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