virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/syscore_ops.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/pgtable.h>
  57
  58 #include "coalesced_mmio.h"
  59 #include "async_pf.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/kvm.h>
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 /*
  68  * Ordering of locks:
  69  *
  70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  71  */
  72
  73 DEFINE_RAW_SPINLOCK(kvm_lock);
  74 LIST_HEAD(vm_list);
  75
  76 static cpumask_var_t cpus_hardware_enabled;
  77 static int kvm_usage_count = 0;
  78 static atomic_t hardware_enable_failed;
  79
  80 struct kmem_cache *kvm_vcpu_cache;
  81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  82
  83 static __read_mostly struct preempt_ops kvm_preempt_ops;
  84
  85 struct dentry *kvm_debugfs_dir;
  86
  87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  88                            unsigned long arg);
  89 #ifdef CONFIG_COMPAT
  90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  91                                   unsigned long arg);
  92 #endif
  93 static int hardware_enable_all(void);
  94 static void hardware_disable_all(void);
  95
  96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  97
  98 bool kvm_rebooting;
  99 EXPORT_SYMBOL_GPL(kvm_rebooting);
 100
 101 static bool largepages_enabled = true;
 102
 103 static struct page *hwpoison_page;
 104 static pfn_t hwpoison_pfn;
 105
 106 struct page *fault_page;
 107 pfn_t fault_pfn;
 108
 109 inline int kvm_is_mmio_pfn(pfn_t pfn)
 110 {
 111         if (pfn_valid(pfn)) {
 112                 int reserved;
 113                 struct page *tail = pfn_to_page(pfn);
 114                 struct page *head = compound_trans_head(tail);
 115                 reserved = PageReserved(head);
 116                 if (head != tail) {
 117                         /*
 118                          * "head" is not a dangling pointer
 119                          * (compound_trans_head takes care of that)
 120                          * but the hugepage may have been splitted
 121                          * from under us (and we may not hold a
 122                          * reference count on the head page so it can
 123                          * be reused before we run PageReferenced), so
 124                          * we've to check PageTail before returning
 125                          * what we just read.
 126                          */
 127                         smp_rmb();
 128                         if (PageTail(tail))
 129                                 return reserved;
 130                 }
 131                 return PageReserved(tail);
 132         }
 133
 134         return true;
 135 }
 136
 137 /*
 138  * Switches to specified vcpu, until a matching vcpu_put()
 139  */
 140 void vcpu_load(struct kvm_vcpu *vcpu)
 141 {
 142         int cpu;
 143
 144         mutex_lock(&vcpu->mutex);
 145         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 146                 /* The thread running this VCPU changed. */
 147                 struct pid *oldpid = vcpu->pid;
 148                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 149                 rcu_assign_pointer(vcpu->pid, newpid);
 150                 synchronize_rcu();
 151                 put_pid(oldpid);
 152         }
 153         cpu = get_cpu();
 154         preempt_notifier_register(&vcpu->preempt_notifier);
 155         kvm_arch_vcpu_load(vcpu, cpu);
 156         put_cpu();
 157 }
 158
 159 void vcpu_put(struct kvm_vcpu *vcpu)
 160 {
 161         preempt_disable();
 162         kvm_arch_vcpu_put(vcpu);
 163         preempt_notifier_unregister(&vcpu->preempt_notifier);
 164         preempt_enable();
 165         mutex_unlock(&vcpu->mutex);
 166 }
 167
 168 static void ack_flush(void *_completed)
 169 {
 170 }
 171
 172 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 173 {
 174         int i, cpu, me;
 175         cpumask_var_t cpus;
 176         bool called = true;
 177         struct kvm_vcpu *vcpu;
 178
 179         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 180
 181         me = get_cpu();
 182         kvm_for_each_vcpu(i, vcpu, kvm) {
 183                 kvm_make_request(req, vcpu);
 184                 cpu = vcpu->cpu;
 185
 186                 /* Set ->requests bit before we read ->mode */
 187                 smp_mb();
 188
 189                 if (cpus != NULL && cpu != -1 && cpu != me &&
 190                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 191                         cpumask_set_cpu(cpu, cpus);
 192         }
 193         if (unlikely(cpus == NULL))
 194                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 195         else if (!cpumask_empty(cpus))
 196                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 197         else
 198                 called = false;
 199         put_cpu();
 200         free_cpumask_var(cpus);
 201         return called;
 202 }
 203
 204 void kvm_flush_remote_tlbs(struct kvm *kvm)
 205 {
 206         int dirty_count = kvm->tlbs_dirty;
 207
 208         smp_mb();
 209         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 210                 ++kvm->stat.remote_tlb_flush;
 211         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 212 }
 213
 214 void kvm_reload_remote_mmus(struct kvm *kvm)
 215 {
 216         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 217 }
 218
 219 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 220 {
 221         struct page *page;
 222         int r;
 223
 224         mutex_init(&vcpu->mutex);
 225         vcpu->cpu = -1;
 226         vcpu->kvm = kvm;
 227         vcpu->vcpu_id = id;
 228         vcpu->pid = NULL;
 229         init_waitqueue_head(&vcpu->wq);
 230         kvm_async_pf_vcpu_init(vcpu);
 231
 232         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 233         if (!page) {
 234                 r = -ENOMEM;
 235                 goto fail;
 236         }
 237         vcpu->run = page_address(page);
 238
 239         r = kvm_arch_vcpu_init(vcpu);
 240         if (r < 0)
 241                 goto fail_free_run;
 242         return 0;
 243
 244 fail_free_run:
 245         free_page((unsigned long)vcpu->run);
 246 fail:
 247         return r;
 248 }
 249 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 250
 251 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 252 {
 253         put_pid(vcpu->pid);
 254         kvm_arch_vcpu_uninit(vcpu);
 255         free_page((unsigned long)vcpu->run);
 256 }
 257 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 258
 259 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 260 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 261 {
 262         return container_of(mn, struct kvm, mmu_notifier);
 263 }
 264
 265 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 266                                              struct mm_struct *mm,
 267                                              unsigned long address)
 268 {
 269         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 270         int need_tlb_flush, idx;
 271
 272         /*
 273          * When ->invalidate_page runs, the linux pte has been zapped
 274          * already but the page is still allocated until
 275          * ->invalidate_page returns. So if we increase the sequence
 276          * here the kvm page fault will notice if the spte can't be
 277          * established because the page is going to be freed. If
 278          * instead the kvm page fault establishes the spte before
 279          * ->invalidate_page runs, kvm_unmap_hva will release it
 280          * before returning.
 281          *
 282          * The sequence increase only need to be seen at spin_unlock
 283          * time, and not at spin_lock time.
 284          *
 285          * Increasing the sequence after the spin_unlock would be
 286          * unsafe because the kvm page fault could then establish the
 287          * pte after kvm_unmap_hva returned, without noticing the page
 288          * is going to be freed.
 289          */
 290         idx = srcu_read_lock(&kvm->srcu);
 291         spin_lock(&kvm->mmu_lock);
 292         kvm->mmu_notifier_seq++;
 293         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 294         spin_unlock(&kvm->mmu_lock);
 295         srcu_read_unlock(&kvm->srcu, idx);
 296
 297         /* we've to flush the tlb before the pages can be freed */
 298         if (need_tlb_flush)
 299                 kvm_flush_remote_tlbs(kvm);
 300
 301 }
 302
 303 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 304                                         struct mm_struct *mm,
 305                                         unsigned long address,
 306                                         pte_t pte)
 307 {
 308         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 309         int idx;
 310
 311         idx = srcu_read_lock(&kvm->srcu);
 312         spin_lock(&kvm->mmu_lock);
 313         kvm->mmu_notifier_seq++;
 314         kvm_set_spte_hva(kvm, address, pte);
 315         spin_unlock(&kvm->mmu_lock);
 316         srcu_read_unlock(&kvm->srcu, idx);
 317 }
 318
 319 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 320                                                     struct mm_struct *mm,
 321                                                     unsigned long start,
 322                                                     unsigned long end)
 323 {
 324         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 325         int need_tlb_flush = 0, idx;
 326
 327         idx = srcu_read_lock(&kvm->srcu);
 328         spin_lock(&kvm->mmu_lock);
 329         /*
 330          * The count increase must become visible at unlock time as no
 331          * spte can be established without taking the mmu_lock and
 332          * count is also read inside the mmu_lock critical section.
 333          */
 334         kvm->mmu_notifier_count++;
 335         for (; start < end; start += PAGE_SIZE)
 336                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 337         need_tlb_flush |= kvm->tlbs_dirty;
 338         spin_unlock(&kvm->mmu_lock);
 339         srcu_read_unlock(&kvm->srcu, idx);
 340
 341         /* we've to flush the tlb before the pages can be freed */
 342         if (need_tlb_flush)
 343                 kvm_flush_remote_tlbs(kvm);
 344 }
 345
 346 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 347                                                   struct mm_struct *mm,
 348                                                   unsigned long start,
 349                                                   unsigned long end)
 350 {
 351         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 352
 353         spin_lock(&kvm->mmu_lock);
 354         /*
 355          * This sequence increase will notify the kvm page fault that
 356          * the page that is going to be mapped in the spte could have
 357          * been freed.
 358          */
 359         kvm->mmu_notifier_seq++;
 360         /*
 361          * The above sequence increase must be visible before the
 362          * below count decrease but both values are read by the kvm
 363          * page fault under mmu_lock spinlock so we don't need to add
 364          * a smb_wmb() here in between the two.
 365          */
 366         kvm->mmu_notifier_count--;
 367         spin_unlock(&kvm->mmu_lock);
 368
 369         BUG_ON(kvm->mmu_notifier_count < 0);
 370 }
 371
 372 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 373                                               struct mm_struct *mm,
 374                                               unsigned long address)
 375 {
 376         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 377         int young, idx;
 378
 379         idx = srcu_read_lock(&kvm->srcu);
 380         spin_lock(&kvm->mmu_lock);
 381         young = kvm_age_hva(kvm, address);
 382         spin_unlock(&kvm->mmu_lock);
 383         srcu_read_unlock(&kvm->srcu, idx);
 384
 385         if (young)
 386                 kvm_flush_remote_tlbs(kvm);
 387
 388         return young;
 389 }
 390
 391 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 392                                        struct mm_struct *mm,
 393                                        unsigned long address)
 394 {
 395         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 396         int young, idx;
 397
 398         idx = srcu_read_lock(&kvm->srcu);
 399         spin_lock(&kvm->mmu_lock);
 400         young = kvm_test_age_hva(kvm, address);
 401         spin_unlock(&kvm->mmu_lock);
 402         srcu_read_unlock(&kvm->srcu, idx);
 403
 404         return young;
 405 }
 406
 407 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 408                                      struct mm_struct *mm)
 409 {
 410         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 411         int idx;
 412
 413         idx = srcu_read_lock(&kvm->srcu);
 414         kvm_arch_flush_shadow(kvm);
 415         srcu_read_unlock(&kvm->srcu, idx);
 416 }
 417
 418 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 419         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 420         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 421         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 422         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 423         .test_young             = kvm_mmu_notifier_test_young,
 424         .change_pte             = kvm_mmu_notifier_change_pte,
 425         .release                = kvm_mmu_notifier_release,
 426 };
 427
 428 static int kvm_init_mmu_notifier(struct kvm *kvm)
 429 {
 430         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 431         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 432 }
 433
 434 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 435
 436 static int kvm_init_mmu_notifier(struct kvm *kvm)
 437 {
 438         return 0;
 439 }
 440
 441 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 442
 443 static 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(void)
 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);
 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 struct kvm_memory_slot *
 644 search_memslots(struct kvm_memslots *slots, gfn_t gfn)
 645 {
 646         struct kvm_memory_slot *memslot;
 647
 648         kvm_for_each_memslot(memslot, slots)
 649                 if (gfn >= memslot->base_gfn &&
 650                       gfn < memslot->base_gfn + memslot->npages)
 651                         return memslot;
 652
 653         return NULL;
 654 }
 655
 656 static int cmp_memslot(const void *slot1, const void *slot2)
 657 {
 658         struct kvm_memory_slot *s1, *s2;
 659
 660         s1 = (struct kvm_memory_slot *)slot1;
 661         s2 = (struct kvm_memory_slot *)slot2;
 662
 663         if (s1->npages < s2->npages)
 664                 return 1;
 665         if (s1->npages > s2->npages)
 666                 return -1;
 667
 668         return 0;
 669 }
 670
 671 /*
 672  * Sort the memslots base on its size, so the larger slots
 673  * will get better fit.
 674  */
 675 static void sort_memslots(struct kvm_memslots *slots)
 676 {
 677         int i;
 678
 679         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 680               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 681
 682         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 683                 slots->id_to_index[slots->memslots[i].id] = i;
 684 }
 685
 686 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
 687 {
 688         if (new) {
 689                 int id = new->id;
 690                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 691                 unsigned long npages = old->npages;
 692
 693                 *old = *new;
 694                 if (new->npages != npages)
 695                         sort_memslots(slots);
 696         }
 697
 698         slots->generation++;
 699 }
 700
 701 /*
 702  * Allocate some memory and give it an address in the guest physical address
 703  * space.
 704  *
 705  * Discontiguous memory is allowed, mostly for framebuffers.
 706  *
 707  * Must be called holding mmap_sem for write.
 708  */
 709 int __kvm_set_memory_region(struct kvm *kvm,
 710                             struct kvm_userspace_memory_region *mem,
 711                             int user_alloc)
 712 {
 713         int r;
 714         gfn_t base_gfn;
 715         unsigned long npages;
 716         unsigned long i;
 717         struct kvm_memory_slot *memslot;
 718         struct kvm_memory_slot old, new;
 719         struct kvm_memslots *slots, *old_memslots;
 720
 721         r = -EINVAL;
 722         /* General sanity checks */
 723         if (mem->memory_size & (PAGE_SIZE - 1))
 724                 goto out;
 725         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 726                 goto out;
 727         /* We can read the guest memory with __xxx_user() later on. */
 728         if (user_alloc &&
 729             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 730              !access_ok(VERIFY_WRITE,
 731                         (void __user *)(unsigned long)mem->userspace_addr,
 732                         mem->memory_size)))
 733                 goto out;
 734         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 735                 goto out;
 736         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 737                 goto out;
 738
 739         memslot = id_to_memslot(kvm->memslots, mem->slot);
 740         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 741         npages = mem->memory_size >> PAGE_SHIFT;
 742
 743         r = -EINVAL;
 744         if (npages > KVM_MEM_MAX_NR_PAGES)
 745                 goto out;
 746
 747         if (!npages)
 748                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 749
 750         new = old = *memslot;
 751
 752         new.id = mem->slot;
 753         new.base_gfn = base_gfn;
 754         new.npages = npages;
 755         new.flags = mem->flags;
 756
 757         /* Disallow changing a memory slot's size. */
 758         r = -EINVAL;
 759         if (npages && old.npages && npages != old.npages)
 760                 goto out_free;
 761
 762         /* Check for overlaps */
 763         r = -EEXIST;
 764         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 765                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 766
 767                 if (s == memslot || !s->npages)
 768                         continue;
 769                 if (!((base_gfn + npages <= s->base_gfn) ||
 770                       (base_gfn >= s->base_gfn + s->npages)))
 771                         goto out_free;
 772         }
 773
 774         /* Free page dirty bitmap if unneeded */
 775         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 776                 new.dirty_bitmap = NULL;
 777
 778         r = -ENOMEM;
 779
 780         /* Allocate if a slot is being created */
 781 #ifndef CONFIG_S390
 782         if (npages && !new.rmap) {
 783                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
 784
 785                 if (!new.rmap)
 786                         goto out_free;
 787
 788                 new.user_alloc = user_alloc;
 789                 new.userspace_addr = mem->userspace_addr;
 790         }
 791         if (!npages)
 792                 goto skip_lpage;
 793
 794         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 795                 unsigned long ugfn;
 796                 unsigned long j;
 797                 int lpages;
 798                 int level = i + 2;
 799
 800                 /* Avoid unused variable warning if no large pages */
 801                 (void)level;
 802
 803                 if (new.lpage_info[i])
 804                         continue;
 805
 806                 lpages = 1 + ((base_gfn + npages - 1)
 807                              >> KVM_HPAGE_GFN_SHIFT(level));
 808                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
 809
 810                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
 811
 812                 if (!new.lpage_info[i])
 813                         goto out_free;
 814
 815                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
 816                         new.lpage_info[i][0].write_count = 1;
 817                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
 818                         new.lpage_info[i][lpages - 1].write_count = 1;
 819                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 820                 /*
 821                  * If the gfn and userspace address are not aligned wrt each
 822                  * other, or if explicitly asked to, disable large page
 823                  * support for this slot
 824                  */
 825                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 826                     !largepages_enabled)
 827                         for (j = 0; j < lpages; ++j)
 828                                 new.lpage_info[i][j].write_count = 1;
 829         }
 830
 831 skip_lpage:
 832
 833         /* Allocate page dirty bitmap if needed */
 834         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 835                 if (kvm_create_dirty_bitmap(&new) < 0)
 836                         goto out_free;
 837                 /* destroy any largepage mappings for dirty tracking */
 838         }
 839 #else  /* not defined CONFIG_S390 */
 840         new.user_alloc = user_alloc;
 841         if (user_alloc)
 842                 new.userspace_addr = mem->userspace_addr;
 843 #endif /* not defined CONFIG_S390 */
 844
 845         if (!npages) {
 846                 struct kvm_memory_slot *slot;
 847
 848                 r = -ENOMEM;
 849                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 850                                 GFP_KERNEL);
 851                 if (!slots)
 852                         goto out_free;
 853                 slot = id_to_memslot(slots, mem->slot);
 854                 slot->flags |= KVM_MEMSLOT_INVALID;
 855
 856                 update_memslots(slots, NULL);
 857
 858                 old_memslots = kvm->memslots;
 859                 rcu_assign_pointer(kvm->memslots, slots);
 860                 synchronize_srcu_expedited(&kvm->srcu);
 861                 /* From this point no new shadow pages pointing to a deleted
 862                  * memslot will be created.
 863                  *
 864                  * validation of sp->gfn happens in:
 865                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 866                  *      - kvm_is_visible_gfn (mmu_check_roots)
 867                  */
 868                 kvm_arch_flush_shadow(kvm);
 869                 kfree(old_memslots);
 870         }
 871
 872         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 873         if (r)
 874                 goto out_free;
 875
 876         /* map the pages in iommu page table */
 877         if (npages) {
 878                 r = kvm_iommu_map_pages(kvm, &new);
 879                 if (r)
 880                         goto out_free;
 881         }
 882
 883         r = -ENOMEM;
 884         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 885                         GFP_KERNEL);
 886         if (!slots)
 887                 goto out_free;
 888
 889         /* actual memory is freed via old in kvm_free_physmem_slot below */
 890         if (!npages) {
 891                 new.rmap = NULL;
 892                 new.dirty_bitmap = NULL;
 893                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 894                         new.lpage_info[i] = NULL;
 895         }
 896
 897         update_memslots(slots, &new);
 898         old_memslots = kvm->memslots;
 899         rcu_assign_pointer(kvm->memslots, slots);
 900         synchronize_srcu_expedited(&kvm->srcu);
 901
 902         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 903
 904         /*
 905          * If the new memory slot is created, we need to clear all
 906          * mmio sptes.
 907          */
 908         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 909                 kvm_arch_flush_shadow(kvm);
 910
 911         kvm_free_physmem_slot(&old, &new);
 912         kfree(old_memslots);
 913
 914         return 0;
 915
 916 out_free:
 917         kvm_free_physmem_slot(&new, &old);
 918 out:
 919         return r;
 920
 921 }
 922 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 923
 924 int kvm_set_memory_region(struct kvm *kvm,
 925                           struct kvm_userspace_memory_region *mem,
 926                           int user_alloc)
 927 {
 928         int r;
 929
 930         mutex_lock(&kvm->slots_lock);
 931         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 932         mutex_unlock(&kvm->slots_lock);
 933         return r;
 934 }
 935 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 936
 937 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 938                                    struct
 939                                    kvm_userspace_memory_region *mem,
 940                                    int user_alloc)
 941 {
 942         if (mem->slot >= KVM_MEMORY_SLOTS)
 943                 return -EINVAL;
 944         return kvm_set_memory_region(kvm, mem, user_alloc);
 945 }
 946
 947 int kvm_get_dirty_log(struct kvm *kvm,
 948                         struct kvm_dirty_log *log, int *is_dirty)
 949 {
 950         struct kvm_memory_slot *memslot;
 951         int r, i;
 952         unsigned long n;
 953         unsigned long any = 0;
 954
 955         r = -EINVAL;
 956         if (log->slot >= KVM_MEMORY_SLOTS)
 957                 goto out;
 958
 959         memslot = id_to_memslot(kvm->memslots, log->slot);
 960         r = -ENOENT;
 961         if (!memslot->dirty_bitmap)
 962                 goto out;
 963
 964         n = kvm_dirty_bitmap_bytes(memslot);
 965
 966         for (i = 0; !any && i < n/sizeof(long); ++i)
 967                 any = memslot->dirty_bitmap[i];
 968
 969         r = -EFAULT;
 970         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 971                 goto out;
 972
 973         if (any)
 974                 *is_dirty = 1;
 975
 976         r = 0;
 977 out:
 978         return r;
 979 }
 980
 981 void kvm_disable_largepages(void)
 982 {
 983         largepages_enabled = false;
 984 }
 985 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 986
 987 int is_error_page(struct page *page)
 988 {
 989         return page == bad_page || page == hwpoison_page || page == fault_page;
 990 }
 991 EXPORT_SYMBOL_GPL(is_error_page);
 992
 993 int is_error_pfn(pfn_t pfn)
 994 {
 995         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 996 }
 997 EXPORT_SYMBOL_GPL(is_error_pfn);
 998
 999 int is_hwpoison_pfn(pfn_t pfn)
1000 {
1001         return pfn == hwpoison_pfn;
1002 }
1003 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
1004
1005 int is_fault_pfn(pfn_t pfn)
1006 {
1007         return pfn == fault_pfn;
1008 }
1009 EXPORT_SYMBOL_GPL(is_fault_pfn);
1010
1011 int is_noslot_pfn(pfn_t pfn)
1012 {
1013         return pfn == bad_pfn;
1014 }
1015 EXPORT_SYMBOL_GPL(is_noslot_pfn);
1016
1017 int is_invalid_pfn(pfn_t pfn)
1018 {
1019         return pfn == hwpoison_pfn || pfn == fault_pfn;
1020 }
1021 EXPORT_SYMBOL_GPL(is_invalid_pfn);
1022
1023 static inline unsigned long bad_hva(void)
1024 {
1025         return PAGE_OFFSET;
1026 }
1027
1028 int kvm_is_error_hva(unsigned long addr)
1029 {
1030         return addr == bad_hva();
1031 }
1032 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1033
1034 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
1035                                                 gfn_t gfn)
1036 {
1037         return search_memslots(slots, gfn);
1038 }
1039
1040 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1041 {
1042         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1043 }
1044 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1045
1046 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1047 {
1048         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1049
1050         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
1051               memslot->flags & KVM_MEMSLOT_INVALID)
1052                 return 0;
1053
1054         return 1;
1055 }
1056 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1057
1058 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1059 {
1060         struct vm_area_struct *vma;
1061         unsigned long addr, size;
1062
1063         size = PAGE_SIZE;
1064
1065         addr = gfn_to_hva(kvm, gfn);
1066         if (kvm_is_error_hva(addr))
1067                 return PAGE_SIZE;
1068
1069         down_read(&current->mm->mmap_sem);
1070         vma = find_vma(current->mm, addr);
1071         if (!vma)
1072                 goto out;
1073
1074         size = vma_kernel_pagesize(vma);
1075
1076 out:
1077         up_read(&current->mm->mmap_sem);
1078
1079         return size;
1080 }
1081
1082 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1083                                      gfn_t *nr_pages)
1084 {
1085         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1086                 return bad_hva();
1087
1088         if (nr_pages)
1089                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1090
1091         return gfn_to_hva_memslot(slot, gfn);
1092 }
1093
1094 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1095 {
1096         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1097 }
1098 EXPORT_SYMBOL_GPL(gfn_to_hva);
1099
1100 static pfn_t get_fault_pfn(void)
1101 {
1102         get_page(fault_page);
1103         return fault_pfn;
1104 }
1105
1106 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1107         unsigned long start, int write, struct page **page)
1108 {
1109         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1110
1111         if (write)
1112                 flags |= FOLL_WRITE;
1113
1114         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1115 }
1116
1117 static inline int check_user_page_hwpoison(unsigned long addr)
1118 {
1119         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1120
1121         rc = __get_user_pages(current, current->mm, addr, 1,
1122                               flags, NULL, NULL, NULL);
1123         return rc == -EHWPOISON;
1124 }
1125
1126 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1127                         bool *async, bool write_fault, bool *writable)
1128 {
1129         struct page *page[1];
1130         int npages = 0;
1131         pfn_t pfn;
1132
1133         /* we can do it either atomically or asynchronously, not both */
1134         BUG_ON(atomic && async);
1135
1136         BUG_ON(!write_fault && !writable);
1137
1138         if (writable)
1139                 *writable = true;
1140
1141         if (atomic || async)
1142                 npages = __get_user_pages_fast(addr, 1, 1, page);
1143
1144         if (unlikely(npages != 1) && !atomic) {
1145                 might_sleep();
1146
1147                 if (writable)
1148                         *writable = write_fault;
1149
1150                 if (async) {
1151                         down_read(&current->mm->mmap_sem);
1152                         npages = get_user_page_nowait(current, current->mm,
1153                                                      addr, write_fault, page);
1154                         up_read(&current->mm->mmap_sem);
1155                 } else
1156                         npages = get_user_pages_fast(addr, 1, write_fault,
1157                                                      page);
1158
1159                 /* map read fault as writable if possible */
1160                 if (unlikely(!write_fault) && npages == 1) {
1161                         struct page *wpage[1];
1162
1163                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1164                         if (npages == 1) {
1165                                 *writable = true;
1166                                 put_page(page[0]);
1167                                 page[0] = wpage[0];
1168                         }
1169                         npages = 1;
1170                 }
1171         }
1172
1173         if (unlikely(npages != 1)) {
1174                 struct vm_area_struct *vma;
1175
1176                 if (atomic)
1177                         return get_fault_pfn();
1178
1179                 down_read(&current->mm->mmap_sem);
1180                 if (npages == -EHWPOISON ||
1181                         (!async && check_user_page_hwpoison(addr))) {
1182                         up_read(&current->mm->mmap_sem);
1183                         get_page(hwpoison_page);
1184                         return page_to_pfn(hwpoison_page);
1185                 }
1186
1187                 vma = find_vma_intersection(current->mm, addr, addr+1);
1188
1189                 if (vma == NULL)
1190                         pfn = get_fault_pfn();
1191                 else if ((vma->vm_flags & VM_PFNMAP)) {
1192                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1193                                 vma->vm_pgoff;
1194                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1195                 } else {
1196                         if (async && (vma->vm_flags & VM_WRITE))
1197                                 *async = true;
1198                         pfn = get_fault_pfn();
1199                 }
1200                 up_read(&current->mm->mmap_sem);
1201         } else
1202                 pfn = page_to_pfn(page[0]);
1203
1204         return pfn;
1205 }
1206
1207 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1208 {
1209         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1210 }
1211 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1212
1213 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1214                           bool write_fault, bool *writable)
1215 {
1216         unsigned long addr;
1217
1218         if (async)
1219                 *async = false;
1220
1221         addr = gfn_to_hva(kvm, gfn);
1222         if (kvm_is_error_hva(addr)) {
1223                 get_page(bad_page);
1224                 return page_to_pfn(bad_page);
1225         }
1226
1227         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1228 }
1229
1230 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1231 {
1232         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1233 }
1234 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1235
1236 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1237                        bool write_fault, bool *writable)
1238 {
1239         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1240 }
1241 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1242
1243 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1244 {
1245         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1246 }
1247 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1248
1249 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1250                       bool *writable)
1251 {
1252         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1253 }
1254 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1255
1256 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1257                          struct kvm_memory_slot *slot, gfn_t gfn)
1258 {
1259         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1260         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1261 }
1262
1263 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1264                                                                   int nr_pages)
1265 {
1266         unsigned long addr;
1267         gfn_t entry;
1268
1269         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1270         if (kvm_is_error_hva(addr))
1271                 return -1;
1272
1273         if (entry < nr_pages)
1274                 return 0;
1275
1276         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1277 }
1278 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1279
1280 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1281 {
1282         pfn_t pfn;
1283
1284         pfn = gfn_to_pfn(kvm, gfn);
1285         if (!kvm_is_mmio_pfn(pfn))
1286                 return pfn_to_page(pfn);
1287
1288         WARN_ON(kvm_is_mmio_pfn(pfn));
1289
1290         get_page(bad_page);
1291         return bad_page;
1292 }
1293
1294 EXPORT_SYMBOL_GPL(gfn_to_page);
1295
1296 void kvm_release_page_clean(struct page *page)
1297 {
1298         kvm_release_pfn_clean(page_to_pfn(page));
1299 }
1300 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1301
1302 void kvm_release_pfn_clean(pfn_t pfn)
1303 {
1304         if (!kvm_is_mmio_pfn(pfn))
1305                 put_page(pfn_to_page(pfn));
1306 }
1307 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1308
1309 void kvm_release_page_dirty(struct page *page)
1310 {
1311         kvm_release_pfn_dirty(page_to_pfn(page));
1312 }
1313 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1314
1315 void kvm_release_pfn_dirty(pfn_t pfn)
1316 {
1317         kvm_set_pfn_dirty(pfn);
1318         kvm_release_pfn_clean(pfn);
1319 }
1320 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1321
1322 void kvm_set_page_dirty(struct page *page)
1323 {
1324         kvm_set_pfn_dirty(page_to_pfn(page));
1325 }
1326 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1327
1328 void kvm_set_pfn_dirty(pfn_t pfn)
1329 {
1330         if (!kvm_is_mmio_pfn(pfn)) {
1331                 struct page *page = pfn_to_page(pfn);
1332                 if (!PageReserved(page))
1333                         SetPageDirty(page);
1334         }
1335 }
1336 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1337
1338 void kvm_set_pfn_accessed(pfn_t pfn)
1339 {
1340         if (!kvm_is_mmio_pfn(pfn))
1341                 mark_page_accessed(pfn_to_page(pfn));
1342 }
1343 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1344
1345 void kvm_get_pfn(pfn_t pfn)
1346 {
1347         if (!kvm_is_mmio_pfn(pfn))
1348                 get_page(pfn_to_page(pfn));
1349 }
1350 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1351
1352 static int next_segment(unsigned long len, int offset)
1353 {
1354         if (len > PAGE_SIZE - offset)
1355                 return PAGE_SIZE - offset;
1356         else
1357                 return len;
1358 }
1359
1360 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1361                         int len)
1362 {
1363         int r;
1364         unsigned long addr;
1365
1366         addr = gfn_to_hva(kvm, gfn);
1367         if (kvm_is_error_hva(addr))
1368                 return -EFAULT;
1369         r = __copy_from_user(data, (void __user *)addr + offset, len);
1370         if (r)
1371                 return -EFAULT;
1372         return 0;
1373 }
1374 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1375
1376 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1377 {
1378         gfn_t gfn = gpa >> PAGE_SHIFT;
1379         int seg;
1380         int offset = offset_in_page(gpa);
1381         int ret;
1382
1383         while ((seg = next_segment(len, offset)) != 0) {
1384                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1385                 if (ret < 0)
1386                         return ret;
1387                 offset = 0;
1388                 len -= seg;
1389                 data += seg;
1390                 ++gfn;
1391         }
1392         return 0;
1393 }
1394 EXPORT_SYMBOL_GPL(kvm_read_guest);
1395
1396 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1397                           unsigned long len)
1398 {
1399         int r;
1400         unsigned long addr;
1401         gfn_t gfn = gpa >> PAGE_SHIFT;
1402         int offset = offset_in_page(gpa);
1403
1404         addr = gfn_to_hva(kvm, gfn);
1405         if (kvm_is_error_hva(addr))
1406                 return -EFAULT;
1407         pagefault_disable();
1408         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1409         pagefault_enable();
1410         if (r)
1411                 return -EFAULT;
1412         return 0;
1413 }
1414 EXPORT_SYMBOL(kvm_read_guest_atomic);
1415
1416 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1417                          int offset, int len)
1418 {
1419         int r;
1420         unsigned long addr;
1421
1422         addr = gfn_to_hva(kvm, gfn);
1423         if (kvm_is_error_hva(addr))
1424                 return -EFAULT;
1425         r = __copy_to_user((void __user *)addr + offset, data, len);
1426         if (r)
1427                 return -EFAULT;
1428         mark_page_dirty(kvm, gfn);
1429         return 0;
1430 }
1431 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1432
1433 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1434                     unsigned long len)
1435 {
1436         gfn_t gfn = gpa >> PAGE_SHIFT;
1437         int seg;
1438         int offset = offset_in_page(gpa);
1439         int ret;
1440
1441         while ((seg = next_segment(len, offset)) != 0) {
1442                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1443                 if (ret < 0)
1444                         return ret;
1445                 offset = 0;
1446                 len -= seg;
1447                 data += seg;
1448                 ++gfn;
1449         }
1450         return 0;
1451 }
1452
1453 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1454                               gpa_t gpa)
1455 {
1456         struct kvm_memslots *slots = kvm_memslots(kvm);
1457         int offset = offset_in_page(gpa);
1458         gfn_t gfn = gpa >> PAGE_SHIFT;
1459
1460         ghc->gpa = gpa;
1461         ghc->generation = slots->generation;
1462         ghc->memslot = __gfn_to_memslot(slots, gfn);
1463         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1464         if (!kvm_is_error_hva(ghc->hva))
1465                 ghc->hva += offset;
1466         else
1467                 return -EFAULT;
1468
1469         return 0;
1470 }
1471 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1472
1473 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1474                            void *data, unsigned long len)
1475 {
1476         struct kvm_memslots *slots = kvm_memslots(kvm);
1477         int r;
1478
1479         if (slots->generation != ghc->generation)
1480                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1481
1482         if (kvm_is_error_hva(ghc->hva))
1483                 return -EFAULT;
1484
1485         r = __copy_to_user((void __user *)ghc->hva, data, len);
1486         if (r)
1487                 return -EFAULT;
1488         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1489
1490         return 0;
1491 }
1492 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1493
1494 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1495                            void *data, unsigned long len)
1496 {
1497         struct kvm_memslots *slots = kvm_memslots(kvm);
1498         int r;
1499
1500         if (slots->generation != ghc->generation)
1501                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1502
1503         if (kvm_is_error_hva(ghc->hva))
1504                 return -EFAULT;
1505
1506         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1507         if (r)
1508                 return -EFAULT;
1509
1510         return 0;
1511 }
1512 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1513
1514 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1515 {
1516         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1517                                     offset, len);
1518 }
1519 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1520
1521 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1522 {
1523         gfn_t gfn = gpa >> PAGE_SHIFT;
1524         int seg;
1525         int offset = offset_in_page(gpa);
1526         int ret;
1527
1528         while ((seg = next_segment(len, offset)) != 0) {
1529                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1530                 if (ret < 0)
1531                         return ret;
1532                 offset = 0;
1533                 len -= seg;
1534                 ++gfn;
1535         }
1536         return 0;
1537 }
1538 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1539
1540 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1541                              gfn_t gfn)
1542 {
1543         if (memslot && memslot->dirty_bitmap) {
1544                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1545
1546                 if (!test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap))
1547                         memslot->nr_dirty_pages++;
1548         }
1549 }
1550
1551 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1552 {
1553         struct kvm_memory_slot *memslot;
1554
1555         memslot = gfn_to_memslot(kvm, gfn);
1556         mark_page_dirty_in_slot(kvm, memslot, gfn);
1557 }
1558
1559 /*
1560  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1561  */
1562 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1563 {
1564         DEFINE_WAIT(wait);
1565
1566         for (;;) {
1567                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1568
1569                 if (kvm_arch_vcpu_runnable(vcpu)) {
1570                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1571                         break;
1572                 }
1573                 if (kvm_cpu_has_pending_timer(vcpu))
1574                         break;
1575                 if (signal_pending(current))
1576                         break;
1577
1578                 schedule();
1579         }
1580
1581         finish_wait(&vcpu->wq, &wait);
1582 }
1583
1584 void kvm_resched(struct kvm_vcpu *vcpu)
1585 {
1586         if (!need_resched())
1587                 return;
1588         cond_resched();
1589 }
1590 EXPORT_SYMBOL_GPL(kvm_resched);
1591
1592 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1593 {
1594         struct kvm *kvm = me->kvm;
1595         struct kvm_vcpu *vcpu;
1596         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1597         int yielded = 0;
1598         int pass;
1599         int i;
1600
1601         /*
1602          * We boost the priority of a VCPU that is runnable but not
1603          * currently running, because it got preempted by something
1604          * else and called schedule in __vcpu_run.  Hopefully that
1605          * VCPU is holding the lock that we need and will release it.
1606          * We approximate round-robin by starting at the last boosted VCPU.
1607          */
1608         for (pass = 0; pass < 2 && !yielded; pass++) {
1609                 kvm_for_each_vcpu(i, vcpu, kvm) {
1610                         struct task_struct *task = NULL;
1611                         struct pid *pid;
1612                         if (!pass && i < last_boosted_vcpu) {
1613                                 i = last_boosted_vcpu;
1614                                 continue;
1615                         } else if (pass && i > last_boosted_vcpu)
1616                                 break;
1617                         if (vcpu == me)
1618                                 continue;
1619                         if (waitqueue_active(&vcpu->wq))
1620                                 continue;
1621                         rcu_read_lock();
1622                         pid = rcu_dereference(vcpu->pid);
1623                         if (pid)
1624                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1625                         rcu_read_unlock();
1626                         if (!task)
1627                                 continue;
1628                         if (task->flags & PF_VCPU) {
1629                                 put_task_struct(task);
1630                                 continue;
1631                         }
1632                         if (yield_to(task, 1)) {
1633                                 put_task_struct(task);
1634                                 kvm->last_boosted_vcpu = i;
1635                                 yielded = 1;
1636                                 break;
1637                         }
1638                         put_task_struct(task);
1639                 }
1640         }
1641 }
1642 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1643
1644 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1645 {
1646         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1647         struct page *page;
1648
1649         if (vmf->pgoff == 0)
1650                 page = virt_to_page(vcpu->run);
1651 #ifdef CONFIG_X86
1652         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1653                 page = virt_to_page(vcpu->arch.pio_data);
1654 #endif
1655 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1656         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1657                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1658 #endif
1659         else
1660                 return VM_FAULT_SIGBUS;
1661         get_page(page);
1662         vmf->page = page;
1663         return 0;
1664 }
1665
1666 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1667         .fault = kvm_vcpu_fault,
1668 };
1669
1670 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1671 {
1672         vma->vm_ops = &kvm_vcpu_vm_ops;
1673         return 0;
1674 }
1675
1676 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1677 {
1678         struct kvm_vcpu *vcpu = filp->private_data;
1679
1680         kvm_put_kvm(vcpu->kvm);
1681         return 0;
1682 }
1683
1684 static struct file_operations kvm_vcpu_fops = {
1685         .release        = kvm_vcpu_release,
1686         .unlocked_ioctl = kvm_vcpu_ioctl,
1687 #ifdef CONFIG_COMPAT
1688         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1689 #endif
1690         .mmap           = kvm_vcpu_mmap,
1691         .llseek         = noop_llseek,
1692 };
1693
1694 /*
1695  * Allocates an inode for the vcpu.
1696  */
1697 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1698 {
1699         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1700 }
1701
1702 /*
1703  * Creates some virtual cpus.  Good luck creating more than one.
1704  */
1705 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1706 {
1707         int r;
1708         struct kvm_vcpu *vcpu, *v;
1709
1710         vcpu = kvm_arch_vcpu_create(kvm, id);
1711         if (IS_ERR(vcpu))
1712                 return PTR_ERR(vcpu);
1713
1714         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1715
1716         r = kvm_arch_vcpu_setup(vcpu);
1717         if (r)
1718                 goto vcpu_destroy;
1719
1720         mutex_lock(&kvm->lock);
1721         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1722                 r = -EINVAL;
1723                 goto unlock_vcpu_destroy;
1724         }
1725
1726         kvm_for_each_vcpu(r, v, kvm)
1727                 if (v->vcpu_id == id) {
1728                         r = -EEXIST;
1729                         goto unlock_vcpu_destroy;
1730                 }
1731
1732         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1733
1734         /* Now it's all set up, let userspace reach it */
1735         kvm_get_kvm(kvm);
1736         r = create_vcpu_fd(vcpu);
1737         if (r < 0) {
1738                 kvm_put_kvm(kvm);
1739                 goto unlock_vcpu_destroy;
1740         }
1741
1742         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1743         smp_wmb();
1744         atomic_inc(&kvm->online_vcpus);
1745
1746         mutex_unlock(&kvm->lock);
1747         return r;
1748
1749 unlock_vcpu_destroy:
1750         mutex_unlock(&kvm->lock);
1751 vcpu_destroy:
1752         kvm_arch_vcpu_destroy(vcpu);
1753         return r;
1754 }
1755
1756 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1757 {
1758         if (sigset) {
1759                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1760                 vcpu->sigset_active = 1;
1761                 vcpu->sigset = *sigset;
1762         } else
1763                 vcpu->sigset_active = 0;
1764         return 0;
1765 }
1766
1767 static long kvm_vcpu_ioctl(struct file *filp,
1768                            unsigned int ioctl, unsigned long arg)
1769 {
1770         struct kvm_vcpu *vcpu = filp->private_data;
1771         void __user *argp = (void __user *)arg;
1772         int r;
1773         struct kvm_fpu *fpu = NULL;
1774         struct kvm_sregs *kvm_sregs = NULL;
1775
1776         if (vcpu->kvm->mm != current->mm)
1777                 return -EIO;
1778
1779 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1780         /*
1781          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1782          * so vcpu_load() would break it.
1783          */
1784         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1785                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1786 #endif
1787
1788
1789         vcpu_load(vcpu);
1790         switch (ioctl) {
1791         case KVM_RUN:
1792                 r = -EINVAL;
1793                 if (arg)
1794                         goto out;
1795                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1796                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1797                 break;
1798         case KVM_GET_REGS: {
1799                 struct kvm_regs *kvm_regs;
1800
1801                 r = -ENOMEM;
1802                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1803                 if (!kvm_regs)
1804                         goto out;
1805                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1806                 if (r)
1807                         goto out_free1;
1808                 r = -EFAULT;
1809                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1810                         goto out_free1;
1811                 r = 0;
1812 out_free1:
1813                 kfree(kvm_regs);
1814                 break;
1815         }
1816         case KVM_SET_REGS: {
1817                 struct kvm_regs *kvm_regs;
1818
1819                 r = -ENOMEM;
1820                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1821                 if (IS_ERR(kvm_regs)) {
1822                         r = PTR_ERR(kvm_regs);
1823                         goto out;
1824                 }
1825                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1826                 if (r)
1827                         goto out_free2;
1828                 r = 0;
1829 out_free2:
1830                 kfree(kvm_regs);
1831                 break;
1832         }
1833         case KVM_GET_SREGS: {
1834                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1835                 r = -ENOMEM;
1836                 if (!kvm_sregs)
1837                         goto out;
1838                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1839                 if (r)
1840                         goto out;
1841                 r = -EFAULT;
1842                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1843                         goto out;
1844                 r = 0;
1845                 break;
1846         }
1847         case KVM_SET_SREGS: {
1848                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1849                 if (IS_ERR(kvm_sregs)) {
1850                         r = PTR_ERR(kvm_sregs);
1851                         goto out;
1852                 }
1853                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1854                 if (r)
1855                         goto out;
1856                 r = 0;
1857                 break;
1858         }
1859         case KVM_GET_MP_STATE: {
1860                 struct kvm_mp_state mp_state;
1861
1862                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1863                 if (r)
1864                         goto out;
1865                 r = -EFAULT;
1866                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1867                         goto out;
1868                 r = 0;
1869                 break;
1870         }
1871         case KVM_SET_MP_STATE: {
1872                 struct kvm_mp_state mp_state;
1873
1874                 r = -EFAULT;
1875                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1876                         goto out;
1877                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1878                 if (r)
1879                         goto out;
1880                 r = 0;
1881                 break;
1882         }
1883         case KVM_TRANSLATE: {
1884                 struct kvm_translation tr;
1885
1886                 r = -EFAULT;
1887                 if (copy_from_user(&tr, argp, sizeof tr))
1888                         goto out;
1889                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1890                 if (r)
1891                         goto out;
1892                 r = -EFAULT;
1893                 if (copy_to_user(argp, &tr, sizeof tr))
1894                         goto out;
1895                 r = 0;
1896                 break;
1897         }
1898         case KVM_SET_GUEST_DEBUG: {
1899                 struct kvm_guest_debug dbg;
1900
1901                 r = -EFAULT;
1902                 if (copy_from_user(&dbg, argp, sizeof dbg))
1903                         goto out;
1904                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1905                 if (r)
1906                         goto out;
1907                 r = 0;
1908                 break;
1909         }
1910         case KVM_SET_SIGNAL_MASK: {
1911                 struct kvm_signal_mask __user *sigmask_arg = argp;
1912                 struct kvm_signal_mask kvm_sigmask;
1913                 sigset_t sigset, *p;
1914
1915                 p = NULL;
1916                 if (argp) {
1917                         r = -EFAULT;
1918                         if (copy_from_user(&kvm_sigmask, argp,
1919                                            sizeof kvm_sigmask))
1920                                 goto out;
1921                         r = -EINVAL;
1922                         if (kvm_sigmask.len != sizeof sigset)
1923                                 goto out;
1924                         r = -EFAULT;
1925                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1926                                            sizeof sigset))
1927                                 goto out;
1928                         p = &sigset;
1929                 }
1930                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1931                 break;
1932         }
1933         case KVM_GET_FPU: {
1934                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1935                 r = -ENOMEM;
1936                 if (!fpu)
1937                         goto out;
1938                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1939                 if (r)
1940                         goto out;
1941                 r = -EFAULT;
1942                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1943                         goto out;
1944                 r = 0;
1945                 break;
1946         }
1947         case KVM_SET_FPU: {
1948                 fpu = memdup_user(argp, sizeof(*fpu));
1949                 if (IS_ERR(fpu)) {
1950                         r = PTR_ERR(fpu);
1951                         goto out;
1952                 }
1953                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1954                 if (r)
1955                         goto out;
1956                 r = 0;
1957                 break;
1958         }
1959         default:
1960                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1961         }
1962 out:
1963         vcpu_put(vcpu);
1964         kfree(fpu);
1965         kfree(kvm_sregs);
1966         return r;
1967 }
1968
1969 #ifdef CONFIG_COMPAT
1970 static long kvm_vcpu_compat_ioctl(struct file *filp,
1971                                   unsigned int ioctl, unsigned long arg)
1972 {
1973         struct kvm_vcpu *vcpu = filp->private_data;
1974         void __user *argp = compat_ptr(arg);
1975         int r;
1976
1977         if (vcpu->kvm->mm != current->mm)
1978                 return -EIO;
1979
1980         switch (ioctl) {
1981         case KVM_SET_SIGNAL_MASK: {
1982                 struct kvm_signal_mask __user *sigmask_arg = argp;
1983                 struct kvm_signal_mask kvm_sigmask;
1984                 compat_sigset_t csigset;
1985                 sigset_t sigset;
1986
1987                 if (argp) {
1988                         r = -EFAULT;
1989                         if (copy_from_user(&kvm_sigmask, argp,
1990                                            sizeof kvm_sigmask))
1991                                 goto out;
1992                         r = -EINVAL;
1993                         if (kvm_sigmask.len != sizeof csigset)
1994                                 goto out;
1995                         r = -EFAULT;
1996                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1997                                            sizeof csigset))
1998                                 goto out;
1999                 }
2000                 sigset_from_compat(&sigset, &csigset);
2001                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2002                 break;
2003         }
2004         default:
2005                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2006         }
2007
2008 out:
2009         return r;
2010 }
2011 #endif
2012
2013 static long kvm_vm_ioctl(struct file *filp,
2014                            unsigned int ioctl, unsigned long arg)
2015 {
2016         struct kvm *kvm = filp->private_data;
2017         void __user *argp = (void __user *)arg;
2018         int r;
2019
2020         if (kvm->mm != current->mm)
2021                 return -EIO;
2022         switch (ioctl) {
2023         case KVM_CREATE_VCPU:
2024                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2025                 if (r < 0)
2026                         goto out;
2027                 break;
2028         case KVM_SET_USER_MEMORY_REGION: {
2029                 struct kvm_userspace_memory_region kvm_userspace_mem;
2030
2031                 r = -EFAULT;
2032                 if (copy_from_user(&kvm_userspace_mem, argp,
2033                                                 sizeof kvm_userspace_mem))
2034                         goto out;
2035
2036                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2037                 if (r)
2038                         goto out;
2039                 break;
2040         }
2041         case KVM_GET_DIRTY_LOG: {
2042                 struct kvm_dirty_log log;
2043
2044                 r = -EFAULT;
2045                 if (copy_from_user(&log, argp, sizeof log))
2046                         goto out;
2047                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2048                 if (r)
2049                         goto out;
2050                 break;
2051         }
2052 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2053         case KVM_REGISTER_COALESCED_MMIO: {
2054                 struct kvm_coalesced_mmio_zone zone;
2055                 r = -EFAULT;
2056                 if (copy_from_user(&zone, argp, sizeof zone))
2057                         goto out;
2058                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2059                 if (r)
2060                         goto out;
2061                 r = 0;
2062                 break;
2063         }
2064         case KVM_UNREGISTER_COALESCED_MMIO: {
2065                 struct kvm_coalesced_mmio_zone zone;
2066                 r = -EFAULT;
2067                 if (copy_from_user(&zone, argp, sizeof zone))
2068                         goto out;
2069                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2070                 if (r)
2071                         goto out;
2072                 r = 0;
2073                 break;
2074         }
2075 #endif
2076         case KVM_IRQFD: {
2077                 struct kvm_irqfd data;
2078
2079                 r = -EFAULT;
2080                 if (copy_from_user(&data, argp, sizeof data))
2081                         goto out;
2082                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2083                 break;
2084         }
2085         case KVM_IOEVENTFD: {
2086                 struct kvm_ioeventfd data;
2087
2088                 r = -EFAULT;
2089                 if (copy_from_user(&data, argp, sizeof data))
2090                         goto out;
2091                 r = kvm_ioeventfd(kvm, &data);
2092                 break;
2093         }
2094 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2095         case KVM_SET_BOOT_CPU_ID:
2096                 r = 0;
2097                 mutex_lock(&kvm->lock);
2098                 if (atomic_read(&kvm->online_vcpus) != 0)
2099                         r = -EBUSY;
2100                 else
2101                         kvm->bsp_vcpu_id = arg;
2102                 mutex_unlock(&kvm->lock);
2103                 break;
2104 #endif
2105         default:
2106                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2107                 if (r == -ENOTTY)
2108                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2109         }
2110 out:
2111         return r;
2112 }
2113
2114 #ifdef CONFIG_COMPAT
2115 struct compat_kvm_dirty_log {
2116         __u32 slot;
2117         __u32 padding1;
2118         union {
2119                 compat_uptr_t dirty_bitmap; /* one bit per page */
2120                 __u64 padding2;
2121         };
2122 };
2123
2124 static long kvm_vm_compat_ioctl(struct file *filp,
2125                            unsigned int ioctl, unsigned long arg)
2126 {
2127         struct kvm *kvm = filp->private_data;
2128         int r;
2129
2130         if (kvm->mm != current->mm)
2131                 return -EIO;
2132         switch (ioctl) {
2133         case KVM_GET_DIRTY_LOG: {
2134                 struct compat_kvm_dirty_log compat_log;
2135                 struct kvm_dirty_log log;
2136
2137                 r = -EFAULT;
2138                 if (copy_from_user(&compat_log, (void __user *)arg,
2139                                    sizeof(compat_log)))
2140                         goto out;
2141                 log.slot         = compat_log.slot;
2142                 log.padding1     = compat_log.padding1;
2143                 log.padding2     = compat_log.padding2;
2144                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2145
2146                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2147                 if (r)
2148                         goto out;
2149                 break;
2150         }
2151         default:
2152                 r = kvm_vm_ioctl(filp, ioctl, arg);
2153         }
2154
2155 out:
2156         return r;
2157 }
2158 #endif
2159
2160 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2161 {
2162         struct page *page[1];
2163         unsigned long addr;
2164         int npages;
2165         gfn_t gfn = vmf->pgoff;
2166         struct kvm *kvm = vma->vm_file->private_data;
2167
2168         addr = gfn_to_hva(kvm, gfn);
2169         if (kvm_is_error_hva(addr))
2170                 return VM_FAULT_SIGBUS;
2171
2172         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2173                                 NULL);
2174         if (unlikely(npages != 1))
2175                 return VM_FAULT_SIGBUS;
2176
2177         vmf->page = page[0];
2178         return 0;
2179 }
2180
2181 static const struct vm_operations_struct kvm_vm_vm_ops = {
2182         .fault = kvm_vm_fault,
2183 };
2184
2185 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2186 {
2187         vma->vm_ops = &kvm_vm_vm_ops;
2188         return 0;
2189 }
2190
2191 static struct file_operations kvm_vm_fops = {
2192         .release        = kvm_vm_release,
2193         .unlocked_ioctl = kvm_vm_ioctl,
2194 #ifdef CONFIG_COMPAT
2195         .compat_ioctl   = kvm_vm_compat_ioctl,
2196 #endif
2197         .mmap           = kvm_vm_mmap,
2198         .llseek         = noop_llseek,
2199 };
2200
2201 static int kvm_dev_ioctl_create_vm(void)
2202 {
2203         int r;
2204         struct kvm *kvm;
2205
2206         kvm = kvm_create_vm();
2207         if (IS_ERR(kvm))
2208                 return PTR_ERR(kvm);
2209 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2210         r = kvm_coalesced_mmio_init(kvm);
2211         if (r < 0) {
2212                 kvm_put_kvm(kvm);
2213                 return r;
2214         }
2215 #endif
2216         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2217         if (r < 0)
2218                 kvm_put_kvm(kvm);
2219
2220         return r;
2221 }
2222
2223 static long kvm_dev_ioctl_check_extension_generic(long arg)
2224 {
2225         switch (arg) {
2226         case KVM_CAP_USER_MEMORY:
2227         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2228         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2229 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2230         case KVM_CAP_SET_BOOT_CPU_ID:
2231 #endif
2232         case KVM_CAP_INTERNAL_ERROR_DATA:
2233                 return 1;
2234 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2235         case KVM_CAP_IRQ_ROUTING:
2236                 return KVM_MAX_IRQ_ROUTES;
2237 #endif
2238         default:
2239                 break;
2240         }
2241         return kvm_dev_ioctl_check_extension(arg);
2242 }
2243
2244 static long kvm_dev_ioctl(struct file *filp,
2245                           unsigned int ioctl, unsigned long arg)
2246 {
2247         long r = -EINVAL;
2248
2249         switch (ioctl) {
2250         case KVM_GET_API_VERSION:
2251                 r = -EINVAL;
2252                 if (arg)
2253                         goto out;
2254                 r = KVM_API_VERSION;
2255                 break;
2256         case KVM_CREATE_VM:
2257                 r = -EINVAL;
2258                 if (arg)
2259                         goto out;
2260                 r = kvm_dev_ioctl_create_vm();
2261                 break;
2262         case KVM_CHECK_EXTENSION:
2263                 r = kvm_dev_ioctl_check_extension_generic(arg);
2264                 break;
2265         case KVM_GET_VCPU_MMAP_SIZE:
2266                 r = -EINVAL;
2267                 if (arg)
2268                         goto out;
2269                 r = PAGE_SIZE;     /* struct kvm_run */
2270 #ifdef CONFIG_X86
2271                 r += PAGE_SIZE;    /* pio data page */
2272 #endif
2273 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2274                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2275 #endif
2276                 break;
2277         case KVM_TRACE_ENABLE:
2278         case KVM_TRACE_PAUSE:
2279         case KVM_TRACE_DISABLE:
2280                 r = -EOPNOTSUPP;
2281                 break;
2282         default:
2283                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2284         }
2285 out:
2286         return r;
2287 }
2288
2289 static struct file_operations kvm_chardev_ops = {
2290         .unlocked_ioctl = kvm_dev_ioctl,
2291         .compat_ioctl   = kvm_dev_ioctl,
2292         .llseek         = noop_llseek,
2293 };
2294
2295 static struct miscdevice kvm_dev = {
2296         KVM_MINOR,
2297         "kvm",
2298         &kvm_chardev_ops,
2299 };
2300
2301 static void hardware_enable_nolock(void *junk)
2302 {
2303         int cpu = raw_smp_processor_id();
2304         int r;
2305
2306         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2307                 return;
2308
2309         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2310
2311         r = kvm_arch_hardware_enable(NULL);
2312
2313         if (r) {
2314                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2315                 atomic_inc(&hardware_enable_failed);
2316                 printk(KERN_INFO "kvm: enabling virtualization on "
2317                                  "CPU%d failed\n", cpu);
2318         }
2319 }
2320
2321 static void hardware_enable(void *junk)
2322 {
2323         raw_spin_lock(&kvm_lock);
2324         hardware_enable_nolock(junk);
2325         raw_spin_unlock(&kvm_lock);
2326 }
2327
2328 static void hardware_disable_nolock(void *junk)
2329 {
2330         int cpu = raw_smp_processor_id();
2331
2332         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2333                 return;
2334         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2335         kvm_arch_hardware_disable(NULL);
2336 }
2337
2338 static void hardware_disable(void *junk)
2339 {
2340         raw_spin_lock(&kvm_lock);
2341         hardware_disable_nolock(junk);
2342         raw_spin_unlock(&kvm_lock);
2343 }
2344
2345 static void hardware_disable_all_nolock(void)
2346 {
2347         BUG_ON(!kvm_usage_count);
2348
2349         kvm_usage_count--;
2350         if (!kvm_usage_count)
2351                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2352 }
2353
2354 static void hardware_disable_all(void)
2355 {
2356         raw_spin_lock(&kvm_lock);
2357         hardware_disable_all_nolock();
2358         raw_spin_unlock(&kvm_lock);
2359 }
2360
2361 static int hardware_enable_all(void)
2362 {
2363         int r = 0;
2364
2365         raw_spin_lock(&kvm_lock);
2366
2367         kvm_usage_count++;
2368         if (kvm_usage_count == 1) {
2369                 atomic_set(&hardware_enable_failed, 0);
2370                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2371
2372                 if (atomic_read(&hardware_enable_failed)) {
2373                         hardware_disable_all_nolock();
2374                         r = -EBUSY;
2375                 }
2376         }
2377
2378         raw_spin_unlock(&kvm_lock);
2379
2380         return r;
2381 }
2382
2383 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2384                            void *v)
2385 {
2386         int cpu = (long)v;
2387
2388         if (!kvm_usage_count)
2389                 return NOTIFY_OK;
2390
2391         val &= ~CPU_TASKS_FROZEN;
2392         switch (val) {
2393         case CPU_DYING:
2394                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2395                        cpu);
2396                 hardware_disable(NULL);
2397                 break;
2398         case CPU_STARTING:
2399                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2400                        cpu);
2401                 hardware_enable(NULL);
2402                 break;
2403         }
2404         return NOTIFY_OK;
2405 }
2406
2407
2408 asmlinkage void kvm_spurious_fault(void)
2409 {
2410         /* Fault while not rebooting.  We want the trace. */
2411         BUG();
2412 }
2413 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2414
2415 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2416                       void *v)
2417 {
2418         /*
2419          * Some (well, at least mine) BIOSes hang on reboot if
2420          * in vmx root mode.
2421          *
2422          * And Intel TXT required VMX off for all cpu when system shutdown.
2423          */
2424         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2425         kvm_rebooting = true;
2426         on_each_cpu(hardware_disable_nolock, NULL, 1);
2427         return NOTIFY_OK;
2428 }
2429
2430 static struct notifier_block kvm_reboot_notifier = {
2431         .notifier_call = kvm_reboot,
2432         .priority = 0,
2433 };
2434
2435 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2436 {
2437         int i;
2438
2439         for (i = 0; i < bus->dev_count; i++) {
2440                 struct kvm_io_device *pos = bus->range[i].dev;
2441
2442                 kvm_iodevice_destructor(pos);
2443         }
2444         kfree(bus);
2445 }
2446
2447 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2448 {
2449         const struct kvm_io_range *r1 = p1;
2450         const struct kvm_io_range *r2 = p2;
2451
2452         if (r1->addr < r2->addr)
2453                 return -1;
2454         if (r1->addr + r1->len > r2->addr + r2->len)
2455                 return 1;
2456         return 0;
2457 }
2458
2459 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2460                           gpa_t addr, int len)
2461 {
2462         if (bus->dev_count == NR_IOBUS_DEVS)
2463                 return -ENOSPC;
2464
2465         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2466                 .addr = addr,
2467                 .len = len,
2468                 .dev = dev,
2469         };
2470
2471         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2472                 kvm_io_bus_sort_cmp, NULL);
2473
2474         return 0;
2475 }
2476
2477 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2478                              gpa_t addr, int len)
2479 {
2480         struct kvm_io_range *range, key;
2481         int off;
2482
2483         key = (struct kvm_io_range) {
2484                 .addr = addr,
2485                 .len = len,
2486         };
2487
2488         range = bsearch(&key, bus->range, bus->dev_count,
2489                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2490         if (range == NULL)
2491                 return -ENOENT;
2492
2493         off = range - bus->range;
2494
2495         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2496                 off--;
2497
2498         return off;
2499 }
2500
2501 /* kvm_io_bus_write - called under kvm->slots_lock */
2502 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2503                      int len, const void *val)
2504 {
2505         int idx;
2506         struct kvm_io_bus *bus;
2507         struct kvm_io_range range;
2508
2509         range = (struct kvm_io_range) {
2510                 .addr = addr,
2511                 .len = len,
2512         };
2513
2514         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2515         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2516         if (idx < 0)
2517                 return -EOPNOTSUPP;
2518
2519         while (idx < bus->dev_count &&
2520                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2521                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2522                         return 0;
2523                 idx++;
2524         }
2525
2526         return -EOPNOTSUPP;
2527 }
2528
2529 /* kvm_io_bus_read - called under kvm->slots_lock */
2530 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2531                     int len, void *val)
2532 {
2533         int idx;
2534         struct kvm_io_bus *bus;
2535         struct kvm_io_range range;
2536
2537         range = (struct kvm_io_range) {
2538                 .addr = addr,
2539                 .len = len,
2540         };
2541
2542         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2543         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2544         if (idx < 0)
2545                 return -EOPNOTSUPP;
2546
2547         while (idx < bus->dev_count &&
2548                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2549                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2550                         return 0;
2551                 idx++;
2552         }
2553
2554         return -EOPNOTSUPP;
2555 }
2556
2557 /* Caller must hold slots_lock. */
2558 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2559                             int len, struct kvm_io_device *dev)
2560 {
2561         struct kvm_io_bus *new_bus, *bus;
2562
2563         bus = kvm->buses[bus_idx];
2564         if (bus->dev_count > NR_IOBUS_DEVS-1)
2565                 return -ENOSPC;
2566
2567         new_bus = kmemdup(bus, sizeof(struct kvm_io_bus), GFP_KERNEL);
2568         if (!new_bus)
2569                 return -ENOMEM;
2570         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2571         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2572         synchronize_srcu_expedited(&kvm->srcu);
2573         kfree(bus);
2574
2575         return 0;
2576 }
2577
2578 /* Caller must hold slots_lock. */
2579 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2580                               struct kvm_io_device *dev)
2581 {
2582         int i, r;
2583         struct kvm_io_bus *new_bus, *bus;
2584
2585         bus = kvm->buses[bus_idx];
2586
2587         new_bus = kmemdup(bus, sizeof(*bus), GFP_KERNEL);
2588         if (!new_bus)
2589                 return -ENOMEM;
2590
2591         r = -ENOENT;
2592         for (i = 0; i < new_bus->dev_count; i++)
2593                 if (new_bus->range[i].dev == dev) {
2594                         r = 0;
2595                         new_bus->dev_count--;
2596                         new_bus->range[i] = new_bus->range[new_bus->dev_count];
2597                         sort(new_bus->range, new_bus->dev_count,
2598                              sizeof(struct kvm_io_range),
2599                              kvm_io_bus_sort_cmp, NULL);
2600                         break;
2601                 }
2602
2603         if (r) {
2604                 kfree(new_bus);
2605                 return r;
2606         }
2607
2608         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2609         synchronize_srcu_expedited(&kvm->srcu);
2610         kfree(bus);
2611         return r;
2612 }
2613
2614 static struct notifier_block kvm_cpu_notifier = {
2615         .notifier_call = kvm_cpu_hotplug,
2616 };
2617
2618 static int vm_stat_get(void *_offset, u64 *val)
2619 {
2620         unsigned offset = (long)_offset;
2621         struct kvm *kvm;
2622
2623         *val = 0;
2624         raw_spin_lock(&kvm_lock);
2625         list_for_each_entry(kvm, &vm_list, vm_list)
2626                 *val += *(u32 *)((void *)kvm + offset);
2627         raw_spin_unlock(&kvm_lock);
2628         return 0;
2629 }
2630
2631 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2632
2633 static int vcpu_stat_get(void *_offset, u64 *val)
2634 {
2635         unsigned offset = (long)_offset;
2636         struct kvm *kvm;
2637         struct kvm_vcpu *vcpu;
2638         int i;
2639
2640         *val = 0;
2641         raw_spin_lock(&kvm_lock);
2642         list_for_each_entry(kvm, &vm_list, vm_list)
2643                 kvm_for_each_vcpu(i, vcpu, kvm)
2644                         *val += *(u32 *)((void *)vcpu + offset);
2645
2646         raw_spin_unlock(&kvm_lock);
2647         return 0;
2648 }
2649
2650 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2651
2652 static const struct file_operations *stat_fops[] = {
2653         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2654         [KVM_STAT_VM]   = &vm_stat_fops,
2655 };
2656
2657 static int kvm_init_debug(void)
2658 {
2659         int r = -EFAULT;
2660         struct kvm_stats_debugfs_item *p;
2661
2662         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2663         if (kvm_debugfs_dir == NULL)
2664                 goto out;
2665
2666         for (p = debugfs_entries; p->name; ++p) {
2667                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2668                                                 (void *)(long)p->offset,
2669                                                 stat_fops[p->kind]);
2670                 if (p->dentry == NULL)
2671                         goto out_dir;
2672         }
2673
2674         return 0;
2675
2676 out_dir:
2677         debugfs_remove_recursive(kvm_debugfs_dir);
2678 out:
2679         return r;
2680 }
2681
2682 static void kvm_exit_debug(void)
2683 {
2684         struct kvm_stats_debugfs_item *p;
2685
2686         for (p = debugfs_entries; p->name; ++p)
2687                 debugfs_remove(p->dentry);
2688         debugfs_remove(kvm_debugfs_dir);
2689 }
2690
2691 static int kvm_suspend(void)
2692 {
2693         if (kvm_usage_count)
2694                 hardware_disable_nolock(NULL);
2695         return 0;
2696 }
2697
2698 static void kvm_resume(void)
2699 {
2700         if (kvm_usage_count) {
2701                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2702                 hardware_enable_nolock(NULL);
2703         }
2704 }
2705
2706 static struct syscore_ops kvm_syscore_ops = {
2707         .suspend = kvm_suspend,
2708         .resume = kvm_resume,
2709 };
2710
2711 struct page *bad_page;
2712 pfn_t bad_pfn;
2713
2714 static inline
2715 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2716 {
2717         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2718 }
2719
2720 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2721 {
2722         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2723
2724         kvm_arch_vcpu_load(vcpu, cpu);
2725 }
2726
2727 static void kvm_sched_out(struct preempt_notifier *pn,
2728                           struct task_struct *next)
2729 {
2730         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2731
2732         kvm_arch_vcpu_put(vcpu);
2733 }
2734
2735 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2736                   struct module *module)
2737 {
2738         int r;
2739         int cpu;
2740
2741         r = kvm_arch_init(opaque);
2742         if (r)
2743                 goto out_fail;
2744
2745         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2746
2747         if (bad_page == NULL) {
2748                 r = -ENOMEM;
2749                 goto out;
2750         }
2751
2752         bad_pfn = page_to_pfn(bad_page);
2753
2754         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2755
2756         if (hwpoison_page == NULL) {
2757                 r = -ENOMEM;
2758                 goto out_free_0;
2759         }
2760
2761         hwpoison_pfn = page_to_pfn(hwpoison_page);
2762
2763         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2764
2765         if (fault_page == NULL) {
2766                 r = -ENOMEM;
2767                 goto out_free_0;
2768         }
2769
2770         fault_pfn = page_to_pfn(fault_page);
2771
2772         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2773                 r = -ENOMEM;
2774                 goto out_free_0;
2775         }
2776
2777         r = kvm_arch_hardware_setup();
2778         if (r < 0)
2779                 goto out_free_0a;
2780
2781         for_each_online_cpu(cpu) {
2782                 smp_call_function_single(cpu,
2783                                 kvm_arch_check_processor_compat,
2784                                 &r, 1);
2785                 if (r < 0)
2786                         goto out_free_1;
2787         }
2788
2789         r = register_cpu_notifier(&kvm_cpu_notifier);
2790         if (r)
2791                 goto out_free_2;
2792         register_reboot_notifier(&kvm_reboot_notifier);
2793
2794         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2795         if (!vcpu_align)
2796                 vcpu_align = __alignof__(struct kvm_vcpu);
2797         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2798                                            0, NULL);
2799         if (!kvm_vcpu_cache) {
2800                 r = -ENOMEM;
2801                 goto out_free_3;
2802         }
2803
2804         r = kvm_async_pf_init();
2805         if (r)
2806                 goto out_free;
2807
2808         kvm_chardev_ops.owner = module;
2809         kvm_vm_fops.owner = module;
2810         kvm_vcpu_fops.owner = module;
2811
2812         r = misc_register(&kvm_dev);
2813         if (r) {
2814                 printk(KERN_ERR "kvm: misc device register failed\n");
2815                 goto out_unreg;
2816         }
2817
2818         register_syscore_ops(&kvm_syscore_ops);
2819
2820         kvm_preempt_ops.sched_in = kvm_sched_in;
2821         kvm_preempt_ops.sched_out = kvm_sched_out;
2822
2823         r = kvm_init_debug();
2824         if (r) {
2825                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2826                 goto out_undebugfs;
2827         }
2828
2829         return 0;
2830
2831 out_undebugfs:
2832         unregister_syscore_ops(&kvm_syscore_ops);
2833 out_unreg:
2834         kvm_async_pf_deinit();
2835 out_free:
2836         kmem_cache_destroy(kvm_vcpu_cache);
2837 out_free_3:
2838         unregister_reboot_notifier(&kvm_reboot_notifier);
2839         unregister_cpu_notifier(&kvm_cpu_notifier);
2840 out_free_2:
2841 out_free_1:
2842         kvm_arch_hardware_unsetup();
2843 out_free_0a:
2844         free_cpumask_var(cpus_hardware_enabled);
2845 out_free_0:
2846         if (fault_page)
2847                 __free_page(fault_page);
2848         if (hwpoison_page)
2849                 __free_page(hwpoison_page);
2850         __free_page(bad_page);
2851 out:
2852         kvm_arch_exit();
2853 out_fail:
2854         return r;
2855 }
2856 EXPORT_SYMBOL_GPL(kvm_init);
2857
2858 void kvm_exit(void)
2859 {
2860         kvm_exit_debug();
2861         misc_deregister(&kvm_dev);
2862         kmem_cache_destroy(kvm_vcpu_cache);
2863         kvm_async_pf_deinit();
2864         unregister_syscore_ops(&kvm_syscore_ops);
2865         unregister_reboot_notifier(&kvm_reboot_notifier);
2866         unregister_cpu_notifier(&kvm_cpu_notifier);
2867         on_each_cpu(hardware_disable_nolock, NULL, 1);
2868         kvm_arch_hardware_unsetup();
2869         kvm_arch_exit();
2870         free_cpumask_var(cpus_hardware_enabled);
2871         __free_page(hwpoison_page);
2872         __free_page(bad_page);
2873 }
2874 EXPORT_SYMBOL_GPL(kvm_exit);