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