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   <[email protected]>
  11  *   Yaniv Kamay  <[email protected]>
  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/mm.h>
  26 #include <linux/miscdevice.h>
  27 #include <linux/vmalloc.h>
  28 #include <linux/reboot.h>
  29 #include <linux/debugfs.h>
  30 #include <linux/highmem.h>
  31 #include <linux/file.h>
  32 #include <linux/sysdev.h>
  33 #include <linux/cpu.h>
  34 #include <linux/sched.h>
  35 #include <linux/cpumask.h>
  36 #include <linux/smp.h>
  37 #include <linux/anon_inodes.h>
  38 #include <linux/profile.h>
  39 #include <linux/kvm_para.h>
  40 #include <linux/pagemap.h>
  41 #include <linux/mman.h>
  42 #include <linux/swap.h>
  43 #include <linux/bitops.h>
  44 #include <linux/spinlock.h>
  45 #include <linux/compat.h>
  46 #include <linux/srcu.h>
  47 #include <linux/hugetlb.h>
  48 #include <linux/slab.h>
  49
  50 #include <asm/processor.h>
  51 #include <asm/io.h>
  52 #include <asm/uaccess.h>
  53 #include <asm/pgtable.h>
  54 #include <asm-generic/bitops/le.h>
  55
  56 #include "coalesced_mmio.h"
  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 void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  90
  91 static bool kvm_rebooting;
  92
  93 static bool largepages_enabled = true;
  94
  95 inline int kvm_is_mmio_pfn(pfn_t pfn)
  96 {
  97         if (pfn_valid(pfn)) {
  98                 struct page *page = compound_head(pfn_to_page(pfn));
  99                 return PageReserved(page);
 100         }
 101
 102         return true;
 103 }
 104
 105 /*
 106  * Switches to specified vcpu, until a matching vcpu_put()
 107  */
 108 void vcpu_load(struct kvm_vcpu *vcpu)
 109 {
 110         int cpu;
 111
 112         mutex_lock(&vcpu->mutex);
 113         cpu = get_cpu();
 114         preempt_notifier_register(&vcpu->preempt_notifier);
 115         kvm_arch_vcpu_load(vcpu, cpu);
 116         put_cpu();
 117 }
 118
 119 void vcpu_put(struct kvm_vcpu *vcpu)
 120 {
 121         preempt_disable();
 122         kvm_arch_vcpu_put(vcpu);
 123         preempt_notifier_unregister(&vcpu->preempt_notifier);
 124         preempt_enable();
 125         mutex_unlock(&vcpu->mutex);
 126 }
 127
 128 static void ack_flush(void *_completed)
 129 {
 130 }
 131
 132 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 133 {
 134         int i, cpu, me;
 135         cpumask_var_t cpus;
 136         bool called = true;
 137         struct kvm_vcpu *vcpu;
 138
 139         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 140
 141         raw_spin_lock(&kvm->requests_lock);
 142         me = smp_processor_id();
 143         kvm_for_each_vcpu(i, vcpu, kvm) {
 144                 if (test_and_set_bit(req, &vcpu->requests))
 145                         continue;
 146                 cpu = vcpu->cpu;
 147                 if (cpus != NULL && cpu != -1 && cpu != me)
 148                         cpumask_set_cpu(cpu, cpus);
 149         }
 150         if (unlikely(cpus == NULL))
 151                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 152         else if (!cpumask_empty(cpus))
 153                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 154         else
 155                 called = false;
 156         raw_spin_unlock(&kvm->requests_lock);
 157         free_cpumask_var(cpus);
 158         return called;
 159 }
 160
 161 void kvm_flush_remote_tlbs(struct kvm *kvm)
 162 {
 163         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 164                 ++kvm->stat.remote_tlb_flush;
 165 }
 166
 167 void kvm_reload_remote_mmus(struct kvm *kvm)
 168 {
 169         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 170 }
 171
 172 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 173 {
 174         struct page *page;
 175         int r;
 176
 177         mutex_init(&vcpu->mutex);
 178         vcpu->cpu = -1;
 179         vcpu->kvm = kvm;
 180         vcpu->vcpu_id = id;
 181         init_waitqueue_head(&vcpu->wq);
 182
 183         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 184         if (!page) {
 185                 r = -ENOMEM;
 186                 goto fail;
 187         }
 188         vcpu->run = page_address(page);
 189
 190         r = kvm_arch_vcpu_init(vcpu);
 191         if (r < 0)
 192                 goto fail_free_run;
 193         return 0;
 194
 195 fail_free_run:
 196         free_page((unsigned long)vcpu->run);
 197 fail:
 198         return r;
 199 }
 200 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 201
 202 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 203 {
 204         kvm_arch_vcpu_uninit(vcpu);
 205         free_page((unsigned long)vcpu->run);
 206 }
 207 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 208
 209 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 210 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 211 {
 212         return container_of(mn, struct kvm, mmu_notifier);
 213 }
 214
 215 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 216                                              struct mm_struct *mm,
 217                                              unsigned long address)
 218 {
 219         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 220         int need_tlb_flush, idx;
 221
 222         /*
 223          * When ->invalidate_page runs, the linux pte has been zapped
 224          * already but the page is still allocated until
 225          * ->invalidate_page returns. So if we increase the sequence
 226          * here the kvm page fault will notice if the spte can't be
 227          * established because the page is going to be freed. If
 228          * instead the kvm page fault establishes the spte before
 229          * ->invalidate_page runs, kvm_unmap_hva will release it
 230          * before returning.
 231          *
 232          * The sequence increase only need to be seen at spin_unlock
 233          * time, and not at spin_lock time.
 234          *
 235          * Increasing the sequence after the spin_unlock would be
 236          * unsafe because the kvm page fault could then establish the
 237          * pte after kvm_unmap_hva returned, without noticing the page
 238          * is going to be freed.
 239          */
 240         idx = srcu_read_lock(&kvm->srcu);
 241         spin_lock(&kvm->mmu_lock);
 242         kvm->mmu_notifier_seq++;
 243         need_tlb_flush = kvm_unmap_hva(kvm, address);
 244         spin_unlock(&kvm->mmu_lock);
 245         srcu_read_unlock(&kvm->srcu, idx);
 246
 247         /* we've to flush the tlb before the pages can be freed */
 248         if (need_tlb_flush)
 249                 kvm_flush_remote_tlbs(kvm);
 250
 251 }
 252
 253 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 254                                         struct mm_struct *mm,
 255                                         unsigned long address,
 256                                         pte_t pte)
 257 {
 258         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 259         int idx;
 260
 261         idx = srcu_read_lock(&kvm->srcu);
 262         spin_lock(&kvm->mmu_lock);
 263         kvm->mmu_notifier_seq++;
 264         kvm_set_spte_hva(kvm, address, pte);
 265         spin_unlock(&kvm->mmu_lock);
 266         srcu_read_unlock(&kvm->srcu, idx);
 267 }
 268
 269 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 270                                                     struct mm_struct *mm,
 271                                                     unsigned long start,
 272                                                     unsigned long end)
 273 {
 274         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 275         int need_tlb_flush = 0, idx;
 276
 277         idx = srcu_read_lock(&kvm->srcu);
 278         spin_lock(&kvm->mmu_lock);
 279         /*
 280          * The count increase must become visible at unlock time as no
 281          * spte can be established without taking the mmu_lock and
 282          * count is also read inside the mmu_lock critical section.
 283          */
 284         kvm->mmu_notifier_count++;
 285         for (; start < end; start += PAGE_SIZE)
 286                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 287         spin_unlock(&kvm->mmu_lock);
 288         srcu_read_unlock(&kvm->srcu, idx);
 289
 290         /* we've to flush the tlb before the pages can be freed */
 291         if (need_tlb_flush)
 292                 kvm_flush_remote_tlbs(kvm);
 293 }
 294
 295 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 296                                                   struct mm_struct *mm,
 297                                                   unsigned long start,
 298                                                   unsigned long end)
 299 {
 300         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 301
 302         spin_lock(&kvm->mmu_lock);
 303         /*
 304          * This sequence increase will notify the kvm page fault that
 305          * the page that is going to be mapped in the spte could have
 306          * been freed.
 307          */
 308         kvm->mmu_notifier_seq++;
 309         /*
 310          * The above sequence increase must be visible before the
 311          * below count decrease but both values are read by the kvm
 312          * page fault under mmu_lock spinlock so we don't need to add
 313          * a smb_wmb() here in between the two.
 314          */
 315         kvm->mmu_notifier_count--;
 316         spin_unlock(&kvm->mmu_lock);
 317
 318         BUG_ON(kvm->mmu_notifier_count < 0);
 319 }
 320
 321 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 322                                               struct mm_struct *mm,
 323                                               unsigned long address)
 324 {
 325         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 326         int young, idx;
 327
 328         idx = srcu_read_lock(&kvm->srcu);
 329         spin_lock(&kvm->mmu_lock);
 330         young = kvm_age_hva(kvm, address);
 331         spin_unlock(&kvm->mmu_lock);
 332         srcu_read_unlock(&kvm->srcu, idx);
 333
 334         if (young)
 335                 kvm_flush_remote_tlbs(kvm);
 336
 337         return young;
 338 }
 339
 340 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 341                                      struct mm_struct *mm)
 342 {
 343         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 344         kvm_arch_flush_shadow(kvm);
 345 }
 346
 347 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 348         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 349         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 350         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 351         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 352         .change_pte             = kvm_mmu_notifier_change_pte,
 353         .release                = kvm_mmu_notifier_release,
 354 };
 355
 356 static int kvm_init_mmu_notifier(struct kvm *kvm)
 357 {
 358         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 359         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 360 }
 361
 362 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 363
 364 static int kvm_init_mmu_notifier(struct kvm *kvm)
 365 {
 366         return 0;
 367 }
 368
 369 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 370
 371 static struct kvm *kvm_create_vm(void)
 372 {
 373         int r = 0, i;
 374         struct kvm *kvm = kvm_arch_create_vm();
 375
 376         if (IS_ERR(kvm))
 377                 goto out;
 378
 379         r = hardware_enable_all();
 380         if (r)
 381                 goto out_err_nodisable;
 382
 383 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 384         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 385         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 386 #endif
 387
 388         r = -ENOMEM;
 389         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 390         if (!kvm->memslots)
 391                 goto out_err;
 392         if (init_srcu_struct(&kvm->srcu))
 393                 goto out_err;
 394         for (i = 0; i < KVM_NR_BUSES; i++) {
 395                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 396                                         GFP_KERNEL);
 397                 if (!kvm->buses[i]) {
 398                         cleanup_srcu_struct(&kvm->srcu);
 399                         goto out_err;
 400                 }
 401         }
 402
 403         r = kvm_init_mmu_notifier(kvm);
 404         if (r) {
 405                 cleanup_srcu_struct(&kvm->srcu);
 406                 goto out_err;
 407         }
 408
 409         kvm->mm = current->mm;
 410         atomic_inc(&kvm->mm->mm_count);
 411         spin_lock_init(&kvm->mmu_lock);
 412         raw_spin_lock_init(&kvm->requests_lock);
 413         kvm_eventfd_init(kvm);
 414         mutex_init(&kvm->lock);
 415         mutex_init(&kvm->irq_lock);
 416         mutex_init(&kvm->slots_lock);
 417         atomic_set(&kvm->users_count, 1);
 418         spin_lock(&kvm_lock);
 419         list_add(&kvm->vm_list, &vm_list);
 420         spin_unlock(&kvm_lock);
 421 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 422         kvm_coalesced_mmio_init(kvm);
 423 #endif
 424 out:
 425         return kvm;
 426
 427 out_err:
 428         hardware_disable_all();
 429 out_err_nodisable:
 430         for (i = 0; i < KVM_NR_BUSES; i++)
 431                 kfree(kvm->buses[i]);
 432         kfree(kvm->memslots);
 433         kfree(kvm);
 434         return ERR_PTR(r);
 435 }
 436
 437 /*
 438  * Free any memory in @free but not in @dont.
 439  */
 440 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 441                                   struct kvm_memory_slot *dont)
 442 {
 443         int i;
 444
 445         if (!dont || free->rmap != dont->rmap)
 446                 vfree(free->rmap);
 447
 448         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 449                 vfree(free->dirty_bitmap);
 450
 451
 452         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 453                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 454                         vfree(free->lpage_info[i]);
 455                         free->lpage_info[i] = NULL;
 456                 }
 457         }
 458
 459         free->npages = 0;
 460         free->dirty_bitmap = NULL;
 461         free->rmap = NULL;
 462 }
 463
 464 void kvm_free_physmem(struct kvm *kvm)
 465 {
 466         int i;
 467         struct kvm_memslots *slots = kvm->memslots;
 468
 469         for (i = 0; i < slots->nmemslots; ++i)
 470                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 471
 472         kfree(kvm->memslots);
 473 }
 474
 475 static void kvm_destroy_vm(struct kvm *kvm)
 476 {
 477         int i;
 478         struct mm_struct *mm = kvm->mm;
 479
 480         kvm_arch_sync_events(kvm);
 481         spin_lock(&kvm_lock);
 482         list_del(&kvm->vm_list);
 483         spin_unlock(&kvm_lock);
 484         kvm_free_irq_routing(kvm);
 485         for (i = 0; i < KVM_NR_BUSES; i++)
 486                 kvm_io_bus_destroy(kvm->buses[i]);
 487         kvm_coalesced_mmio_free(kvm);
 488 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 489         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 490 #else
 491         kvm_arch_flush_shadow(kvm);
 492 #endif
 493         kvm_arch_destroy_vm(kvm);
 494         hardware_disable_all();
 495         mmdrop(mm);
 496 }
 497
 498 void kvm_get_kvm(struct kvm *kvm)
 499 {
 500         atomic_inc(&kvm->users_count);
 501 }
 502 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 503
 504 void kvm_put_kvm(struct kvm *kvm)
 505 {
 506         if (atomic_dec_and_test(&kvm->users_count))
 507                 kvm_destroy_vm(kvm);
 508 }
 509 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 510
 511
 512 static int kvm_vm_release(struct inode *inode, struct file *filp)
 513 {
 514         struct kvm *kvm = filp->private_data;
 515
 516         kvm_irqfd_release(kvm);
 517
 518         kvm_put_kvm(kvm);
 519         return 0;
 520 }
 521
 522 /*
 523  * Allocate some memory and give it an address in the guest physical address
 524  * space.
 525  *
 526  * Discontiguous memory is allowed, mostly for framebuffers.
 527  *
 528  * Must be called holding mmap_sem for write.
 529  */
 530 int __kvm_set_memory_region(struct kvm *kvm,
 531                             struct kvm_userspace_memory_region *mem,
 532                             int user_alloc)
 533 {
 534         int r, flush_shadow = 0;
 535         gfn_t base_gfn;
 536         unsigned long npages;
 537         unsigned long i;
 538         struct kvm_memory_slot *memslot;
 539         struct kvm_memory_slot old, new;
 540         struct kvm_memslots *slots, *old_memslots;
 541
 542         r = -EINVAL;
 543         /* General sanity checks */
 544         if (mem->memory_size & (PAGE_SIZE - 1))
 545                 goto out;
 546         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 547                 goto out;
 548         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 549                 goto out;
 550         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 551                 goto out;
 552         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 553                 goto out;
 554
 555         memslot = &kvm->memslots->memslots[mem->slot];
 556         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 557         npages = mem->memory_size >> PAGE_SHIFT;
 558
 559         if (!npages)
 560                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 561
 562         new = old = *memslot;
 563
 564         new.base_gfn = base_gfn;
 565         new.npages = npages;
 566         new.flags = mem->flags;
 567
 568         /* Disallow changing a memory slot's size. */
 569         r = -EINVAL;
 570         if (npages && old.npages && npages != old.npages)
 571                 goto out_free;
 572
 573         /* Check for overlaps */
 574         r = -EEXIST;
 575         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 576                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 577
 578                 if (s == memslot || !s->npages)
 579                         continue;
 580                 if (!((base_gfn + npages <= s->base_gfn) ||
 581                       (base_gfn >= s->base_gfn + s->npages)))
 582                         goto out_free;
 583         }
 584
 585         /* Free page dirty bitmap if unneeded */
 586         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 587                 new.dirty_bitmap = NULL;
 588
 589         r = -ENOMEM;
 590
 591         /* Allocate if a slot is being created */
 592 #ifndef CONFIG_S390
 593         if (npages && !new.rmap) {
 594                 new.rmap = vmalloc(npages * sizeof(struct page *));
 595
 596                 if (!new.rmap)
 597                         goto out_free;
 598
 599                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 600
 601                 new.user_alloc = user_alloc;
 602                 new.userspace_addr = mem->userspace_addr;
 603         }
 604         if (!npages)
 605                 goto skip_lpage;
 606
 607         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 608                 unsigned long ugfn;
 609                 unsigned long j;
 610                 int lpages;
 611                 int level = i + 2;
 612
 613                 /* Avoid unused variable warning if no large pages */
 614                 (void)level;
 615
 616                 if (new.lpage_info[i])
 617                         continue;
 618
 619                 lpages = 1 + (base_gfn + npages - 1) /
 620                              KVM_PAGES_PER_HPAGE(level);
 621                 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
 622
 623                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
 624
 625                 if (!new.lpage_info[i])
 626                         goto out_free;
 627
 628                 memset(new.lpage_info[i], 0,
 629                        lpages * sizeof(*new.lpage_info[i]));
 630
 631                 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
 632                         new.lpage_info[i][0].write_count = 1;
 633                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
 634                         new.lpage_info[i][lpages - 1].write_count = 1;
 635                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 636                 /*
 637                  * If the gfn and userspace address are not aligned wrt each
 638                  * other, or if explicitly asked to, disable large page
 639                  * support for this slot
 640                  */
 641                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 642                     !largepages_enabled)
 643                         for (j = 0; j < lpages; ++j)
 644                                 new.lpage_info[i][j].write_count = 1;
 645         }
 646
 647 skip_lpage:
 648
 649         /* Allocate page dirty bitmap if needed */
 650         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 651                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 652
 653                 new.dirty_bitmap = vmalloc(dirty_bytes);
 654                 if (!new.dirty_bitmap)
 655                         goto out_free;
 656                 memset(new.dirty_bitmap, 0, dirty_bytes);
 657                 /* destroy any largepage mappings for dirty tracking */
 658                 if (old.npages)
 659                         flush_shadow = 1;
 660         }
 661 #else  /* not defined CONFIG_S390 */
 662         new.user_alloc = user_alloc;
 663         if (user_alloc)
 664                 new.userspace_addr = mem->userspace_addr;
 665 #endif /* not defined CONFIG_S390 */
 666
 667         if (!npages) {
 668                 r = -ENOMEM;
 669                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 670                 if (!slots)
 671                         goto out_free;
 672                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 673                 if (mem->slot >= slots->nmemslots)
 674                         slots->nmemslots = mem->slot + 1;
 675                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 676
 677                 old_memslots = kvm->memslots;
 678                 rcu_assign_pointer(kvm->memslots, slots);
 679                 synchronize_srcu_expedited(&kvm->srcu);
 680                 /* From this point no new shadow pages pointing to a deleted
 681                  * memslot will be created.
 682                  *
 683                  * validation of sp->gfn happens in:
 684                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 685                  *      - kvm_is_visible_gfn (mmu_check_roots)
 686                  */
 687                 kvm_arch_flush_shadow(kvm);
 688                 kfree(old_memslots);
 689         }
 690
 691         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 692         if (r)
 693                 goto out_free;
 694
 695 #ifdef CONFIG_DMAR
 696         /* map the pages in iommu page table */
 697         if (npages) {
 698                 r = kvm_iommu_map_pages(kvm, &new);
 699                 if (r)
 700                         goto out_free;
 701         }
 702 #endif
 703
 704         r = -ENOMEM;
 705         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 706         if (!slots)
 707                 goto out_free;
 708         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 709         if (mem->slot >= slots->nmemslots)
 710                 slots->nmemslots = mem->slot + 1;
 711
 712         /* actual memory is freed via old in kvm_free_physmem_slot below */
 713         if (!npages) {
 714                 new.rmap = NULL;
 715                 new.dirty_bitmap = NULL;
 716                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 717                         new.lpage_info[i] = NULL;
 718         }
 719
 720         slots->memslots[mem->slot] = new;
 721         old_memslots = kvm->memslots;
 722         rcu_assign_pointer(kvm->memslots, slots);
 723         synchronize_srcu_expedited(&kvm->srcu);
 724
 725         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 726
 727         kvm_free_physmem_slot(&old, &new);
 728         kfree(old_memslots);
 729
 730         if (flush_shadow)
 731                 kvm_arch_flush_shadow(kvm);
 732
 733         return 0;
 734
 735 out_free:
 736         kvm_free_physmem_slot(&new, &old);
 737 out:
 738         return r;
 739
 740 }
 741 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 742
 743 int kvm_set_memory_region(struct kvm *kvm,
 744                           struct kvm_userspace_memory_region *mem,
 745                           int user_alloc)
 746 {
 747         int r;
 748
 749         mutex_lock(&kvm->slots_lock);
 750         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 751         mutex_unlock(&kvm->slots_lock);
 752         return r;
 753 }
 754 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 755
 756 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 757                                    struct
 758                                    kvm_userspace_memory_region *mem,
 759                                    int user_alloc)
 760 {
 761         if (mem->slot >= KVM_MEMORY_SLOTS)
 762                 return -EINVAL;
 763         return kvm_set_memory_region(kvm, mem, user_alloc);
 764 }
 765
 766 int kvm_get_dirty_log(struct kvm *kvm,
 767                         struct kvm_dirty_log *log, int *is_dirty)
 768 {
 769         struct kvm_memory_slot *memslot;
 770         int r, i;
 771         int n;
 772         unsigned long any = 0;
 773
 774         r = -EINVAL;
 775         if (log->slot >= KVM_MEMORY_SLOTS)
 776                 goto out;
 777
 778         memslot = &kvm->memslots->memslots[log->slot];
 779         r = -ENOENT;
 780         if (!memslot->dirty_bitmap)
 781                 goto out;
 782
 783         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 784
 785         for (i = 0; !any && i < n/sizeof(long); ++i)
 786                 any = memslot->dirty_bitmap[i];
 787
 788         r = -EFAULT;
 789         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 790                 goto out;
 791
 792         if (any)
 793                 *is_dirty = 1;
 794
 795         r = 0;
 796 out:
 797         return r;
 798 }
 799
 800 void kvm_disable_largepages(void)
 801 {
 802         largepages_enabled = false;
 803 }
 804 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 805
 806 int is_error_page(struct page *page)
 807 {
 808         return page == bad_page;
 809 }
 810 EXPORT_SYMBOL_GPL(is_error_page);
 811
 812 int is_error_pfn(pfn_t pfn)
 813 {
 814         return pfn == bad_pfn;
 815 }
 816 EXPORT_SYMBOL_GPL(is_error_pfn);
 817
 818 static inline unsigned long bad_hva(void)
 819 {
 820         return PAGE_OFFSET;
 821 }
 822
 823 int kvm_is_error_hva(unsigned long addr)
 824 {
 825         return addr == bad_hva();
 826 }
 827 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 828
 829 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
 830 {
 831         int i;
 832         struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
 833
 834         for (i = 0; i < slots->nmemslots; ++i) {
 835                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 836
 837                 if (gfn >= memslot->base_gfn
 838                     && gfn < memslot->base_gfn + memslot->npages)
 839                         return memslot;
 840         }
 841         return NULL;
 842 }
 843 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
 844
 845 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 846 {
 847         gfn = unalias_gfn(kvm, gfn);
 848         return gfn_to_memslot_unaliased(kvm, gfn);
 849 }
 850
 851 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 852 {
 853         int i;
 854         struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
 855
 856         gfn = unalias_gfn_instantiation(kvm, gfn);
 857         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 858                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 859
 860                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 861                         continue;
 862
 863                 if (gfn >= memslot->base_gfn
 864                     && gfn < memslot->base_gfn + memslot->npages)
 865                         return 1;
 866         }
 867         return 0;
 868 }
 869 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 870
 871 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 872 {
 873         struct vm_area_struct *vma;
 874         unsigned long addr, size;
 875
 876         size = PAGE_SIZE;
 877
 878         addr = gfn_to_hva(kvm, gfn);
 879         if (kvm_is_error_hva(addr))
 880                 return PAGE_SIZE;
 881
 882         down_read(&current->mm->mmap_sem);
 883         vma = find_vma(current->mm, addr);
 884         if (!vma)
 885                 goto out;
 886
 887         size = vma_kernel_pagesize(vma);
 888
 889 out:
 890         up_read(&current->mm->mmap_sem);
 891
 892         return size;
 893 }
 894
 895 int memslot_id(struct kvm *kvm, gfn_t gfn)
 896 {
 897         int i;
 898         struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
 899         struct kvm_memory_slot *memslot = NULL;
 900
 901         gfn = unalias_gfn(kvm, gfn);
 902         for (i = 0; i < slots->nmemslots; ++i) {
 903                 memslot = &slots->memslots[i];
 904
 905                 if (gfn >= memslot->base_gfn
 906                     && gfn < memslot->base_gfn + memslot->npages)
 907                         break;
 908         }
 909
 910         return memslot - slots->memslots;
 911 }
 912
 913 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 914 {
 915         struct kvm_memory_slot *slot;
 916
 917         gfn = unalias_gfn_instantiation(kvm, gfn);
 918         slot = gfn_to_memslot_unaliased(kvm, gfn);
 919         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 920                 return bad_hva();
 921         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 922 }
 923 EXPORT_SYMBOL_GPL(gfn_to_hva);
 924
 925 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr)
 926 {
 927         struct page *page[1];
 928         int npages;
 929         pfn_t pfn;
 930
 931         might_sleep();
 932
 933         npages = get_user_pages_fast(addr, 1, 1, page);
 934
 935         if (unlikely(npages != 1)) {
 936                 struct vm_area_struct *vma;
 937
 938                 down_read(&current->mm->mmap_sem);
 939                 vma = find_vma(current->mm, addr);
 940
 941                 if (vma == NULL || addr < vma->vm_start ||
 942                     !(vma->vm_flags & VM_PFNMAP)) {
 943                         up_read(&current->mm->mmap_sem);
 944                         get_page(bad_page);
 945                         return page_to_pfn(bad_page);
 946                 }
 947
 948                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 949                 up_read(&current->mm->mmap_sem);
 950                 BUG_ON(!kvm_is_mmio_pfn(pfn));
 951         } else
 952                 pfn = page_to_pfn(page[0]);
 953
 954         return pfn;
 955 }
 956
 957 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 958 {
 959         unsigned long addr;
 960
 961         addr = gfn_to_hva(kvm, gfn);
 962         if (kvm_is_error_hva(addr)) {
 963                 get_page(bad_page);
 964                 return page_to_pfn(bad_page);
 965         }
 966
 967         return hva_to_pfn(kvm, addr);
 968 }
 969 EXPORT_SYMBOL_GPL(gfn_to_pfn);
 970
 971 static unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
 972 {
 973         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 974 }
 975
 976 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
 977                          struct kvm_memory_slot *slot, gfn_t gfn)
 978 {
 979         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
 980         return hva_to_pfn(kvm, addr);
 981 }
 982
 983 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 984 {
 985         pfn_t pfn;
 986
 987         pfn = gfn_to_pfn(kvm, gfn);
 988         if (!kvm_is_mmio_pfn(pfn))
 989                 return pfn_to_page(pfn);
 990
 991         WARN_ON(kvm_is_mmio_pfn(pfn));
 992
 993         get_page(bad_page);
 994         return bad_page;
 995 }
 996
 997 EXPORT_SYMBOL_GPL(gfn_to_page);
 998
 999 void kvm_release_page_clean(struct page *page)
1000 {
1001         kvm_release_pfn_clean(page_to_pfn(page));
1002 }
1003 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1004
1005 void kvm_release_pfn_clean(pfn_t pfn)
1006 {
1007         if (!kvm_is_mmio_pfn(pfn))
1008                 put_page(pfn_to_page(pfn));
1009 }
1010 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1011
1012 void kvm_release_page_dirty(struct page *page)
1013 {
1014         kvm_release_pfn_dirty(page_to_pfn(page));
1015 }
1016 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1017
1018 void kvm_release_pfn_dirty(pfn_t pfn)
1019 {
1020         kvm_set_pfn_dirty(pfn);
1021         kvm_release_pfn_clean(pfn);
1022 }
1023 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1024
1025 void kvm_set_page_dirty(struct page *page)
1026 {
1027         kvm_set_pfn_dirty(page_to_pfn(page));
1028 }
1029 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1030
1031 void kvm_set_pfn_dirty(pfn_t pfn)
1032 {
1033         if (!kvm_is_mmio_pfn(pfn)) {
1034                 struct page *page = pfn_to_page(pfn);
1035                 if (!PageReserved(page))
1036                         SetPageDirty(page);
1037         }
1038 }
1039 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1040
1041 void kvm_set_pfn_accessed(pfn_t pfn)
1042 {
1043         if (!kvm_is_mmio_pfn(pfn))
1044                 mark_page_accessed(pfn_to_page(pfn));
1045 }
1046 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1047
1048 void kvm_get_pfn(pfn_t pfn)
1049 {
1050         if (!kvm_is_mmio_pfn(pfn))
1051                 get_page(pfn_to_page(pfn));
1052 }
1053 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1054
1055 static int next_segment(unsigned long len, int offset)
1056 {
1057         if (len > PAGE_SIZE - offset)
1058                 return PAGE_SIZE - offset;
1059         else
1060                 return len;
1061 }
1062
1063 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1064                         int len)
1065 {
1066         int r;
1067         unsigned long addr;
1068
1069         addr = gfn_to_hva(kvm, gfn);
1070         if (kvm_is_error_hva(addr))
1071                 return -EFAULT;
1072         r = copy_from_user(data, (void __user *)addr + offset, len);
1073         if (r)
1074                 return -EFAULT;
1075         return 0;
1076 }
1077 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1078
1079 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1080 {
1081         gfn_t gfn = gpa >> PAGE_SHIFT;
1082         int seg;
1083         int offset = offset_in_page(gpa);
1084         int ret;
1085
1086         while ((seg = next_segment(len, offset)) != 0) {
1087                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1088                 if (ret < 0)
1089                         return ret;
1090                 offset = 0;
1091                 len -= seg;
1092                 data += seg;
1093                 ++gfn;
1094         }
1095         return 0;
1096 }
1097 EXPORT_SYMBOL_GPL(kvm_read_guest);
1098
1099 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1100                           unsigned long len)
1101 {
1102         int r;
1103         unsigned long addr;
1104         gfn_t gfn = gpa >> PAGE_SHIFT;
1105         int offset = offset_in_page(gpa);
1106
1107         addr = gfn_to_hva(kvm, gfn);
1108         if (kvm_is_error_hva(addr))
1109                 return -EFAULT;
1110         pagefault_disable();
1111         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1112         pagefault_enable();
1113         if (r)
1114                 return -EFAULT;
1115         return 0;
1116 }
1117 EXPORT_SYMBOL(kvm_read_guest_atomic);
1118
1119 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1120                          int offset, int len)
1121 {
1122         int r;
1123         unsigned long addr;
1124
1125         addr = gfn_to_hva(kvm, gfn);
1126         if (kvm_is_error_hva(addr))
1127                 return -EFAULT;
1128         r = copy_to_user((void __user *)addr + offset, data, len);
1129         if (r)
1130                 return -EFAULT;
1131         mark_page_dirty(kvm, gfn);
1132         return 0;
1133 }
1134 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1135
1136 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1137                     unsigned long len)
1138 {
1139         gfn_t gfn = gpa >> PAGE_SHIFT;
1140         int seg;
1141         int offset = offset_in_page(gpa);
1142         int ret;
1143
1144         while ((seg = next_segment(len, offset)) != 0) {
1145                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1146                 if (ret < 0)
1147                         return ret;
1148                 offset = 0;
1149                 len -= seg;
1150                 data += seg;
1151                 ++gfn;
1152         }
1153         return 0;
1154 }
1155
1156 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1157 {
1158         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1159 }
1160 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1161
1162 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1163 {
1164         gfn_t gfn = gpa >> PAGE_SHIFT;
1165         int seg;
1166         int offset = offset_in_page(gpa);
1167         int ret;
1168
1169         while ((seg = next_segment(len, offset)) != 0) {
1170                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1171                 if (ret < 0)
1172                         return ret;
1173                 offset = 0;
1174                 len -= seg;
1175                 ++gfn;
1176         }
1177         return 0;
1178 }
1179 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1180
1181 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1182 {
1183         struct kvm_memory_slot *memslot;
1184
1185         gfn = unalias_gfn(kvm, gfn);
1186         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1187         if (memslot && memslot->dirty_bitmap) {
1188                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1189
1190                 /* avoid RMW */
1191                 if (!generic_test_le_bit(rel_gfn, memslot->dirty_bitmap))
1192                         generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1193         }
1194 }
1195
1196 /*
1197  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1198  */
1199 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1200 {
1201         DEFINE_WAIT(wait);
1202
1203         for (;;) {
1204                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1205
1206                 if (kvm_arch_vcpu_runnable(vcpu)) {
1207                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1208                         break;
1209                 }
1210                 if (kvm_cpu_has_pending_timer(vcpu))
1211                         break;
1212                 if (signal_pending(current))
1213                         break;
1214
1215                 schedule();
1216         }
1217
1218         finish_wait(&vcpu->wq, &wait);
1219 }
1220
1221 void kvm_resched(struct kvm_vcpu *vcpu)
1222 {
1223         if (!need_resched())
1224                 return;
1225         cond_resched();
1226 }
1227 EXPORT_SYMBOL_GPL(kvm_resched);
1228
1229 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1230 {
1231         ktime_t expires;
1232         DEFINE_WAIT(wait);
1233
1234         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1235
1236         /* Sleep for 100 us, and hope lock-holder got scheduled */
1237         expires = ktime_add_ns(ktime_get(), 100000UL);
1238         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1239
1240         finish_wait(&vcpu->wq, &wait);
1241 }
1242 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1243
1244 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1245 {
1246         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1247         struct page *page;
1248
1249         if (vmf->pgoff == 0)
1250                 page = virt_to_page(vcpu->run);
1251 #ifdef CONFIG_X86
1252         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1253                 page = virt_to_page(vcpu->arch.pio_data);
1254 #endif
1255 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1256         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1257                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1258 #endif
1259         else
1260                 return VM_FAULT_SIGBUS;
1261         get_page(page);
1262         vmf->page = page;
1263         return 0;
1264 }
1265
1266 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1267         .fault = kvm_vcpu_fault,
1268 };
1269
1270 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1271 {
1272         vma->vm_ops = &kvm_vcpu_vm_ops;
1273         return 0;
1274 }
1275
1276 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1277 {
1278         struct kvm_vcpu *vcpu = filp->private_data;
1279
1280         kvm_put_kvm(vcpu->kvm);
1281         return 0;
1282 }
1283
1284 static struct file_operations kvm_vcpu_fops = {
1285         .release        = kvm_vcpu_release,
1286         .unlocked_ioctl = kvm_vcpu_ioctl,
1287         .compat_ioctl   = kvm_vcpu_ioctl,
1288         .mmap           = kvm_vcpu_mmap,
1289 };
1290
1291 /*
1292  * Allocates an inode for the vcpu.
1293  */
1294 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1295 {
1296         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1297 }
1298
1299 /*
1300  * Creates some virtual cpus.  Good luck creating more than one.
1301  */
1302 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1303 {
1304         int r;
1305         struct kvm_vcpu *vcpu, *v;
1306
1307         vcpu = kvm_arch_vcpu_create(kvm, id);
1308         if (IS_ERR(vcpu))
1309                 return PTR_ERR(vcpu);
1310
1311         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1312
1313         r = kvm_arch_vcpu_setup(vcpu);
1314         if (r)
1315                 return r;
1316
1317         mutex_lock(&kvm->lock);
1318         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1319                 r = -EINVAL;
1320                 goto vcpu_destroy;
1321         }
1322
1323         kvm_for_each_vcpu(r, v, kvm)
1324                 if (v->vcpu_id == id) {
1325                         r = -EEXIST;
1326                         goto vcpu_destroy;
1327                 }
1328
1329         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1330
1331         /* Now it's all set up, let userspace reach it */
1332         kvm_get_kvm(kvm);
1333         r = create_vcpu_fd(vcpu);
1334         if (r < 0) {
1335                 kvm_put_kvm(kvm);
1336                 goto vcpu_destroy;
1337         }
1338
1339         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1340         smp_wmb();
1341         atomic_inc(&kvm->online_vcpus);
1342
1343 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1344         if (kvm->bsp_vcpu_id == id)
1345                 kvm->bsp_vcpu = vcpu;
1346 #endif
1347         mutex_unlock(&kvm->lock);
1348         return r;
1349
1350 vcpu_destroy:
1351         mutex_unlock(&kvm->lock);
1352         kvm_arch_vcpu_destroy(vcpu);
1353         return r;
1354 }
1355
1356 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1357 {
1358         if (sigset) {
1359                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1360                 vcpu->sigset_active = 1;
1361                 vcpu->sigset = *sigset;
1362         } else
1363                 vcpu->sigset_active = 0;
1364         return 0;
1365 }
1366
1367 static long kvm_vcpu_ioctl(struct file *filp,
1368                            unsigned int ioctl, unsigned long arg)
1369 {
1370         struct kvm_vcpu *vcpu = filp->private_data;
1371         void __user *argp = (void __user *)arg;
1372         int r;
1373         struct kvm_fpu *fpu = NULL;
1374         struct kvm_sregs *kvm_sregs = NULL;
1375
1376         if (vcpu->kvm->mm != current->mm)
1377                 return -EIO;
1378         switch (ioctl) {
1379         case KVM_RUN:
1380                 r = -EINVAL;
1381                 if (arg)
1382                         goto out;
1383                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1384                 break;
1385         case KVM_GET_REGS: {
1386                 struct kvm_regs *kvm_regs;
1387
1388                 r = -ENOMEM;
1389                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1390                 if (!kvm_regs)
1391                         goto out;
1392                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1393                 if (r)
1394                         goto out_free1;
1395                 r = -EFAULT;
1396                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1397                         goto out_free1;
1398                 r = 0;
1399 out_free1:
1400                 kfree(kvm_regs);
1401                 break;
1402         }
1403         case KVM_SET_REGS: {
1404                 struct kvm_regs *kvm_regs;
1405
1406                 r = -ENOMEM;
1407                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1408                 if (!kvm_regs)
1409                         goto out;
1410                 r = -EFAULT;
1411                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1412                         goto out_free2;
1413                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1414                 if (r)
1415                         goto out_free2;
1416                 r = 0;
1417 out_free2:
1418                 kfree(kvm_regs);
1419                 break;
1420         }
1421         case KVM_GET_SREGS: {
1422                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1423                 r = -ENOMEM;
1424                 if (!kvm_sregs)
1425                         goto out;
1426                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1427                 if (r)
1428                         goto out;
1429                 r = -EFAULT;
1430                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1431                         goto out;
1432                 r = 0;
1433                 break;
1434         }
1435         case KVM_SET_SREGS: {
1436                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1437                 r = -ENOMEM;
1438                 if (!kvm_sregs)
1439                         goto out;
1440                 r = -EFAULT;
1441                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1442                         goto out;
1443                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1444                 if (r)
1445                         goto out;
1446                 r = 0;
1447                 break;
1448         }
1449         case KVM_GET_MP_STATE: {
1450                 struct kvm_mp_state mp_state;
1451
1452                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1453                 if (r)
1454                         goto out;
1455                 r = -EFAULT;
1456                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1457                         goto out;
1458                 r = 0;
1459                 break;
1460         }
1461         case KVM_SET_MP_STATE: {
1462                 struct kvm_mp_state mp_state;
1463
1464                 r = -EFAULT;
1465                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1466                         goto out;
1467                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1468                 if (r)
1469                         goto out;
1470                 r = 0;
1471                 break;
1472         }
1473         case KVM_TRANSLATE: {
1474                 struct kvm_translation tr;
1475
1476                 r = -EFAULT;
1477                 if (copy_from_user(&tr, argp, sizeof tr))
1478                         goto out;
1479                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1480                 if (r)
1481                         goto out;
1482                 r = -EFAULT;
1483                 if (copy_to_user(argp, &tr, sizeof tr))
1484                         goto out;
1485                 r = 0;
1486                 break;
1487         }
1488         case KVM_SET_GUEST_DEBUG: {
1489                 struct kvm_guest_debug dbg;
1490
1491                 r = -EFAULT;
1492                 if (copy_from_user(&dbg, argp, sizeof dbg))
1493                         goto out;
1494                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1495                 if (r)
1496                         goto out;
1497                 r = 0;
1498                 break;
1499         }
1500         case KVM_SET_SIGNAL_MASK: {
1501                 struct kvm_signal_mask __user *sigmask_arg = argp;
1502                 struct kvm_signal_mask kvm_sigmask;
1503                 sigset_t sigset, *p;
1504
1505                 p = NULL;
1506                 if (argp) {
1507                         r = -EFAULT;
1508                         if (copy_from_user(&kvm_sigmask, argp,
1509                                            sizeof kvm_sigmask))
1510                                 goto out;
1511                         r = -EINVAL;
1512                         if (kvm_sigmask.len != sizeof sigset)
1513                                 goto out;
1514                         r = -EFAULT;
1515                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1516                                            sizeof sigset))
1517                                 goto out;
1518                         p = &sigset;
1519                 }
1520                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1521                 break;
1522         }
1523         case KVM_GET_FPU: {
1524                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1525                 r = -ENOMEM;
1526                 if (!fpu)
1527                         goto out;
1528                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1529                 if (r)
1530                         goto out;
1531                 r = -EFAULT;
1532                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1533                         goto out;
1534                 r = 0;
1535                 break;
1536         }
1537         case KVM_SET_FPU: {
1538                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1539                 r = -ENOMEM;
1540                 if (!fpu)
1541                         goto out;
1542                 r = -EFAULT;
1543                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1544                         goto out;
1545                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1546                 if (r)
1547                         goto out;
1548                 r = 0;
1549                 break;
1550         }
1551         default:
1552                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1553         }
1554 out:
1555         kfree(fpu);
1556         kfree(kvm_sregs);
1557         return r;
1558 }
1559
1560 static long kvm_vm_ioctl(struct file *filp,
1561                            unsigned int ioctl, unsigned long arg)
1562 {
1563         struct kvm *kvm = filp->private_data;
1564         void __user *argp = (void __user *)arg;
1565         int r;
1566
1567         if (kvm->mm != current->mm)
1568                 return -EIO;
1569         switch (ioctl) {
1570         case KVM_CREATE_VCPU:
1571                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1572                 if (r < 0)
1573                         goto out;
1574                 break;
1575         case KVM_SET_USER_MEMORY_REGION: {
1576                 struct kvm_userspace_memory_region kvm_userspace_mem;
1577
1578                 r = -EFAULT;
1579                 if (copy_from_user(&kvm_userspace_mem, argp,
1580                                                 sizeof kvm_userspace_mem))
1581                         goto out;
1582
1583                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1584                 if (r)
1585                         goto out;
1586                 break;
1587         }
1588         case KVM_GET_DIRTY_LOG: {
1589                 struct kvm_dirty_log log;
1590
1591                 r = -EFAULT;
1592                 if (copy_from_user(&log, argp, sizeof log))
1593                         goto out;
1594                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1595                 if (r)
1596                         goto out;
1597                 break;
1598         }
1599 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1600         case KVM_REGISTER_COALESCED_MMIO: {
1601                 struct kvm_coalesced_mmio_zone zone;
1602                 r = -EFAULT;
1603                 if (copy_from_user(&zone, argp, sizeof zone))
1604                         goto out;
1605                 r = -ENXIO;
1606                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1607                 if (r)
1608                         goto out;
1609                 r = 0;
1610                 break;
1611         }
1612         case KVM_UNREGISTER_COALESCED_MMIO: {
1613                 struct kvm_coalesced_mmio_zone zone;
1614                 r = -EFAULT;
1615                 if (copy_from_user(&zone, argp, sizeof zone))
1616                         goto out;
1617                 r = -ENXIO;
1618                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1619                 if (r)
1620                         goto out;
1621                 r = 0;
1622                 break;
1623         }
1624 #endif
1625         case KVM_IRQFD: {
1626                 struct kvm_irqfd data;
1627
1628                 r = -EFAULT;
1629                 if (copy_from_user(&data, argp, sizeof data))
1630                         goto out;
1631                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1632                 break;
1633         }
1634         case KVM_IOEVENTFD: {
1635                 struct kvm_ioeventfd data;
1636
1637                 r = -EFAULT;
1638                 if (copy_from_user(&data, argp, sizeof data))
1639                         goto out;
1640                 r = kvm_ioeventfd(kvm, &data);
1641                 break;
1642         }
1643 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1644         case KVM_SET_BOOT_CPU_ID:
1645                 r = 0;
1646                 mutex_lock(&kvm->lock);
1647                 if (atomic_read(&kvm->online_vcpus) != 0)
1648                         r = -EBUSY;
1649                 else
1650                         kvm->bsp_vcpu_id = arg;
1651                 mutex_unlock(&kvm->lock);
1652                 break;
1653 #endif
1654         default:
1655                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1656                 if (r == -ENOTTY)
1657                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1658         }
1659 out:
1660         return r;
1661 }
1662
1663 #ifdef CONFIG_COMPAT
1664 struct compat_kvm_dirty_log {
1665         __u32 slot;
1666         __u32 padding1;
1667         union {
1668                 compat_uptr_t dirty_bitmap; /* one bit per page */
1669                 __u64 padding2;
1670         };
1671 };
1672
1673 static long kvm_vm_compat_ioctl(struct file *filp,
1674                            unsigned int ioctl, unsigned long arg)
1675 {
1676         struct kvm *kvm = filp->private_data;
1677         int r;
1678
1679         if (kvm->mm != current->mm)
1680                 return -EIO;
1681         switch (ioctl) {
1682         case KVM_GET_DIRTY_LOG: {
1683                 struct compat_kvm_dirty_log compat_log;
1684                 struct kvm_dirty_log log;
1685
1686                 r = -EFAULT;
1687                 if (copy_from_user(&compat_log, (void __user *)arg,
1688                                    sizeof(compat_log)))
1689                         goto out;
1690                 log.slot         = compat_log.slot;
1691                 log.padding1     = compat_log.padding1;
1692                 log.padding2     = compat_log.padding2;
1693                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1694
1695                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1696                 if (r)
1697                         goto out;
1698                 break;
1699         }
1700         default:
1701                 r = kvm_vm_ioctl(filp, ioctl, arg);
1702         }
1703
1704 out:
1705         return r;
1706 }
1707 #endif
1708
1709 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1710 {
1711         struct page *page[1];
1712         unsigned long addr;
1713         int npages;
1714         gfn_t gfn = vmf->pgoff;
1715         struct kvm *kvm = vma->vm_file->private_data;
1716
1717         addr = gfn_to_hva(kvm, gfn);
1718         if (kvm_is_error_hva(addr))
1719                 return VM_FAULT_SIGBUS;
1720
1721         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1722                                 NULL);
1723         if (unlikely(npages != 1))
1724                 return VM_FAULT_SIGBUS;
1725
1726         vmf->page = page[0];
1727         return 0;
1728 }
1729
1730 static const struct vm_operations_struct kvm_vm_vm_ops = {
1731         .fault = kvm_vm_fault,
1732 };
1733
1734 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1735 {
1736         vma->vm_ops = &kvm_vm_vm_ops;
1737         return 0;
1738 }
1739
1740 static struct file_operations kvm_vm_fops = {
1741         .release        = kvm_vm_release,
1742         .unlocked_ioctl = kvm_vm_ioctl,
1743 #ifdef CONFIG_COMPAT
1744         .compat_ioctl   = kvm_vm_compat_ioctl,
1745 #endif
1746         .mmap           = kvm_vm_mmap,
1747 };
1748
1749 static int kvm_dev_ioctl_create_vm(void)
1750 {
1751         int fd;
1752         struct kvm *kvm;
1753
1754         kvm = kvm_create_vm();
1755         if (IS_ERR(kvm))
1756                 return PTR_ERR(kvm);
1757         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1758         if (fd < 0)
1759                 kvm_put_kvm(kvm);
1760
1761         return fd;
1762 }
1763
1764 static long kvm_dev_ioctl_check_extension_generic(long arg)
1765 {
1766         switch (arg) {
1767         case KVM_CAP_USER_MEMORY:
1768         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1769         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1770 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1771         case KVM_CAP_SET_BOOT_CPU_ID:
1772 #endif
1773         case KVM_CAP_INTERNAL_ERROR_DATA:
1774                 return 1;
1775 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1776         case KVM_CAP_IRQ_ROUTING:
1777                 return KVM_MAX_IRQ_ROUTES;
1778 #endif
1779         default:
1780                 break;
1781         }
1782         return kvm_dev_ioctl_check_extension(arg);
1783 }
1784
1785 static long kvm_dev_ioctl(struct file *filp,
1786                           unsigned int ioctl, unsigned long arg)
1787 {
1788         long r = -EINVAL;
1789
1790         switch (ioctl) {
1791         case KVM_GET_API_VERSION:
1792                 r = -EINVAL;
1793                 if (arg)
1794                         goto out;
1795                 r = KVM_API_VERSION;
1796                 break;
1797         case KVM_CREATE_VM:
1798                 r = -EINVAL;
1799                 if (arg)
1800                         goto out;
1801                 r = kvm_dev_ioctl_create_vm();
1802                 break;
1803         case KVM_CHECK_EXTENSION:
1804                 r = kvm_dev_ioctl_check_extension_generic(arg);
1805                 break;
1806         case KVM_GET_VCPU_MMAP_SIZE:
1807                 r = -EINVAL;
1808                 if (arg)
1809                         goto out;
1810                 r = PAGE_SIZE;     /* struct kvm_run */
1811 #ifdef CONFIG_X86
1812                 r += PAGE_SIZE;    /* pio data page */
1813 #endif
1814 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1815                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1816 #endif
1817                 break;
1818         case KVM_TRACE_ENABLE:
1819         case KVM_TRACE_PAUSE:
1820         case KVM_TRACE_DISABLE:
1821                 r = -EOPNOTSUPP;
1822                 break;
1823         default:
1824                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1825         }
1826 out:
1827         return r;
1828 }
1829
1830 static struct file_operations kvm_chardev_ops = {
1831         .unlocked_ioctl = kvm_dev_ioctl,
1832         .compat_ioctl   = kvm_dev_ioctl,
1833 };
1834
1835 static struct miscdevice kvm_dev = {
1836         KVM_MINOR,
1837         "kvm",
1838         &kvm_chardev_ops,
1839 };
1840
1841 static void hardware_enable(void *junk)
1842 {
1843         int cpu = raw_smp_processor_id();
1844         int r;
1845
1846         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1847                 return;
1848
1849         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1850
1851         r = kvm_arch_hardware_enable(NULL);
1852
1853         if (r) {
1854                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1855                 atomic_inc(&hardware_enable_failed);
1856                 printk(KERN_INFO "kvm: enabling virtualization on "
1857                                  "CPU%d failed\n", cpu);
1858         }
1859 }
1860
1861 static void hardware_disable(void *junk)
1862 {
1863         int cpu = raw_smp_processor_id();
1864
1865         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1866                 return;
1867         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1868         kvm_arch_hardware_disable(NULL);
1869 }
1870
1871 static void hardware_disable_all_nolock(void)
1872 {
1873         BUG_ON(!kvm_usage_count);
1874
1875         kvm_usage_count--;
1876         if (!kvm_usage_count)
1877                 on_each_cpu(hardware_disable, NULL, 1);
1878 }
1879
1880 static void hardware_disable_all(void)
1881 {
1882         spin_lock(&kvm_lock);
1883         hardware_disable_all_nolock();
1884         spin_unlock(&kvm_lock);
1885 }
1886
1887 static int hardware_enable_all(void)
1888 {
1889         int r = 0;
1890
1891         spin_lock(&kvm_lock);
1892
1893         kvm_usage_count++;
1894         if (kvm_usage_count == 1) {
1895                 atomic_set(&hardware_enable_failed, 0);
1896                 on_each_cpu(hardware_enable, NULL, 1);
1897
1898                 if (atomic_read(&hardware_enable_failed)) {
1899                         hardware_disable_all_nolock();
1900                         r = -EBUSY;
1901                 }
1902         }
1903
1904         spin_unlock(&kvm_lock);
1905
1906         return r;
1907 }
1908
1909 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1910                            void *v)
1911 {
1912         int cpu = (long)v;
1913
1914         if (!kvm_usage_count)
1915                 return NOTIFY_OK;
1916
1917         val &= ~CPU_TASKS_FROZEN;
1918         switch (val) {
1919         case CPU_DYING:
1920                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1921                        cpu);
1922                 hardware_disable(NULL);
1923                 break;
1924         case CPU_UP_CANCELED:
1925                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1926                        cpu);
1927                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1928                 break;
1929         case CPU_ONLINE:
1930                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1931                        cpu);
1932                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1933                 break;
1934         }
1935         return NOTIFY_OK;
1936 }
1937
1938
1939 asmlinkage void kvm_handle_fault_on_reboot(void)
1940 {
1941         if (kvm_rebooting)
1942                 /* spin while reset goes on */
1943                 while (true)
1944                         ;
1945         /* Fault while not rebooting.  We want the trace. */
1946         BUG();
1947 }
1948 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1949
1950 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1951                       void *v)
1952 {
1953         /*
1954          * Some (well, at least mine) BIOSes hang on reboot if
1955          * in vmx root mode.
1956          *
1957          * And Intel TXT required VMX off for all cpu when system shutdown.
1958          */
1959         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1960         kvm_rebooting = true;
1961         on_each_cpu(hardware_disable, NULL, 1);
1962         return NOTIFY_OK;
1963 }
1964
1965 static struct notifier_block kvm_reboot_notifier = {
1966         .notifier_call = kvm_reboot,
1967         .priority = 0,
1968 };
1969
1970 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1971 {
1972         int i;
1973
1974         for (i = 0; i < bus->dev_count; i++) {
1975                 struct kvm_io_device *pos = bus->devs[i];
1976
1977                 kvm_iodevice_destructor(pos);
1978         }
1979         kfree(bus);
1980 }
1981
1982 /* kvm_io_bus_write - called under kvm->slots_lock */
1983 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
1984                      int len, const void *val)
1985 {
1986         int i;
1987         struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
1988         for (i = 0; i < bus->dev_count; i++)
1989                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
1990                         return 0;
1991         return -EOPNOTSUPP;
1992 }
1993
1994 /* kvm_io_bus_read - called under kvm->slots_lock */
1995 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
1996                     int len, void *val)
1997 {
1998         int i;
1999         struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
2000
2001         for (i = 0; i < bus->dev_count; i++)
2002                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2003                         return 0;
2004         return -EOPNOTSUPP;
2005 }
2006
2007 /* Caller must hold slots_lock. */
2008 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2009                             struct kvm_io_device *dev)
2010 {
2011         struct kvm_io_bus *new_bus, *bus;
2012
2013         bus = kvm->buses[bus_idx];
2014         if (bus->dev_count > NR_IOBUS_DEVS-1)
2015                 return -ENOSPC;
2016
2017         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2018         if (!new_bus)
2019                 return -ENOMEM;
2020         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2021         new_bus->devs[new_bus->dev_count++] = dev;
2022         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2023         synchronize_srcu_expedited(&kvm->srcu);
2024         kfree(bus);
2025
2026         return 0;
2027 }
2028
2029 /* Caller must hold slots_lock. */
2030 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2031                               struct kvm_io_device *dev)
2032 {
2033         int i, r;
2034         struct kvm_io_bus *new_bus, *bus;
2035
2036         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2037         if (!new_bus)
2038                 return -ENOMEM;
2039
2040         bus = kvm->buses[bus_idx];
2041         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2042
2043         r = -ENOENT;
2044         for (i = 0; i < new_bus->dev_count; i++)
2045                 if (new_bus->devs[i] == dev) {
2046                         r = 0;
2047                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2048                         break;
2049                 }
2050
2051         if (r) {
2052                 kfree(new_bus);
2053                 return r;
2054         }
2055
2056         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2057         synchronize_srcu_expedited(&kvm->srcu);
2058         kfree(bus);
2059         return r;
2060 }
2061
2062 static struct notifier_block kvm_cpu_notifier = {
2063         .notifier_call = kvm_cpu_hotplug,
2064         .priority = 20, /* must be > scheduler priority */
2065 };
2066
2067 static int vm_stat_get(void *_offset, u64 *val)
2068 {
2069         unsigned offset = (long)_offset;
2070         struct kvm *kvm;
2071
2072         *val = 0;
2073         spin_lock(&kvm_lock);
2074         list_for_each_entry(kvm, &vm_list, vm_list)
2075                 *val += *(u32 *)((void *)kvm + offset);
2076         spin_unlock(&kvm_lock);
2077         return 0;
2078 }
2079
2080 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2081
2082 static int vcpu_stat_get(void *_offset, u64 *val)
2083 {
2084         unsigned offset = (long)_offset;
2085         struct kvm *kvm;
2086         struct kvm_vcpu *vcpu;
2087         int i;
2088
2089         *val = 0;
2090         spin_lock(&kvm_lock);
2091         list_for_each_entry(kvm, &vm_list, vm_list)
2092                 kvm_for_each_vcpu(i, vcpu, kvm)
2093                         *val += *(u32 *)((void *)vcpu + offset);
2094
2095         spin_unlock(&kvm_lock);
2096         return 0;
2097 }
2098
2099 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2100
2101 static const struct file_operations *stat_fops[] = {
2102         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2103         [KVM_STAT_VM]   = &vm_stat_fops,
2104 };
2105
2106 static void kvm_init_debug(void)
2107 {
2108         struct kvm_stats_debugfs_item *p;
2109
2110         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2111         for (p = debugfs_entries; p->name; ++p)
2112                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2113                                                 (void *)(long)p->offset,
2114                                                 stat_fops[p->kind]);
2115 }
2116
2117 static void kvm_exit_debug(void)
2118 {
2119         struct kvm_stats_debugfs_item *p;
2120
2121         for (p = debugfs_entries; p->name; ++p)
2122                 debugfs_remove(p->dentry);
2123         debugfs_remove(kvm_debugfs_dir);
2124 }
2125
2126 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2127 {
2128         if (kvm_usage_count)
2129                 hardware_disable(NULL);
2130         return 0;
2131 }
2132
2133 static int kvm_resume(struct sys_device *dev)
2134 {
2135         if (kvm_usage_count)
2136                 hardware_enable(NULL);
2137         return 0;
2138 }
2139
2140 static struct sysdev_class kvm_sysdev_class = {
2141         .name = "kvm",
2142         .suspend = kvm_suspend,
2143         .resume = kvm_resume,
2144 };
2145
2146 static struct sys_device kvm_sysdev = {
2147         .id = 0,
2148         .cls = &kvm_sysdev_class,
2149 };
2150
2151 struct page *bad_page;
2152 pfn_t bad_pfn;
2153
2154 static inline
2155 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2156 {
2157         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2158 }
2159
2160 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2161 {
2162         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2163
2164         kvm_arch_vcpu_load(vcpu, cpu);
2165 }
2166
2167 static void kvm_sched_out(struct preempt_notifier *pn,
2168                           struct task_struct *next)
2169 {
2170         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2171
2172         kvm_arch_vcpu_put(vcpu);
2173 }
2174
2175 int kvm_init(void *opaque, unsigned int vcpu_size,
2176                   struct module *module)
2177 {
2178         int r;
2179         int cpu;
2180
2181         r = kvm_arch_init(opaque);
2182         if (r)
2183                 goto out_fail;
2184
2185         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2186
2187         if (bad_page == NULL) {
2188                 r = -ENOMEM;
2189                 goto out;
2190         }
2191
2192         bad_pfn = page_to_pfn(bad_page);
2193
2194         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2195                 r = -ENOMEM;
2196                 goto out_free_0;
2197         }
2198
2199         r = kvm_arch_hardware_setup();
2200         if (r < 0)
2201                 goto out_free_0a;
2202
2203         for_each_online_cpu(cpu) {
2204                 smp_call_function_single(cpu,
2205                                 kvm_arch_check_processor_compat,
2206                                 &r, 1);
2207                 if (r < 0)
2208                         goto out_free_1;
2209         }
2210
2211         r = register_cpu_notifier(&kvm_cpu_notifier);
2212         if (r)
2213                 goto out_free_2;
2214         register_reboot_notifier(&kvm_reboot_notifier);
2215
2216         r = sysdev_class_register(&kvm_sysdev_class);
2217         if (r)
2218                 goto out_free_3;
2219
2220         r = sysdev_register(&kvm_sysdev);
2221         if (r)
2222                 goto out_free_4;
2223
2224         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2225         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2226                                            __alignof__(struct kvm_vcpu),
2227                                            0, NULL);
2228         if (!kvm_vcpu_cache) {
2229                 r = -ENOMEM;
2230                 goto out_free_5;
2231         }
2232
2233         kvm_chardev_ops.owner = module;
2234         kvm_vm_fops.owner = module;
2235         kvm_vcpu_fops.owner = module;
2236
2237         r = misc_register(&kvm_dev);
2238         if (r) {
2239                 printk(KERN_ERR "kvm: misc device register failed\n");
2240                 goto out_free;
2241         }
2242
2243         kvm_preempt_ops.sched_in = kvm_sched_in;
2244         kvm_preempt_ops.sched_out = kvm_sched_out;
2245
2246         kvm_init_debug();
2247
2248         return 0;
2249
2250 out_free:
2251         kmem_cache_destroy(kvm_vcpu_cache);
2252 out_free_5:
2253         sysdev_unregister(&kvm_sysdev);
2254 out_free_4:
2255         sysdev_class_unregister(&kvm_sysdev_class);
2256 out_free_3:
2257         unregister_reboot_notifier(&kvm_reboot_notifier);
2258         unregister_cpu_notifier(&kvm_cpu_notifier);
2259 out_free_2:
2260 out_free_1:
2261         kvm_arch_hardware_unsetup();
2262 out_free_0a:
2263         free_cpumask_var(cpus_hardware_enabled);
2264 out_free_0:
2265         __free_page(bad_page);
2266 out:
2267         kvm_arch_exit();
2268 out_fail:
2269         return r;
2270 }
2271 EXPORT_SYMBOL_GPL(kvm_init);
2272
2273 void kvm_exit(void)
2274 {
2275         tracepoint_synchronize_unregister();
2276         kvm_exit_debug();
2277         misc_deregister(&kvm_dev);
2278         kmem_cache_destroy(kvm_vcpu_cache);
2279         sysdev_unregister(&kvm_sysdev);
2280         sysdev_class_unregister(&kvm_sysdev_class);
2281         unregister_reboot_notifier(&kvm_reboot_notifier);
2282         unregister_cpu_notifier(&kvm_cpu_notifier);
2283         on_each_cpu(hardware_disable, NULL, 1);
2284         kvm_arch_hardware_unsetup();
2285         kvm_arch_exit();
2286         free_cpumask_var(cpus_hardware_enabled);
2287         __free_page(bad_page);
2288 }
2289 EXPORT_SYMBOL_GPL(kvm_exit);