
1557                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1558                 if (ret < 0)
1559                         return ret;
1560                 offset = 0;
1561                 len -= seg;
1562                 data += seg;
1563                 ++gfn;
1564         }
1565         return 0;
1566 }
1567 EXPORT_SYMBOL_GPL(kvm_read_guest);
1568
1569 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1570                           unsigned long len)
1571 {
1572         int r;
1573         unsigned long addr;
1574         gfn_t gfn = gpa >> PAGE_SHIFT;
1575         int offset = offset_in_page(gpa);
1576
1577         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1578         if (kvm_is_error_hva(addr))
1579                 return -EFAULT;
1580         pagefault_disable();
1581         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1582         pagefault_enable();
1583         if (r)
1584                 return -EFAULT;
1585         return 0;
1586 }
1587 EXPORT_SYMBOL(kvm_read_guest_atomic);
1588
1589 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1590                          int offset, int len)
1591 {
1592         int r;
1593         struct kvm_memory_slot *memslot;
1594         unsigned long addr;
1595
1596         memslot = gfn_to_memslot(kvm, gfn);
1597         addr = gfn_to_hva_memslot(memslot, gfn);
1598         if (kvm_is_error_hva(addr))
1599                 return -EFAULT;
1600         r = __copy_to_user((void __user *)addr + offset, data, len);
1601         if (r)
1602                 return -EFAULT;
1603         mark_page_dirty_in_slot(kvm, memslot, gfn);
1604         return 0;
1605 }
1606 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1607
1608 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1609                     unsigned long len)
1610 {
1611         gfn_t gfn = gpa >> PAGE_SHIFT;
1612         int seg;
1613         int offset = offset_in_page(gpa);
1614         int ret;
1615
1616         while ((seg = next_segment(len, offset)) != 0) {
1617                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1618                 if (ret < 0)
1619                         return ret;
1620                 offset = 0;
1621                 len -= seg;
1622                 data += seg;
1623                 ++gfn;
1624         }
1625         return 0;
1626 }
1627 EXPORT_SYMBOL_GPL(kvm_write_guest);
1628
1629 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1630                               gpa_t gpa, unsigned long len)
1631 {
1632         struct kvm_memslots *slots = kvm_memslots(kvm);
1633         int offset = offset_in_page(gpa);
1634         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1635         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1636         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1637         gfn_t nr_pages_avail;
1638
1639         ghc->gpa = gpa;
1640         ghc->generation = slots->generation;
1641         ghc->len = len;
1642         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1643         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1644         if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1645                 ghc->hva += offset;
1646         } else {
1647                 /*
1648                  * If the requested region crosses two memslots, we still
1649                  * verify that the entire region is valid here.
1650                  */
1651                 while (start_gfn <= end_gfn) {
1652                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1653                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1654                                                    &nr_pages_avail);
1655                         if (kvm_is_error_hva(ghc->hva))
1656                                 return -EFAULT;
1657                         start_gfn += nr_pages_avail;
1658                 }
1659                 /* Use the slow path for cross page reads and writes. */
1660                 ghc->memslot = NULL;
1661         }
1662         return 0;
1663 }
1664 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1665
1666 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1667                            void *data, unsigned long len)
1668 {
1669         struct kvm_memslots *slots = kvm_memslots(kvm);
1670         int r;
1671
1672         BUG_ON(len > ghc->len);
1673
1674         if (slots->generation != ghc->generation)
1675                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1676
1677         if (unlikely(!ghc->memslot))
1678                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1679
1680         if (kvm_is_error_hva(ghc->hva))
1681                 return -EFAULT;
1682
1683         r = __copy_to_user((void __user *)ghc->hva, data, len);
1684         if (r)
1685                 return -EFAULT;
1686         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1687
1688         return 0;
1689 }
1690 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1691
1692 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1693                            void *data, unsigned long len)
1694 {
1695         struct kvm_memslots *slots = kvm_memslots(kvm);
1696         int r;
1697
1698         BUG_ON(len > ghc->len);
1699
1700         if (slots->generation != ghc->generation)
1701                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1702
1703         if (unlikely(!ghc->memslot))
1704                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1705
1706         if (kvm_is_error_hva(ghc->hva))
1707                 return -EFAULT;
1708
1709         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1710         if (r)
1711                 return -EFAULT;
1712
1713         return 0;
1714 }
1715 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1716
1717 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1718 {
1719         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1720
1721         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1722 }
1723 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1724
1725 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1726 {
1727         gfn_t gfn = gpa >> PAGE_SHIFT;
1728         int seg;
1729         int offset = offset_in_page(gpa);
1730         int ret;
1731
1732         while ((seg = next_segment(len, offset)) != 0) {
1733                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1734                 if (ret < 0)
1735                         return ret;
1736                 offset = 0;
1737                 len -= seg;
1738                 ++gfn;
1739         }
1740         return 0;
1741 }
1742 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1743
1744 static void mark_page_dirty_in_slot(struct kvm *kvm,
1745                                     struct kvm_memory_slot *memslot,
1746                                     gfn_t gfn)
1747 {
1748         if (memslot && memslot->dirty_bitmap) {
1749                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1750
1751                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1752         }
1753 }
1754
1755 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1756 {
1757         struct kvm_memory_slot *memslot;
1758
1759         memslot = gfn_to_memslot(kvm, gfn);
1760         mark_page_dirty_in_slot(kvm, memslot, gfn);
1761 }
1762 EXPORT_SYMBOL_GPL(mark_page_dirty);
1763
1764 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1765 {
1766         if (kvm_arch_vcpu_runnable(vcpu)) {
1767                 kvm_make_request(KVM_REQ_UNHALT, vcpu);
1768                 return -EINTR;
1769         }
1770         if (kvm_cpu_has_pending_timer(vcpu))
1771                 return -EINTR;
1772         if (signal_pending(current))
1773                 return -EINTR;
1774
1775         return 0;
1776 }
1777
1778 /*
1779  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1780  */
1781 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1782 {
1783         ktime_t start, cur;
1784         DEFINE_WAIT(wait);
1785         bool waited = false;
1786
1787         start = cur = ktime_get();
1788         if (halt_poll_ns) {
1789                 ktime_t stop = ktime_add_ns(ktime_get(), halt_poll_ns);
1790
1791                 do {
1792                         /*
1793                          * This sets KVM_REQ_UNHALT if an interrupt
1794                          * arrives.
1795                          */
1796                         if (kvm_vcpu_check_block(vcpu) < 0) {
1797                                 ++vcpu->stat.halt_successful_poll;
1798                                 goto out;
1799                         }
1800                         cur = ktime_get();
1801                 } while (single_task_running() && ktime_before(cur, stop));
1802         }
1803
1804         for (;;) {
1805                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1806
1807                 if (kvm_vcpu_check_block(vcpu) < 0)
1808                         break;
1809
1810                 waited = true;
1811                 schedule();
1812         }
1813
1814         finish_wait(&vcpu->wq, &wait);
1815         cur = ktime_get();
1816
1817 out:
1818         trace_kvm_vcpu_wakeup(ktime_to_ns(cur) - ktime_to_ns(start), waited);
1819 }
1820 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1821
1822 #ifndef CONFIG_S390
1823 /*
1824  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1825  */
1826 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1827 {
1828         int me;
1829         int cpu = vcpu->cpu;
1830         wait_queue_head_t *wqp;
1831
1832         wqp = kvm_arch_vcpu_wq(vcpu);
1833         if (waitqueue_active(wqp)) {
1834                 wake_up_interruptible(wqp);
1835                 ++vcpu->stat.halt_wakeup;
1836         }
1837
1838         me = get_cpu();
1839         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1840                 if (kvm_arch_vcpu_should_kick(vcpu))
1841                         smp_send_reschedule(cpu);
1842         put_cpu();
1843 }
1844 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1845 #endif /* !CONFIG_S390 */
1846
1847 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
1848 {
1849         struct pid *pid;
1850         struct task_struct *task = NULL;
1851         int ret = 0;
1852
1853         rcu_read_lock();
1854         pid = rcu_dereference(target->pid);
1855         if (pid)
1856                 task = get_pid_task(pid, PIDTYPE_PID);
1857         rcu_read_unlock();
1858         if (!task)
1859                 return ret;
1860         ret = yield_to(task, 1);
1861         put_task_struct(task);
1862
1863         return ret;
1864 }
1865 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1866
1867 /*
1868  * Helper that checks whether a VCPU is eligible for directed yield.
1869  * Most eligible candidate to yield is decided by following heuristics:
1870  *
1871  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1872  *  (preempted lock holder), indicated by @in_spin_loop.
1873  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1874  *
1875  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1876  *  chance last time (mostly it has become eligible now since we have probably
1877  *  yielded to lockholder in last iteration. This is done by toggling
1878  *  @dy_eligible each time a VCPU checked for eligibility.)
1879  *
1880  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1881  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1882  *  burning. Giving priority for a potential lock-holder increases lock
1883  *  progress.
1884  *
1885  *  Since algorithm is based on heuristics, accessing another VCPU data without
1886  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1887  *  and continue with next VCPU and so on.
1888  */
1889 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1890 {
1891 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1892         bool eligible;
1893
1894         eligible = !vcpu->spin_loop.in_spin_loop ||
1895                     vcpu->spin_loop.dy_eligible;
1896
1897         if (vcpu->spin_loop.in_spin_loop)
1898                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1899
1900         return eligible;
1901 #else
1902         return true;
1903 #endif
1904 }
1905
1906 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1907 {
1908         struct kvm *kvm = me->kvm;
1909         struct kvm_vcpu *vcpu;
1910         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1911         int yielded = 0;
1912         int try = 3;
1913         int pass;
1914         int i;
1915
1916         kvm_vcpu_set_in_spin_loop(me, true);
1917         /*
1918          * We boost the priority of a VCPU that is runnable but not
1919          * currently running, because it got preempted by something
1920          * else and called schedule in __vcpu_run.  Hopefully that
1921          * VCPU is holding the lock that we need and will release it.
1922          * We approximate round-robin by starting at the last boosted VCPU.
1923          */
1924         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1925                 kvm_for_each_vcpu(i, vcpu, kvm) {
1926                         if (!pass && i <= last_boosted_vcpu) {
1927                                 i = last_boosted_vcpu;
1928                                 continue;
1929                         } else if (pass && i > last_boosted_vcpu)
1930                                 break;
1931                         if (!ACCESS_ONCE(vcpu->preempted))
1932                                 continue;
1933                         if (vcpu == me)
1934                                 continue;
1935                         if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1936                                 continue;
1937                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1938                                 continue;
1939
1940                         yielded = kvm_vcpu_yield_to(vcpu);
1941                         if (yielded > 0) {
1942                                 kvm->last_boosted_vcpu = i;
1943                                 break;
1944                         } else if (yielded < 0) {
1945                                 try--;
1946                                 if (!try)
1947                                         break;
1948                         }
1949                 }
1950         }
1951         kvm_vcpu_set_in_spin_loop(me, false);
1952
1953         /* Ensure vcpu is not eligible during next spinloop */
1954         kvm_vcpu_set_dy_eligible(me, false);
1955 }
1956 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1957
1958 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1959 {
1960         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1961         struct page *page;
1962
1963         if (vmf->pgoff == 0)
1964                 page = virt_to_page(vcpu->run);
1965 #ifdef CONFIG_X86
1966         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1967                 page = virt_to_page(vcpu->arch.pio_data);
1968 #endif
1969 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1970         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1971                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1972 #endif
1973         else
1974                 return kvm_arch_vcpu_fault(vcpu, vmf);
1975         get_page(page);
1976         vmf->page = page;
1977         return 0;
1978 }
1979
1980 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1981         .fault = kvm_vcpu_fault,
1982 };
1983
1984 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1985 {
1986         vma->vm_ops = &kvm_vcpu_vm_ops;
1987         return 0;
1988 }
1989
1990 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1991 {
1992         struct kvm_vcpu *vcpu = filp->private_data;
1993
1994         kvm_put_kvm(vcpu->kvm);
1995         return 0;
1996 }
1997
1998 static struct file_operations kvm_vcpu_fops = {
1999         .release        = kvm_vcpu_release,
2000         .unlocked_ioctl = kvm_vcpu_ioctl,
2001 #ifdef CONFIG_KVM_COMPAT
2002         .compat_ioctl   = kvm_vcpu_compat_ioctl,
2003 #endif
2004         .mmap           = kvm_vcpu_mmap,
2005         .llseek         = noop_llseek,
2006 };
2007
2008 /*
2009  * Allocates an inode for the vcpu.
2010  */
2011 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2012 {
2013         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2014 }
2015
2016 /*
2017  * Creates some virtual cpus.  Good luck creating more than one.
2018  */
2019 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2020 {
2021         int r;
2022         struct kvm_vcpu *vcpu, *v;
2023
2024         if (id >= KVM_MAX_VCPUS)
2025                 return -EINVAL;
2026
2027         vcpu = kvm_arch_vcpu_create(kvm, id);
2028         if (IS_ERR(vcpu))
2029                 return PTR_ERR(vcpu);
2030
2031         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2032
2033         r = kvm_arch_vcpu_setup(vcpu);
2034         if (r)
2035                 goto vcpu_destroy;
2036
2037         mutex_lock(&kvm->lock);
2038         if (!kvm_vcpu_compatible(vcpu)) {
2039                 r = -EINVAL;
2040                 goto unlock_vcpu_destroy;
2041         }
2042         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2043                 r = -EINVAL;
2044                 goto unlock_vcpu_destroy;
2045         }
2046
2047         kvm_for_each_vcpu(r, v, kvm)
2048                 if (v->vcpu_id == id) {
2049                         r = -EEXIST;
2050                         goto unlock_vcpu_destroy;
2051                 }
2052
2053         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2054
2055         /* Now it's all set up, let userspace reach it */
2056         kvm_get_kvm(kvm);
2057         r = create_vcpu_fd(vcpu);
2058         if (r < 0) {
2059                 kvm_put_kvm(kvm);
2060                 goto unlock_vcpu_destroy;
2061         }
2062
2063         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2064         smp_wmb();
2065         atomic_inc(&kvm->online_vcpus);
2066
2067         mutex_unlock(&kvm->lock);
2068         kvm_arch_vcpu_postcreate(vcpu);
2069         return r;
2070
2071 unlock_vcpu_destroy:
2072         mutex_unlock(&kvm->lock);
2073 vcpu_destroy:
2074         kvm_arch_vcpu_destroy(vcpu);
2075         return r;
2076 }
2077
2078 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2079 {
2080         if (sigset) {
2081                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2082                 vcpu->sigset_active = 1;
2083                 vcpu->sigset = *sigset;
2084         } else
2085                 vcpu->sigset_active = 0;
2086         return 0;
2087 }
2088
2089 static long kvm_vcpu_ioctl(struct file *filp,
2090                            unsigned int ioctl, unsigned long arg)
2091 {
2092         struct kvm_vcpu *vcpu = filp->private_data;
2093         void __user *argp = (void __user *)arg;
2094         int r;
2095         struct kvm_fpu *fpu = NULL;
2096         struct kvm_sregs *kvm_sregs = NULL;
2097
2098         if (vcpu->kvm->mm != current->mm)
2099                 return -EIO;
2100
2101         if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2102                 return -EINVAL;
2103
2104 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2105         /*
2106          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2107          * so vcpu_load() would break it.
2108          */
2109         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2110                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2111 #endif
2112
2113
2114         r = vcpu_load(vcpu);
2115         if (r)
2116                 return r;
2117         switch (ioctl) {
2118         case KVM_RUN:
2119                 r = -EINVAL;
2120                 if (arg)
2121                         goto out;
2122                 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2123                         /* The thread running this VCPU changed. */
2124                         struct pid *oldpid = vcpu->pid;
2125                         struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2126
2127                         rcu_assign_pointer(vcpu->pid, newpid);
2128                         if (oldpid)
2129                                 synchronize_rcu();
2130                         put_pid(oldpid);
2131                 }
2132                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2133                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2134                 break;
2135         case KVM_GET_REGS: {
2136                 struct kvm_regs *kvm_regs;
2137
2138                 r = -ENOMEM;
2139                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2140                 if (!kvm_regs)
2141                         goto out;
2142                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2143                 if (r)
2144                         goto out_free1;
2145                 r = -EFAULT;
2146                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2147                         goto out_free1;
2148                 r = 0;
2149 out_free1:
2150                 kfree(kvm_regs);
2151                 break;
2152         }
2153         case KVM_SET_REGS: {
2154                 struct kvm_regs *kvm_regs;
2155
2156                 r = -ENOMEM;
2157                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2158                 if (IS_ERR(kvm_regs)) {
2159                         r = PTR_ERR(kvm_regs);
2160                         goto out;
2161                 }
2162                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2163                 kfree(kvm_regs);
2164                 break;
2165         }
2166         case KVM_GET_SREGS: {
2167                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2168                 r = -ENOMEM;
2169                 if (!kvm_sregs)
2170                         goto out;
2171                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2172                 if (r)
2173                         goto out;
2174                 r = -EFAULT;
2175                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2176                         goto out;
2177                 r = 0;
2178                 break;
2179         }
2180         case KVM_SET_SREGS: {
2181                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2182                 if (IS_ERR(kvm_sregs)) {
2183                         r = PTR_ERR(kvm_sregs);
2184                         kvm_sregs = NULL;
2185                         goto out;
2186                 }
2187                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2188                 break;
2189         }
2190         case KVM_GET_MP_STATE: {
2191                 struct kvm_mp_state mp_state;
2192
2193                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2194                 if (r)
2195                         goto out;
2196                 r = -EFAULT;
2197                 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2198                         goto out;
2199                 r = 0;
2200                 break;
2201         }
2202         case KVM_SET_MP_STATE: {
2203                 struct kvm_mp_state mp_state;
2204
2205                 r = -EFAULT;
2206                 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2207                         goto out;
2208                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2209                 break;
2210         }
2211         case KVM_TRANSLATE: {
2212                 struct kvm_translation tr;
2213
2214                 r = -EFAULT;
2215                 if (copy_from_user(&tr, argp, sizeof(tr)))
2216                         goto out;
2217                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2218                 if (r)
2219                         goto out;
2220                 r = -EFAULT;
2221                 if (copy_to_user(argp, &tr, sizeof(tr)))
2222                         goto out;
2223                 r = 0;
2224                 break;
2225         }
2226         case KVM_SET_GUEST_DEBUG: {
2227                 struct kvm_guest_debug dbg;
2228
2229                 r = -EFAULT;
2230                 if (copy_from_user(&dbg, argp, sizeof(dbg)))
2231                         goto out;
2232                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2233                 break;
2234         }
2235         case KVM_SET_SIGNAL_MASK: {
2236                 struct kvm_signal_mask __user *sigmask_arg = argp;
2237                 struct kvm_signal_mask kvm_sigmask;
2238                 sigset_t sigset, *p;
2239
2240                 p = NULL;
2241                 if (argp) {
2242                         r = -EFAULT;
2243                         if (copy_from_user(&kvm_sigmask, argp,
2244                                            sizeof(kvm_sigmask)))
2245                                 goto out;
2246                         r = -EINVAL;
2247                         if (kvm_sigmask.len != sizeof(sigset))
2248                                 goto out;
2249                         r = -EFAULT;
2250                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2251                                            sizeof(sigset)))
2252                                 goto out;
2253                         p = &sigset;
2254                 }
2255                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2256                 break;
2257         }
2258         case KVM_GET_FPU: {
2259                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2260                 r = -ENOMEM;
2261                 if (!fpu)
2262                         goto out;
2263                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2264                 if (r)
2265                         goto out;
2266                 r = -EFAULT;
2267                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2268                         goto out;
2269                 r = 0;
2270                 break;
2271         }
2272         case KVM_SET_FPU: {
2273                 fpu = memdup_user(argp, sizeof(*fpu));
2274                 if (IS_ERR(fpu)) {
2275                         r = PTR_ERR(fpu);
2276                         fpu = NULL;
2277                         goto out;
2278                 }
2279                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2280                 break;
2281         }
2282         default:
2283                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2284         }
2285 out:
2286         vcpu_put(vcpu);
2287         kfree(fpu);
2288         kfree(kvm_sregs);
2289         return r;
2290 }
2291
2292 #ifdef CONFIG_KVM_COMPAT
2293 static long kvm_vcpu_compat_ioctl(struct file *filp,
2294                                   unsigned int ioctl, unsigned long arg)
2295 {
2296         struct kvm_vcpu *vcpu = filp->private_data;
2297         void __user *argp = compat_ptr(arg);
2298         int r;
2299
2300         if (vcpu->kvm->mm != current->mm)
2301                 return -EIO;
2302
2303         switch (ioctl) {
2304         case KVM_SET_SIGNAL_MASK: {
2305                 struct kvm_signal_mask __user *sigmask_arg = argp;
2306                 struct kvm_signal_mask kvm_sigmask;
2307                 compat_sigset_t csigset;
2308                 sigset_t sigset;
2309
2310                 if (argp) {
2311                         r = -EFAULT;
2312                         if (copy_from_user(&kvm_sigmask, argp,
2313                                            sizeof(kvm_sigmask)))
2314                                 goto out;
2315                         r = -EINVAL;
2316                         if (kvm_sigmask.len != sizeof(csigset))
2317                                 goto out;
2318                         r = -EFAULT;
2319                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2320                                            sizeof(csigset)))
2321                                 goto out;
2322                         sigset_from_compat(&sigset, &csigset);
2323                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2324                 } else
2325                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2326                 break;
2327         }
2328         default:
2329                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2330         }
2331
2332 out:
2333         return r;
2334 }
2335 #endif
2336
2337 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2338                                  int (*accessor)(struct kvm_device *dev,
2339                                                  struct kvm_device_attr *attr),
2340                                  unsigned long arg)
2341 {
2342         struct kvm_device_attr attr;
2343
2344         if (!accessor)
2345                 return -EPERM;
2346
2347         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2348                 return -EFAULT;
2349
2350         return accessor(dev, &attr);
2351 }
2352
2353 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2354                              unsigned long arg)
2355 {
2356         struct kvm_device *dev = filp->private_data;
2357
2358         switch (ioctl) {
2359         case KVM_SET_DEVICE_ATTR:
2360                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2361         case KVM_GET_DEVICE_ATTR:
2362                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2363         case KVM_HAS_DEVICE_ATTR:
2364                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2365         default:
2366                 if (dev->ops->ioctl)
2367                         return dev->ops->ioctl(dev, ioctl, arg);
2368
2369                 return -ENOTTY;
2370         }
2371 }
2372
2373 static int kvm_device_release(struct inode *inode, struct file *filp)
2374 {
2375         struct kvm_device *dev = filp->private_data;
2376         struct kvm *kvm = dev->kvm;
2377
2378         kvm_put_kvm(kvm);
2379         return 0;
2380 }
2381
2382 static const struct file_operations kvm_device_fops = {
2383         .unlocked_ioctl = kvm_device_ioctl,
2384 #ifdef CONFIG_KVM_COMPAT
2385         .compat_ioctl = kvm_device_ioctl,
2386 #endif
2387         .release = kvm_device_release,
2388 };
2389
2390 struct kvm_device *kvm_device_from_filp(struct file *filp)
2391 {
2392         if (filp->f_op != &kvm_device_fops)
2393                 return NULL;
2394
2395         return filp->private_data;
2396 }
2397
2398 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2399 #ifdef CONFIG_KVM_MPIC
2400         [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
2401         [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
2402 #endif
2403
2404 #ifdef CONFIG_KVM_XICS
2405         [KVM_DEV_TYPE_XICS]             = &kvm_xics_ops,
2406 #endif
2407 };
2408
2409 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2410 {
2411         if (type >= ARRAY_SIZE(kvm_device_ops_table))
2412                 return -ENOSPC;
2413
2414         if (kvm_device_ops_table[type] != NULL)
2415                 return -EEXIST;
2416
2417         kvm_device_ops_table[type] = ops;
2418         return 0;
2419 }
2420
2421 void kvm_unregister_device_ops(u32 type)
2422 {
2423         if (kvm_device_ops_table[type] != NULL)
2424                 kvm_device_ops_table[type] = NULL;
2425 }
2426
2427 static int kvm_ioctl_create_device(struct kvm *kvm,
2428                                    struct kvm_create_device *cd)
2429 {
2430         struct kvm_device_ops *ops = NULL;
2431         struct kvm_device *dev;
2432         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2433         int ret;
2434
2435         if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2436                 return -ENODEV;
2437
2438         ops = kvm_device_ops_table[cd->type];
2439         if (ops == NULL)
2440                 return -ENODEV;
2441
2442         if (test)
2443                 return 0;
2444
2445         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2446         if (!dev)
2447                 return -ENOMEM;
2448
2449         dev->ops = ops;
2450         dev->kvm = kvm;
2451
2452         ret = ops->create(dev, cd->type);
2453         if (ret < 0) {
2454                 kfree(dev);
2455                 return ret;
2456         }
2457
2458         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2459         if (ret < 0) {
2460                 ops->destroy(dev);
2461                 return ret;
2462         }
2463
2464         list_add(&dev->vm_node, &kvm->devices);
2465         kvm_get_kvm(kvm);
2466         cd->fd = ret;
2467         return 0;
2468 }
2469
2470 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2471 {
2472         switch (arg) {
2473         case KVM_CAP_USER_MEMORY:
2474         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2475         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2476 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2477         case KVM_CAP_SET_BOOT_CPU_ID:
2478 #endif
2479         case KVM_CAP_INTERNAL_ERROR_DATA:
2480 #ifdef CONFIG_HAVE_KVM_MSI
2481         case KVM_CAP_SIGNAL_MSI:
2482 #endif
2483 #ifdef CONFIG_HAVE_KVM_IRQFD
2484         case KVM_CAP_IRQFD:
2485         case KVM_CAP_IRQFD_RESAMPLE:
2486 #endif
2487         case KVM_CAP_CHECK_EXTENSION_VM:
2488                 return 1;
2489 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2490         case KVM_CAP_IRQ_ROUTING:
2491                 return KVM_MAX_IRQ_ROUTES;
2492 #endif
2493         default:
2494                 break;
2495         }
2496         return kvm_vm_ioctl_check_extension(kvm, arg);
2497 }
2498
2499 static long kvm_vm_ioctl(struct file *filp,
2500                            unsigned int ioctl, unsigned long arg)
2501 {
2502         struct kvm *kvm = filp->private_data;
2503         void __user *argp = (void __user *)arg;
2504         int r;
2505
2506         if (kvm->mm != current->mm)
2507                 return -EIO;
2508         switch (ioctl) {
2509         case KVM_CREATE_VCPU:
2510                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2511                 break;
2512         case KVM_SET_USER_MEMORY_REGION: {
2513                 struct kvm_userspace_memory_region kvm_userspace_mem;
2514
2515                 r = -EFAULT;
2516                 if (copy_from_user(&kvm_userspace_mem, argp,
2517                                                 sizeof(kvm_userspace_mem)))
2518                         goto out;
2519
2520                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2521                 break;
2522         }
2523         case KVM_GET_DIRTY_LOG: {
2524                 struct kvm_dirty_log log;
2525
2526                 r = -EFAULT;
2527                 if (copy_from_user(&log, argp, sizeof(log)))
2528                         goto out;
2529                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2530                 break;
2531         }
2532 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2533         case KVM_REGISTER_COALESCED_MMIO: {
2534                 struct kvm_coalesced_mmio_zone zone;
2535
2536                 r = -EFAULT;
2537                 if (copy_from_user(&zone, argp, sizeof(zone)))
2538                         goto out;
2539                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2540                 break;
2541         }
2542         case KVM_UNREGISTER_COALESCED_MMIO: {
2543                 struct kvm_coalesced_mmio_zone zone;
2544
2545                 r = -EFAULT;
2546                 if (copy_from_user(&zone, argp, sizeof(zone)))
2547                         goto out;
2548                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2549                 break;
2550         }
2551 #endif
2552         case KVM_IRQFD: {
2553                 struct kvm_irqfd data;
2554
2555                 r = -EFAULT;
2556                 if (copy_from_user(&data, argp, sizeof(data)))
2557                         goto out;
2558                 r = kvm_irqfd(kvm, &data);
2559                 break;
2560         }
2561         case KVM_IOEVENTFD: {
2562                 struct kvm_ioeventfd data;
2563
2564                 r = -EFAULT;
2565                 if (copy_from_user(&data, argp, sizeof(data)))
2566                         goto out;
2567                 r = kvm_ioeventfd(kvm, &data);
2568                 break;
2569         }
2570 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2571         case KVM_SET_BOOT_CPU_ID:
2572                 r = 0;
2573                 mutex_lock(&kvm->lock);
2574                 if (atomic_read(&kvm->online_vcpus) != 0)
2575                         r = -EBUSY;
2576                 else
2577                         kvm->bsp_vcpu_id = arg;
2578                 mutex_unlock(&kvm->lock);
2579                 break;
2580 #endif
2581 #ifdef CONFIG_HAVE_KVM_MSI
2582         case KVM_SIGNAL_MSI: {
2583                 struct kvm_msi msi;
2584
2585                 r = -EFAULT;
2586                 if (copy_from_user(&msi, argp, sizeof(msi)))
2587                         goto out;
2588                 r = kvm_send_userspace_msi(kvm, &msi);
2589                 break;
2590         }
2591 #endif
2592 #ifdef __KVM_HAVE_IRQ_LINE
2593         case KVM_IRQ_LINE_STATUS:
2594         case KVM_IRQ_LINE: {
2595                 struct kvm_irq_level irq_event;
2596
2597                 r = -EFAULT;
2598                 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2599                         goto out;
2600
2601                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2602                                         ioctl == KVM_IRQ_LINE_STATUS);
2603                 if (r)
2604                         goto out;
2605
2606                 r = -EFAULT;
2607                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2608                         if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2609                                 goto out;
2610                 }
2611
2612                 r = 0;
2613                 break;
2614         }
2615 #endif
2616 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2617         case KVM_SET_GSI_ROUTING: {
2618                 struct kvm_irq_routing routing;
2619                 struct kvm_irq_routing __user *urouting;
2620                 struct kvm_irq_routing_entry *entries;
2621
2622                 r = -EFAULT;
2623                 if (copy_from_user(&routing, argp, sizeof(routing)))
2624                         goto out;
2625                 r = -EINVAL;
2626                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2627                         goto out;
2628                 if (routing.flags)
2629                         goto out;
2630                 r = -ENOMEM;
2631                 entries = vmalloc(routing.nr * sizeof(*entries));
2632                 if (!entries)
2633                         goto out;
2634                 r = -EFAULT;
2635                 urouting = argp;
2636                 if (copy_from_user(entries, urouting->entries,
2637                                    routing.nr * sizeof(*entries)))
2638                         goto out_free_irq_routing;
2639                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2640                                         routing.flags);
2641 out_free_irq_routing:
2642                 vfree(entries);
2643                 break;
2644         }
2645 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2646         case KVM_CREATE_DEVICE: {
2647                 struct kvm_create_device cd;
2648
2649                 r = -EFAULT;
2650                 if (copy_from_user(&cd, argp, sizeof(cd)))
2651                         goto out;
2652
2653                 r = kvm_ioctl_create_device(kvm, &cd);
2654                 if (r)
2655                         goto out;
2656
2657                 r = -EFAULT;
2658                 if (copy_to_user(argp, &cd, sizeof(cd)))
2659                         goto out;
2660
2661                 r = 0;
2662                 break;
2663         }
2664         case KVM_CHECK_EXTENSION:
2665                 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2666                 break;
2667         default:
2668                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2669         }
2670 out:
2671         return r;
2672 }
2673
2674 #ifdef CONFIG_KVM_COMPAT
2675 struct compat_kvm_dirty_log {
2676         __u32 slot;
2677         __u32 padding1;
2678         union {
2679                 compat_uptr_t dirty_bitmap; /* one bit per page */
2680                 __u64 padding2;
2681         };
2682 };
2683
2684 static long kvm_vm_compat_ioctl(struct file *filp,
2685                            unsigned int ioctl, unsigned long arg)
2686 {
2687         struct kvm *kvm = filp->private_data;
2688         int r;
2689
2690         if (kvm->mm != current->mm)
2691                 return -EIO;
2692         switch (ioctl) {
2693         case KVM_GET_DIRTY_LOG: {
2694                 struct compat_kvm_dirty_log compat_log;
2695                 struct kvm_dirty_log log;
2696
2697                 r = -EFAULT;
2698                 if (copy_from_user(&compat_log, (void __user *)arg,
2699                                    sizeof(compat_log)))
2700                         goto out;
2701                 log.slot         = compat_log.slot;
2702                 log.padding1     = compat_log.padding1;
2703                 log.padding2     = compat_log.padding2;
2704                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2705
2706                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2707                 break;
2708         }
2709         default:
2710                 r = kvm_vm_ioctl(filp, ioctl, arg);
2711         }
2712
2713 out:
2714         return r;
2715 }
2716 #endif
2717
2718 static struct file_operations kvm_vm_fops = {
2719         .release        = kvm_vm_release,
2720         .unlocked_ioctl = kvm_vm_ioctl,
2721 #ifdef CONFIG_KVM_COMPAT
2722         .compat_ioctl   = kvm_vm_compat_ioctl,
2723 #endif
2724         .llseek         = noop_llseek,
2725 };
2726
2727 static int kvm_dev_ioctl_create_vm(unsigned long type)
2728 {
2729         int r;
2730         struct kvm *kvm;
2731
2732         kvm = kvm_create_vm(type);
2733         if (IS_ERR(kvm))
2734                 return PTR_ERR(kvm);
2735 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2736         r = kvm_coalesced_mmio_init(kvm);
2737         if (r < 0) {
2738                 kvm_put_kvm(kvm);
2739                 return r;
2740         }
2741 #endif
2742         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2743         if (r < 0)
2744                 kvm_put_kvm(kvm);
2745
2746         return r;
2747 }
2748
2749 static long kvm_dev_ioctl(struct file *filp,
2750                           unsigned int ioctl, unsigned long arg)
2751 {
2752         long r = -EINVAL;
2753
2754         switch (ioctl) {
2755         case KVM_GET_API_VERSION:
2756                 if (arg)
2757                         goto out;
2758                 r = KVM_API_VERSION;
2759                 break;
2760         case KVM_CREATE_VM:
2761                 r = kvm_dev_ioctl_create_vm(arg);
2762                 break;
2763         case KVM_CHECK_EXTENSION:
2764                 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2765                 break;
2766         case KVM_GET_VCPU_MMAP_SIZE:
2767                 if (arg)
2768                         goto out;
2769                 r = PAGE_SIZE;     /* struct kvm_run */
2770 #ifdef CONFIG_X86
2771                 r += PAGE_SIZE;    /* pio data page */
2772 #endif
2773 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2774                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2775 #endif
2776                 break;
2777         case KVM_TRACE_ENABLE:
2778         case KVM_TRACE_PAUSE:
2779         case KVM_TRACE_DISABLE:
2780                 r = -EOPNOTSUPP;
2781                 break;
2782         default:
2783                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2784         }
2785 out:
2786         return r;
2787 }
2788
2789 static struct file_operations kvm_chardev_ops = {
2790         .unlocked_ioctl = kvm_dev_ioctl,
2791         .compat_ioctl   = kvm_dev_ioctl,
2792         .llseek         = noop_llseek,
2793 };
2794
2795 static struct miscdevice kvm_dev = {
2796         KVM_MINOR,
2797         "kvm",
2798         &kvm_chardev_ops,
2799 };
2800
2801 static void hardware_enable_nolock(void *junk)
2802 {
2803         int cpu = raw_smp_processor_id();
2804         int r;
2805
2806         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2807                 return;
2808
2809         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2810
2811         r = kvm_arch_hardware_enable();
2812
2813         if (r) {
2814                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2815                 atomic_inc(&hardware_enable_failed);
2816                 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
2817         }
2818 }
2819
2820 static void hardware_enable(void)
2821 {
2822         raw_spin_lock(&kvm_count_lock);
2823         if (kvm_usage_count)
2824                 hardware_enable_nolock(NULL);
2825         raw_spin_unlock(&kvm_count_lock);
2826 }
2827
2828 static void hardware_disable_nolock(void *junk)
2829 {
2830         int cpu = raw_smp_processor_id();
2831
2832         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2833                 return;
2834         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2835         kvm_arch_hardware_disable();
2836 }
2837
2838 static void hardware_disable(void)
2839 {
2840         raw_spin_lock(&kvm_count_lock);
2841         if (kvm_usage_count)
2842                 hardware_disable_nolock(NULL);
2843         raw_spin_unlock(&kvm_count_lock);
2844 }
2845
2846 static void hardware_disable_all_nolock(void)
2847 {
2848         BUG_ON(!kvm_usage_count);
2849
2850         kvm_usage_count--;
2851         if (!kvm_usage_count)
2852                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2853 }
2854
2855 static void hardware_disable_all(void)
2856 {
2857         raw_spin_lock(&kvm_count_lock);
2858         hardware_disable_all_nolock();
2859         raw_spin_unlock(&kvm_count_lock);
2860 }
2861
2862 static int hardware_enable_all(void)
2863 {
2864         int r = 0;
2865
2866         raw_spin_lock(&kvm_count_lock);
2867
2868         kvm_usage_count++;
2869         if (kvm_usage_count == 1) {
2870                 atomic_set(&hardware_enable_failed, 0);
2871                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2872
2873                 if (atomic_read(&hardware_enable_failed)) {
2874                         hardware_disable_all_nolock();
2875                         r = -EBUSY;
2876                 }
2877         }
2878
2879         raw_spin_unlock(&kvm_count_lock);
2880
2881         return r;
2882 }
2883
2884 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2885                            void *v)
2886 {
2887         val &= ~CPU_TASKS_FROZEN;
2888         switch (val) {
2889         case CPU_DYING:
2890                 hardware_disable();
2891                 break;
2892         case CPU_STARTING:
2893                 hardware_enable();
2894                 break;
2895         }
2896         return NOTIFY_OK;
2897 }
2898
2899 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2900                       void *v)
2901 {
2902         /*
2903          * Some (well, at least mine) BIOSes hang on reboot if
2904          * in vmx root mode.
2905          *
2906          * And Intel TXT required VMX off for all cpu when system shutdown.
2907          */
2908         pr_info("kvm: exiting hardware virtualization\n");
2909         kvm_rebooting = true;
2910         on_each_cpu(hardware_disable_nolock, NULL, 1);
2911         return NOTIFY_OK;
2912 }
2913
2914 static struct notifier_block kvm_reboot_notifier = {
2915         .notifier_call = kvm_reboot,
2916         .priority = 0,
2917 };
2918
2919 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2920 {
2921         int i;
2922
2923         for (i = 0; i < bus->dev_count; i++) {
2924                 struct kvm_io_device *pos = bus->range[i].dev;
2925
2926                 kvm_iodevice_destructor(pos);
2927         }
2928         kfree(bus);
2929 }
2930
2931 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2932                                  const struct kvm_io_range *r2)
2933 {
2934         if (r1->addr < r2->addr)
2935                 return -1;
2936         if (r1->addr + r1->len > r2->addr + r2->len)
2937                 return 1;
2938         return 0;
2939 }
2940
2941 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2942 {
2943         return kvm_io_bus_cmp(p1, p2);
2944 }
2945
2946 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2947                           gpa_t addr, int len)
2948 {
2949         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2950                 .addr = addr,
2951                 .len = len,
2952                 .dev = dev,
2953         };
2954
2955         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2956                 kvm_io_bus_sort_cmp, NULL);
2957
2958         return 0;
2959 }
2960
2961 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2962                              gpa_t addr, int len)
2963 {
2964         struct kvm_io_range *range, key;
2965         int off;
2966
2967         key = (struct kvm_io_range) {
2968                 .addr = addr,
2969                 .len = len,
2970         };
2971
2972         range = bsearch(&key, bus->range, bus->dev_count,
2973                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2974         if (range == NULL)
2975                 return -ENOENT;
2976
2977         off = range - bus->range;
2978
2979         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2980                 off--;
2981
2982         return off;
2983 }
2984
2985 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
2986                               struct kvm_io_range *range, const void *val)
2987 {
2988         int idx;
2989
2990         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2991         if (idx < 0)
2992                 return -EOPNOTSUPP;
2993
2994         while (idx < bus->dev_count &&
2995                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2996                 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
2997                                         range->len, val))
2998                         return idx;
2999                 idx++;
3000         }
3001
3002         return -EOPNOTSUPP;
3003 }
3004
3005 /* kvm_io_bus_write - called under kvm->slots_lock */
3006 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3007                      int len, const void *val)
3008 {
3009         struct kvm_io_bus *bus;
3010         struct kvm_io_range range;
3011         int r;
3012
3013         range = (struct kvm_io_range) {
3014                 .addr = addr,
3015                 .len = len,
3016         };
3017
3018         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3019         r = __kvm_io_bus_write(vcpu, bus, &range, val);
3020         return r < 0 ? r : 0;
3021 }
3022
3023 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3024 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3025                             gpa_t addr, int len, const void *val, long cookie)
3026 {
3027         struct kvm_io_bus *bus;
3028         struct kvm_io_range range;
3029
3030         range = (struct kvm_io_range) {
3031                 .addr = addr,
3032                 .len = len,
3033         };
3034
3035         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3036
3037         /* First try the device referenced by cookie. */
3038         if ((cookie >= 0) && (cookie < bus->dev_count) &&
3039             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3040                 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3041                                         val))
3042                         return cookie;
3043
3044         /*
3045          * cookie contained garbage; fall back to search and return the
3046          * correct cookie value.
3047          */
3048         return __kvm_io_bus_write(vcpu, bus, &range, val);
3049 }
3050
3051 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3052                              struct kvm_io_range *range, void *val)
3053 {
3054         int idx;
3055
3056         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3057         if (idx < 0)
3058                 return -EOPNOTSUPP;
3059
3060         while (idx < bus->dev_count &&
3061                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3062                 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3063                                        range->len, val))
3064                         return idx;
3065                 idx++;
3066         }
3067
3068         return -EOPNOTSUPP;
3069 }
3070 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3071
3072 /* kvm_io_bus_read - called under kvm->slots_lock */
3073 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3074                     int len, void *val)
3075 {
3076         struct kvm_io_bus *bus;
3077         struct kvm_io_range range;
3078         int r;
3079
3080         range = (struct kvm_io_range) {
3081                 .addr = addr,
3082                 .len = len,
3083         };
3084
3085         bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3086         r = __kvm_io_bus_read(vcpu, bus, &range, val);
3087         return r < 0 ? r : 0;
3088 }
3089
3090
3091 /* Caller must hold slots_lock. */
3092 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3093                             int len, struct kvm_io_device *dev)
3094 {
3095         struct kvm_io_bus *new_bus, *bus;
3096
3097         bus = kvm->buses[bus_idx];
3098         /* exclude ioeventfd which is limited by maximum fd */
3099         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3100                 return -ENOSPC;
3101
3102         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3103                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3104         if (!new_bus)
3105                 return -ENOMEM;
3106         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3107                sizeof(struct kvm_io_range)));
3108         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3109         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3110         synchronize_srcu_expedited(&kvm->srcu);
3111         kfree(bus);
3112
3113         return 0;
3114 }
3115
3116 /* Caller must hold slots_lock. */
3117 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3118                               struct kvm_io_device *dev)
3119 {
3120         int i, r;
3121         struct kvm_io_bus *new_bus, *bus;
3122
3123         bus = kvm->buses[bus_idx];
3124         r = -ENOENT;
3125         for (i = 0; i < bus->dev_count; i++)
3126                 if (bus->range[i].dev == dev) {
3127                         r = 0;
3128                         break;
3129                 }
3130
3131         if (r)
3132                 return r;
3133
3134         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3135                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3136         if (!new_bus)
3137                 return -ENOMEM;
3138
3139         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3140         new_bus->dev_count--;
3141         memcpy(new_bus->range + i, bus->range + i + 1,
3142                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3143
3144         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3145         synchronize_srcu_expedited(&kvm->srcu);
3146         kfree(bus);
3147         return r;
3148 }
3149
3150 static struct notifier_block kvm_cpu_notifier = {
3151         .notifier_call = kvm_cpu_hotplug,
3152 };
3153
3154 static int vm_stat_get(void *_offset, u64 *val)
3155 {
3156         unsigned offset = (long)_offset;
3157         struct kvm *kvm;
3158
3159         *val = 0;
3160         spin_lock(&kvm_lock);
3161         list_for_each_entry(kvm, &vm_list, vm_list)
3162                 *val += *(u32 *)((void *)kvm + offset);
3163         spin_unlock(&kvm_lock);
3164         return 0;
3165 }
3166
3167 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3168
3169 static int vcpu_stat_get(void *_offset, u64 *val)
3170 {
3171         unsigned offset = (long)_offset;
3172         struct kvm *kvm;
3173         struct kvm_vcpu *vcpu;
3174         int i;
3175
3176         *val = 0;
3177         spin_lock(&kvm_lock);
3178         list_for_each_entry(kvm, &vm_list, vm_list)
3179                 kvm_for_each_vcpu(i, vcpu, kvm)
3180                         *val += *(u32 *)((void *)vcpu + offset);
3181
3182         spin_unlock(&kvm_lock);
3183         return 0;
3184 }
3185
3186 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3187
3188 static const struct file_operations *stat_fops[] = {
3189         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3190         [KVM_STAT_VM]   = &vm_stat_fops,
3191 };
3192
3193 static int kvm_init_debug(void)
3194 {
3195         int r = -EEXIST;
3196         struct kvm_stats_debugfs_item *p;
3197
3198         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3199         if (kvm_debugfs_dir == NULL)
3200                 goto out;
3201
3202         for (p = debugfs_entries; p->name; ++p) {
3203                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3204                                                 (void *)(long)p->offset,
3205                                                 stat_fops[p->kind]);
3206                 if (p->dentry == NULL)
3207                         goto out_dir;
3208         }
3209
3210         return 0;
3211
3212 out_dir:
3213         debugfs_remove_recursive(kvm_debugfs_dir);
3214 out:
3215         return r;
3216 }
3217
3218 static void kvm_exit_debug(void)
3219 {
3220         struct kvm_stats_debugfs_item *p;
3221
3222         for (p = debugfs_entries; p->name; ++p)
3223                 debugfs_remove(p->dentry);
3224         debugfs_remove(kvm_debugfs_dir);
3225 }
3226
3227 static int kvm_suspend(void)
3228 {
3229         if (kvm_usage_count)
3230                 hardware_disable_nolock(NULL);
3231         return 0;
3232 }
3233
3234 static void kvm_resume(void)
3235 {
3236         if (kvm_usage_count) {
3237                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3238                 hardware_enable_nolock(NULL);
3239         }
3240 }
3241
3242 static struct syscore_ops kvm_syscore_ops = {
3243         .suspend = kvm_suspend,
3244         .resume = kvm_resume,
3245 };
3246
3247 static inline
3248 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3249 {
3250         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3251 }
3252
3253 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3254 {
3255         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3256
3257         if (vcpu->preempted)
3258                 vcpu->preempted = false;
3259
3260         kvm_arch_sched_in(vcpu, cpu);
3261
3262         kvm_arch_vcpu_load(vcpu, cpu);
3263 }
3264
3265 static void kvm_sched_out(struct preempt_notifier *pn,
3266                           struct task_struct *next)
3267 {
3268         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3269
3270         if (current->state == TASK_RUNNING)
3271                 vcpu->preempted = true;
3272         kvm_arch_vcpu_put(vcpu);
3273 }
3274
3275 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3276                   struct module *module)
3277 {
3278         int r;
3279         int cpu;
3280
3281         r = kvm_arch_init(opaque);
3282         if (r)
3283                 goto out_fail;
3284
3285         /*
3286          * kvm_arch_init makes sure there's at most one caller
3287          * for architectures that support multiple implementations,
3288          * like intel and amd on x86.
3289          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3290          * conflicts in case kvm is already setup for another implementation.
3291          */
3292         r = kvm_irqfd_init();
3293         if (r)
3294                 goto out_irqfd;
3295
3296         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3297                 r = -ENOMEM;
3298                 goto out_free_0;
3299         }
3300
3301         r = kvm_arch_hardware_setup();
3302         if (r < 0)
3303                 goto out_free_0a;
3304
3305         for_each_online_cpu(cpu) {
3306                 smp_call_function_single(cpu,
3307                                 kvm_arch_check_processor_compat,
3308                                 &r, 1);
3309                 if (r < 0)
3310                         goto out_free_1;
3311         }
3312
3313         r = register_cpu_notifier(&kvm_cpu_notifier);
3314         if (r)
3315                 goto out_free_2;
3316         register_reboot_notifier(&kvm_reboot_notifier);
3317
3318         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3319         if (!vcpu_align)
3320                 vcpu_align = __alignof__(struct kvm_vcpu);
3321         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3322                                            0, NULL);
3323         if (!kvm_vcpu_cache) {
3324                 r = -ENOMEM;
3325                 goto out_free_3;
3326         }
3327
3328         r = kvm_async_pf_init();
3329         if (r)
3330                 goto out_free;
3331
3332         kvm_chardev_ops.owner = module;
3333         kvm_vm_fops.owner = module;
3334         kvm_vcpu_fops.owner = module;
3335
3336         r = misc_register(&kvm_dev);
3337         if (r) {
3338                 pr_err("kvm: misc device register failed\n");
3339                 goto out_unreg;
3340         }
3341
3342         register_syscore_ops(&kvm_syscore_ops);
3343
3344         kvm_preempt_ops.sched_in = kvm_sched_in;
3345         kvm_preempt_ops.sched_out = kvm_sched_out;
3346
3347         r = kvm_init_debug();
3348         if (r) {
3349                 pr_err("kvm: create debugfs files failed\n");
3350                 goto out_undebugfs;
3351         }
3352
3353         r = kvm_vfio_ops_init();
3354         WARN_ON(r);
3355
3356         return 0;
3357
3358 out_undebugfs:
3359         unregister_syscore_ops(&kvm_syscore_ops);
3360         misc_deregister(&kvm_dev);
3361 out_unreg:
3362         kvm_async_pf_deinit();
3363 out_free:
3364         kmem_cache_destroy(kvm_vcpu_cache);
3365 out_free_3:
3366         unregister_reboot_notifier(&kvm_reboot_notifier);
3367         unregister_cpu_notifier(&kvm_cpu_notifier);
3368 out_free_2:
3369 out_free_1:
3370         kvm_arch_hardware_unsetup();
3371 out_free_0a:
3372         free_cpumask_var(cpus_hardware_enabled);
3373 out_free_0:
3374         kvm_irqfd_exit();
3375 out_irqfd:
3376         kvm_arch_exit();
3377 out_fail:
3378         return r;
3379 }
3380 EXPORT_SYMBOL_GPL(kvm_init);
3381
3382 void kvm_exit(void)
3383 {
3384         kvm_exit_debug();
3385         misc_deregister(&kvm_dev);
3386         kmem_cache_destroy(kvm_vcpu_cache);
3387         kvm_async_pf_deinit();
3388         unregister_syscore_ops(&kvm_syscore_ops);
3389         unregister_reboot_notifier(&kvm_reboot_notifier);
3390         unregister_cpu_notifier(&kvm_cpu_notifier);
3391         on_each_cpu(hardware_disable_nolock, NULL, 1);
3392         kvm_arch_hardware_unsetup();
3393         kvm_arch_exit();
3394         kvm_irqfd_exit();
3395         free_cpumask_var(cpus_hardware_enabled);
3396         kvm_vfio_ops_exit();
3397 }
3398 EXPORT_SYMBOL_GPL(kvm_exit);