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         struct kvm_memslots *slots = kvm->memslots;
 551         struct kvm_memory_slot *memslot;
 552
 553         kvm_for_each_memslot(memslot, slots)
 554                 kvm_free_physmem_slot(memslot, 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                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 638
 639                 *old = *new;
 640                 if (id >= slots->nmemslots)
 641                         slots->nmemslots = id + 1;
 642         }
 643
 644         slots->generation++;
 645 }
 646
 647 /*
 648  * Allocate some memory and give it an address in the guest physical address
 649  * space.
 650  *
 651  * Discontiguous memory is allowed, mostly for framebuffers.
 652  *
 653  * Must be called holding mmap_sem for write.
 654  */
 655 int __kvm_set_memory_region(struct kvm *kvm,
 656                             struct kvm_userspace_memory_region *mem,
 657                             int user_alloc)
 658 {
 659         int r;
 660         gfn_t base_gfn;
 661         unsigned long npages;
 662         unsigned long i;
 663         struct kvm_memory_slot *memslot;
 664         struct kvm_memory_slot old, new;
 665         struct kvm_memslots *slots, *old_memslots;
 666
 667         r = -EINVAL;
 668         /* General sanity checks */
 669         if (mem->memory_size & (PAGE_SIZE - 1))
 670                 goto out;
 671         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 672                 goto out;
 673         /* We can read the guest memory with __xxx_user() later on. */
 674         if (user_alloc &&
 675             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 676              !access_ok(VERIFY_WRITE,
 677                         (void __user *)(unsigned long)mem->userspace_addr,
 678                         mem->memory_size)))
 679                 goto out;
 680         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 681                 goto out;
 682         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 683                 goto out;
 684
 685         memslot = id_to_memslot(kvm->memslots, mem->slot);
 686         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 687         npages = mem->memory_size >> PAGE_SHIFT;
 688
 689         r = -EINVAL;
 690         if (npages > KVM_MEM_MAX_NR_PAGES)
 691                 goto out;
 692
 693         if (!npages)
 694                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 695
 696         new = old = *memslot;
 697
 698         new.id = mem->slot;
 699         new.base_gfn = base_gfn;
 700         new.npages = npages;
 701         new.flags = mem->flags;
 702
 703         /* Disallow changing a memory slot's size. */
 704         r = -EINVAL;
 705         if (npages && old.npages && npages != old.npages)
 706                 goto out_free;
 707
 708         /* Check for overlaps */
 709         r = -EEXIST;
 710         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 711                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 712
 713                 if (s == memslot || !s->npages)
 714                         continue;
 715                 if (!((base_gfn + npages <= s->base_gfn) ||
 716                       (base_gfn >= s->base_gfn + s->npages)))
 717                         goto out_free;
 718         }
 719
 720         /* Free page dirty bitmap if unneeded */
 721         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 722                 new.dirty_bitmap = NULL;
 723
 724         r = -ENOMEM;
 725
 726         /* Allocate if a slot is being created */
 727 #ifndef CONFIG_S390
 728         if (npages && !new.rmap) {
 729                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
 730
 731                 if (!new.rmap)
 732                         goto out_free;
 733
 734                 new.user_alloc = user_alloc;
 735                 new.userspace_addr = mem->userspace_addr;
 736         }
 737         if (!npages)
 738                 goto skip_lpage;
 739
 740         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 741                 unsigned long ugfn;
 742                 unsigned long j;
 743                 int lpages;
 744                 int level = i + 2;
 745
 746                 /* Avoid unused variable warning if no large pages */
 747                 (void)level;
 748
 749                 if (new.lpage_info[i])
 750                         continue;
 751
 752                 lpages = 1 + ((base_gfn + npages - 1)
 753                              >> KVM_HPAGE_GFN_SHIFT(level));
 754                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
 755
 756                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
 757
 758                 if (!new.lpage_info[i])
 759                         goto out_free;
 760
 761                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
 762                         new.lpage_info[i][0].write_count = 1;
 763                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
 764                         new.lpage_info[i][lpages - 1].write_count = 1;
 765                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 766                 /*
 767                  * If the gfn and userspace address are not aligned wrt each
 768                  * other, or if explicitly asked to, disable large page
 769                  * support for this slot
 770                  */
 771                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 772                     !largepages_enabled)
 773                         for (j = 0; j < lpages; ++j)
 774                                 new.lpage_info[i][j].write_count = 1;
 775         }
 776
 777 skip_lpage:
 778
 779         /* Allocate page dirty bitmap if needed */
 780         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 781                 if (kvm_create_dirty_bitmap(&new) < 0)
 782                         goto out_free;
 783                 /* destroy any largepage mappings for dirty tracking */
 784         }
 785 #else  /* not defined CONFIG_S390 */
 786         new.user_alloc = user_alloc;
 787         if (user_alloc)
 788                 new.userspace_addr = mem->userspace_addr;
 789 #endif /* not defined CONFIG_S390 */
 790
 791         if (!npages) {
 792                 struct kvm_memory_slot *slot;
 793
 794                 r = -ENOMEM;
 795                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 796                                 GFP_KERNEL);
 797                 if (!slots)
 798                         goto out_free;
 799                 slot = id_to_memslot(slots, mem->slot);
 800                 slot->flags |= KVM_MEMSLOT_INVALID;
 801
 802                 update_memslots(slots, NULL);
 803
 804                 old_memslots = kvm->memslots;
 805                 rcu_assign_pointer(kvm->memslots, slots);
 806                 synchronize_srcu_expedited(&kvm->srcu);
 807                 /* From this point no new shadow pages pointing to a deleted
 808                  * memslot will be created.
 809                  *
 810                  * validation of sp->gfn happens in:
 811                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 812                  *      - kvm_is_visible_gfn (mmu_check_roots)
 813                  */
 814                 kvm_arch_flush_shadow(kvm);
 815                 kfree(old_memslots);
 816         }
 817
 818         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 819         if (r)
 820                 goto out_free;
 821
 822         /* map the pages in iommu page table */
 823         if (npages) {
 824                 r = kvm_iommu_map_pages(kvm, &new);
 825                 if (r)
 826                         goto out_free;
 827         }
 828
 829         r = -ENOMEM;
 830         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 831                         GFP_KERNEL);
 832         if (!slots)
 833                 goto out_free;
 834
 835         /* actual memory is freed via old in kvm_free_physmem_slot below */
 836         if (!npages) {
 837                 new.rmap = NULL;
 838                 new.dirty_bitmap = NULL;
 839                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 840                         new.lpage_info[i] = NULL;
 841         }
 842
 843         update_memslots(slots, &new);
 844         old_memslots = kvm->memslots;
 845         rcu_assign_pointer(kvm->memslots, slots);
 846         synchronize_srcu_expedited(&kvm->srcu);
 847
 848         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 849
 850         /*
 851          * If the new memory slot is created, we need to clear all
 852          * mmio sptes.
 853          */
 854         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 855                 kvm_arch_flush_shadow(kvm);
 856
 857         kvm_free_physmem_slot(&old, &new);
 858         kfree(old_memslots);
 859
 860         return 0;
 861
 862 out_free:
 863         kvm_free_physmem_slot(&new, &old);
 864 out:
 865         return r;
 866
 867 }
 868 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 869
 870 int kvm_set_memory_region(struct kvm *kvm,
 871                           struct kvm_userspace_memory_region *mem,
 872                           int user_alloc)
 873 {
 874         int r;
 875
 876         mutex_lock(&kvm->slots_lock);
 877         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 878         mutex_unlock(&kvm->slots_lock);
 879         return r;
 880 }
 881 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 882
 883 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 884                                    struct
 885                                    kvm_userspace_memory_region *mem,
 886                                    int user_alloc)
 887 {
 888         if (mem->slot >= KVM_MEMORY_SLOTS)
 889                 return -EINVAL;
 890         return kvm_set_memory_region(kvm, mem, user_alloc);
 891 }
 892
 893 int kvm_get_dirty_log(struct kvm *kvm,
 894                         struct kvm_dirty_log *log, int *is_dirty)
 895 {
 896         struct kvm_memory_slot *memslot;
 897         int r, i;
 898         unsigned long n;
 899         unsigned long any = 0;
 900
 901         r = -EINVAL;
 902         if (log->slot >= KVM_MEMORY_SLOTS)
 903                 goto out;
 904
 905         memslot = id_to_memslot(kvm->memslots, log->slot);
 906         r = -ENOENT;
 907         if (!memslot->dirty_bitmap)
 908                 goto out;
 909
 910         n = kvm_dirty_bitmap_bytes(memslot);
 911
 912         for (i = 0; !any && i < n/sizeof(long); ++i)
 913                 any = memslot->dirty_bitmap[i];
 914
 915         r = -EFAULT;
 916         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 917                 goto out;
 918
 919         if (any)
 920                 *is_dirty = 1;
 921
 922         r = 0;
 923 out:
 924         return r;
 925 }
 926
 927 void kvm_disable_largepages(void)
 928 {
 929         largepages_enabled = false;
 930 }
 931 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 932
 933 int is_error_page(struct page *page)
 934 {
 935         return page == bad_page || page == hwpoison_page || page == fault_page;
 936 }
 937 EXPORT_SYMBOL_GPL(is_error_page);
 938
 939 int is_error_pfn(pfn_t pfn)
 940 {
 941         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 942 }
 943 EXPORT_SYMBOL_GPL(is_error_pfn);
 944
 945 int is_hwpoison_pfn(pfn_t pfn)
 946 {
 947         return pfn == hwpoison_pfn;
 948 }
 949 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 950
 951 int is_fault_pfn(pfn_t pfn)
 952 {
 953         return pfn == fault_pfn;
 954 }
 955 EXPORT_SYMBOL_GPL(is_fault_pfn);
 956
 957 int is_noslot_pfn(pfn_t pfn)
 958 {
 959         return pfn == bad_pfn;
 960 }
 961 EXPORT_SYMBOL_GPL(is_noslot_pfn);
 962
 963 int is_invalid_pfn(pfn_t pfn)
 964 {
 965         return pfn == hwpoison_pfn || pfn == fault_pfn;
 966 }
 967 EXPORT_SYMBOL_GPL(is_invalid_pfn);
 968
 969 static inline unsigned long bad_hva(void)
 970 {
 971         return PAGE_OFFSET;
 972 }
 973
 974 int kvm_is_error_hva(unsigned long addr)
 975 {
 976         return addr == bad_hva();
 977 }
 978 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 979
 980 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
 981                                                 gfn_t gfn)
 982 {
 983         struct kvm_memory_slot *memslot;
 984
 985         kvm_for_each_memslot(memslot, slots)
 986                 if (gfn >= memslot->base_gfn
 987                     && gfn < memslot->base_gfn + memslot->npages)
 988                         return memslot;
 989
 990         return NULL;
 991 }
 992
 993 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 994 {
 995         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 996 }
 997 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 998
 999 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1000 {
1001         int i;
1002         struct kvm_memslots *slots = kvm_memslots(kvm);
1003
1004         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1005                 struct kvm_memory_slot *memslot = &slots->memslots[i];
1006
1007                 if (memslot->flags & KVM_MEMSLOT_INVALID)
1008                         continue;
1009
1010                 if (gfn >= memslot->base_gfn
1011                     && gfn < memslot->base_gfn + memslot->npages)
1012                         return 1;
1013         }
1014         return 0;
1015 }
1016 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1017
1018 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1019 {
1020         struct vm_area_struct *vma;
1021         unsigned long addr, size;
1022
1023         size = PAGE_SIZE;
1024
1025         addr = gfn_to_hva(kvm, gfn);
1026         if (kvm_is_error_hva(addr))
1027                 return PAGE_SIZE;
1028
1029         down_read(&current->mm->mmap_sem);
1030         vma = find_vma(current->mm, addr);
1031         if (!vma)
1032                 goto out;
1033
1034         size = vma_kernel_pagesize(vma);
1035
1036 out:
1037         up_read(&current->mm->mmap_sem);
1038
1039         return size;
1040 }
1041
1042 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1043                                      gfn_t *nr_pages)
1044 {
1045         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1046                 return bad_hva();
1047
1048         if (nr_pages)
1049                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1050
1051         return gfn_to_hva_memslot(slot, gfn);
1052 }
1053
1054 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1055 {
1056         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1057 }
1058 EXPORT_SYMBOL_GPL(gfn_to_hva);
1059
1060 static pfn_t get_fault_pfn(void)
1061 {
1062         get_page(fault_page);
1063         return fault_pfn;
1064 }
1065
1066 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1067         unsigned long start, int write, struct page **page)
1068 {
1069         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1070
1071         if (write)
1072                 flags |= FOLL_WRITE;
1073
1074         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1075 }
1076
1077 static inline int check_user_page_hwpoison(unsigned long addr)
1078 {
1079         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1080
1081         rc = __get_user_pages(current, current->mm, addr, 1,
1082                               flags, NULL, NULL, NULL);
1083         return rc == -EHWPOISON;
1084 }
1085
1086 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1087                         bool *async, bool write_fault, bool *writable)
1088 {
1089         struct page *page[1];
1090         int npages = 0;
1091         pfn_t pfn;
1092
1093         /* we can do it either atomically or asynchronously, not both */
1094         BUG_ON(atomic && async);
1095
1096         BUG_ON(!write_fault && !writable);
1097
1098         if (writable)
1099                 *writable = true;
1100
1101         if (atomic || async)
1102                 npages = __get_user_pages_fast(addr, 1, 1, page);
1103
1104         if (unlikely(npages != 1) && !atomic) {
1105                 might_sleep();
1106
1107                 if (writable)
1108                         *writable = write_fault;
1109
1110                 if (async) {
1111                         down_read(&current->mm->mmap_sem);
1112                         npages = get_user_page_nowait(current, current->mm,
1113                                                      addr, write_fault, page);
1114                         up_read(&current->mm->mmap_sem);
1115                 } else
1116                         npages = get_user_pages_fast(addr, 1, write_fault,
1117                                                      page);
1118
1119                 /* map read fault as writable if possible */
1120                 if (unlikely(!write_fault) && npages == 1) {
1121                         struct page *wpage[1];
1122
1123                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1124                         if (npages == 1) {
1125                                 *writable = true;
1126                                 put_page(page[0]);
1127                                 page[0] = wpage[0];
1128                         }
1129                         npages = 1;
1130                 }
1131         }
1132
1133         if (unlikely(npages != 1)) {
1134                 struct vm_area_struct *vma;
1135
1136                 if (atomic)
1137                         return get_fault_pfn();
1138
1139                 down_read(&current->mm->mmap_sem);
1140                 if (npages == -EHWPOISON ||
1141                         (!async && check_user_page_hwpoison(addr))) {
1142                         up_read(&current->mm->mmap_sem);
1143                         get_page(hwpoison_page);
1144                         return page_to_pfn(hwpoison_page);
1145                 }
1146
1147                 vma = find_vma_intersection(current->mm, addr, addr+1);
1148
1149                 if (vma == NULL)
1150                         pfn = get_fault_pfn();
1151                 else if ((vma->vm_flags & VM_PFNMAP)) {
1152                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1153                                 vma->vm_pgoff;
1154                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1155                 } else {
1156                         if (async && (vma->vm_flags & VM_WRITE))
1157                                 *async = true;
1158                         pfn = get_fault_pfn();
1159                 }
1160                 up_read(&current->mm->mmap_sem);
1161         } else
1162                 pfn = page_to_pfn(page[0]);
1163
1164         return pfn;
1165 }
1166
1167 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1168 {
1169         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1170 }
1171 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1172
1173 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1174                           bool write_fault, bool *writable)
1175 {
1176         unsigned long addr;
1177
1178         if (async)
1179                 *async = false;
1180
1181         addr = gfn_to_hva(kvm, gfn);
1182         if (kvm_is_error_hva(addr)) {
1183                 get_page(bad_page);
1184                 return page_to_pfn(bad_page);
1185         }
1186
1187         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1188 }
1189
1190 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1191 {
1192         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1193 }
1194 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1195
1196 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1197                        bool write_fault, bool *writable)
1198 {
1199         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1200 }
1201 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1202
1203 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1204 {
1205         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1206 }
1207 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1208
1209 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1210                       bool *writable)
1211 {
1212         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1213 }
1214 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1215
1216 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1217                          struct kvm_memory_slot *slot, gfn_t gfn)
1218 {
1219         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1220         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1221 }
1222
1223 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1224                                                                   int nr_pages)
1225 {
1226         unsigned long addr;
1227         gfn_t entry;
1228
1229         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1230         if (kvm_is_error_hva(addr))
1231                 return -1;
1232
1233         if (entry < nr_pages)
1234                 return 0;
1235
1236         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1237 }
1238 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1239
1240 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1241 {
1242         pfn_t pfn;
1243
1244         pfn = gfn_to_pfn(kvm, gfn);
1245         if (!kvm_is_mmio_pfn(pfn))
1246                 return pfn_to_page(pfn);
1247
1248         WARN_ON(kvm_is_mmio_pfn(pfn));
1249
1250         get_page(bad_page);
1251         return bad_page;
1252 }
1253
1254 EXPORT_SYMBOL_GPL(gfn_to_page);
1255
1256 void kvm_release_page_clean(struct page *page)
1257 {
1258         kvm_release_pfn_clean(page_to_pfn(page));
1259 }
1260 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1261
1262 void kvm_release_pfn_clean(pfn_t pfn)
1263 {
1264         if (!kvm_is_mmio_pfn(pfn))
1265                 put_page(pfn_to_page(pfn));
1266 }
1267 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1268
1269 void kvm_release_page_dirty(struct page *page)
1270 {
1271         kvm_release_pfn_dirty(page_to_pfn(page));
1272 }
1273 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1274
1275 void kvm_release_pfn_dirty(pfn_t pfn)
1276 {
1277         kvm_set_pfn_dirty(pfn);
1278         kvm_release_pfn_clean(pfn);
1279 }
1280 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1281
1282 void kvm_set_page_dirty(struct page *page)
1283 {
1284         kvm_set_pfn_dirty(page_to_pfn(page));
1285 }
1286 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1287
1288 void kvm_set_pfn_dirty(pfn_t pfn)
1289 {
1290         if (!kvm_is_mmio_pfn(pfn)) {
1291                 struct page *page = pfn_to_page(pfn);
1292                 if (!PageReserved(page))
1293                         SetPageDirty(page);
1294         }
1295 }
1296 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1297
1298 void kvm_set_pfn_accessed(pfn_t pfn)
1299 {
1300         if (!kvm_is_mmio_pfn(pfn))
1301                 mark_page_accessed(pfn_to_page(pfn));
1302 }
1303 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1304
1305 void kvm_get_pfn(pfn_t pfn)
1306 {
1307         if (!kvm_is_mmio_pfn(pfn))
1308                 get_page(pfn_to_page(pfn));
1309 }
1310 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1311
1312 static int next_segment(unsigned long len, int offset)
1313 {
1314         if (len > PAGE_SIZE - offset)
1315                 return PAGE_SIZE - offset;
1316         else
1317                 return len;
1318 }
1319
1320 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1321                         int len)
1322 {
1323         int r;
1324         unsigned long addr;
1325
1326         addr = gfn_to_hva(kvm, gfn);
1327         if (kvm_is_error_hva(addr))
1328                 return -EFAULT;
1329         r = __copy_from_user(data, (void __user *)addr + offset, len);
1330         if (r)
1331                 return -EFAULT;
1332         return 0;
1333 }
1334 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1335
1336 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1337 {
1338         gfn_t gfn = gpa >> PAGE_SHIFT;
1339         int seg;
1340         int offset = offset_in_page(gpa);
1341         int ret;
1342
1343         while ((seg = next_segment(len, offset)) != 0) {
1344                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1345                 if (ret < 0)
1346                         return ret;
1347                 offset = 0;
1348                 len -= seg;
1349                 data += seg;
1350                 ++gfn;
1351         }
1352         return 0;
1353 }
1354 EXPORT_SYMBOL_GPL(kvm_read_guest);
1355
1356 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1357                           unsigned long len)
1358 {
1359         int r;
1360         unsigned long addr;
1361         gfn_t gfn = gpa >> PAGE_SHIFT;
1362         int offset = offset_in_page(gpa);
1363
1364         addr = gfn_to_hva(kvm, gfn);
1365         if (kvm_is_error_hva(addr))
1366                 return -EFAULT;
1367         pagefault_disable();
1368         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1369         pagefault_enable();
1370         if (r)
1371                 return -EFAULT;
1372         return 0;
1373 }
1374 EXPORT_SYMBOL(kvm_read_guest_atomic);
1375
1376 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1377                          int offset, int len)
1378 {
1379         int r;
1380         unsigned long addr;
1381
1382         addr = gfn_to_hva(kvm, gfn);
1383         if (kvm_is_error_hva(addr))
1384                 return -EFAULT;
1385         r = __copy_to_user((void __user *)addr + offset, data, len);
1386         if (r)
1387                 return -EFAULT;
1388         mark_page_dirty(kvm, gfn);
1389         return 0;
1390 }
1391 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1392
1393 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1394                     unsigned long len)
1395 {
1396         gfn_t gfn = gpa >> PAGE_SHIFT;
1397         int seg;
1398         int offset = offset_in_page(gpa);
1399         int ret;
1400
1401         while ((seg = next_segment(len, offset)) != 0) {
1402                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1403                 if (ret < 0)
1404                         return ret;
1405                 offset = 0;
1406                 len -= seg;
1407                 data += seg;
1408                 ++gfn;
1409         }
1410         return 0;
1411 }
1412
1413 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1414                               gpa_t gpa)
1415 {
1416         struct kvm_memslots *slots = kvm_memslots(kvm);
1417         int offset = offset_in_page(gpa);
1418         gfn_t gfn = gpa >> PAGE_SHIFT;
1419
1420         ghc->gpa = gpa;
1421         ghc->generation = slots->generation;
1422         ghc->memslot = __gfn_to_memslot(slots, gfn);
1423         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1424         if (!kvm_is_error_hva(ghc->hva))
1425                 ghc->hva += offset;
1426         else
1427                 return -EFAULT;
1428
1429         return 0;
1430 }
1431 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1432
1433 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1434                            void *data, unsigned long len)
1435 {
1436         struct kvm_memslots *slots = kvm_memslots(kvm);
1437         int r;
1438
1439         if (slots->generation != ghc->generation)
1440                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1441
1442         if (kvm_is_error_hva(ghc->hva))
1443                 return -EFAULT;
1444
1445         r = __copy_to_user((void __user *)ghc->hva, data, len);
1446         if (r)
1447                 return -EFAULT;
1448         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1449
1450         return 0;
1451 }
1452 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1453
1454 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1455                            void *data, unsigned long len)
1456 {
1457         struct kvm_memslots *slots = kvm_memslots(kvm);
1458         int r;
1459
1460         if (slots->generation != ghc->generation)
1461                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1462
1463         if (kvm_is_error_hva(ghc->hva))
1464                 return -EFAULT;
1465
1466         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1467         if (r)
1468                 return -EFAULT;
1469
1470         return 0;
1471 }
1472 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1473
1474 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1475 {
1476         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1477                                     offset, len);
1478 }
1479 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1480
1481 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1482 {
1483         gfn_t gfn = gpa >> PAGE_SHIFT;
1484         int seg;
1485         int offset = offset_in_page(gpa);
1486         int ret;
1487
1488         while ((seg = next_segment(len, offset)) != 0) {
1489                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1490                 if (ret < 0)
1491                         return ret;
1492                 offset = 0;
1493                 len -= seg;
1494                 ++gfn;
1495         }
1496         return 0;
1497 }
1498 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1499
1500 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1501                              gfn_t gfn)
1502 {
1503         if (memslot && memslot->dirty_bitmap) {
1504                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1505
1506                 if (!__test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap))
1507                         memslot->nr_dirty_pages++;
1508         }
1509 }
1510
1511 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1512 {
1513         struct kvm_memory_slot *memslot;
1514
1515         memslot = gfn_to_memslot(kvm, gfn);
1516         mark_page_dirty_in_slot(kvm, memslot, gfn);
1517 }
1518
1519 /*
1520  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1521  */
1522 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1523 {
1524         DEFINE_WAIT(wait);
1525
1526         for (;;) {
1527                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1528
1529                 if (kvm_arch_vcpu_runnable(vcpu)) {
1530                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1531                         break;
1532                 }
1533                 if (kvm_cpu_has_pending_timer(vcpu))
1534                         break;
1535                 if (signal_pending(current))
1536                         break;
1537
1538                 schedule();
1539         }
1540
1541         finish_wait(&vcpu->wq, &wait);
1542 }
1543
1544 void kvm_resched(struct kvm_vcpu *vcpu)
1545 {
1546         if (!need_resched())
1547                 return;
1548         cond_resched();
1549 }
1550 EXPORT_SYMBOL_GPL(kvm_resched);
1551
1552 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1553 {
1554         struct kvm *kvm = me->kvm;
1555         struct kvm_vcpu *vcpu;
1556         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1557         int yielded = 0;
1558         int pass;
1559         int i;
1560
1561         /*
1562          * We boost the priority of a VCPU that is runnable but not
1563          * currently running, because it got preempted by something
1564          * else and called schedule in __vcpu_run.  Hopefully that
1565          * VCPU is holding the lock that we need and will release it.
1566          * We approximate round-robin by starting at the last boosted VCPU.
1567          */
1568         for (pass = 0; pass < 2 && !yielded; pass++) {
1569                 kvm_for_each_vcpu(i, vcpu, kvm) {
1570                         struct task_struct *task = NULL;
1571                         struct pid *pid;
1572                         if (!pass && i < last_boosted_vcpu) {
1573                                 i = last_boosted_vcpu;
1574                                 continue;
1575                         } else if (pass && i > last_boosted_vcpu)
1576                                 break;
1577                         if (vcpu == me)
1578                                 continue;
1579                         if (waitqueue_active(&vcpu->wq))
1580                                 continue;
1581                         rcu_read_lock();
1582                         pid = rcu_dereference(vcpu->pid);
1583                         if (pid)
1584                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1585                         rcu_read_unlock();
1586                         if (!task)
1587                                 continue;
1588                         if (task->flags & PF_VCPU) {
1589                                 put_task_struct(task);
1590                                 continue;
1591                         }
1592                         if (yield_to(task, 1)) {
1593                                 put_task_struct(task);
1594                                 kvm->last_boosted_vcpu = i;
1595                                 yielded = 1;
1596                                 break;
1597                         }
1598                         put_task_struct(task);
1599                 }
1600         }
1601 }
1602 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1603
1604 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1605 {
1606         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1607         struct page *page;
1608
1609         if (vmf->pgoff == 0)
1610                 page = virt_to_page(vcpu->run);
1611 #ifdef CONFIG_X86
1612         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1613                 page = virt_to_page(vcpu->arch.pio_data);
1614 #endif
1615 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1616         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1617                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1618 #endif
1619         else
1620                 return VM_FAULT_SIGBUS;
1621         get_page(page);
1622         vmf->page = page;
1623         return 0;
1624 }
1625
1626 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1627         .fault = kvm_vcpu_fault,
1628 };
1629
1630 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1631 {
1632         vma->vm_ops = &kvm_vcpu_vm_ops;
1633         return 0;
1634 }
1635
1636 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1637 {
1638         struct kvm_vcpu *vcpu = filp->private_data;
1639
1640         kvm_put_kvm(vcpu->kvm);
1641         return 0;
1642 }
1643
1644 static struct file_operations kvm_vcpu_fops = {
1645         .release        = kvm_vcpu_release,
1646         .unlocked_ioctl = kvm_vcpu_ioctl,
1647 #ifdef CONFIG_COMPAT
1648         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1649 #endif
1650         .mmap           = kvm_vcpu_mmap,
1651         .llseek         = noop_llseek,
1652 };
1653
1654 /*
1655  * Allocates an inode for the vcpu.
1656  */
1657 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1658 {
1659         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1660 }
1661
1662 /*
1663  * Creates some virtual cpus.  Good luck creating more than one.
1664  */
1665 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1666 {
1667         int r;
1668         struct kvm_vcpu *vcpu, *v;
1669
1670         vcpu = kvm_arch_vcpu_create(kvm, id);
1671         if (IS_ERR(vcpu))
1672                 return PTR_ERR(vcpu);
1673
1674         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1675
1676         r = kvm_arch_vcpu_setup(vcpu);
1677         if (r)
1678                 goto vcpu_destroy;
1679
1680         mutex_lock(&kvm->lock);
1681         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1682                 r = -EINVAL;
1683                 goto unlock_vcpu_destroy;
1684         }
1685
1686         kvm_for_each_vcpu(r, v, kvm)
1687                 if (v->vcpu_id == id) {
1688                         r = -EEXIST;
1689                         goto unlock_vcpu_destroy;
1690                 }
1691
1692         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1693
1694         /* Now it's all set up, let userspace reach it */
1695         kvm_get_kvm(kvm);
1696         r = create_vcpu_fd(vcpu);
1697         if (r < 0) {
1698                 kvm_put_kvm(kvm);
1699                 goto unlock_vcpu_destroy;
1700         }
1701
1702         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1703         smp_wmb();
1704         atomic_inc(&kvm->online_vcpus);
1705
1706 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1707         if (kvm->bsp_vcpu_id == id)
1708                 kvm->bsp_vcpu = vcpu;
1709 #endif
1710         mutex_unlock(&kvm->lock);
1711         return r;
1712
1713 unlock_vcpu_destroy:
1714         mutex_unlock(&kvm->lock);
1715 vcpu_destroy:
1716         kvm_arch_vcpu_destroy(vcpu);
1717         return r;
1718 }
1719
1720 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1721 {
1722         if (sigset) {
1723                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1724                 vcpu->sigset_active = 1;
1725                 vcpu->sigset = *sigset;
1726         } else
1727                 vcpu->sigset_active = 0;
1728         return 0;
1729 }
1730
1731 static long kvm_vcpu_ioctl(struct file *filp,
1732                            unsigned int ioctl, unsigned long arg)
1733 {
1734         struct kvm_vcpu *vcpu = filp->private_data;
1735         void __user *argp = (void __user *)arg;
1736         int r;
1737         struct kvm_fpu *fpu = NULL;
1738         struct kvm_sregs *kvm_sregs = NULL;
1739
1740         if (vcpu->kvm->mm != current->mm)
1741                 return -EIO;
1742
1743 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1744         /*
1745          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1746          * so vcpu_load() would break it.
1747          */
1748         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1749                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1750 #endif
1751
1752
1753         vcpu_load(vcpu);
1754         switch (ioctl) {
1755         case KVM_RUN:
1756                 r = -EINVAL;
1757                 if (arg)
1758                         goto out;
1759                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1760                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1761                 break;
1762         case KVM_GET_REGS: {
1763                 struct kvm_regs *kvm_regs;
1764
1765                 r = -ENOMEM;
1766                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1767                 if (!kvm_regs)
1768                         goto out;
1769                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1770                 if (r)
1771                         goto out_free1;
1772                 r = -EFAULT;
1773                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1774                         goto out_free1;
1775                 r = 0;
1776 out_free1:
1777                 kfree(kvm_regs);
1778                 break;
1779         }
1780         case KVM_SET_REGS: {
1781                 struct kvm_regs *kvm_regs;
1782
1783                 r = -ENOMEM;
1784                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1785                 if (!kvm_regs)
1786                         goto out;
1787                 r = -EFAULT;
1788                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1789                         goto out_free2;
1790                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1791                 if (r)
1792                         goto out_free2;
1793                 r = 0;
1794 out_free2:
1795                 kfree(kvm_regs);
1796                 break;
1797         }
1798         case KVM_GET_SREGS: {
1799                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1800                 r = -ENOMEM;
1801                 if (!kvm_sregs)
1802                         goto out;
1803                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1804                 if (r)
1805                         goto out;
1806                 r = -EFAULT;
1807                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1808                         goto out;
1809                 r = 0;
1810                 break;
1811         }
1812         case KVM_SET_SREGS: {
1813                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1814                 r = -ENOMEM;
1815                 if (!kvm_sregs)
1816                         goto out;
1817                 r = -EFAULT;
1818                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1819                         goto out;
1820                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1821                 if (r)
1822                         goto out;
1823                 r = 0;
1824                 break;
1825         }
1826         case KVM_GET_MP_STATE: {
1827                 struct kvm_mp_state mp_state;
1828
1829                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1830                 if (r)
1831                         goto out;
1832                 r = -EFAULT;
1833                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1834                         goto out;
1835                 r = 0;
1836                 break;
1837         }
1838         case KVM_SET_MP_STATE: {
1839                 struct kvm_mp_state mp_state;
1840
1841                 r = -EFAULT;
1842                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1843                         goto out;
1844                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1845                 if (r)
1846                         goto out;
1847                 r = 0;
1848                 break;
1849         }
1850         case KVM_TRANSLATE: {
1851                 struct kvm_translation tr;
1852
1853                 r = -EFAULT;
1854                 if (copy_from_user(&tr, argp, sizeof tr))
1855                         goto out;
1856                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1857                 if (r)
1858                         goto out;
1859                 r = -EFAULT;
1860                 if (copy_to_user(argp, &tr, sizeof tr))
1861                         goto out;
1862                 r = 0;
1863                 break;
1864         }
1865         case KVM_SET_GUEST_DEBUG: {
1866                 struct kvm_guest_debug dbg;
1867
1868                 r = -EFAULT;
1869                 if (copy_from_user(&dbg, argp, sizeof dbg))
1870                         goto out;
1871                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1872                 if (r)
1873                         goto out;
1874                 r = 0;
1875                 break;
1876         }
1877         case KVM_SET_SIGNAL_MASK: {
1878                 struct kvm_signal_mask __user *sigmask_arg = argp;
1879                 struct kvm_signal_mask kvm_sigmask;
1880                 sigset_t sigset, *p;
1881
1882                 p = NULL;
1883                 if (argp) {
1884                         r = -EFAULT;
1885                         if (copy_from_user(&kvm_sigmask, argp,
1886                                            sizeof kvm_sigmask))
1887                                 goto out;
1888                         r = -EINVAL;
1889                         if (kvm_sigmask.len != sizeof sigset)
1890                                 goto out;
1891                         r = -EFAULT;
1892                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1893                                            sizeof sigset))
1894                                 goto out;
1895                         p = &sigset;
1896                 }
1897                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1898                 break;
1899         }
1900         case KVM_GET_FPU: {
1901                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1902                 r = -ENOMEM;
1903                 if (!fpu)
1904                         goto out;
1905                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1906                 if (r)
1907                         goto out;
1908                 r = -EFAULT;
1909                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1910                         goto out;
1911                 r = 0;
1912                 break;
1913         }
1914         case KVM_SET_FPU: {
1915                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1916                 r = -ENOMEM;
1917                 if (!fpu)
1918                         goto out;
1919                 r = -EFAULT;
1920                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1921                         goto out;
1922                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1923                 if (r)
1924                         goto out;
1925                 r = 0;
1926                 break;
1927         }
1928         default:
1929                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1930         }
1931 out:
1932         vcpu_put(vcpu);
1933         kfree(fpu);
1934         kfree(kvm_sregs);
1935         return r;
1936 }
1937
1938 #ifdef CONFIG_COMPAT
1939 static long kvm_vcpu_compat_ioctl(struct file *filp,
1940                                   unsigned int ioctl, unsigned long arg)
1941 {
1942         struct kvm_vcpu *vcpu = filp->private_data;
1943         void __user *argp = compat_ptr(arg);
1944         int r;
1945
1946         if (vcpu->kvm->mm != current->mm)
1947                 return -EIO;
1948
1949         switch (ioctl) {
1950         case KVM_SET_SIGNAL_MASK: {
1951                 struct kvm_signal_mask __user *sigmask_arg = argp;
1952                 struct kvm_signal_mask kvm_sigmask;
1953                 compat_sigset_t csigset;
1954                 sigset_t sigset;
1955
1956                 if (argp) {
1957                         r = -EFAULT;
1958                         if (copy_from_user(&kvm_sigmask, argp,
1959                                            sizeof kvm_sigmask))
1960                                 goto out;
1961                         r = -EINVAL;
1962                         if (kvm_sigmask.len != sizeof csigset)
1963                                 goto out;
1964                         r = -EFAULT;
1965                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1966                                            sizeof csigset))
1967                                 goto out;
1968                 }
1969                 sigset_from_compat(&sigset, &csigset);
1970                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1971                 break;
1972         }
1973         default:
1974                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1975         }
1976
1977 out:
1978         return r;
1979 }
1980 #endif
1981
1982 static long kvm_vm_ioctl(struct file *filp,
1983                            unsigned int ioctl, unsigned long arg)
1984 {
1985         struct kvm *kvm = filp->private_data;
1986         void __user *argp = (void __user *)arg;
1987         int r;
1988
1989         if (kvm->mm != current->mm)
1990                 return -EIO;
1991         switch (ioctl) {
1992         case KVM_CREATE_VCPU:
1993                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1994                 if (r < 0)
1995                         goto out;
1996                 break;
1997         case KVM_SET_USER_MEMORY_REGION: {
1998                 struct kvm_userspace_memory_region kvm_userspace_mem;
1999
2000                 r = -EFAULT;
2001                 if (copy_from_user(&kvm_userspace_mem, argp,
2002                                                 sizeof kvm_userspace_mem))
2003                         goto out;
2004
2005                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2006                 if (r)
2007                         goto out;
2008                 break;
2009         }
2010         case KVM_GET_DIRTY_LOG: {
2011                 struct kvm_dirty_log log;
2012
2013                 r = -EFAULT;
2014                 if (copy_from_user(&log, argp, sizeof log))
2015                         goto out;
2016                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2017                 if (r)
2018                         goto out;
2019                 break;
2020         }
2021 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2022         case KVM_REGISTER_COALESCED_MMIO: {
2023                 struct kvm_coalesced_mmio_zone zone;
2024                 r = -EFAULT;
2025                 if (copy_from_user(&zone, argp, sizeof zone))
2026                         goto out;
2027                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2028                 if (r)
2029                         goto out;
2030                 r = 0;
2031                 break;
2032         }
2033         case KVM_UNREGISTER_COALESCED_MMIO: {
2034                 struct kvm_coalesced_mmio_zone zone;
2035                 r = -EFAULT;
2036                 if (copy_from_user(&zone, argp, sizeof zone))
2037                         goto out;
2038                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2039                 if (r)
2040                         goto out;
2041                 r = 0;
2042                 break;
2043         }
2044 #endif
2045         case KVM_IRQFD: {
2046                 struct kvm_irqfd data;
2047
2048                 r = -EFAULT;
2049                 if (copy_from_user(&data, argp, sizeof data))
2050                         goto out;
2051                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2052                 break;
2053         }
2054         case KVM_IOEVENTFD: {
2055                 struct kvm_ioeventfd data;
2056
2057                 r = -EFAULT;
2058                 if (copy_from_user(&data, argp, sizeof data))
2059                         goto out;
2060                 r = kvm_ioeventfd(kvm, &data);
2061                 break;
2062         }
2063 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2064         case KVM_SET_BOOT_CPU_ID:
2065                 r = 0;
2066                 mutex_lock(&kvm->lock);
2067                 if (atomic_read(&kvm->online_vcpus) != 0)
2068                         r = -EBUSY;
2069                 else
2070                         kvm->bsp_vcpu_id = arg;
2071                 mutex_unlock(&kvm->lock);
2072                 break;
2073 #endif
2074         default:
2075                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2076                 if (r == -ENOTTY)
2077                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2078         }
2079 out:
2080         return r;
2081 }
2082
2083 #ifdef CONFIG_COMPAT
2084 struct compat_kvm_dirty_log {
2085         __u32 slot;
2086         __u32 padding1;
2087         union {
2088                 compat_uptr_t dirty_bitmap; /* one bit per page */
2089                 __u64 padding2;
2090         };
2091 };
2092
2093 static long kvm_vm_compat_ioctl(struct file *filp,
2094                            unsigned int ioctl, unsigned long arg)
2095 {
2096         struct kvm *kvm = filp->private_data;
2097         int r;
2098
2099         if (kvm->mm != current->mm)
2100                 return -EIO;
2101         switch (ioctl) {
2102         case KVM_GET_DIRTY_LOG: {
2103                 struct compat_kvm_dirty_log compat_log;
2104                 struct kvm_dirty_log log;
2105
2106                 r = -EFAULT;
2107                 if (copy_from_user(&compat_log, (void __user *)arg,
2108                                    sizeof(compat_log)))
2109                         goto out;
2110                 log.slot         = compat_log.slot;
2111                 log.padding1     = compat_log.padding1;
2112                 log.padding2     = compat_log.padding2;
2113                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2114
2115                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2116                 if (r)
2117                         goto out;
2118                 break;
2119         }
2120         default:
2121                 r = kvm_vm_ioctl(filp, ioctl, arg);
2122         }
2123
2124 out:
2125         return r;
2126 }
2127 #endif
2128
2129 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2130 {
2131         struct page *page[1];
2132         unsigned long addr;
2133         int npages;
2134         gfn_t gfn = vmf->pgoff;
2135         struct kvm *kvm = vma->vm_file->private_data;
2136
2137         addr = gfn_to_hva(kvm, gfn);
2138         if (kvm_is_error_hva(addr))
2139                 return VM_FAULT_SIGBUS;
2140
2141         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2142                                 NULL);
2143         if (unlikely(npages != 1))
2144                 return VM_FAULT_SIGBUS;
2145
2146         vmf->page = page[0];
2147         return 0;
2148 }
2149
2150 static const struct vm_operations_struct kvm_vm_vm_ops = {
2151         .fault = kvm_vm_fault,
2152 };
2153
2154 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2155 {
2156         vma->vm_ops = &kvm_vm_vm_ops;
2157         return 0;
2158 }
2159
2160 static struct file_operations kvm_vm_fops = {
2161         .release        = kvm_vm_release,
2162         .unlocked_ioctl = kvm_vm_ioctl,
2163 #ifdef CONFIG_COMPAT
2164         .compat_ioctl   = kvm_vm_compat_ioctl,
2165 #endif
2166         .mmap           = kvm_vm_mmap,
2167         .llseek         = noop_llseek,
2168 };
2169
2170 static int kvm_dev_ioctl_create_vm(void)
2171 {
2172         int r;
2173         struct kvm *kvm;
2174
2175         kvm = kvm_create_vm();
2176         if (IS_ERR(kvm))
2177                 return PTR_ERR(kvm);
2178 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2179         r = kvm_coalesced_mmio_init(kvm);
2180         if (r < 0) {
2181                 kvm_put_kvm(kvm);
2182                 return r;
2183         }
2184 #endif
2185         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2186         if (r < 0)
2187                 kvm_put_kvm(kvm);
2188
2189         return r;
2190 }
2191
2192 static long kvm_dev_ioctl_check_extension_generic(long arg)
2193 {
2194         switch (arg) {
2195         case KVM_CAP_USER_MEMORY:
2196         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2197         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2198 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2199         case KVM_CAP_SET_BOOT_CPU_ID:
2200 #endif
2201         case KVM_CAP_INTERNAL_ERROR_DATA:
2202                 return 1;
2203 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2204         case KVM_CAP_IRQ_ROUTING:
2205                 return KVM_MAX_IRQ_ROUTES;
2206 #endif
2207         default:
2208                 break;
2209         }
2210         return kvm_dev_ioctl_check_extension(arg);
2211 }
2212
2213 static long kvm_dev_ioctl(struct file *filp,
2214                           unsigned int ioctl, unsigned long arg)
2215 {
2216         long r = -EINVAL;
2217
2218         switch (ioctl) {
2219         case KVM_GET_API_VERSION:
2220                 r = -EINVAL;
2221                 if (arg)
2222                         goto out;
2223                 r = KVM_API_VERSION;
2224                 break;
2225         case KVM_CREATE_VM:
2226                 r = -EINVAL;
2227                 if (arg)
2228                         goto out;
2229                 r = kvm_dev_ioctl_create_vm();
2230                 break;
2231         case KVM_CHECK_EXTENSION:
2232                 r = kvm_dev_ioctl_check_extension_generic(arg);
2233                 break;
2234         case KVM_GET_VCPU_MMAP_SIZE:
2235                 r = -EINVAL;
2236                 if (arg)
2237                         goto out;
2238                 r = PAGE_SIZE;     /* struct kvm_run */
2239 #ifdef CONFIG_X86
2240                 r += PAGE_SIZE;    /* pio data page */
2241 #endif
2242 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2243                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2244 #endif
2245                 break;
2246         case KVM_TRACE_ENABLE:
2247         case KVM_TRACE_PAUSE:
2248         case KVM_TRACE_DISABLE:
2249                 r = -EOPNOTSUPP;
2250                 break;
2251         default:
2252                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2253         }
2254 out:
2255         return r;
2256 }
2257
2258 static struct file_operations kvm_chardev_ops = {
2259         .unlocked_ioctl = kvm_dev_ioctl,
2260         .compat_ioctl   = kvm_dev_ioctl,
2261         .llseek         = noop_llseek,
2262 };
2263
2264 static struct miscdevice kvm_dev = {
2265         KVM_MINOR,
2266         "kvm",
2267         &kvm_chardev_ops,
2268 };
2269
2270 static void hardware_enable_nolock(void *junk)
2271 {
2272         int cpu = raw_smp_processor_id();
2273         int r;
2274
2275         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2276                 return;
2277
2278         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2279
2280         r = kvm_arch_hardware_enable(NULL);
2281
2282         if (r) {
2283                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2284                 atomic_inc(&hardware_enable_failed);
2285                 printk(KERN_INFO "kvm: enabling virtualization on "
2286                                  "CPU%d failed\n", cpu);
2287         }
2288 }
2289
2290 static void hardware_enable(void *junk)
2291 {
2292         raw_spin_lock(&kvm_lock);
2293         hardware_enable_nolock(junk);
2294         raw_spin_unlock(&kvm_lock);
2295 }
2296
2297 static void hardware_disable_nolock(void *junk)
2298 {
2299         int cpu = raw_smp_processor_id();
2300
2301         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2302                 return;
2303         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2304         kvm_arch_hardware_disable(NULL);
2305 }
2306
2307 static void hardware_disable(void *junk)
2308 {
2309         raw_spin_lock(&kvm_lock);
2310         hardware_disable_nolock(junk);
2311         raw_spin_unlock(&kvm_lock);
2312 }
2313
2314 static void hardware_disable_all_nolock(void)
2315 {
2316         BUG_ON(!kvm_usage_count);
2317
2318         kvm_usage_count--;
2319         if (!kvm_usage_count)
2320                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2321 }
2322
2323 static void hardware_disable_all(void)
2324 {
2325         raw_spin_lock(&kvm_lock);
2326         hardware_disable_all_nolock();
2327         raw_spin_unlock(&kvm_lock);
2328 }
2329
2330 static int hardware_enable_all(void)
2331 {
2332         int r = 0;
2333
2334         raw_spin_lock(&kvm_lock);
2335
2336         kvm_usage_count++;
2337         if (kvm_usage_count == 1) {
2338                 atomic_set(&hardware_enable_failed, 0);
2339                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2340
2341                 if (atomic_read(&hardware_enable_failed)) {
2342                         hardware_disable_all_nolock();
2343                         r = -EBUSY;
2344                 }
2345         }
2346
2347         raw_spin_unlock(&kvm_lock);
2348
2349         return r;
2350 }
2351
2352 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2353                            void *v)
2354 {
2355         int cpu = (long)v;
2356
2357         if (!kvm_usage_count)
2358                 return NOTIFY_OK;
2359
2360         val &= ~CPU_TASKS_FROZEN;
2361         switch (val) {
2362         case CPU_DYING:
2363                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2364                        cpu);
2365                 hardware_disable(NULL);
2366                 break;
2367         case CPU_STARTING:
2368                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2369                        cpu);
2370                 hardware_enable(NULL);
2371                 break;
2372         }
2373         return NOTIFY_OK;
2374 }
2375
2376
2377 asmlinkage void kvm_spurious_fault(void)
2378 {
2379         /* Fault while not rebooting.  We want the trace. */
2380         BUG();
2381 }
2382 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2383
2384 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2385                       void *v)
2386 {
2387         /*
2388          * Some (well, at least mine) BIOSes hang on reboot if
2389          * in vmx root mode.
2390          *
2391          * And Intel TXT required VMX off for all cpu when system shutdown.
2392          */
2393         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2394         kvm_rebooting = true;
2395         on_each_cpu(hardware_disable_nolock, NULL, 1);
2396         return NOTIFY_OK;
2397 }
2398
2399 static struct notifier_block kvm_reboot_notifier = {
2400         .notifier_call = kvm_reboot,
2401         .priority = 0,
2402 };
2403
2404 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2405 {
2406         int i;
2407
2408         for (i = 0; i < bus->dev_count; i++) {
2409                 struct kvm_io_device *pos = bus->range[i].dev;
2410
2411                 kvm_iodevice_destructor(pos);
2412         }
2413         kfree(bus);
2414 }
2415
2416 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2417 {
2418         const struct kvm_io_range *r1 = p1;
2419         const struct kvm_io_range *r2 = p2;
2420
2421         if (r1->addr < r2->addr)
2422                 return -1;
2423         if (r1->addr + r1->len > r2->addr + r2->len)
2424                 return 1;
2425         return 0;
2426 }
2427
2428 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2429                           gpa_t addr, int len)
2430 {
2431         if (bus->dev_count == NR_IOBUS_DEVS)
2432                 return -ENOSPC;
2433
2434         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2435                 .addr = addr,
2436                 .len = len,
2437                 .dev = dev,
2438         };
2439
2440         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2441                 kvm_io_bus_sort_cmp, NULL);
2442
2443         return 0;
2444 }
2445
2446 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2447                              gpa_t addr, int len)
2448 {
2449         struct kvm_io_range *range, key;
2450         int off;
2451
2452         key = (struct kvm_io_range) {
2453                 .addr = addr,
2454                 .len = len,
2455         };
2456
2457         range = bsearch(&key, bus->range, bus->dev_count,
2458                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2459         if (range == NULL)
2460                 return -ENOENT;
2461
2462         off = range - bus->range;
2463
2464         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2465                 off--;
2466
2467         return off;
2468 }
2469
2470 /* kvm_io_bus_write - called under kvm->slots_lock */
2471 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2472                      int len, const void *val)
2473 {
2474         int idx;
2475         struct kvm_io_bus *bus;
2476         struct kvm_io_range range;
2477
2478         range = (struct kvm_io_range) {
2479                 .addr = addr,
2480                 .len = len,
2481         };
2482
2483         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2484         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2485         if (idx < 0)
2486                 return -EOPNOTSUPP;
2487
2488         while (idx < bus->dev_count &&
2489                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2490                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2491                         return 0;
2492                 idx++;
2493         }
2494
2495         return -EOPNOTSUPP;
2496 }
2497
2498 /* kvm_io_bus_read - called under kvm->slots_lock */
2499 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2500                     int len, void *val)
2501 {
2502         int idx;
2503         struct kvm_io_bus *bus;
2504         struct kvm_io_range range;
2505
2506         range = (struct kvm_io_range) {
2507                 .addr = addr,
2508                 .len = len,
2509         };
2510
2511         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2512         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2513         if (idx < 0)
2514                 return -EOPNOTSUPP;
2515
2516         while (idx < bus->dev_count &&
2517                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2518                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2519                         return 0;
2520                 idx++;
2521         }
2522
2523         return -EOPNOTSUPP;
2524 }
2525
2526 /* Caller must hold slots_lock. */
2527 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2528                             int len, struct kvm_io_device *dev)
2529 {
2530         struct kvm_io_bus *new_bus, *bus;
2531
2532         bus = kvm->buses[bus_idx];
2533         if (bus->dev_count > NR_IOBUS_DEVS-1)
2534                 return -ENOSPC;
2535
2536         new_bus = kmemdup(bus, sizeof(struct kvm_io_bus), GFP_KERNEL);
2537         if (!new_bus)
2538                 return -ENOMEM;
2539         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2540         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2541         synchronize_srcu_expedited(&kvm->srcu);
2542         kfree(bus);
2543
2544         return 0;
2545 }
2546
2547 /* Caller must hold slots_lock. */
2548 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2549                               struct kvm_io_device *dev)
2550 {
2551         int i, r;
2552         struct kvm_io_bus *new_bus, *bus;
2553
2554         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2555         if (!new_bus)
2556                 return -ENOMEM;
2557
2558         bus = kvm->buses[bus_idx];
2559         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2560
2561         r = -ENOENT;
2562         for (i = 0; i < new_bus->dev_count; i++)
2563                 if (new_bus->range[i].dev == dev) {
2564                         r = 0;
2565                         new_bus->dev_count--;
2566                         new_bus->range[i] = new_bus->range[new_bus->dev_count];
2567                         sort(new_bus->range, new_bus->dev_count,
2568                              sizeof(struct kvm_io_range),
2569                              kvm_io_bus_sort_cmp, NULL);
2570                         break;
2571                 }
2572
2573         if (r) {
2574                 kfree(new_bus);
2575                 return r;
2576         }
2577
2578         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2579         synchronize_srcu_expedited(&kvm->srcu);
2580         kfree(bus);
2581         return r;
2582 }
2583
2584 static struct notifier_block kvm_cpu_notifier = {
2585         .notifier_call = kvm_cpu_hotplug,
2586 };
2587
2588 static int vm_stat_get(void *_offset, u64 *val)
2589 {
2590         unsigned offset = (long)_offset;
2591         struct kvm *kvm;
2592
2593         *val = 0;
2594         raw_spin_lock(&kvm_lock);
2595         list_for_each_entry(kvm, &vm_list, vm_list)
2596                 *val += *(u32 *)((void *)kvm + offset);
2597         raw_spin_unlock(&kvm_lock);
2598         return 0;
2599 }
2600
2601 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2602
2603 static int vcpu_stat_get(void *_offset, u64 *val)
2604 {
2605         unsigned offset = (long)_offset;
2606         struct kvm *kvm;
2607         struct kvm_vcpu *vcpu;
2608         int i;
2609
2610         *val = 0;
2611         raw_spin_lock(&kvm_lock);
2612         list_for_each_entry(kvm, &vm_list, vm_list)
2613                 kvm_for_each_vcpu(i, vcpu, kvm)
2614                         *val += *(u32 *)((void *)vcpu + offset);
2615
2616         raw_spin_unlock(&kvm_lock);
2617         return 0;
2618 }
2619
2620 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2621
2622 static const struct file_operations *stat_fops[] = {
2623         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2624         [KVM_STAT_VM]   = &vm_stat_fops,
2625 };
2626
2627 static void kvm_init_debug(void)
2628 {
2629         struct kvm_stats_debugfs_item *p;
2630
2631         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2632         for (p = debugfs_entries; p->name; ++p)
2633                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2634                                                 (void *)(long)p->offset,
2635                                                 stat_fops[p->kind]);
2636 }
2637
2638 static void kvm_exit_debug(void)
2639 {
2640         struct kvm_stats_debugfs_item *p;
2641
2642         for (p = debugfs_entries; p->name; ++p)
2643                 debugfs_remove(p->dentry);
2644         debugfs_remove(kvm_debugfs_dir);
2645 }
2646
2647 static int kvm_suspend(void)
2648 {
2649         if (kvm_usage_count)
2650                 hardware_disable_nolock(NULL);
2651         return 0;
2652 }
2653
2654 static void kvm_resume(void)
2655 {
2656         if (kvm_usage_count) {
2657                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2658                 hardware_enable_nolock(NULL);
2659         }
2660 }
2661
2662 static struct syscore_ops kvm_syscore_ops = {
2663         .suspend = kvm_suspend,
2664         .resume = kvm_resume,
2665 };
2666
2667 struct page *bad_page;
2668 pfn_t bad_pfn;
2669
2670 static inline
2671 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2672 {
2673         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2674 }
2675
2676 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2677 {
2678         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2679
2680         kvm_arch_vcpu_load(vcpu, cpu);
2681 }
2682
2683 static void kvm_sched_out(struct preempt_notifier *pn,
2684                           struct task_struct *next)
2685 {
2686         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2687
2688         kvm_arch_vcpu_put(vcpu);
2689 }
2690
2691 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2692                   struct module *module)
2693 {
2694         int r;
2695         int cpu;
2696
2697         r = kvm_arch_init(opaque);
2698         if (r)
2699                 goto out_fail;
2700
2701         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2702
2703         if (bad_page == NULL) {
2704                 r = -ENOMEM;
2705                 goto out;
2706         }
2707
2708         bad_pfn = page_to_pfn(bad_page);
2709
2710         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2711
2712         if (hwpoison_page == NULL) {
2713                 r = -ENOMEM;
2714                 goto out_free_0;
2715         }
2716
2717         hwpoison_pfn = page_to_pfn(hwpoison_page);
2718
2719         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2720
2721         if (fault_page == NULL) {
2722                 r = -ENOMEM;
2723                 goto out_free_0;
2724         }
2725
2726         fault_pfn = page_to_pfn(fault_page);
2727
2728         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2729                 r = -ENOMEM;
2730                 goto out_free_0;
2731         }
2732
2733         r = kvm_arch_hardware_setup();
2734         if (r < 0)
2735                 goto out_free_0a;
2736
2737         for_each_online_cpu(cpu) {
2738                 smp_call_function_single(cpu,
2739                                 kvm_arch_check_processor_compat,
2740                                 &r, 1);
2741                 if (r < 0)
2742                         goto out_free_1;
2743         }
2744
2745         r = register_cpu_notifier(&kvm_cpu_notifier);
2746         if (r)
2747                 goto out_free_2;
2748         register_reboot_notifier(&kvm_reboot_notifier);
2749
2750         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2751         if (!vcpu_align)
2752                 vcpu_align = __alignof__(struct kvm_vcpu);
2753         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2754                                            0, NULL);
2755         if (!kvm_vcpu_cache) {
2756                 r = -ENOMEM;
2757                 goto out_free_3;
2758         }
2759
2760         r = kvm_async_pf_init();
2761         if (r)
2762                 goto out_free;
2763
2764         kvm_chardev_ops.owner = module;
2765         kvm_vm_fops.owner = module;
2766         kvm_vcpu_fops.owner = module;
2767
2768         r = misc_register(&kvm_dev);
2769         if (r) {
2770                 printk(KERN_ERR "kvm: misc device register failed\n");
2771                 goto out_unreg;
2772         }
2773
2774         register_syscore_ops(&kvm_syscore_ops);
2775
2776         kvm_preempt_ops.sched_in = kvm_sched_in;
2777         kvm_preempt_ops.sched_out = kvm_sched_out;
2778
2779         kvm_init_debug();
2780
2781         return 0;
2782
2783 out_unreg:
2784         kvm_async_pf_deinit();
2785 out_free:
2786         kmem_cache_destroy(kvm_vcpu_cache);
2787 out_free_3:
2788         unregister_reboot_notifier(&kvm_reboot_notifier);
2789         unregister_cpu_notifier(&kvm_cpu_notifier);
2790 out_free_2:
2791 out_free_1:
2792         kvm_arch_hardware_unsetup();
2793 out_free_0a:
2794         free_cpumask_var(cpus_hardware_enabled);
2795 out_free_0:
2796         if (fault_page)
2797                 __free_page(fault_page);
2798         if (hwpoison_page)
2799                 __free_page(hwpoison_page);
2800         __free_page(bad_page);
2801 out:
2802         kvm_arch_exit();
2803 out_fail:
2804         return r;
2805 }
2806 EXPORT_SYMBOL_GPL(kvm_init);
2807
2808 void kvm_exit(void)
2809 {
2810         kvm_exit_debug();
2811         misc_deregister(&kvm_dev);
2812         kmem_cache_destroy(kvm_vcpu_cache);
2813         kvm_async_pf_deinit();
2814         unregister_syscore_ops(&kvm_syscore_ops);
2815         unregister_reboot_notifier(&kvm_reboot_notifier);
2816         unregister_cpu_notifier(&kvm_cpu_notifier);
2817         on_each_cpu(hardware_disable_nolock, NULL, 1);
2818         kvm_arch_hardware_unsetup();
2819         kvm_arch_exit();
2820         free_cpumask_var(cpus_hardware_enabled);
2821         __free_page(hwpoison_page);
2822         __free_page(bad_page);
2823 }
2824 EXPORT_SYMBOL_GPL(kvm_exit);