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  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "iodev.h"
  19
  20 #include <linux/kvm_host.h>
  21 #include <linux/kvm.h>
  22 #include <linux/module.h>
  23 #include <linux/errno.h>
  24 #include <linux/percpu.h>
  25 #include <linux/gfp.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/sysdev.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
  48 #include <asm/processor.h>
  49 #include <asm/io.h>
  50 #include <asm/uaccess.h>
  51 #include <asm/pgtable.h>
  52 #include <asm-generic/bitops/le.h>
  53
  54 #include "coalesced_mmio.h"
  55
  56 #define CREATE_TRACE_POINTS
  57 #include <trace/events/kvm.h>
  58
  59 MODULE_AUTHOR("Qumranet");
  60 MODULE_LICENSE("GPL");
  61
  62 /*
  63  * Ordering of locks:
  64  *
  65  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  66  */
  67
  68 DEFINE_SPINLOCK(kvm_lock);
  69 LIST_HEAD(vm_list);
  70
  71 static cpumask_var_t cpus_hardware_enabled;
  72 static int kvm_usage_count = 0;
  73 static atomic_t hardware_enable_failed;
  74
  75 struct kmem_cache *kvm_vcpu_cache;
  76 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  77
  78 static __read_mostly struct preempt_ops kvm_preempt_ops;
  79
  80 struct dentry *kvm_debugfs_dir;
  81
  82 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  83                            unsigned long arg);
  84 static int hardware_enable_all(void);
  85 static void hardware_disable_all(void);
  86
  87 static bool kvm_rebooting;
  88
  89 static bool largepages_enabled = true;
  90
  91 inline int kvm_is_mmio_pfn(pfn_t pfn)
  92 {
  93         if (pfn_valid(pfn)) {
  94                 struct page *page = compound_head(pfn_to_page(pfn));
  95                 return PageReserved(page);
  96         }
  97
  98         return true;
  99 }
 100
 101 /*
 102  * Switches to specified vcpu, until a matching vcpu_put()
 103  */
 104 void vcpu_load(struct kvm_vcpu *vcpu)
 105 {
 106         int cpu;
 107
 108         mutex_lock(&vcpu->mutex);
 109         cpu = get_cpu();
 110         preempt_notifier_register(&vcpu->preempt_notifier);
 111         kvm_arch_vcpu_load(vcpu, cpu);
 112         put_cpu();
 113 }
 114
 115 void vcpu_put(struct kvm_vcpu *vcpu)
 116 {
 117         preempt_disable();
 118         kvm_arch_vcpu_put(vcpu);
 119         preempt_notifier_unregister(&vcpu->preempt_notifier);
 120         preempt_enable();
 121         mutex_unlock(&vcpu->mutex);
 122 }
 123
 124 static void ack_flush(void *_completed)
 125 {
 126 }
 127
 128 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 129 {
 130         int i, cpu, me;
 131         cpumask_var_t cpus;
 132         bool called = true;
 133         struct kvm_vcpu *vcpu;
 134
 135         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 136
 137         spin_lock(&kvm->requests_lock);
 138         me = smp_processor_id();
 139         kvm_for_each_vcpu(i, vcpu, kvm) {
 140                 if (test_and_set_bit(req, &vcpu->requests))
 141                         continue;
 142                 cpu = vcpu->cpu;
 143                 if (cpus != NULL && cpu != -1 && cpu != me)
 144                         cpumask_set_cpu(cpu, cpus);
 145         }
 146         if (unlikely(cpus == NULL))
 147                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 148         else if (!cpumask_empty(cpus))
 149                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 150         else
 151                 called = false;
 152         spin_unlock(&kvm->requests_lock);
 153         free_cpumask_var(cpus);
 154         return called;
 155 }
 156
 157 void kvm_flush_remote_tlbs(struct kvm *kvm)
 158 {
 159         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 160                 ++kvm->stat.remote_tlb_flush;
 161 }
 162
 163 void kvm_reload_remote_mmus(struct kvm *kvm)
 164 {
 165         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 166 }
 167
 168 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 169 {
 170         struct page *page;
 171         int r;
 172
 173         mutex_init(&vcpu->mutex);
 174         vcpu->cpu = -1;
 175         vcpu->kvm = kvm;
 176         vcpu->vcpu_id = id;
 177         init_waitqueue_head(&vcpu->wq);
 178
 179         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 180         if (!page) {
 181                 r = -ENOMEM;
 182                 goto fail;
 183         }
 184         vcpu->run = page_address(page);
 185
 186         r = kvm_arch_vcpu_init(vcpu);
 187         if (r < 0)
 188                 goto fail_free_run;
 189         return 0;
 190
 191 fail_free_run:
 192         free_page((unsigned long)vcpu->run);
 193 fail:
 194         return r;
 195 }
 196 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 197
 198 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 199 {
 200         kvm_arch_vcpu_uninit(vcpu);
 201         free_page((unsigned long)vcpu->run);
 202 }
 203 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 204
 205 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 206 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 207 {
 208         return container_of(mn, struct kvm, mmu_notifier);
 209 }
 210
 211 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 212                                              struct mm_struct *mm,
 213                                              unsigned long address)
 214 {
 215         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 216         int need_tlb_flush;
 217
 218         /*
 219          * When ->invalidate_page runs, the linux pte has been zapped
 220          * already but the page is still allocated until
 221          * ->invalidate_page returns. So if we increase the sequence
 222          * here the kvm page fault will notice if the spte can't be
 223          * established because the page is going to be freed. If
 224          * instead the kvm page fault establishes the spte before
 225          * ->invalidate_page runs, kvm_unmap_hva will release it
 226          * before returning.
 227          *
 228          * The sequence increase only need to be seen at spin_unlock
 229          * time, and not at spin_lock time.
 230          *
 231          * Increasing the sequence after the spin_unlock would be
 232          * unsafe because the kvm page fault could then establish the
 233          * pte after kvm_unmap_hva returned, without noticing the page
 234          * is going to be freed.
 235          */
 236         spin_lock(&kvm->mmu_lock);
 237         kvm->mmu_notifier_seq++;
 238         need_tlb_flush = kvm_unmap_hva(kvm, address);
 239         spin_unlock(&kvm->mmu_lock);
 240
 241         /* we've to flush the tlb before the pages can be freed */
 242         if (need_tlb_flush)
 243                 kvm_flush_remote_tlbs(kvm);
 244
 245 }
 246
 247 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 248                                         struct mm_struct *mm,
 249                                         unsigned long address,
 250                                         pte_t pte)
 251 {
 252         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 253
 254         spin_lock(&kvm->mmu_lock);
 255         kvm->mmu_notifier_seq++;
 256         kvm_set_spte_hva(kvm, address, pte);
 257         spin_unlock(&kvm->mmu_lock);
 258 }
 259
 260 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 261                                                     struct mm_struct *mm,
 262                                                     unsigned long start,
 263                                                     unsigned long end)
 264 {
 265         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 266         int need_tlb_flush = 0;
 267
 268         spin_lock(&kvm->mmu_lock);
 269         /*
 270          * The count increase must become visible at unlock time as no
 271          * spte can be established without taking the mmu_lock and
 272          * count is also read inside the mmu_lock critical section.
 273          */
 274         kvm->mmu_notifier_count++;
 275         for (; start < end; start += PAGE_SIZE)
 276                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 277         spin_unlock(&kvm->mmu_lock);
 278
 279         /* we've to flush the tlb before the pages can be freed */
 280         if (need_tlb_flush)
 281                 kvm_flush_remote_tlbs(kvm);
 282 }
 283
 284 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 285                                                   struct mm_struct *mm,
 286                                                   unsigned long start,
 287                                                   unsigned long end)
 288 {
 289         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 290
 291         spin_lock(&kvm->mmu_lock);
 292         /*
 293          * This sequence increase will notify the kvm page fault that
 294          * the page that is going to be mapped in the spte could have
 295          * been freed.
 296          */
 297         kvm->mmu_notifier_seq++;
 298         /*
 299          * The above sequence increase must be visible before the
 300          * below count decrease but both values are read by the kvm
 301          * page fault under mmu_lock spinlock so we don't need to add
 302          * a smb_wmb() here in between the two.
 303          */
 304         kvm->mmu_notifier_count--;
 305         spin_unlock(&kvm->mmu_lock);
 306
 307         BUG_ON(kvm->mmu_notifier_count < 0);
 308 }
 309
 310 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 311                                               struct mm_struct *mm,
 312                                               unsigned long address)
 313 {
 314         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 315         int young;
 316
 317         spin_lock(&kvm->mmu_lock);
 318         young = kvm_age_hva(kvm, address);
 319         spin_unlock(&kvm->mmu_lock);
 320
 321         if (young)
 322                 kvm_flush_remote_tlbs(kvm);
 323
 324         return young;
 325 }
 326
 327 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 328                                      struct mm_struct *mm)
 329 {
 330         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 331         kvm_arch_flush_shadow(kvm);
 332 }
 333
 334 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 335         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 336         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 337         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 338         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 339         .change_pte             = kvm_mmu_notifier_change_pte,
 340         .release                = kvm_mmu_notifier_release,
 341 };
 342
 343 static int kvm_init_mmu_notifier(struct kvm *kvm)
 344 {
 345         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 346         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 347 }
 348
 349 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 350
 351 static int kvm_init_mmu_notifier(struct kvm *kvm)
 352 {
 353         return 0;
 354 }
 355
 356 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 357
 358 static struct kvm *kvm_create_vm(void)
 359 {
 360         int r = 0;
 361         struct kvm *kvm = kvm_arch_create_vm();
 362 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 363         struct page *page;
 364 #endif
 365
 366         if (IS_ERR(kvm))
 367                 goto out;
 368
 369         r = hardware_enable_all();
 370         if (r)
 371                 goto out_err_nodisable;
 372
 373 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 374         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 375         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 376 #endif
 377
 378         r = -ENOMEM;
 379         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 380         if (!kvm->memslots)
 381                 goto out_err;
 382
 383 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 384         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 385         if (!page)
 386                 goto out_err;
 387
 388         kvm->coalesced_mmio_ring =
 389                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 390 #endif
 391
 392         r = kvm_init_mmu_notifier(kvm);
 393         if (r) {
 394 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 395                 put_page(page);
 396 #endif
 397                 goto out_err;
 398         }
 399
 400         kvm->mm = current->mm;
 401         atomic_inc(&kvm->mm->mm_count);
 402         spin_lock_init(&kvm->mmu_lock);
 403         spin_lock_init(&kvm->requests_lock);
 404         kvm_io_bus_init(&kvm->pio_bus);
 405         kvm_eventfd_init(kvm);
 406         mutex_init(&kvm->lock);
 407         mutex_init(&kvm->irq_lock);
 408         kvm_io_bus_init(&kvm->mmio_bus);
 409         init_rwsem(&kvm->slots_lock);
 410         atomic_set(&kvm->users_count, 1);
 411         spin_lock(&kvm_lock);
 412         list_add(&kvm->vm_list, &vm_list);
 413         spin_unlock(&kvm_lock);
 414 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 415         kvm_coalesced_mmio_init(kvm);
 416 #endif
 417 out:
 418         return kvm;
 419
 420 out_err:
 421         hardware_disable_all();
 422 out_err_nodisable:
 423         kfree(kvm->memslots);
 424         kfree(kvm);
 425         return ERR_PTR(r);
 426 }
 427
 428 /*
 429  * Free any memory in @free but not in @dont.
 430  */
 431 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 432                                   struct kvm_memory_slot *dont)
 433 {
 434         int i;
 435
 436         if (!dont || free->rmap != dont->rmap)
 437                 vfree(free->rmap);
 438
 439         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 440                 vfree(free->dirty_bitmap);
 441
 442
 443         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 444                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 445                         vfree(free->lpage_info[i]);
 446                         free->lpage_info[i] = NULL;
 447                 }
 448         }
 449
 450         free->npages = 0;
 451         free->dirty_bitmap = NULL;
 452         free->rmap = NULL;
 453 }
 454
 455 void kvm_free_physmem(struct kvm *kvm)
 456 {
 457         int i;
 458         struct kvm_memslots *slots = kvm->memslots;
 459
 460         for (i = 0; i < slots->nmemslots; ++i)
 461                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 462
 463         kfree(kvm->memslots);
 464 }
 465
 466 static void kvm_destroy_vm(struct kvm *kvm)
 467 {
 468         struct mm_struct *mm = kvm->mm;
 469
 470         kvm_arch_sync_events(kvm);
 471         spin_lock(&kvm_lock);
 472         list_del(&kvm->vm_list);
 473         spin_unlock(&kvm_lock);
 474         kvm_free_irq_routing(kvm);
 475         kvm_io_bus_destroy(&kvm->pio_bus);
 476         kvm_io_bus_destroy(&kvm->mmio_bus);
 477         kvm_coalesced_mmio_free(kvm);
 478 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 479         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 480 #else
 481         kvm_arch_flush_shadow(kvm);
 482 #endif
 483         kvm_arch_destroy_vm(kvm);
 484         hardware_disable_all();
 485         mmdrop(mm);
 486 }
 487
 488 void kvm_get_kvm(struct kvm *kvm)
 489 {
 490         atomic_inc(&kvm->users_count);
 491 }
 492 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 493
 494 void kvm_put_kvm(struct kvm *kvm)
 495 {
 496         if (atomic_dec_and_test(&kvm->users_count))
 497                 kvm_destroy_vm(kvm);
 498 }
 499 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 500
 501
 502 static int kvm_vm_release(struct inode *inode, struct file *filp)
 503 {
 504         struct kvm *kvm = filp->private_data;
 505
 506         kvm_irqfd_release(kvm);
 507
 508         kvm_put_kvm(kvm);
 509         return 0;
 510 }
 511
 512 /*
 513  * Allocate some memory and give it an address in the guest physical address
 514  * space.
 515  *
 516  * Discontiguous memory is allowed, mostly for framebuffers.
 517  *
 518  * Must be called holding mmap_sem for write.
 519  */
 520 int __kvm_set_memory_region(struct kvm *kvm,
 521                             struct kvm_userspace_memory_region *mem,
 522                             int user_alloc)
 523 {
 524         int r;
 525         gfn_t base_gfn;
 526         unsigned long npages;
 527         unsigned long i;
 528         struct kvm_memory_slot *memslot;
 529         struct kvm_memory_slot old, new;
 530
 531         r = -EINVAL;
 532         /* General sanity checks */
 533         if (mem->memory_size & (PAGE_SIZE - 1))
 534                 goto out;
 535         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 536                 goto out;
 537         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 538                 goto out;
 539         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 540                 goto out;
 541         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 542                 goto out;
 543
 544         memslot = &kvm->memslots->memslots[mem->slot];
 545         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 546         npages = mem->memory_size >> PAGE_SHIFT;
 547
 548         if (!npages)
 549                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 550
 551         new = old = *memslot;
 552
 553         new.base_gfn = base_gfn;
 554         new.npages = npages;
 555         new.flags = mem->flags;
 556
 557         /* Disallow changing a memory slot's size. */
 558         r = -EINVAL;
 559         if (npages && old.npages && npages != old.npages)
 560                 goto out_free;
 561
 562         /* Check for overlaps */
 563         r = -EEXIST;
 564         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 565                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 566
 567                 if (s == memslot || !s->npages)
 568                         continue;
 569                 if (!((base_gfn + npages <= s->base_gfn) ||
 570                       (base_gfn >= s->base_gfn + s->npages)))
 571                         goto out_free;
 572         }
 573
 574         /* Free page dirty bitmap if unneeded */
 575         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 576                 new.dirty_bitmap = NULL;
 577
 578         r = -ENOMEM;
 579
 580         /* Allocate if a slot is being created */
 581 #ifndef CONFIG_S390
 582         if (npages && !new.rmap) {
 583                 new.rmap = vmalloc(npages * sizeof(struct page *));
 584
 585                 if (!new.rmap)
 586                         goto out_free;
 587
 588                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 589
 590                 new.user_alloc = user_alloc;
 591                 /*
 592                  * hva_to_rmmap() serialzies with the mmu_lock and to be
 593                  * safe it has to ignore memslots with !user_alloc &&
 594                  * !userspace_addr.
 595                  */
 596                 if (user_alloc)
 597                         new.userspace_addr = mem->userspace_addr;
 598                 else
 599                         new.userspace_addr = 0;
 600         }
 601         if (!npages)
 602                 goto skip_lpage;
 603
 604         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 605                 unsigned long ugfn;
 606                 unsigned long j;
 607                 int lpages;
 608                 int level = i + 2;
 609
 610                 /* Avoid unused variable warning if no large pages */
 611                 (void)level;
 612
 613                 if (new.lpage_info[i])
 614                         continue;
 615
 616                 lpages = 1 + (base_gfn + npages - 1) /
 617                              KVM_PAGES_PER_HPAGE(level);
 618                 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
 619
 620                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
 621
 622                 if (!new.lpage_info[i])
 623                         goto out_free;
 624
 625                 memset(new.lpage_info[i], 0,
 626                        lpages * sizeof(*new.lpage_info[i]));
 627
 628                 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
 629                         new.lpage_info[i][0].write_count = 1;
 630                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
 631                         new.lpage_info[i][lpages - 1].write_count = 1;
 632                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 633                 /*
 634                  * If the gfn and userspace address are not aligned wrt each
 635                  * other, or if explicitly asked to, disable large page
 636                  * support for this slot
 637                  */
 638                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 639                     !largepages_enabled)
 640                         for (j = 0; j < lpages; ++j)
 641                                 new.lpage_info[i][j].write_count = 1;
 642         }
 643
 644 skip_lpage:
 645
 646         /* Allocate page dirty bitmap if needed */
 647         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 648                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 649
 650                 new.dirty_bitmap = vmalloc(dirty_bytes);
 651                 if (!new.dirty_bitmap)
 652                         goto out_free;
 653                 memset(new.dirty_bitmap, 0, dirty_bytes);
 654                 if (old.npages)
 655                         kvm_arch_flush_shadow(kvm);
 656         }
 657 #else  /* not defined CONFIG_S390 */
 658         new.user_alloc = user_alloc;
 659         if (user_alloc)
 660                 new.userspace_addr = mem->userspace_addr;
 661 #endif /* not defined CONFIG_S390 */
 662
 663         if (!npages)
 664                 kvm_arch_flush_shadow(kvm);
 665
 666         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 667         if (r)
 668                 goto out_free;
 669
 670         spin_lock(&kvm->mmu_lock);
 671         if (mem->slot >= kvm->memslots->nmemslots)
 672                 kvm->memslots->nmemslots = mem->slot + 1;
 673
 674         *memslot = new;
 675         spin_unlock(&kvm->mmu_lock);
 676
 677         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 678
 679         kvm_free_physmem_slot(&old, npages ? &new : NULL);
 680         /* Slot deletion case: we have to update the current slot */
 681         spin_lock(&kvm->mmu_lock);
 682         if (!npages)
 683                 *memslot = old;
 684         spin_unlock(&kvm->mmu_lock);
 685 #ifdef CONFIG_DMAR
 686         /* map the pages in iommu page table */
 687         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
 688         if (r)
 689                 goto out;
 690 #endif
 691         return 0;
 692
 693 out_free:
 694         kvm_free_physmem_slot(&new, &old);
 695 out:
 696         return r;
 697
 698 }
 699 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 700
 701 int kvm_set_memory_region(struct kvm *kvm,
 702                           struct kvm_userspace_memory_region *mem,
 703                           int user_alloc)
 704 {
 705         int r;
 706
 707         down_write(&kvm->slots_lock);
 708         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 709         up_write(&kvm->slots_lock);
 710         return r;
 711 }
 712 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 713
 714 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 715                                    struct
 716                                    kvm_userspace_memory_region *mem,
 717                                    int user_alloc)
 718 {
 719         if (mem->slot >= KVM_MEMORY_SLOTS)
 720                 return -EINVAL;
 721         return kvm_set_memory_region(kvm, mem, user_alloc);
 722 }
 723
 724 int kvm_get_dirty_log(struct kvm *kvm,
 725                         struct kvm_dirty_log *log, int *is_dirty)
 726 {
 727         struct kvm_memory_slot *memslot;
 728         int r, i;
 729         int n;
 730         unsigned long any = 0;
 731
 732         r = -EINVAL;
 733         if (log->slot >= KVM_MEMORY_SLOTS)
 734                 goto out;
 735
 736         memslot = &kvm->memslots->memslots[log->slot];
 737         r = -ENOENT;
 738         if (!memslot->dirty_bitmap)
 739                 goto out;
 740
 741         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 742
 743         for (i = 0; !any && i < n/sizeof(long); ++i)
 744                 any = memslot->dirty_bitmap[i];
 745
 746         r = -EFAULT;
 747         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 748                 goto out;
 749
 750         if (any)
 751                 *is_dirty = 1;
 752
 753         r = 0;
 754 out:
 755         return r;
 756 }
 757
 758 void kvm_disable_largepages(void)
 759 {
 760         largepages_enabled = false;
 761 }
 762 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 763
 764 int is_error_page(struct page *page)
 765 {
 766         return page == bad_page;
 767 }
 768 EXPORT_SYMBOL_GPL(is_error_page);
 769
 770 int is_error_pfn(pfn_t pfn)
 771 {
 772         return pfn == bad_pfn;
 773 }
 774 EXPORT_SYMBOL_GPL(is_error_pfn);
 775
 776 static inline unsigned long bad_hva(void)
 777 {
 778         return PAGE_OFFSET;
 779 }
 780
 781 int kvm_is_error_hva(unsigned long addr)
 782 {
 783         return addr == bad_hva();
 784 }
 785 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 786
 787 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
 788 {
 789         int i;
 790         struct kvm_memslots *slots = kvm->memslots;
 791
 792         for (i = 0; i < slots->nmemslots; ++i) {
 793                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 794
 795                 if (gfn >= memslot->base_gfn
 796                     && gfn < memslot->base_gfn + memslot->npages)
 797                         return memslot;
 798         }
 799         return NULL;
 800 }
 801 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
 802
 803 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 804 {
 805         gfn = unalias_gfn(kvm, gfn);
 806         return gfn_to_memslot_unaliased(kvm, gfn);
 807 }
 808
 809 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 810 {
 811         int i;
 812         struct kvm_memslots *slots = kvm->memslots;
 813
 814         gfn = unalias_gfn(kvm, gfn);
 815         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 816                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 817
 818                 if (gfn >= memslot->base_gfn
 819                     && gfn < memslot->base_gfn + memslot->npages)
 820                         return 1;
 821         }
 822         return 0;
 823 }
 824 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 825
 826 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 827 {
 828         struct kvm_memory_slot *slot;
 829
 830         gfn = unalias_gfn(kvm, gfn);
 831         slot = gfn_to_memslot_unaliased(kvm, gfn);
 832         if (!slot)
 833                 return bad_hva();
 834         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 835 }
 836 EXPORT_SYMBOL_GPL(gfn_to_hva);
 837
 838 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 839 {
 840         struct page *page[1];
 841         unsigned long addr;
 842         int npages;
 843         pfn_t pfn;
 844
 845         might_sleep();
 846
 847         addr = gfn_to_hva(kvm, gfn);
 848         if (kvm_is_error_hva(addr)) {
 849                 get_page(bad_page);
 850                 return page_to_pfn(bad_page);
 851         }
 852
 853         npages = get_user_pages_fast(addr, 1, 1, page);
 854
 855         if (unlikely(npages != 1)) {
 856                 struct vm_area_struct *vma;
 857
 858                 down_read(&current->mm->mmap_sem);
 859                 vma = find_vma(current->mm, addr);
 860
 861                 if (vma == NULL || addr < vma->vm_start ||
 862                     !(vma->vm_flags & VM_PFNMAP)) {
 863                         up_read(&current->mm->mmap_sem);
 864                         get_page(bad_page);
 865                         return page_to_pfn(bad_page);
 866                 }
 867
 868                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 869                 up_read(&current->mm->mmap_sem);
 870                 BUG_ON(!kvm_is_mmio_pfn(pfn));
 871         } else
 872                 pfn = page_to_pfn(page[0]);
 873
 874         return pfn;
 875 }
 876
 877 EXPORT_SYMBOL_GPL(gfn_to_pfn);
 878
 879 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 880 {
 881         pfn_t pfn;
 882
 883         pfn = gfn_to_pfn(kvm, gfn);
 884         if (!kvm_is_mmio_pfn(pfn))
 885                 return pfn_to_page(pfn);
 886
 887         WARN_ON(kvm_is_mmio_pfn(pfn));
 888
 889         get_page(bad_page);
 890         return bad_page;
 891 }
 892
 893 EXPORT_SYMBOL_GPL(gfn_to_page);
 894
 895 void kvm_release_page_clean(struct page *page)
 896 {
 897         kvm_release_pfn_clean(page_to_pfn(page));
 898 }
 899 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
 900
 901 void kvm_release_pfn_clean(pfn_t pfn)
 902 {
 903         if (!kvm_is_mmio_pfn(pfn))
 904                 put_page(pfn_to_page(pfn));
 905 }
 906 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
 907
 908 void kvm_release_page_dirty(struct page *page)
 909 {
 910         kvm_release_pfn_dirty(page_to_pfn(page));
 911 }
 912 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
 913
 914 void kvm_release_pfn_dirty(pfn_t pfn)
 915 {
 916         kvm_set_pfn_dirty(pfn);
 917         kvm_release_pfn_clean(pfn);
 918 }
 919 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
 920
 921 void kvm_set_page_dirty(struct page *page)
 922 {
 923         kvm_set_pfn_dirty(page_to_pfn(page));
 924 }
 925 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
 926
 927 void kvm_set_pfn_dirty(pfn_t pfn)
 928 {
 929         if (!kvm_is_mmio_pfn(pfn)) {
 930                 struct page *page = pfn_to_page(pfn);
 931                 if (!PageReserved(page))
 932                         SetPageDirty(page);
 933         }
 934 }
 935 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
 936
 937 void kvm_set_pfn_accessed(pfn_t pfn)
 938 {
 939         if (!kvm_is_mmio_pfn(pfn))
 940                 mark_page_accessed(pfn_to_page(pfn));
 941 }
 942 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
 943
 944 void kvm_get_pfn(pfn_t pfn)
 945 {
 946         if (!kvm_is_mmio_pfn(pfn))
 947                 get_page(pfn_to_page(pfn));
 948 }
 949 EXPORT_SYMBOL_GPL(kvm_get_pfn);
 950
 951 static int next_segment(unsigned long len, int offset)
 952 {
 953         if (len > PAGE_SIZE - offset)
 954                 return PAGE_SIZE - offset;
 955         else
 956                 return len;
 957 }
 958
 959 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 960                         int len)
 961 {
 962         int r;
 963         unsigned long addr;
 964
 965         addr = gfn_to_hva(kvm, gfn);
 966         if (kvm_is_error_hva(addr))
 967                 return -EFAULT;
 968         r = copy_from_user(data, (void __user *)addr + offset, len);
 969         if (r)
 970                 return -EFAULT;
 971         return 0;
 972 }
 973 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 974
 975 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 976 {
 977         gfn_t gfn = gpa >> PAGE_SHIFT;
 978         int seg;
 979         int offset = offset_in_page(gpa);
 980         int ret;
 981
 982         while ((seg = next_segment(len, offset)) != 0) {
 983                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 984                 if (ret < 0)
 985                         return ret;
 986                 offset = 0;
 987                 len -= seg;
 988                 data += seg;
 989                 ++gfn;
 990         }
 991         return 0;
 992 }
 993 EXPORT_SYMBOL_GPL(kvm_read_guest);
 994
 995 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
 996                           unsigned long len)
 997 {
 998         int r;
 999         unsigned long addr;
1000         gfn_t gfn = gpa >> PAGE_SHIFT;
1001         int offset = offset_in_page(gpa);
1002
1003         addr = gfn_to_hva(kvm, gfn);
1004         if (kvm_is_error_hva(addr))
1005                 return -EFAULT;
1006         pagefault_disable();
1007         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1008         pagefault_enable();
1009         if (r)
1010                 return -EFAULT;
1011         return 0;
1012 }
1013 EXPORT_SYMBOL(kvm_read_guest_atomic);
1014
1015 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1016                          int offset, int len)
1017 {
1018         int r;
1019         unsigned long addr;
1020
1021         addr = gfn_to_hva(kvm, gfn);
1022         if (kvm_is_error_hva(addr))
1023                 return -EFAULT;
1024         r = copy_to_user((void __user *)addr + offset, data, len);
1025         if (r)
1026                 return -EFAULT;
1027         mark_page_dirty(kvm, gfn);
1028         return 0;
1029 }
1030 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1031
1032 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1033                     unsigned long len)
1034 {
1035         gfn_t gfn = gpa >> PAGE_SHIFT;
1036         int seg;
1037         int offset = offset_in_page(gpa);
1038         int ret;
1039
1040         while ((seg = next_segment(len, offset)) != 0) {
1041                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1042                 if (ret < 0)
1043                         return ret;
1044                 offset = 0;
1045                 len -= seg;
1046                 data += seg;
1047                 ++gfn;
1048         }
1049         return 0;
1050 }
1051
1052 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1053 {
1054         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1055 }
1056 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1057
1058 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1059 {
1060         gfn_t gfn = gpa >> PAGE_SHIFT;
1061         int seg;
1062         int offset = offset_in_page(gpa);
1063         int ret;
1064
1065         while ((seg = next_segment(len, offset)) != 0) {
1066                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1067                 if (ret < 0)
1068                         return ret;
1069                 offset = 0;
1070                 len -= seg;
1071                 ++gfn;
1072         }
1073         return 0;
1074 }
1075 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1076
1077 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1078 {
1079         struct kvm_memory_slot *memslot;
1080
1081         gfn = unalias_gfn(kvm, gfn);
1082         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1083         if (memslot && memslot->dirty_bitmap) {
1084                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1085
1086                 /* avoid RMW */
1087                 if (!generic_test_le_bit(rel_gfn, memslot->dirty_bitmap))
1088                         generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1089         }
1090 }
1091
1092 /*
1093  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1094  */
1095 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1096 {
1097         DEFINE_WAIT(wait);
1098
1099         for (;;) {
1100                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1101
1102                 if (kvm_arch_vcpu_runnable(vcpu)) {
1103                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1104                         break;
1105                 }
1106                 if (kvm_cpu_has_pending_timer(vcpu))
1107                         break;
1108                 if (signal_pending(current))
1109                         break;
1110
1111                 schedule();
1112         }
1113
1114         finish_wait(&vcpu->wq, &wait);
1115 }
1116
1117 void kvm_resched(struct kvm_vcpu *vcpu)
1118 {
1119         if (!need_resched())
1120                 return;
1121         cond_resched();
1122 }
1123 EXPORT_SYMBOL_GPL(kvm_resched);
1124
1125 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1126 {
1127         ktime_t expires;
1128         DEFINE_WAIT(wait);
1129
1130         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1131
1132         /* Sleep for 100 us, and hope lock-holder got scheduled */
1133         expires = ktime_add_ns(ktime_get(), 100000UL);
1134         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1135
1136         finish_wait(&vcpu->wq, &wait);
1137 }
1138 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1139
1140 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1141 {
1142         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1143         struct page *page;
1144
1145         if (vmf->pgoff == 0)
1146                 page = virt_to_page(vcpu->run);
1147 #ifdef CONFIG_X86
1148         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1149                 page = virt_to_page(vcpu->arch.pio_data);
1150 #endif
1151 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1152         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1153                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1154 #endif
1155         else
1156                 return VM_FAULT_SIGBUS;
1157         get_page(page);
1158         vmf->page = page;
1159         return 0;
1160 }
1161
1162 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1163         .fault = kvm_vcpu_fault,
1164 };
1165
1166 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1167 {
1168         vma->vm_ops = &kvm_vcpu_vm_ops;
1169         return 0;
1170 }
1171
1172 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1173 {
1174         struct kvm_vcpu *vcpu = filp->private_data;
1175
1176         kvm_put_kvm(vcpu->kvm);
1177         return 0;
1178 }
1179
1180 static struct file_operations kvm_vcpu_fops = {
1181         .release        = kvm_vcpu_release,
1182         .unlocked_ioctl = kvm_vcpu_ioctl,
1183         .compat_ioctl   = kvm_vcpu_ioctl,
1184         .mmap           = kvm_vcpu_mmap,
1185 };
1186
1187 /*
1188  * Allocates an inode for the vcpu.
1189  */
1190 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1191 {
1192         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1193 }
1194
1195 /*
1196  * Creates some virtual cpus.  Good luck creating more than one.
1197  */
1198 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1199 {
1200         int r;
1201         struct kvm_vcpu *vcpu, *v;
1202
1203         vcpu = kvm_arch_vcpu_create(kvm, id);
1204         if (IS_ERR(vcpu))
1205                 return PTR_ERR(vcpu);
1206
1207         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1208
1209         r = kvm_arch_vcpu_setup(vcpu);
1210         if (r)
1211                 return r;
1212
1213         mutex_lock(&kvm->lock);
1214         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1215                 r = -EINVAL;
1216                 goto vcpu_destroy;
1217         }
1218
1219         kvm_for_each_vcpu(r, v, kvm)
1220                 if (v->vcpu_id == id) {
1221                         r = -EEXIST;
1222                         goto vcpu_destroy;
1223                 }
1224
1225         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1226
1227         /* Now it's all set up, let userspace reach it */
1228         kvm_get_kvm(kvm);
1229         r = create_vcpu_fd(vcpu);
1230         if (r < 0) {
1231                 kvm_put_kvm(kvm);
1232                 goto vcpu_destroy;
1233         }
1234
1235         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1236         smp_wmb();
1237         atomic_inc(&kvm->online_vcpus);
1238
1239 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1240         if (kvm->bsp_vcpu_id == id)
1241                 kvm->bsp_vcpu = vcpu;
1242 #endif
1243         mutex_unlock(&kvm->lock);
1244         return r;
1245
1246 vcpu_destroy:
1247         mutex_unlock(&kvm->lock);
1248         kvm_arch_vcpu_destroy(vcpu);
1249         return r;
1250 }
1251
1252 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1253 {
1254         if (sigset) {
1255                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1256                 vcpu->sigset_active = 1;
1257                 vcpu->sigset = *sigset;
1258         } else
1259                 vcpu->sigset_active = 0;
1260         return 0;
1261 }
1262
1263 static long kvm_vcpu_ioctl(struct file *filp,
1264                            unsigned int ioctl, unsigned long arg)
1265 {
1266         struct kvm_vcpu *vcpu = filp->private_data;
1267         void __user *argp = (void __user *)arg;
1268         int r;
1269         struct kvm_fpu *fpu = NULL;
1270         struct kvm_sregs *kvm_sregs = NULL;
1271
1272         if (vcpu->kvm->mm != current->mm)
1273                 return -EIO;
1274         switch (ioctl) {
1275         case KVM_RUN:
1276                 r = -EINVAL;
1277                 if (arg)
1278                         goto out;
1279                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1280                 break;
1281         case KVM_GET_REGS: {
1282                 struct kvm_regs *kvm_regs;
1283
1284                 r = -ENOMEM;
1285                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1286                 if (!kvm_regs)
1287                         goto out;
1288                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1289                 if (r)
1290                         goto out_free1;
1291                 r = -EFAULT;
1292                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1293                         goto out_free1;
1294                 r = 0;
1295 out_free1:
1296                 kfree(kvm_regs);
1297                 break;
1298         }
1299         case KVM_SET_REGS: {
1300                 struct kvm_regs *kvm_regs;
1301
1302                 r = -ENOMEM;
1303                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1304                 if (!kvm_regs)
1305                         goto out;
1306                 r = -EFAULT;
1307                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1308                         goto out_free2;
1309                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1310                 if (r)
1311                         goto out_free2;
1312                 r = 0;
1313 out_free2:
1314                 kfree(kvm_regs);
1315                 break;
1316         }
1317         case KVM_GET_SREGS: {
1318                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1319                 r = -ENOMEM;
1320                 if (!kvm_sregs)
1321                         goto out;
1322                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1323                 if (r)
1324                         goto out;
1325                 r = -EFAULT;
1326                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1327                         goto out;
1328                 r = 0;
1329                 break;
1330         }
1331         case KVM_SET_SREGS: {
1332                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1333                 r = -ENOMEM;
1334                 if (!kvm_sregs)
1335                         goto out;
1336                 r = -EFAULT;
1337                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1338                         goto out;
1339                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1340                 if (r)
1341                         goto out;
1342                 r = 0;
1343                 break;
1344         }
1345         case KVM_GET_MP_STATE: {
1346                 struct kvm_mp_state mp_state;
1347
1348                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1349                 if (r)
1350                         goto out;
1351                 r = -EFAULT;
1352                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1353                         goto out;
1354                 r = 0;
1355                 break;
1356         }
1357         case KVM_SET_MP_STATE: {
1358                 struct kvm_mp_state mp_state;
1359
1360                 r = -EFAULT;
1361                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1362                         goto out;
1363                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1364                 if (r)
1365                         goto out;
1366                 r = 0;
1367                 break;
1368         }
1369         case KVM_TRANSLATE: {
1370                 struct kvm_translation tr;
1371
1372                 r = -EFAULT;
1373                 if (copy_from_user(&tr, argp, sizeof tr))
1374                         goto out;
1375                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1376                 if (r)
1377                         goto out;
1378                 r = -EFAULT;
1379                 if (copy_to_user(argp, &tr, sizeof tr))
1380                         goto out;
1381                 r = 0;
1382                 break;
1383         }
1384         case KVM_SET_GUEST_DEBUG: {
1385                 struct kvm_guest_debug dbg;
1386
1387                 r = -EFAULT;
1388                 if (copy_from_user(&dbg, argp, sizeof dbg))
1389                         goto out;
1390                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1391                 if (r)
1392                         goto out;
1393                 r = 0;
1394                 break;
1395         }
1396         case KVM_SET_SIGNAL_MASK: {
1397                 struct kvm_signal_mask __user *sigmask_arg = argp;
1398                 struct kvm_signal_mask kvm_sigmask;
1399                 sigset_t sigset, *p;
1400
1401                 p = NULL;
1402                 if (argp) {
1403                         r = -EFAULT;
1404                         if (copy_from_user(&kvm_sigmask, argp,
1405                                            sizeof kvm_sigmask))
1406                                 goto out;
1407                         r = -EINVAL;
1408                         if (kvm_sigmask.len != sizeof sigset)
1409                                 goto out;
1410                         r = -EFAULT;
1411                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1412                                            sizeof sigset))
1413                                 goto out;
1414                         p = &sigset;
1415                 }
1416                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1417                 break;
1418         }
1419         case KVM_GET_FPU: {
1420                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1421                 r = -ENOMEM;
1422                 if (!fpu)
1423                         goto out;
1424                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1425                 if (r)
1426                         goto out;
1427                 r = -EFAULT;
1428                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1429                         goto out;
1430                 r = 0;
1431                 break;
1432         }
1433         case KVM_SET_FPU: {
1434                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1435                 r = -ENOMEM;
1436                 if (!fpu)
1437                         goto out;
1438                 r = -EFAULT;
1439                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1440                         goto out;
1441                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1442                 if (r)
1443                         goto out;
1444                 r = 0;
1445                 break;
1446         }
1447         default:
1448                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1449         }
1450 out:
1451         kfree(fpu);
1452         kfree(kvm_sregs);
1453         return r;
1454 }
1455
1456 static long kvm_vm_ioctl(struct file *filp,
1457                            unsigned int ioctl, unsigned long arg)
1458 {
1459         struct kvm *kvm = filp->private_data;
1460         void __user *argp = (void __user *)arg;
1461         int r;
1462
1463         if (kvm->mm != current->mm)
1464                 return -EIO;
1465         switch (ioctl) {
1466         case KVM_CREATE_VCPU:
1467                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1468                 if (r < 0)
1469                         goto out;
1470                 break;
1471         case KVM_SET_USER_MEMORY_REGION: {
1472                 struct kvm_userspace_memory_region kvm_userspace_mem;
1473
1474                 r = -EFAULT;
1475                 if (copy_from_user(&kvm_userspace_mem, argp,
1476                                                 sizeof kvm_userspace_mem))
1477                         goto out;
1478
1479                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1480                 if (r)
1481                         goto out;
1482                 break;
1483         }
1484         case KVM_GET_DIRTY_LOG: {
1485                 struct kvm_dirty_log log;
1486
1487                 r = -EFAULT;
1488                 if (copy_from_user(&log, argp, sizeof log))
1489                         goto out;
1490                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1491                 if (r)
1492                         goto out;
1493                 break;
1494         }
1495 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1496         case KVM_REGISTER_COALESCED_MMIO: {
1497                 struct kvm_coalesced_mmio_zone zone;
1498                 r = -EFAULT;
1499                 if (copy_from_user(&zone, argp, sizeof zone))
1500                         goto out;
1501                 r = -ENXIO;
1502                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1503                 if (r)
1504                         goto out;
1505                 r = 0;
1506                 break;
1507         }
1508         case KVM_UNREGISTER_COALESCED_MMIO: {
1509                 struct kvm_coalesced_mmio_zone zone;
1510                 r = -EFAULT;
1511                 if (copy_from_user(&zone, argp, sizeof zone))
1512                         goto out;
1513                 r = -ENXIO;
1514                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1515                 if (r)
1516                         goto out;
1517                 r = 0;
1518                 break;
1519         }
1520 #endif
1521         case KVM_IRQFD: {
1522                 struct kvm_irqfd data;
1523
1524                 r = -EFAULT;
1525                 if (copy_from_user(&data, argp, sizeof data))
1526                         goto out;
1527                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1528                 break;
1529         }
1530         case KVM_IOEVENTFD: {
1531                 struct kvm_ioeventfd data;
1532
1533                 r = -EFAULT;
1534                 if (copy_from_user(&data, argp, sizeof data))
1535                         goto out;
1536                 r = kvm_ioeventfd(kvm, &data);
1537                 break;
1538         }
1539 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1540         case KVM_SET_BOOT_CPU_ID:
1541                 r = 0;
1542                 mutex_lock(&kvm->lock);
1543                 if (atomic_read(&kvm->online_vcpus) != 0)
1544                         r = -EBUSY;
1545                 else
1546                         kvm->bsp_vcpu_id = arg;
1547                 mutex_unlock(&kvm->lock);
1548                 break;
1549 #endif
1550         default:
1551                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1552                 if (r == -ENOTTY)
1553                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1554         }
1555 out:
1556         return r;
1557 }
1558
1559 #ifdef CONFIG_COMPAT
1560 struct compat_kvm_dirty_log {
1561         __u32 slot;
1562         __u32 padding1;
1563         union {
1564                 compat_uptr_t dirty_bitmap; /* one bit per page */
1565                 __u64 padding2;
1566         };
1567 };
1568
1569 static long kvm_vm_compat_ioctl(struct file *filp,
1570                            unsigned int ioctl, unsigned long arg)
1571 {
1572         struct kvm *kvm = filp->private_data;
1573         int r;
1574
1575         if (kvm->mm != current->mm)
1576                 return -EIO;
1577         switch (ioctl) {
1578         case KVM_GET_DIRTY_LOG: {
1579                 struct compat_kvm_dirty_log compat_log;
1580                 struct kvm_dirty_log log;
1581
1582                 r = -EFAULT;
1583                 if (copy_from_user(&compat_log, (void __user *)arg,
1584                                    sizeof(compat_log)))
1585                         goto out;
1586                 log.slot         = compat_log.slot;
1587                 log.padding1     = compat_log.padding1;
1588                 log.padding2     = compat_log.padding2;
1589                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1590
1591                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1592                 if (r)
1593                         goto out;
1594                 break;
1595         }
1596         default:
1597                 r = kvm_vm_ioctl(filp, ioctl, arg);
1598         }
1599
1600 out:
1601         return r;
1602 }
1603 #endif
1604
1605 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1606 {
1607         struct page *page[1];
1608         unsigned long addr;
1609         int npages;
1610         gfn_t gfn = vmf->pgoff;
1611         struct kvm *kvm = vma->vm_file->private_data;
1612
1613         addr = gfn_to_hva(kvm, gfn);
1614         if (kvm_is_error_hva(addr))
1615                 return VM_FAULT_SIGBUS;
1616
1617         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1618                                 NULL);
1619         if (unlikely(npages != 1))
1620                 return VM_FAULT_SIGBUS;
1621
1622         vmf->page = page[0];
1623         return 0;
1624 }
1625
1626 static const struct vm_operations_struct kvm_vm_vm_ops = {
1627         .fault = kvm_vm_fault,
1628 };
1629
1630 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1631 {
1632         vma->vm_ops = &kvm_vm_vm_ops;
1633         return 0;
1634 }
1635
1636 static struct file_operations kvm_vm_fops = {
1637         .release        = kvm_vm_release,
1638         .unlocked_ioctl = kvm_vm_ioctl,
1639 #ifdef CONFIG_COMPAT
1640         .compat_ioctl   = kvm_vm_compat_ioctl,
1641 #endif
1642         .mmap           = kvm_vm_mmap,
1643 };
1644
1645 static int kvm_dev_ioctl_create_vm(void)
1646 {
1647         int fd;
1648         struct kvm *kvm;
1649
1650         kvm = kvm_create_vm();
1651         if (IS_ERR(kvm))
1652                 return PTR_ERR(kvm);
1653         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1654         if (fd < 0)
1655                 kvm_put_kvm(kvm);
1656
1657         return fd;
1658 }
1659
1660 static long kvm_dev_ioctl_check_extension_generic(long arg)
1661 {
1662         switch (arg) {
1663         case KVM_CAP_USER_MEMORY:
1664         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1665         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1666 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1667         case KVM_CAP_SET_BOOT_CPU_ID:
1668 #endif
1669         case KVM_CAP_INTERNAL_ERROR_DATA:
1670                 return 1;
1671 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1672         case KVM_CAP_IRQ_ROUTING:
1673                 return KVM_MAX_IRQ_ROUTES;
1674 #endif
1675         default:
1676                 break;
1677         }
1678         return kvm_dev_ioctl_check_extension(arg);
1679 }
1680
1681 static long kvm_dev_ioctl(struct file *filp,
1682                           unsigned int ioctl, unsigned long arg)
1683 {
1684         long r = -EINVAL;
1685
1686         switch (ioctl) {
1687         case KVM_GET_API_VERSION:
1688                 r = -EINVAL;
1689                 if (arg)
1690                         goto out;
1691                 r = KVM_API_VERSION;
1692                 break;
1693         case KVM_CREATE_VM:
1694                 r = -EINVAL;
1695                 if (arg)
1696                         goto out;
1697                 r = kvm_dev_ioctl_create_vm();
1698                 break;
1699         case KVM_CHECK_EXTENSION:
1700                 r = kvm_dev_ioctl_check_extension_generic(arg);
1701                 break;
1702         case KVM_GET_VCPU_MMAP_SIZE:
1703                 r = -EINVAL;
1704                 if (arg)
1705                         goto out;
1706                 r = PAGE_SIZE;     /* struct kvm_run */
1707 #ifdef CONFIG_X86
1708                 r += PAGE_SIZE;    /* pio data page */
1709 #endif
1710 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1711                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1712 #endif
1713                 break;
1714         case KVM_TRACE_ENABLE:
1715         case KVM_TRACE_PAUSE:
1716         case KVM_TRACE_DISABLE:
1717                 r = -EOPNOTSUPP;
1718                 break;
1719         default:
1720                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1721         }
1722 out:
1723         return r;
1724 }
1725
1726 static struct file_operations kvm_chardev_ops = {
1727         .unlocked_ioctl = kvm_dev_ioctl,
1728         .compat_ioctl   = kvm_dev_ioctl,
1729 };
1730
1731 static struct miscdevice kvm_dev = {
1732         KVM_MINOR,
1733         "kvm",
1734         &kvm_chardev_ops,
1735 };
1736
1737 static void hardware_enable(void *junk)
1738 {
1739         int cpu = raw_smp_processor_id();
1740         int r;
1741
1742         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1743                 return;
1744
1745         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1746
1747         r = kvm_arch_hardware_enable(NULL);
1748
1749         if (r) {
1750                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1751                 atomic_inc(&hardware_enable_failed);
1752                 printk(KERN_INFO "kvm: enabling virtualization on "
1753                                  "CPU%d failed\n", cpu);
1754         }
1755 }
1756
1757 static void hardware_disable(void *junk)
1758 {
1759         int cpu = raw_smp_processor_id();
1760
1761         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1762                 return;
1763         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1764         kvm_arch_hardware_disable(NULL);
1765 }
1766
1767 static void hardware_disable_all_nolock(void)
1768 {
1769         BUG_ON(!kvm_usage_count);
1770
1771         kvm_usage_count--;
1772         if (!kvm_usage_count)
1773                 on_each_cpu(hardware_disable, NULL, 1);
1774 }
1775
1776 static void hardware_disable_all(void)
1777 {
1778         spin_lock(&kvm_lock);
1779         hardware_disable_all_nolock();
1780         spin_unlock(&kvm_lock);
1781 }
1782
1783 static int hardware_enable_all(void)
1784 {
1785         int r = 0;
1786
1787         spin_lock(&kvm_lock);
1788
1789         kvm_usage_count++;
1790         if (kvm_usage_count == 1) {
1791                 atomic_set(&hardware_enable_failed, 0);
1792                 on_each_cpu(hardware_enable, NULL, 1);
1793
1794                 if (atomic_read(&hardware_enable_failed)) {
1795                         hardware_disable_all_nolock();
1796                         r = -EBUSY;
1797                 }
1798         }
1799
1800         spin_unlock(&kvm_lock);
1801
1802         return r;
1803 }
1804
1805 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1806                            void *v)
1807 {
1808         int cpu = (long)v;
1809
1810         if (!kvm_usage_count)
1811                 return NOTIFY_OK;
1812
1813         val &= ~CPU_TASKS_FROZEN;
1814         switch (val) {
1815         case CPU_DYING:
1816                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1817                        cpu);
1818                 hardware_disable(NULL);
1819                 break;
1820         case CPU_UP_CANCELED:
1821                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1822                        cpu);
1823                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1824                 break;
1825         case CPU_ONLINE:
1826                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1827                        cpu);
1828                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1829                 break;
1830         }
1831         return NOTIFY_OK;
1832 }
1833
1834
1835 asmlinkage void kvm_handle_fault_on_reboot(void)
1836 {
1837         if (kvm_rebooting)
1838                 /* spin while reset goes on */
1839                 while (true)
1840                         ;
1841         /* Fault while not rebooting.  We want the trace. */
1842         BUG();
1843 }
1844 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1845
1846 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1847                       void *v)
1848 {
1849         /*
1850          * Some (well, at least mine) BIOSes hang on reboot if
1851          * in vmx root mode.
1852          *
1853          * And Intel TXT required VMX off for all cpu when system shutdown.
1854          */
1855         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1856         kvm_rebooting = true;
1857         on_each_cpu(hardware_disable, NULL, 1);
1858         return NOTIFY_OK;
1859 }
1860
1861 static struct notifier_block kvm_reboot_notifier = {
1862         .notifier_call = kvm_reboot,
1863         .priority = 0,
1864 };
1865
1866 void kvm_io_bus_init(struct kvm_io_bus *bus)
1867 {
1868         memset(bus, 0, sizeof(*bus));
1869 }
1870
1871 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1872 {
1873         int i;
1874
1875         for (i = 0; i < bus->dev_count; i++) {
1876                 struct kvm_io_device *pos = bus->devs[i];
1877
1878                 kvm_iodevice_destructor(pos);
1879         }
1880 }
1881
1882 /* kvm_io_bus_write - called under kvm->slots_lock */
1883 int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr,
1884                      int len, const void *val)
1885 {
1886         int i;
1887         for (i = 0; i < bus->dev_count; i++)
1888                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
1889                         return 0;
1890         return -EOPNOTSUPP;
1891 }
1892
1893 /* kvm_io_bus_read - called under kvm->slots_lock */
1894 int kvm_io_bus_read(struct kvm_io_bus *bus, gpa_t addr, int len, void *val)
1895 {
1896         int i;
1897         for (i = 0; i < bus->dev_count; i++)
1898                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
1899                         return 0;
1900         return -EOPNOTSUPP;
1901 }
1902
1903 int kvm_io_bus_register_dev(struct kvm *kvm, struct kvm_io_bus *bus,
1904                              struct kvm_io_device *dev)
1905 {
1906         int ret;
1907
1908         down_write(&kvm->slots_lock);
1909         ret = __kvm_io_bus_register_dev(bus, dev);
1910         up_write(&kvm->slots_lock);
1911
1912         return ret;
1913 }
1914
1915 /* An unlocked version. Caller must have write lock on slots_lock. */
1916 int __kvm_io_bus_register_dev(struct kvm_io_bus *bus,
1917                               struct kvm_io_device *dev)
1918 {
1919         if (bus->dev_count > NR_IOBUS_DEVS-1)
1920                 return -ENOSPC;
1921
1922         bus->devs[bus->dev_count++] = dev;
1923
1924         return 0;
1925 }
1926
1927 void kvm_io_bus_unregister_dev(struct kvm *kvm,
1928                                struct kvm_io_bus *bus,
1929                                struct kvm_io_device *dev)
1930 {
1931         down_write(&kvm->slots_lock);
1932         __kvm_io_bus_unregister_dev(bus, dev);
1933         up_write(&kvm->slots_lock);
1934 }
1935
1936 /* An unlocked version. Caller must have write lock on slots_lock. */
1937 void __kvm_io_bus_unregister_dev(struct kvm_io_bus *bus,
1938                                  struct kvm_io_device *dev)
1939 {
1940         int i;
1941
1942         for (i = 0; i < bus->dev_count; i++)
1943                 if (bus->devs[i] == dev) {
1944                         bus->devs[i] = bus->devs[--bus->dev_count];
1945                         break;
1946                 }
1947 }
1948
1949 static struct notifier_block kvm_cpu_notifier = {
1950         .notifier_call = kvm_cpu_hotplug,
1951         .priority = 20, /* must be > scheduler priority */
1952 };
1953
1954 static int vm_stat_get(void *_offset, u64 *val)
1955 {
1956         unsigned offset = (long)_offset;
1957         struct kvm *kvm;
1958
1959         *val = 0;
1960         spin_lock(&kvm_lock);
1961         list_for_each_entry(kvm, &vm_list, vm_list)
1962                 *val += *(u32 *)((void *)kvm + offset);
1963         spin_unlock(&kvm_lock);
1964         return 0;
1965 }
1966
1967 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1968
1969 static int vcpu_stat_get(void *_offset, u64 *val)
1970 {
1971         unsigned offset = (long)_offset;
1972         struct kvm *kvm;
1973         struct kvm_vcpu *vcpu;
1974         int i;
1975
1976         *val = 0;
1977         spin_lock(&kvm_lock);
1978         list_for_each_entry(kvm, &vm_list, vm_list)
1979                 kvm_for_each_vcpu(i, vcpu, kvm)
1980                         *val += *(u32 *)((void *)vcpu + offset);
1981
1982         spin_unlock(&kvm_lock);
1983         return 0;
1984 }
1985
1986 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1987
1988 static const struct file_operations *stat_fops[] = {
1989         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1990         [KVM_STAT_VM]   = &vm_stat_fops,
1991 };
1992
1993 static void kvm_init_debug(void)
1994 {
1995         struct kvm_stats_debugfs_item *p;
1996
1997         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1998         for (p = debugfs_entries; p->name; ++p)
1999                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2000                                                 (void *)(long)p->offset,
2001                                                 stat_fops[p->kind]);
2002 }
2003
2004 static void kvm_exit_debug(void)
2005 {
2006         struct kvm_stats_debugfs_item *p;
2007
2008         for (p = debugfs_entries; p->name; ++p)
2009                 debugfs_remove(p->dentry);
2010         debugfs_remove(kvm_debugfs_dir);
2011 }
2012
2013 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2014 {
2015         if (kvm_usage_count)
2016                 hardware_disable(NULL);
2017         return 0;
2018 }
2019
2020 static int kvm_resume(struct sys_device *dev)
2021 {
2022         if (kvm_usage_count)
2023                 hardware_enable(NULL);
2024         return 0;
2025 }
2026
2027 static struct sysdev_class kvm_sysdev_class = {
2028         .name = "kvm",
2029         .suspend = kvm_suspend,
2030         .resume = kvm_resume,
2031 };
2032
2033 static struct sys_device kvm_sysdev = {
2034         .id = 0,
2035         .cls = &kvm_sysdev_class,
2036 };
2037
2038 struct page *bad_page;
2039 pfn_t bad_pfn;
2040
2041 static inline
2042 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2043 {
2044         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2045 }
2046
2047 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2048 {
2049         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2050
2051         kvm_arch_vcpu_load(vcpu, cpu);
2052 }
2053
2054 static void kvm_sched_out(struct preempt_notifier *pn,
2055                           struct task_struct *next)
2056 {
2057         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2058
2059         kvm_arch_vcpu_put(vcpu);
2060 }
2061
2062 int kvm_init(void *opaque, unsigned int vcpu_size,
2063                   struct module *module)
2064 {
2065         int r;
2066         int cpu;
2067
2068         r = kvm_arch_init(opaque);
2069         if (r)
2070                 goto out_fail;
2071
2072         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2073
2074         if (bad_page == NULL) {
2075                 r = -ENOMEM;
2076                 goto out;
2077         }
2078
2079         bad_pfn = page_to_pfn(bad_page);
2080
2081         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2082                 r = -ENOMEM;
2083                 goto out_free_0;
2084         }
2085
2086         r = kvm_arch_hardware_setup();
2087         if (r < 0)
2088                 goto out_free_0a;
2089
2090         for_each_online_cpu(cpu) {
2091                 smp_call_function_single(cpu,
2092                                 kvm_arch_check_processor_compat,
2093                                 &r, 1);
2094                 if (r < 0)
2095                         goto out_free_1;
2096         }
2097
2098         r = register_cpu_notifier(&kvm_cpu_notifier);
2099         if (r)
2100                 goto out_free_2;
2101         register_reboot_notifier(&kvm_reboot_notifier);
2102
2103         r = sysdev_class_register(&kvm_sysdev_class);
2104         if (r)
2105                 goto out_free_3;
2106
2107         r = sysdev_register(&kvm_sysdev);
2108         if (r)
2109                 goto out_free_4;
2110
2111         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2112         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2113                                            __alignof__(struct kvm_vcpu),
2114                                            0, NULL);
2115         if (!kvm_vcpu_cache) {
2116                 r = -ENOMEM;
2117                 goto out_free_5;
2118         }
2119
2120         kvm_chardev_ops.owner = module;
2121         kvm_vm_fops.owner = module;
2122         kvm_vcpu_fops.owner = module;
2123
2124         r = misc_register(&kvm_dev);
2125         if (r) {
2126                 printk(KERN_ERR "kvm: misc device register failed\n");
2127                 goto out_free;
2128         }
2129
2130         kvm_preempt_ops.sched_in = kvm_sched_in;
2131         kvm_preempt_ops.sched_out = kvm_sched_out;
2132
2133         kvm_init_debug();
2134
2135         return 0;
2136
2137 out_free:
2138         kmem_cache_destroy(kvm_vcpu_cache);
2139 out_free_5:
2140         sysdev_unregister(&kvm_sysdev);
2141 out_free_4:
2142         sysdev_class_unregister(&kvm_sysdev_class);
2143 out_free_3:
2144         unregister_reboot_notifier(&kvm_reboot_notifier);
2145         unregister_cpu_notifier(&kvm_cpu_notifier);
2146 out_free_2:
2147 out_free_1:
2148         kvm_arch_hardware_unsetup();
2149 out_free_0a:
2150         free_cpumask_var(cpus_hardware_enabled);
2151 out_free_0:
2152         __free_page(bad_page);
2153 out:
2154         kvm_arch_exit();
2155 out_fail:
2156         return r;
2157 }
2158 EXPORT_SYMBOL_GPL(kvm_init);
2159
2160 void kvm_exit(void)
2161 {
2162         tracepoint_synchronize_unregister();
2163         kvm_exit_debug();
2164         misc_deregister(&kvm_dev);
2165         kmem_cache_destroy(kvm_vcpu_cache);
2166         sysdev_unregister(&kvm_sysdev);
2167         sysdev_class_unregister(&kvm_sysdev_class);
2168         unregister_reboot_notifier(&kvm_reboot_notifier);
2169         unregister_cpu_notifier(&kvm_cpu_notifier);
2170         on_each_cpu(hardware_disable, NULL, 1);
2171         kvm_arch_hardware_unsetup();
2172         kvm_arch_exit();
2173         free_cpumask_var(cpus_hardware_enabled);
2174         __free_page(bad_page);
2175 }
2176 EXPORT_SYMBOL_GPL(kvm_exit);