virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/syscore_ops.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/pgtable.h>
  57
  58 #include "coalesced_mmio.h"
  59 #include "async_pf.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/kvm.h>
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 /*
  68  * Ordering of locks:
  69  *
  70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  71  */
  72
  73 DEFINE_RAW_SPINLOCK(kvm_lock);
  74 LIST_HEAD(vm_list);
  75
  76 static cpumask_var_t cpus_hardware_enabled;
  77 static int kvm_usage_count = 0;
  78 static atomic_t hardware_enable_failed;
  79
  80 struct kmem_cache *kvm_vcpu_cache;
  81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  82
  83 static __read_mostly struct preempt_ops kvm_preempt_ops;
  84
  85 struct dentry *kvm_debugfs_dir;
  86
  87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  88                            unsigned long arg);
  89 #ifdef CONFIG_COMPAT
  90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  91                                   unsigned long arg);
  92 #endif
  93 static int hardware_enable_all(void);
  94 static void hardware_disable_all(void);
  95
  96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  97
  98 bool kvm_rebooting;
  99 EXPORT_SYMBOL_GPL(kvm_rebooting);
 100
 101 static bool largepages_enabled = true;
 102
 103 static struct page *hwpoison_page;
 104 static pfn_t hwpoison_pfn;
 105
 106 struct page *fault_page;
 107 pfn_t fault_pfn;
 108
 109 inline int kvm_is_mmio_pfn(pfn_t pfn)
 110 {
 111         if (pfn_valid(pfn)) {
 112                 int reserved;
 113                 struct page *tail = pfn_to_page(pfn);
 114                 struct page *head = compound_trans_head(tail);
 115                 reserved = PageReserved(head);
 116                 if (head != tail) {
 117                         /*
 118                          * "head" is not a dangling pointer
 119                          * (compound_trans_head takes care of that)
 120                          * but the hugepage may have been splitted
 121                          * from under us (and we may not hold a
 122                          * reference count on the head page so it can
 123                          * be reused before we run PageReferenced), so
 124                          * we've to check PageTail before returning
 125                          * what we just read.
 126                          */
 127                         smp_rmb();
 128                         if (PageTail(tail))
 129                                 return reserved;
 130                 }
 131                 return PageReserved(tail);
 132         }
 133
 134         return true;
 135 }
 136
 137 /*
 138  * Switches to specified vcpu, until a matching vcpu_put()
 139  */
 140 void vcpu_load(struct kvm_vcpu *vcpu)
 141 {
 142         int cpu;
 143
 144         mutex_lock(&vcpu->mutex);
 145         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 146                 /* The thread running this VCPU changed. */
 147                 struct pid *oldpid = vcpu->pid;
 148                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 149                 rcu_assign_pointer(vcpu->pid, newpid);
 150                 synchronize_rcu();
 151                 put_pid(oldpid);
 152         }
 153         cpu = get_cpu();
 154         preempt_notifier_register(&vcpu->preempt_notifier);
 155         kvm_arch_vcpu_load(vcpu, cpu);
 156         put_cpu();
 157 }
 158
 159 void vcpu_put(struct kvm_vcpu *vcpu)
 160 {
 161         preempt_disable();
 162         kvm_arch_vcpu_put(vcpu);
 163         preempt_notifier_unregister(&vcpu->preempt_notifier);
 164         preempt_enable();
 165         mutex_unlock(&vcpu->mutex);
 166 }
 167
 168 static void ack_flush(void *_completed)
 169 {
 170 }
 171
 172 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 173 {
 174         int i, cpu, me;
 175         cpumask_var_t cpus;
 176         bool called = true;
 177         struct kvm_vcpu *vcpu;
 178
 179         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 180
 181         me = get_cpu();
 182         kvm_for_each_vcpu(i, vcpu, kvm) {
 183                 kvm_make_request(req, vcpu);
 184                 cpu = vcpu->cpu;
 185
 186                 /* Set ->requests bit before we read ->mode */
 187                 smp_mb();
 188
 189                 if (cpus != NULL && cpu != -1 && cpu != me &&
 190                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 191                         cpumask_set_cpu(cpu, cpus);
 192         }
 193         if (unlikely(cpus == NULL))
 194                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 195         else if (!cpumask_empty(cpus))
 196                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 197         else
 198                 called = false;
 199         put_cpu();
 200         free_cpumask_var(cpus);
 201         return called;
 202 }
 203
 204 void kvm_flush_remote_tlbs(struct kvm *kvm)
 205 {
 206         int dirty_count = kvm->tlbs_dirty;
 207
 208         smp_mb();
 209         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 210                 ++kvm->stat.remote_tlb_flush;
 211         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 212 }
 213
 214 void kvm_reload_remote_mmus(struct kvm *kvm)
 215 {
 216         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 217 }
 218
 219 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 220 {
 221         struct page *page;
 222         int r;
 223
 224         mutex_init(&vcpu->mutex);
 225         vcpu->cpu = -1;
 226         vcpu->kvm = kvm;
 227         vcpu->vcpu_id = id;
 228         vcpu->pid = NULL;
 229         init_waitqueue_head(&vcpu->wq);
 230         kvm_async_pf_vcpu_init(vcpu);
 231
 232         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 233         if (!page) {
 234                 r = -ENOMEM;
 235                 goto fail;
 236         }
 237         vcpu->run = page_address(page);
 238
 239         r = kvm_arch_vcpu_init(vcpu);
 240         if (r < 0)
 241                 goto fail_free_run;
 242         return 0;
 243
 244 fail_free_run:
 245         free_page((unsigned long)vcpu->run);
 246 fail:
 247         return r;
 248 }
 249 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 250
 251 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 252 {
 253         put_pid(vcpu->pid);
 254         kvm_arch_vcpu_uninit(vcpu);
 255         free_page((unsigned long)vcpu->run);
 256 }
 257 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 258
 259 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 260 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 261 {
 262         return container_of(mn, struct kvm, mmu_notifier);
 263 }
 264
 265 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 266                                              struct mm_struct *mm,
 267                                              unsigned long address)
 268 {
 269         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 270         int need_tlb_flush, idx;
 271
 272         /*
 273          * When ->invalidate_page runs, the linux pte has been zapped
 274          * already but the page is still allocated until
 275          * ->invalidate_page returns. So if we increase the sequence
 276          * here the kvm page fault will notice if the spte can't be
 277          * established because the page is going to be freed. If
 278          * instead the kvm page fault establishes the spte before
 279          * ->invalidate_page runs, kvm_unmap_hva will release it
 280          * before returning.
 281          *
 282          * The sequence increase only need to be seen at spin_unlock
 283          * time, and not at spin_lock time.
 284          *
 285          * Increasing the sequence after the spin_unlock would be
 286          * unsafe because the kvm page fault could then establish the
 287          * pte after kvm_unmap_hva returned, without noticing the page
 288          * is going to be freed.
 289          */
 290         idx = srcu_read_lock(&kvm->srcu);
 291         spin_lock(&kvm->mmu_lock);
 292         kvm->mmu_notifier_seq++;
 293         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 294         spin_unlock(&kvm->mmu_lock);
 295         srcu_read_unlock(&kvm->srcu, idx);
 296
 297         /* we've to flush the tlb before the pages can be freed */
 298         if (need_tlb_flush)
 299                 kvm_flush_remote_tlbs(kvm);
 300
 301 }
 302
 303 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 304                                         struct mm_struct *mm,
 305                                         unsigned long address,
 306                                         pte_t pte)
 307 {
 308         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 309         int idx;
 310
 311         idx = srcu_read_lock(&kvm->srcu);
 312         spin_lock(&kvm->mmu_lock);
 313         kvm->mmu_notifier_seq++;
 314         kvm_set_spte_hva(kvm, address, pte);
 315         spin_unlock(&kvm->mmu_lock);
 316         srcu_read_unlock(&kvm->srcu, idx);
 317 }
 318
 319 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 320                                                     struct mm_struct *mm,
 321                                                     unsigned long start,
 322                                                     unsigned long end)
 323 {
 324         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 325         int need_tlb_flush = 0, idx;
 326
 327         idx = srcu_read_lock(&kvm->srcu);
 328         spin_lock(&kvm->mmu_lock);
 329         /*
 330          * The count increase must become visible at unlock time as no
 331          * spte can be established without taking the mmu_lock and
 332          * count is also read inside the mmu_lock critical section.
 333          */
 334         kvm->mmu_notifier_count++;
 335         for (; start < end; start += PAGE_SIZE)
 336                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 337         need_tlb_flush |= kvm->tlbs_dirty;
 338         spin_unlock(&kvm->mmu_lock);
 339         srcu_read_unlock(&kvm->srcu, idx);
 340
 341         /* we've to flush the tlb before the pages can be freed */
 342         if (need_tlb_flush)
 343                 kvm_flush_remote_tlbs(kvm);
 344 }
 345
 346 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 347                                                   struct mm_struct *mm,
 348                                                   unsigned long start,
 349                                                   unsigned long end)
 350 {
 351         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 352
 353         spin_lock(&kvm->mmu_lock);
 354         /*
 355          * This sequence increase will notify the kvm page fault that
 356          * the page that is going to be mapped in the spte could have
 357          * been freed.
 358          */
 359         kvm->mmu_notifier_seq++;
 360         /*
 361          * The above sequence increase must be visible before the
 362          * below count decrease but both values are read by the kvm
 363          * page fault under mmu_lock spinlock so we don't need to add
 364          * a smb_wmb() here in between the two.
 365          */
 366         kvm->mmu_notifier_count--;
 367         spin_unlock(&kvm->mmu_lock);
 368
 369         BUG_ON(kvm->mmu_notifier_count < 0);
 370 }
 371
 372 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 373                                               struct mm_struct *mm,
 374                                               unsigned long address)
 375 {
 376         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 377         int young, idx;
 378
 379         idx = srcu_read_lock(&kvm->srcu);
 380         spin_lock(&kvm->mmu_lock);
 381         young = kvm_age_hva(kvm, address);
 382         spin_unlock(&kvm->mmu_lock);
 383         srcu_read_unlock(&kvm->srcu, idx);
 384
 385         if (young)
 386                 kvm_flush_remote_tlbs(kvm);
 387
 388         return young;
 389 }
 390
 391 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 392                                        struct mm_struct *mm,
 393                                        unsigned long address)
 394 {
 395         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 396         int young, idx;
 397
 398         idx = srcu_read_lock(&kvm->srcu);
 399         spin_lock(&kvm->mmu_lock);
 400         young = kvm_test_age_hva(kvm, address);
 401         spin_unlock(&kvm->mmu_lock);
 402         srcu_read_unlock(&kvm->srcu, idx);
 403
 404         return young;
 405 }
 406
 407 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 408                                      struct mm_struct *mm)
 409 {
 410         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 411         int idx;
 412
 413         idx = srcu_read_lock(&kvm->srcu);
 414         kvm_arch_flush_shadow(kvm);
 415         srcu_read_unlock(&kvm->srcu, idx);
 416 }
 417
 418 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 419         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 420         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 421         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 422         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 423         .test_young             = kvm_mmu_notifier_test_young,
 424         .change_pte             = kvm_mmu_notifier_change_pte,
 425         .release                = kvm_mmu_notifier_release,
 426 };
 427
 428 static int kvm_init_mmu_notifier(struct kvm *kvm)
 429 {
 430         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 431         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 432 }
 433
 434 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 435
 436 static int kvm_init_mmu_notifier(struct kvm *kvm)
 437 {
 438         return 0;
 439 }
 440
 441 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 442
 443 static struct kvm *kvm_create_vm(void)
 444 {
 445         int r, i;
 446         struct kvm *kvm = kvm_arch_alloc_vm();
 447
 448         if (!kvm)
 449                 return ERR_PTR(-ENOMEM);
 450
 451         r = kvm_arch_init_vm(kvm);
 452         if (r)
 453                 goto out_err_nodisable;
 454
 455         r = hardware_enable_all();
 456         if (r)
 457                 goto out_err_nodisable;
 458
 459 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 460         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 461         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 462 #endif
 463
 464         r = -ENOMEM;
 465         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 466         if (!kvm->memslots)
 467                 goto out_err_nosrcu;
 468         if (init_srcu_struct(&kvm->srcu))
 469                 goto out_err_nosrcu;
 470         for (i = 0; i < KVM_NR_BUSES; i++) {
 471                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 472                                         GFP_KERNEL);
 473                 if (!kvm->buses[i])
 474                         goto out_err;
 475         }
 476
 477         spin_lock_init(&kvm->mmu_lock);
 478         kvm->mm = current->mm;
 479         atomic_inc(&kvm->mm->mm_count);
 480         kvm_eventfd_init(kvm);
 481         mutex_init(&kvm->lock);
 482         mutex_init(&kvm->irq_lock);
 483         mutex_init(&kvm->slots_lock);
 484         atomic_set(&kvm->users_count, 1);
 485
 486         r = kvm_init_mmu_notifier(kvm);
 487         if (r)
 488                 goto out_err;
 489
 490         raw_spin_lock(&kvm_lock);
 491         list_add(&kvm->vm_list, &vm_list);
 492         raw_spin_unlock(&kvm_lock);
 493
 494         return kvm;
 495
 496 out_err:
 497         cleanup_srcu_struct(&kvm->srcu);
 498 out_err_nosrcu:
 499         hardware_disable_all();
 500 out_err_nodisable:
 501         for (i = 0; i < KVM_NR_BUSES; i++)
 502                 kfree(kvm->buses[i]);
 503         kfree(kvm->memslots);
 504         kvm_arch_free_vm(kvm);
 505         return ERR_PTR(r);
 506 }
 507
 508 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 509 {
 510         if (!memslot->dirty_bitmap)
 511                 return;
 512
 513         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
 514                 vfree(memslot->dirty_bitmap_head);
 515         else
 516                 kfree(memslot->dirty_bitmap_head);
 517
 518         memslot->dirty_bitmap = NULL;
 519         memslot->dirty_bitmap_head = NULL;
 520 }
 521
 522 /*
 523  * Free any memory in @free but not in @dont.
 524  */
 525 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 526                                   struct kvm_memory_slot *dont)
 527 {
 528         int i;
 529
 530         if (!dont || free->rmap != dont->rmap)
 531                 vfree(free->rmap);
 532
 533         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 534                 kvm_destroy_dirty_bitmap(free);
 535
 536
 537         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 538                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 539                         vfree(free->lpage_info[i]);
 540                         free->lpage_info[i] = NULL;
 541                 }
 542         }
 543
 544         free->npages = 0;
 545         free->rmap = NULL;
 546 }
 547
 548 void kvm_free_physmem(struct kvm *kvm)
 549 {
 550         int i;
 551         struct kvm_memslots *slots = kvm->memslots;
 552
 553         for (i = 0; i < slots->nmemslots; ++i)
 554                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 555
 556         kfree(kvm->memslots);
 557 }
 558
 559 static void kvm_destroy_vm(struct kvm *kvm)
 560 {
 561         int i;
 562         struct mm_struct *mm = kvm->mm;
 563
 564         kvm_arch_sync_events(kvm);
 565         raw_spin_lock(&kvm_lock);
 566         list_del(&kvm->vm_list);
 567         raw_spin_unlock(&kvm_lock);
 568         kvm_free_irq_routing(kvm);
 569         for (i = 0; i < KVM_NR_BUSES; i++)
 570                 kvm_io_bus_destroy(kvm->buses[i]);
 571         kvm_coalesced_mmio_free(kvm);
 572 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 573         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 574 #else
 575         kvm_arch_flush_shadow(kvm);
 576 #endif
 577         kvm_arch_destroy_vm(kvm);
 578         kvm_free_physmem(kvm);
 579         cleanup_srcu_struct(&kvm->srcu);
 580         kvm_arch_free_vm(kvm);
 581         hardware_disable_all();
 582         mmdrop(mm);
 583 }
 584
 585 void kvm_get_kvm(struct kvm *kvm)
 586 {
 587         atomic_inc(&kvm->users_count);
 588 }
 589 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 590
 591 void kvm_put_kvm(struct kvm *kvm)
 592 {
 593         if (atomic_dec_and_test(&kvm->users_count))
 594                 kvm_destroy_vm(kvm);
 595 }
 596 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 597
 598
 599 static int kvm_vm_release(struct inode *inode, struct file *filp)
 600 {
 601         struct kvm *kvm = filp->private_data;
 602
 603         kvm_irqfd_release(kvm);
 604
 605         kvm_put_kvm(kvm);
 606         return 0;
 607 }
 608
 609 #ifndef CONFIG_S390
 610 /*
 611  * Allocation size is twice as large as the actual dirty bitmap size.
 612  * This makes it possible to do double buffering: see x86's
 613  * kvm_vm_ioctl_get_dirty_log().
 614  */
 615 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 616 {
 617         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 618
 619         if (dirty_bytes > PAGE_SIZE)
 620                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
 621         else
 622                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
 623
 624         if (!memslot->dirty_bitmap)
 625                 return -ENOMEM;
 626
 627         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
 628         return 0;
 629 }
 630 #endif /* !CONFIG_S390 */
 631
 632 /*
 633  * Allocate some memory and give it an address in the guest physical address
 634  * space.
 635  *
 636  * Discontiguous memory is allowed, mostly for framebuffers.
 637  *
 638  * Must be called holding mmap_sem for write.
 639  */
 640 int __kvm_set_memory_region(struct kvm *kvm,
 641                             struct kvm_userspace_memory_region *mem,
 642                             int user_alloc)
 643 {
 644         int r;
 645         gfn_t base_gfn;
 646         unsigned long npages;
 647         unsigned long i;
 648         struct kvm_memory_slot *memslot;
 649         struct kvm_memory_slot old, new;
 650         struct kvm_memslots *slots, *old_memslots;
 651
 652         r = -EINVAL;
 653         /* General sanity checks */
 654         if (mem->memory_size & (PAGE_SIZE - 1))
 655                 goto out;
 656         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 657                 goto out;
 658         /* We can read the guest memory with __xxx_user() later on. */
 659         if (user_alloc &&
 660             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 661              !access_ok(VERIFY_WRITE,
 662                         (void __user *)(unsigned long)mem->userspace_addr,
 663                         mem->memory_size)))
 664                 goto out;
 665         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 666                 goto out;
 667         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 668                 goto out;
 669
 670         memslot = &kvm->memslots->memslots[mem->slot];
 671         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 672         npages = mem->memory_size >> PAGE_SHIFT;
 673
 674         r = -EINVAL;
 675         if (npages > KVM_MEM_MAX_NR_PAGES)
 676                 goto out;
 677
 678         if (!npages)
 679                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 680
 681         new = old = *memslot;
 682
 683         new.id = mem->slot;
 684         new.base_gfn = base_gfn;
 685         new.npages = npages;
 686         new.flags = mem->flags;
 687
 688         /* Disallow changing a memory slot's size. */
 689         r = -EINVAL;
 690         if (npages && old.npages && npages != old.npages)
 691                 goto out_free;
 692
 693         /* Check for overlaps */
 694         r = -EEXIST;
 695         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 696                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 697
 698                 if (s == memslot || !s->npages)
 699                         continue;
 700                 if (!((base_gfn + npages <= s->base_gfn) ||
 701                       (base_gfn >= s->base_gfn + s->npages)))
 702                         goto out_free;
 703         }
 704
 705         /* Free page dirty bitmap if unneeded */
 706         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 707                 new.dirty_bitmap = NULL;
 708
 709         r = -ENOMEM;
 710
 711         /* Allocate if a slot is being created */
 712 #ifndef CONFIG_S390
 713         if (npages && !new.rmap) {
 714                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
 715
 716                 if (!new.rmap)
 717                         goto out_free;
 718
 719                 new.user_alloc = user_alloc;
 720                 new.userspace_addr = mem->userspace_addr;
 721         }
 722         if (!npages)
 723                 goto skip_lpage;
 724
 725         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 726                 unsigned long ugfn;
 727                 unsigned long j;
 728                 int lpages;
 729                 int level = i + 2;
 730
 731                 /* Avoid unused variable warning if no large pages */
 732                 (void)level;
 733
 734                 if (new.lpage_info[i])
 735                         continue;
 736
 737                 lpages = 1 + ((base_gfn + npages - 1)
 738                              >> KVM_HPAGE_GFN_SHIFT(level));
 739                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
 740
 741                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
 742
 743                 if (!new.lpage_info[i])
 744                         goto out_free;
 745
 746                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
 747                         new.lpage_info[i][0].write_count = 1;
 748                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
 749                         new.lpage_info[i][lpages - 1].write_count = 1;
 750                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 751                 /*
 752                  * If the gfn and userspace address are not aligned wrt each
 753                  * other, or if explicitly asked to, disable large page
 754                  * support for this slot
 755                  */
 756                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 757                     !largepages_enabled)
 758                         for (j = 0; j < lpages; ++j)
 759                                 new.lpage_info[i][j].write_count = 1;
 760         }
 761
 762 skip_lpage:
 763
 764         /* Allocate page dirty bitmap if needed */
 765         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 766                 if (kvm_create_dirty_bitmap(&new) < 0)
 767                         goto out_free;
 768                 /* destroy any largepage mappings for dirty tracking */
 769         }
 770 #else  /* not defined CONFIG_S390 */
 771         new.user_alloc = user_alloc;
 772         if (user_alloc)
 773                 new.userspace_addr = mem->userspace_addr;
 774 #endif /* not defined CONFIG_S390 */
 775
 776         if (!npages) {
 777                 r = -ENOMEM;
 778                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 779                 if (!slots)
 780                         goto out_free;
 781                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 782                 if (mem->slot >= slots->nmemslots)
 783                         slots->nmemslots = mem->slot + 1;
 784                 slots->generation++;
 785                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 786
 787                 old_memslots = kvm->memslots;
 788                 rcu_assign_pointer(kvm->memslots, slots);
 789                 synchronize_srcu_expedited(&kvm->srcu);
 790                 /* From this point no new shadow pages pointing to a deleted
 791                  * memslot will be created.
 792                  *
 793                  * validation of sp->gfn happens in:
 794                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 795                  *      - kvm_is_visible_gfn (mmu_check_roots)
 796                  */
 797                 kvm_arch_flush_shadow(kvm);
 798                 kfree(old_memslots);
 799         }
 800
 801         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 802         if (r)
 803                 goto out_free;
 804
 805         /* map/unmap the pages in iommu page table */
 806         if (npages) {
 807                 r = kvm_iommu_map_pages(kvm, &new);
 808                 if (r)
 809                         goto out_free;
 810         } else
 811                 kvm_iommu_unmap_pages(kvm, &old);
 812
 813         r = -ENOMEM;
 814         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 815         if (!slots)
 816                 goto out_free;
 817         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 818         if (mem->slot >= slots->nmemslots)
 819                 slots->nmemslots = mem->slot + 1;
 820         slots->generation++;
 821
 822         /* actual memory is freed via old in kvm_free_physmem_slot below */
 823         if (!npages) {
 824                 new.rmap = NULL;
 825                 new.dirty_bitmap = NULL;
 826                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 827                         new.lpage_info[i] = NULL;
 828         }
 829
 830         slots->memslots[mem->slot] = new;
 831         old_memslots = kvm->memslots;
 832         rcu_assign_pointer(kvm->memslots, slots);
 833         synchronize_srcu_expedited(&kvm->srcu);
 834
 835         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 836
 837         /*
 838          * If the new memory slot is created, we need to clear all
 839          * mmio sptes.
 840          */
 841         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 842                 kvm_arch_flush_shadow(kvm);
 843
 844         kvm_free_physmem_slot(&old, &new);
 845         kfree(old_memslots);
 846
 847         return 0;
 848
 849 out_free:
 850         kvm_free_physmem_slot(&new, &old);
 851 out:
 852         return r;
 853
 854 }
 855 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 856
 857 int kvm_set_memory_region(struct kvm *kvm,
 858                           struct kvm_userspace_memory_region *mem,
 859                           int user_alloc)
 860 {
 861         int r;
 862
 863         mutex_lock(&kvm->slots_lock);
 864         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 865         mutex_unlock(&kvm->slots_lock);
 866         return r;
 867 }
 868 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 869
 870 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 871                                    struct
 872                                    kvm_userspace_memory_region *mem,
 873                                    int user_alloc)
 874 {
 875         if (mem->slot >= KVM_MEMORY_SLOTS)
 876                 return -EINVAL;
 877         return kvm_set_memory_region(kvm, mem, user_alloc);
 878 }
 879
 880 int kvm_get_dirty_log(struct kvm *kvm,
 881                         struct kvm_dirty_log *log, int *is_dirty)
 882 {
 883         struct kvm_memory_slot *memslot;
 884         int r, i;
 885         unsigned long n;
 886         unsigned long any = 0;
 887
 888         r = -EINVAL;
 889         if (log->slot >= KVM_MEMORY_SLOTS)
 890                 goto out;
 891
 892         memslot = &kvm->memslots->memslots[log->slot];
 893         r = -ENOENT;
 894         if (!memslot->dirty_bitmap)
 895                 goto out;
 896
 897         n = kvm_dirty_bitmap_bytes(memslot);
 898
 899         for (i = 0; !any && i < n/sizeof(long); ++i)
 900                 any = memslot->dirty_bitmap[i];
 901
 902         r = -EFAULT;
 903         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 904                 goto out;
 905
 906         if (any)
 907                 *is_dirty = 1;
 908
 909         r = 0;
 910 out:
 911         return r;
 912 }
 913
 914 void kvm_disable_largepages(void)
 915 {
 916         largepages_enabled = false;
 917 }
 918 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 919
 920 int is_error_page(struct page *page)
 921 {
 922         return page == bad_page || page == hwpoison_page || page == fault_page;
 923 }
 924 EXPORT_SYMBOL_GPL(is_error_page);
 925
 926 int is_error_pfn(pfn_t pfn)
 927 {
 928         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 929 }
 930 EXPORT_SYMBOL_GPL(is_error_pfn);
 931
 932 int is_hwpoison_pfn(pfn_t pfn)
 933 {
 934         return pfn == hwpoison_pfn;
 935 }
 936 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 937
 938 int is_fault_pfn(pfn_t pfn)
 939 {
 940         return pfn == fault_pfn;
 941 }
 942 EXPORT_SYMBOL_GPL(is_fault_pfn);
 943
 944 int is_noslot_pfn(pfn_t pfn)
 945 {
 946         return pfn == bad_pfn;
 947 }
 948 EXPORT_SYMBOL_GPL(is_noslot_pfn);
 949
 950 int is_invalid_pfn(pfn_t pfn)
 951 {
 952         return pfn == hwpoison_pfn || pfn == fault_pfn;
 953 }
 954 EXPORT_SYMBOL_GPL(is_invalid_pfn);
 955
 956 static inline unsigned long bad_hva(void)
 957 {
 958         return PAGE_OFFSET;
 959 }
 960
 961 int kvm_is_error_hva(unsigned long addr)
 962 {
 963         return addr == bad_hva();
 964 }
 965 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 966
 967 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
 968                                                 gfn_t gfn)
 969 {
 970         int i;
 971
 972         for (i = 0; i < slots->nmemslots; ++i) {
 973                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 974
 975                 if (gfn >= memslot->base_gfn
 976                     && gfn < memslot->base_gfn + memslot->npages)
 977                         return memslot;
 978         }
 979         return NULL;
 980 }
 981
 982 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 983 {
 984         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 985 }
 986 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 987
 988 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 989 {
 990         int i;
 991         struct kvm_memslots *slots = kvm_memslots(kvm);
 992
 993         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 994                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 995
 996                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 997                         continue;
 998
 999                 if (gfn >= memslot->base_gfn
1000                     && gfn < memslot->base_gfn + memslot->npages)
1001                         return 1;
1002         }
1003         return 0;
1004 }
1005 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1006
1007 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1008 {
1009         struct vm_area_struct *vma;
1010         unsigned long addr, size;
1011
1012         size = PAGE_SIZE;
1013
1014         addr = gfn_to_hva(kvm, gfn);
1015         if (kvm_is_error_hva(addr))
1016                 return PAGE_SIZE;
1017
1018         down_read(&current->mm->mmap_sem);
1019         vma = find_vma(current->mm, addr);
1020         if (!vma)
1021                 goto out;
1022
1023         size = vma_kernel_pagesize(vma);
1024
1025 out:
1026         up_read(&current->mm->mmap_sem);
1027
1028         return size;
1029 }
1030
1031 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1032                                      gfn_t *nr_pages)
1033 {
1034         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1035                 return bad_hva();
1036
1037         if (nr_pages)
1038                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1039
1040         return gfn_to_hva_memslot(slot, gfn);
1041 }
1042
1043 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1044 {
1045         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1046 }
1047 EXPORT_SYMBOL_GPL(gfn_to_hva);
1048
1049 static pfn_t get_fault_pfn(void)
1050 {
1051         get_page(fault_page);
1052         return fault_pfn;
1053 }
1054
1055 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1056         unsigned long start, int write, struct page **page)
1057 {
1058         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1059
1060         if (write)
1061                 flags |= FOLL_WRITE;
1062
1063         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1064 }
1065
1066 static inline int check_user_page_hwpoison(unsigned long addr)
1067 {
1068         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1069
1070         rc = __get_user_pages(current, current->mm, addr, 1,
1071                               flags, NULL, NULL, NULL);
1072         return rc == -EHWPOISON;
1073 }
1074
1075 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1076                         bool *async, bool write_fault, bool *writable)
1077 {
1078         struct page *page[1];
1079         int npages = 0;
1080         pfn_t pfn;
1081
1082         /* we can do it either atomically or asynchronously, not both */
1083         BUG_ON(atomic && async);
1084
1085         BUG_ON(!write_fault && !writable);
1086
1087         if (writable)
1088                 *writable = true;
1089
1090         if (atomic || async)
1091                 npages = __get_user_pages_fast(addr, 1, 1, page);
1092
1093         if (unlikely(npages != 1) && !atomic) {
1094                 might_sleep();
1095
1096                 if (writable)
1097                         *writable = write_fault;
1098
1099                 if (async) {
1100                         down_read(&current->mm->mmap_sem);
1101                         npages = get_user_page_nowait(current, current->mm,
1102                                                      addr, write_fault, page);
1103                         up_read(&current->mm->mmap_sem);
1104                 } else
1105                         npages = get_user_pages_fast(addr, 1, write_fault,
1106                                                      page);
1107
1108                 /* map read fault as writable if possible */
1109                 if (unlikely(!write_fault) && npages == 1) {
1110                         struct page *wpage[1];
1111
1112                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1113                         if (npages == 1) {
1114                                 *writable = true;
1115                                 put_page(page[0]);
1116                                 page[0] = wpage[0];
1117                         }
1118                         npages = 1;
1119                 }
1120         }
1121
1122         if (unlikely(npages != 1)) {
1123                 struct vm_area_struct *vma;
1124
1125                 if (atomic)
1126                         return get_fault_pfn();
1127
1128                 down_read(&current->mm->mmap_sem);
1129                 if (npages == -EHWPOISON ||
1130                         (!async && check_user_page_hwpoison(addr))) {
1131                         up_read(&current->mm->mmap_sem);
1132                         get_page(hwpoison_page);
1133                         return page_to_pfn(hwpoison_page);
1134                 }
1135
1136                 vma = find_vma_intersection(current->mm, addr, addr+1);
1137
1138                 if (vma == NULL)
1139                         pfn = get_fault_pfn();
1140                 else if ((vma->vm_flags & VM_PFNMAP)) {
1141                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1142                                 vma->vm_pgoff;
1143                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1144                 } else {
1145                         if (async && (vma->vm_flags & VM_WRITE))
1146                                 *async = true;
1147                         pfn = get_fault_pfn();
1148                 }
1149                 up_read(&current->mm->mmap_sem);
1150         } else
1151                 pfn = page_to_pfn(page[0]);
1152
1153         return pfn;
1154 }
1155
1156 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1157 {
1158         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1159 }
1160 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1161
1162 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1163                           bool write_fault, bool *writable)
1164 {
1165         unsigned long addr;
1166
1167         if (async)
1168                 *async = false;
1169
1170         addr = gfn_to_hva(kvm, gfn);
1171         if (kvm_is_error_hva(addr)) {
1172                 get_page(bad_page);
1173                 return page_to_pfn(bad_page);
1174         }
1175
1176         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1177 }
1178
1179 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1180 {
1181         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1182 }
1183 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1184
1185 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1186                        bool write_fault, bool *writable)
1187 {
1188         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1189 }
1190 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1191
1192 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1193 {
1194         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1195 }
1196 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1197
1198 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1199                       bool *writable)
1200 {
1201         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1202 }
1203 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1204
1205 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1206                          struct kvm_memory_slot *slot, gfn_t gfn)
1207 {
1208         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1209         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1210 }
1211
1212 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1213                                                                   int nr_pages)
1214 {
1215         unsigned long addr;
1216         gfn_t entry;
1217
1218         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1219         if (kvm_is_error_hva(addr))
1220                 return -1;
1221
1222         if (entry < nr_pages)
1223                 return 0;
1224
1225         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1226 }
1227 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1228
1229 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1230 {
1231         pfn_t pfn;
1232
1233         pfn = gfn_to_pfn(kvm, gfn);
1234         if (!kvm_is_mmio_pfn(pfn))
1235                 return pfn_to_page(pfn);
1236
1237         WARN_ON(kvm_is_mmio_pfn(pfn));
1238
1239         get_page(bad_page);
1240         return bad_page;
1241 }
1242
1243 EXPORT_SYMBOL_GPL(gfn_to_page);
1244
1245 void kvm_release_page_clean(struct page *page)
1246 {
1247         kvm_release_pfn_clean(page_to_pfn(page));
1248 }
1249 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1250
1251 void kvm_release_pfn_clean(pfn_t pfn)
1252 {
1253         if (!kvm_is_mmio_pfn(pfn))
1254                 put_page(pfn_to_page(pfn));
1255 }
1256 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1257
1258 void kvm_release_page_dirty(struct page *page)
1259 {
1260         kvm_release_pfn_dirty(page_to_pfn(page));
1261 }
1262 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1263
1264 void kvm_release_pfn_dirty(pfn_t pfn)
1265 {
1266         kvm_set_pfn_dirty(pfn);
1267         kvm_release_pfn_clean(pfn);
1268 }
1269 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1270
1271 void kvm_set_page_dirty(struct page *page)
1272 {
1273         kvm_set_pfn_dirty(page_to_pfn(page));
1274 }
1275 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1276
1277 void kvm_set_pfn_dirty(pfn_t pfn)
1278 {
1279         if (!kvm_is_mmio_pfn(pfn)) {
1280                 struct page *page = pfn_to_page(pfn);
1281                 if (!PageReserved(page))
1282                         SetPageDirty(page);
1283         }
1284 }
1285 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1286
1287 void kvm_set_pfn_accessed(pfn_t pfn)
1288 {
1289         if (!kvm_is_mmio_pfn(pfn))
1290                 mark_page_accessed(pfn_to_page(pfn));
1291 }
1292 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1293
1294 void kvm_get_pfn(pfn_t pfn)
1295 {
1296         if (!kvm_is_mmio_pfn(pfn))
1297                 get_page(pfn_to_page(pfn));
1298 }
1299 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1300
1301 static int next_segment(unsigned long len, int offset)
1302 {
1303         if (len > PAGE_SIZE - offset)
1304                 return PAGE_SIZE - offset;
1305         else
1306                 return len;
1307 }
1308
1309 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1310                         int len)
1311 {
1312         int r;
1313         unsigned long addr;
1314
1315         addr = gfn_to_hva(kvm, gfn);
1316         if (kvm_is_error_hva(addr))
1317                 return -EFAULT;
1318         r = __copy_from_user(data, (void __user *)addr + offset, len);
1319         if (r)
1320                 return -EFAULT;
1321         return 0;
1322 }
1323 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1324
1325 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1326 {
1327         gfn_t gfn = gpa >> PAGE_SHIFT;
1328         int seg;
1329         int offset = offset_in_page(gpa);
1330         int ret;
1331
1332         while ((seg = next_segment(len, offset)) != 0) {
1333                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1334                 if (ret < 0)
1335                         return ret;
1336                 offset = 0;
1337                 len -= seg;
1338                 data += seg;
1339                 ++gfn;
1340         }
1341         return 0;
1342 }
1343 EXPORT_SYMBOL_GPL(kvm_read_guest);
1344
1345 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1346                           unsigned long len)
1347 {
1348         int r;
1349         unsigned long addr;
1350         gfn_t gfn = gpa >> PAGE_SHIFT;
1351         int offset = offset_in_page(gpa);
1352
1353         addr = gfn_to_hva(kvm, gfn);
1354         if (kvm_is_error_hva(addr))
1355                 return -EFAULT;
1356         pagefault_disable();
1357         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1358         pagefault_enable();
1359         if (r)
1360                 return -EFAULT;
1361         return 0;
1362 }
1363 EXPORT_SYMBOL(kvm_read_guest_atomic);
1364
1365 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1366                          int offset, int len)
1367 {
1368         int r;
1369         unsigned long addr;
1370
1371         addr = gfn_to_hva(kvm, gfn);
1372         if (kvm_is_error_hva(addr))
1373                 return -EFAULT;
1374         r = __copy_to_user((void __user *)addr + offset, data, len);
1375         if (r)
1376                 return -EFAULT;
1377         mark_page_dirty(kvm, gfn);
1378         return 0;
1379 }
1380 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1381
1382 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1383                     unsigned long len)
1384 {
1385         gfn_t gfn = gpa >> PAGE_SHIFT;
1386         int seg;
1387         int offset = offset_in_page(gpa);
1388         int ret;
1389
1390         while ((seg = next_segment(len, offset)) != 0) {
1391                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1392                 if (ret < 0)
1393                         return ret;
1394                 offset = 0;
1395                 len -= seg;
1396                 data += seg;
1397                 ++gfn;
1398         }
1399         return 0;
1400 }
1401
1402 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1403                               gpa_t gpa)
1404 {
1405         struct kvm_memslots *slots = kvm_memslots(kvm);
1406         int offset = offset_in_page(gpa);
1407         gfn_t gfn = gpa >> PAGE_SHIFT;
1408
1409         ghc->gpa = gpa;
1410         ghc->generation = slots->generation;
1411         ghc->memslot = __gfn_to_memslot(slots, gfn);
1412         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1413         if (!kvm_is_error_hva(ghc->hva))
1414                 ghc->hva += offset;
1415         else
1416                 return -EFAULT;
1417
1418         return 0;
1419 }
1420 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1421
1422 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1423                            void *data, unsigned long len)
1424 {
1425         struct kvm_memslots *slots = kvm_memslots(kvm);
1426         int r;
1427
1428         if (slots->generation != ghc->generation)
1429                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1430
1431         if (kvm_is_error_hva(ghc->hva))
1432                 return -EFAULT;
1433
1434         r = __copy_to_user((void __user *)ghc->hva, data, len);
1435         if (r)
1436                 return -EFAULT;
1437         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1438
1439         return 0;
1440 }
1441 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1442
1443 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1444                            void *data, unsigned long len)
1445 {
1446         struct kvm_memslots *slots = kvm_memslots(kvm);
1447         int r;
1448
1449         if (slots->generation != ghc->generation)
1450                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1451
1452         if (kvm_is_error_hva(ghc->hva))
1453                 return -EFAULT;
1454
1455         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1456         if (r)
1457                 return -EFAULT;
1458
1459         return 0;
1460 }
1461 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1462
1463 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1464 {
1465         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1466                                     offset, len);
1467 }
1468 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1469
1470 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1471 {
1472         gfn_t gfn = gpa >> PAGE_SHIFT;
1473         int seg;
1474         int offset = offset_in_page(gpa);
1475         int ret;
1476
1477         while ((seg = next_segment(len, offset)) != 0) {
1478                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1479                 if (ret < 0)
1480                         return ret;
1481                 offset = 0;
1482                 len -= seg;
1483                 ++gfn;
1484         }
1485         return 0;
1486 }
1487 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1488
1489 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1490                              gfn_t gfn)
1491 {
1492         if (memslot && memslot->dirty_bitmap) {
1493                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1494
1495                 __set_bit_le(rel_gfn, memslot->dirty_bitmap);
1496         }
1497 }
1498
1499 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1500 {
1501         struct kvm_memory_slot *memslot;
1502
1503         memslot = gfn_to_memslot(kvm, gfn);
1504         mark_page_dirty_in_slot(kvm, memslot, gfn);
1505 }
1506
1507 /*
1508  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1509  */
1510 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1511 {
1512         DEFINE_WAIT(wait);
1513
1514         for (;;) {
1515                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1516
1517                 if (kvm_arch_vcpu_runnable(vcpu)) {
1518                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1519                         break;
1520                 }
1521                 if (kvm_cpu_has_pending_timer(vcpu))
1522                         break;
1523                 if (signal_pending(current))
1524                         break;
1525
1526                 schedule();
1527         }
1528
1529         finish_wait(&vcpu->wq, &wait);
1530 }
1531
1532 void kvm_resched(struct kvm_vcpu *vcpu)
1533 {
1534         if (!need_resched())
1535                 return;
1536         cond_resched();
1537 }
1538 EXPORT_SYMBOL_GPL(kvm_resched);
1539
1540 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1541 {
1542         struct kvm *kvm = me->kvm;
1543         struct kvm_vcpu *vcpu;
1544         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1545         int yielded = 0;
1546         int pass;
1547         int i;
1548
1549         /*
1550          * We boost the priority of a VCPU that is runnable but not
1551          * currently running, because it got preempted by something
1552          * else and called schedule in __vcpu_run.  Hopefully that
1553          * VCPU is holding the lock that we need and will release it.
1554          * We approximate round-robin by starting at the last boosted VCPU.
1555          */
1556         for (pass = 0; pass < 2 && !yielded; pass++) {
1557                 kvm_for_each_vcpu(i, vcpu, kvm) {
1558                         struct task_struct *task = NULL;
1559                         struct pid *pid;
1560                         if (!pass && i < last_boosted_vcpu) {
1561                                 i = last_boosted_vcpu;
1562                                 continue;
1563                         } else if (pass && i > last_boosted_vcpu)
1564                                 break;
1565                         if (vcpu == me)
1566                                 continue;
1567                         if (waitqueue_active(&vcpu->wq))
1568                                 continue;
1569                         rcu_read_lock();
1570                         pid = rcu_dereference(vcpu->pid);
1571                         if (pid)
1572                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1573                         rcu_read_unlock();
1574                         if (!task)
1575                                 continue;
1576                         if (task->flags & PF_VCPU) {
1577                                 put_task_struct(task);
1578                                 continue;
1579                         }
1580                         if (yield_to(task, 1)) {
1581                                 put_task_struct(task);
1582                                 kvm->last_boosted_vcpu = i;
1583                                 yielded = 1;
1584                                 break;
1585                         }
1586                         put_task_struct(task);
1587                 }
1588         }
1589 }
1590 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1591
1592 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1593 {
1594         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1595         struct page *page;
1596
1597         if (vmf->pgoff == 0)
1598                 page = virt_to_page(vcpu->run);
1599 #ifdef CONFIG_X86
1600         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1601                 page = virt_to_page(vcpu->arch.pio_data);
1602 #endif
1603 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1604         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1605                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1606 #endif
1607         else
1608                 return VM_FAULT_SIGBUS;
1609         get_page(page);
1610         vmf->page = page;
1611         return 0;
1612 }
1613
1614 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1615         .fault = kvm_vcpu_fault,
1616 };
1617
1618 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1619 {
1620         vma->vm_ops = &kvm_vcpu_vm_ops;
1621         return 0;
1622 }
1623
1624 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1625 {
1626         struct kvm_vcpu *vcpu = filp->private_data;
1627
1628         kvm_put_kvm(vcpu->kvm);
1629         return 0;
1630 }
1631
1632 static struct file_operations kvm_vcpu_fops = {
1633         .release        = kvm_vcpu_release,
1634         .unlocked_ioctl = kvm_vcpu_ioctl,
1635 #ifdef CONFIG_COMPAT
1636         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1637 #endif
1638         .mmap           = kvm_vcpu_mmap,
1639         .llseek         = noop_llseek,
1640 };
1641
1642 /*
1643  * Allocates an inode for the vcpu.
1644  */
1645 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1646 {
1647         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1648 }
1649
1650 /*
1651  * Creates some virtual cpus.  Good luck creating more than one.
1652  */
1653 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1654 {
1655         int r;
1656         struct kvm_vcpu *vcpu, *v;
1657
1658         vcpu = kvm_arch_vcpu_create(kvm, id);
1659         if (IS_ERR(vcpu))
1660                 return PTR_ERR(vcpu);
1661
1662         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1663
1664         r = kvm_arch_vcpu_setup(vcpu);
1665         if (r)
1666                 goto vcpu_destroy;
1667
1668         mutex_lock(&kvm->lock);
1669         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1670                 r = -EINVAL;
1671                 goto unlock_vcpu_destroy;
1672         }
1673
1674         kvm_for_each_vcpu(r, v, kvm)
1675                 if (v->vcpu_id == id) {
1676                         r = -EEXIST;
1677                         goto unlock_vcpu_destroy;
1678                 }
1679
1680         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1681
1682         /* Now it's all set up, let userspace reach it */
1683         kvm_get_kvm(kvm);
1684         r = create_vcpu_fd(vcpu);
1685         if (r < 0) {
1686                 kvm_put_kvm(kvm);
1687                 goto unlock_vcpu_destroy;
1688         }
1689
1690         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1691         smp_wmb();
1692         atomic_inc(&kvm->online_vcpus);
1693
1694 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1695         if (kvm->bsp_vcpu_id == id)
1696                 kvm->bsp_vcpu = vcpu;
1697 #endif
1698         mutex_unlock(&kvm->lock);
1699         return r;
1700
1701 unlock_vcpu_destroy:
1702         mutex_unlock(&kvm->lock);
1703 vcpu_destroy:
1704         kvm_arch_vcpu_destroy(vcpu);
1705         return r;
1706 }
1707
1708 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1709 {
1710         if (sigset) {
1711                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1712                 vcpu->sigset_active = 1;
1713                 vcpu->sigset = *sigset;
1714         } else
1715                 vcpu->sigset_active = 0;
1716         return 0;
1717 }
1718
1719 static long kvm_vcpu_ioctl(struct file *filp,
1720                            unsigned int ioctl, unsigned long arg)
1721 {
1722         struct kvm_vcpu *vcpu = filp->private_data;
1723         void __user *argp = (void __user *)arg;
1724         int r;
1725         struct kvm_fpu *fpu = NULL;
1726         struct kvm_sregs *kvm_sregs = NULL;
1727
1728         if (vcpu->kvm->mm != current->mm)
1729                 return -EIO;
1730
1731 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1732         /*
1733          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1734          * so vcpu_load() would break it.
1735          */
1736         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1737                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1738 #endif
1739
1740
1741         vcpu_load(vcpu);
1742         switch (ioctl) {
1743         case KVM_RUN:
1744                 r = -EINVAL;
1745                 if (arg)
1746                         goto out;
1747                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1748                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1749                 break;
1750         case KVM_GET_REGS: {
1751                 struct kvm_regs *kvm_regs;
1752
1753                 r = -ENOMEM;
1754                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1755                 if (!kvm_regs)
1756                         goto out;
1757                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1758                 if (r)
1759                         goto out_free1;
1760                 r = -EFAULT;
1761                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1762                         goto out_free1;
1763                 r = 0;
1764 out_free1:
1765                 kfree(kvm_regs);
1766                 break;
1767         }
1768         case KVM_SET_REGS: {
1769                 struct kvm_regs *kvm_regs;
1770
1771                 r = -ENOMEM;
1772                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1773                 if (!kvm_regs)
1774                         goto out;
1775                 r = -EFAULT;
1776                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1777                         goto out_free2;
1778                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1779                 if (r)
1780                         goto out_free2;
1781                 r = 0;
1782 out_free2:
1783                 kfree(kvm_regs);
1784                 break;
1785         }
1786         case KVM_GET_SREGS: {
1787                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1788                 r = -ENOMEM;
1789                 if (!kvm_sregs)
1790                         goto out;
1791                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1792                 if (r)
1793                         goto out;
1794                 r = -EFAULT;
1795                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1796                         goto out;
1797                 r = 0;
1798                 break;
1799         }
1800         case KVM_SET_SREGS: {
1801                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1802                 r = -ENOMEM;
1803                 if (!kvm_sregs)
1804                         goto out;
1805                 r = -EFAULT;
1806                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1807                         goto out;
1808                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1809                 if (r)
1810                         goto out;
1811                 r = 0;
1812                 break;
1813         }
1814         case KVM_GET_MP_STATE: {
1815                 struct kvm_mp_state mp_state;
1816
1817                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1818                 if (r)
1819                         goto out;
1820                 r = -EFAULT;
1821                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1822                         goto out;
1823                 r = 0;
1824                 break;
1825         }
1826         case KVM_SET_MP_STATE: {
1827                 struct kvm_mp_state mp_state;
1828
1829                 r = -EFAULT;
1830                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1831                         goto out;
1832                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1833                 if (r)
1834                         goto out;
1835                 r = 0;
1836                 break;
1837         }
1838         case KVM_TRANSLATE: {
1839                 struct kvm_translation tr;
1840
1841                 r = -EFAULT;
1842                 if (copy_from_user(&tr, argp, sizeof tr))
1843                         goto out;
1844                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1845                 if (r)
1846                         goto out;
1847                 r = -EFAULT;
1848                 if (copy_to_user(argp, &tr, sizeof tr))
1849                         goto out;
1850                 r = 0;
1851                 break;
1852         }
1853         case KVM_SET_GUEST_DEBUG: {
1854                 struct kvm_guest_debug dbg;
1855
1856                 r = -EFAULT;
1857                 if (copy_from_user(&dbg, argp, sizeof dbg))
1858                         goto out;
1859                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1860                 if (r)
1861                         goto out;
1862                 r = 0;
1863                 break;
1864         }
1865         case KVM_SET_SIGNAL_MASK: {
1866                 struct kvm_signal_mask __user *sigmask_arg = argp;
1867                 struct kvm_signal_mask kvm_sigmask;
1868                 sigset_t sigset, *p;
1869
1870                 p = NULL;
1871                 if (argp) {
1872                         r = -EFAULT;
1873                         if (copy_from_user(&kvm_sigmask, argp,
1874                                            sizeof kvm_sigmask))
1875                                 goto out;
1876                         r = -EINVAL;
1877                         if (kvm_sigmask.len != sizeof sigset)
1878                                 goto out;
1879                         r = -EFAULT;
1880                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1881                                            sizeof sigset))
1882                                 goto out;
1883                         p = &sigset;
1884                 }
1885                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1886                 break;
1887         }
1888         case KVM_GET_FPU: {
1889                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1890                 r = -ENOMEM;
1891                 if (!fpu)
1892                         goto out;
1893                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1894                 if (r)
1895                         goto out;
1896                 r = -EFAULT;
1897                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1898                         goto out;
1899                 r = 0;
1900                 break;
1901         }
1902         case KVM_SET_FPU: {
1903                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1904                 r = -ENOMEM;
1905                 if (!fpu)
1906                         goto out;
1907                 r = -EFAULT;
1908                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1909                         goto out;
1910                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1911                 if (r)
1912                         goto out;
1913                 r = 0;
1914                 break;
1915         }
1916         default:
1917                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1918         }
1919 out:
1920         vcpu_put(vcpu);
1921         kfree(fpu);
1922         kfree(kvm_sregs);
1923         return r;
1924 }
1925
1926 #ifdef CONFIG_COMPAT
1927 static long kvm_vcpu_compat_ioctl(struct file *filp,
1928                                   unsigned int ioctl, unsigned long arg)
1929 {
1930         struct kvm_vcpu *vcpu = filp->private_data;
1931         void __user *argp = compat_ptr(arg);
1932         int r;
1933
1934         if (vcpu->kvm->mm != current->mm)
1935                 return -EIO;
1936
1937         switch (ioctl) {
1938         case KVM_SET_SIGNAL_MASK: {
1939                 struct kvm_signal_mask __user *sigmask_arg = argp;
1940                 struct kvm_signal_mask kvm_sigmask;
1941                 compat_sigset_t csigset;
1942                 sigset_t sigset;
1943
1944                 if (argp) {
1945                         r = -EFAULT;
1946                         if (copy_from_user(&kvm_sigmask, argp,
1947                                            sizeof kvm_sigmask))
1948                                 goto out;
1949                         r = -EINVAL;
1950                         if (kvm_sigmask.len != sizeof csigset)
1951                                 goto out;
1952                         r = -EFAULT;
1953                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1954                                            sizeof csigset))
1955                                 goto out;
1956                 }
1957                 sigset_from_compat(&sigset, &csigset);
1958                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1959                 break;
1960         }
1961         default:
1962                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1963         }
1964
1965 out:
1966         return r;
1967 }
1968 #endif
1969
1970 static long kvm_vm_ioctl(struct file *filp,
1971                            unsigned int ioctl, unsigned long arg)
1972 {
1973         struct kvm *kvm = filp->private_data;
1974         void __user *argp = (void __user *)arg;
1975         int r;
1976
1977         if (kvm->mm != current->mm)
1978                 return -EIO;
1979         switch (ioctl) {
1980         case KVM_CREATE_VCPU:
1981                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1982                 if (r < 0)
1983                         goto out;
1984                 break;
1985         case KVM_SET_USER_MEMORY_REGION: {
1986                 struct kvm_userspace_memory_region kvm_userspace_mem;
1987
1988                 r = -EFAULT;
1989                 if (copy_from_user(&kvm_userspace_mem, argp,
1990                                                 sizeof kvm_userspace_mem))
1991                         goto out;
1992
1993                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1994                 if (r)
1995                         goto out;
1996                 break;
1997         }
1998         case KVM_GET_DIRTY_LOG: {
1999                 struct kvm_dirty_log log;
2000
2001                 r = -EFAULT;
2002                 if (copy_from_user(&log, argp, sizeof log))
2003                         goto out;
2004                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2005                 if (r)
2006                         goto out;
2007                 break;
2008         }
2009 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2010         case KVM_REGISTER_COALESCED_MMIO: {
2011                 struct kvm_coalesced_mmio_zone zone;
2012                 r = -EFAULT;
2013                 if (copy_from_user(&zone, argp, sizeof zone))
2014                         goto out;
2015                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2016                 if (r)
2017                         goto out;
2018                 r = 0;
2019                 break;
2020         }
2021         case KVM_UNREGISTER_COALESCED_MMIO: {
2022                 struct kvm_coalesced_mmio_zone zone;
2023                 r = -EFAULT;
2024                 if (copy_from_user(&zone, argp, sizeof zone))
2025                         goto out;
2026                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2027                 if (r)
2028                         goto out;
2029                 r = 0;
2030                 break;
2031         }
2032 #endif
2033         case KVM_IRQFD: {
2034                 struct kvm_irqfd data;
2035
2036                 r = -EFAULT;
2037                 if (copy_from_user(&data, argp, sizeof data))
2038                         goto out;
2039                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2040                 break;
2041         }
2042         case KVM_IOEVENTFD: {
2043                 struct kvm_ioeventfd data;
2044
2045                 r = -EFAULT;
2046                 if (copy_from_user(&data, argp, sizeof data))
2047                         goto out;
2048                 r = kvm_ioeventfd(kvm, &data);
2049                 break;
2050         }
2051 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2052         case KVM_SET_BOOT_CPU_ID:
2053                 r = 0;
2054                 mutex_lock(&kvm->lock);
2055                 if (atomic_read(&kvm->online_vcpus) != 0)
2056                         r = -EBUSY;
2057                 else
2058                         kvm->bsp_vcpu_id = arg;
2059                 mutex_unlock(&kvm->lock);
2060                 break;
2061 #endif
2062         default:
2063                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2064                 if (r == -ENOTTY)
2065                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2066         }
2067 out:
2068         return r;
2069 }
2070
2071 #ifdef CONFIG_COMPAT
2072 struct compat_kvm_dirty_log {
2073         __u32 slot;
2074         __u32 padding1;
2075         union {
2076                 compat_uptr_t dirty_bitmap; /* one bit per page */
2077                 __u64 padding2;
2078         };
2079 };
2080
2081 static long kvm_vm_compat_ioctl(struct file *filp,
2082                            unsigned int ioctl, unsigned long arg)
2083 {
2084         struct kvm *kvm = filp->private_data;
2085         int r;
2086
2087         if (kvm->mm != current->mm)
2088                 return -EIO;
2089         switch (ioctl) {
2090         case KVM_GET_DIRTY_LOG: {
2091                 struct compat_kvm_dirty_log compat_log;
2092                 struct kvm_dirty_log log;
2093
2094                 r = -EFAULT;
2095                 if (copy_from_user(&compat_log, (void __user *)arg,
2096                                    sizeof(compat_log)))
2097                         goto out;
2098                 log.slot         = compat_log.slot;
2099                 log.padding1     = compat_log.padding1;
2100                 log.padding2     = compat_log.padding2;
2101                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2102
2103                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2104                 if (r)
2105                         goto out;
2106                 break;
2107         }
2108         default:
2109                 r = kvm_vm_ioctl(filp, ioctl, arg);
2110         }
2111
2112 out:
2113         return r;
2114 }
2115 #endif
2116
2117 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2118 {
2119         struct page *page[1];
2120         unsigned long addr;
2121         int npages;
2122         gfn_t gfn = vmf->pgoff;
2123         struct kvm *kvm = vma->vm_file->private_data;
2124
2125         addr = gfn_to_hva(kvm, gfn);
2126         if (kvm_is_error_hva(addr))
2127                 return VM_FAULT_SIGBUS;
2128
2129         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2130                                 NULL);
2131         if (unlikely(npages != 1))
2132                 return VM_FAULT_SIGBUS;
2133
2134         vmf->page = page[0];
2135         return 0;
2136 }
2137
2138 static const struct vm_operations_struct kvm_vm_vm_ops = {
2139         .fault = kvm_vm_fault,
2140 };
2141
2142 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2143 {
2144         vma->vm_ops = &kvm_vm_vm_ops;
2145         return 0;
2146 }
2147
2148 static struct file_operations kvm_vm_fops = {
2149         .release        = kvm_vm_release,
2150         .unlocked_ioctl = kvm_vm_ioctl,
2151 #ifdef CONFIG_COMPAT
2152         .compat_ioctl   = kvm_vm_compat_ioctl,
2153 #endif
2154         .mmap           = kvm_vm_mmap,
2155         .llseek         = noop_llseek,
2156 };
2157
2158 static int kvm_dev_ioctl_create_vm(void)
2159 {
2160         int r;
2161         struct kvm *kvm;
2162
2163         kvm = kvm_create_vm();
2164         if (IS_ERR(kvm))
2165                 return PTR_ERR(kvm);
2166 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2167         r = kvm_coalesced_mmio_init(kvm);
2168         if (r < 0) {
2169                 kvm_put_kvm(kvm);
2170                 return r;
2171         }
2172 #endif
2173         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2174         if (r < 0)
2175                 kvm_put_kvm(kvm);
2176
2177         return r;
2178 }
2179
2180 static long kvm_dev_ioctl_check_extension_generic(long arg)
2181 {
2182         switch (arg) {
2183         case KVM_CAP_USER_MEMORY:
2184         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2185         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2186 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2187         case KVM_CAP_SET_BOOT_CPU_ID:
2188 #endif
2189         case KVM_CAP_INTERNAL_ERROR_DATA:
2190                 return 1;
2191 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2192         case KVM_CAP_IRQ_ROUTING:
2193                 return KVM_MAX_IRQ_ROUTES;
2194 #endif
2195         default:
2196                 break;
2197         }
2198         return kvm_dev_ioctl_check_extension(arg);
2199 }
2200
2201 static long kvm_dev_ioctl(struct file *filp,
2202                           unsigned int ioctl, unsigned long arg)
2203 {
2204         long r = -EINVAL;
2205
2206         switch (ioctl) {
2207         case KVM_GET_API_VERSION:
2208                 r = -EINVAL;
2209                 if (arg)
2210                         goto out;
2211                 r = KVM_API_VERSION;
2212                 break;
2213         case KVM_CREATE_VM:
2214                 r = -EINVAL;
2215                 if (arg)
2216                         goto out;
2217                 r = kvm_dev_ioctl_create_vm();
2218                 break;
2219         case KVM_CHECK_EXTENSION:
2220                 r = kvm_dev_ioctl_check_extension_generic(arg);
2221                 break;
2222         case KVM_GET_VCPU_MMAP_SIZE:
2223                 r = -EINVAL;
2224                 if (arg)
2225                         goto out;
2226                 r = PAGE_SIZE;     /* struct kvm_run */
2227 #ifdef CONFIG_X86
2228                 r += PAGE_SIZE;    /* pio data page */
2229 #endif
2230 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2231                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2232 #endif
2233                 break;
2234         case KVM_TRACE_ENABLE:
2235         case KVM_TRACE_PAUSE:
2236         case KVM_TRACE_DISABLE:
2237                 r = -EOPNOTSUPP;
2238                 break;
2239         default:
2240                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2241         }
2242 out:
2243         return r;
2244 }
2245
2246 static struct file_operations kvm_chardev_ops = {
2247         .unlocked_ioctl = kvm_dev_ioctl,
2248         .compat_ioctl   = kvm_dev_ioctl,
2249         .llseek         = noop_llseek,
2250 };
2251
2252 static struct miscdevice kvm_dev = {
2253         KVM_MINOR,
2254         "kvm",
2255         &kvm_chardev_ops,
2256 };
2257
2258 static void hardware_enable_nolock(void *junk)
2259 {
2260         int cpu = raw_smp_processor_id();
2261         int r;
2262
2263         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2264                 return;
2265
2266         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2267
2268         r = kvm_arch_hardware_enable(NULL);
2269
2270         if (r) {
2271                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2272                 atomic_inc(&hardware_enable_failed);
2273                 printk(KERN_INFO "kvm: enabling virtualization on "
2274                                  "CPU%d failed\n", cpu);
2275         }
2276 }
2277
2278 static void hardware_enable(void *junk)
2279 {
2280         raw_spin_lock(&kvm_lock);
2281         hardware_enable_nolock(junk);
2282         raw_spin_unlock(&kvm_lock);
2283 }
2284
2285 static void hardware_disable_nolock(void *junk)
2286 {
2287         int cpu = raw_smp_processor_id();
2288
2289         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2290                 return;
2291         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2292         kvm_arch_hardware_disable(NULL);
2293 }
2294
2295 static void hardware_disable(void *junk)
2296 {
2297         raw_spin_lock(&kvm_lock);
2298         hardware_disable_nolock(junk);
2299         raw_spin_unlock(&kvm_lock);
2300 }
2301
2302 static void hardware_disable_all_nolock(void)
2303 {
2304         BUG_ON(!kvm_usage_count);
2305
2306         kvm_usage_count--;
2307         if (!kvm_usage_count)
2308                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2309 }
2310
2311 static void hardware_disable_all(void)
2312 {
2313         raw_spin_lock(&kvm_lock);
2314         hardware_disable_all_nolock();
2315         raw_spin_unlock(&kvm_lock);
2316 }
2317
2318 static int hardware_enable_all(void)
2319 {
2320         int r = 0;
2321
2322         raw_spin_lock(&kvm_lock);
2323
2324         kvm_usage_count++;
2325         if (kvm_usage_count == 1) {
2326                 atomic_set(&hardware_enable_failed, 0);
2327                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2328
2329                 if (atomic_read(&hardware_enable_failed)) {
2330                         hardware_disable_all_nolock();
2331                         r = -EBUSY;
2332                 }
2333         }
2334
2335         raw_spin_unlock(&kvm_lock);
2336
2337         return r;
2338 }
2339
2340 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2341                            void *v)
2342 {
2343         int cpu = (long)v;
2344
2345         if (!kvm_usage_count)
2346                 return NOTIFY_OK;
2347
2348         val &= ~CPU_TASKS_FROZEN;
2349         switch (val) {
2350         case CPU_DYING:
2351                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2352                        cpu);
2353                 hardware_disable(NULL);
2354                 break;
2355         case CPU_STARTING:
2356                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2357                        cpu);
2358                 hardware_enable(NULL);
2359                 break;
2360         }
2361         return NOTIFY_OK;
2362 }
2363
2364
2365 asmlinkage void kvm_spurious_fault(void)
2366 {
2367         /* Fault while not rebooting.  We want the trace. */
2368         BUG();
2369 }
2370 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2371
2372 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2373                       void *v)
2374 {
2375         /*
2376          * Some (well, at least mine) BIOSes hang on reboot if
2377          * in vmx root mode.
2378          *
2379          * And Intel TXT required VMX off for all cpu when system shutdown.
2380          */
2381         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2382         kvm_rebooting = true;
2383         on_each_cpu(hardware_disable_nolock, NULL, 1);
2384         return NOTIFY_OK;
2385 }
2386
2387 static struct notifier_block kvm_reboot_notifier = {
2388         .notifier_call = kvm_reboot,
2389         .priority = 0,
2390 };
2391
2392 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2393 {
2394         int i;
2395
2396         for (i = 0; i < bus->dev_count; i++) {
2397                 struct kvm_io_device *pos = bus->range[i].dev;
2398
2399                 kvm_iodevice_destructor(pos);
2400         }
2401         kfree(bus);
2402 }
2403
2404 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2405 {
2406         const struct kvm_io_range *r1 = p1;
2407         const struct kvm_io_range *r2 = p2;
2408
2409         if (r1->addr < r2->addr)
2410                 return -1;
2411         if (r1->addr + r1->len > r2->addr + r2->len)
2412                 return 1;
2413         return 0;
2414 }
2415
2416 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2417                           gpa_t addr, int len)
2418 {
2419         if (bus->dev_count == NR_IOBUS_DEVS)
2420                 return -ENOSPC;
2421
2422         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2423                 .addr = addr,
2424                 .len = len,
2425                 .dev = dev,
2426         };
2427
2428         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2429                 kvm_io_bus_sort_cmp, NULL);
2430
2431         return 0;
2432 }
2433
2434 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2435                              gpa_t addr, int len)
2436 {
2437         struct kvm_io_range *range, key;
2438         int off;
2439
2440         key = (struct kvm_io_range) {
2441                 .addr = addr,
2442                 .len = len,
2443         };
2444
2445         range = bsearch(&key, bus->range, bus->dev_count,
2446                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2447         if (range == NULL)
2448                 return -ENOENT;
2449
2450         off = range - bus->range;
2451
2452         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2453                 off--;
2454
2455         return off;
2456 }
2457
2458 /* kvm_io_bus_write - called under kvm->slots_lock */
2459 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2460                      int len, const void *val)
2461 {
2462         int idx;
2463         struct kvm_io_bus *bus;
2464         struct kvm_io_range range;
2465
2466         range = (struct kvm_io_range) {
2467                 .addr = addr,
2468                 .len = len,
2469         };
2470
2471         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2472         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2473         if (idx < 0)
2474                 return -EOPNOTSUPP;
2475
2476         while (idx < bus->dev_count &&
2477                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2478                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2479                         return 0;
2480                 idx++;
2481         }
2482
2483         return -EOPNOTSUPP;
2484 }
2485
2486 /* kvm_io_bus_read - called under kvm->slots_lock */
2487 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2488                     int len, void *val)
2489 {
2490         int idx;
2491         struct kvm_io_bus *bus;
2492         struct kvm_io_range range;
2493
2494         range = (struct kvm_io_range) {
2495                 .addr = addr,
2496                 .len = len,
2497         };
2498
2499         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2500         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2501         if (idx < 0)
2502                 return -EOPNOTSUPP;
2503
2504         while (idx < bus->dev_count &&
2505                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2506                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2507                         return 0;
2508                 idx++;
2509         }
2510
2511         return -EOPNOTSUPP;
2512 }
2513
2514 /* Caller must hold slots_lock. */
2515 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2516                             int len, struct kvm_io_device *dev)
2517 {
2518         struct kvm_io_bus *new_bus, *bus;
2519
2520         bus = kvm->buses[bus_idx];
2521         if (bus->dev_count > NR_IOBUS_DEVS-1)
2522                 return -ENOSPC;
2523
2524         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2525         if (!new_bus)
2526                 return -ENOMEM;
2527         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2528         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2529         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2530         synchronize_srcu_expedited(&kvm->srcu);
2531         kfree(bus);
2532
2533         return 0;
2534 }
2535
2536 /* Caller must hold slots_lock. */
2537 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2538                               struct kvm_io_device *dev)
2539 {
2540         int i, r;
2541         struct kvm_io_bus *new_bus, *bus;
2542
2543         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2544         if (!new_bus)
2545                 return -ENOMEM;
2546
2547         bus = kvm->buses[bus_idx];
2548         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2549
2550         r = -ENOENT;
2551         for (i = 0; i < new_bus->dev_count; i++)
2552                 if (new_bus->range[i].dev == dev) {
2553                         r = 0;
2554                         new_bus->dev_count--;
2555                         new_bus->range[i] = new_bus->range[new_bus->dev_count];
2556                         sort(new_bus->range, new_bus->dev_count,
2557                              sizeof(struct kvm_io_range),
2558                              kvm_io_bus_sort_cmp, NULL);
2559                         break;
2560                 }
2561
2562         if (r) {
2563                 kfree(new_bus);
2564                 return r;
2565         }
2566
2567         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2568         synchronize_srcu_expedited(&kvm->srcu);
2569         kfree(bus);
2570         return r;
2571 }
2572
2573 static struct notifier_block kvm_cpu_notifier = {
2574         .notifier_call = kvm_cpu_hotplug,
2575 };
2576
2577 static int vm_stat_get(void *_offset, u64 *val)
2578 {
2579         unsigned offset = (long)_offset;
2580         struct kvm *kvm;
2581
2582         *val = 0;
2583         raw_spin_lock(&kvm_lock);
2584         list_for_each_entry(kvm, &vm_list, vm_list)
2585                 *val += *(u32 *)((void *)kvm + offset);
2586         raw_spin_unlock(&kvm_lock);
2587         return 0;
2588 }
2589
2590 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2591
2592 static int vcpu_stat_get(void *_offset, u64 *val)
2593 {
2594         unsigned offset = (long)_offset;
2595         struct kvm *kvm;
2596         struct kvm_vcpu *vcpu;
2597         int i;
2598
2599         *val = 0;
2600         raw_spin_lock(&kvm_lock);
2601         list_for_each_entry(kvm, &vm_list, vm_list)
2602                 kvm_for_each_vcpu(i, vcpu, kvm)
2603                         *val += *(u32 *)((void *)vcpu + offset);
2604
2605         raw_spin_unlock(&kvm_lock);
2606         return 0;
2607 }
2608
2609 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2610
2611 static const struct file_operations *stat_fops[] = {
2612         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2613         [KVM_STAT_VM]   = &vm_stat_fops,
2614 };
2615
2616 static void kvm_init_debug(void)
2617 {
2618         struct kvm_stats_debugfs_item *p;
2619
2620         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2621         for (p = debugfs_entries; p->name; ++p)
2622                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2623                                                 (void *)(long)p->offset,
2624                                                 stat_fops[p->kind]);
2625 }
2626
2627 static void kvm_exit_debug(void)
2628 {
2629         struct kvm_stats_debugfs_item *p;
2630
2631         for (p = debugfs_entries; p->name; ++p)
2632                 debugfs_remove(p->dentry);
2633         debugfs_remove(kvm_debugfs_dir);
2634 }
2635
2636 static int kvm_suspend(void)
2637 {
2638         if (kvm_usage_count)
2639                 hardware_disable_nolock(NULL);
2640         return 0;
2641 }
2642
2643 static void kvm_resume(void)
2644 {
2645         if (kvm_usage_count) {
2646                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2647                 hardware_enable_nolock(NULL);
2648         }
2649 }
2650
2651 static struct syscore_ops kvm_syscore_ops = {
2652         .suspend = kvm_suspend,
2653         .resume = kvm_resume,
2654 };
2655
2656 struct page *bad_page;
2657 pfn_t bad_pfn;
2658
2659 static inline
2660 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2661 {
2662         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2663 }
2664
2665 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2666 {
2667         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2668
2669         kvm_arch_vcpu_load(vcpu, cpu);
2670 }
2671
2672 static void kvm_sched_out(struct preempt_notifier *pn,
2673                           struct task_struct *next)
2674 {
2675         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2676
2677         kvm_arch_vcpu_put(vcpu);
2678 }
2679
2680 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2681                   struct module *module)
2682 {
2683         int r;
2684         int cpu;
2685
2686         r = kvm_arch_init(opaque);
2687         if (r)
2688                 goto out_fail;
2689
2690         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2691
2692         if (bad_page == NULL) {
2693                 r = -ENOMEM;
2694                 goto out;
2695         }
2696
2697         bad_pfn = page_to_pfn(bad_page);
2698
2699         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2700
2701         if (hwpoison_page == NULL) {
2702                 r = -ENOMEM;
2703                 goto out_free_0;
2704         }
2705
2706         hwpoison_pfn = page_to_pfn(hwpoison_page);
2707
2708         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2709
2710         if (fault_page == NULL) {
2711                 r = -ENOMEM;
2712                 goto out_free_0;
2713         }
2714
2715         fault_pfn = page_to_pfn(fault_page);
2716
2717         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2718                 r = -ENOMEM;
2719                 goto out_free_0;
2720         }
2721
2722         r = kvm_arch_hardware_setup();
2723         if (r < 0)
2724                 goto out_free_0a;
2725
2726         for_each_online_cpu(cpu) {
2727                 smp_call_function_single(cpu,
2728                                 kvm_arch_check_processor_compat,
2729                                 &r, 1);
2730                 if (r < 0)
2731                         goto out_free_1;
2732         }
2733
2734         r = register_cpu_notifier(&kvm_cpu_notifier);
2735         if (r)
2736                 goto out_free_2;
2737         register_reboot_notifier(&kvm_reboot_notifier);
2738
2739         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2740         if (!vcpu_align)
2741                 vcpu_align = __alignof__(struct kvm_vcpu);
2742         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2743                                            0, NULL);
2744         if (!kvm_vcpu_cache) {
2745                 r = -ENOMEM;
2746                 goto out_free_3;
2747         }
2748
2749         r = kvm_async_pf_init();
2750         if (r)
2751                 goto out_free;
2752
2753         kvm_chardev_ops.owner = module;
2754         kvm_vm_fops.owner = module;
2755         kvm_vcpu_fops.owner = module;
2756
2757         r = misc_register(&kvm_dev);
2758         if (r) {
2759                 printk(KERN_ERR "kvm: misc device register failed\n");
2760                 goto out_unreg;
2761         }
2762
2763         register_syscore_ops(&kvm_syscore_ops);
2764
2765         kvm_preempt_ops.sched_in = kvm_sched_in;
2766         kvm_preempt_ops.sched_out = kvm_sched_out;
2767
2768         kvm_init_debug();
2769
2770         return 0;
2771
2772 out_unreg:
2773         kvm_async_pf_deinit();
2774 out_free:
2775         kmem_cache_destroy(kvm_vcpu_cache);
2776 out_free_3:
2777         unregister_reboot_notifier(&kvm_reboot_notifier);
2778         unregister_cpu_notifier(&kvm_cpu_notifier);
2779 out_free_2:
2780 out_free_1:
2781         kvm_arch_hardware_unsetup();
2782 out_free_0a:
2783         free_cpumask_var(cpus_hardware_enabled);
2784 out_free_0:
2785         if (fault_page)
2786                 __free_page(fault_page);
2787         if (hwpoison_page)
2788                 __free_page(hwpoison_page);
2789         __free_page(bad_page);
2790 out:
2791         kvm_arch_exit();
2792 out_fail:
2793         return r;
2794 }
2795 EXPORT_SYMBOL_GPL(kvm_init);
2796
2797 void kvm_exit(void)
2798 {
2799         kvm_exit_debug();
2800         misc_deregister(&kvm_dev);
2801         kmem_cache_destroy(kvm_vcpu_cache);
2802         kvm_async_pf_deinit();
2803         unregister_syscore_ops(&kvm_syscore_ops);
2804         unregister_reboot_notifier(&kvm_reboot_notifier);
2805         unregister_cpu_notifier(&kvm_cpu_notifier);
2806         on_each_cpu(hardware_disable_nolock, NULL, 1);
2807         kvm_arch_hardware_unsetup();
2808         kvm_arch_exit();
2809         free_cpumask_var(cpus_hardware_enabled);
2810         __free_page(hwpoison_page);
2811         __free_page(bad_page);
2812 }
2813 EXPORT_SYMBOL_GPL(kvm_exit);