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