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  * Copyright 2010 Red Hat, Inc. and/or its affilates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.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 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50
  51 #include <asm/processor.h>
  52 #include <asm/io.h>
  53 #include <asm/uaccess.h>
  54 #include <asm/pgtable.h>
  55 #include <asm-generic/bitops/le.h>
  56
  57 #include "coalesced_mmio.h"
  58
  59 #define CREATE_TRACE_POINTS
  60 #include <trace/events/kvm.h>
  61
  62 MODULE_AUTHOR("Qumranet");
  63 MODULE_LICENSE("GPL");
  64
  65 /*
  66  * Ordering of locks:
  67  *
  68  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  69  */
  70
  71 DEFINE_SPINLOCK(kvm_lock);
  72 LIST_HEAD(vm_list);
  73
  74 static cpumask_var_t cpus_hardware_enabled;
  75 static int kvm_usage_count = 0;
  76 static atomic_t hardware_enable_failed;
  77
  78 struct kmem_cache *kvm_vcpu_cache;
  79 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  80
  81 static __read_mostly struct preempt_ops kvm_preempt_ops;
  82
  83 struct dentry *kvm_debugfs_dir;
  84
  85 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  86                            unsigned long arg);
  87 static int hardware_enable_all(void);
  88 static void hardware_disable_all(void);
  89
  90 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  91
  92 static bool kvm_rebooting;
  93
  94 static bool largepages_enabled = true;
  95
  96 struct page *hwpoison_page;
  97 pfn_t hwpoison_pfn;
  98
  99 inline int kvm_is_mmio_pfn(pfn_t pfn)
 100 {
 101         if (pfn_valid(pfn)) {
 102                 struct page *page = compound_head(pfn_to_page(pfn));
 103                 return PageReserved(page);
 104         }
 105
 106         return true;
 107 }
 108
 109 /*
 110  * Switches to specified vcpu, until a matching vcpu_put()
 111  */
 112 void vcpu_load(struct kvm_vcpu *vcpu)
 113 {
 114         int cpu;
 115
 116         mutex_lock(&vcpu->mutex);
 117         cpu = get_cpu();
 118         preempt_notifier_register(&vcpu->preempt_notifier);
 119         kvm_arch_vcpu_load(vcpu, cpu);
 120         put_cpu();
 121 }
 122
 123 void vcpu_put(struct kvm_vcpu *vcpu)
 124 {
 125         preempt_disable();
 126         kvm_arch_vcpu_put(vcpu);
 127         preempt_notifier_unregister(&vcpu->preempt_notifier);
 128         preempt_enable();
 129         mutex_unlock(&vcpu->mutex);
 130 }
 131
 132 static void ack_flush(void *_completed)
 133 {
 134 }
 135
 136 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 137 {
 138         int i, cpu, me;
 139         cpumask_var_t cpus;
 140         bool called = true;
 141         struct kvm_vcpu *vcpu;
 142
 143         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 144
 145         raw_spin_lock(&kvm->requests_lock);
 146         me = smp_processor_id();
 147         kvm_for_each_vcpu(i, vcpu, kvm) {
 148                 if (test_and_set_bit(req, &vcpu->requests))
 149                         continue;
 150                 cpu = vcpu->cpu;
 151                 if (cpus != NULL && cpu != -1 && cpu != me)
 152                         cpumask_set_cpu(cpu, cpus);
 153         }
 154         if (unlikely(cpus == NULL))
 155                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 156         else if (!cpumask_empty(cpus))
 157                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 158         else
 159                 called = false;
 160         raw_spin_unlock(&kvm->requests_lock);
 161         free_cpumask_var(cpus);
 162         return called;
 163 }
 164
 165 void kvm_flush_remote_tlbs(struct kvm *kvm)
 166 {
 167         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 168                 ++kvm->stat.remote_tlb_flush;
 169 }
 170
 171 void kvm_reload_remote_mmus(struct kvm *kvm)
 172 {
 173         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 174 }
 175
 176 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 177 {
 178         struct page *page;
 179         int r;
 180
 181         mutex_init(&vcpu->mutex);
 182         vcpu->cpu = -1;
 183         vcpu->kvm = kvm;
 184         vcpu->vcpu_id = id;
 185         init_waitqueue_head(&vcpu->wq);
 186
 187         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 188         if (!page) {
 189                 r = -ENOMEM;
 190                 goto fail;
 191         }
 192         vcpu->run = page_address(page);
 193
 194         r = kvm_arch_vcpu_init(vcpu);
 195         if (r < 0)
 196                 goto fail_free_run;
 197         return 0;
 198
 199 fail_free_run:
 200         free_page((unsigned long)vcpu->run);
 201 fail:
 202         return r;
 203 }
 204 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 205
 206 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 207 {
 208         kvm_arch_vcpu_uninit(vcpu);
 209         free_page((unsigned long)vcpu->run);
 210 }
 211 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 212
 213 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 214 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 215 {
 216         return container_of(mn, struct kvm, mmu_notifier);
 217 }
 218
 219 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 220                                              struct mm_struct *mm,
 221                                              unsigned long address)
 222 {
 223         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 224         int need_tlb_flush, idx;
 225
 226         /*
 227          * When ->invalidate_page runs, the linux pte has been zapped
 228          * already but the page is still allocated until
 229          * ->invalidate_page returns. So if we increase the sequence
 230          * here the kvm page fault will notice if the spte can't be
 231          * established because the page is going to be freed. If
 232          * instead the kvm page fault establishes the spte before
 233          * ->invalidate_page runs, kvm_unmap_hva will release it
 234          * before returning.
 235          *
 236          * The sequence increase only need to be seen at spin_unlock
 237          * time, and not at spin_lock time.
 238          *
 239          * Increasing the sequence after the spin_unlock would be
 240          * unsafe because the kvm page fault could then establish the
 241          * pte after kvm_unmap_hva returned, without noticing the page
 242          * is going to be freed.
 243          */
 244         idx = srcu_read_lock(&kvm->srcu);
 245         spin_lock(&kvm->mmu_lock);
 246         kvm->mmu_notifier_seq++;
 247         need_tlb_flush = kvm_unmap_hva(kvm, address);
 248         spin_unlock(&kvm->mmu_lock);
 249         srcu_read_unlock(&kvm->srcu, idx);
 250
 251         /* we've to flush the tlb before the pages can be freed */
 252         if (need_tlb_flush)
 253                 kvm_flush_remote_tlbs(kvm);
 254
 255 }
 256
 257 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 258                                         struct mm_struct *mm,
 259                                         unsigned long address,
 260                                         pte_t pte)
 261 {
 262         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 263         int idx;
 264
 265         idx = srcu_read_lock(&kvm->srcu);
 266         spin_lock(&kvm->mmu_lock);
 267         kvm->mmu_notifier_seq++;
 268         kvm_set_spte_hva(kvm, address, pte);
 269         spin_unlock(&kvm->mmu_lock);
 270         srcu_read_unlock(&kvm->srcu, idx);
 271 }
 272
 273 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 274                                                     struct mm_struct *mm,
 275                                                     unsigned long start,
 276                                                     unsigned long end)
 277 {
 278         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 279         int need_tlb_flush = 0, idx;
 280
 281         idx = srcu_read_lock(&kvm->srcu);
 282         spin_lock(&kvm->mmu_lock);
 283         /*
 284          * The count increase must become visible at unlock time as no
 285          * spte can be established without taking the mmu_lock and
 286          * count is also read inside the mmu_lock critical section.
 287          */
 288         kvm->mmu_notifier_count++;
 289         for (; start < end; start += PAGE_SIZE)
 290                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 291         spin_unlock(&kvm->mmu_lock);
 292         srcu_read_unlock(&kvm->srcu, idx);
 293
 294         /* we've to flush the tlb before the pages can be freed */
 295         if (need_tlb_flush)
 296                 kvm_flush_remote_tlbs(kvm);
 297 }
 298
 299 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 300                                                   struct mm_struct *mm,
 301                                                   unsigned long start,
 302                                                   unsigned long end)
 303 {
 304         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 305
 306         spin_lock(&kvm->mmu_lock);
 307         /*
 308          * This sequence increase will notify the kvm page fault that
 309          * the page that is going to be mapped in the spte could have
 310          * been freed.
 311          */
 312         kvm->mmu_notifier_seq++;
 313         /*
 314          * The above sequence increase must be visible before the
 315          * below count decrease but both values are read by the kvm
 316          * page fault under mmu_lock spinlock so we don't need to add
 317          * a smb_wmb() here in between the two.
 318          */
 319         kvm->mmu_notifier_count--;
 320         spin_unlock(&kvm->mmu_lock);
 321
 322         BUG_ON(kvm->mmu_notifier_count < 0);
 323 }
 324
 325 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 326                                               struct mm_struct *mm,
 327                                               unsigned long address)
 328 {
 329         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 330         int young, idx;
 331
 332         idx = srcu_read_lock(&kvm->srcu);
 333         spin_lock(&kvm->mmu_lock);
 334         young = kvm_age_hva(kvm, address);
 335         spin_unlock(&kvm->mmu_lock);
 336         srcu_read_unlock(&kvm->srcu, idx);
 337
 338         if (young)
 339                 kvm_flush_remote_tlbs(kvm);
 340
 341         return young;
 342 }
 343
 344 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 345                                      struct mm_struct *mm)
 346 {
 347         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 348         int idx;
 349
 350         idx = srcu_read_lock(&kvm->srcu);
 351         kvm_arch_flush_shadow(kvm);
 352         srcu_read_unlock(&kvm->srcu, idx);
 353 }
 354
 355 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 356         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 357         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 358         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 359         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 360         .change_pte             = kvm_mmu_notifier_change_pte,
 361         .release                = kvm_mmu_notifier_release,
 362 };
 363
 364 static int kvm_init_mmu_notifier(struct kvm *kvm)
 365 {
 366         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 367         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 368 }
 369
 370 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 371
 372 static int kvm_init_mmu_notifier(struct kvm *kvm)
 373 {
 374         return 0;
 375 }
 376
 377 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 378
 379 static struct kvm *kvm_create_vm(void)
 380 {
 381         int r = 0, i;
 382         struct kvm *kvm = kvm_arch_create_vm();
 383
 384         if (IS_ERR(kvm))
 385                 goto out;
 386
 387         r = hardware_enable_all();
 388         if (r)
 389                 goto out_err_nodisable;
 390
 391 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 392         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 393         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 394 #endif
 395
 396         r = -ENOMEM;
 397         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 398         if (!kvm->memslots)
 399                 goto out_err;
 400         if (init_srcu_struct(&kvm->srcu))
 401                 goto out_err;
 402         for (i = 0; i < KVM_NR_BUSES; i++) {
 403                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 404                                         GFP_KERNEL);
 405                 if (!kvm->buses[i]) {
 406                         cleanup_srcu_struct(&kvm->srcu);
 407                         goto out_err;
 408                 }
 409         }
 410
 411         r = kvm_init_mmu_notifier(kvm);
 412         if (r) {
 413                 cleanup_srcu_struct(&kvm->srcu);
 414                 goto out_err;
 415         }
 416
 417         kvm->mm = current->mm;
 418         atomic_inc(&kvm->mm->mm_count);
 419         spin_lock_init(&kvm->mmu_lock);
 420         raw_spin_lock_init(&kvm->requests_lock);
 421         kvm_eventfd_init(kvm);
 422         mutex_init(&kvm->lock);
 423         mutex_init(&kvm->irq_lock);
 424         mutex_init(&kvm->slots_lock);
 425         atomic_set(&kvm->users_count, 1);
 426         spin_lock(&kvm_lock);
 427         list_add(&kvm->vm_list, &vm_list);
 428         spin_unlock(&kvm_lock);
 429 out:
 430         return kvm;
 431
 432 out_err:
 433         hardware_disable_all();
 434 out_err_nodisable:
 435         for (i = 0; i < KVM_NR_BUSES; i++)
 436                 kfree(kvm->buses[i]);
 437         kfree(kvm->memslots);
 438         kfree(kvm);
 439         return ERR_PTR(r);
 440 }
 441
 442 /*
 443  * Free any memory in @free but not in @dont.
 444  */
 445 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 446                                   struct kvm_memory_slot *dont)
 447 {
 448         int i;
 449
 450         if (!dont || free->rmap != dont->rmap)
 451                 vfree(free->rmap);
 452
 453         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 454                 vfree(free->dirty_bitmap);
 455
 456
 457         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 458                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 459                         vfree(free->lpage_info[i]);
 460                         free->lpage_info[i] = NULL;
 461                 }
 462         }
 463
 464         free->npages = 0;
 465         free->dirty_bitmap = NULL;
 466         free->rmap = NULL;
 467 }
 468
 469 void kvm_free_physmem(struct kvm *kvm)
 470 {
 471         int i;
 472         struct kvm_memslots *slots = kvm->memslots;
 473
 474         for (i = 0; i < slots->nmemslots; ++i)
 475                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 476
 477         kfree(kvm->memslots);
 478 }
 479
 480 static void kvm_destroy_vm(struct kvm *kvm)
 481 {
 482         int i;
 483         struct mm_struct *mm = kvm->mm;
 484
 485         kvm_arch_sync_events(kvm);
 486         spin_lock(&kvm_lock);
 487         list_del(&kvm->vm_list);
 488         spin_unlock(&kvm_lock);
 489         kvm_free_irq_routing(kvm);
 490         for (i = 0; i < KVM_NR_BUSES; i++)
 491                 kvm_io_bus_destroy(kvm->buses[i]);
 492         kvm_coalesced_mmio_free(kvm);
 493 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 494         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 495 #else
 496         kvm_arch_flush_shadow(kvm);
 497 #endif
 498         kvm_arch_destroy_vm(kvm);
 499         hardware_disable_all();
 500         mmdrop(mm);
 501 }
 502
 503 void kvm_get_kvm(struct kvm *kvm)
 504 {
 505         atomic_inc(&kvm->users_count);
 506 }
 507 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 508
 509 void kvm_put_kvm(struct kvm *kvm)
 510 {
 511         if (atomic_dec_and_test(&kvm->users_count))
 512                 kvm_destroy_vm(kvm);
 513 }
 514 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 515
 516
 517 static int kvm_vm_release(struct inode *inode, struct file *filp)
 518 {
 519         struct kvm *kvm = filp->private_data;
 520
 521         kvm_irqfd_release(kvm);
 522
 523         kvm_put_kvm(kvm);
 524         return 0;
 525 }
 526
 527 /*
 528  * Allocate some memory and give it an address in the guest physical address
 529  * space.
 530  *
 531  * Discontiguous memory is allowed, mostly for framebuffers.
 532  *
 533  * Must be called holding mmap_sem for write.
 534  */
 535 int __kvm_set_memory_region(struct kvm *kvm,
 536                             struct kvm_userspace_memory_region *mem,
 537                             int user_alloc)
 538 {
 539         int r, flush_shadow = 0;
 540         gfn_t base_gfn;
 541         unsigned long npages;
 542         unsigned long i;
 543         struct kvm_memory_slot *memslot;
 544         struct kvm_memory_slot old, new;
 545         struct kvm_memslots *slots, *old_memslots;
 546
 547         r = -EINVAL;
 548         /* General sanity checks */
 549         if (mem->memory_size & (PAGE_SIZE - 1))
 550                 goto out;
 551         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 552                 goto out;
 553         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 554                 goto out;
 555         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 556                 goto out;
 557         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 558                 goto out;
 559
 560         memslot = &kvm->memslots->memslots[mem->slot];
 561         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 562         npages = mem->memory_size >> PAGE_SHIFT;
 563
 564         r = -EINVAL;
 565         if (npages > KVM_MEM_MAX_NR_PAGES)
 566                 goto out;
 567
 568         if (!npages)
 569                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 570
 571         new = old = *memslot;
 572
 573         new.base_gfn = base_gfn;
 574         new.npages = npages;
 575         new.flags = mem->flags;
 576
 577         /* Disallow changing a memory slot's size. */
 578         r = -EINVAL;
 579         if (npages && old.npages && npages != old.npages)
 580                 goto out_free;
 581
 582         /* Check for overlaps */
 583         r = -EEXIST;
 584         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 585                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 586
 587                 if (s == memslot || !s->npages)
 588                         continue;
 589                 if (!((base_gfn + npages <= s->base_gfn) ||
 590                       (base_gfn >= s->base_gfn + s->npages)))
 591                         goto out_free;
 592         }
 593
 594         /* Free page dirty bitmap if unneeded */
 595         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 596                 new.dirty_bitmap = NULL;
 597
 598         r = -ENOMEM;
 599
 600         /* Allocate if a slot is being created */
 601 #ifndef CONFIG_S390
 602         if (npages && !new.rmap) {
 603                 new.rmap = vmalloc(npages * sizeof(struct page *));
 604
 605                 if (!new.rmap)
 606                         goto out_free;
 607
 608                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 609
 610                 new.user_alloc = user_alloc;
 611                 new.userspace_addr = mem->userspace_addr;
 612         }
 613         if (!npages)
 614                 goto skip_lpage;
 615
 616         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 617                 unsigned long ugfn;
 618                 unsigned long j;
 619                 int lpages;
 620                 int level = i + 2;
 621
 622                 /* Avoid unused variable warning if no large pages */
 623                 (void)level;
 624
 625                 if (new.lpage_info[i])
 626                         continue;
 627
 628                 lpages = 1 + (base_gfn + npages - 1) /
 629                              KVM_PAGES_PER_HPAGE(level);
 630                 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
 631
 632                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
 633
 634                 if (!new.lpage_info[i])
 635                         goto out_free;
 636
 637                 memset(new.lpage_info[i], 0,
 638                        lpages * sizeof(*new.lpage_info[i]));
 639
 640                 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
 641                         new.lpage_info[i][0].write_count = 1;
 642                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
 643                         new.lpage_info[i][lpages - 1].write_count = 1;
 644                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 645                 /*
 646                  * If the gfn and userspace address are not aligned wrt each
 647                  * other, or if explicitly asked to, disable large page
 648                  * support for this slot
 649                  */
 650                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 651                     !largepages_enabled)
 652                         for (j = 0; j < lpages; ++j)
 653                                 new.lpage_info[i][j].write_count = 1;
 654         }
 655
 656 skip_lpage:
 657
 658         /* Allocate page dirty bitmap if needed */
 659         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 660                 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(&new);
 661
 662                 new.dirty_bitmap = vmalloc(dirty_bytes);
 663                 if (!new.dirty_bitmap)
 664                         goto out_free;
 665                 memset(new.dirty_bitmap, 0, dirty_bytes);
 666                 /* destroy any largepage mappings for dirty tracking */
 667                 if (old.npages)
 668                         flush_shadow = 1;
 669         }
 670 #else  /* not defined CONFIG_S390 */
 671         new.user_alloc = user_alloc;
 672         if (user_alloc)
 673                 new.userspace_addr = mem->userspace_addr;
 674 #endif /* not defined CONFIG_S390 */
 675
 676         if (!npages) {
 677                 r = -ENOMEM;
 678                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 679                 if (!slots)
 680                         goto out_free;
 681                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 682                 if (mem->slot >= slots->nmemslots)
 683                         slots->nmemslots = mem->slot + 1;
 684                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 685
 686                 old_memslots = kvm->memslots;
 687                 rcu_assign_pointer(kvm->memslots, slots);
 688                 synchronize_srcu_expedited(&kvm->srcu);
 689                 /* From this point no new shadow pages pointing to a deleted
 690                  * memslot will be created.
 691                  *
 692                  * validation of sp->gfn happens in:
 693                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 694                  *      - kvm_is_visible_gfn (mmu_check_roots)
 695                  */
 696                 kvm_arch_flush_shadow(kvm);
 697                 kfree(old_memslots);
 698         }
 699
 700         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 701         if (r)
 702                 goto out_free;
 703
 704 #ifdef CONFIG_DMAR
 705         /* map the pages in iommu page table */
 706         if (npages) {
 707                 r = kvm_iommu_map_pages(kvm, &new);
 708                 if (r)
 709                         goto out_free;
 710         }
 711 #endif
 712
 713         r = -ENOMEM;
 714         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 715         if (!slots)
 716                 goto out_free;
 717         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 718         if (mem->slot >= slots->nmemslots)
 719                 slots->nmemslots = mem->slot + 1;
 720
 721         /* actual memory is freed via old in kvm_free_physmem_slot below */
 722         if (!npages) {
 723                 new.rmap = NULL;
 724                 new.dirty_bitmap = NULL;
 725                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 726                         new.lpage_info[i] = NULL;
 727         }
 728
 729         slots->memslots[mem->slot] = new;
 730         old_memslots = kvm->memslots;
 731         rcu_assign_pointer(kvm->memslots, slots);
 732         synchronize_srcu_expedited(&kvm->srcu);
 733
 734         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 735
 736         kvm_free_physmem_slot(&old, &new);
 737         kfree(old_memslots);
 738
 739         if (flush_shadow)
 740                 kvm_arch_flush_shadow(kvm);
 741
 742         return 0;
 743
 744 out_free:
 745         kvm_free_physmem_slot(&new, &old);
 746 out:
 747         return r;
 748
 749 }
 750 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 751
 752 int kvm_set_memory_region(struct kvm *kvm,
 753                           struct kvm_userspace_memory_region *mem,
 754                           int user_alloc)
 755 {
 756         int r;
 757
 758         mutex_lock(&kvm->slots_lock);
 759         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 760         mutex_unlock(&kvm->slots_lock);
 761         return r;
 762 }
 763 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 764
 765 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 766                                    struct
 767                                    kvm_userspace_memory_region *mem,
 768                                    int user_alloc)
 769 {
 770         if (mem->slot >= KVM_MEMORY_SLOTS)
 771                 return -EINVAL;
 772         return kvm_set_memory_region(kvm, mem, user_alloc);
 773 }
 774
 775 int kvm_get_dirty_log(struct kvm *kvm,
 776                         struct kvm_dirty_log *log, int *is_dirty)
 777 {
 778         struct kvm_memory_slot *memslot;
 779         int r, i;
 780         unsigned long n;
 781         unsigned long any = 0;
 782
 783         r = -EINVAL;
 784         if (log->slot >= KVM_MEMORY_SLOTS)
 785                 goto out;
 786
 787         memslot = &kvm->memslots->memslots[log->slot];
 788         r = -ENOENT;
 789         if (!memslot->dirty_bitmap)
 790                 goto out;
 791
 792         n = kvm_dirty_bitmap_bytes(memslot);
 793
 794         for (i = 0; !any && i < n/sizeof(long); ++i)
 795                 any = memslot->dirty_bitmap[i];
 796
 797         r = -EFAULT;
 798         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 799                 goto out;
 800
 801         if (any)
 802                 *is_dirty = 1;
 803
 804         r = 0;
 805 out:
 806         return r;
 807 }
 808
 809 void kvm_disable_largepages(void)
 810 {
 811         largepages_enabled = false;
 812 }
 813 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 814
 815 int is_error_page(struct page *page)
 816 {
 817         return page == bad_page || page == hwpoison_page;
 818 }
 819 EXPORT_SYMBOL_GPL(is_error_page);
 820
 821 int is_error_pfn(pfn_t pfn)
 822 {
 823         return pfn == bad_pfn || pfn == hwpoison_pfn;
 824 }
 825 EXPORT_SYMBOL_GPL(is_error_pfn);
 826
 827 int is_hwpoison_pfn(pfn_t pfn)
 828 {
 829         return pfn == hwpoison_pfn;
 830 }
 831 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 832
 833 static inline unsigned long bad_hva(void)
 834 {
 835         return PAGE_OFFSET;
 836 }
 837
 838 int kvm_is_error_hva(unsigned long addr)
 839 {
 840         return addr == bad_hva();
 841 }
 842 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 843
 844 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
 845 {
 846         int i;
 847         struct kvm_memslots *slots = kvm_memslots(kvm);
 848
 849         for (i = 0; i < slots->nmemslots; ++i) {
 850                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 851
 852                 if (gfn >= memslot->base_gfn
 853                     && gfn < memslot->base_gfn + memslot->npages)
 854                         return memslot;
 855         }
 856         return NULL;
 857 }
 858 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
 859
 860 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 861 {
 862         gfn = unalias_gfn(kvm, gfn);
 863         return gfn_to_memslot_unaliased(kvm, gfn);
 864 }
 865
 866 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 867 {
 868         int i;
 869         struct kvm_memslots *slots = kvm_memslots(kvm);
 870
 871         gfn = unalias_gfn_instantiation(kvm, gfn);
 872         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 873                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 874
 875                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 876                         continue;
 877
 878                 if (gfn >= memslot->base_gfn
 879                     && gfn < memslot->base_gfn + memslot->npages)
 880                         return 1;
 881         }
 882         return 0;
 883 }
 884 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 885
 886 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 887 {
 888         struct vm_area_struct *vma;
 889         unsigned long addr, size;
 890
 891         size = PAGE_SIZE;
 892
 893         addr = gfn_to_hva(kvm, gfn);
 894         if (kvm_is_error_hva(addr))
 895                 return PAGE_SIZE;
 896
 897         down_read(&current->mm->mmap_sem);
 898         vma = find_vma(current->mm, addr);
 899         if (!vma)
 900                 goto out;
 901
 902         size = vma_kernel_pagesize(vma);
 903
 904 out:
 905         up_read(&current->mm->mmap_sem);
 906
 907         return size;
 908 }
 909
 910 int memslot_id(struct kvm *kvm, gfn_t gfn)
 911 {
 912         int i;
 913         struct kvm_memslots *slots = kvm_memslots(kvm);
 914         struct kvm_memory_slot *memslot = NULL;
 915
 916         gfn = unalias_gfn(kvm, gfn);
 917         for (i = 0; i < slots->nmemslots; ++i) {
 918                 memslot = &slots->memslots[i];
 919
 920                 if (gfn >= memslot->base_gfn
 921                     && gfn < memslot->base_gfn + memslot->npages)
 922                         break;
 923         }
 924
 925         return memslot - slots->memslots;
 926 }
 927
 928 static unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
 929 {
 930         return slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE;
 931 }
 932
 933 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 934 {
 935         struct kvm_memory_slot *slot;
 936
 937         gfn = unalias_gfn_instantiation(kvm, gfn);
 938         slot = gfn_to_memslot_unaliased(kvm, gfn);
 939         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 940                 return bad_hva();
 941         return gfn_to_hva_memslot(slot, gfn);
 942 }
 943 EXPORT_SYMBOL_GPL(gfn_to_hva);
 944
 945 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr)
 946 {
 947         struct page *page[1];
 948         int npages;
 949         pfn_t pfn;
 950
 951         might_sleep();
 952
 953         npages = get_user_pages_fast(addr, 1, 1, page);
 954
 955         if (unlikely(npages != 1)) {
 956                 struct vm_area_struct *vma;
 957
 958                 if (is_hwpoison_address(addr)) {
 959                         get_page(hwpoison_page);
 960                         return page_to_pfn(hwpoison_page);
 961                 }
 962
 963                 down_read(&current->mm->mmap_sem);
 964                 vma = find_vma(current->mm, addr);
 965
 966                 if (vma == NULL || addr < vma->vm_start ||
 967                     !(vma->vm_flags & VM_PFNMAP)) {
 968                         up_read(&current->mm->mmap_sem);
 969                         get_page(bad_page);
 970                         return page_to_pfn(bad_page);
 971                 }
 972
 973                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 974                 up_read(&current->mm->mmap_sem);
 975                 BUG_ON(!kvm_is_mmio_pfn(pfn));
 976         } else
 977                 pfn = page_to_pfn(page[0]);
 978
 979         return pfn;
 980 }
 981
 982 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 983 {
 984         unsigned long addr;
 985
 986         addr = gfn_to_hva(kvm, gfn);
 987         if (kvm_is_error_hva(addr)) {
 988                 get_page(bad_page);
 989                 return page_to_pfn(bad_page);
 990         }
 991
 992         return hva_to_pfn(kvm, addr);
 993 }
 994 EXPORT_SYMBOL_GPL(gfn_to_pfn);
 995
 996 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
 997                          struct kvm_memory_slot *slot, gfn_t gfn)
 998 {
 999         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1000         return hva_to_pfn(kvm, addr);
1001 }
1002
1003 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1004 {
1005         pfn_t pfn;
1006
1007         pfn = gfn_to_pfn(kvm, gfn);
1008         if (!kvm_is_mmio_pfn(pfn))
1009                 return pfn_to_page(pfn);
1010
1011         WARN_ON(kvm_is_mmio_pfn(pfn));
1012
1013         get_page(bad_page);
1014         return bad_page;
1015 }
1016
1017 EXPORT_SYMBOL_GPL(gfn_to_page);
1018
1019 void kvm_release_page_clean(struct page *page)
1020 {
1021         kvm_release_pfn_clean(page_to_pfn(page));
1022 }
1023 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1024
1025 void kvm_release_pfn_clean(pfn_t pfn)
1026 {
1027         if (!kvm_is_mmio_pfn(pfn))
1028                 put_page(pfn_to_page(pfn));
1029 }
1030 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1031
1032 void kvm_release_page_dirty(struct page *page)
1033 {
1034         kvm_release_pfn_dirty(page_to_pfn(page));
1035 }
1036 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1037
1038 void kvm_release_pfn_dirty(pfn_t pfn)
1039 {
1040         kvm_set_pfn_dirty(pfn);
1041         kvm_release_pfn_clean(pfn);
1042 }
1043 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1044
1045 void kvm_set_page_dirty(struct page *page)
1046 {
1047         kvm_set_pfn_dirty(page_to_pfn(page));
1048 }
1049 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1050
1051 void kvm_set_pfn_dirty(pfn_t pfn)
1052 {
1053         if (!kvm_is_mmio_pfn(pfn)) {
1054                 struct page *page = pfn_to_page(pfn);
1055                 if (!PageReserved(page))
1056                         SetPageDirty(page);
1057         }
1058 }
1059 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1060
1061 void kvm_set_pfn_accessed(pfn_t pfn)
1062 {
1063         if (!kvm_is_mmio_pfn(pfn))
1064                 mark_page_accessed(pfn_to_page(pfn));
1065 }
1066 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1067
1068 void kvm_get_pfn(pfn_t pfn)
1069 {
1070         if (!kvm_is_mmio_pfn(pfn))
1071                 get_page(pfn_to_page(pfn));
1072 }
1073 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1074
1075 static int next_segment(unsigned long len, int offset)
1076 {
1077         if (len > PAGE_SIZE - offset)
1078                 return PAGE_SIZE - offset;
1079         else
1080                 return len;
1081 }
1082
1083 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1084                         int len)
1085 {
1086         int r;
1087         unsigned long addr;
1088
1089         addr = gfn_to_hva(kvm, gfn);
1090         if (kvm_is_error_hva(addr))
1091                 return -EFAULT;
1092         r = copy_from_user(data, (void __user *)addr + offset, len);
1093         if (r)
1094                 return -EFAULT;
1095         return 0;
1096 }
1097 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1098
1099 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1100 {
1101         gfn_t gfn = gpa >> PAGE_SHIFT;
1102         int seg;
1103         int offset = offset_in_page(gpa);
1104         int ret;
1105
1106         while ((seg = next_segment(len, offset)) != 0) {
1107                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1108                 if (ret < 0)
1109                         return ret;
1110                 offset = 0;
1111                 len -= seg;
1112                 data += seg;
1113                 ++gfn;
1114         }
1115         return 0;
1116 }
1117 EXPORT_SYMBOL_GPL(kvm_read_guest);
1118
1119 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1120                           unsigned long len)
1121 {
1122         int r;
1123         unsigned long addr;
1124         gfn_t gfn = gpa >> PAGE_SHIFT;
1125         int offset = offset_in_page(gpa);
1126
1127         addr = gfn_to_hva(kvm, gfn);
1128         if (kvm_is_error_hva(addr))
1129                 return -EFAULT;
1130         pagefault_disable();
1131         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1132         pagefault_enable();
1133         if (r)
1134                 return -EFAULT;
1135         return 0;
1136 }
1137 EXPORT_SYMBOL(kvm_read_guest_atomic);
1138
1139 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1140                          int offset, int len)
1141 {
1142         int r;
1143         unsigned long addr;
1144
1145         addr = gfn_to_hva(kvm, gfn);
1146         if (kvm_is_error_hva(addr))
1147                 return -EFAULT;
1148         r = copy_to_user((void __user *)addr + offset, data, len);
1149         if (r)
1150                 return -EFAULT;
1151         mark_page_dirty(kvm, gfn);
1152         return 0;
1153 }
1154 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1155
1156 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1157                     unsigned long len)
1158 {
1159         gfn_t gfn = gpa >> PAGE_SHIFT;
1160         int seg;
1161         int offset = offset_in_page(gpa);
1162         int ret;
1163
1164         while ((seg = next_segment(len, offset)) != 0) {
1165                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1166                 if (ret < 0)
1167                         return ret;
1168                 offset = 0;
1169                 len -= seg;
1170                 data += seg;
1171                 ++gfn;
1172         }
1173         return 0;
1174 }
1175
1176 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1177 {
1178         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1179 }
1180 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1181
1182 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1183 {
1184         gfn_t gfn = gpa >> PAGE_SHIFT;
1185         int seg;
1186         int offset = offset_in_page(gpa);
1187         int ret;
1188
1189         while ((seg = next_segment(len, offset)) != 0) {
1190                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1191                 if (ret < 0)
1192                         return ret;
1193                 offset = 0;
1194                 len -= seg;
1195                 ++gfn;
1196         }
1197         return 0;
1198 }
1199 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1200
1201 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1202 {
1203         struct kvm_memory_slot *memslot;
1204
1205         gfn = unalias_gfn(kvm, gfn);
1206         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1207         if (memslot && memslot->dirty_bitmap) {
1208                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1209
1210                 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1211         }
1212 }
1213
1214 /*
1215  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1216  */
1217 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1218 {
1219         DEFINE_WAIT(wait);
1220
1221         for (;;) {
1222                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1223
1224                 if (kvm_arch_vcpu_runnable(vcpu)) {
1225                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1226                         break;
1227                 }
1228                 if (kvm_cpu_has_pending_timer(vcpu))
1229                         break;
1230                 if (signal_pending(current))
1231                         break;
1232
1233                 schedule();
1234         }
1235
1236         finish_wait(&vcpu->wq, &wait);
1237 }
1238
1239 void kvm_resched(struct kvm_vcpu *vcpu)
1240 {
1241         if (!need_resched())
1242                 return;
1243         cond_resched();
1244 }
1245 EXPORT_SYMBOL_GPL(kvm_resched);
1246
1247 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1248 {
1249         ktime_t expires;
1250         DEFINE_WAIT(wait);
1251
1252         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1253
1254         /* Sleep for 100 us, and hope lock-holder got scheduled */
1255         expires = ktime_add_ns(ktime_get(), 100000UL);
1256         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1257
1258         finish_wait(&vcpu->wq, &wait);
1259 }
1260 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1261
1262 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1263 {
1264         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1265         struct page *page;
1266
1267         if (vmf->pgoff == 0)
1268                 page = virt_to_page(vcpu->run);
1269 #ifdef CONFIG_X86
1270         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1271                 page = virt_to_page(vcpu->arch.pio_data);
1272 #endif
1273 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1274         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1275                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1276 #endif
1277         else
1278                 return VM_FAULT_SIGBUS;
1279         get_page(page);
1280         vmf->page = page;
1281         return 0;
1282 }
1283
1284 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1285         .fault = kvm_vcpu_fault,
1286 };
1287
1288 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1289 {
1290         vma->vm_ops = &kvm_vcpu_vm_ops;
1291         return 0;
1292 }
1293
1294 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1295 {
1296         struct kvm_vcpu *vcpu = filp->private_data;
1297
1298         kvm_put_kvm(vcpu->kvm);
1299         return 0;
1300 }
1301
1302 static struct file_operations kvm_vcpu_fops = {
1303         .release        = kvm_vcpu_release,
1304         .unlocked_ioctl = kvm_vcpu_ioctl,
1305         .compat_ioctl   = kvm_vcpu_ioctl,
1306         .mmap           = kvm_vcpu_mmap,
1307 };
1308
1309 /*
1310  * Allocates an inode for the vcpu.
1311  */
1312 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1313 {
1314         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1315 }
1316
1317 /*
1318  * Creates some virtual cpus.  Good luck creating more than one.
1319  */
1320 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1321 {
1322         int r;
1323         struct kvm_vcpu *vcpu, *v;
1324
1325         vcpu = kvm_arch_vcpu_create(kvm, id);
1326         if (IS_ERR(vcpu))
1327                 return PTR_ERR(vcpu);
1328
1329         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1330
1331         r = kvm_arch_vcpu_setup(vcpu);
1332         if (r)
1333                 return r;
1334
1335         mutex_lock(&kvm->lock);
1336         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1337                 r = -EINVAL;
1338                 goto vcpu_destroy;
1339         }
1340
1341         kvm_for_each_vcpu(r, v, kvm)
1342                 if (v->vcpu_id == id) {
1343                         r = -EEXIST;
1344                         goto vcpu_destroy;
1345                 }
1346
1347         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1348
1349         /* Now it's all set up, let userspace reach it */
1350         kvm_get_kvm(kvm);
1351         r = create_vcpu_fd(vcpu);
1352         if (r < 0) {
1353                 kvm_put_kvm(kvm);
1354                 goto vcpu_destroy;
1355         }
1356
1357         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1358         smp_wmb();
1359         atomic_inc(&kvm->online_vcpus);
1360
1361 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1362         if (kvm->bsp_vcpu_id == id)
1363                 kvm->bsp_vcpu = vcpu;
1364 #endif
1365         mutex_unlock(&kvm->lock);
1366         return r;
1367
1368 vcpu_destroy:
1369         mutex_unlock(&kvm->lock);
1370         kvm_arch_vcpu_destroy(vcpu);
1371         return r;
1372 }
1373
1374 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1375 {
1376         if (sigset) {
1377                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1378                 vcpu->sigset_active = 1;
1379                 vcpu->sigset = *sigset;
1380         } else
1381                 vcpu->sigset_active = 0;
1382         return 0;
1383 }
1384
1385 static long kvm_vcpu_ioctl(struct file *filp,
1386                            unsigned int ioctl, unsigned long arg)
1387 {
1388         struct kvm_vcpu *vcpu = filp->private_data;
1389         void __user *argp = (void __user *)arg;
1390         int r;
1391         struct kvm_fpu *fpu = NULL;
1392         struct kvm_sregs *kvm_sregs = NULL;
1393
1394         if (vcpu->kvm->mm != current->mm)
1395                 return -EIO;
1396
1397 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1398         /*
1399          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1400          * so vcpu_load() would break it.
1401          */
1402         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1403                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1404 #endif
1405
1406
1407         vcpu_load(vcpu);
1408         switch (ioctl) {
1409         case KVM_RUN:
1410                 r = -EINVAL;
1411                 if (arg)
1412                         goto out;
1413                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1414                 break;
1415         case KVM_GET_REGS: {
1416                 struct kvm_regs *kvm_regs;
1417
1418                 r = -ENOMEM;
1419                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1420                 if (!kvm_regs)
1421                         goto out;
1422                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1423                 if (r)
1424                         goto out_free1;
1425                 r = -EFAULT;
1426                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1427                         goto out_free1;
1428                 r = 0;
1429 out_free1:
1430                 kfree(kvm_regs);
1431                 break;
1432         }
1433         case KVM_SET_REGS: {
1434                 struct kvm_regs *kvm_regs;
1435
1436                 r = -ENOMEM;
1437                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1438                 if (!kvm_regs)
1439                         goto out;
1440                 r = -EFAULT;
1441                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1442                         goto out_free2;
1443                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1444                 if (r)
1445                         goto out_free2;
1446                 r = 0;
1447 out_free2:
1448                 kfree(kvm_regs);
1449                 break;
1450         }
1451         case KVM_GET_SREGS: {
1452                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1453                 r = -ENOMEM;
1454                 if (!kvm_sregs)
1455                         goto out;
1456                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1457                 if (r)
1458                         goto out;
1459                 r = -EFAULT;
1460                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1461                         goto out;
1462                 r = 0;
1463                 break;
1464         }
1465         case KVM_SET_SREGS: {
1466                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1467                 r = -ENOMEM;
1468                 if (!kvm_sregs)
1469                         goto out;
1470                 r = -EFAULT;
1471                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1472                         goto out;
1473                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1474                 if (r)
1475                         goto out;
1476                 r = 0;
1477                 break;
1478         }
1479         case KVM_GET_MP_STATE: {
1480                 struct kvm_mp_state mp_state;
1481
1482                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1483                 if (r)
1484                         goto out;
1485                 r = -EFAULT;
1486                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1487                         goto out;
1488                 r = 0;
1489                 break;
1490         }
1491         case KVM_SET_MP_STATE: {
1492                 struct kvm_mp_state mp_state;
1493
1494                 r = -EFAULT;
1495                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1496                         goto out;
1497                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1498                 if (r)
1499                         goto out;
1500                 r = 0;
1501                 break;
1502         }
1503         case KVM_TRANSLATE: {
1504                 struct kvm_translation tr;
1505
1506                 r = -EFAULT;
1507                 if (copy_from_user(&tr, argp, sizeof tr))
1508                         goto out;
1509                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1510                 if (r)
1511                         goto out;
1512                 r = -EFAULT;
1513                 if (copy_to_user(argp, &tr, sizeof tr))
1514                         goto out;
1515                 r = 0;
1516                 break;
1517         }
1518         case KVM_SET_GUEST_DEBUG: {
1519                 struct kvm_guest_debug dbg;
1520
1521                 r = -EFAULT;
1522                 if (copy_from_user(&dbg, argp, sizeof dbg))
1523                         goto out;
1524                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1525                 if (r)
1526                         goto out;
1527                 r = 0;
1528                 break;
1529         }
1530         case KVM_SET_SIGNAL_MASK: {
1531                 struct kvm_signal_mask __user *sigmask_arg = argp;
1532                 struct kvm_signal_mask kvm_sigmask;
1533                 sigset_t sigset, *p;
1534
1535                 p = NULL;
1536                 if (argp) {
1537                         r = -EFAULT;
1538                         if (copy_from_user(&kvm_sigmask, argp,
1539                                            sizeof kvm_sigmask))
1540                                 goto out;
1541                         r = -EINVAL;
1542                         if (kvm_sigmask.len != sizeof sigset)
1543                                 goto out;
1544                         r = -EFAULT;
1545                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1546                                            sizeof sigset))
1547                                 goto out;
1548                         p = &sigset;
1549                 }
1550                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1551                 break;
1552         }
1553         case KVM_GET_FPU: {
1554                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1555                 r = -ENOMEM;
1556                 if (!fpu)
1557                         goto out;
1558                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1559                 if (r)
1560                         goto out;
1561                 r = -EFAULT;
1562                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1563                         goto out;
1564                 r = 0;
1565                 break;
1566         }
1567         case KVM_SET_FPU: {
1568                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1569                 r = -ENOMEM;
1570                 if (!fpu)
1571                         goto out;
1572                 r = -EFAULT;
1573                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1574                         goto out;
1575                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1576                 if (r)
1577                         goto out;
1578                 r = 0;
1579                 break;
1580         }
1581         default:
1582                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1583         }
1584 out:
1585         vcpu_put(vcpu);
1586         kfree(fpu);
1587         kfree(kvm_sregs);
1588         return r;
1589 }
1590
1591 static long kvm_vm_ioctl(struct file *filp,
1592                            unsigned int ioctl, unsigned long arg)
1593 {
1594         struct kvm *kvm = filp->private_data;
1595         void __user *argp = (void __user *)arg;
1596         int r;
1597
1598         if (kvm->mm != current->mm)
1599                 return -EIO;
1600         switch (ioctl) {
1601         case KVM_CREATE_VCPU:
1602                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1603                 if (r < 0)
1604                         goto out;
1605                 break;
1606         case KVM_SET_USER_MEMORY_REGION: {
1607                 struct kvm_userspace_memory_region kvm_userspace_mem;
1608
1609                 r = -EFAULT;
1610                 if (copy_from_user(&kvm_userspace_mem, argp,
1611                                                 sizeof kvm_userspace_mem))
1612                         goto out;
1613
1614                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1615                 if (r)
1616                         goto out;
1617                 break;
1618         }
1619         case KVM_GET_DIRTY_LOG: {
1620                 struct kvm_dirty_log log;
1621
1622                 r = -EFAULT;
1623                 if (copy_from_user(&log, argp, sizeof log))
1624                         goto out;
1625                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1626                 if (r)
1627                         goto out;
1628                 break;
1629         }
1630 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1631         case KVM_REGISTER_COALESCED_MMIO: {
1632                 struct kvm_coalesced_mmio_zone zone;
1633                 r = -EFAULT;
1634                 if (copy_from_user(&zone, argp, sizeof zone))
1635                         goto out;
1636                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1637                 if (r)
1638                         goto out;
1639                 r = 0;
1640                 break;
1641         }
1642         case KVM_UNREGISTER_COALESCED_MMIO: {
1643                 struct kvm_coalesced_mmio_zone zone;
1644                 r = -EFAULT;
1645                 if (copy_from_user(&zone, argp, sizeof zone))
1646                         goto out;
1647                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1648                 if (r)
1649                         goto out;
1650                 r = 0;
1651                 break;
1652         }
1653 #endif
1654         case KVM_IRQFD: {
1655                 struct kvm_irqfd data;
1656
1657                 r = -EFAULT;
1658                 if (copy_from_user(&data, argp, sizeof data))
1659                         goto out;
1660                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1661                 break;
1662         }
1663         case KVM_IOEVENTFD: {
1664                 struct kvm_ioeventfd data;
1665
1666                 r = -EFAULT;
1667                 if (copy_from_user(&data, argp, sizeof data))
1668                         goto out;
1669                 r = kvm_ioeventfd(kvm, &data);
1670                 break;
1671         }
1672 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1673         case KVM_SET_BOOT_CPU_ID:
1674                 r = 0;
1675                 mutex_lock(&kvm->lock);
1676                 if (atomic_read(&kvm->online_vcpus) != 0)
1677                         r = -EBUSY;
1678                 else
1679                         kvm->bsp_vcpu_id = arg;
1680                 mutex_unlock(&kvm->lock);
1681                 break;
1682 #endif
1683         default:
1684                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1685                 if (r == -ENOTTY)
1686                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1687         }
1688 out:
1689         return r;
1690 }
1691
1692 #ifdef CONFIG_COMPAT
1693 struct compat_kvm_dirty_log {
1694         __u32 slot;
1695         __u32 padding1;
1696         union {
1697                 compat_uptr_t dirty_bitmap; /* one bit per page */
1698                 __u64 padding2;
1699         };
1700 };
1701
1702 static long kvm_vm_compat_ioctl(struct file *filp,
1703                            unsigned int ioctl, unsigned long arg)
1704 {
1705         struct kvm *kvm = filp->private_data;
1706         int r;
1707
1708         if (kvm->mm != current->mm)
1709                 return -EIO;
1710         switch (ioctl) {
1711         case KVM_GET_DIRTY_LOG: {
1712                 struct compat_kvm_dirty_log compat_log;
1713                 struct kvm_dirty_log log;
1714
1715                 r = -EFAULT;
1716                 if (copy_from_user(&compat_log, (void __user *)arg,
1717                                    sizeof(compat_log)))
1718                         goto out;
1719                 log.slot         = compat_log.slot;
1720                 log.padding1     = compat_log.padding1;
1721                 log.padding2     = compat_log.padding2;
1722                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1723
1724                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1725                 if (r)
1726                         goto out;
1727                 break;
1728         }
1729         default:
1730                 r = kvm_vm_ioctl(filp, ioctl, arg);
1731         }
1732
1733 out:
1734         return r;
1735 }
1736 #endif
1737
1738 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1739 {
1740         struct page *page[1];
1741         unsigned long addr;
1742         int npages;
1743         gfn_t gfn = vmf->pgoff;
1744         struct kvm *kvm = vma->vm_file->private_data;
1745
1746         addr = gfn_to_hva(kvm, gfn);
1747         if (kvm_is_error_hva(addr))
1748                 return VM_FAULT_SIGBUS;
1749
1750         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1751                                 NULL);
1752         if (unlikely(npages != 1))
1753                 return VM_FAULT_SIGBUS;
1754
1755         vmf->page = page[0];
1756         return 0;
1757 }
1758
1759 static const struct vm_operations_struct kvm_vm_vm_ops = {
1760         .fault = kvm_vm_fault,
1761 };
1762
1763 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1764 {
1765         vma->vm_ops = &kvm_vm_vm_ops;
1766         return 0;
1767 }
1768
1769 static struct file_operations kvm_vm_fops = {
1770         .release        = kvm_vm_release,
1771         .unlocked_ioctl = kvm_vm_ioctl,
1772 #ifdef CONFIG_COMPAT
1773         .compat_ioctl   = kvm_vm_compat_ioctl,
1774 #endif
1775         .mmap           = kvm_vm_mmap,
1776 };
1777
1778 static int kvm_dev_ioctl_create_vm(void)
1779 {
1780         int fd, r;
1781         struct kvm *kvm;
1782
1783         kvm = kvm_create_vm();
1784         if (IS_ERR(kvm))
1785                 return PTR_ERR(kvm);
1786 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1787         r = kvm_coalesced_mmio_init(kvm);
1788         if (r < 0) {
1789                 kvm_put_kvm(kvm);
1790                 return r;
1791         }
1792 #endif
1793         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1794         if (fd < 0)
1795                 kvm_put_kvm(kvm);
1796
1797         return fd;
1798 }
1799
1800 static long kvm_dev_ioctl_check_extension_generic(long arg)
1801 {
1802         switch (arg) {
1803         case KVM_CAP_USER_MEMORY:
1804         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1805         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1806 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1807         case KVM_CAP_SET_BOOT_CPU_ID:
1808 #endif
1809         case KVM_CAP_INTERNAL_ERROR_DATA:
1810                 return 1;
1811 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1812         case KVM_CAP_IRQ_ROUTING:
1813                 return KVM_MAX_IRQ_ROUTES;
1814 #endif
1815         default:
1816                 break;
1817         }
1818         return kvm_dev_ioctl_check_extension(arg);
1819 }
1820
1821 static long kvm_dev_ioctl(struct file *filp,
1822                           unsigned int ioctl, unsigned long arg)
1823 {
1824         long r = -EINVAL;
1825
1826         switch (ioctl) {
1827         case KVM_GET_API_VERSION:
1828                 r = -EINVAL;
1829                 if (arg)
1830                         goto out;
1831                 r = KVM_API_VERSION;
1832                 break;
1833         case KVM_CREATE_VM:
1834                 r = -EINVAL;
1835                 if (arg)
1836                         goto out;
1837                 r = kvm_dev_ioctl_create_vm();
1838                 break;
1839         case KVM_CHECK_EXTENSION:
1840                 r = kvm_dev_ioctl_check_extension_generic(arg);
1841                 break;
1842         case KVM_GET_VCPU_MMAP_SIZE:
1843                 r = -EINVAL;
1844                 if (arg)
1845                         goto out;
1846                 r = PAGE_SIZE;     /* struct kvm_run */
1847 #ifdef CONFIG_X86
1848                 r += PAGE_SIZE;    /* pio data page */
1849 #endif
1850 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1851                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1852 #endif
1853                 break;
1854         case KVM_TRACE_ENABLE:
1855         case KVM_TRACE_PAUSE:
1856         case KVM_TRACE_DISABLE:
1857                 r = -EOPNOTSUPP;
1858                 break;
1859         default:
1860                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1861         }
1862 out:
1863         return r;
1864 }
1865
1866 static struct file_operations kvm_chardev_ops = {
1867         .unlocked_ioctl = kvm_dev_ioctl,
1868         .compat_ioctl   = kvm_dev_ioctl,
1869 };
1870
1871 static struct miscdevice kvm_dev = {
1872         KVM_MINOR,
1873         "kvm",
1874         &kvm_chardev_ops,
1875 };
1876
1877 static void hardware_enable(void *junk)
1878 {
1879         int cpu = raw_smp_processor_id();
1880         int r;
1881
1882         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1883                 return;
1884
1885         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1886
1887         r = kvm_arch_hardware_enable(NULL);
1888
1889         if (r) {
1890                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1891                 atomic_inc(&hardware_enable_failed);
1892                 printk(KERN_INFO "kvm: enabling virtualization on "
1893                                  "CPU%d failed\n", cpu);
1894         }
1895 }
1896
1897 static void hardware_disable(void *junk)
1898 {
1899         int cpu = raw_smp_processor_id();
1900
1901         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1902                 return;
1903         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1904         kvm_arch_hardware_disable(NULL);
1905 }
1906
1907 static void hardware_disable_all_nolock(void)
1908 {
1909         BUG_ON(!kvm_usage_count);
1910
1911         kvm_usage_count--;
1912         if (!kvm_usage_count)
1913                 on_each_cpu(hardware_disable, NULL, 1);
1914 }
1915
1916 static void hardware_disable_all(void)
1917 {
1918         spin_lock(&kvm_lock);
1919         hardware_disable_all_nolock();
1920         spin_unlock(&kvm_lock);
1921 }
1922
1923 static int hardware_enable_all(void)
1924 {
1925         int r = 0;
1926
1927         spin_lock(&kvm_lock);
1928
1929         kvm_usage_count++;
1930         if (kvm_usage_count == 1) {
1931                 atomic_set(&hardware_enable_failed, 0);
1932                 on_each_cpu(hardware_enable, NULL, 1);
1933
1934                 if (atomic_read(&hardware_enable_failed)) {
1935                         hardware_disable_all_nolock();
1936                         r = -EBUSY;
1937                 }
1938         }
1939
1940         spin_unlock(&kvm_lock);
1941
1942         return r;
1943 }
1944
1945 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1946                            void *v)
1947 {
1948         int cpu = (long)v;
1949
1950         if (!kvm_usage_count)
1951                 return NOTIFY_OK;
1952
1953         val &= ~CPU_TASKS_FROZEN;
1954         switch (val) {
1955         case CPU_DYING:
1956                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1957                        cpu);
1958                 hardware_disable(NULL);
1959                 break;
1960         case CPU_ONLINE:
1961                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1962                        cpu);
1963                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1964                 break;
1965         }
1966         return NOTIFY_OK;
1967 }
1968
1969
1970 asmlinkage void kvm_handle_fault_on_reboot(void)
1971 {
1972         if (kvm_rebooting)
1973                 /* spin while reset goes on */
1974                 while (true)
1975                         ;
1976         /* Fault while not rebooting.  We want the trace. */
1977         BUG();
1978 }
1979 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1980
1981 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1982                       void *v)
1983 {
1984         /*
1985          * Some (well, at least mine) BIOSes hang on reboot if
1986          * in vmx root mode.
1987          *
1988          * And Intel TXT required VMX off for all cpu when system shutdown.
1989          */
1990         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1991         kvm_rebooting = true;
1992         on_each_cpu(hardware_disable, NULL, 1);
1993         return NOTIFY_OK;
1994 }
1995
1996 static struct notifier_block kvm_reboot_notifier = {
1997         .notifier_call = kvm_reboot,
1998         .priority = 0,
1999 };
2000
2001 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2002 {
2003         int i;
2004
2005         for (i = 0; i < bus->dev_count; i++) {
2006                 struct kvm_io_device *pos = bus->devs[i];
2007
2008                 kvm_iodevice_destructor(pos);
2009         }
2010         kfree(bus);
2011 }
2012
2013 /* kvm_io_bus_write - called under kvm->slots_lock */
2014 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2015                      int len, const void *val)
2016 {
2017         int i;
2018         struct kvm_io_bus *bus;
2019
2020         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2021         for (i = 0; i < bus->dev_count; i++)
2022                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2023                         return 0;
2024         return -EOPNOTSUPP;
2025 }
2026
2027 /* kvm_io_bus_read - called under kvm->slots_lock */
2028 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2029                     int len, void *val)
2030 {
2031         int i;
2032         struct kvm_io_bus *bus;
2033
2034         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2035         for (i = 0; i < bus->dev_count; i++)
2036                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2037                         return 0;
2038         return -EOPNOTSUPP;
2039 }
2040
2041 /* Caller must hold slots_lock. */
2042 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2043                             struct kvm_io_device *dev)
2044 {
2045         struct kvm_io_bus *new_bus, *bus;
2046
2047         bus = kvm->buses[bus_idx];
2048         if (bus->dev_count > NR_IOBUS_DEVS-1)
2049                 return -ENOSPC;
2050
2051         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2052         if (!new_bus)
2053                 return -ENOMEM;
2054         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2055         new_bus->devs[new_bus->dev_count++] = dev;
2056         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2057         synchronize_srcu_expedited(&kvm->srcu);
2058         kfree(bus);
2059
2060         return 0;
2061 }
2062
2063 /* Caller must hold slots_lock. */
2064 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2065                               struct kvm_io_device *dev)
2066 {
2067         int i, r;
2068         struct kvm_io_bus *new_bus, *bus;
2069
2070         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2071         if (!new_bus)
2072                 return -ENOMEM;
2073
2074         bus = kvm->buses[bus_idx];
2075         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2076
2077         r = -ENOENT;
2078         for (i = 0; i < new_bus->dev_count; i++)
2079                 if (new_bus->devs[i] == dev) {
2080                         r = 0;
2081                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2082                         break;
2083                 }
2084
2085         if (r) {
2086                 kfree(new_bus);
2087                 return r;
2088         }
2089
2090         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2091         synchronize_srcu_expedited(&kvm->srcu);
2092         kfree(bus);
2093         return r;
2094 }
2095
2096 static struct notifier_block kvm_cpu_notifier = {
2097         .notifier_call = kvm_cpu_hotplug,
2098         .priority = 20, /* must be > scheduler priority */
2099 };
2100
2101 static int vm_stat_get(void *_offset, u64 *val)
2102 {
2103         unsigned offset = (long)_offset;
2104         struct kvm *kvm;
2105
2106         *val = 0;
2107         spin_lock(&kvm_lock);
2108         list_for_each_entry(kvm, &vm_list, vm_list)
2109                 *val += *(u32 *)((void *)kvm + offset);
2110         spin_unlock(&kvm_lock);
2111         return 0;
2112 }
2113
2114 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2115
2116 static int vcpu_stat_get(void *_offset, u64 *val)
2117 {
2118         unsigned offset = (long)_offset;
2119         struct kvm *kvm;
2120         struct kvm_vcpu *vcpu;
2121         int i;
2122
2123         *val = 0;
2124         spin_lock(&kvm_lock);
2125         list_for_each_entry(kvm, &vm_list, vm_list)
2126                 kvm_for_each_vcpu(i, vcpu, kvm)
2127                         *val += *(u32 *)((void *)vcpu + offset);
2128
2129         spin_unlock(&kvm_lock);
2130         return 0;
2131 }
2132
2133 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2134
2135 static const struct file_operations *stat_fops[] = {
2136         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2137         [KVM_STAT_VM]   = &vm_stat_fops,
2138 };
2139
2140 static void kvm_init_debug(void)
2141 {
2142         struct kvm_stats_debugfs_item *p;
2143
2144         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2145         for (p = debugfs_entries; p->name; ++p)
2146                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2147                                                 (void *)(long)p->offset,
2148                                                 stat_fops[p->kind]);
2149 }
2150
2151 static void kvm_exit_debug(void)
2152 {
2153         struct kvm_stats_debugfs_item *p;
2154
2155         for (p = debugfs_entries; p->name; ++p)
2156                 debugfs_remove(p->dentry);
2157         debugfs_remove(kvm_debugfs_dir);
2158 }
2159
2160 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2161 {
2162         if (kvm_usage_count)
2163                 hardware_disable(NULL);
2164         return 0;
2165 }
2166
2167 static int kvm_resume(struct sys_device *dev)
2168 {
2169         if (kvm_usage_count)
2170                 hardware_enable(NULL);
2171         return 0;
2172 }
2173
2174 static struct sysdev_class kvm_sysdev_class = {
2175         .name = "kvm",
2176         .suspend = kvm_suspend,
2177         .resume = kvm_resume,
2178 };
2179
2180 static struct sys_device kvm_sysdev = {
2181         .id = 0,
2182         .cls = &kvm_sysdev_class,
2183 };
2184
2185 struct page *bad_page;
2186 pfn_t bad_pfn;
2187
2188 static inline
2189 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2190 {
2191         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2192 }
2193
2194 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2195 {
2196         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2197
2198         kvm_arch_vcpu_load(vcpu, cpu);
2199 }
2200
2201 static void kvm_sched_out(struct preempt_notifier *pn,
2202                           struct task_struct *next)
2203 {
2204         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2205
2206         kvm_arch_vcpu_put(vcpu);
2207 }
2208
2209 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2210                   struct module *module)
2211 {
2212         int r;
2213         int cpu;
2214
2215         r = kvm_arch_init(opaque);
2216         if (r)
2217                 goto out_fail;
2218
2219         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2220
2221         if (bad_page == NULL) {
2222                 r = -ENOMEM;
2223                 goto out;
2224         }
2225
2226         bad_pfn = page_to_pfn(bad_page);
2227
2228         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2229
2230         if (hwpoison_page == NULL) {
2231                 r = -ENOMEM;
2232                 goto out_free_0;
2233         }
2234
2235         hwpoison_pfn = page_to_pfn(hwpoison_page);
2236
2237         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2238                 r = -ENOMEM;
2239                 goto out_free_0;
2240         }
2241
2242         r = kvm_arch_hardware_setup();
2243         if (r < 0)
2244                 goto out_free_0a;
2245
2246         for_each_online_cpu(cpu) {
2247                 smp_call_function_single(cpu,
2248                                 kvm_arch_check_processor_compat,
2249                                 &r, 1);
2250                 if (r < 0)
2251                         goto out_free_1;
2252         }
2253
2254         r = register_cpu_notifier(&kvm_cpu_notifier);
2255         if (r)
2256                 goto out_free_2;
2257         register_reboot_notifier(&kvm_reboot_notifier);
2258
2259         r = sysdev_class_register(&kvm_sysdev_class);
2260         if (r)
2261                 goto out_free_3;
2262
2263         r = sysdev_register(&kvm_sysdev);
2264         if (r)
2265                 goto out_free_4;
2266
2267         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2268         if (!vcpu_align)
2269                 vcpu_align = __alignof__(struct kvm_vcpu);
2270         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2271                                            0, NULL);
2272         if (!kvm_vcpu_cache) {
2273                 r = -ENOMEM;
2274                 goto out_free_5;
2275         }
2276
2277         kvm_chardev_ops.owner = module;
2278         kvm_vm_fops.owner = module;
2279         kvm_vcpu_fops.owner = module;
2280
2281         r = misc_register(&kvm_dev);
2282         if (r) {
2283                 printk(KERN_ERR "kvm: misc device register failed\n");
2284                 goto out_free;
2285         }
2286
2287         kvm_preempt_ops.sched_in = kvm_sched_in;
2288         kvm_preempt_ops.sched_out = kvm_sched_out;
2289
2290         kvm_init_debug();
2291
2292         return 0;
2293
2294 out_free:
2295         kmem_cache_destroy(kvm_vcpu_cache);
2296 out_free_5:
2297         sysdev_unregister(&kvm_sysdev);
2298 out_free_4:
2299         sysdev_class_unregister(&kvm_sysdev_class);
2300 out_free_3:
2301         unregister_reboot_notifier(&kvm_reboot_notifier);
2302         unregister_cpu_notifier(&kvm_cpu_notifier);
2303 out_free_2:
2304 out_free_1:
2305         kvm_arch_hardware_unsetup();
2306 out_free_0a:
2307         free_cpumask_var(cpus_hardware_enabled);
2308 out_free_0:
2309         if (hwpoison_page)
2310                 __free_page(hwpoison_page);
2311         __free_page(bad_page);
2312 out:
2313         kvm_arch_exit();
2314 out_fail:
2315         return r;
2316 }
2317 EXPORT_SYMBOL_GPL(kvm_init);
2318
2319 void kvm_exit(void)
2320 {
2321         kvm_exit_debug();
2322         misc_deregister(&kvm_dev);
2323         kmem_cache_destroy(kvm_vcpu_cache);
2324         sysdev_unregister(&kvm_sysdev);
2325         sysdev_class_unregister(&kvm_sysdev_class);
2326         unregister_reboot_notifier(&kvm_reboot_notifier);
2327         unregister_cpu_notifier(&kvm_cpu_notifier);
2328         on_each_cpu(hardware_disable, NULL, 1);
2329         kvm_arch_hardware_unsetup();
2330         kvm_arch_exit();
2331         free_cpumask_var(cpus_hardware_enabled);
2332         __free_page(hwpoison_page);
2333         __free_page(bad_page);
2334 }
2335 EXPORT_SYMBOL_GPL(kvm_exit);