SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 379         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 380         if (!page) {
 381                 r = -ENOMEM;
 382                 goto out_err;
 383         }
 384         kvm->coalesced_mmio_ring =
 385                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 386 #endif
 387
 388         r = kvm_init_mmu_notifier(kvm);
 389         if (r) {
 390 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 391                 put_page(page);
 392 #endif
 393                 goto out_err;
 394         }
 395
 396         kvm->mm = current->mm;
 397         atomic_inc(&kvm->mm->mm_count);
 398         spin_lock_init(&kvm->mmu_lock);
 399         spin_lock_init(&kvm->requests_lock);
 400         kvm_io_bus_init(&kvm->pio_bus);
 401         kvm_eventfd_init(kvm);
 402         mutex_init(&kvm->lock);
 403         mutex_init(&kvm->irq_lock);
 404         kvm_io_bus_init(&kvm->mmio_bus);
 405         init_rwsem(&kvm->slots_lock);
 406         atomic_set(&kvm->users_count, 1);
 407         spin_lock(&kvm_lock);
 408         list_add(&kvm->vm_list, &vm_list);
 409         spin_unlock(&kvm_lock);
 410 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 411         kvm_coalesced_mmio_init(kvm);
 412 #endif
 413 out:
 414         return kvm;
 415
 416 out_err:
 417         hardware_disable_all();
 418 out_err_nodisable:
 419         kfree(kvm);
 420         return ERR_PTR(r);
 421 }
 422
 423 /*
 424  * Free any memory in @free but not in @dont.
 425  */
 426 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 427                                   struct kvm_memory_slot *dont)
 428 {
 429         int i;
 430
 431         if (!dont || free->rmap != dont->rmap)
 432                 vfree(free->rmap);
 433
 434         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 435                 vfree(free->dirty_bitmap);
 436
 437
 438         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 439                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 440                         vfree(free->lpage_info[i]);
 441                         free->lpage_info[i] = NULL;
 442                 }
 443         }
 444
 445         free->npages = 0;
 446         free->dirty_bitmap = NULL;
 447         free->rmap = NULL;
 448 }
 449
 450 void kvm_free_physmem(struct kvm *kvm)
 451 {
 452         int i;
 453
 454         for (i = 0; i < kvm->nmemslots; ++i)
 455                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 456 }
 457
 458 static void kvm_destroy_vm(struct kvm *kvm)
 459 {
 460         struct mm_struct *mm = kvm->mm;
 461
 462         kvm_arch_sync_events(kvm);
 463         spin_lock(&kvm_lock);
 464         list_del(&kvm->vm_list);
 465         spin_unlock(&kvm_lock);
 466         kvm_free_irq_routing(kvm);
 467         kvm_io_bus_destroy(&kvm->pio_bus);
 468         kvm_io_bus_destroy(&kvm->mmio_bus);
 469         kvm_coalesced_mmio_free(kvm);
 470 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 471         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 472 #else
 473         kvm_arch_flush_shadow(kvm);
 474 #endif
 475         kvm_arch_destroy_vm(kvm);
 476         hardware_disable_all();
 477         mmdrop(mm);
 478 }
 479
 480 void kvm_get_kvm(struct kvm *kvm)
 481 {
 482         atomic_inc(&kvm->users_count);
 483 }
 484 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 485
 486 void kvm_put_kvm(struct kvm *kvm)
 487 {
 488         if (atomic_dec_and_test(&kvm->users_count))
 489                 kvm_destroy_vm(kvm);
 490 }
 491 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 492
 493
 494 static int kvm_vm_release(struct inode *inode, struct file *filp)
 495 {
 496         struct kvm *kvm = filp->private_data;
 497
 498         kvm_irqfd_release(kvm);
 499
 500         kvm_put_kvm(kvm);
 501         return 0;
 502 }
 503
 504 /*
 505  * Allocate some memory and give it an address in the guest physical address
 506  * space.
 507  *
 508  * Discontiguous memory is allowed, mostly for framebuffers.
 509  *
 510  * Must be called holding mmap_sem for write.
 511  */
 512 int __kvm_set_memory_region(struct kvm *kvm,
 513                             struct kvm_userspace_memory_region *mem,
 514                             int user_alloc)
 515 {
 516         int r;
 517         gfn_t base_gfn;
 518         unsigned long npages;
 519         unsigned long i;
 520         struct kvm_memory_slot *memslot;
 521         struct kvm_memory_slot old, new;
 522
 523         r = -EINVAL;
 524         /* General sanity checks */
 525         if (mem->memory_size & (PAGE_SIZE - 1))
 526                 goto out;
 527         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 528                 goto out;
 529         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 530                 goto out;
 531         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 532                 goto out;
 533         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 534                 goto out;
 535
 536         memslot = &kvm->memslots[mem->slot];
 537         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 538         npages = mem->memory_size >> PAGE_SHIFT;
 539
 540         if (!npages)
 541                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 542
 543         new = old = *memslot;
 544
 545         new.base_gfn = base_gfn;
 546         new.npages = npages;
 547         new.flags = mem->flags;
 548
 549         /* Disallow changing a memory slot's size. */
 550         r = -EINVAL;
 551         if (npages && old.npages && npages != old.npages)
 552                 goto out_free;
 553
 554         /* Check for overlaps */
 555         r = -EEXIST;
 556         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 557                 struct kvm_memory_slot *s = &kvm->memslots[i];
 558
 559                 if (s == memslot || !s->npages)
 560                         continue;
 561                 if (!((base_gfn + npages <= s->base_gfn) ||
 562                       (base_gfn >= s->base_gfn + s->npages)))
 563                         goto out_free;
 564         }
 565
 566         /* Free page dirty bitmap if unneeded */
 567         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 568                 new.dirty_bitmap = NULL;
 569
 570         r = -ENOMEM;
 571
 572         /* Allocate if a slot is being created */
 573 #ifndef CONFIG_S390
 574         if (npages && !new.rmap) {
 575                 new.rmap = vmalloc(npages * sizeof(struct page *));
 576
 577                 if (!new.rmap)
 578                         goto out_free;
 579
 580                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 581
 582                 new.user_alloc = user_alloc;
 583                 /*
 584                  * hva_to_rmmap() serialzies with the mmu_lock and to be
 585                  * safe it has to ignore memslots with !user_alloc &&
 586                  * !userspace_addr.
 587                  */
 588                 if (user_alloc)
 589                         new.userspace_addr = mem->userspace_addr;
 590                 else
 591                         new.userspace_addr = 0;
 592         }
 593         if (!npages)
 594                 goto skip_lpage;
 595
 596         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 597                 unsigned long ugfn;
 598                 unsigned long j;
 599                 int lpages;
 600                 int level = i + 2;
 601
 602                 /* Avoid unused variable warning if no large pages */
 603                 (void)level;
 604
 605                 if (new.lpage_info[i])
 606                         continue;
 607
 608                 lpages = 1 + (base_gfn + npages - 1) /
 609                              KVM_PAGES_PER_HPAGE(level);
 610                 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
 611
 612                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
 613
 614                 if (!new.lpage_info[i])
 615                         goto out_free;
 616
 617                 memset(new.lpage_info[i], 0,
 618                        lpages * sizeof(*new.lpage_info[i]));
 619
 620                 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
 621                         new.lpage_info[i][0].write_count = 1;
 622                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
 623                         new.lpage_info[i][lpages - 1].write_count = 1;
 624                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 625                 /*
 626                  * If the gfn and userspace address are not aligned wrt each
 627                  * other, or if explicitly asked to, disable large page
 628                  * support for this slot
 629                  */
 630                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 631                     !largepages_enabled)
 632                         for (j = 0; j < lpages; ++j)
 633                                 new.lpage_info[i][j].write_count = 1;
 634         }
 635
 636 skip_lpage:
 637
 638         /* Allocate page dirty bitmap if needed */
 639         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 640                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 641
 642                 new.dirty_bitmap = vmalloc(dirty_bytes);
 643                 if (!new.dirty_bitmap)
 644                         goto out_free;
 645                 memset(new.dirty_bitmap, 0, dirty_bytes);
 646                 if (old.npages)
 647                         kvm_arch_flush_shadow(kvm);
 648         }
 649 #else  /* not defined CONFIG_S390 */
 650         new.user_alloc = user_alloc;
 651         if (user_alloc)
 652                 new.userspace_addr = mem->userspace_addr;
 653 #endif /* not defined CONFIG_S390 */
 654
 655         if (!npages)
 656                 kvm_arch_flush_shadow(kvm);
 657
 658         spin_lock(&kvm->mmu_lock);
 659         if (mem->slot >= kvm->nmemslots)
 660                 kvm->nmemslots = mem->slot + 1;
 661
 662         *memslot = new;
 663         spin_unlock(&kvm->mmu_lock);
 664
 665         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
 666         if (r) {
 667                 spin_lock(&kvm->mmu_lock);
 668                 *memslot = old;
 669                 spin_unlock(&kvm->mmu_lock);
 670                 goto out_free;
 671         }
 672
 673         kvm_free_physmem_slot(&old, npages ? &new : NULL);
 674         /* Slot deletion case: we have to update the current slot */
 675         spin_lock(&kvm->mmu_lock);
 676         if (!npages)
 677                 *memslot = old;
 678         spin_unlock(&kvm->mmu_lock);
 679 #ifdef CONFIG_DMAR
 680         /* map the pages in iommu page table */
 681         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
 682         if (r)
 683                 goto out;
 684 #endif
 685         return 0;
 686
 687 out_free:
 688         kvm_free_physmem_slot(&new, &old);
 689 out:
 690         return r;
 691
 692 }
 693 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 694
 695 int kvm_set_memory_region(struct kvm *kvm,
 696                           struct kvm_userspace_memory_region *mem,
 697                           int user_alloc)
 698 {
 699         int r;
 700
 701         down_write(&kvm->slots_lock);
 702         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 703         up_write(&kvm->slots_lock);
 704         return r;
 705 }
 706 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 707
 708 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 709                                    struct
 710                                    kvm_userspace_memory_region *mem,
 711                                    int user_alloc)
 712 {
 713         if (mem->slot >= KVM_MEMORY_SLOTS)
 714                 return -EINVAL;
 715         return kvm_set_memory_region(kvm, mem, user_alloc);
 716 }
 717
 718 int kvm_get_dirty_log(struct kvm *kvm,
 719                         struct kvm_dirty_log *log, int *is_dirty)
 720 {
 721         struct kvm_memory_slot *memslot;
 722         int r, i;
 723         int n;
 724         unsigned long any = 0;
 725
 726         r = -EINVAL;
 727         if (log->slot >= KVM_MEMORY_SLOTS)
 728                 goto out;
 729
 730         memslot = &kvm->memslots[log->slot];
 731         r = -ENOENT;
 732         if (!memslot->dirty_bitmap)
 733                 goto out;
 734
 735         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 736
 737         for (i = 0; !any && i < n/sizeof(long); ++i)
 738                 any = memslot->dirty_bitmap[i];
 739
 740         r = -EFAULT;
 741         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 742                 goto out;
 743
 744         if (any)
 745                 *is_dirty = 1;
 746
 747         r = 0;
 748 out:
 749         return r;
 750 }
 751
 752 void kvm_disable_largepages(void)
 753 {
 754         largepages_enabled = false;
 755 }
 756 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 757
 758 int is_error_page(struct page *page)
 759 {
 760         return page == bad_page;
 761 }
 762 EXPORT_SYMBOL_GPL(is_error_page);
 763
 764 int is_error_pfn(pfn_t pfn)
 765 {
 766         return pfn == bad_pfn;
 767 }
 768 EXPORT_SYMBOL_GPL(is_error_pfn);
 769
 770 static inline unsigned long bad_hva(void)
 771 {
 772         return PAGE_OFFSET;
 773 }
 774
 775 int kvm_is_error_hva(unsigned long addr)
 776 {
 777         return addr == bad_hva();
 778 }
 779 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 780
 781 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
 782 {
 783         int i;
 784
 785         for (i = 0; i < kvm->nmemslots; ++i) {
 786                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 787
 788                 if (gfn >= memslot->base_gfn
 789                     && gfn < memslot->base_gfn + memslot->npages)
 790                         return memslot;
 791         }
 792         return NULL;
 793 }
 794 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
 795
 796 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 797 {
 798         gfn = unalias_gfn(kvm, gfn);
 799         return gfn_to_memslot_unaliased(kvm, gfn);
 800 }
 801
 802 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 803 {
 804         int i;
 805
 806         gfn = unalias_gfn(kvm, gfn);
 807         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 808                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 809
 810                 if (gfn >= memslot->base_gfn
 811                     && gfn < memslot->base_gfn + memslot->npages)
 812                         return 1;
 813         }
 814         return 0;
 815 }
 816 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 817
 818 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 819 {
 820         struct kvm_memory_slot *slot;
 821
 822         gfn = unalias_gfn(kvm, gfn);
 823         slot = gfn_to_memslot_unaliased(kvm, gfn);
 824         if (!slot)
 825                 return bad_hva();
 826         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 827 }
 828 EXPORT_SYMBOL_GPL(gfn_to_hva);
 829
 830 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 831 {
 832         struct page *page[1];
 833         unsigned long addr;
 834         int npages;
 835         pfn_t pfn;
 836
 837         might_sleep();
 838
 839         addr = gfn_to_hva(kvm, gfn);
 840         if (kvm_is_error_hva(addr)) {
 841                 get_page(bad_page);
 842                 return page_to_pfn(bad_page);
 843         }
 844
 845         npages = get_user_pages_fast(addr, 1, 1, page);
 846
 847         if (unlikely(npages != 1)) {
 848                 struct vm_area_struct *vma;
 849
 850                 down_read(&current->mm->mmap_sem);
 851                 vma = find_vma(current->mm, addr);
 852
 853                 if (vma == NULL || addr < vma->vm_start ||
 854                     !(vma->vm_flags & VM_PFNMAP)) {
 855                         up_read(&current->mm->mmap_sem);
 856                         get_page(bad_page);
 857                         return page_to_pfn(bad_page);
 858                 }
 859
 860                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 861                 up_read(&current->mm->mmap_sem);
 862                 BUG_ON(!kvm_is_mmio_pfn(pfn));
 863         } else
 864                 pfn = page_to_pfn(page[0]);
 865
 866         return pfn;
 867 }
 868
 869 EXPORT_SYMBOL_GPL(gfn_to_pfn);
 870
 871 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 872 {
 873         pfn_t pfn;
 874
 875         pfn = gfn_to_pfn(kvm, gfn);
 876         if (!kvm_is_mmio_pfn(pfn))
 877                 return pfn_to_page(pfn);
 878
 879         WARN_ON(kvm_is_mmio_pfn(pfn));
 880
 881         get_page(bad_page);
 882         return bad_page;
 883 }
 884
 885 EXPORT_SYMBOL_GPL(gfn_to_page);
 886
 887 void kvm_release_page_clean(struct page *page)
 888 {
 889         kvm_release_pfn_clean(page_to_pfn(page));
 890 }
 891 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
 892
 893 void kvm_release_pfn_clean(pfn_t pfn)
 894 {
 895         if (!kvm_is_mmio_pfn(pfn))
 896                 put_page(pfn_to_page(pfn));
 897 }
 898 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
 899
 900 void kvm_release_page_dirty(struct page *page)
 901 {
 902         kvm_release_pfn_dirty(page_to_pfn(page));
 903 }
 904 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
 905
 906 void kvm_release_pfn_dirty(pfn_t pfn)
 907 {
 908         kvm_set_pfn_dirty(pfn);
 909         kvm_release_pfn_clean(pfn);
 910 }
 911 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
 912
 913 void kvm_set_page_dirty(struct page *page)
 914 {
 915         kvm_set_pfn_dirty(page_to_pfn(page));
 916 }
 917 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
 918
 919 void kvm_set_pfn_dirty(pfn_t pfn)
 920 {
 921         if (!kvm_is_mmio_pfn(pfn)) {
 922                 struct page *page = pfn_to_page(pfn);
 923                 if (!PageReserved(page))
 924                         SetPageDirty(page);
 925         }
 926 }
 927 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
 928
 929 void kvm_set_pfn_accessed(pfn_t pfn)
 930 {
 931         if (!kvm_is_mmio_pfn(pfn))
 932                 mark_page_accessed(pfn_to_page(pfn));
 933 }
 934 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
 935
 936 void kvm_get_pfn(pfn_t pfn)
 937 {
 938         if (!kvm_is_mmio_pfn(pfn))
 939                 get_page(pfn_to_page(pfn));
 940 }
 941 EXPORT_SYMBOL_GPL(kvm_get_pfn);
 942
 943 static int next_segment(unsigned long len, int offset)
 944 {
 945         if (len > PAGE_SIZE - offset)
 946                 return PAGE_SIZE - offset;
 947         else
 948                 return len;
 949 }
 950
 951 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 952                         int len)
 953 {
 954         int r;
 955         unsigned long addr;
 956
 957         addr = gfn_to_hva(kvm, gfn);
 958         if (kvm_is_error_hva(addr))
 959                 return -EFAULT;
 960         r = copy_from_user(data, (void __user *)addr + offset, len);
 961         if (r)
 962                 return -EFAULT;
 963         return 0;
 964 }
 965 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 966
 967 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 968 {
 969         gfn_t gfn = gpa >> PAGE_SHIFT;
 970         int seg;
 971         int offset = offset_in_page(gpa);
 972         int ret;
 973
 974         while ((seg = next_segment(len, offset)) != 0) {
 975                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 976                 if (ret < 0)
 977                         return ret;
 978                 offset = 0;
 979                 len -= seg;
 980                 data += seg;
 981                 ++gfn;
 982         }
 983         return 0;
 984 }
 985 EXPORT_SYMBOL_GPL(kvm_read_guest);
 986
 987 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
 988                           unsigned long len)
 989 {
 990         int r;
 991         unsigned long addr;
 992         gfn_t gfn = gpa >> PAGE_SHIFT;
 993         int offset = offset_in_page(gpa);
 994
 995         addr = gfn_to_hva(kvm, gfn);
 996         if (kvm_is_error_hva(addr))
 997                 return -EFAULT;
 998         pagefault_disable();
 999         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1000         pagefault_enable();
1001         if (r)
1002                 return -EFAULT;
1003         return 0;
1004 }
1005 EXPORT_SYMBOL(kvm_read_guest_atomic);
1006
1007 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1008                          int offset, int len)
1009 {
1010         int r;
1011         unsigned long addr;
1012
1013         addr = gfn_to_hva(kvm, gfn);
1014         if (kvm_is_error_hva(addr))
1015                 return -EFAULT;
1016         r = copy_to_user((void __user *)addr + offset, data, len);
1017         if (r)
1018                 return -EFAULT;
1019         mark_page_dirty(kvm, gfn);
1020         return 0;
1021 }
1022 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1023
1024 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1025                     unsigned long len)
1026 {
1027         gfn_t gfn = gpa >> PAGE_SHIFT;
1028         int seg;
1029         int offset = offset_in_page(gpa);
1030         int ret;
1031
1032         while ((seg = next_segment(len, offset)) != 0) {
1033                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1034                 if (ret < 0)
1035                         return ret;
1036                 offset = 0;
1037                 len -= seg;
1038                 data += seg;
1039                 ++gfn;
1040         }
1041         return 0;
1042 }
1043
1044 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1045 {
1046         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1047 }
1048 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1049
1050 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1051 {
1052         gfn_t gfn = gpa >> PAGE_SHIFT;
1053         int seg;
1054         int offset = offset_in_page(gpa);
1055         int ret;
1056
1057         while ((seg = next_segment(len, offset)) != 0) {
1058                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1059                 if (ret < 0)
1060                         return ret;
1061                 offset = 0;
1062                 len -= seg;
1063                 ++gfn;
1064         }
1065         return 0;
1066 }
1067 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1068
1069 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1070 {
1071         struct kvm_memory_slot *memslot;
1072
1073         gfn = unalias_gfn(kvm, gfn);
1074         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1075         if (memslot && memslot->dirty_bitmap) {
1076                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1077
1078                 /* avoid RMW */
1079                 if (!generic_test_le_bit(rel_gfn, memslot->dirty_bitmap))
1080                         generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1081         }
1082 }
1083
1084 /*
1085  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1086  */
1087 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1088 {
1089         DEFINE_WAIT(wait);
1090
1091         for (;;) {
1092                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1093
1094                 if (kvm_arch_vcpu_runnable(vcpu)) {
1095                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1096                         break;
1097                 }
1098                 if (kvm_cpu_has_pending_timer(vcpu))
1099                         break;
1100                 if (signal_pending(current))
1101                         break;
1102
1103                 schedule();
1104         }
1105
1106         finish_wait(&vcpu->wq, &wait);
1107 }
1108
1109 void kvm_resched(struct kvm_vcpu *vcpu)
1110 {
1111         if (!need_resched())
1112                 return;
1113         cond_resched();
1114 }
1115 EXPORT_SYMBOL_GPL(kvm_resched);
1116
1117 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1118 {
1119         ktime_t expires;
1120         DEFINE_WAIT(wait);
1121
1122         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1123
1124         /* Sleep for 100 us, and hope lock-holder got scheduled */
1125         expires = ktime_add_ns(ktime_get(), 100000UL);
1126         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1127
1128         finish_wait(&vcpu->wq, &wait);
1129 }
1130 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1131
1132 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1133 {
1134         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1135         struct page *page;
1136
1137         if (vmf->pgoff == 0)
1138                 page = virt_to_page(vcpu->run);
1139 #ifdef CONFIG_X86
1140         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1141                 page = virt_to_page(vcpu->arch.pio_data);
1142 #endif
1143 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1144         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1145                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1146 #endif
1147         else
1148                 return VM_FAULT_SIGBUS;
1149         get_page(page);
1150         vmf->page = page;
1151         return 0;
1152 }
1153
1154 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1155         .fault = kvm_vcpu_fault,
1156 };
1157
1158 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1159 {
1160         vma->vm_ops = &kvm_vcpu_vm_ops;
1161         return 0;
1162 }
1163
1164 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1165 {
1166         struct kvm_vcpu *vcpu = filp->private_data;
1167
1168         kvm_put_kvm(vcpu->kvm);
1169         return 0;
1170 }
1171
1172 static struct file_operations kvm_vcpu_fops = {
1173         .release        = kvm_vcpu_release,
1174         .unlocked_ioctl = kvm_vcpu_ioctl,
1175         .compat_ioctl   = kvm_vcpu_ioctl,
1176         .mmap           = kvm_vcpu_mmap,
1177 };
1178
1179 /*
1180  * Allocates an inode for the vcpu.
1181  */
1182 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1183 {
1184         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1185 }
1186
1187 /*
1188  * Creates some virtual cpus.  Good luck creating more than one.
1189  */
1190 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1191 {
1192         int r;
1193         struct kvm_vcpu *vcpu, *v;
1194
1195         vcpu = kvm_arch_vcpu_create(kvm, id);
1196         if (IS_ERR(vcpu))
1197                 return PTR_ERR(vcpu);
1198
1199         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1200
1201         r = kvm_arch_vcpu_setup(vcpu);
1202         if (r)
1203                 return r;
1204
1205         mutex_lock(&kvm->lock);
1206         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1207                 r = -EINVAL;
1208                 goto vcpu_destroy;
1209         }
1210
1211         kvm_for_each_vcpu(r, v, kvm)
1212                 if (v->vcpu_id == id) {
1213                         r = -EEXIST;
1214                         goto vcpu_destroy;
1215                 }
1216
1217         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1218
1219         /* Now it's all set up, let userspace reach it */
1220         kvm_get_kvm(kvm);
1221         r = create_vcpu_fd(vcpu);
1222         if (r < 0) {
1223                 kvm_put_kvm(kvm);
1224                 goto vcpu_destroy;
1225         }
1226
1227         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1228         smp_wmb();
1229         atomic_inc(&kvm->online_vcpus);
1230
1231 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1232         if (kvm->bsp_vcpu_id == id)
1233                 kvm->bsp_vcpu = vcpu;
1234 #endif
1235         mutex_unlock(&kvm->lock);
1236         return r;
1237
1238 vcpu_destroy:
1239         mutex_unlock(&kvm->lock);
1240         kvm_arch_vcpu_destroy(vcpu);
1241         return r;
1242 }
1243
1244 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1245 {
1246         if (sigset) {
1247                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1248                 vcpu->sigset_active = 1;
1249                 vcpu->sigset = *sigset;
1250         } else
1251                 vcpu->sigset_active = 0;
1252         return 0;
1253 }
1254
1255 static long kvm_vcpu_ioctl(struct file *filp,
1256                            unsigned int ioctl, unsigned long arg)
1257 {
1258         struct kvm_vcpu *vcpu = filp->private_data;
1259         void __user *argp = (void __user *)arg;
1260         int r;
1261         struct kvm_fpu *fpu = NULL;
1262         struct kvm_sregs *kvm_sregs = NULL;
1263
1264         if (vcpu->kvm->mm != current->mm)
1265                 return -EIO;
1266         switch (ioctl) {
1267         case KVM_RUN:
1268                 r = -EINVAL;
1269                 if (arg)
1270                         goto out;
1271                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1272                 break;
1273         case KVM_GET_REGS: {
1274                 struct kvm_regs *kvm_regs;
1275
1276                 r = -ENOMEM;
1277                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1278                 if (!kvm_regs)
1279                         goto out;
1280                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1281                 if (r)
1282                         goto out_free1;
1283                 r = -EFAULT;
1284                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1285                         goto out_free1;
1286                 r = 0;
1287 out_free1:
1288                 kfree(kvm_regs);
1289                 break;
1290         }
1291         case KVM_SET_REGS: {
1292                 struct kvm_regs *kvm_regs;
1293
1294                 r = -ENOMEM;
1295                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1296                 if (!kvm_regs)
1297                         goto out;
1298                 r = -EFAULT;
1299                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1300                         goto out_free2;
1301                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1302                 if (r)
1303                         goto out_free2;
1304                 r = 0;
1305 out_free2:
1306                 kfree(kvm_regs);
1307                 break;
1308         }
1309         case KVM_GET_SREGS: {
1310                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1311                 r = -ENOMEM;
1312                 if (!kvm_sregs)
1313                         goto out;
1314                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1315                 if (r)
1316                         goto out;
1317                 r = -EFAULT;
1318                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1319                         goto out;
1320                 r = 0;
1321                 break;
1322         }
1323         case KVM_SET_SREGS: {
1324                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1325                 r = -ENOMEM;
1326                 if (!kvm_sregs)
1327                         goto out;
1328                 r = -EFAULT;
1329                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1330                         goto out;
1331                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1332                 if (r)
1333                         goto out;
1334                 r = 0;
1335                 break;
1336         }
1337         case KVM_GET_MP_STATE: {
1338                 struct kvm_mp_state mp_state;
1339
1340                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1341                 if (r)
1342                         goto out;
1343                 r = -EFAULT;
1344                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1345                         goto out;
1346                 r = 0;
1347                 break;
1348         }
1349         case KVM_SET_MP_STATE: {
1350                 struct kvm_mp_state mp_state;
1351
1352                 r = -EFAULT;
1353                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1354                         goto out;
1355                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1356                 if (r)
1357                         goto out;
1358                 r = 0;
1359                 break;
1360         }
1361         case KVM_TRANSLATE: {
1362                 struct kvm_translation tr;
1363
1364                 r = -EFAULT;
1365                 if (copy_from_user(&tr, argp, sizeof tr))
1366                         goto out;
1367                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1368                 if (r)
1369                         goto out;
1370                 r = -EFAULT;
1371                 if (copy_to_user(argp, &tr, sizeof tr))
1372                         goto out;
1373                 r = 0;
1374                 break;
1375         }
1376         case KVM_SET_GUEST_DEBUG: {
1377                 struct kvm_guest_debug dbg;
1378
1379                 r = -EFAULT;
1380                 if (copy_from_user(&dbg, argp, sizeof dbg))
1381                         goto out;
1382                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1383                 if (r)
1384                         goto out;
1385                 r = 0;
1386                 break;
1387         }
1388         case KVM_SET_SIGNAL_MASK: {
1389                 struct kvm_signal_mask __user *sigmask_arg = argp;
1390                 struct kvm_signal_mask kvm_sigmask;
1391                 sigset_t sigset, *p;
1392
1393                 p = NULL;
1394                 if (argp) {
1395                         r = -EFAULT;
1396                         if (copy_from_user(&kvm_sigmask, argp,
1397                                            sizeof kvm_sigmask))
1398                                 goto out;
1399                         r = -EINVAL;
1400                         if (kvm_sigmask.len != sizeof sigset)
1401                                 goto out;
1402                         r = -EFAULT;
1403                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1404                                            sizeof sigset))
1405                                 goto out;
1406                         p = &sigset;
1407                 }
1408                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1409                 break;
1410         }
1411         case KVM_GET_FPU: {
1412                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1413                 r = -ENOMEM;
1414                 if (!fpu)
1415                         goto out;
1416                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1417                 if (r)
1418                         goto out;
1419                 r = -EFAULT;
1420                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1421                         goto out;
1422                 r = 0;
1423                 break;
1424         }
1425         case KVM_SET_FPU: {
1426                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1427                 r = -ENOMEM;
1428                 if (!fpu)
1429                         goto out;
1430                 r = -EFAULT;
1431                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1432                         goto out;
1433                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1434                 if (r)
1435                         goto out;
1436                 r = 0;
1437                 break;
1438         }
1439         default:
1440                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1441         }
1442 out:
1443         kfree(fpu);
1444         kfree(kvm_sregs);
1445         return r;
1446 }
1447
1448 static long kvm_vm_ioctl(struct file *filp,
1449                            unsigned int ioctl, unsigned long arg)
1450 {
1451         struct kvm *kvm = filp->private_data;
1452         void __user *argp = (void __user *)arg;
1453         int r;
1454
1455         if (kvm->mm != current->mm)
1456                 return -EIO;
1457         switch (ioctl) {
1458         case KVM_CREATE_VCPU:
1459                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1460                 if (r < 0)
1461                         goto out;
1462                 break;
1463         case KVM_SET_USER_MEMORY_REGION: {
1464                 struct kvm_userspace_memory_region kvm_userspace_mem;
1465
1466                 r = -EFAULT;
1467                 if (copy_from_user(&kvm_userspace_mem, argp,
1468                                                 sizeof kvm_userspace_mem))
1469                         goto out;
1470
1471                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1472                 if (r)
1473                         goto out;
1474                 break;
1475         }
1476         case KVM_GET_DIRTY_LOG: {
1477                 struct kvm_dirty_log log;
1478
1479                 r = -EFAULT;
1480                 if (copy_from_user(&log, argp, sizeof log))
1481                         goto out;
1482                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1483                 if (r)
1484                         goto out;
1485                 break;
1486         }
1487 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1488         case KVM_REGISTER_COALESCED_MMIO: {
1489                 struct kvm_coalesced_mmio_zone zone;
1490                 r = -EFAULT;
1491                 if (copy_from_user(&zone, argp, sizeof zone))
1492                         goto out;
1493                 r = -ENXIO;
1494                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1495                 if (r)
1496                         goto out;
1497                 r = 0;
1498                 break;
1499         }
1500         case KVM_UNREGISTER_COALESCED_MMIO: {
1501                 struct kvm_coalesced_mmio_zone zone;
1502                 r = -EFAULT;
1503                 if (copy_from_user(&zone, argp, sizeof zone))
1504                         goto out;
1505                 r = -ENXIO;
1506                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1507                 if (r)
1508                         goto out;
1509                 r = 0;
1510                 break;
1511         }
1512 #endif
1513         case KVM_IRQFD: {
1514                 struct kvm_irqfd data;
1515
1516                 r = -EFAULT;
1517                 if (copy_from_user(&data, argp, sizeof data))
1518                         goto out;
1519                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1520                 break;
1521         }
1522         case KVM_IOEVENTFD: {
1523                 struct kvm_ioeventfd data;
1524
1525                 r = -EFAULT;
1526                 if (copy_from_user(&data, argp, sizeof data))
1527                         goto out;
1528                 r = kvm_ioeventfd(kvm, &data);
1529                 break;
1530         }
1531 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1532         case KVM_SET_BOOT_CPU_ID:
1533                 r = 0;
1534                 mutex_lock(&kvm->lock);
1535                 if (atomic_read(&kvm->online_vcpus) != 0)
1536                         r = -EBUSY;
1537                 else
1538                         kvm->bsp_vcpu_id = arg;
1539                 mutex_unlock(&kvm->lock);
1540                 break;
1541 #endif
1542         default:
1543                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1544                 if (r == -ENOTTY)
1545                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1546         }
1547 out:
1548         return r;
1549 }
1550
1551 #ifdef CONFIG_COMPAT
1552 struct compat_kvm_dirty_log {
1553         __u32 slot;
1554         __u32 padding1;
1555         union {
1556                 compat_uptr_t dirty_bitmap; /* one bit per page */
1557                 __u64 padding2;
1558         };
1559 };
1560
1561 static long kvm_vm_compat_ioctl(struct file *filp,
1562                            unsigned int ioctl, unsigned long arg)
1563 {
1564         struct kvm *kvm = filp->private_data;
1565         int r;
1566
1567         if (kvm->mm != current->mm)
1568                 return -EIO;
1569         switch (ioctl) {
1570         case KVM_GET_DIRTY_LOG: {
1571                 struct compat_kvm_dirty_log compat_log;
1572                 struct kvm_dirty_log log;
1573
1574                 r = -EFAULT;
1575                 if (copy_from_user(&compat_log, (void __user *)arg,
1576                                    sizeof(compat_log)))
1577                         goto out;
1578                 log.slot         = compat_log.slot;
1579                 log.padding1     = compat_log.padding1;
1580                 log.padding2     = compat_log.padding2;
1581                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1582
1583                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1584                 if (r)
1585                         goto out;
1586                 break;
1587         }
1588         default:
1589                 r = kvm_vm_ioctl(filp, ioctl, arg);
1590         }
1591
1592 out:
1593         return r;
1594 }
1595 #endif
1596
1597 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1598 {
1599         struct page *page[1];
1600         unsigned long addr;
1601         int npages;
1602         gfn_t gfn = vmf->pgoff;
1603         struct kvm *kvm = vma->vm_file->private_data;
1604
1605         addr = gfn_to_hva(kvm, gfn);
1606         if (kvm_is_error_hva(addr))
1607                 return VM_FAULT_SIGBUS;
1608
1609         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1610                                 NULL);
1611         if (unlikely(npages != 1))
1612                 return VM_FAULT_SIGBUS;
1613
1614         vmf->page = page[0];
1615         return 0;
1616 }
1617
1618 static const struct vm_operations_struct kvm_vm_vm_ops = {
1619         .fault = kvm_vm_fault,
1620 };
1621
1622 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1623 {
1624         vma->vm_ops = &kvm_vm_vm_ops;
1625         return 0;
1626 }
1627
1628 static struct file_operations kvm_vm_fops = {
1629         .release        = kvm_vm_release,
1630         .unlocked_ioctl = kvm_vm_ioctl,
1631 #ifdef CONFIG_COMPAT
1632         .compat_ioctl   = kvm_vm_compat_ioctl,
1633 #endif
1634         .mmap           = kvm_vm_mmap,
1635 };
1636
1637 static int kvm_dev_ioctl_create_vm(void)
1638 {
1639         int fd;
1640         struct kvm *kvm;
1641
1642         kvm = kvm_create_vm();
1643         if (IS_ERR(kvm))
1644                 return PTR_ERR(kvm);
1645         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1646         if (fd < 0)
1647                 kvm_put_kvm(kvm);
1648
1649         return fd;
1650 }
1651
1652 static long kvm_dev_ioctl_check_extension_generic(long arg)
1653 {
1654         switch (arg) {
1655         case KVM_CAP_USER_MEMORY:
1656         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1657         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1658 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1659         case KVM_CAP_SET_BOOT_CPU_ID:
1660 #endif
1661         case KVM_CAP_INTERNAL_ERROR_DATA:
1662                 return 1;
1663 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1664         case KVM_CAP_IRQ_ROUTING:
1665                 return KVM_MAX_IRQ_ROUTES;
1666 #endif
1667         default:
1668                 break;
1669         }
1670         return kvm_dev_ioctl_check_extension(arg);
1671 }
1672
1673 static long kvm_dev_ioctl(struct file *filp,
1674                           unsigned int ioctl, unsigned long arg)
1675 {
1676         long r = -EINVAL;
1677
1678         switch (ioctl) {
1679         case KVM_GET_API_VERSION:
1680                 r = -EINVAL;
1681                 if (arg)
1682                         goto out;
1683                 r = KVM_API_VERSION;
1684                 break;
1685         case KVM_CREATE_VM:
1686                 r = -EINVAL;
1687                 if (arg)
1688                         goto out;
1689                 r = kvm_dev_ioctl_create_vm();
1690                 break;
1691         case KVM_CHECK_EXTENSION:
1692                 r = kvm_dev_ioctl_check_extension_generic(arg);
1693                 break;
1694         case KVM_GET_VCPU_MMAP_SIZE:
1695                 r = -EINVAL;
1696                 if (arg)
1697                         goto out;
1698                 r = PAGE_SIZE;     /* struct kvm_run */
1699 #ifdef CONFIG_X86
1700                 r += PAGE_SIZE;    /* pio data page */
1701 #endif
1702 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1703                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1704 #endif
1705                 break;
1706         case KVM_TRACE_ENABLE:
1707         case KVM_TRACE_PAUSE:
1708         case KVM_TRACE_DISABLE:
1709                 r = -EOPNOTSUPP;
1710                 break;
1711         default:
1712                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1713         }
1714 out:
1715         return r;
1716 }
1717
1718 static struct file_operations kvm_chardev_ops = {
1719         .unlocked_ioctl = kvm_dev_ioctl,
1720         .compat_ioctl   = kvm_dev_ioctl,
1721 };
1722
1723 static struct miscdevice kvm_dev = {
1724         KVM_MINOR,
1725         "kvm",
1726         &kvm_chardev_ops,
1727 };
1728
1729 static void hardware_enable(void *junk)
1730 {
1731         int cpu = raw_smp_processor_id();
1732         int r;
1733
1734         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1735                 return;
1736
1737         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1738
1739         r = kvm_arch_hardware_enable(NULL);
1740
1741         if (r) {
1742                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1743                 atomic_inc(&hardware_enable_failed);
1744                 printk(KERN_INFO "kvm: enabling virtualization on "
1745                                  "CPU%d failed\n", cpu);
1746         }
1747 }
1748
1749 static void hardware_disable(void *junk)
1750 {
1751         int cpu = raw_smp_processor_id();
1752
1753         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1754                 return;
1755         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1756         kvm_arch_hardware_disable(NULL);
1757 }
1758
1759 static void hardware_disable_all_nolock(void)
1760 {
1761         BUG_ON(!kvm_usage_count);
1762
1763         kvm_usage_count--;
1764         if (!kvm_usage_count)
1765                 on_each_cpu(hardware_disable, NULL, 1);
1766 }
1767
1768 static void hardware_disable_all(void)
1769 {
1770         spin_lock(&kvm_lock);
1771         hardware_disable_all_nolock();
1772         spin_unlock(&kvm_lock);
1773 }
1774
1775 static int hardware_enable_all(void)
1776 {
1777         int r = 0;
1778
1779         spin_lock(&kvm_lock);
1780
1781         kvm_usage_count++;
1782         if (kvm_usage_count == 1) {
1783                 atomic_set(&hardware_enable_failed, 0);
1784                 on_each_cpu(hardware_enable, NULL, 1);
1785
1786                 if (atomic_read(&hardware_enable_failed)) {
1787                         hardware_disable_all_nolock();
1788                         r = -EBUSY;
1789                 }
1790         }
1791
1792         spin_unlock(&kvm_lock);
1793
1794         return r;
1795 }
1796
1797 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1798                            void *v)
1799 {
1800         int cpu = (long)v;
1801
1802         if (!kvm_usage_count)
1803                 return NOTIFY_OK;
1804
1805         val &= ~CPU_TASKS_FROZEN;
1806         switch (val) {
1807         case CPU_DYING:
1808                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1809                        cpu);
1810                 hardware_disable(NULL);
1811                 break;
1812         case CPU_UP_CANCELED:
1813                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1814                        cpu);
1815                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1816                 break;
1817         case CPU_ONLINE:
1818                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1819                        cpu);
1820                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1821                 break;
1822         }
1823         return NOTIFY_OK;
1824 }
1825
1826
1827 asmlinkage void kvm_handle_fault_on_reboot(void)
1828 {
1829         if (kvm_rebooting)
1830                 /* spin while reset goes on */
1831                 while (true)
1832                         ;
1833         /* Fault while not rebooting.  We want the trace. */
1834         BUG();
1835 }
1836 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1837
1838 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1839                       void *v)
1840 {
1841         /*
1842          * Some (well, at least mine) BIOSes hang on reboot if
1843          * in vmx root mode.
1844          *
1845          * And Intel TXT required VMX off for all cpu when system shutdown.
1846          */
1847         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1848         kvm_rebooting = true;
1849         on_each_cpu(hardware_disable, NULL, 1);
1850         return NOTIFY_OK;
1851 }
1852
1853 static struct notifier_block kvm_reboot_notifier = {
1854         .notifier_call = kvm_reboot,
1855         .priority = 0,
1856 };
1857
1858 void kvm_io_bus_init(struct kvm_io_bus *bus)
1859 {
1860         memset(bus, 0, sizeof(*bus));
1861 }
1862
1863 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1864 {
1865         int i;
1866
1867         for (i = 0; i < bus->dev_count; i++) {
1868                 struct kvm_io_device *pos = bus->devs[i];
1869
1870                 kvm_iodevice_destructor(pos);
1871         }
1872 }
1873
1874 /* kvm_io_bus_write - called under kvm->slots_lock */
1875 int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr,
1876                      int len, const void *val)
1877 {
1878         int i;
1879         for (i = 0; i < bus->dev_count; i++)
1880                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
1881                         return 0;
1882         return -EOPNOTSUPP;
1883 }
1884
1885 /* kvm_io_bus_read - called under kvm->slots_lock */
1886 int kvm_io_bus_read(struct kvm_io_bus *bus, gpa_t addr, int len, void *val)
1887 {
1888         int i;
1889         for (i = 0; i < bus->dev_count; i++)
1890                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
1891                         return 0;
1892         return -EOPNOTSUPP;
1893 }
1894
1895 int kvm_io_bus_register_dev(struct kvm *kvm, struct kvm_io_bus *bus,
1896                              struct kvm_io_device *dev)
1897 {
1898         int ret;
1899
1900         down_write(&kvm->slots_lock);
1901         ret = __kvm_io_bus_register_dev(bus, dev);
1902         up_write(&kvm->slots_lock);
1903
1904         return ret;
1905 }
1906
1907 /* An unlocked version. Caller must have write lock on slots_lock. */
1908 int __kvm_io_bus_register_dev(struct kvm_io_bus *bus,
1909                               struct kvm_io_device *dev)
1910 {
1911         if (bus->dev_count > NR_IOBUS_DEVS-1)
1912                 return -ENOSPC;
1913
1914         bus->devs[bus->dev_count++] = dev;
1915
1916         return 0;
1917 }
1918
1919 void kvm_io_bus_unregister_dev(struct kvm *kvm,
1920                                struct kvm_io_bus *bus,
1921                                struct kvm_io_device *dev)
1922 {
1923         down_write(&kvm->slots_lock);
1924         __kvm_io_bus_unregister_dev(bus, dev);
1925         up_write(&kvm->slots_lock);
1926 }
1927
1928 /* An unlocked version. Caller must have write lock on slots_lock. */
1929 void __kvm_io_bus_unregister_dev(struct kvm_io_bus *bus,
1930                                  struct kvm_io_device *dev)
1931 {
1932         int i;
1933
1934         for (i = 0; i < bus->dev_count; i++)
1935                 if (bus->devs[i] == dev) {
1936                         bus->devs[i] = bus->devs[--bus->dev_count];
1937                         break;
1938                 }
1939 }
1940
1941 static struct notifier_block kvm_cpu_notifier = {
1942         .notifier_call = kvm_cpu_hotplug,
1943         .priority = 20, /* must be > scheduler priority */
1944 };
1945
1946 static int vm_stat_get(void *_offset, u64 *val)
1947 {
1948         unsigned offset = (long)_offset;
1949         struct kvm *kvm;
1950
1951         *val = 0;
1952         spin_lock(&kvm_lock);
1953         list_for_each_entry(kvm, &vm_list, vm_list)
1954                 *val += *(u32 *)((void *)kvm + offset);
1955         spin_unlock(&kvm_lock);
1956         return 0;
1957 }
1958
1959 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1960
1961 static int vcpu_stat_get(void *_offset, u64 *val)
1962 {
1963         unsigned offset = (long)_offset;
1964         struct kvm *kvm;
1965         struct kvm_vcpu *vcpu;
1966         int i;
1967
1968         *val = 0;
1969         spin_lock(&kvm_lock);
1970         list_for_each_entry(kvm, &vm_list, vm_list)
1971                 kvm_for_each_vcpu(i, vcpu, kvm)
1972                         *val += *(u32 *)((void *)vcpu + offset);
1973
1974         spin_unlock(&kvm_lock);
1975         return 0;
1976 }
1977
1978 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1979
1980 static const struct file_operations *stat_fops[] = {
1981         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1982         [KVM_STAT_VM]   = &vm_stat_fops,
1983 };
1984
1985 static void kvm_init_debug(void)
1986 {
1987         struct kvm_stats_debugfs_item *p;
1988
1989         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1990         for (p = debugfs_entries; p->name; ++p)
1991                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1992                                                 (void *)(long)p->offset,
1993                                                 stat_fops[p->kind]);
1994 }
1995
1996 static void kvm_exit_debug(void)
1997 {
1998         struct kvm_stats_debugfs_item *p;
1999
2000         for (p = debugfs_entries; p->name; ++p)
2001                 debugfs_remove(p->dentry);
2002         debugfs_remove(kvm_debugfs_dir);
2003 }
2004
2005 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2006 {
2007         if (kvm_usage_count)
2008                 hardware_disable(NULL);
2009         return 0;
2010 }
2011
2012 static int kvm_resume(struct sys_device *dev)
2013 {
2014         if (kvm_usage_count)
2015                 hardware_enable(NULL);
2016         return 0;
2017 }
2018
2019 static struct sysdev_class kvm_sysdev_class = {
2020         .name = "kvm",
2021         .suspend = kvm_suspend,
2022         .resume = kvm_resume,
2023 };
2024
2025 static struct sys_device kvm_sysdev = {
2026         .id = 0,
2027         .cls = &kvm_sysdev_class,
2028 };
2029
2030 struct page *bad_page;
2031 pfn_t bad_pfn;
2032
2033 static inline
2034 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2035 {
2036         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2037 }
2038
2039 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2040 {
2041         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2042
2043         kvm_arch_vcpu_load(vcpu, cpu);
2044 }
2045
2046 static void kvm_sched_out(struct preempt_notifier *pn,
2047                           struct task_struct *next)
2048 {
2049         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2050
2051         kvm_arch_vcpu_put(vcpu);
2052 }
2053
2054 int kvm_init(void *opaque, unsigned int vcpu_size,
2055                   struct module *module)
2056 {
2057         int r;
2058         int cpu;
2059
2060         r = kvm_arch_init(opaque);
2061         if (r)
2062                 goto out_fail;
2063
2064         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2065
2066         if (bad_page == NULL) {
2067                 r = -ENOMEM;
2068                 goto out;
2069         }
2070
2071         bad_pfn = page_to_pfn(bad_page);
2072
2073         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2074                 r = -ENOMEM;
2075                 goto out_free_0;
2076         }
2077
2078         r = kvm_arch_hardware_setup();
2079         if (r < 0)
2080                 goto out_free_0a;
2081
2082         for_each_online_cpu(cpu) {
2083                 smp_call_function_single(cpu,
2084                                 kvm_arch_check_processor_compat,
2085                                 &r, 1);
2086                 if (r < 0)
2087                         goto out_free_1;
2088         }
2089
2090         r = register_cpu_notifier(&kvm_cpu_notifier);
2091         if (r)
2092                 goto out_free_2;
2093         register_reboot_notifier(&kvm_reboot_notifier);
2094
2095         r = sysdev_class_register(&kvm_sysdev_class);
2096         if (r)
2097                 goto out_free_3;
2098
2099         r = sysdev_register(&kvm_sysdev);
2100         if (r)
2101                 goto out_free_4;
2102
2103         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2104         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2105                                            __alignof__(struct kvm_vcpu),
2106                                            0, NULL);
2107         if (!kvm_vcpu_cache) {
2108                 r = -ENOMEM;
2109                 goto out_free_5;
2110         }
2111
2112         kvm_chardev_ops.owner = module;
2113         kvm_vm_fops.owner = module;
2114         kvm_vcpu_fops.owner = module;
2115
2116         r = misc_register(&kvm_dev);
2117         if (r) {
2118                 printk(KERN_ERR "kvm: misc device register failed\n");
2119                 goto out_free;
2120         }
2121
2122         kvm_preempt_ops.sched_in = kvm_sched_in;
2123         kvm_preempt_ops.sched_out = kvm_sched_out;
2124
2125         kvm_init_debug();
2126
2127         return 0;
2128
2129 out_free:
2130         kmem_cache_destroy(kvm_vcpu_cache);
2131 out_free_5:
2132         sysdev_unregister(&kvm_sysdev);
2133 out_free_4:
2134         sysdev_class_unregister(&kvm_sysdev_class);
2135 out_free_3:
2136         unregister_reboot_notifier(&kvm_reboot_notifier);
2137         unregister_cpu_notifier(&kvm_cpu_notifier);
2138 out_free_2:
2139 out_free_1:
2140         kvm_arch_hardware_unsetup();
2141 out_free_0a:
2142         free_cpumask_var(cpus_hardware_enabled);
2143 out_free_0:
2144         __free_page(bad_page);
2145 out:
2146         kvm_arch_exit();
2147 out_fail:
2148         return r;
2149 }
2150 EXPORT_SYMBOL_GPL(kvm_init);
2151
2152 void kvm_exit(void)
2153 {
2154         tracepoint_synchronize_unregister();
2155         kvm_exit_debug();
2156         misc_deregister(&kvm_dev);
2157         kmem_cache_destroy(kvm_vcpu_cache);
2158         sysdev_unregister(&kvm_sysdev);
2159         sysdev_class_unregister(&kvm_sysdev_class);
2160         unregister_reboot_notifier(&kvm_reboot_notifier);
2161         unregister_cpu_notifier(&kvm_cpu_notifier);
2162         on_each_cpu(hardware_disable, NULL, 1);
2163         kvm_arch_hardware_unsetup();
2164         kvm_arch_exit();
2165         free_cpumask_var(cpus_hardware_enabled);
2166         __free_page(bad_page);
2167 }
2168 EXPORT_SYMBOL_GPL(kvm_exit);