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         smp_wmb();
 361         /*
 362          * The above sequence increase must be visible before the
 363          * below count decrease, which is ensured by the smp_wmb above
 364          * in conjunction with the smp_rmb in mmu_notifier_retry().
 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 void kvm_init_memslots_id(struct kvm *kvm)
 444 {
 445         int i;
 446         struct kvm_memslots *slots = kvm->memslots;
 447
 448         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 449                 slots->id_to_index[i] = slots->memslots[i].id = i;
 450 }
 451
 452 static struct kvm *kvm_create_vm(unsigned long type)
 453 {
 454         int r, i;
 455         struct kvm *kvm = kvm_arch_alloc_vm();
 456
 457         if (!kvm)
 458                 return ERR_PTR(-ENOMEM);
 459
 460         r = kvm_arch_init_vm(kvm, type);
 461         if (r)
 462                 goto out_err_nodisable;
 463
 464         r = hardware_enable_all();
 465         if (r)
 466                 goto out_err_nodisable;
 467
 468 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 469         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 470         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 471 #endif
 472
 473         r = -ENOMEM;
 474         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 475         if (!kvm->memslots)
 476                 goto out_err_nosrcu;
 477         kvm_init_memslots_id(kvm);
 478         if (init_srcu_struct(&kvm->srcu))
 479                 goto out_err_nosrcu;
 480         for (i = 0; i < KVM_NR_BUSES; i++) {
 481                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 482                                         GFP_KERNEL);
 483                 if (!kvm->buses[i])
 484                         goto out_err;
 485         }
 486
 487         spin_lock_init(&kvm->mmu_lock);
 488         kvm->mm = current->mm;
 489         atomic_inc(&kvm->mm->mm_count);
 490         kvm_eventfd_init(kvm);
 491         mutex_init(&kvm->lock);
 492         mutex_init(&kvm->irq_lock);
 493         mutex_init(&kvm->slots_lock);
 494         atomic_set(&kvm->users_count, 1);
 495
 496         r = kvm_init_mmu_notifier(kvm);
 497         if (r)
 498                 goto out_err;
 499
 500         raw_spin_lock(&kvm_lock);
 501         list_add(&kvm->vm_list, &vm_list);
 502         raw_spin_unlock(&kvm_lock);
 503
 504         return kvm;
 505
 506 out_err:
 507         cleanup_srcu_struct(&kvm->srcu);
 508 out_err_nosrcu:
 509         hardware_disable_all();
 510 out_err_nodisable:
 511         for (i = 0; i < KVM_NR_BUSES; i++)
 512                 kfree(kvm->buses[i]);
 513         kfree(kvm->memslots);
 514         kvm_arch_free_vm(kvm);
 515         return ERR_PTR(r);
 516 }
 517
 518 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 519 {
 520         if (!memslot->dirty_bitmap)
 521                 return;
 522
 523         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
 524                 vfree(memslot->dirty_bitmap_head);
 525         else
 526                 kfree(memslot->dirty_bitmap_head);
 527
 528         memslot->dirty_bitmap = NULL;
 529         memslot->dirty_bitmap_head = NULL;
 530 }
 531
 532 /*
 533  * Free any memory in @free but not in @dont.
 534  */
 535 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 536                                   struct kvm_memory_slot *dont)
 537 {
 538         int i;
 539
 540         if (!dont || free->rmap != dont->rmap)
 541                 vfree(free->rmap);
 542
 543         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 544                 kvm_destroy_dirty_bitmap(free);
 545
 546
 547         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 548                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 549                         vfree(free->lpage_info[i]);
 550                         free->lpage_info[i] = NULL;
 551                 }
 552         }
 553
 554         free->npages = 0;
 555         free->rmap = NULL;
 556 }
 557
 558 void kvm_free_physmem(struct kvm *kvm)
 559 {
 560         struct kvm_memslots *slots = kvm->memslots;
 561         struct kvm_memory_slot *memslot;
 562
 563         kvm_for_each_memslot(memslot, slots)
 564                 kvm_free_physmem_slot(memslot, NULL);
 565
 566         kfree(kvm->memslots);
 567 }
 568
 569 static void kvm_destroy_vm(struct kvm *kvm)
 570 {
 571         int i;
 572         struct mm_struct *mm = kvm->mm;
 573
 574         kvm_arch_sync_events(kvm);
 575         raw_spin_lock(&kvm_lock);
 576         list_del(&kvm->vm_list);
 577         raw_spin_unlock(&kvm_lock);
 578         kvm_free_irq_routing(kvm);
 579         for (i = 0; i < KVM_NR_BUSES; i++)
 580                 kvm_io_bus_destroy(kvm->buses[i]);
 581         kvm_coalesced_mmio_free(kvm);
 582 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 583         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 584 #else
 585         kvm_arch_flush_shadow(kvm);
 586 #endif
 587         kvm_arch_destroy_vm(kvm);
 588         kvm_free_physmem(kvm);
 589         cleanup_srcu_struct(&kvm->srcu);
 590         kvm_arch_free_vm(kvm);
 591         hardware_disable_all();
 592         mmdrop(mm);
 593 }
 594
 595 void kvm_get_kvm(struct kvm *kvm)
 596 {
 597         atomic_inc(&kvm->users_count);
 598 }
 599 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 600
 601 void kvm_put_kvm(struct kvm *kvm)
 602 {
 603         if (atomic_dec_and_test(&kvm->users_count))
 604                 kvm_destroy_vm(kvm);
 605 }
 606 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 607
 608
 609 static int kvm_vm_release(struct inode *inode, struct file *filp)
 610 {
 611         struct kvm *kvm = filp->private_data;
 612
 613         kvm_irqfd_release(kvm);
 614
 615         kvm_put_kvm(kvm);
 616         return 0;
 617 }
 618
 619 #ifndef CONFIG_S390
 620 /*
 621  * Allocation size is twice as large as the actual dirty bitmap size.
 622  * This makes it possible to do double buffering: see x86's
 623  * kvm_vm_ioctl_get_dirty_log().
 624  */
 625 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 626 {
 627         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 628
 629         if (dirty_bytes > PAGE_SIZE)
 630                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
 631         else
 632                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
 633
 634         if (!memslot->dirty_bitmap)
 635                 return -ENOMEM;
 636
 637         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
 638         memslot->nr_dirty_pages = 0;
 639         return 0;
 640 }
 641 #endif /* !CONFIG_S390 */
 642
 643 static int cmp_memslot(const void *slot1, const void *slot2)
 644 {
 645         struct kvm_memory_slot *s1, *s2;
 646
 647         s1 = (struct kvm_memory_slot *)slot1;
 648         s2 = (struct kvm_memory_slot *)slot2;
 649
 650         if (s1->npages < s2->npages)
 651                 return 1;
 652         if (s1->npages > s2->npages)
 653                 return -1;
 654
 655         return 0;
 656 }
 657
 658 /*
 659  * Sort the memslots base on its size, so the larger slots
 660  * will get better fit.
 661  */
 662 static void sort_memslots(struct kvm_memslots *slots)
 663 {
 664         int i;
 665
 666         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 667               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 668
 669         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 670                 slots->id_to_index[slots->memslots[i].id] = i;
 671 }
 672
 673 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
 674 {
 675         if (new) {
 676                 int id = new->id;
 677                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 678                 unsigned long npages = old->npages;
 679
 680                 *old = *new;
 681                 if (new->npages != npages)
 682                         sort_memslots(slots);
 683         }
 684
 685         slots->generation++;
 686 }
 687
 688 /*
 689  * Allocate some memory and give it an address in the guest physical address
 690  * space.
 691  *
 692  * Discontiguous memory is allowed, mostly for framebuffers.
 693  *
 694  * Must be called holding mmap_sem for write.
 695  */
 696 int __kvm_set_memory_region(struct kvm *kvm,
 697                             struct kvm_userspace_memory_region *mem,
 698                             int user_alloc)
 699 {
 700         int r;
 701         gfn_t base_gfn;
 702         unsigned long npages;
 703         unsigned long i;
 704         struct kvm_memory_slot *memslot;
 705         struct kvm_memory_slot old, new;
 706         struct kvm_memslots *slots, *old_memslots;
 707
 708         r = -EINVAL;
 709         /* General sanity checks */
 710         if (mem->memory_size & (PAGE_SIZE - 1))
 711                 goto out;
 712         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 713                 goto out;
 714         /* We can read the guest memory with __xxx_user() later on. */
 715         if (user_alloc &&
 716             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 717              !access_ok(VERIFY_WRITE,
 718                         (void __user *)(unsigned long)mem->userspace_addr,
 719                         mem->memory_size)))
 720                 goto out;
 721         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 722                 goto out;
 723         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 724                 goto out;
 725
 726         memslot = id_to_memslot(kvm->memslots, mem->slot);
 727         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 728         npages = mem->memory_size >> PAGE_SHIFT;
 729
 730         r = -EINVAL;
 731         if (npages > KVM_MEM_MAX_NR_PAGES)
 732                 goto out;
 733
 734         if (!npages)
 735                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 736
 737         new = old = *memslot;
 738
 739         new.id = mem->slot;
 740         new.base_gfn = base_gfn;
 741         new.npages = npages;
 742         new.flags = mem->flags;
 743
 744         /* Disallow changing a memory slot's size. */
 745         r = -EINVAL;
 746         if (npages && old.npages && npages != old.npages)
 747                 goto out_free;
 748
 749         /* Check for overlaps */
 750         r = -EEXIST;
 751         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 752                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 753
 754                 if (s == memslot || !s->npages)
 755                         continue;
 756                 if (!((base_gfn + npages <= s->base_gfn) ||
 757                       (base_gfn >= s->base_gfn + s->npages)))
 758                         goto out_free;
 759         }
 760
 761         /* Free page dirty bitmap if unneeded */
 762         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 763                 new.dirty_bitmap = NULL;
 764
 765         r = -ENOMEM;
 766
 767         /* Allocate if a slot is being created */
 768 #ifndef CONFIG_S390
 769         if (npages && !new.rmap) {
 770                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
 771
 772                 if (!new.rmap)
 773                         goto out_free;
 774
 775                 new.user_alloc = user_alloc;
 776                 new.userspace_addr = mem->userspace_addr;
 777         }
 778         if (!npages)
 779                 goto skip_lpage;
 780
 781         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 782                 unsigned long ugfn;
 783                 unsigned long j;
 784                 int lpages;
 785                 int level = i + 2;
 786
 787                 if (new.lpage_info[i])
 788                         continue;
 789
 790                 lpages = gfn_to_index(base_gfn + npages - 1, base_gfn, level) + 1;
 791
 792                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
 793
 794                 if (!new.lpage_info[i])
 795                         goto out_free;
 796
 797                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
 798                         new.lpage_info[i][0].write_count = 1;
 799                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
 800                         new.lpage_info[i][lpages - 1].write_count = 1;
 801                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 802                 /*
 803                  * If the gfn and userspace address are not aligned wrt each
 804                  * other, or if explicitly asked to, disable large page
 805                  * support for this slot
 806                  */
 807                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 808                     !largepages_enabled)
 809                         for (j = 0; j < lpages; ++j)
 810                                 new.lpage_info[i][j].write_count = 1;
 811         }
 812
 813 skip_lpage:
 814
 815         /* Allocate page dirty bitmap if needed */
 816         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 817                 if (kvm_create_dirty_bitmap(&new) < 0)
 818                         goto out_free;
 819                 /* destroy any largepage mappings for dirty tracking */
 820         }
 821 #else  /* not defined CONFIG_S390 */
 822         new.user_alloc = user_alloc;
 823         if (user_alloc)
 824                 new.userspace_addr = mem->userspace_addr;
 825 #endif /* not defined CONFIG_S390 */
 826
 827         if (!npages) {
 828                 struct kvm_memory_slot *slot;
 829
 830                 r = -ENOMEM;
 831                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 832                                 GFP_KERNEL);
 833                 if (!slots)
 834                         goto out_free;
 835                 slot = id_to_memslot(slots, mem->slot);
 836                 slot->flags |= KVM_MEMSLOT_INVALID;
 837
 838                 update_memslots(slots, NULL);
 839
 840                 old_memslots = kvm->memslots;
 841                 rcu_assign_pointer(kvm->memslots, slots);
 842                 synchronize_srcu_expedited(&kvm->srcu);
 843                 /* From this point no new shadow pages pointing to a deleted
 844                  * memslot will be created.
 845                  *
 846                  * validation of sp->gfn happens in:
 847                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 848                  *      - kvm_is_visible_gfn (mmu_check_roots)
 849                  */
 850                 kvm_arch_flush_shadow(kvm);
 851                 kfree(old_memslots);
 852         }
 853
 854         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 855         if (r)
 856                 goto out_free;
 857
 858         /* map the pages in iommu page table */
 859         if (npages) {
 860                 r = kvm_iommu_map_pages(kvm, &new);
 861                 if (r)
 862                         goto out_free;
 863         }
 864
 865         r = -ENOMEM;
 866         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 867                         GFP_KERNEL);
 868         if (!slots)
 869                 goto out_free;
 870
 871         /* actual memory is freed via old in kvm_free_physmem_slot below */
 872         if (!npages) {
 873                 new.rmap = NULL;
 874                 new.dirty_bitmap = NULL;
 875                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 876                         new.lpage_info[i] = NULL;
 877         }
 878
 879         update_memslots(slots, &new);
 880         old_memslots = kvm->memslots;
 881         rcu_assign_pointer(kvm->memslots, slots);
 882         synchronize_srcu_expedited(&kvm->srcu);
 883
 884         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 885
 886         /*
 887          * If the new memory slot is created, we need to clear all
 888          * mmio sptes.
 889          */
 890         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 891                 kvm_arch_flush_shadow(kvm);
 892
 893         kvm_free_physmem_slot(&old, &new);
 894         kfree(old_memslots);
 895
 896         return 0;
 897
 898 out_free:
 899         kvm_free_physmem_slot(&new, &old);
 900 out:
 901         return r;
 902
 903 }
 904 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 905
 906 int kvm_set_memory_region(struct kvm *kvm,
 907                           struct kvm_userspace_memory_region *mem,
 908                           int user_alloc)
 909 {
 910         int r;
 911
 912         mutex_lock(&kvm->slots_lock);
 913         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 914         mutex_unlock(&kvm->slots_lock);
 915         return r;
 916 }
 917 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 918
 919 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 920                                    struct
 921                                    kvm_userspace_memory_region *mem,
 922                                    int user_alloc)
 923 {
 924         if (mem->slot >= KVM_MEMORY_SLOTS)
 925                 return -EINVAL;
 926         return kvm_set_memory_region(kvm, mem, user_alloc);
 927 }
 928
 929 int kvm_get_dirty_log(struct kvm *kvm,
 930                         struct kvm_dirty_log *log, int *is_dirty)
 931 {
 932         struct kvm_memory_slot *memslot;
 933         int r, i;
 934         unsigned long n;
 935         unsigned long any = 0;
 936
 937         r = -EINVAL;
 938         if (log->slot >= KVM_MEMORY_SLOTS)
 939                 goto out;
 940
 941         memslot = id_to_memslot(kvm->memslots, log->slot);
 942         r = -ENOENT;
 943         if (!memslot->dirty_bitmap)
 944                 goto out;
 945
 946         n = kvm_dirty_bitmap_bytes(memslot);
 947
 948         for (i = 0; !any && i < n/sizeof(long); ++i)
 949                 any = memslot->dirty_bitmap[i];
 950
 951         r = -EFAULT;
 952         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 953                 goto out;
 954
 955         if (any)
 956                 *is_dirty = 1;
 957
 958         r = 0;
 959 out:
 960         return r;
 961 }
 962
 963 void kvm_disable_largepages(void)
 964 {
 965         largepages_enabled = false;
 966 }
 967 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 968
 969 int is_error_page(struct page *page)
 970 {
 971         return page == bad_page || page == hwpoison_page || page == fault_page;
 972 }
 973 EXPORT_SYMBOL_GPL(is_error_page);
 974
 975 int is_error_pfn(pfn_t pfn)
 976 {
 977         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 978 }
 979 EXPORT_SYMBOL_GPL(is_error_pfn);
 980
 981 int is_hwpoison_pfn(pfn_t pfn)
 982 {
 983         return pfn == hwpoison_pfn;
 984 }
 985 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 986
 987 int is_fault_pfn(pfn_t pfn)
 988 {
 989         return pfn == fault_pfn;
 990 }
 991 EXPORT_SYMBOL_GPL(is_fault_pfn);
 992
 993 int is_noslot_pfn(pfn_t pfn)
 994 {
 995         return pfn == bad_pfn;
 996 }
 997 EXPORT_SYMBOL_GPL(is_noslot_pfn);
 998
 999 int is_invalid_pfn(pfn_t pfn)
1000 {
1001         return pfn == hwpoison_pfn || pfn == fault_pfn;
1002 }
1003 EXPORT_SYMBOL_GPL(is_invalid_pfn);
1004
1005 static inline unsigned long bad_hva(void)
1006 {
1007         return PAGE_OFFSET;
1008 }
1009
1010 int kvm_is_error_hva(unsigned long addr)
1011 {
1012         return addr == bad_hva();
1013 }
1014 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1015
1016 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1017 {
1018         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1019 }
1020 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1021
1022 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1023 {
1024         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1025
1026         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
1027               memslot->flags & KVM_MEMSLOT_INVALID)
1028                 return 0;
1029
1030         return 1;
1031 }
1032 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1033
1034 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1035 {
1036         struct vm_area_struct *vma;
1037         unsigned long addr, size;
1038
1039         size = PAGE_SIZE;
1040
1041         addr = gfn_to_hva(kvm, gfn);
1042         if (kvm_is_error_hva(addr))
1043                 return PAGE_SIZE;
1044
1045         down_read(&current->mm->mmap_sem);
1046         vma = find_vma(current->mm, addr);
1047         if (!vma)
1048                 goto out;
1049
1050         size = vma_kernel_pagesize(vma);
1051
1052 out:
1053         up_read(&current->mm->mmap_sem);
1054
1055         return size;
1056 }
1057
1058 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1059                                      gfn_t *nr_pages)
1060 {
1061         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1062                 return bad_hva();
1063
1064         if (nr_pages)
1065                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1066
1067         return gfn_to_hva_memslot(slot, gfn);
1068 }
1069
1070 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1071 {
1072         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1073 }
1074 EXPORT_SYMBOL_GPL(gfn_to_hva);
1075
1076 static pfn_t get_fault_pfn(void)
1077 {
1078         get_page(fault_page);
1079         return fault_pfn;
1080 }
1081
1082 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1083         unsigned long start, int write, struct page **page)
1084 {
1085         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1086
1087         if (write)
1088                 flags |= FOLL_WRITE;
1089
1090         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1091 }
1092
1093 static inline int check_user_page_hwpoison(unsigned long addr)
1094 {
1095         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1096
1097         rc = __get_user_pages(current, current->mm, addr, 1,
1098                               flags, NULL, NULL, NULL);
1099         return rc == -EHWPOISON;
1100 }
1101
1102 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1103                         bool *async, bool write_fault, bool *writable)
1104 {
1105         struct page *page[1];
1106         int npages = 0;
1107         pfn_t pfn;
1108
1109         /* we can do it either atomically or asynchronously, not both */
1110         BUG_ON(atomic && async);
1111
1112         BUG_ON(!write_fault && !writable);
1113
1114         if (writable)
1115                 *writable = true;
1116
1117         if (atomic || async)
1118                 npages = __get_user_pages_fast(addr, 1, 1, page);
1119
1120         if (unlikely(npages != 1) && !atomic) {
1121                 might_sleep();
1122
1123                 if (writable)
1124                         *writable = write_fault;
1125
1126                 if (async) {
1127                         down_read(&current->mm->mmap_sem);
1128                         npages = get_user_page_nowait(current, current->mm,
1129                                                      addr, write_fault, page);
1130                         up_read(&current->mm->mmap_sem);
1131                 } else
1132                         npages = get_user_pages_fast(addr, 1, write_fault,
1133                                                      page);
1134
1135                 /* map read fault as writable if possible */
1136                 if (unlikely(!write_fault) && npages == 1) {
1137                         struct page *wpage[1];
1138
1139                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1140                         if (npages == 1) {
1141                                 *writable = true;
1142                                 put_page(page[0]);
1143                                 page[0] = wpage[0];
1144                         }
1145                         npages = 1;
1146                 }
1147         }
1148
1149         if (unlikely(npages != 1)) {
1150                 struct vm_area_struct *vma;
1151
1152                 if (atomic)
1153                         return get_fault_pfn();
1154
1155                 down_read(&current->mm->mmap_sem);
1156                 if (npages == -EHWPOISON ||
1157                         (!async && check_user_page_hwpoison(addr))) {
1158                         up_read(&current->mm->mmap_sem);
1159                         get_page(hwpoison_page);
1160                         return page_to_pfn(hwpoison_page);
1161                 }
1162
1163                 vma = find_vma_intersection(current->mm, addr, addr+1);
1164
1165                 if (vma == NULL)
1166                         pfn = get_fault_pfn();
1167                 else if ((vma->vm_flags & VM_PFNMAP)) {
1168                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1169                                 vma->vm_pgoff;
1170                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1171                 } else {
1172                         if (async && (vma->vm_flags & VM_WRITE))
1173                                 *async = true;
1174                         pfn = get_fault_pfn();
1175                 }
1176                 up_read(&current->mm->mmap_sem);
1177         } else
1178                 pfn = page_to_pfn(page[0]);
1179
1180         return pfn;
1181 }
1182
1183 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1184 {
1185         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1186 }
1187 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1188
1189 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1190                           bool write_fault, bool *writable)
1191 {
1192         unsigned long addr;
1193
1194         if (async)
1195                 *async = false;
1196
1197         addr = gfn_to_hva(kvm, gfn);
1198         if (kvm_is_error_hva(addr)) {
1199                 get_page(bad_page);
1200                 return page_to_pfn(bad_page);
1201         }
1202
1203         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1204 }
1205
1206 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1207 {
1208         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1209 }
1210 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1211
1212 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1213                        bool write_fault, bool *writable)
1214 {
1215         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1216 }
1217 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1218
1219 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1220 {
1221         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1222 }
1223 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1224
1225 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1226                       bool *writable)
1227 {
1228         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1229 }
1230 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1231
1232 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1233                          struct kvm_memory_slot *slot, gfn_t gfn)
1234 {
1235         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1236         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1237 }
1238
1239 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1240                                                                   int nr_pages)
1241 {
1242         unsigned long addr;
1243         gfn_t entry;
1244
1245         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1246         if (kvm_is_error_hva(addr))
1247                 return -1;
1248
1249         if (entry < nr_pages)
1250                 return 0;
1251
1252         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1253 }
1254 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1255
1256 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1257 {
1258         pfn_t pfn;
1259
1260         pfn = gfn_to_pfn(kvm, gfn);
1261         if (!kvm_is_mmio_pfn(pfn))
1262                 return pfn_to_page(pfn);
1263
1264         WARN_ON(kvm_is_mmio_pfn(pfn));
1265
1266         get_page(bad_page);
1267         return bad_page;
1268 }
1269
1270 EXPORT_SYMBOL_GPL(gfn_to_page);
1271
1272 void kvm_release_page_clean(struct page *page)
1273 {
1274         kvm_release_pfn_clean(page_to_pfn(page));
1275 }
1276 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1277
1278 void kvm_release_pfn_clean(pfn_t pfn)
1279 {
1280         if (!kvm_is_mmio_pfn(pfn))
1281                 put_page(pfn_to_page(pfn));
1282 }
1283 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1284
1285 void kvm_release_page_dirty(struct page *page)
1286 {
1287         kvm_release_pfn_dirty(page_to_pfn(page));
1288 }
1289 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1290
1291 void kvm_release_pfn_dirty(pfn_t pfn)
1292 {
1293         kvm_set_pfn_dirty(pfn);
1294         kvm_release_pfn_clean(pfn);
1295 }
1296 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1297
1298 void kvm_set_page_dirty(struct page *page)
1299 {
1300         kvm_set_pfn_dirty(page_to_pfn(page));
1301 }
1302 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1303
1304 void kvm_set_pfn_dirty(pfn_t pfn)
1305 {
1306         if (!kvm_is_mmio_pfn(pfn)) {
1307                 struct page *page = pfn_to_page(pfn);
1308                 if (!PageReserved(page))
1309                         SetPageDirty(page);
1310         }
1311 }
1312 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1313
1314 void kvm_set_pfn_accessed(pfn_t pfn)
1315 {
1316         if (!kvm_is_mmio_pfn(pfn))
1317                 mark_page_accessed(pfn_to_page(pfn));
1318 }
1319 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1320
1321 void kvm_get_pfn(pfn_t pfn)
1322 {
1323         if (!kvm_is_mmio_pfn(pfn))
1324                 get_page(pfn_to_page(pfn));
1325 }
1326 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1327
1328 static int next_segment(unsigned long len, int offset)
1329 {
1330         if (len > PAGE_SIZE - offset)
1331                 return PAGE_SIZE - offset;
1332         else
1333                 return len;
1334 }
1335
1336 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1337                         int len)
1338 {
1339         int r;
1340         unsigned long addr;
1341
1342         addr = gfn_to_hva(kvm, gfn);
1343         if (kvm_is_error_hva(addr))
1344                 return -EFAULT;
1345         r = __copy_from_user(data, (void __user *)addr + offset, len);
1346         if (r)
1347                 return -EFAULT;
1348         return 0;
1349 }
1350 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1351
1352 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1353 {
1354         gfn_t gfn = gpa >> PAGE_SHIFT;
1355         int seg;
1356         int offset = offset_in_page(gpa);
1357         int ret;
1358
1359         while ((seg = next_segment(len, offset)) != 0) {
1360                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1361                 if (ret < 0)
1362                         return ret;
1363                 offset = 0;
1364                 len -= seg;
1365                 data += seg;
1366                 ++gfn;
1367         }
1368         return 0;
1369 }
1370 EXPORT_SYMBOL_GPL(kvm_read_guest);
1371
1372 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1373                           unsigned long len)
1374 {
1375         int r;
1376         unsigned long addr;
1377         gfn_t gfn = gpa >> PAGE_SHIFT;
1378         int offset = offset_in_page(gpa);
1379
1380         addr = gfn_to_hva(kvm, gfn);
1381         if (kvm_is_error_hva(addr))
1382                 return -EFAULT;
1383         pagefault_disable();
1384         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1385         pagefault_enable();
1386         if (r)
1387                 return -EFAULT;
1388         return 0;
1389 }
1390 EXPORT_SYMBOL(kvm_read_guest_atomic);
1391
1392 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1393                          int offset, int len)
1394 {
1395         int r;
1396         unsigned long addr;
1397
1398         addr = gfn_to_hva(kvm, gfn);
1399         if (kvm_is_error_hva(addr))
1400                 return -EFAULT;
1401         r = __copy_to_user((void __user *)addr + offset, data, len);
1402         if (r)
1403                 return -EFAULT;
1404         mark_page_dirty(kvm, gfn);
1405         return 0;
1406 }
1407 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1408
1409 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1410                     unsigned long len)
1411 {
1412         gfn_t gfn = gpa >> PAGE_SHIFT;
1413         int seg;
1414         int offset = offset_in_page(gpa);
1415         int ret;
1416
1417         while ((seg = next_segment(len, offset)) != 0) {
1418                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1419                 if (ret < 0)
1420                         return ret;
1421                 offset = 0;
1422                 len -= seg;
1423                 data += seg;
1424                 ++gfn;
1425         }
1426         return 0;
1427 }
1428
1429 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1430                               gpa_t gpa)
1431 {
1432         struct kvm_memslots *slots = kvm_memslots(kvm);
1433         int offset = offset_in_page(gpa);
1434         gfn_t gfn = gpa >> PAGE_SHIFT;
1435
1436         ghc->gpa = gpa;
1437         ghc->generation = slots->generation;
1438         ghc->memslot = gfn_to_memslot(kvm, gfn);
1439         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1440         if (!kvm_is_error_hva(ghc->hva))
1441                 ghc->hva += offset;
1442         else
1443                 return -EFAULT;
1444
1445         return 0;
1446 }
1447 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1448
1449 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1450                            void *data, unsigned long len)
1451 {
1452         struct kvm_memslots *slots = kvm_memslots(kvm);
1453         int r;
1454
1455         if (slots->generation != ghc->generation)
1456                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1457
1458         if (kvm_is_error_hva(ghc->hva))
1459                 return -EFAULT;
1460
1461         r = __copy_to_user((void __user *)ghc->hva, data, len);
1462         if (r)
1463                 return -EFAULT;
1464         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1465
1466         return 0;
1467 }
1468 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1469
1470 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1471                            void *data, unsigned long len)
1472 {
1473         struct kvm_memslots *slots = kvm_memslots(kvm);
1474         int r;
1475
1476         if (slots->generation != ghc->generation)
1477                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1478
1479         if (kvm_is_error_hva(ghc->hva))
1480                 return -EFAULT;
1481
1482         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1483         if (r)
1484                 return -EFAULT;
1485
1486         return 0;
1487 }
1488 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1489
1490 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1491 {
1492         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1493                                     offset, len);
1494 }
1495 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1496
1497 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1498 {
1499         gfn_t gfn = gpa >> PAGE_SHIFT;
1500         int seg;
1501         int offset = offset_in_page(gpa);
1502         int ret;
1503
1504         while ((seg = next_segment(len, offset)) != 0) {
1505                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1506                 if (ret < 0)
1507                         return ret;
1508                 offset = 0;
1509                 len -= seg;
1510                 ++gfn;
1511         }
1512         return 0;
1513 }
1514 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1515
1516 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1517                              gfn_t gfn)
1518 {
1519         if (memslot && memslot->dirty_bitmap) {
1520                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1521
1522                 if (!test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap))
1523                         memslot->nr_dirty_pages++;
1524         }
1525 }
1526
1527 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1528 {
1529         struct kvm_memory_slot *memslot;
1530
1531         memslot = gfn_to_memslot(kvm, gfn);
1532         mark_page_dirty_in_slot(kvm, memslot, gfn);
1533 }
1534
1535 /*
1536  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1537  */
1538 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1539 {
1540         DEFINE_WAIT(wait);
1541
1542         for (;;) {
1543                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1544
1545                 if (kvm_arch_vcpu_runnable(vcpu)) {
1546                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1547                         break;
1548                 }
1549                 if (kvm_cpu_has_pending_timer(vcpu))
1550                         break;
1551                 if (signal_pending(current))
1552                         break;
1553
1554                 schedule();
1555         }
1556
1557         finish_wait(&vcpu->wq, &wait);
1558 }
1559
1560 void kvm_resched(struct kvm_vcpu *vcpu)
1561 {
1562         if (!need_resched())
1563                 return;
1564         cond_resched();
1565 }
1566 EXPORT_SYMBOL_GPL(kvm_resched);
1567
1568 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1569 {
1570         struct kvm *kvm = me->kvm;
1571         struct kvm_vcpu *vcpu;
1572         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1573         int yielded = 0;
1574         int pass;
1575         int i;
1576
1577         /*
1578          * We boost the priority of a VCPU that is runnable but not
1579          * currently running, because it got preempted by something
1580          * else and called schedule in __vcpu_run.  Hopefully that
1581          * VCPU is holding the lock that we need and will release it.
1582          * We approximate round-robin by starting at the last boosted VCPU.
1583          */
1584         for (pass = 0; pass < 2 && !yielded; pass++) {
1585                 kvm_for_each_vcpu(i, vcpu, kvm) {
1586                         struct task_struct *task = NULL;
1587                         struct pid *pid;
1588                         if (!pass && i < last_boosted_vcpu) {
1589                                 i = last_boosted_vcpu;
1590                                 continue;
1591                         } else if (pass && i > last_boosted_vcpu)
1592                                 break;
1593                         if (vcpu == me)
1594                                 continue;
1595                         if (waitqueue_active(&vcpu->wq))
1596                                 continue;
1597                         rcu_read_lock();
1598                         pid = rcu_dereference(vcpu->pid);
1599                         if (pid)
1600                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1601                         rcu_read_unlock();
1602                         if (!task)
1603                                 continue;
1604                         if (task->flags & PF_VCPU) {
1605                                 put_task_struct(task);
1606                                 continue;
1607                         }
1608                         if (yield_to(task, 1)) {
1609                                 put_task_struct(task);
1610                                 kvm->last_boosted_vcpu = i;
1611                                 yielded = 1;
1612                                 break;
1613                         }
1614                         put_task_struct(task);
1615                 }
1616         }
1617 }
1618 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1619
1620 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1621 {
1622         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1623         struct page *page;
1624
1625         if (vmf->pgoff == 0)
1626                 page = virt_to_page(vcpu->run);
1627 #ifdef CONFIG_X86
1628         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1629                 page = virt_to_page(vcpu->arch.pio_data);
1630 #endif
1631 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1632         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1633                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1634 #endif
1635         else
1636                 return kvm_arch_vcpu_fault(vcpu, vmf);
1637         get_page(page);
1638         vmf->page = page;
1639         return 0;
1640 }
1641
1642 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1643         .fault = kvm_vcpu_fault,
1644 };
1645
1646 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1647 {
1648         vma->vm_ops = &kvm_vcpu_vm_ops;
1649         return 0;
1650 }
1651
1652 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1653 {
1654         struct kvm_vcpu *vcpu = filp->private_data;
1655
1656         kvm_put_kvm(vcpu->kvm);
1657         return 0;
1658 }
1659
1660 static struct file_operations kvm_vcpu_fops = {
1661         .release        = kvm_vcpu_release,
1662         .unlocked_ioctl = kvm_vcpu_ioctl,
1663 #ifdef CONFIG_COMPAT
1664         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1665 #endif
1666         .mmap           = kvm_vcpu_mmap,
1667         .llseek         = noop_llseek,
1668 };
1669
1670 /*
1671  * Allocates an inode for the vcpu.
1672  */
1673 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1674 {
1675         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1676 }
1677
1678 /*
1679  * Creates some virtual cpus.  Good luck creating more than one.
1680  */
1681 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1682 {
1683         int r;
1684         struct kvm_vcpu *vcpu, *v;
1685
1686         vcpu = kvm_arch_vcpu_create(kvm, id);
1687         if (IS_ERR(vcpu))
1688                 return PTR_ERR(vcpu);
1689
1690         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1691
1692         r = kvm_arch_vcpu_setup(vcpu);
1693         if (r)
1694                 goto vcpu_destroy;
1695
1696         mutex_lock(&kvm->lock);
1697         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1698                 r = -EINVAL;
1699                 goto unlock_vcpu_destroy;
1700         }
1701
1702         kvm_for_each_vcpu(r, v, kvm)
1703                 if (v->vcpu_id == id) {
1704                         r = -EEXIST;
1705                         goto unlock_vcpu_destroy;
1706                 }
1707
1708         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1709
1710         /* Now it's all set up, let userspace reach it */
1711         kvm_get_kvm(kvm);
1712         r = create_vcpu_fd(vcpu);
1713         if (r < 0) {
1714                 kvm_put_kvm(kvm);
1715                 goto unlock_vcpu_destroy;
1716         }
1717
1718         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1719         smp_wmb();
1720         atomic_inc(&kvm->online_vcpus);
1721
1722         mutex_unlock(&kvm->lock);
1723         return r;
1724
1725 unlock_vcpu_destroy:
1726         mutex_unlock(&kvm->lock);
1727 vcpu_destroy:
1728         kvm_arch_vcpu_destroy(vcpu);
1729         return r;
1730 }
1731
1732 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1733 {
1734         if (sigset) {
1735                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1736                 vcpu->sigset_active = 1;
1737                 vcpu->sigset = *sigset;
1738         } else
1739                 vcpu->sigset_active = 0;
1740         return 0;
1741 }
1742
1743 static long kvm_vcpu_ioctl(struct file *filp,
1744                            unsigned int ioctl, unsigned long arg)
1745 {
1746         struct kvm_vcpu *vcpu = filp->private_data;
1747         void __user *argp = (void __user *)arg;
1748         int r;
1749         struct kvm_fpu *fpu = NULL;
1750         struct kvm_sregs *kvm_sregs = NULL;
1751
1752         if (vcpu->kvm->mm != current->mm)
1753                 return -EIO;
1754
1755 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1756         /*
1757          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1758          * so vcpu_load() would break it.
1759          */
1760         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1761                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1762 #endif
1763
1764
1765         vcpu_load(vcpu);
1766         switch (ioctl) {
1767         case KVM_RUN:
1768                 r = -EINVAL;
1769                 if (arg)
1770                         goto out;
1771                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1772                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1773                 break;
1774         case KVM_GET_REGS: {
1775                 struct kvm_regs *kvm_regs;
1776
1777                 r = -ENOMEM;
1778                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1779                 if (!kvm_regs)
1780                         goto out;
1781                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1782                 if (r)
1783                         goto out_free1;
1784                 r = -EFAULT;
1785                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1786                         goto out_free1;
1787                 r = 0;
1788 out_free1:
1789                 kfree(kvm_regs);
1790                 break;
1791         }
1792         case KVM_SET_REGS: {
1793                 struct kvm_regs *kvm_regs;
1794
1795                 r = -ENOMEM;
1796                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1797                 if (IS_ERR(kvm_regs)) {
1798                         r = PTR_ERR(kvm_regs);
1799                         goto out;
1800                 }
1801                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1802                 if (r)
1803                         goto out_free2;
1804                 r = 0;
1805 out_free2:
1806                 kfree(kvm_regs);
1807                 break;
1808         }
1809         case KVM_GET_SREGS: {
1810                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1811                 r = -ENOMEM;
1812                 if (!kvm_sregs)
1813                         goto out;
1814                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1815                 if (r)
1816                         goto out;
1817                 r = -EFAULT;
1818                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1819                         goto out;
1820                 r = 0;
1821                 break;
1822         }
1823         case KVM_SET_SREGS: {
1824                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1825                 if (IS_ERR(kvm_sregs)) {
1826                         r = PTR_ERR(kvm_sregs);
1827                         goto out;
1828                 }
1829                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1830                 if (r)
1831                         goto out;
1832                 r = 0;
1833                 break;
1834         }
1835         case KVM_GET_MP_STATE: {
1836                 struct kvm_mp_state mp_state;
1837
1838                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1839                 if (r)
1840                         goto out;
1841                 r = -EFAULT;
1842                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1843                         goto out;
1844                 r = 0;
1845                 break;
1846         }
1847         case KVM_SET_MP_STATE: {
1848                 struct kvm_mp_state mp_state;
1849
1850                 r = -EFAULT;
1851                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1852                         goto out;
1853                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1854                 if (r)
1855                         goto out;
1856                 r = 0;
1857                 break;
1858         }
1859         case KVM_TRANSLATE: {
1860                 struct kvm_translation tr;
1861
1862                 r = -EFAULT;
1863                 if (copy_from_user(&tr, argp, sizeof tr))
1864                         goto out;
1865                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1866                 if (r)
1867                         goto out;
1868                 r = -EFAULT;
1869                 if (copy_to_user(argp, &tr, sizeof tr))
1870                         goto out;
1871                 r = 0;
1872                 break;
1873         }
1874         case KVM_SET_GUEST_DEBUG: {
1875                 struct kvm_guest_debug dbg;
1876
1877                 r = -EFAULT;
1878                 if (copy_from_user(&dbg, argp, sizeof dbg))
1879                         goto out;
1880                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1881                 if (r)
1882                         goto out;
1883                 r = 0;
1884                 break;
1885         }
1886         case KVM_SET_SIGNAL_MASK: {
1887                 struct kvm_signal_mask __user *sigmask_arg = argp;
1888                 struct kvm_signal_mask kvm_sigmask;
1889                 sigset_t sigset, *p;
1890
1891                 p = NULL;
1892                 if (argp) {
1893                         r = -EFAULT;
1894                         if (copy_from_user(&kvm_sigmask, argp,
1895                                            sizeof kvm_sigmask))
1896                                 goto out;
1897                         r = -EINVAL;
1898                         if (kvm_sigmask.len != sizeof sigset)
1899                                 goto out;
1900                         r = -EFAULT;
1901                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1902                                            sizeof sigset))
1903                                 goto out;
1904                         p = &sigset;
1905                 }
1906                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1907                 break;
1908         }
1909         case KVM_GET_FPU: {
1910                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1911                 r = -ENOMEM;
1912                 if (!fpu)
1913                         goto out;
1914                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1915                 if (r)
1916                         goto out;
1917                 r = -EFAULT;
1918                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1919                         goto out;
1920                 r = 0;
1921                 break;
1922         }
1923         case KVM_SET_FPU: {
1924                 fpu = memdup_user(argp, sizeof(*fpu));
1925                 if (IS_ERR(fpu)) {
1926                         r = PTR_ERR(fpu);
1927                         goto out;
1928                 }
1929                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1930                 if (r)
1931                         goto out;
1932                 r = 0;
1933                 break;
1934         }
1935         default:
1936                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1937         }
1938 out:
1939         vcpu_put(vcpu);
1940         kfree(fpu);
1941         kfree(kvm_sregs);
1942         return r;
1943 }
1944
1945 #ifdef CONFIG_COMPAT
1946 static long kvm_vcpu_compat_ioctl(struct file *filp,
1947                                   unsigned int ioctl, unsigned long arg)
1948 {
1949         struct kvm_vcpu *vcpu = filp->private_data;
1950         void __user *argp = compat_ptr(arg);
1951         int r;
1952
1953         if (vcpu->kvm->mm != current->mm)
1954                 return -EIO;
1955
1956         switch (ioctl) {
1957         case KVM_SET_SIGNAL_MASK: {
1958                 struct kvm_signal_mask __user *sigmask_arg = argp;
1959                 struct kvm_signal_mask kvm_sigmask;
1960                 compat_sigset_t csigset;
1961                 sigset_t sigset;
1962
1963                 if (argp) {
1964                         r = -EFAULT;
1965                         if (copy_from_user(&kvm_sigmask, argp,
1966                                            sizeof kvm_sigmask))
1967                                 goto out;
1968                         r = -EINVAL;
1969                         if (kvm_sigmask.len != sizeof csigset)
1970                                 goto out;
1971                         r = -EFAULT;
1972                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1973                                            sizeof csigset))
1974                                 goto out;
1975                 }
1976                 sigset_from_compat(&sigset, &csigset);
1977                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1978                 break;
1979         }
1980         default:
1981                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1982         }
1983
1984 out:
1985         return r;
1986 }
1987 #endif
1988
1989 static long kvm_vm_ioctl(struct file *filp,
1990                            unsigned int ioctl, unsigned long arg)
1991 {
1992         struct kvm *kvm = filp->private_data;
1993         void __user *argp = (void __user *)arg;
1994         int r;
1995
1996         if (kvm->mm != current->mm)
1997                 return -EIO;
1998         switch (ioctl) {
1999         case KVM_CREATE_VCPU:
2000                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2001                 if (r < 0)
2002                         goto out;
2003                 break;
2004         case KVM_SET_USER_MEMORY_REGION: {
2005                 struct kvm_userspace_memory_region kvm_userspace_mem;
2006
2007                 r = -EFAULT;
2008                 if (copy_from_user(&kvm_userspace_mem, argp,
2009                                                 sizeof kvm_userspace_mem))
2010                         goto out;
2011
2012                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2013                 if (r)
2014                         goto out;
2015                 break;
2016         }
2017         case KVM_GET_DIRTY_LOG: {
2018                 struct kvm_dirty_log log;
2019
2020                 r = -EFAULT;
2021                 if (copy_from_user(&log, argp, sizeof log))
2022                         goto out;
2023                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2024                 if (r)
2025                         goto out;
2026                 break;
2027         }
2028 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2029         case KVM_REGISTER_COALESCED_MMIO: {
2030                 struct kvm_coalesced_mmio_zone zone;
2031                 r = -EFAULT;
2032                 if (copy_from_user(&zone, argp, sizeof zone))
2033                         goto out;
2034                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2035                 if (r)
2036                         goto out;
2037                 r = 0;
2038                 break;
2039         }
2040         case KVM_UNREGISTER_COALESCED_MMIO: {
2041                 struct kvm_coalesced_mmio_zone zone;
2042                 r = -EFAULT;
2043                 if (copy_from_user(&zone, argp, sizeof zone))
2044                         goto out;
2045                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2046                 if (r)
2047                         goto out;
2048                 r = 0;
2049                 break;
2050         }
2051 #endif
2052         case KVM_IRQFD: {
2053                 struct kvm_irqfd data;
2054
2055                 r = -EFAULT;
2056                 if (copy_from_user(&data, argp, sizeof data))
2057                         goto out;
2058                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2059                 break;
2060         }
2061         case KVM_IOEVENTFD: {
2062                 struct kvm_ioeventfd data;
2063
2064                 r = -EFAULT;
2065                 if (copy_from_user(&data, argp, sizeof data))
2066                         goto out;
2067                 r = kvm_ioeventfd(kvm, &data);
2068                 break;
2069         }
2070 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2071         case KVM_SET_BOOT_CPU_ID:
2072                 r = 0;
2073                 mutex_lock(&kvm->lock);
2074                 if (atomic_read(&kvm->online_vcpus) != 0)
2075                         r = -EBUSY;
2076                 else
2077                         kvm->bsp_vcpu_id = arg;
2078                 mutex_unlock(&kvm->lock);
2079                 break;
2080 #endif
2081         default:
2082                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2083                 if (r == -ENOTTY)
2084                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2085         }
2086 out:
2087         return r;
2088 }
2089
2090 #ifdef CONFIG_COMPAT
2091 struct compat_kvm_dirty_log {
2092         __u32 slot;
2093         __u32 padding1;
2094         union {
2095                 compat_uptr_t dirty_bitmap; /* one bit per page */
2096                 __u64 padding2;
2097         };
2098 };
2099
2100 static long kvm_vm_compat_ioctl(struct file *filp,
2101                            unsigned int ioctl, unsigned long arg)
2102 {
2103         struct kvm *kvm = filp->private_data;
2104         int r;
2105
2106         if (kvm->mm != current->mm)
2107                 return -EIO;
2108         switch (ioctl) {
2109         case KVM_GET_DIRTY_LOG: {
2110                 struct compat_kvm_dirty_log compat_log;
2111                 struct kvm_dirty_log log;
2112
2113                 r = -EFAULT;
2114                 if (copy_from_user(&compat_log, (void __user *)arg,
2115                                    sizeof(compat_log)))
2116                         goto out;
2117                 log.slot         = compat_log.slot;
2118                 log.padding1     = compat_log.padding1;
2119                 log.padding2     = compat_log.padding2;
2120                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2121
2122                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2123                 if (r)
2124                         goto out;
2125                 break;
2126         }
2127         default:
2128                 r = kvm_vm_ioctl(filp, ioctl, arg);
2129         }
2130
2131 out:
2132         return r;
2133 }
2134 #endif
2135
2136 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2137 {
2138         struct page *page[1];
2139         unsigned long addr;
2140         int npages;
2141         gfn_t gfn = vmf->pgoff;
2142         struct kvm *kvm = vma->vm_file->private_data;
2143
2144         addr = gfn_to_hva(kvm, gfn);
2145         if (kvm_is_error_hva(addr))
2146                 return VM_FAULT_SIGBUS;
2147
2148         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2149                                 NULL);
2150         if (unlikely(npages != 1))
2151                 return VM_FAULT_SIGBUS;
2152
2153         vmf->page = page[0];
2154         return 0;
2155 }
2156
2157 static const struct vm_operations_struct kvm_vm_vm_ops = {
2158         .fault = kvm_vm_fault,
2159 };
2160
2161 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2162 {
2163         vma->vm_ops = &kvm_vm_vm_ops;
2164         return 0;
2165 }
2166
2167 static struct file_operations kvm_vm_fops = {
2168         .release        = kvm_vm_release,
2169         .unlocked_ioctl = kvm_vm_ioctl,
2170 #ifdef CONFIG_COMPAT
2171         .compat_ioctl   = kvm_vm_compat_ioctl,
2172 #endif
2173         .mmap           = kvm_vm_mmap,
2174         .llseek         = noop_llseek,
2175 };
2176
2177 static int kvm_dev_ioctl_create_vm(unsigned long type)
2178 {
2179         int r;
2180         struct kvm *kvm;
2181
2182         kvm = kvm_create_vm(type);
2183         if (IS_ERR(kvm))
2184                 return PTR_ERR(kvm);
2185 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2186         r = kvm_coalesced_mmio_init(kvm);
2187         if (r < 0) {
2188                 kvm_put_kvm(kvm);
2189                 return r;
2190         }
2191 #endif
2192         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2193         if (r < 0)
2194                 kvm_put_kvm(kvm);
2195
2196         return r;
2197 }
2198
2199 static long kvm_dev_ioctl_check_extension_generic(long arg)
2200 {
2201         switch (arg) {
2202         case KVM_CAP_USER_MEMORY:
2203         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2204         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2205 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2206         case KVM_CAP_SET_BOOT_CPU_ID:
2207 #endif
2208         case KVM_CAP_INTERNAL_ERROR_DATA:
2209                 return 1;
2210 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2211         case KVM_CAP_IRQ_ROUTING:
2212                 return KVM_MAX_IRQ_ROUTES;
2213 #endif
2214         default:
2215                 break;
2216         }
2217         return kvm_dev_ioctl_check_extension(arg);
2218 }
2219
2220 static long kvm_dev_ioctl(struct file *filp,
2221                           unsigned int ioctl, unsigned long arg)
2222 {
2223         long r = -EINVAL;
2224
2225         switch (ioctl) {
2226         case KVM_GET_API_VERSION:
2227                 r = -EINVAL;
2228                 if (arg)
2229                         goto out;
2230                 r = KVM_API_VERSION;
2231                 break;
2232         case KVM_CREATE_VM:
2233                 r = kvm_dev_ioctl_create_vm(arg);
2234                 break;
2235         case KVM_CHECK_EXTENSION:
2236                 r = kvm_dev_ioctl_check_extension_generic(arg);
2237                 break;
2238         case KVM_GET_VCPU_MMAP_SIZE:
2239                 r = -EINVAL;
2240                 if (arg)
2241                         goto out;
2242                 r = PAGE_SIZE;     /* struct kvm_run */
2243 #ifdef CONFIG_X86
2244                 r += PAGE_SIZE;    /* pio data page */
2245 #endif
2246 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2247                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2248 #endif
2249                 break;
2250         case KVM_TRACE_ENABLE:
2251         case KVM_TRACE_PAUSE:
2252         case KVM_TRACE_DISABLE:
2253                 r = -EOPNOTSUPP;
2254                 break;
2255         default:
2256                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2257         }
2258 out:
2259         return r;
2260 }
2261
2262 static struct file_operations kvm_chardev_ops = {
2263         .unlocked_ioctl = kvm_dev_ioctl,
2264         .compat_ioctl   = kvm_dev_ioctl,
2265         .llseek         = noop_llseek,
2266 };
2267
2268 static struct miscdevice kvm_dev = {
2269         KVM_MINOR,
2270         "kvm",
2271         &kvm_chardev_ops,
2272 };
2273
2274 static void hardware_enable_nolock(void *junk)
2275 {
2276         int cpu = raw_smp_processor_id();
2277         int r;
2278
2279         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2280                 return;
2281
2282         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2283
2284         r = kvm_arch_hardware_enable(NULL);
2285
2286         if (r) {
2287                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2288                 atomic_inc(&hardware_enable_failed);
2289                 printk(KERN_INFO "kvm: enabling virtualization on "
2290                                  "CPU%d failed\n", cpu);
2291         }
2292 }
2293
2294 static void hardware_enable(void *junk)
2295 {
2296         raw_spin_lock(&kvm_lock);
2297         hardware_enable_nolock(junk);
2298         raw_spin_unlock(&kvm_lock);
2299 }
2300
2301 static void hardware_disable_nolock(void *junk)
2302 {
2303         int cpu = raw_smp_processor_id();
2304
2305         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2306                 return;
2307         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2308         kvm_arch_hardware_disable(NULL);
2309 }
2310
2311 static void hardware_disable(void *junk)
2312 {
2313         raw_spin_lock(&kvm_lock);
2314         hardware_disable_nolock(junk);
2315         raw_spin_unlock(&kvm_lock);
2316 }
2317
2318 static void hardware_disable_all_nolock(void)
2319 {
2320         BUG_ON(!kvm_usage_count);
2321
2322         kvm_usage_count--;
2323         if (!kvm_usage_count)
2324                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2325 }
2326
2327 static void hardware_disable_all(void)
2328 {
2329         raw_spin_lock(&kvm_lock);
2330         hardware_disable_all_nolock();
2331         raw_spin_unlock(&kvm_lock);
2332 }
2333
2334 static int hardware_enable_all(void)
2335 {
2336         int r = 0;
2337
2338         raw_spin_lock(&kvm_lock);
2339
2340         kvm_usage_count++;
2341         if (kvm_usage_count == 1) {
2342                 atomic_set(&hardware_enable_failed, 0);
2343                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2344
2345                 if (atomic_read(&hardware_enable_failed)) {
2346                         hardware_disable_all_nolock();
2347                         r = -EBUSY;
2348                 }
2349         }
2350
2351         raw_spin_unlock(&kvm_lock);
2352
2353         return r;
2354 }
2355
2356 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2357                            void *v)
2358 {
2359         int cpu = (long)v;
2360
2361         if (!kvm_usage_count)
2362                 return NOTIFY_OK;
2363
2364         val &= ~CPU_TASKS_FROZEN;
2365         switch (val) {
2366         case CPU_DYING:
2367                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2368                        cpu);
2369                 hardware_disable(NULL);
2370                 break;
2371         case CPU_STARTING:
2372                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2373                        cpu);
2374                 hardware_enable(NULL);
2375                 break;
2376         }
2377         return NOTIFY_OK;
2378 }
2379
2380
2381 asmlinkage void kvm_spurious_fault(void)
2382 {
2383         /* Fault while not rebooting.  We want the trace. */
2384         BUG();
2385 }
2386 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2387
2388 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2389                       void *v)
2390 {
2391         /*
2392          * Some (well, at least mine) BIOSes hang on reboot if
2393          * in vmx root mode.
2394          *
2395          * And Intel TXT required VMX off for all cpu when system shutdown.
2396          */
2397         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2398         kvm_rebooting = true;
2399         on_each_cpu(hardware_disable_nolock, NULL, 1);
2400         return NOTIFY_OK;
2401 }
2402
2403 static struct notifier_block kvm_reboot_notifier = {
2404         .notifier_call = kvm_reboot,
2405         .priority = 0,
2406 };
2407
2408 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2409 {
2410         int i;
2411
2412         for (i = 0; i < bus->dev_count; i++) {
2413                 struct kvm_io_device *pos = bus->range[i].dev;
2414
2415                 kvm_iodevice_destructor(pos);
2416         }
2417         kfree(bus);
2418 }
2419
2420 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2421 {
2422         const struct kvm_io_range *r1 = p1;
2423         const struct kvm_io_range *r2 = p2;
2424
2425         if (r1->addr < r2->addr)
2426                 return -1;
2427         if (r1->addr + r1->len > r2->addr + r2->len)
2428                 return 1;
2429         return 0;
2430 }
2431
2432 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2433                           gpa_t addr, int len)
2434 {
2435         if (bus->dev_count == NR_IOBUS_DEVS)
2436                 return -ENOSPC;
2437
2438         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2439                 .addr = addr,
2440                 .len = len,
2441                 .dev = dev,
2442         };
2443
2444         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2445                 kvm_io_bus_sort_cmp, NULL);
2446
2447         return 0;
2448 }
2449
2450 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2451                              gpa_t addr, int len)
2452 {
2453         struct kvm_io_range *range, key;
2454         int off;
2455
2456         key = (struct kvm_io_range) {
2457                 .addr = addr,
2458                 .len = len,
2459         };
2460
2461         range = bsearch(&key, bus->range, bus->dev_count,
2462                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2463         if (range == NULL)
2464                 return -ENOENT;
2465
2466         off = range - bus->range;
2467
2468         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2469                 off--;
2470
2471         return off;
2472 }
2473
2474 /* kvm_io_bus_write - called under kvm->slots_lock */
2475 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2476                      int len, const void *val)
2477 {
2478         int idx;
2479         struct kvm_io_bus *bus;
2480         struct kvm_io_range range;
2481
2482         range = (struct kvm_io_range) {
2483                 .addr = addr,
2484                 .len = len,
2485         };
2486
2487         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2488         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2489         if (idx < 0)
2490                 return -EOPNOTSUPP;
2491
2492         while (idx < bus->dev_count &&
2493                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2494                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2495                         return 0;
2496                 idx++;
2497         }
2498
2499         return -EOPNOTSUPP;
2500 }
2501
2502 /* kvm_io_bus_read - called under kvm->slots_lock */
2503 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2504                     int len, void *val)
2505 {
2506         int idx;
2507         struct kvm_io_bus *bus;
2508         struct kvm_io_range range;
2509
2510         range = (struct kvm_io_range) {
2511                 .addr = addr,
2512                 .len = len,
2513         };
2514
2515         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2516         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2517         if (idx < 0)
2518                 return -EOPNOTSUPP;
2519
2520         while (idx < bus->dev_count &&
2521                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2522                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2523                         return 0;
2524                 idx++;
2525         }
2526
2527         return -EOPNOTSUPP;
2528 }
2529
2530 /* Caller must hold slots_lock. */
2531 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2532                             int len, struct kvm_io_device *dev)
2533 {
2534         struct kvm_io_bus *new_bus, *bus;
2535
2536         bus = kvm->buses[bus_idx];
2537         if (bus->dev_count > NR_IOBUS_DEVS-1)
2538                 return -ENOSPC;
2539
2540         new_bus = kmemdup(bus, sizeof(struct kvm_io_bus), GFP_KERNEL);
2541         if (!new_bus)
2542                 return -ENOMEM;
2543         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2544         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2545         synchronize_srcu_expedited(&kvm->srcu);
2546         kfree(bus);
2547
2548         return 0;
2549 }
2550
2551 /* Caller must hold slots_lock. */
2552 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2553                               struct kvm_io_device *dev)
2554 {
2555         int i, r;
2556         struct kvm_io_bus *new_bus, *bus;
2557
2558         bus = kvm->buses[bus_idx];
2559
2560         new_bus = kmemdup(bus, sizeof(*bus), GFP_KERNEL);
2561         if (!new_bus)
2562                 return -ENOMEM;
2563
2564         r = -ENOENT;
2565         for (i = 0; i < new_bus->dev_count; i++)
2566                 if (new_bus->range[i].dev == dev) {
2567                         r = 0;
2568                         new_bus->dev_count--;
2569                         new_bus->range[i] = new_bus->range[new_bus->dev_count];
2570                         sort(new_bus->range, new_bus->dev_count,
2571                              sizeof(struct kvm_io_range),
2572                              kvm_io_bus_sort_cmp, NULL);
2573                         break;
2574                 }
2575
2576         if (r) {
2577                 kfree(new_bus);
2578                 return r;
2579         }
2580
2581         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2582         synchronize_srcu_expedited(&kvm->srcu);
2583         kfree(bus);
2584         return r;
2585 }
2586
2587 static struct notifier_block kvm_cpu_notifier = {
2588         .notifier_call = kvm_cpu_hotplug,
2589 };
2590
2591 static int vm_stat_get(void *_offset, u64 *val)
2592 {
2593         unsigned offset = (long)_offset;
2594         struct kvm *kvm;
2595
2596         *val = 0;
2597         raw_spin_lock(&kvm_lock);
2598         list_for_each_entry(kvm, &vm_list, vm_list)
2599                 *val += *(u32 *)((void *)kvm + offset);
2600         raw_spin_unlock(&kvm_lock);
2601         return 0;
2602 }
2603
2604 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2605
2606 static int vcpu_stat_get(void *_offset, u64 *val)
2607 {
2608         unsigned offset = (long)_offset;
2609         struct kvm *kvm;
2610         struct kvm_vcpu *vcpu;
2611         int i;
2612
2613         *val = 0;
2614         raw_spin_lock(&kvm_lock);
2615         list_for_each_entry(kvm, &vm_list, vm_list)
2616                 kvm_for_each_vcpu(i, vcpu, kvm)
2617                         *val += *(u32 *)((void *)vcpu + offset);
2618
2619         raw_spin_unlock(&kvm_lock);
2620         return 0;
2621 }
2622
2623 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2624
2625 static const struct file_operations *stat_fops[] = {
2626         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2627         [KVM_STAT_VM]   = &vm_stat_fops,
2628 };
2629
2630 static int kvm_init_debug(void)
2631 {
2632         int r = -EFAULT;
2633         struct kvm_stats_debugfs_item *p;
2634
2635         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2636         if (kvm_debugfs_dir == NULL)
2637                 goto out;
2638
2639         for (p = debugfs_entries; p->name; ++p) {
2640                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2641                                                 (void *)(long)p->offset,
2642                                                 stat_fops[p->kind]);
2643                 if (p->dentry == NULL)
2644                         goto out_dir;
2645         }
2646
2647         return 0;
2648
2649 out_dir:
2650         debugfs_remove_recursive(kvm_debugfs_dir);
2651 out:
2652         return r;
2653 }
2654
2655 static void kvm_exit_debug(void)
2656 {
2657         struct kvm_stats_debugfs_item *p;
2658
2659         for (p = debugfs_entries; p->name; ++p)
2660                 debugfs_remove(p->dentry);
2661         debugfs_remove(kvm_debugfs_dir);
2662 }
2663
2664 static int kvm_suspend(void)
2665 {
2666         if (kvm_usage_count)
2667                 hardware_disable_nolock(NULL);
2668         return 0;
2669 }
2670
2671 static void kvm_resume(void)
2672 {
2673         if (kvm_usage_count) {
2674                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2675                 hardware_enable_nolock(NULL);
2676         }
2677 }
2678
2679 static struct syscore_ops kvm_syscore_ops = {
2680         .suspend = kvm_suspend,
2681         .resume = kvm_resume,
2682 };
2683
2684 struct page *bad_page;
2685 pfn_t bad_pfn;
2686
2687 static inline
2688 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2689 {
2690         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2691 }
2692
2693 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2694 {
2695         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2696
2697         kvm_arch_vcpu_load(vcpu, cpu);
2698 }
2699
2700 static void kvm_sched_out(struct preempt_notifier *pn,
2701                           struct task_struct *next)
2702 {
2703         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2704
2705         kvm_arch_vcpu_put(vcpu);
2706 }
2707
2708 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2709                   struct module *module)
2710 {
2711         int r;
2712         int cpu;
2713
2714         r = kvm_arch_init(opaque);
2715         if (r)
2716                 goto out_fail;
2717
2718         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2719
2720         if (bad_page == NULL) {
2721                 r = -ENOMEM;
2722                 goto out;
2723         }
2724
2725         bad_pfn = page_to_pfn(bad_page);
2726
2727         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2728
2729         if (hwpoison_page == NULL) {
2730                 r = -ENOMEM;
2731                 goto out_free_0;
2732         }
2733
2734         hwpoison_pfn = page_to_pfn(hwpoison_page);
2735
2736         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2737
2738         if (fault_page == NULL) {
2739                 r = -ENOMEM;
2740                 goto out_free_0;
2741         }
2742
2743         fault_pfn = page_to_pfn(fault_page);
2744
2745         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2746                 r = -ENOMEM;
2747                 goto out_free_0;
2748         }
2749
2750         r = kvm_arch_hardware_setup();
2751         if (r < 0)
2752                 goto out_free_0a;
2753
2754         for_each_online_cpu(cpu) {
2755                 smp_call_function_single(cpu,
2756                                 kvm_arch_check_processor_compat,
2757                                 &r, 1);
2758                 if (r < 0)
2759                         goto out_free_1;
2760         }
2761
2762         r = register_cpu_notifier(&kvm_cpu_notifier);
2763         if (r)
2764                 goto out_free_2;
2765         register_reboot_notifier(&kvm_reboot_notifier);
2766
2767         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2768         if (!vcpu_align)
2769                 vcpu_align = __alignof__(struct kvm_vcpu);
2770         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2771                                            0, NULL);
2772         if (!kvm_vcpu_cache) {
2773                 r = -ENOMEM;
2774                 goto out_free_3;
2775         }
2776
2777         r = kvm_async_pf_init();
2778         if (r)
2779                 goto out_free;
2780
2781         kvm_chardev_ops.owner = module;
2782         kvm_vm_fops.owner = module;
2783         kvm_vcpu_fops.owner = module;
2784
2785         r = misc_register(&kvm_dev);
2786         if (r) {
2787                 printk(KERN_ERR "kvm: misc device register failed\n");
2788                 goto out_unreg;
2789         }
2790
2791         register_syscore_ops(&kvm_syscore_ops);
2792
2793         kvm_preempt_ops.sched_in = kvm_sched_in;
2794         kvm_preempt_ops.sched_out = kvm_sched_out;
2795
2796         r = kvm_init_debug();
2797         if (r) {
2798                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2799                 goto out_undebugfs;
2800         }
2801
2802         return 0;
2803
2804 out_undebugfs:
2805         unregister_syscore_ops(&kvm_syscore_ops);
2806 out_unreg:
2807         kvm_async_pf_deinit();
2808 out_free:
2809         kmem_cache_destroy(kvm_vcpu_cache);
2810 out_free_3:
2811         unregister_reboot_notifier(&kvm_reboot_notifier);
2812         unregister_cpu_notifier(&kvm_cpu_notifier);
2813 out_free_2:
2814 out_free_1:
2815         kvm_arch_hardware_unsetup();
2816 out_free_0a:
2817         free_cpumask_var(cpus_hardware_enabled);
2818 out_free_0:
2819         if (fault_page)
2820                 __free_page(fault_page);
2821         if (hwpoison_page)
2822                 __free_page(hwpoison_page);
2823         __free_page(bad_page);
2824 out:
2825         kvm_arch_exit();
2826 out_fail:
2827         return r;
2828 }
2829 EXPORT_SYMBOL_GPL(kvm_init);
2830
2831 void kvm_exit(void)
2832 {
2833         kvm_exit_debug();
2834         misc_deregister(&kvm_dev);
2835         kmem_cache_destroy(kvm_vcpu_cache);
2836         kvm_async_pf_deinit();
2837         unregister_syscore_ops(&kvm_syscore_ops);
2838         unregister_reboot_notifier(&kvm_reboot_notifier);
2839         unregister_cpu_notifier(&kvm_cpu_notifier);
2840         on_each_cpu(hardware_disable_nolock, NULL, 1);
2841         kvm_arch_hardware_unsetup();
2842         kvm_arch_exit();
2843         free_cpumask_var(cpus_hardware_enabled);
2844         __free_page(hwpoison_page);
2845         __free_page(bad_page);
2846 }
2847 EXPORT_SYMBOL_GPL(kvm_exit);