virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/syscore_ops.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/pgtable.h>
  57
  58 #include "coalesced_mmio.h"
  59 #include "async_pf.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/kvm.h>
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 /*
  68  * Ordering of locks:
  69  *
  70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  71  */
  72
  73 DEFINE_RAW_SPINLOCK(kvm_lock);
  74 LIST_HEAD(vm_list);
  75
  76 static cpumask_var_t cpus_hardware_enabled;
  77 static int kvm_usage_count = 0;
  78 static atomic_t hardware_enable_failed;
  79
  80 struct kmem_cache *kvm_vcpu_cache;
  81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  82
  83 static __read_mostly struct preempt_ops kvm_preempt_ops;
  84
  85 struct dentry *kvm_debugfs_dir;
  86
  87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  88                            unsigned long arg);
  89 #ifdef CONFIG_COMPAT
  90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  91                                   unsigned long arg);
  92 #endif
  93 static int hardware_enable_all(void);
  94 static void hardware_disable_all(void);
  95
  96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  97
  98 bool kvm_rebooting;
  99 EXPORT_SYMBOL_GPL(kvm_rebooting);
 100
 101 static bool largepages_enabled = true;
 102
 103 static struct page *hwpoison_page;
 104 static pfn_t hwpoison_pfn;
 105
 106 struct page *fault_page;
 107 pfn_t fault_pfn;
 108
 109 inline int kvm_is_mmio_pfn(pfn_t pfn)
 110 {
 111         if (pfn_valid(pfn)) {
 112                 int reserved;
 113                 struct page *tail = pfn_to_page(pfn);
 114                 struct page *head = compound_trans_head(tail);
 115                 reserved = PageReserved(head);
 116                 if (head != tail) {
 117                         /*
 118                          * "head" is not a dangling pointer
 119                          * (compound_trans_head takes care of that)
 120                          * but the hugepage may have been splitted
 121                          * from under us (and we may not hold a
 122                          * reference count on the head page so it can
 123                          * be reused before we run PageReferenced), so
 124                          * we've to check PageTail before returning
 125                          * what we just read.
 126                          */
 127                         smp_rmb();
 128                         if (PageTail(tail))
 129                                 return reserved;
 130                 }
 131                 return PageReserved(tail);
 132         }
 133
 134         return true;
 135 }
 136
 137 /*
 138  * Switches to specified vcpu, until a matching vcpu_put()
 139  */
 140 void vcpu_load(struct kvm_vcpu *vcpu)
 141 {
 142         int cpu;
 143
 144         mutex_lock(&vcpu->mutex);
 145         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 146                 /* The thread running this VCPU changed. */
 147                 struct pid *oldpid = vcpu->pid;
 148                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 149                 rcu_assign_pointer(vcpu->pid, newpid);
 150                 synchronize_rcu();
 151                 put_pid(oldpid);
 152         }
 153         cpu = get_cpu();
 154         preempt_notifier_register(&vcpu->preempt_notifier);
 155         kvm_arch_vcpu_load(vcpu, cpu);
 156         put_cpu();
 157 }
 158
 159 void vcpu_put(struct kvm_vcpu *vcpu)
 160 {
 161         preempt_disable();
 162         kvm_arch_vcpu_put(vcpu);
 163         preempt_notifier_unregister(&vcpu->preempt_notifier);
 164         preempt_enable();
 165         mutex_unlock(&vcpu->mutex);
 166 }
 167
 168 static void ack_flush(void *_completed)
 169 {
 170 }
 171
 172 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 173 {
 174         int i, cpu, me;
 175         cpumask_var_t cpus;
 176         bool called = true;
 177         struct kvm_vcpu *vcpu;
 178
 179         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 180
 181         me = get_cpu();
 182         kvm_for_each_vcpu(i, vcpu, kvm) {
 183                 kvm_make_request(req, vcpu);
 184                 cpu = vcpu->cpu;
 185
 186                 /* Set ->requests bit before we read ->mode */
 187                 smp_mb();
 188
 189                 if (cpus != NULL && cpu != -1 && cpu != me &&
 190                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 191                         cpumask_set_cpu(cpu, cpus);
 192         }
 193         if (unlikely(cpus == NULL))
 194                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 195         else if (!cpumask_empty(cpus))
 196                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 197         else
 198                 called = false;
 199         put_cpu();
 200         free_cpumask_var(cpus);
 201         return called;
 202 }
 203
 204 void kvm_flush_remote_tlbs(struct kvm *kvm)
 205 {
 206         int dirty_count = kvm->tlbs_dirty;
 207
 208         smp_mb();
 209         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 210                 ++kvm->stat.remote_tlb_flush;
 211         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 212 }
 213
 214 void kvm_reload_remote_mmus(struct kvm *kvm)
 215 {
 216         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 217 }
 218
 219 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 220 {
 221         struct page *page;
 222         int r;
 223
 224         mutex_init(&vcpu->mutex);
 225         vcpu->cpu = -1;
 226         vcpu->kvm = kvm;
 227         vcpu->vcpu_id = id;
 228         vcpu->pid = NULL;
 229         init_waitqueue_head(&vcpu->wq);
 230         kvm_async_pf_vcpu_init(vcpu);
 231
 232         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 233         if (!page) {
 234                 r = -ENOMEM;
 235                 goto fail;
 236         }
 237         vcpu->run = page_address(page);
 238
 239         r = kvm_arch_vcpu_init(vcpu);
 240         if (r < 0)
 241                 goto fail_free_run;
 242         return 0;
 243
 244 fail_free_run:
 245         free_page((unsigned long)vcpu->run);
 246 fail:
 247         return r;
 248 }
 249 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 250
 251 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 252 {
 253         put_pid(vcpu->pid);
 254         kvm_arch_vcpu_uninit(vcpu);
 255         free_page((unsigned long)vcpu->run);
 256 }
 257 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 258
 259 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 260 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 261 {
 262         return container_of(mn, struct kvm, mmu_notifier);
 263 }
 264
 265 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 266                                              struct mm_struct *mm,
 267                                              unsigned long address)
 268 {
 269         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 270         int need_tlb_flush, idx;
 271
 272         /*
 273          * When ->invalidate_page runs, the linux pte has been zapped
 274          * already but the page is still allocated until
 275          * ->invalidate_page returns. So if we increase the sequence
 276          * here the kvm page fault will notice if the spte can't be
 277          * established because the page is going to be freed. If
 278          * instead the kvm page fault establishes the spte before
 279          * ->invalidate_page runs, kvm_unmap_hva will release it
 280          * before returning.
 281          *
 282          * The sequence increase only need to be seen at spin_unlock
 283          * time, and not at spin_lock time.
 284          *
 285          * Increasing the sequence after the spin_unlock would be
 286          * unsafe because the kvm page fault could then establish the
 287          * pte after kvm_unmap_hva returned, without noticing the page
 288          * is going to be freed.
 289          */
 290         idx = srcu_read_lock(&kvm->srcu);
 291         spin_lock(&kvm->mmu_lock);
 292         kvm->mmu_notifier_seq++;
 293         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 294         spin_unlock(&kvm->mmu_lock);
 295         srcu_read_unlock(&kvm->srcu, idx);
 296
 297         /* we've to flush the tlb before the pages can be freed */
 298         if (need_tlb_flush)
 299                 kvm_flush_remote_tlbs(kvm);
 300
 301 }
 302
 303 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 304                                         struct mm_struct *mm,
 305                                         unsigned long address,
 306                                         pte_t pte)
 307 {
 308         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 309         int idx;
 310
 311         idx = srcu_read_lock(&kvm->srcu);
 312         spin_lock(&kvm->mmu_lock);
 313         kvm->mmu_notifier_seq++;
 314         kvm_set_spte_hva(kvm, address, pte);
 315         spin_unlock(&kvm->mmu_lock);
 316         srcu_read_unlock(&kvm->srcu, idx);
 317 }
 318
 319 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 320                                                     struct mm_struct *mm,
 321                                                     unsigned long start,
 322                                                     unsigned long end)
 323 {
 324         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 325         int need_tlb_flush = 0, idx;
 326
 327         idx = srcu_read_lock(&kvm->srcu);
 328         spin_lock(&kvm->mmu_lock);
 329         /*
 330          * The count increase must become visible at unlock time as no
 331          * spte can be established without taking the mmu_lock and
 332          * count is also read inside the mmu_lock critical section.
 333          */
 334         kvm->mmu_notifier_count++;
 335         for (; start < end; start += PAGE_SIZE)
 336                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 337         need_tlb_flush |= kvm->tlbs_dirty;
 338         spin_unlock(&kvm->mmu_lock);
 339         srcu_read_unlock(&kvm->srcu, idx);
 340
 341         /* we've to flush the tlb before the pages can be freed */
 342         if (need_tlb_flush)
 343                 kvm_flush_remote_tlbs(kvm);
 344 }
 345
 346 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 347                                                   struct mm_struct *mm,
 348                                                   unsigned long start,
 349                                                   unsigned long end)
 350 {
 351         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 352
 353         spin_lock(&kvm->mmu_lock);
 354         /*
 355          * This sequence increase will notify the kvm page fault that
 356          * the page that is going to be mapped in the spte could have
 357          * been freed.
 358          */
 359         kvm->mmu_notifier_seq++;
 360         /*
 361          * The above sequence increase must be visible before the
 362          * below count decrease but both values are read by the kvm
 363          * page fault under mmu_lock spinlock so we don't need to add
 364          * a smb_wmb() here in between the two.
 365          */
 366         kvm->mmu_notifier_count--;
 367         spin_unlock(&kvm->mmu_lock);
 368
 369         BUG_ON(kvm->mmu_notifier_count < 0);
 370 }
 371
 372 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 373                                               struct mm_struct *mm,
 374                                               unsigned long address)
 375 {
 376         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 377         int young, idx;
 378
 379         idx = srcu_read_lock(&kvm->srcu);
 380         spin_lock(&kvm->mmu_lock);
 381         young = kvm_age_hva(kvm, address);
 382         spin_unlock(&kvm->mmu_lock);
 383         srcu_read_unlock(&kvm->srcu, idx);
 384
 385         if (young)
 386                 kvm_flush_remote_tlbs(kvm);
 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(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 *kvm_create_vm(void)
 444 {
 445         int r, i;
 446         struct kvm *kvm = kvm_arch_alloc_vm();
 447
 448         if (!kvm)
 449                 return ERR_PTR(-ENOMEM);
 450
 451         r = kvm_arch_init_vm(kvm);
 452         if (r)
 453                 goto out_err_nodisable;
 454
 455         r = hardware_enable_all();
 456         if (r)
 457                 goto out_err_nodisable;
 458
 459 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 460         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 461         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 462 #endif
 463
 464         r = -ENOMEM;
 465         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 466         if (!kvm->memslots)
 467                 goto out_err_nosrcu;
 468         if (init_srcu_struct(&kvm->srcu))
 469                 goto out_err_nosrcu;
 470         for (i = 0; i < KVM_NR_BUSES; i++) {
 471                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 472                                         GFP_KERNEL);
 473                 if (!kvm->buses[i])
 474                         goto out_err;
 475         }
 476
 477         spin_lock_init(&kvm->mmu_lock);
 478         kvm->mm = current->mm;
 479         atomic_inc(&kvm->mm->mm_count);
 480         kvm_eventfd_init(kvm);
 481         mutex_init(&kvm->lock);
 482         mutex_init(&kvm->irq_lock);
 483         mutex_init(&kvm->slots_lock);
 484         atomic_set(&kvm->users_count, 1);
 485
 486         r = kvm_init_mmu_notifier(kvm);
 487         if (r)
 488                 goto out_err;
 489
 490         raw_spin_lock(&kvm_lock);
 491         list_add(&kvm->vm_list, &vm_list);
 492         raw_spin_unlock(&kvm_lock);
 493
 494         return kvm;
 495
 496 out_err:
 497         cleanup_srcu_struct(&kvm->srcu);
 498 out_err_nosrcu:
 499         hardware_disable_all();
 500 out_err_nodisable:
 501         for (i = 0; i < KVM_NR_BUSES; i++)
 502                 kfree(kvm->buses[i]);
 503         kfree(kvm->memslots);
 504         kvm_arch_free_vm(kvm);
 505         return ERR_PTR(r);
 506 }
 507
 508 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 509 {
 510         if (!memslot->dirty_bitmap)
 511                 return;
 512
 513         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
 514                 vfree(memslot->dirty_bitmap_head);
 515         else
 516                 kfree(memslot->dirty_bitmap_head);
 517
 518         memslot->dirty_bitmap = NULL;
 519         memslot->dirty_bitmap_head = NULL;
 520 }
 521
 522 /*
 523  * Free any memory in @free but not in @dont.
 524  */
 525 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 526                                   struct kvm_memory_slot *dont)
 527 {
 528         int i;
 529
 530         if (!dont || free->rmap != dont->rmap)
 531                 vfree(free->rmap);
 532
 533         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 534                 kvm_destroy_dirty_bitmap(free);
 535
 536
 537         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 538                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 539                         vfree(free->lpage_info[i]);
 540                         free->lpage_info[i] = NULL;
 541                 }
 542         }
 543
 544         free->npages = 0;
 545         free->rmap = NULL;
 546 }
 547
 548 void kvm_free_physmem(struct kvm *kvm)
 549 {
 550         int i;
 551         struct kvm_memslots *slots = kvm->memslots;
 552
 553         for (i = 0; i < slots->nmemslots; ++i)
 554                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 555
 556         kfree(kvm->memslots);
 557 }
 558
 559 static void kvm_destroy_vm(struct kvm *kvm)
 560 {
 561         int i;
 562         struct mm_struct *mm = kvm->mm;
 563
 564         kvm_arch_sync_events(kvm);
 565         raw_spin_lock(&kvm_lock);
 566         list_del(&kvm->vm_list);
 567         raw_spin_unlock(&kvm_lock);
 568         kvm_free_irq_routing(kvm);
 569         for (i = 0; i < KVM_NR_BUSES; i++)
 570                 kvm_io_bus_destroy(kvm->buses[i]);
 571         kvm_coalesced_mmio_free(kvm);
 572 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 573         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 574 #else
 575         kvm_arch_flush_shadow(kvm);
 576 #endif
 577         kvm_arch_destroy_vm(kvm);
 578         kvm_free_physmem(kvm);
 579         cleanup_srcu_struct(&kvm->srcu);
 580         kvm_arch_free_vm(kvm);
 581         hardware_disable_all();
 582         mmdrop(mm);
 583 }
 584
 585 void kvm_get_kvm(struct kvm *kvm)
 586 {
 587         atomic_inc(&kvm->users_count);
 588 }
 589 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 590
 591 void kvm_put_kvm(struct kvm *kvm)
 592 {
 593         if (atomic_dec_and_test(&kvm->users_count))
 594                 kvm_destroy_vm(kvm);
 595 }
 596 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 597
 598
 599 static int kvm_vm_release(struct inode *inode, struct file *filp)
 600 {
 601         struct kvm *kvm = filp->private_data;
 602
 603         kvm_irqfd_release(kvm);
 604
 605         kvm_put_kvm(kvm);
 606         return 0;
 607 }
 608
 609 #ifndef CONFIG_S390
 610 /*
 611  * Allocation size is twice as large as the actual dirty bitmap size.
 612  * This makes it possible to do double buffering: see x86's
 613  * kvm_vm_ioctl_get_dirty_log().
 614  */
 615 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 616 {
 617         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 618
 619         if (dirty_bytes > PAGE_SIZE)
 620                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
 621         else
 622                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
 623
 624         if (!memslot->dirty_bitmap)
 625                 return -ENOMEM;
 626
 627         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
 628         memslot->nr_dirty_pages = 0;
 629         return 0;
 630 }
 631 #endif /* !CONFIG_S390 */
 632
 633 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
 634 {
 635         if (new) {
 636                 int id = new->id;
 637
 638                 slots->memslots[id] = *new;
 639                 if (id >= slots->nmemslots)
 640                         slots->nmemslots = id + 1;
 641         }
 642
 643         slots->generation++;
 644 }
 645
 646 /*
 647  * Allocate some memory and give it an address in the guest physical address
 648  * space.
 649  *
 650  * Discontiguous memory is allowed, mostly for framebuffers.
 651  *
 652  * Must be called holding mmap_sem for write.
 653  */
 654 int __kvm_set_memory_region(struct kvm *kvm,
 655                             struct kvm_userspace_memory_region *mem,
 656                             int user_alloc)
 657 {
 658         int r;
 659         gfn_t base_gfn;
 660         unsigned long npages;
 661         unsigned long i;
 662         struct kvm_memory_slot *memslot;
 663         struct kvm_memory_slot old, new;
 664         struct kvm_memslots *slots, *old_memslots;
 665
 666         r = -EINVAL;
 667         /* General sanity checks */
 668         if (mem->memory_size & (PAGE_SIZE - 1))
 669                 goto out;
 670         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 671                 goto out;
 672         /* We can read the guest memory with __xxx_user() later on. */
 673         if (user_alloc &&
 674             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 675              !access_ok(VERIFY_WRITE,
 676                         (void __user *)(unsigned long)mem->userspace_addr,
 677                         mem->memory_size)))
 678                 goto out;
 679         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 680                 goto out;
 681         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 682                 goto out;
 683
 684         memslot = &kvm->memslots->memslots[mem->slot];
 685         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 686         npages = mem->memory_size >> PAGE_SHIFT;
 687
 688         r = -EINVAL;
 689         if (npages > KVM_MEM_MAX_NR_PAGES)
 690                 goto out;
 691
 692         if (!npages)
 693                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 694
 695         new = old = *memslot;
 696
 697         new.id = mem->slot;
 698         new.base_gfn = base_gfn;
 699         new.npages = npages;
 700         new.flags = mem->flags;
 701
 702         /* Disallow changing a memory slot's size. */
 703         r = -EINVAL;
 704         if (npages && old.npages && npages != old.npages)
 705                 goto out_free;
 706
 707         /* Check for overlaps */
 708         r = -EEXIST;
 709         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 710                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 711
 712                 if (s == memslot || !s->npages)
 713                         continue;
 714                 if (!((base_gfn + npages <= s->base_gfn) ||
 715                       (base_gfn >= s->base_gfn + s->npages)))
 716                         goto out_free;
 717         }
 718
 719         /* Free page dirty bitmap if unneeded */
 720         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 721                 new.dirty_bitmap = NULL;
 722
 723         r = -ENOMEM;
 724
 725         /* Allocate if a slot is being created */
 726 #ifndef CONFIG_S390
 727         if (npages && !new.rmap) {
 728                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
 729
 730                 if (!new.rmap)
 731                         goto out_free;
 732
 733                 new.user_alloc = user_alloc;
 734                 new.userspace_addr = mem->userspace_addr;
 735         }
 736         if (!npages)
 737                 goto skip_lpage;
 738
 739         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 740                 unsigned long ugfn;
 741                 unsigned long j;
 742                 int lpages;
 743                 int level = i + 2;
 744
 745                 /* Avoid unused variable warning if no large pages */
 746                 (void)level;
 747
 748                 if (new.lpage_info[i])
 749                         continue;
 750
 751                 lpages = 1 + ((base_gfn + npages - 1)
 752                              >> KVM_HPAGE_GFN_SHIFT(level));
 753                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
 754
 755                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
 756
 757                 if (!new.lpage_info[i])
 758                         goto out_free;
 759
 760                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
 761                         new.lpage_info[i][0].write_count = 1;
 762                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
 763                         new.lpage_info[i][lpages - 1].write_count = 1;
 764                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 765                 /*
 766                  * If the gfn and userspace address are not aligned wrt each
 767                  * other, or if explicitly asked to, disable large page
 768                  * support for this slot
 769                  */
 770                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 771                     !largepages_enabled)
 772                         for (j = 0; j < lpages; ++j)
 773                                 new.lpage_info[i][j].write_count = 1;
 774         }
 775
 776 skip_lpage:
 777
 778         /* Allocate page dirty bitmap if needed */
 779         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 780                 if (kvm_create_dirty_bitmap(&new) < 0)
 781                         goto out_free;
 782                 /* destroy any largepage mappings for dirty tracking */
 783         }
 784 #else  /* not defined CONFIG_S390 */
 785         new.user_alloc = user_alloc;
 786         if (user_alloc)
 787                 new.userspace_addr = mem->userspace_addr;
 788 #endif /* not defined CONFIG_S390 */
 789
 790         if (!npages) {
 791                 r = -ENOMEM;
 792                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 793                                 GFP_KERNEL);
 794                 if (!slots)
 795                         goto out_free;
 796                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 797                 update_memslots(slots, NULL);
 798
 799                 old_memslots = kvm->memslots;
 800                 rcu_assign_pointer(kvm->memslots, slots);
 801                 synchronize_srcu_expedited(&kvm->srcu);
 802                 /* From this point no new shadow pages pointing to a deleted
 803                  * memslot will be created.
 804                  *
 805                  * validation of sp->gfn happens in:
 806                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 807                  *      - kvm_is_visible_gfn (mmu_check_roots)
 808                  */
 809                 kvm_arch_flush_shadow(kvm);
 810                 kfree(old_memslots);
 811         }
 812
 813         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 814         if (r)
 815                 goto out_free;
 816
 817         /* map the pages in iommu page table */
 818         if (npages) {
 819                 r = kvm_iommu_map_pages(kvm, &new);
 820                 if (r)
 821                         goto out_free;
 822         }
 823
 824         r = -ENOMEM;
 825         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 826                         GFP_KERNEL);
 827         if (!slots)
 828                 goto out_free;
 829
 830         /* actual memory is freed via old in kvm_free_physmem_slot below */
 831         if (!npages) {
 832                 new.rmap = NULL;
 833                 new.dirty_bitmap = NULL;
 834                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 835                         new.lpage_info[i] = NULL;
 836         }
 837
 838         update_memslots(slots, &new);
 839         old_memslots = kvm->memslots;
 840         rcu_assign_pointer(kvm->memslots, slots);
 841         synchronize_srcu_expedited(&kvm->srcu);
 842
 843         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 844
 845         /*
 846          * If the new memory slot is created, we need to clear all
 847          * mmio sptes.
 848          */
 849         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 850                 kvm_arch_flush_shadow(kvm);
 851
 852         kvm_free_physmem_slot(&old, &new);
 853         kfree(old_memslots);
 854
 855         return 0;
 856
 857 out_free:
 858         kvm_free_physmem_slot(&new, &old);
 859 out:
 860         return r;
 861
 862 }
 863 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 864
 865 int kvm_set_memory_region(struct kvm *kvm,
 866                           struct kvm_userspace_memory_region *mem,
 867                           int user_alloc)
 868 {
 869         int r;
 870
 871         mutex_lock(&kvm->slots_lock);
 872         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 873         mutex_unlock(&kvm->slots_lock);
 874         return r;
 875 }
 876 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 877
 878 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 879                                    struct
 880                                    kvm_userspace_memory_region *mem,
 881                                    int user_alloc)
 882 {
 883         if (mem->slot >= KVM_MEMORY_SLOTS)
 884                 return -EINVAL;
 885         return kvm_set_memory_region(kvm, mem, user_alloc);
 886 }
 887
 888 int kvm_get_dirty_log(struct kvm *kvm,
 889                         struct kvm_dirty_log *log, int *is_dirty)
 890 {
 891         struct kvm_memory_slot *memslot;
 892         int r, i;
 893         unsigned long n;
 894         unsigned long any = 0;
 895
 896         r = -EINVAL;
 897         if (log->slot >= KVM_MEMORY_SLOTS)
 898                 goto out;
 899
 900         memslot = &kvm->memslots->memslots[log->slot];
 901         r = -ENOENT;
 902         if (!memslot->dirty_bitmap)
 903                 goto out;
 904
 905         n = kvm_dirty_bitmap_bytes(memslot);
 906
 907         for (i = 0; !any && i < n/sizeof(long); ++i)
 908                 any = memslot->dirty_bitmap[i];
 909
 910         r = -EFAULT;
 911         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 912                 goto out;
 913
 914         if (any)
 915                 *is_dirty = 1;
 916
 917         r = 0;
 918 out:
 919         return r;
 920 }
 921
 922 void kvm_disable_largepages(void)
 923 {
 924         largepages_enabled = false;
 925 }
 926 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 927
 928 int is_error_page(struct page *page)
 929 {
 930         return page == bad_page || page == hwpoison_page || page == fault_page;
 931 }
 932 EXPORT_SYMBOL_GPL(is_error_page);
 933
 934 int is_error_pfn(pfn_t pfn)
 935 {
 936         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 937 }
 938 EXPORT_SYMBOL_GPL(is_error_pfn);
 939
 940 int is_hwpoison_pfn(pfn_t pfn)
 941 {
 942         return pfn == hwpoison_pfn;
 943 }
 944 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 945
 946 int is_fault_pfn(pfn_t pfn)
 947 {
 948         return pfn == fault_pfn;
 949 }
 950 EXPORT_SYMBOL_GPL(is_fault_pfn);
 951
 952 int is_noslot_pfn(pfn_t pfn)
 953 {
 954         return pfn == bad_pfn;
 955 }
 956 EXPORT_SYMBOL_GPL(is_noslot_pfn);
 957
 958 int is_invalid_pfn(pfn_t pfn)
 959 {
 960         return pfn == hwpoison_pfn || pfn == fault_pfn;
 961 }
 962 EXPORT_SYMBOL_GPL(is_invalid_pfn);
 963
 964 static inline unsigned long bad_hva(void)
 965 {
 966         return PAGE_OFFSET;
 967 }
 968
 969 int kvm_is_error_hva(unsigned long addr)
 970 {
 971         return addr == bad_hva();
 972 }
 973 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 974
 975 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
 976                                                 gfn_t gfn)
 977 {
 978         int i;
 979
 980         for (i = 0; i < slots->nmemslots; ++i) {
 981                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 982
 983                 if (gfn >= memslot->base_gfn
 984                     && gfn < memslot->base_gfn + memslot->npages)
 985                         return memslot;
 986         }
 987         return NULL;
 988 }
 989
 990 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 991 {
 992         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 993 }
 994 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 995
 996 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 997 {
 998         int i;
 999         struct kvm_memslots *slots = kvm_memslots(kvm);
1000
1001         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1002                 struct kvm_memory_slot *memslot = &slots->memslots[i];
1003
1004                 if (memslot->flags & KVM_MEMSLOT_INVALID)
1005                         continue;
1006
1007                 if (gfn >= memslot->base_gfn
1008                     && gfn < memslot->base_gfn + memslot->npages)
1009                         return 1;
1010         }
1011         return 0;
1012 }
1013 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1014
1015 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1016 {
1017         struct vm_area_struct *vma;
1018         unsigned long addr, size;
1019
1020         size = PAGE_SIZE;
1021
1022         addr = gfn_to_hva(kvm, gfn);
1023         if (kvm_is_error_hva(addr))
1024                 return PAGE_SIZE;
1025
1026         down_read(&current->mm->mmap_sem);
1027         vma = find_vma(current->mm, addr);
1028         if (!vma)
1029                 goto out;
1030
1031         size = vma_kernel_pagesize(vma);
1032
1033 out:
1034         up_read(&current->mm->mmap_sem);
1035
1036         return size;
1037 }
1038
1039 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1040                                      gfn_t *nr_pages)
1041 {
1042         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1043                 return bad_hva();
1044
1045         if (nr_pages)
1046                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1047
1048         return gfn_to_hva_memslot(slot, gfn);
1049 }
1050
1051 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1052 {
1053         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1054 }
1055 EXPORT_SYMBOL_GPL(gfn_to_hva);
1056
1057 static pfn_t get_fault_pfn(void)
1058 {
1059         get_page(fault_page);
1060         return fault_pfn;
1061 }
1062
1063 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1064         unsigned long start, int write, struct page **page)
1065 {
1066         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1067
1068         if (write)
1069                 flags |= FOLL_WRITE;
1070
1071         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1072 }
1073
1074 static inline int check_user_page_hwpoison(unsigned long addr)
1075 {
1076         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1077
1078         rc = __get_user_pages(current, current->mm, addr, 1,
1079                               flags, NULL, NULL, NULL);
1080         return rc == -EHWPOISON;
1081 }
1082
1083 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1084                         bool *async, bool write_fault, bool *writable)
1085 {
1086         struct page *page[1];
1087         int npages = 0;
1088         pfn_t pfn;
1089
1090         /* we can do it either atomically or asynchronously, not both */
1091         BUG_ON(atomic && async);
1092
1093         BUG_ON(!write_fault && !writable);
1094
1095         if (writable)
1096                 *writable = true;
1097
1098         if (atomic || async)
1099                 npages = __get_user_pages_fast(addr, 1, 1, page);
1100
1101         if (unlikely(npages != 1) && !atomic) {
1102                 might_sleep();
1103
1104                 if (writable)
1105                         *writable = write_fault;
1106
1107                 if (async) {
1108                         down_read(&current->mm->mmap_sem);
1109                         npages = get_user_page_nowait(current, current->mm,
1110                                                      addr, write_fault, page);
1111                         up_read(&current->mm->mmap_sem);
1112                 } else
1113                         npages = get_user_pages_fast(addr, 1, write_fault,
1114                                                      page);
1115
1116                 /* map read fault as writable if possible */
1117                 if (unlikely(!write_fault) && npages == 1) {
1118                         struct page *wpage[1];
1119
1120                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1121                         if (npages == 1) {
1122                                 *writable = true;
1123                                 put_page(page[0]);
1124                                 page[0] = wpage[0];
1125                         }
1126                         npages = 1;
1127                 }
1128         }
1129
1130         if (unlikely(npages != 1)) {
1131                 struct vm_area_struct *vma;
1132
1133                 if (atomic)
1134                         return get_fault_pfn();
1135
1136                 down_read(&current->mm->mmap_sem);
1137                 if (npages == -EHWPOISON ||
1138                         (!async && check_user_page_hwpoison(addr))) {
1139                         up_read(&current->mm->mmap_sem);
1140                         get_page(hwpoison_page);
1141                         return page_to_pfn(hwpoison_page);
1142                 }
1143
1144                 vma = find_vma_intersection(current->mm, addr, addr+1);
1145
1146                 if (vma == NULL)
1147                         pfn = get_fault_pfn();
1148                 else if ((vma->vm_flags & VM_PFNMAP)) {
1149                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1150                                 vma->vm_pgoff;
1151                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1152                 } else {
1153                         if (async && (vma->vm_flags & VM_WRITE))
1154                                 *async = true;
1155                         pfn = get_fault_pfn();
1156                 }
1157                 up_read(&current->mm->mmap_sem);
1158         } else
1159                 pfn = page_to_pfn(page[0]);
1160
1161         return pfn;
1162 }
1163
1164 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1165 {
1166         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1167 }
1168 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1169
1170 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1171                           bool write_fault, bool *writable)
1172 {
1173         unsigned long addr;
1174
1175         if (async)
1176                 *async = false;
1177
1178         addr = gfn_to_hva(kvm, gfn);
1179         if (kvm_is_error_hva(addr)) {
1180                 get_page(bad_page);
1181                 return page_to_pfn(bad_page);
1182         }
1183
1184         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1185 }
1186
1187 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1188 {
1189         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1190 }
1191 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1192
1193 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1194                        bool write_fault, bool *writable)
1195 {
1196         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1197 }
1198 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1199
1200 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1201 {
1202         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1203 }
1204 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1205
1206 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1207                       bool *writable)
1208 {
1209         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1210 }
1211 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1212
1213 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1214                          struct kvm_memory_slot *slot, gfn_t gfn)
1215 {
1216         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1217         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1218 }
1219
1220 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1221                                                                   int nr_pages)
1222 {
1223         unsigned long addr;
1224         gfn_t entry;
1225
1226         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1227         if (kvm_is_error_hva(addr))
1228                 return -1;
1229
1230         if (entry < nr_pages)
1231                 return 0;
1232
1233         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1234 }
1235 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1236
1237 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1238 {
1239         pfn_t pfn;
1240
1241         pfn = gfn_to_pfn(kvm, gfn);
1242         if (!kvm_is_mmio_pfn(pfn))
1243                 return pfn_to_page(pfn);
1244
1245         WARN_ON(kvm_is_mmio_pfn(pfn));
1246
1247         get_page(bad_page);
1248         return bad_page;
1249 }
1250
1251 EXPORT_SYMBOL_GPL(gfn_to_page);
1252
1253 void kvm_release_page_clean(struct page *page)
1254 {
1255         kvm_release_pfn_clean(page_to_pfn(page));
1256 }
1257 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1258
1259 void kvm_release_pfn_clean(pfn_t pfn)
1260 {
1261         if (!kvm_is_mmio_pfn(pfn))
1262                 put_page(pfn_to_page(pfn));
1263 }
1264 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1265
1266 void kvm_release_page_dirty(struct page *page)
1267 {
1268         kvm_release_pfn_dirty(page_to_pfn(page));
1269 }
1270 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1271
1272 void kvm_release_pfn_dirty(pfn_t pfn)
1273 {
1274         kvm_set_pfn_dirty(pfn);
1275         kvm_release_pfn_clean(pfn);
1276 }
1277 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1278
1279 void kvm_set_page_dirty(struct page *page)
1280 {
1281         kvm_set_pfn_dirty(page_to_pfn(page));
1282 }
1283 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1284
1285 void kvm_set_pfn_dirty(pfn_t pfn)
1286 {
1287         if (!kvm_is_mmio_pfn(pfn)) {
1288                 struct page *page = pfn_to_page(pfn);
1289                 if (!PageReserved(page))
1290                         SetPageDirty(page);
1291         }
1292 }
1293 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1294
1295 void kvm_set_pfn_accessed(pfn_t pfn)
1296 {
1297         if (!kvm_is_mmio_pfn(pfn))
1298                 mark_page_accessed(pfn_to_page(pfn));
1299 }
1300 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1301
1302 void kvm_get_pfn(pfn_t pfn)
1303 {
1304         if (!kvm_is_mmio_pfn(pfn))
1305                 get_page(pfn_to_page(pfn));
1306 }
1307 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1308
1309 static int next_segment(unsigned long len, int offset)
1310 {
1311         if (len > PAGE_SIZE - offset)
1312                 return PAGE_SIZE - offset;
1313         else
1314                 return len;
1315 }
1316
1317 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1318                         int len)
1319 {
1320         int r;
1321         unsigned long addr;
1322
1323         addr = gfn_to_hva(kvm, gfn);
1324         if (kvm_is_error_hva(addr))
1325                 return -EFAULT;
1326         r = __copy_from_user(data, (void __user *)addr + offset, len);
1327         if (r)
1328                 return -EFAULT;
1329         return 0;
1330 }
1331 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1332
1333 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1334 {
1335         gfn_t gfn = gpa >> PAGE_SHIFT;
1336         int seg;
1337         int offset = offset_in_page(gpa);
1338         int ret;
1339
1340         while ((seg = next_segment(len, offset)) != 0) {
1341                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1342                 if (ret < 0)
1343                         return ret;
1344                 offset = 0;
1345                 len -= seg;
1346                 data += seg;
1347                 ++gfn;
1348         }
1349         return 0;
1350 }
1351 EXPORT_SYMBOL_GPL(kvm_read_guest);
1352
1353 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1354                           unsigned long len)
1355 {
1356         int r;
1357         unsigned long addr;
1358         gfn_t gfn = gpa >> PAGE_SHIFT;
1359         int offset = offset_in_page(gpa);
1360
1361         addr = gfn_to_hva(kvm, gfn);
1362         if (kvm_is_error_hva(addr))
1363                 return -EFAULT;
1364         pagefault_disable();
1365         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1366         pagefault_enable();
1367         if (r)
1368                 return -EFAULT;
1369         return 0;
1370 }
1371 EXPORT_SYMBOL(kvm_read_guest_atomic);
1372
1373 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1374                          int offset, int len)
1375 {
1376         int r;
1377         unsigned long addr;
1378
1379         addr = gfn_to_hva(kvm, gfn);
1380         if (kvm_is_error_hva(addr))
1381                 return -EFAULT;
1382         r = __copy_to_user((void __user *)addr + offset, data, len);
1383         if (r)
1384                 return -EFAULT;
1385         mark_page_dirty(kvm, gfn);
1386         return 0;
1387 }
1388 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1389
1390 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1391                     unsigned long len)
1392 {
1393         gfn_t gfn = gpa >> PAGE_SHIFT;
1394         int seg;
1395         int offset = offset_in_page(gpa);
1396         int ret;
1397
1398         while ((seg = next_segment(len, offset)) != 0) {
1399                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1400                 if (ret < 0)
1401                         return ret;
1402                 offset = 0;
1403                 len -= seg;
1404                 data += seg;
1405                 ++gfn;
1406         }
1407         return 0;
1408 }
1409
1410 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1411                               gpa_t gpa)
1412 {
1413         struct kvm_memslots *slots = kvm_memslots(kvm);
1414         int offset = offset_in_page(gpa);
1415         gfn_t gfn = gpa >> PAGE_SHIFT;
1416
1417         ghc->gpa = gpa;
1418         ghc->generation = slots->generation;
1419         ghc->memslot = __gfn_to_memslot(slots, gfn);
1420         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1421         if (!kvm_is_error_hva(ghc->hva))
1422                 ghc->hva += offset;
1423         else
1424                 return -EFAULT;
1425
1426         return 0;
1427 }
1428 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1429
1430 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1431                            void *data, unsigned long len)
1432 {
1433         struct kvm_memslots *slots = kvm_memslots(kvm);
1434         int r;
1435
1436         if (slots->generation != ghc->generation)
1437                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1438
1439         if (kvm_is_error_hva(ghc->hva))
1440                 return -EFAULT;
1441
1442         r = __copy_to_user((void __user *)ghc->hva, data, len);
1443         if (r)
1444                 return -EFAULT;
1445         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1446
1447         return 0;
1448 }
1449 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1450
1451 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1452                            void *data, unsigned long len)
1453 {
1454         struct kvm_memslots *slots = kvm_memslots(kvm);
1455         int r;
1456
1457         if (slots->generation != ghc->generation)
1458                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1459
1460         if (kvm_is_error_hva(ghc->hva))
1461                 return -EFAULT;
1462
1463         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1464         if (r)
1465                 return -EFAULT;
1466
1467         return 0;
1468 }
1469 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1470
1471 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1472 {
1473         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1474                                     offset, len);
1475 }
1476 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1477
1478 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1479 {
1480         gfn_t gfn = gpa >> PAGE_SHIFT;
1481         int seg;
1482         int offset = offset_in_page(gpa);
1483         int ret;
1484
1485         while ((seg = next_segment(len, offset)) != 0) {
1486                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1487                 if (ret < 0)
1488                         return ret;
1489                 offset = 0;
1490                 len -= seg;
1491                 ++gfn;
1492         }
1493         return 0;
1494 }
1495 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1496
1497 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1498                              gfn_t gfn)
1499 {
1500         if (memslot && memslot->dirty_bitmap) {
1501                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1502
1503                 if (!__test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap))
1504                         memslot->nr_dirty_pages++;
1505         }
1506 }
1507
1508 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1509 {
1510         struct kvm_memory_slot *memslot;
1511
1512         memslot = gfn_to_memslot(kvm, gfn);
1513         mark_page_dirty_in_slot(kvm, memslot, gfn);
1514 }
1515
1516 /*
1517  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1518  */
1519 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1520 {
1521         DEFINE_WAIT(wait);
1522
1523         for (;;) {
1524                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1525
1526                 if (kvm_arch_vcpu_runnable(vcpu)) {
1527                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1528                         break;
1529                 }
1530                 if (kvm_cpu_has_pending_timer(vcpu))
1531                         break;
1532                 if (signal_pending(current))
1533                         break;
1534
1535                 schedule();
1536         }
1537
1538         finish_wait(&vcpu->wq, &wait);
1539 }
1540
1541 void kvm_resched(struct kvm_vcpu *vcpu)
1542 {
1543         if (!need_resched())
1544                 return;
1545         cond_resched();
1546 }
1547 EXPORT_SYMBOL_GPL(kvm_resched);
1548
1549 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1550 {
1551         struct kvm *kvm = me->kvm;
1552         struct kvm_vcpu *vcpu;
1553         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1554         int yielded = 0;
1555         int pass;
1556         int i;
1557
1558         /*
1559          * We boost the priority of a VCPU that is runnable but not
1560          * currently running, because it got preempted by something
1561          * else and called schedule in __vcpu_run.  Hopefully that
1562          * VCPU is holding the lock that we need and will release it.
1563          * We approximate round-robin by starting at the last boosted VCPU.
1564          */
1565         for (pass = 0; pass < 2 && !yielded; pass++) {
1566                 kvm_for_each_vcpu(i, vcpu, kvm) {
1567                         struct task_struct *task = NULL;
1568                         struct pid *pid;
1569                         if (!pass && i < last_boosted_vcpu) {
1570                                 i = last_boosted_vcpu;
1571                                 continue;
1572                         } else if (pass && i > last_boosted_vcpu)
1573                                 break;
1574                         if (vcpu == me)
1575                                 continue;
1576                         if (waitqueue_active(&vcpu->wq))
1577                                 continue;
1578                         rcu_read_lock();
1579                         pid = rcu_dereference(vcpu->pid);
1580                         if (pid)
1581                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1582                         rcu_read_unlock();
1583                         if (!task)
1584                                 continue;
1585                         if (task->flags & PF_VCPU) {
1586                                 put_task_struct(task);
1587                                 continue;
1588                         }
1589                         if (yield_to(task, 1)) {
1590                                 put_task_struct(task);
1591                                 kvm->last_boosted_vcpu = i;
1592                                 yielded = 1;
1593                                 break;
1594                         }
1595                         put_task_struct(task);
1596                 }
1597         }
1598 }
1599 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1600
1601 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1602 {
1603         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1604         struct page *page;
1605
1606         if (vmf->pgoff == 0)
1607                 page = virt_to_page(vcpu->run);
1608 #ifdef CONFIG_X86
1609         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1610                 page = virt_to_page(vcpu->arch.pio_data);
1611 #endif
1612 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1613         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1614                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1615 #endif
1616         else
1617                 return VM_FAULT_SIGBUS;
1618         get_page(page);
1619         vmf->page = page;
1620         return 0;
1621 }
1622
1623 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1624         .fault = kvm_vcpu_fault,
1625 };
1626
1627 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1628 {
1629         vma->vm_ops = &kvm_vcpu_vm_ops;
1630         return 0;
1631 }
1632
1633 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1634 {
1635         struct kvm_vcpu *vcpu = filp->private_data;
1636
1637         kvm_put_kvm(vcpu->kvm);
1638         return 0;
1639 }
1640
1641 static struct file_operations kvm_vcpu_fops = {
1642         .release        = kvm_vcpu_release,
1643         .unlocked_ioctl = kvm_vcpu_ioctl,
1644 #ifdef CONFIG_COMPAT
1645         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1646 #endif
1647         .mmap           = kvm_vcpu_mmap,
1648         .llseek         = noop_llseek,
1649 };
1650
1651 /*
1652  * Allocates an inode for the vcpu.
1653  */
1654 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1655 {
1656         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1657 }
1658
1659 /*
1660  * Creates some virtual cpus.  Good luck creating more than one.
1661  */
1662 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1663 {
1664         int r;
1665         struct kvm_vcpu *vcpu, *v;
1666
1667         vcpu = kvm_arch_vcpu_create(kvm, id);
1668         if (IS_ERR(vcpu))
1669                 return PTR_ERR(vcpu);
1670
1671         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1672
1673         r = kvm_arch_vcpu_setup(vcpu);
1674         if (r)
1675                 goto vcpu_destroy;
1676
1677         mutex_lock(&kvm->lock);
1678         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1679                 r = -EINVAL;
1680                 goto unlock_vcpu_destroy;
1681         }
1682
1683         kvm_for_each_vcpu(r, v, kvm)
1684                 if (v->vcpu_id == id) {
1685                         r = -EEXIST;
1686                         goto unlock_vcpu_destroy;
1687                 }
1688
1689         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1690
1691         /* Now it's all set up, let userspace reach it */
1692         kvm_get_kvm(kvm);
1693         r = create_vcpu_fd(vcpu);
1694         if (r < 0) {
1695                 kvm_put_kvm(kvm);
1696                 goto unlock_vcpu_destroy;
1697         }
1698
1699         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1700         smp_wmb();
1701         atomic_inc(&kvm->online_vcpus);
1702
1703 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1704         if (kvm->bsp_vcpu_id == id)
1705                 kvm->bsp_vcpu = vcpu;
1706 #endif
1707         mutex_unlock(&kvm->lock);
1708         return r;
1709
1710 unlock_vcpu_destroy:
1711         mutex_unlock(&kvm->lock);
1712 vcpu_destroy:
1713         kvm_arch_vcpu_destroy(vcpu);
1714         return r;
1715 }
1716
1717 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1718 {
1719         if (sigset) {
1720                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1721                 vcpu->sigset_active = 1;
1722                 vcpu->sigset = *sigset;
1723         } else
1724                 vcpu->sigset_active = 0;
1725         return 0;
1726 }
1727
1728 static long kvm_vcpu_ioctl(struct file *filp,
1729                            unsigned int ioctl, unsigned long arg)
1730 {
1731         struct kvm_vcpu *vcpu = filp->private_data;
1732         void __user *argp = (void __user *)arg;
1733         int r;
1734         struct kvm_fpu *fpu = NULL;
1735         struct kvm_sregs *kvm_sregs = NULL;
1736
1737         if (vcpu->kvm->mm != current->mm)
1738                 return -EIO;
1739
1740 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1741         /*
1742          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1743          * so vcpu_load() would break it.
1744          */
1745         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1746                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1747 #endif
1748
1749
1750         vcpu_load(vcpu);
1751         switch (ioctl) {
1752         case KVM_RUN:
1753                 r = -EINVAL;
1754                 if (arg)
1755                         goto out;
1756                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1757                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1758                 break;
1759         case KVM_GET_REGS: {
1760                 struct kvm_regs *kvm_regs;
1761
1762                 r = -ENOMEM;
1763                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1764                 if (!kvm_regs)
1765                         goto out;
1766                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1767                 if (r)
1768                         goto out_free1;
1769                 r = -EFAULT;
1770                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1771                         goto out_free1;
1772                 r = 0;
1773 out_free1:
1774                 kfree(kvm_regs);
1775                 break;
1776         }
1777         case KVM_SET_REGS: {
1778                 struct kvm_regs *kvm_regs;
1779
1780                 r = -ENOMEM;
1781                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1782                 if (!kvm_regs)
1783                         goto out;
1784                 r = -EFAULT;
1785                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1786                         goto out_free2;
1787                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1788                 if (r)
1789                         goto out_free2;
1790                 r = 0;
1791 out_free2:
1792                 kfree(kvm_regs);
1793                 break;
1794         }
1795         case KVM_GET_SREGS: {
1796                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1797                 r = -ENOMEM;
1798                 if (!kvm_sregs)
1799                         goto out;
1800                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1801                 if (r)
1802                         goto out;
1803                 r = -EFAULT;
1804                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1805                         goto out;
1806                 r = 0;
1807                 break;
1808         }
1809         case KVM_SET_SREGS: {
1810                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1811                 r = -ENOMEM;
1812                 if (!kvm_sregs)
1813                         goto out;
1814                 r = -EFAULT;
1815                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1816                         goto out;
1817                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1818                 if (r)
1819                         goto out;
1820                 r = 0;
1821                 break;
1822         }
1823         case KVM_GET_MP_STATE: {
1824                 struct kvm_mp_state mp_state;
1825
1826                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1827                 if (r)
1828                         goto out;
1829                 r = -EFAULT;
1830                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1831                         goto out;
1832                 r = 0;
1833                 break;
1834         }
1835         case KVM_SET_MP_STATE: {
1836                 struct kvm_mp_state mp_state;
1837
1838                 r = -EFAULT;
1839                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1840                         goto out;
1841                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1842                 if (r)
1843                         goto out;
1844                 r = 0;
1845                 break;
1846         }
1847         case KVM_TRANSLATE: {
1848                 struct kvm_translation tr;
1849
1850                 r = -EFAULT;
1851                 if (copy_from_user(&tr, argp, sizeof tr))
1852                         goto out;
1853                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1854                 if (r)
1855                         goto out;
1856                 r = -EFAULT;
1857                 if (copy_to_user(argp, &tr, sizeof tr))
1858                         goto out;
1859                 r = 0;
1860                 break;
1861         }
1862         case KVM_SET_GUEST_DEBUG: {
1863                 struct kvm_guest_debug dbg;
1864
1865                 r = -EFAULT;
1866                 if (copy_from_user(&dbg, argp, sizeof dbg))
1867                         goto out;
1868                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1869                 if (r)
1870                         goto out;
1871                 r = 0;
1872                 break;
1873         }
1874         case KVM_SET_SIGNAL_MASK: {
1875                 struct kvm_signal_mask __user *sigmask_arg = argp;
1876                 struct kvm_signal_mask kvm_sigmask;
1877                 sigset_t sigset, *p;
1878
1879                 p = NULL;
1880                 if (argp) {
1881                         r = -EFAULT;
1882                         if (copy_from_user(&kvm_sigmask, argp,
1883                                            sizeof kvm_sigmask))
1884                                 goto out;
1885                         r = -EINVAL;
1886                         if (kvm_sigmask.len != sizeof sigset)
1887                                 goto out;
1888                         r = -EFAULT;
1889                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1890                                            sizeof sigset))
1891                                 goto out;
1892                         p = &sigset;
1893                 }
1894                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1895                 break;
1896         }
1897         case KVM_GET_FPU: {
1898                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1899                 r = -ENOMEM;
1900                 if (!fpu)
1901                         goto out;
1902                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1903                 if (r)
1904                         goto out;
1905                 r = -EFAULT;
1906                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1907                         goto out;
1908                 r = 0;
1909                 break;
1910         }
1911         case KVM_SET_FPU: {
1912                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1913                 r = -ENOMEM;
1914                 if (!fpu)
1915                         goto out;
1916                 r = -EFAULT;
1917                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1918                         goto out;
1919                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1920                 if (r)
1921                         goto out;
1922                 r = 0;
1923                 break;
1924         }
1925         default:
1926                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1927         }
1928 out:
1929         vcpu_put(vcpu);
1930         kfree(fpu);
1931         kfree(kvm_sregs);
1932         return r;
1933 }
1934
1935 #ifdef CONFIG_COMPAT
1936 static long kvm_vcpu_compat_ioctl(struct file *filp,
1937                                   unsigned int ioctl, unsigned long arg)
1938 {
1939         struct kvm_vcpu *vcpu = filp->private_data;
1940         void __user *argp = compat_ptr(arg);
1941         int r;
1942
1943         if (vcpu->kvm->mm != current->mm)
1944                 return -EIO;
1945
1946         switch (ioctl) {
1947         case KVM_SET_SIGNAL_MASK: {
1948                 struct kvm_signal_mask __user *sigmask_arg = argp;
1949                 struct kvm_signal_mask kvm_sigmask;
1950                 compat_sigset_t csigset;
1951                 sigset_t sigset;
1952
1953                 if (argp) {
1954                         r = -EFAULT;
1955                         if (copy_from_user(&kvm_sigmask, argp,
1956                                            sizeof kvm_sigmask))
1957                                 goto out;
1958                         r = -EINVAL;
1959                         if (kvm_sigmask.len != sizeof csigset)
1960                                 goto out;
1961                         r = -EFAULT;
1962                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1963                                            sizeof csigset))
1964                                 goto out;
1965                 }
1966                 sigset_from_compat(&sigset, &csigset);
1967                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1968                 break;
1969         }
1970         default:
1971                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1972         }
1973
1974 out:
1975         return r;
1976 }
1977 #endif
1978
1979 static long kvm_vm_ioctl(struct file *filp,
1980                            unsigned int ioctl, unsigned long arg)
1981 {
1982         struct kvm *kvm = filp->private_data;
1983         void __user *argp = (void __user *)arg;
1984         int r;
1985
1986         if (kvm->mm != current->mm)
1987                 return -EIO;
1988         switch (ioctl) {
1989         case KVM_CREATE_VCPU:
1990                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1991                 if (r < 0)
1992                         goto out;
1993                 break;
1994         case KVM_SET_USER_MEMORY_REGION: {
1995                 struct kvm_userspace_memory_region kvm_userspace_mem;
1996
1997                 r = -EFAULT;
1998                 if (copy_from_user(&kvm_userspace_mem, argp,
1999                                                 sizeof kvm_userspace_mem))
2000                         goto out;
2001
2002                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2003                 if (r)
2004                         goto out;
2005                 break;
2006         }
2007         case KVM_GET_DIRTY_LOG: {
2008                 struct kvm_dirty_log log;
2009
2010                 r = -EFAULT;
2011                 if (copy_from_user(&log, argp, sizeof log))
2012                         goto out;
2013                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2014                 if (r)
2015                         goto out;
2016                 break;
2017         }
2018 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2019         case KVM_REGISTER_COALESCED_MMIO: {
2020                 struct kvm_coalesced_mmio_zone zone;
2021                 r = -EFAULT;
2022                 if (copy_from_user(&zone, argp, sizeof zone))
2023                         goto out;
2024                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2025                 if (r)
2026                         goto out;
2027                 r = 0;
2028                 break;
2029         }
2030         case KVM_UNREGISTER_COALESCED_MMIO: {
2031                 struct kvm_coalesced_mmio_zone zone;
2032                 r = -EFAULT;
2033                 if (copy_from_user(&zone, argp, sizeof zone))
2034                         goto out;
2035                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2036                 if (r)
2037                         goto out;
2038                 r = 0;
2039                 break;
2040         }
2041 #endif
2042         case KVM_IRQFD: {
2043                 struct kvm_irqfd data;
2044
2045                 r = -EFAULT;
2046                 if (copy_from_user(&data, argp, sizeof data))
2047                         goto out;
2048                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2049                 break;
2050         }
2051         case KVM_IOEVENTFD: {
2052                 struct kvm_ioeventfd data;
2053
2054                 r = -EFAULT;
2055                 if (copy_from_user(&data, argp, sizeof data))
2056                         goto out;
2057                 r = kvm_ioeventfd(kvm, &data);
2058                 break;
2059         }
2060 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2061         case KVM_SET_BOOT_CPU_ID:
2062                 r = 0;
2063                 mutex_lock(&kvm->lock);
2064                 if (atomic_read(&kvm->online_vcpus) != 0)
2065                         r = -EBUSY;
2066                 else
2067                         kvm->bsp_vcpu_id = arg;
2068                 mutex_unlock(&kvm->lock);
2069                 break;
2070 #endif
2071         default:
2072                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2073                 if (r == -ENOTTY)
2074                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2075         }
2076 out:
2077         return r;
2078 }
2079
2080 #ifdef CONFIG_COMPAT
2081 struct compat_kvm_dirty_log {
2082         __u32 slot;
2083         __u32 padding1;
2084         union {
2085                 compat_uptr_t dirty_bitmap; /* one bit per page */
2086                 __u64 padding2;
2087         };
2088 };
2089
2090 static long kvm_vm_compat_ioctl(struct file *filp,
2091                            unsigned int ioctl, unsigned long arg)
2092 {
2093         struct kvm *kvm = filp->private_data;
2094         int r;
2095
2096         if (kvm->mm != current->mm)
2097                 return -EIO;
2098         switch (ioctl) {
2099         case KVM_GET_DIRTY_LOG: {
2100                 struct compat_kvm_dirty_log compat_log;
2101                 struct kvm_dirty_log log;
2102
2103                 r = -EFAULT;
2104                 if (copy_from_user(&compat_log, (void __user *)arg,
2105                                    sizeof(compat_log)))
2106                         goto out;
2107                 log.slot         = compat_log.slot;
2108                 log.padding1     = compat_log.padding1;
2109                 log.padding2     = compat_log.padding2;
2110                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2111
2112                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2113                 if (r)
2114                         goto out;
2115                 break;
2116         }
2117         default:
2118                 r = kvm_vm_ioctl(filp, ioctl, arg);
2119         }
2120
2121 out:
2122         return r;
2123 }
2124 #endif
2125
2126 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2127 {
2128         struct page *page[1];
2129         unsigned long addr;
2130         int npages;
2131         gfn_t gfn = vmf->pgoff;
2132         struct kvm *kvm = vma->vm_file->private_data;
2133
2134         addr = gfn_to_hva(kvm, gfn);
2135         if (kvm_is_error_hva(addr))
2136                 return VM_FAULT_SIGBUS;
2137
2138         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2139                                 NULL);
2140         if (unlikely(npages != 1))
2141                 return VM_FAULT_SIGBUS;
2142
2143         vmf->page = page[0];
2144         return 0;
2145 }
2146
2147 static const struct vm_operations_struct kvm_vm_vm_ops = {
2148         .fault = kvm_vm_fault,
2149 };
2150
2151 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2152 {
2153         vma->vm_ops = &kvm_vm_vm_ops;
2154         return 0;
2155 }
2156
2157 static struct file_operations kvm_vm_fops = {
2158         .release        = kvm_vm_release,
2159         .unlocked_ioctl = kvm_vm_ioctl,
2160 #ifdef CONFIG_COMPAT
2161         .compat_ioctl   = kvm_vm_compat_ioctl,
2162 #endif
2163         .mmap           = kvm_vm_mmap,
2164         .llseek         = noop_llseek,
2165 };
2166
2167 static int kvm_dev_ioctl_create_vm(void)
2168 {
2169         int r;
2170         struct kvm *kvm;
2171
2172         kvm = kvm_create_vm();
2173         if (IS_ERR(kvm))
2174                 return PTR_ERR(kvm);
2175 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2176         r = kvm_coalesced_mmio_init(kvm);
2177         if (r < 0) {
2178                 kvm_put_kvm(kvm);
2179                 return r;
2180         }
2181 #endif
2182         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2183         if (r < 0)
2184                 kvm_put_kvm(kvm);
2185
2186         return r;
2187 }
2188
2189 static long kvm_dev_ioctl_check_extension_generic(long arg)
2190 {
2191         switch (arg) {
2192         case KVM_CAP_USER_MEMORY:
2193         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2194         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2195 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2196         case KVM_CAP_SET_BOOT_CPU_ID:
2197 #endif
2198         case KVM_CAP_INTERNAL_ERROR_DATA:
2199                 return 1;
2200 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2201         case KVM_CAP_IRQ_ROUTING:
2202                 return KVM_MAX_IRQ_ROUTES;
2203 #endif
2204         default:
2205                 break;
2206         }
2207         return kvm_dev_ioctl_check_extension(arg);
2208 }
2209
2210 static long kvm_dev_ioctl(struct file *filp,
2211                           unsigned int ioctl, unsigned long arg)
2212 {
2213         long r = -EINVAL;
2214
2215         switch (ioctl) {
2216         case KVM_GET_API_VERSION:
2217                 r = -EINVAL;
2218                 if (arg)
2219                         goto out;
2220                 r = KVM_API_VERSION;
2221                 break;
2222         case KVM_CREATE_VM:
2223                 r = -EINVAL;
2224                 if (arg)
2225                         goto out;
2226                 r = kvm_dev_ioctl_create_vm();
2227                 break;
2228         case KVM_CHECK_EXTENSION:
2229                 r = kvm_dev_ioctl_check_extension_generic(arg);
2230                 break;
2231         case KVM_GET_VCPU_MMAP_SIZE:
2232                 r = -EINVAL;
2233                 if (arg)
2234                         goto out;
2235                 r = PAGE_SIZE;     /* struct kvm_run */
2236 #ifdef CONFIG_X86
2237                 r += PAGE_SIZE;    /* pio data page */
2238 #endif
2239 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2240                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2241 #endif
2242                 break;
2243         case KVM_TRACE_ENABLE:
2244         case KVM_TRACE_PAUSE:
2245         case KVM_TRACE_DISABLE:
2246                 r = -EOPNOTSUPP;
2247                 break;
2248         default:
2249                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2250         }
2251 out:
2252         return r;
2253 }
2254
2255 static struct file_operations kvm_chardev_ops = {
2256         .unlocked_ioctl = kvm_dev_ioctl,
2257         .compat_ioctl   = kvm_dev_ioctl,
2258         .llseek         = noop_llseek,
2259 };
2260
2261 static struct miscdevice kvm_dev = {
2262         KVM_MINOR,
2263         "kvm",
2264         &kvm_chardev_ops,
2265 };
2266
2267 static void hardware_enable_nolock(void *junk)
2268 {
2269         int cpu = raw_smp_processor_id();
2270         int r;
2271
2272         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2273                 return;
2274
2275         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2276
2277         r = kvm_arch_hardware_enable(NULL);
2278
2279         if (r) {
2280                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2281                 atomic_inc(&hardware_enable_failed);
2282                 printk(KERN_INFO "kvm: enabling virtualization on "
2283                                  "CPU%d failed\n", cpu);
2284         }
2285 }
2286
2287 static void hardware_enable(void *junk)
2288 {
2289         raw_spin_lock(&kvm_lock);
2290         hardware_enable_nolock(junk);
2291         raw_spin_unlock(&kvm_lock);
2292 }
2293
2294 static void hardware_disable_nolock(void *junk)
2295 {
2296         int cpu = raw_smp_processor_id();
2297
2298         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2299                 return;
2300         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2301         kvm_arch_hardware_disable(NULL);
2302 }
2303
2304 static void hardware_disable(void *junk)
2305 {
2306         raw_spin_lock(&kvm_lock);
2307         hardware_disable_nolock(junk);
2308         raw_spin_unlock(&kvm_lock);
2309 }
2310
2311 static void hardware_disable_all_nolock(void)
2312 {
2313         BUG_ON(!kvm_usage_count);
2314
2315         kvm_usage_count--;
2316         if (!kvm_usage_count)
2317                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2318 }
2319
2320 static void hardware_disable_all(void)
2321 {
2322         raw_spin_lock(&kvm_lock);
2323         hardware_disable_all_nolock();
2324         raw_spin_unlock(&kvm_lock);
2325 }
2326
2327 static int hardware_enable_all(void)
2328 {
2329         int r = 0;
2330
2331         raw_spin_lock(&kvm_lock);
2332
2333         kvm_usage_count++;
2334         if (kvm_usage_count == 1) {
2335                 atomic_set(&hardware_enable_failed, 0);
2336                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2337
2338                 if (atomic_read(&hardware_enable_failed)) {
2339                         hardware_disable_all_nolock();
2340                         r = -EBUSY;
2341                 }
2342         }
2343
2344         raw_spin_unlock(&kvm_lock);
2345
2346         return r;
2347 }
2348
2349 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2350                            void *v)
2351 {
2352         int cpu = (long)v;
2353
2354         if (!kvm_usage_count)
2355                 return NOTIFY_OK;
2356
2357         val &= ~CPU_TASKS_FROZEN;
2358         switch (val) {
2359         case CPU_DYING:
2360                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2361                        cpu);
2362                 hardware_disable(NULL);
2363                 break;
2364         case CPU_STARTING:
2365                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2366                        cpu);
2367                 hardware_enable(NULL);
2368                 break;
2369         }
2370         return NOTIFY_OK;
2371 }
2372
2373
2374 asmlinkage void kvm_spurious_fault(void)
2375 {
2376         /* Fault while not rebooting.  We want the trace. */
2377         BUG();
2378 }
2379 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2380
2381 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2382                       void *v)
2383 {
2384         /*
2385          * Some (well, at least mine) BIOSes hang on reboot if
2386          * in vmx root mode.
2387          *
2388          * And Intel TXT required VMX off for all cpu when system shutdown.
2389          */
2390         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2391         kvm_rebooting = true;
2392         on_each_cpu(hardware_disable_nolock, NULL, 1);
2393         return NOTIFY_OK;
2394 }
2395
2396 static struct notifier_block kvm_reboot_notifier = {
2397         .notifier_call = kvm_reboot,
2398         .priority = 0,
2399 };
2400
2401 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2402 {
2403         int i;
2404
2405         for (i = 0; i < bus->dev_count; i++) {
2406                 struct kvm_io_device *pos = bus->range[i].dev;
2407
2408                 kvm_iodevice_destructor(pos);
2409         }
2410         kfree(bus);
2411 }
2412
2413 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2414 {
2415         const struct kvm_io_range *r1 = p1;
2416         const struct kvm_io_range *r2 = p2;
2417
2418         if (r1->addr < r2->addr)
2419                 return -1;
2420         if (r1->addr + r1->len > r2->addr + r2->len)
2421                 return 1;
2422         return 0;
2423 }
2424
2425 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2426                           gpa_t addr, int len)
2427 {
2428         if (bus->dev_count == NR_IOBUS_DEVS)
2429                 return -ENOSPC;
2430
2431         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2432                 .addr = addr,
2433                 .len = len,
2434                 .dev = dev,
2435         };
2436
2437         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2438                 kvm_io_bus_sort_cmp, NULL);
2439
2440         return 0;
2441 }
2442
2443 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2444                              gpa_t addr, int len)
2445 {
2446         struct kvm_io_range *range, key;
2447         int off;
2448
2449         key = (struct kvm_io_range) {
2450                 .addr = addr,
2451                 .len = len,
2452         };
2453
2454         range = bsearch(&key, bus->range, bus->dev_count,
2455                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2456         if (range == NULL)
2457                 return -ENOENT;
2458
2459         off = range - bus->range;
2460
2461         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2462                 off--;
2463
2464         return off;
2465 }
2466
2467 /* kvm_io_bus_write - called under kvm->slots_lock */
2468 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2469                      int len, const void *val)
2470 {
2471         int idx;
2472         struct kvm_io_bus *bus;
2473         struct kvm_io_range range;
2474
2475         range = (struct kvm_io_range) {
2476                 .addr = addr,
2477                 .len = len,
2478         };
2479
2480         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2481         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2482         if (idx < 0)
2483                 return -EOPNOTSUPP;
2484
2485         while (idx < bus->dev_count &&
2486                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2487                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2488                         return 0;
2489                 idx++;
2490         }
2491
2492         return -EOPNOTSUPP;
2493 }
2494
2495 /* kvm_io_bus_read - called under kvm->slots_lock */
2496 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2497                     int len, void *val)
2498 {
2499         int idx;
2500         struct kvm_io_bus *bus;
2501         struct kvm_io_range range;
2502
2503         range = (struct kvm_io_range) {
2504                 .addr = addr,
2505                 .len = len,
2506         };
2507
2508         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2509         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2510         if (idx < 0)
2511                 return -EOPNOTSUPP;
2512
2513         while (idx < bus->dev_count &&
2514                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2515                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2516                         return 0;
2517                 idx++;
2518         }
2519
2520         return -EOPNOTSUPP;
2521 }
2522
2523 /* Caller must hold slots_lock. */
2524 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2525                             int len, struct kvm_io_device *dev)
2526 {
2527         struct kvm_io_bus *new_bus, *bus;
2528
2529         bus = kvm->buses[bus_idx];
2530         if (bus->dev_count > NR_IOBUS_DEVS-1)
2531                 return -ENOSPC;
2532
2533         new_bus = kmemdup(bus, sizeof(struct kvm_io_bus), GFP_KERNEL);
2534         if (!new_bus)
2535                 return -ENOMEM;
2536         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2537         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2538         synchronize_srcu_expedited(&kvm->srcu);
2539         kfree(bus);
2540
2541         return 0;
2542 }
2543
2544 /* Caller must hold slots_lock. */
2545 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2546                               struct kvm_io_device *dev)
2547 {
2548         int i, r;
2549         struct kvm_io_bus *new_bus, *bus;
2550
2551         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2552         if (!new_bus)
2553                 return -ENOMEM;
2554
2555         bus = kvm->buses[bus_idx];
2556         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2557
2558         r = -ENOENT;
2559         for (i = 0; i < new_bus->dev_count; i++)
2560                 if (new_bus->range[i].dev == dev) {
2561                         r = 0;
2562                         new_bus->dev_count--;
2563                         new_bus->range[i] = new_bus->range[new_bus->dev_count];
2564                         sort(new_bus->range, new_bus->dev_count,
2565                              sizeof(struct kvm_io_range),
2566                              kvm_io_bus_sort_cmp, NULL);
2567                         break;
2568                 }
2569
2570         if (r) {
2571                 kfree(new_bus);
2572                 return r;
2573         }
2574
2575         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2576         synchronize_srcu_expedited(&kvm->srcu);
2577         kfree(bus);
2578         return r;
2579 }
2580
2581 static struct notifier_block kvm_cpu_notifier = {
2582         .notifier_call = kvm_cpu_hotplug,
2583 };
2584
2585 static int vm_stat_get(void *_offset, u64 *val)
2586 {
2587         unsigned offset = (long)_offset;
2588         struct kvm *kvm;
2589
2590         *val = 0;
2591         raw_spin_lock(&kvm_lock);
2592         list_for_each_entry(kvm, &vm_list, vm_list)
2593                 *val += *(u32 *)((void *)kvm + offset);
2594         raw_spin_unlock(&kvm_lock);
2595         return 0;
2596 }
2597
2598 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2599
2600 static int vcpu_stat_get(void *_offset, u64 *val)
2601 {
2602         unsigned offset = (long)_offset;
2603         struct kvm *kvm;
2604         struct kvm_vcpu *vcpu;
2605         int i;
2606
2607         *val = 0;
2608         raw_spin_lock(&kvm_lock);
2609         list_for_each_entry(kvm, &vm_list, vm_list)
2610                 kvm_for_each_vcpu(i, vcpu, kvm)
2611                         *val += *(u32 *)((void *)vcpu + offset);
2612
2613         raw_spin_unlock(&kvm_lock);
2614         return 0;
2615 }
2616
2617 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2618
2619 static const struct file_operations *stat_fops[] = {
2620         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2621         [KVM_STAT_VM]   = &vm_stat_fops,
2622 };
2623
2624 static void kvm_init_debug(void)
2625 {
2626         struct kvm_stats_debugfs_item *p;
2627
2628         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2629         for (p = debugfs_entries; p->name; ++p)
2630                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2631                                                 (void *)(long)p->offset,
2632                                                 stat_fops[p->kind]);
2633 }
2634
2635 static void kvm_exit_debug(void)
2636 {
2637         struct kvm_stats_debugfs_item *p;
2638
2639         for (p = debugfs_entries; p->name; ++p)
2640                 debugfs_remove(p->dentry);
2641         debugfs_remove(kvm_debugfs_dir);
2642 }
2643
2644 static int kvm_suspend(void)
2645 {
2646         if (kvm_usage_count)
2647                 hardware_disable_nolock(NULL);
2648         return 0;
2649 }
2650
2651 static void kvm_resume(void)
2652 {
2653         if (kvm_usage_count) {
2654                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2655                 hardware_enable_nolock(NULL);
2656         }
2657 }
2658
2659 static struct syscore_ops kvm_syscore_ops = {
2660         .suspend = kvm_suspend,
2661         .resume = kvm_resume,
2662 };
2663
2664 struct page *bad_page;
2665 pfn_t bad_pfn;
2666
2667 static inline
2668 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2669 {
2670         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2671 }
2672
2673 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2674 {
2675         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2676
2677         kvm_arch_vcpu_load(vcpu, cpu);
2678 }
2679
2680 static void kvm_sched_out(struct preempt_notifier *pn,
2681                           struct task_struct *next)
2682 {
2683         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2684
2685         kvm_arch_vcpu_put(vcpu);
2686 }
2687
2688 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2689                   struct module *module)
2690 {
2691         int r;
2692         int cpu;
2693
2694         r = kvm_arch_init(opaque);
2695         if (r)
2696                 goto out_fail;
2697
2698         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2699
2700         if (bad_page == NULL) {
2701                 r = -ENOMEM;
2702                 goto out;
2703         }
2704
2705         bad_pfn = page_to_pfn(bad_page);
2706
2707         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2708
2709         if (hwpoison_page == NULL) {
2710                 r = -ENOMEM;
2711                 goto out_free_0;
2712         }
2713
2714         hwpoison_pfn = page_to_pfn(hwpoison_page);
2715
2716         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2717
2718         if (fault_page == NULL) {
2719                 r = -ENOMEM;
2720                 goto out_free_0;
2721         }
2722
2723         fault_pfn = page_to_pfn(fault_page);
2724
2725         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2726                 r = -ENOMEM;
2727                 goto out_free_0;
2728         }
2729
2730         r = kvm_arch_hardware_setup();
2731         if (r < 0)
2732                 goto out_free_0a;
2733
2734         for_each_online_cpu(cpu) {
2735                 smp_call_function_single(cpu,
2736                                 kvm_arch_check_processor_compat,
2737                                 &r, 1);
2738                 if (r < 0)
2739                         goto out_free_1;
2740         }
2741
2742         r = register_cpu_notifier(&kvm_cpu_notifier);
2743         if (r)
2744                 goto out_free_2;
2745         register_reboot_notifier(&kvm_reboot_notifier);
2746
2747         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2748         if (!vcpu_align)
2749                 vcpu_align = __alignof__(struct kvm_vcpu);
2750         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2751                                            0, NULL);
2752         if (!kvm_vcpu_cache) {
2753                 r = -ENOMEM;
2754                 goto out_free_3;
2755         }
2756
2757         r = kvm_async_pf_init();
2758         if (r)
2759                 goto out_free;
2760
2761         kvm_chardev_ops.owner = module;
2762         kvm_vm_fops.owner = module;
2763         kvm_vcpu_fops.owner = module;
2764
2765         r = misc_register(&kvm_dev);
2766         if (r) {
2767                 printk(KERN_ERR "kvm: misc device register failed\n");
2768                 goto out_unreg;
2769         }
2770
2771         register_syscore_ops(&kvm_syscore_ops);
2772
2773         kvm_preempt_ops.sched_in = kvm_sched_in;
2774         kvm_preempt_ops.sched_out = kvm_sched_out;
2775
2776         kvm_init_debug();
2777
2778         return 0;
2779
2780 out_unreg:
2781         kvm_async_pf_deinit();
2782 out_free:
2783         kmem_cache_destroy(kvm_vcpu_cache);
2784 out_free_3:
2785         unregister_reboot_notifier(&kvm_reboot_notifier);
2786         unregister_cpu_notifier(&kvm_cpu_notifier);
2787 out_free_2:
2788 out_free_1:
2789         kvm_arch_hardware_unsetup();
2790 out_free_0a:
2791         free_cpumask_var(cpus_hardware_enabled);
2792 out_free_0:
2793         if (fault_page)
2794                 __free_page(fault_page);
2795         if (hwpoison_page)
2796                 __free_page(hwpoison_page);
2797         __free_page(bad_page);
2798 out:
2799         kvm_arch_exit();
2800 out_fail:
2801         return r;
2802 }
2803 EXPORT_SYMBOL_GPL(kvm_init);
2804
2805 void kvm_exit(void)
2806 {
2807         kvm_exit_debug();
2808         misc_deregister(&kvm_dev);
2809         kmem_cache_destroy(kvm_vcpu_cache);
2810         kvm_async_pf_deinit();
2811         unregister_syscore_ops(&kvm_syscore_ops);
2812         unregister_reboot_notifier(&kvm_reboot_notifier);
2813         unregister_cpu_notifier(&kvm_cpu_notifier);
2814         on_each_cpu(hardware_disable_nolock, NULL, 1);
2815         kvm_arch_hardware_unsetup();
2816         kvm_arch_exit();
2817         free_cpumask_var(cpus_hardware_enabled);
2818         __free_page(hwpoison_page);
2819         __free_page(bad_page);
2820 }
2821 EXPORT_SYMBOL_GPL(kvm_exit);