SAFE public projects git trees. - safe/jmp/linux-2.6/blob - virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "iodev.h"
  19
  20 #include <linux/kvm_host.h>
  21 #include <linux/kvm.h>
  22 #include <linux/module.h>
  23 #include <linux/errno.h>
  24 #include <linux/percpu.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/sysdev.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48
  49 #include <asm/processor.h>
  50 #include <asm/io.h>
  51 #include <asm/uaccess.h>
  52 #include <asm/pgtable.h>
  53 #include <asm-generic/bitops/le.h>
  54
  55 #include "coalesced_mmio.h"
  56
  57 #define CREATE_TRACE_POINTS
  58 #include <trace/events/kvm.h>
  59
  60 MODULE_AUTHOR("Qumranet");
  61 MODULE_LICENSE("GPL");
  62
  63 /*
  64  * Ordering of locks:
  65  *
  66  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  67  */
  68
  69 DEFINE_SPINLOCK(kvm_lock);
  70 LIST_HEAD(vm_list);
  71
  72 static cpumask_var_t cpus_hardware_enabled;
  73 static int kvm_usage_count = 0;
  74 static atomic_t hardware_enable_failed;
  75
  76 struct kmem_cache *kvm_vcpu_cache;
  77 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  78
  79 static __read_mostly struct preempt_ops kvm_preempt_ops;
  80
  81 struct dentry *kvm_debugfs_dir;
  82
  83 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  84                            unsigned long arg);
  85 static int hardware_enable_all(void);
  86 static void hardware_disable_all(void);
  87
  88 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  89
  90 static bool kvm_rebooting;
  91
  92 static bool largepages_enabled = true;
  93
  94 inline int kvm_is_mmio_pfn(pfn_t pfn)
  95 {
  96         if (pfn_valid(pfn)) {
  97                 struct page *page = compound_head(pfn_to_page(pfn));
  98                 return PageReserved(page);
  99         }
 100
 101         return true;
 102 }
 103
 104 /*
 105  * Switches to specified vcpu, until a matching vcpu_put()
 106  */
 107 void vcpu_load(struct kvm_vcpu *vcpu)
 108 {
 109         int cpu;
 110
 111         mutex_lock(&vcpu->mutex);
 112         cpu = get_cpu();
 113         preempt_notifier_register(&vcpu->preempt_notifier);
 114         kvm_arch_vcpu_load(vcpu, cpu);
 115         put_cpu();
 116 }
 117
 118 void vcpu_put(struct kvm_vcpu *vcpu)
 119 {
 120         preempt_disable();
 121         kvm_arch_vcpu_put(vcpu);
 122         preempt_notifier_unregister(&vcpu->preempt_notifier);
 123         preempt_enable();
 124         mutex_unlock(&vcpu->mutex);
 125 }
 126
 127 static void ack_flush(void *_completed)
 128 {
 129 }
 130
 131 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 132 {
 133         int i, cpu, me;
 134         cpumask_var_t cpus;
 135         bool called = true;
 136         struct kvm_vcpu *vcpu;
 137
 138         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 139
 140         spin_lock(&kvm->requests_lock);
 141         me = smp_processor_id();
 142         kvm_for_each_vcpu(i, vcpu, kvm) {
 143                 if (test_and_set_bit(req, &vcpu->requests))
 144                         continue;
 145                 cpu = vcpu->cpu;
 146                 if (cpus != NULL && cpu != -1 && cpu != me)
 147                         cpumask_set_cpu(cpu, cpus);
 148         }
 149         if (unlikely(cpus == NULL))
 150                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 151         else if (!cpumask_empty(cpus))
 152                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 153         else
 154                 called = false;
 155         spin_unlock(&kvm->requests_lock);
 156         free_cpumask_var(cpus);
 157         return called;
 158 }
 159
 160 void kvm_flush_remote_tlbs(struct kvm *kvm)
 161 {
 162         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 163                 ++kvm->stat.remote_tlb_flush;
 164 }
 165
 166 void kvm_reload_remote_mmus(struct kvm *kvm)
 167 {
 168         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 169 }
 170
 171 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 172 {
 173         struct page *page;
 174         int r;
 175
 176         mutex_init(&vcpu->mutex);
 177         vcpu->cpu = -1;
 178         vcpu->kvm = kvm;
 179         vcpu->vcpu_id = id;
 180         init_waitqueue_head(&vcpu->wq);
 181
 182         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 183         if (!page) {
 184                 r = -ENOMEM;
 185                 goto fail;
 186         }
 187         vcpu->run = page_address(page);
 188
 189         r = kvm_arch_vcpu_init(vcpu);
 190         if (r < 0)
 191                 goto fail_free_run;
 192         return 0;
 193
 194 fail_free_run:
 195         free_page((unsigned long)vcpu->run);
 196 fail:
 197         return r;
 198 }
 199 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 200
 201 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 202 {
 203         kvm_arch_vcpu_uninit(vcpu);
 204         free_page((unsigned long)vcpu->run);
 205 }
 206 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 207
 208 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 209 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 210 {
 211         return container_of(mn, struct kvm, mmu_notifier);
 212 }
 213
 214 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 215                                              struct mm_struct *mm,
 216                                              unsigned long address)
 217 {
 218         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 219         int need_tlb_flush, idx;
 220
 221         /*
 222          * When ->invalidate_page runs, the linux pte has been zapped
 223          * already but the page is still allocated until
 224          * ->invalidate_page returns. So if we increase the sequence
 225          * here the kvm page fault will notice if the spte can't be
 226          * established because the page is going to be freed. If
 227          * instead the kvm page fault establishes the spte before
 228          * ->invalidate_page runs, kvm_unmap_hva will release it
 229          * before returning.
 230          *
 231          * The sequence increase only need to be seen at spin_unlock
 232          * time, and not at spin_lock time.
 233          *
 234          * Increasing the sequence after the spin_unlock would be
 235          * unsafe because the kvm page fault could then establish the
 236          * pte after kvm_unmap_hva returned, without noticing the page
 237          * is going to be freed.
 238          */
 239         idx = srcu_read_lock(&kvm->srcu);
 240         spin_lock(&kvm->mmu_lock);
 241         kvm->mmu_notifier_seq++;
 242         need_tlb_flush = kvm_unmap_hva(kvm, address);
 243         spin_unlock(&kvm->mmu_lock);
 244         srcu_read_unlock(&kvm->srcu, idx);
 245
 246         /* we've to flush the tlb before the pages can be freed */
 247         if (need_tlb_flush)
 248                 kvm_flush_remote_tlbs(kvm);
 249
 250 }
 251
 252 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 253                                         struct mm_struct *mm,
 254                                         unsigned long address,
 255                                         pte_t pte)
 256 {
 257         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 258         int idx;
 259
 260         idx = srcu_read_lock(&kvm->srcu);
 261         spin_lock(&kvm->mmu_lock);
 262         kvm->mmu_notifier_seq++;
 263         kvm_set_spte_hva(kvm, address, pte);
 264         spin_unlock(&kvm->mmu_lock);
 265         srcu_read_unlock(&kvm->srcu, idx);
 266 }
 267
 268 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 269                                                     struct mm_struct *mm,
 270                                                     unsigned long start,
 271                                                     unsigned long end)
 272 {
 273         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 274         int need_tlb_flush = 0, idx;
 275
 276         idx = srcu_read_lock(&kvm->srcu);
 277         spin_lock(&kvm->mmu_lock);
 278         /*
 279          * The count increase must become visible at unlock time as no
 280          * spte can be established without taking the mmu_lock and
 281          * count is also read inside the mmu_lock critical section.
 282          */
 283         kvm->mmu_notifier_count++;
 284         for (; start < end; start += PAGE_SIZE)
 285                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 286         spin_unlock(&kvm->mmu_lock);
 287         srcu_read_unlock(&kvm->srcu, idx);
 288
 289         /* we've to flush the tlb before the pages can be freed */
 290         if (need_tlb_flush)
 291                 kvm_flush_remote_tlbs(kvm);
 292 }
 293
 294 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 295                                                   struct mm_struct *mm,
 296                                                   unsigned long start,
 297                                                   unsigned long end)
 298 {
 299         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 300
 301         spin_lock(&kvm->mmu_lock);
 302         /*
 303          * This sequence increase will notify the kvm page fault that
 304          * the page that is going to be mapped in the spte could have
 305          * been freed.
 306          */
 307         kvm->mmu_notifier_seq++;
 308         /*
 309          * The above sequence increase must be visible before the
 310          * below count decrease but both values are read by the kvm
 311          * page fault under mmu_lock spinlock so we don't need to add
 312          * a smb_wmb() here in between the two.
 313          */
 314         kvm->mmu_notifier_count--;
 315         spin_unlock(&kvm->mmu_lock);
 316
 317         BUG_ON(kvm->mmu_notifier_count < 0);
 318 }
 319
 320 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 321                                               struct mm_struct *mm,
 322                                               unsigned long address)
 323 {
 324         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 325         int young, idx;
 326
 327         idx = srcu_read_lock(&kvm->srcu);
 328         spin_lock(&kvm->mmu_lock);
 329         young = kvm_age_hva(kvm, address);
 330         spin_unlock(&kvm->mmu_lock);
 331         srcu_read_unlock(&kvm->srcu, idx);
 332
 333         if (young)
 334                 kvm_flush_remote_tlbs(kvm);
 335
 336         return young;
 337 }
 338
 339 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 340                                      struct mm_struct *mm)
 341 {
 342         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 343         kvm_arch_flush_shadow(kvm);
 344 }
 345
 346 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 347         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 348         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 349         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 350         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 351         .change_pte             = kvm_mmu_notifier_change_pte,
 352         .release                = kvm_mmu_notifier_release,
 353 };
 354
 355 static int kvm_init_mmu_notifier(struct kvm *kvm)
 356 {
 357         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 358         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 359 }
 360
 361 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 362
 363 static int kvm_init_mmu_notifier(struct kvm *kvm)
 364 {
 365         return 0;
 366 }
 367
 368 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 369
 370 static struct kvm *kvm_create_vm(void)
 371 {
 372         int r = 0, i;
 373         struct kvm *kvm = kvm_arch_create_vm();
 374
 375         if (IS_ERR(kvm))
 376                 goto out;
 377
 378         r = hardware_enable_all();
 379         if (r)
 380                 goto out_err_nodisable;
 381
 382 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 383         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 384         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 385 #endif
 386
 387         r = -ENOMEM;
 388         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 389         if (!kvm->memslots)
 390                 goto out_err;
 391         if (init_srcu_struct(&kvm->srcu))
 392                 goto out_err;
 393         for (i = 0; i < KVM_NR_BUSES; i++) {
 394                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 395                                         GFP_KERNEL);
 396                 if (!kvm->buses[i]) {
 397                         cleanup_srcu_struct(&kvm->srcu);
 398                         goto out_err;
 399                 }
 400         }
 401
 402         r = kvm_init_mmu_notifier(kvm);
 403         if (r) {
 404                 cleanup_srcu_struct(&kvm->srcu);
 405                 goto out_err;
 406         }
 407
 408         kvm->mm = current->mm;
 409         atomic_inc(&kvm->mm->mm_count);
 410         spin_lock_init(&kvm->mmu_lock);
 411         spin_lock_init(&kvm->requests_lock);
 412         kvm_eventfd_init(kvm);
 413         mutex_init(&kvm->lock);
 414         mutex_init(&kvm->irq_lock);
 415         mutex_init(&kvm->slots_lock);
 416         atomic_set(&kvm->users_count, 1);
 417         spin_lock(&kvm_lock);
 418         list_add(&kvm->vm_list, &vm_list);
 419         spin_unlock(&kvm_lock);
 420 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 421         kvm_coalesced_mmio_init(kvm);
 422 #endif
 423 out:
 424         return kvm;
 425
 426 out_err:
 427         hardware_disable_all();
 428 out_err_nodisable:
 429         for (i = 0; i < KVM_NR_BUSES; i++)
 430                 kfree(kvm->buses[i]);
 431         kfree(kvm->memslots);
 432         kfree(kvm);
 433         return ERR_PTR(r);
 434 }
 435
 436 /*
 437  * Free any memory in @free but not in @dont.
 438  */
 439 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 440                                   struct kvm_memory_slot *dont)
 441 {
 442         int i;
 443
 444         if (!dont || free->rmap != dont->rmap)
 445                 vfree(free->rmap);
 446
 447         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 448                 vfree(free->dirty_bitmap);
 449
 450
 451         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 452                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 453                         vfree(free->lpage_info[i]);
 454                         free->lpage_info[i] = NULL;
 455                 }
 456         }
 457
 458         free->npages = 0;
 459         free->dirty_bitmap = NULL;
 460         free->rmap = NULL;
 461 }
 462
 463 void kvm_free_physmem(struct kvm *kvm)
 464 {
 465         int i;
 466         struct kvm_memslots *slots = kvm->memslots;
 467
 468         for (i = 0; i < slots->nmemslots; ++i)
 469                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 470
 471         kfree(kvm->memslots);
 472 }
 473
 474 static void kvm_destroy_vm(struct kvm *kvm)
 475 {
 476         int i;
 477         struct mm_struct *mm = kvm->mm;
 478
 479         kvm_arch_sync_events(kvm);
 480         spin_lock(&kvm_lock);
 481         list_del(&kvm->vm_list);
 482         spin_unlock(&kvm_lock);
 483         kvm_free_irq_routing(kvm);
 484         for (i = 0; i < KVM_NR_BUSES; i++)
 485                 kvm_io_bus_destroy(kvm->buses[i]);
 486         kvm_coalesced_mmio_free(kvm);
 487 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 488         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 489 #else
 490         kvm_arch_flush_shadow(kvm);
 491 #endif
 492         kvm_arch_destroy_vm(kvm);
 493         hardware_disable_all();
 494         mmdrop(mm);
 495 }
 496
 497 void kvm_get_kvm(struct kvm *kvm)
 498 {
 499         atomic_inc(&kvm->users_count);
 500 }
 501 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 502
 503 void kvm_put_kvm(struct kvm *kvm)
 504 {
 505         if (atomic_dec_and_test(&kvm->users_count))
 506                 kvm_destroy_vm(kvm);
 507 }
 508 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 509
 510
 511 static int kvm_vm_release(struct inode *inode, struct file *filp)
 512 {
 513         struct kvm *kvm = filp->private_data;
 514
 515         kvm_irqfd_release(kvm);
 516
 517         kvm_put_kvm(kvm);
 518         return 0;
 519 }
 520
 521 /*
 522  * Allocate some memory and give it an address in the guest physical address
 523  * space.
 524  *
 525  * Discontiguous memory is allowed, mostly for framebuffers.
 526  *
 527  * Must be called holding mmap_sem for write.
 528  */
 529 int __kvm_set_memory_region(struct kvm *kvm,
 530                             struct kvm_userspace_memory_region *mem,
 531                             int user_alloc)
 532 {
 533         int r, flush_shadow = 0;
 534         gfn_t base_gfn;
 535         unsigned long npages;
 536         unsigned long i;
 537         struct kvm_memory_slot *memslot;
 538         struct kvm_memory_slot old, new;
 539         struct kvm_memslots *slots, *old_memslots;
 540
 541         r = -EINVAL;
 542         /* General sanity checks */
 543         if (mem->memory_size & (PAGE_SIZE - 1))
 544                 goto out;
 545         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 546                 goto out;
 547         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 548                 goto out;
 549         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 550                 goto out;
 551         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 552                 goto out;
 553
 554         memslot = &kvm->memslots->memslots[mem->slot];
 555         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 556         npages = mem->memory_size >> PAGE_SHIFT;
 557
 558         if (!npages)
 559                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 560
 561         new = old = *memslot;
 562
 563         new.base_gfn = base_gfn;
 564         new.npages = npages;
 565         new.flags = mem->flags;
 566
 567         /* Disallow changing a memory slot's size. */
 568         r = -EINVAL;
 569         if (npages && old.npages && npages != old.npages)
 570                 goto out_free;
 571
 572         /* Check for overlaps */
 573         r = -EEXIST;
 574         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 575                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 576
 577                 if (s == memslot || !s->npages)
 578                         continue;
 579                 if (!((base_gfn + npages <= s->base_gfn) ||
 580                       (base_gfn >= s->base_gfn + s->npages)))
 581                         goto out_free;
 582         }
 583
 584         /* Free page dirty bitmap if unneeded */
 585         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 586                 new.dirty_bitmap = NULL;
 587
 588         r = -ENOMEM;
 589
 590         /* Allocate if a slot is being created */
 591 #ifndef CONFIG_S390
 592         if (npages && !new.rmap) {
 593                 new.rmap = vmalloc(npages * sizeof(struct page *));
 594
 595                 if (!new.rmap)
 596                         goto out_free;
 597
 598                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 599
 600                 new.user_alloc = user_alloc;
 601                 new.userspace_addr = mem->userspace_addr;
 602         }
 603         if (!npages)
 604                 goto skip_lpage;
 605
 606         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 607                 unsigned long ugfn;
 608                 unsigned long j;
 609                 int lpages;
 610                 int level = i + 2;
 611
 612                 /* Avoid unused variable warning if no large pages */
 613                 (void)level;
 614
 615                 if (new.lpage_info[i])
 616                         continue;
 617
 618                 lpages = 1 + (base_gfn + npages - 1) /
 619                              KVM_PAGES_PER_HPAGE(level);
 620                 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
 621
 622                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
 623
 624                 if (!new.lpage_info[i])
 625                         goto out_free;
 626
 627                 memset(new.lpage_info[i], 0,
 628                        lpages * sizeof(*new.lpage_info[i]));
 629
 630                 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
 631                         new.lpage_info[i][0].write_count = 1;
 632                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
 633                         new.lpage_info[i][lpages - 1].write_count = 1;
 634                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 635                 /*
 636                  * If the gfn and userspace address are not aligned wrt each
 637                  * other, or if explicitly asked to, disable large page
 638                  * support for this slot
 639                  */
 640                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 641                     !largepages_enabled)
 642                         for (j = 0; j < lpages; ++j)
 643                                 new.lpage_info[i][j].write_count = 1;
 644         }
 645
 646 skip_lpage:
 647
 648         /* Allocate page dirty bitmap if needed */
 649         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 650                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 651
 652                 new.dirty_bitmap = vmalloc(dirty_bytes);
 653                 if (!new.dirty_bitmap)
 654                         goto out_free;
 655                 memset(new.dirty_bitmap, 0, dirty_bytes);
 656                 /* destroy any largepage mappings for dirty tracking */
 657                 if (old.npages)
 658                         flush_shadow = 1;
 659         }
 660 #else  /* not defined CONFIG_S390 */
 661         new.user_alloc = user_alloc;
 662         if (user_alloc)
 663                 new.userspace_addr = mem->userspace_addr;
 664 #endif /* not defined CONFIG_S390 */
 665
 666         if (!npages) {
 667                 r = -ENOMEM;
 668                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 669                 if (!slots)
 670                         goto out_free;
 671                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 672                 if (mem->slot >= slots->nmemslots)
 673                         slots->nmemslots = mem->slot + 1;
 674                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 675
 676                 old_memslots = kvm->memslots;
 677                 rcu_assign_pointer(kvm->memslots, slots);
 678                 synchronize_srcu_expedited(&kvm->srcu);
 679                 /* From this point no new shadow pages pointing to a deleted
 680                  * memslot will be created.
 681                  *
 682                  * validation of sp->gfn happens in:
 683                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 684                  *      - kvm_is_visible_gfn (mmu_check_roots)
 685                  */
 686                 kvm_arch_flush_shadow(kvm);
 687                 kfree(old_memslots);
 688         }
 689
 690         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 691         if (r)
 692                 goto out_free;
 693
 694 #ifdef CONFIG_DMAR
 695         /* map the pages in iommu page table */
 696         if (npages) {
 697                 r = kvm_iommu_map_pages(kvm, &new);
 698                 if (r)
 699                         goto out_free;
 700         }
 701 #endif
 702
 703         r = -ENOMEM;
 704         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 705         if (!slots)
 706                 goto out_free;
 707         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 708         if (mem->slot >= slots->nmemslots)
 709                 slots->nmemslots = mem->slot + 1;
 710
 711         /* actual memory is freed via old in kvm_free_physmem_slot below */
 712         if (!npages) {
 713                 new.rmap = NULL;
 714                 new.dirty_bitmap = NULL;
 715                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 716                         new.lpage_info[i] = NULL;
 717         }
 718
 719         slots->memslots[mem->slot] = new;
 720         old_memslots = kvm->memslots;
 721         rcu_assign_pointer(kvm->memslots, slots);
 722         synchronize_srcu_expedited(&kvm->srcu);
 723
 724         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 725
 726         kvm_free_physmem_slot(&old, &new);
 727         kfree(old_memslots);
 728
 729         if (flush_shadow)
 730                 kvm_arch_flush_shadow(kvm);
 731
 732         return 0;
 733
 734 out_free:
 735         kvm_free_physmem_slot(&new, &old);
 736 out:
 737         return r;
 738
 739 }
 740 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 741
 742 int kvm_set_memory_region(struct kvm *kvm,
 743                           struct kvm_userspace_memory_region *mem,
 744                           int user_alloc)
 745 {
 746         int r;
 747
 748         mutex_lock(&kvm->slots_lock);
 749         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 750         mutex_unlock(&kvm->slots_lock);
 751         return r;
 752 }
 753 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 754
 755 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 756                                    struct
 757                                    kvm_userspace_memory_region *mem,
 758                                    int user_alloc)
 759 {
 760         if (mem->slot >= KVM_MEMORY_SLOTS)
 761                 return -EINVAL;
 762         return kvm_set_memory_region(kvm, mem, user_alloc);
 763 }
 764
 765 int kvm_get_dirty_log(struct kvm *kvm,
 766                         struct kvm_dirty_log *log, int *is_dirty)
 767 {
 768         struct kvm_memory_slot *memslot;
 769         int r, i;
 770         int n;
 771         unsigned long any = 0;
 772
 773         r = -EINVAL;
 774         if (log->slot >= KVM_MEMORY_SLOTS)
 775                 goto out;
 776
 777         memslot = &kvm->memslots->memslots[log->slot];
 778         r = -ENOENT;
 779         if (!memslot->dirty_bitmap)
 780                 goto out;
 781
 782         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 783
 784         for (i = 0; !any && i < n/sizeof(long); ++i)
 785                 any = memslot->dirty_bitmap[i];
 786
 787         r = -EFAULT;
 788         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 789                 goto out;
 790
 791         if (any)
 792                 *is_dirty = 1;
 793
 794         r = 0;
 795 out:
 796         return r;
 797 }
 798
 799 void kvm_disable_largepages(void)
 800 {
 801         largepages_enabled = false;
 802 }
 803 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 804
 805 int is_error_page(struct page *page)
 806 {
 807         return page == bad_page;
 808 }
 809 EXPORT_SYMBOL_GPL(is_error_page);
 810
 811 int is_error_pfn(pfn_t pfn)
 812 {
 813         return pfn == bad_pfn;
 814 }
 815 EXPORT_SYMBOL_GPL(is_error_pfn);
 816
 817 static inline unsigned long bad_hva(void)
 818 {
 819         return PAGE_OFFSET;
 820 }
 821
 822 int kvm_is_error_hva(unsigned long addr)
 823 {
 824         return addr == bad_hva();
 825 }
 826 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 827
 828 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
 829 {
 830         int i;
 831         struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
 832
 833         for (i = 0; i < slots->nmemslots; ++i) {
 834                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 835
 836                 if (gfn >= memslot->base_gfn
 837                     && gfn < memslot->base_gfn + memslot->npages)
 838                         return memslot;
 839         }
 840         return NULL;
 841 }
 842 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
 843
 844 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 845 {
 846         gfn = unalias_gfn(kvm, gfn);
 847         return gfn_to_memslot_unaliased(kvm, gfn);
 848 }
 849
 850 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 851 {
 852         int i;
 853         struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
 854
 855         gfn = unalias_gfn_instantiation(kvm, gfn);
 856         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 857                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 858
 859                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 860                         continue;
 861
 862                 if (gfn >= memslot->base_gfn
 863                     && gfn < memslot->base_gfn + memslot->npages)
 864                         return 1;
 865         }
 866         return 0;
 867 }
 868 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 869
 870 int memslot_id(struct kvm *kvm, gfn_t gfn)
 871 {
 872         int i;
 873         struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
 874         struct kvm_memory_slot *memslot = NULL;
 875
 876         gfn = unalias_gfn(kvm, gfn);
 877         for (i = 0; i < slots->nmemslots; ++i) {
 878                 memslot = &slots->memslots[i];
 879
 880                 if (gfn >= memslot->base_gfn
 881                     && gfn < memslot->base_gfn + memslot->npages)
 882                         break;
 883         }
 884
 885         return memslot - slots->memslots;
 886 }
 887
 888 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 889 {
 890         struct kvm_memory_slot *slot;
 891
 892         gfn = unalias_gfn_instantiation(kvm, gfn);
 893         slot = gfn_to_memslot_unaliased(kvm, gfn);
 894         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 895                 return bad_hva();
 896         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 897 }
 898 EXPORT_SYMBOL_GPL(gfn_to_hva);
 899
 900 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr)
 901 {
 902         struct page *page[1];
 903         int npages;
 904         pfn_t pfn;
 905
 906         might_sleep();
 907
 908         npages = get_user_pages_fast(addr, 1, 1, page);
 909
 910         if (unlikely(npages != 1)) {
 911                 struct vm_area_struct *vma;
 912
 913                 down_read(&current->mm->mmap_sem);
 914                 vma = find_vma(current->mm, addr);
 915
 916                 if (vma == NULL || addr < vma->vm_start ||
 917                     !(vma->vm_flags & VM_PFNMAP)) {
 918                         up_read(&current->mm->mmap_sem);
 919                         get_page(bad_page);
 920                         return page_to_pfn(bad_page);
 921                 }
 922
 923                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 924                 up_read(&current->mm->mmap_sem);
 925                 BUG_ON(!kvm_is_mmio_pfn(pfn));
 926         } else
 927                 pfn = page_to_pfn(page[0]);
 928
 929         return pfn;
 930 }
 931
 932 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 933 {
 934         unsigned long addr;
 935
 936         addr = gfn_to_hva(kvm, gfn);
 937         if (kvm_is_error_hva(addr)) {
 938                 get_page(bad_page);
 939                 return page_to_pfn(bad_page);
 940         }
 941
 942         return hva_to_pfn(kvm, addr);
 943 }
 944 EXPORT_SYMBOL_GPL(gfn_to_pfn);
 945
 946 static unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
 947 {
 948         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 949 }
 950
 951 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
 952                          struct kvm_memory_slot *slot, gfn_t gfn)
 953 {
 954         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
 955         return hva_to_pfn(kvm, addr);
 956 }
 957
 958 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 959 {
 960         pfn_t pfn;
 961
 962         pfn = gfn_to_pfn(kvm, gfn);
 963         if (!kvm_is_mmio_pfn(pfn))
 964                 return pfn_to_page(pfn);
 965
 966         WARN_ON(kvm_is_mmio_pfn(pfn));
 967
 968         get_page(bad_page);
 969         return bad_page;
 970 }
 971
 972 EXPORT_SYMBOL_GPL(gfn_to_page);
 973
 974 void kvm_release_page_clean(struct page *page)
 975 {
 976         kvm_release_pfn_clean(page_to_pfn(page));
 977 }
 978 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
 979
 980 void kvm_release_pfn_clean(pfn_t pfn)
 981 {
 982         if (!kvm_is_mmio_pfn(pfn))
 983                 put_page(pfn_to_page(pfn));
 984 }
 985 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
 986
 987 void kvm_release_page_dirty(struct page *page)
 988 {
 989         kvm_release_pfn_dirty(page_to_pfn(page));
 990 }
 991 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
 992
 993 void kvm_release_pfn_dirty(pfn_t pfn)
 994 {
 995         kvm_set_pfn_dirty(pfn);
 996         kvm_release_pfn_clean(pfn);
 997 }
 998 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
 999
1000 void kvm_set_page_dirty(struct page *page)
1001 {
1002         kvm_set_pfn_dirty(page_to_pfn(page));
1003 }
1004 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1005
1006 void kvm_set_pfn_dirty(pfn_t pfn)
1007 {
1008         if (!kvm_is_mmio_pfn(pfn)) {
1009                 struct page *page = pfn_to_page(pfn);
1010                 if (!PageReserved(page))
1011                         SetPageDirty(page);
1012         }
1013 }
1014 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1015
1016 void kvm_set_pfn_accessed(pfn_t pfn)
1017 {
1018         if (!kvm_is_mmio_pfn(pfn))
1019                 mark_page_accessed(pfn_to_page(pfn));
1020 }
1021 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1022
1023 void kvm_get_pfn(pfn_t pfn)
1024 {
1025         if (!kvm_is_mmio_pfn(pfn))
1026                 get_page(pfn_to_page(pfn));
1027 }
1028 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1029
1030 static int next_segment(unsigned long len, int offset)
1031 {
1032         if (len > PAGE_SIZE - offset)
1033                 return PAGE_SIZE - offset;
1034         else
1035                 return len;
1036 }
1037
1038 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1039                         int len)
1040 {
1041         int r;
1042         unsigned long addr;
1043
1044         addr = gfn_to_hva(kvm, gfn);
1045         if (kvm_is_error_hva(addr))
1046                 return -EFAULT;
1047         r = copy_from_user(data, (void __user *)addr + offset, len);
1048         if (r)
1049                 return -EFAULT;
1050         return 0;
1051 }
1052 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1053
1054 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1055 {
1056         gfn_t gfn = gpa >> PAGE_SHIFT;
1057         int seg;
1058         int offset = offset_in_page(gpa);
1059         int ret;
1060
1061         while ((seg = next_segment(len, offset)) != 0) {
1062                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1063                 if (ret < 0)
1064                         return ret;
1065                 offset = 0;
1066                 len -= seg;
1067                 data += seg;
1068                 ++gfn;
1069         }
1070         return 0;
1071 }
1072 EXPORT_SYMBOL_GPL(kvm_read_guest);
1073
1074 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1075                           unsigned long len)
1076 {
1077         int r;
1078         unsigned long addr;
1079         gfn_t gfn = gpa >> PAGE_SHIFT;
1080         int offset = offset_in_page(gpa);
1081
1082         addr = gfn_to_hva(kvm, gfn);
1083         if (kvm_is_error_hva(addr))
1084                 return -EFAULT;
1085         pagefault_disable();
1086         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1087         pagefault_enable();
1088         if (r)
1089                 return -EFAULT;
1090         return 0;
1091 }
1092 EXPORT_SYMBOL(kvm_read_guest_atomic);
1093
1094 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1095                          int offset, int len)
1096 {
1097         int r;
1098         unsigned long addr;
1099
1100         addr = gfn_to_hva(kvm, gfn);
1101         if (kvm_is_error_hva(addr))
1102                 return -EFAULT;
1103         r = copy_to_user((void __user *)addr + offset, data, len);
1104         if (r)
1105                 return -EFAULT;
1106         mark_page_dirty(kvm, gfn);
1107         return 0;
1108 }
1109 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1110
1111 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1112                     unsigned long len)
1113 {
1114         gfn_t gfn = gpa >> PAGE_SHIFT;
1115         int seg;
1116         int offset = offset_in_page(gpa);
1117         int ret;
1118
1119         while ((seg = next_segment(len, offset)) != 0) {
1120                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1121                 if (ret < 0)
1122                         return ret;
1123                 offset = 0;
1124                 len -= seg;
1125                 data += seg;
1126                 ++gfn;
1127         }
1128         return 0;
1129 }
1130
1131 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1132 {
1133         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1134 }
1135 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1136
1137 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, 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_clear_guest_page(kvm, gfn, offset, seg);
1146                 if (ret < 0)
1147                         return ret;
1148                 offset = 0;
1149                 len -= seg;
1150                 ++gfn;
1151         }
1152         return 0;
1153 }
1154 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1155
1156 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1157 {
1158         struct kvm_memory_slot *memslot;
1159
1160         gfn = unalias_gfn(kvm, gfn);
1161         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1162         if (memslot && memslot->dirty_bitmap) {
1163                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1164
1165                 /* avoid RMW */
1166                 if (!generic_test_le_bit(rel_gfn, memslot->dirty_bitmap))
1167                         generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1168         }
1169 }
1170
1171 /*
1172  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1173  */
1174 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1175 {
1176         DEFINE_WAIT(wait);
1177
1178         for (;;) {
1179                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1180
1181                 if (kvm_arch_vcpu_runnable(vcpu)) {
1182                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1183                         break;
1184                 }
1185                 if (kvm_cpu_has_pending_timer(vcpu))
1186                         break;
1187                 if (signal_pending(current))
1188                         break;
1189
1190                 schedule();
1191         }
1192
1193         finish_wait(&vcpu->wq, &wait);
1194 }
1195
1196 void kvm_resched(struct kvm_vcpu *vcpu)
1197 {
1198         if (!need_resched())
1199                 return;
1200         cond_resched();
1201 }
1202 EXPORT_SYMBOL_GPL(kvm_resched);
1203
1204 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1205 {
1206         ktime_t expires;
1207         DEFINE_WAIT(wait);
1208
1209         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1210
1211         /* Sleep for 100 us, and hope lock-holder got scheduled */
1212         expires = ktime_add_ns(ktime_get(), 100000UL);
1213         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1214
1215         finish_wait(&vcpu->wq, &wait);
1216 }
1217 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1218
1219 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1220 {
1221         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1222         struct page *page;
1223
1224         if (vmf->pgoff == 0)
1225                 page = virt_to_page(vcpu->run);
1226 #ifdef CONFIG_X86
1227         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1228                 page = virt_to_page(vcpu->arch.pio_data);
1229 #endif
1230 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1231         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1232                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1233 #endif
1234         else
1235                 return VM_FAULT_SIGBUS;
1236         get_page(page);
1237         vmf->page = page;
1238         return 0;
1239 }
1240
1241 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1242         .fault = kvm_vcpu_fault,
1243 };
1244
1245 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1246 {
1247         vma->vm_ops = &kvm_vcpu_vm_ops;
1248         return 0;
1249 }
1250
1251 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1252 {
1253         struct kvm_vcpu *vcpu = filp->private_data;
1254
1255         kvm_put_kvm(vcpu->kvm);
1256         return 0;
1257 }
1258
1259 static struct file_operations kvm_vcpu_fops = {
1260         .release        = kvm_vcpu_release,
1261         .unlocked_ioctl = kvm_vcpu_ioctl,
1262         .compat_ioctl   = kvm_vcpu_ioctl,
1263         .mmap           = kvm_vcpu_mmap,
1264 };
1265
1266 /*
1267  * Allocates an inode for the vcpu.
1268  */
1269 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1270 {
1271         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1272 }
1273
1274 /*
1275  * Creates some virtual cpus.  Good luck creating more than one.
1276  */
1277 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1278 {
1279         int r;
1280         struct kvm_vcpu *vcpu, *v;
1281
1282         vcpu = kvm_arch_vcpu_create(kvm, id);
1283         if (IS_ERR(vcpu))
1284                 return PTR_ERR(vcpu);
1285
1286         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1287
1288         r = kvm_arch_vcpu_setup(vcpu);
1289         if (r)
1290                 return r;
1291
1292         mutex_lock(&kvm->lock);
1293         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1294                 r = -EINVAL;
1295                 goto vcpu_destroy;
1296         }
1297
1298         kvm_for_each_vcpu(r, v, kvm)
1299                 if (v->vcpu_id == id) {
1300                         r = -EEXIST;
1301                         goto vcpu_destroy;
1302                 }
1303
1304         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1305
1306         /* Now it's all set up, let userspace reach it */
1307         kvm_get_kvm(kvm);
1308         r = create_vcpu_fd(vcpu);
1309         if (r < 0) {
1310                 kvm_put_kvm(kvm);
1311                 goto vcpu_destroy;
1312         }
1313
1314         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1315         smp_wmb();
1316         atomic_inc(&kvm->online_vcpus);
1317
1318 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1319         if (kvm->bsp_vcpu_id == id)
1320                 kvm->bsp_vcpu = vcpu;
1321 #endif
1322         mutex_unlock(&kvm->lock);
1323         return r;
1324
1325 vcpu_destroy:
1326         mutex_unlock(&kvm->lock);
1327         kvm_arch_vcpu_destroy(vcpu);
1328         return r;
1329 }
1330
1331 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1332 {
1333         if (sigset) {
1334                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1335                 vcpu->sigset_active = 1;
1336                 vcpu->sigset = *sigset;
1337         } else
1338                 vcpu->sigset_active = 0;
1339         return 0;
1340 }
1341
1342 static long kvm_vcpu_ioctl(struct file *filp,
1343                            unsigned int ioctl, unsigned long arg)
1344 {
1345         struct kvm_vcpu *vcpu = filp->private_data;
1346         void __user *argp = (void __user *)arg;
1347         int r;
1348         struct kvm_fpu *fpu = NULL;
1349         struct kvm_sregs *kvm_sregs = NULL;
1350
1351         if (vcpu->kvm->mm != current->mm)
1352                 return -EIO;
1353         switch (ioctl) {
1354         case KVM_RUN:
1355                 r = -EINVAL;
1356                 if (arg)
1357                         goto out;
1358                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1359                 break;
1360         case KVM_GET_REGS: {
1361                 struct kvm_regs *kvm_regs;
1362
1363                 r = -ENOMEM;
1364                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1365                 if (!kvm_regs)
1366                         goto out;
1367                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1368                 if (r)
1369                         goto out_free1;
1370                 r = -EFAULT;
1371                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1372                         goto out_free1;
1373                 r = 0;
1374 out_free1:
1375                 kfree(kvm_regs);
1376                 break;
1377         }
1378         case KVM_SET_REGS: {
1379                 struct kvm_regs *kvm_regs;
1380
1381                 r = -ENOMEM;
1382                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1383                 if (!kvm_regs)
1384                         goto out;
1385                 r = -EFAULT;
1386                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1387                         goto out_free2;
1388                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1389                 if (r)
1390                         goto out_free2;
1391                 r = 0;
1392 out_free2:
1393                 kfree(kvm_regs);
1394                 break;
1395         }
1396         case KVM_GET_SREGS: {
1397                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1398                 r = -ENOMEM;
1399                 if (!kvm_sregs)
1400                         goto out;
1401                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1402                 if (r)
1403                         goto out;
1404                 r = -EFAULT;
1405                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1406                         goto out;
1407                 r = 0;
1408                 break;
1409         }
1410         case KVM_SET_SREGS: {
1411                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1412                 r = -ENOMEM;
1413                 if (!kvm_sregs)
1414                         goto out;
1415                 r = -EFAULT;
1416                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1417                         goto out;
1418                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1419                 if (r)
1420                         goto out;
1421                 r = 0;
1422                 break;
1423         }
1424         case KVM_GET_MP_STATE: {
1425                 struct kvm_mp_state mp_state;
1426
1427                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1428                 if (r)
1429                         goto out;
1430                 r = -EFAULT;
1431                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1432                         goto out;
1433                 r = 0;
1434                 break;
1435         }
1436         case KVM_SET_MP_STATE: {
1437                 struct kvm_mp_state mp_state;
1438
1439                 r = -EFAULT;
1440                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1441                         goto out;
1442                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1443                 if (r)
1444                         goto out;
1445                 r = 0;
1446                 break;
1447         }
1448         case KVM_TRANSLATE: {
1449                 struct kvm_translation tr;
1450
1451                 r = -EFAULT;
1452                 if (copy_from_user(&tr, argp, sizeof tr))
1453                         goto out;
1454                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1455                 if (r)
1456                         goto out;
1457                 r = -EFAULT;
1458                 if (copy_to_user(argp, &tr, sizeof tr))
1459                         goto out;
1460                 r = 0;
1461                 break;
1462         }
1463         case KVM_SET_GUEST_DEBUG: {
1464                 struct kvm_guest_debug dbg;
1465
1466                 r = -EFAULT;
1467                 if (copy_from_user(&dbg, argp, sizeof dbg))
1468                         goto out;
1469                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1470                 if (r)
1471                         goto out;
1472                 r = 0;
1473                 break;
1474         }
1475         case KVM_SET_SIGNAL_MASK: {
1476                 struct kvm_signal_mask __user *sigmask_arg = argp;
1477                 struct kvm_signal_mask kvm_sigmask;
1478                 sigset_t sigset, *p;
1479
1480                 p = NULL;
1481                 if (argp) {
1482                         r = -EFAULT;
1483                         if (copy_from_user(&kvm_sigmask, argp,
1484                                            sizeof kvm_sigmask))
1485                                 goto out;
1486                         r = -EINVAL;
1487                         if (kvm_sigmask.len != sizeof sigset)
1488                                 goto out;
1489                         r = -EFAULT;
1490                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1491                                            sizeof sigset))
1492                                 goto out;
1493                         p = &sigset;
1494                 }
1495                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1496                 break;
1497         }
1498         case KVM_GET_FPU: {
1499                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1500                 r = -ENOMEM;
1501                 if (!fpu)
1502                         goto out;
1503                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1504                 if (r)
1505                         goto out;
1506                 r = -EFAULT;
1507                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1508                         goto out;
1509                 r = 0;
1510                 break;
1511         }
1512         case KVM_SET_FPU: {
1513                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1514                 r = -ENOMEM;
1515                 if (!fpu)
1516                         goto out;
1517                 r = -EFAULT;
1518                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1519                         goto out;
1520                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1521                 if (r)
1522                         goto out;
1523                 r = 0;
1524                 break;
1525         }
1526         default:
1527                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1528         }
1529 out:
1530         kfree(fpu);
1531         kfree(kvm_sregs);
1532         return r;
1533 }
1534
1535 static long kvm_vm_ioctl(struct file *filp,
1536                            unsigned int ioctl, unsigned long arg)
1537 {
1538         struct kvm *kvm = filp->private_data;
1539         void __user *argp = (void __user *)arg;
1540         int r;
1541
1542         if (kvm->mm != current->mm)
1543                 return -EIO;
1544         switch (ioctl) {
1545         case KVM_CREATE_VCPU:
1546                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1547                 if (r < 0)
1548                         goto out;
1549                 break;
1550         case KVM_SET_USER_MEMORY_REGION: {
1551                 struct kvm_userspace_memory_region kvm_userspace_mem;
1552
1553                 r = -EFAULT;
1554                 if (copy_from_user(&kvm_userspace_mem, argp,
1555                                                 sizeof kvm_userspace_mem))
1556                         goto out;
1557
1558                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1559                 if (r)
1560                         goto out;
1561                 break;
1562         }
1563         case KVM_GET_DIRTY_LOG: {
1564                 struct kvm_dirty_log log;
1565
1566                 r = -EFAULT;
1567                 if (copy_from_user(&log, argp, sizeof log))
1568                         goto out;
1569                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1570                 if (r)
1571                         goto out;
1572                 break;
1573         }
1574 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1575         case KVM_REGISTER_COALESCED_MMIO: {
1576                 struct kvm_coalesced_mmio_zone zone;
1577                 r = -EFAULT;
1578                 if (copy_from_user(&zone, argp, sizeof zone))
1579                         goto out;
1580                 r = -ENXIO;
1581                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1582                 if (r)
1583                         goto out;
1584                 r = 0;
1585                 break;
1586         }
1587         case KVM_UNREGISTER_COALESCED_MMIO: {
1588                 struct kvm_coalesced_mmio_zone zone;
1589                 r = -EFAULT;
1590                 if (copy_from_user(&zone, argp, sizeof zone))
1591                         goto out;
1592                 r = -ENXIO;
1593                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1594                 if (r)
1595                         goto out;
1596                 r = 0;
1597                 break;
1598         }
1599 #endif
1600         case KVM_IRQFD: {
1601                 struct kvm_irqfd data;
1602
1603                 r = -EFAULT;
1604                 if (copy_from_user(&data, argp, sizeof data))
1605                         goto out;
1606                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1607                 break;
1608         }
1609         case KVM_IOEVENTFD: {
1610                 struct kvm_ioeventfd data;
1611
1612                 r = -EFAULT;
1613                 if (copy_from_user(&data, argp, sizeof data))
1614                         goto out;
1615                 r = kvm_ioeventfd(kvm, &data);
1616                 break;
1617         }
1618 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1619         case KVM_SET_BOOT_CPU_ID:
1620                 r = 0;
1621                 mutex_lock(&kvm->lock);
1622                 if (atomic_read(&kvm->online_vcpus) != 0)
1623                         r = -EBUSY;
1624                 else
1625                         kvm->bsp_vcpu_id = arg;
1626                 mutex_unlock(&kvm->lock);
1627                 break;
1628 #endif
1629         default:
1630                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1631                 if (r == -ENOTTY)
1632                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1633         }
1634 out:
1635         return r;
1636 }
1637
1638 #ifdef CONFIG_COMPAT
1639 struct compat_kvm_dirty_log {
1640         __u32 slot;
1641         __u32 padding1;
1642         union {
1643                 compat_uptr_t dirty_bitmap; /* one bit per page */
1644                 __u64 padding2;
1645         };
1646 };
1647
1648 static long kvm_vm_compat_ioctl(struct file *filp,
1649                            unsigned int ioctl, unsigned long arg)
1650 {
1651         struct kvm *kvm = filp->private_data;
1652         int r;
1653
1654         if (kvm->mm != current->mm)
1655                 return -EIO;
1656         switch (ioctl) {
1657         case KVM_GET_DIRTY_LOG: {
1658                 struct compat_kvm_dirty_log compat_log;
1659                 struct kvm_dirty_log log;
1660
1661                 r = -EFAULT;
1662                 if (copy_from_user(&compat_log, (void __user *)arg,
1663                                    sizeof(compat_log)))
1664                         goto out;
1665                 log.slot         = compat_log.slot;
1666                 log.padding1     = compat_log.padding1;
1667                 log.padding2     = compat_log.padding2;
1668                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1669
1670                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1671                 if (r)
1672                         goto out;
1673                 break;
1674         }
1675         default:
1676                 r = kvm_vm_ioctl(filp, ioctl, arg);
1677         }
1678
1679 out:
1680         return r;
1681 }
1682 #endif
1683
1684 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1685 {
1686         struct page *page[1];
1687         unsigned long addr;
1688         int npages;
1689         gfn_t gfn = vmf->pgoff;
1690         struct kvm *kvm = vma->vm_file->private_data;
1691
1692         addr = gfn_to_hva(kvm, gfn);
1693         if (kvm_is_error_hva(addr))
1694                 return VM_FAULT_SIGBUS;
1695
1696         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1697                                 NULL);
1698         if (unlikely(npages != 1))
1699                 return VM_FAULT_SIGBUS;
1700
1701         vmf->page = page[0];
1702         return 0;
1703 }
1704
1705 static const struct vm_operations_struct kvm_vm_vm_ops = {
1706         .fault = kvm_vm_fault,
1707 };
1708
1709 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1710 {
1711         vma->vm_ops = &kvm_vm_vm_ops;
1712         return 0;
1713 }
1714
1715 static struct file_operations kvm_vm_fops = {
1716         .release        = kvm_vm_release,
1717         .unlocked_ioctl = kvm_vm_ioctl,
1718 #ifdef CONFIG_COMPAT
1719         .compat_ioctl   = kvm_vm_compat_ioctl,
1720 #endif
1721         .mmap           = kvm_vm_mmap,
1722 };
1723
1724 static int kvm_dev_ioctl_create_vm(void)
1725 {
1726         int fd;
1727         struct kvm *kvm;
1728
1729         kvm = kvm_create_vm();
1730         if (IS_ERR(kvm))
1731                 return PTR_ERR(kvm);
1732         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1733         if (fd < 0)
1734                 kvm_put_kvm(kvm);
1735
1736         return fd;
1737 }
1738
1739 static long kvm_dev_ioctl_check_extension_generic(long arg)
1740 {
1741         switch (arg) {
1742         case KVM_CAP_USER_MEMORY:
1743         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1744         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1745 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1746         case KVM_CAP_SET_BOOT_CPU_ID:
1747 #endif
1748         case KVM_CAP_INTERNAL_ERROR_DATA:
1749                 return 1;
1750 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1751         case KVM_CAP_IRQ_ROUTING:
1752                 return KVM_MAX_IRQ_ROUTES;
1753 #endif
1754         default:
1755                 break;
1756         }
1757         return kvm_dev_ioctl_check_extension(arg);
1758 }
1759
1760 static long kvm_dev_ioctl(struct file *filp,
1761                           unsigned int ioctl, unsigned long arg)
1762 {
1763         long r = -EINVAL;
1764
1765         switch (ioctl) {
1766         case KVM_GET_API_VERSION:
1767                 r = -EINVAL;
1768                 if (arg)
1769                         goto out;
1770                 r = KVM_API_VERSION;
1771                 break;
1772         case KVM_CREATE_VM:
1773                 r = -EINVAL;
1774                 if (arg)
1775                         goto out;
1776                 r = kvm_dev_ioctl_create_vm();
1777                 break;
1778         case KVM_CHECK_EXTENSION:
1779                 r = kvm_dev_ioctl_check_extension_generic(arg);
1780                 break;
1781         case KVM_GET_VCPU_MMAP_SIZE:
1782                 r = -EINVAL;
1783                 if (arg)
1784                         goto out;
1785                 r = PAGE_SIZE;     /* struct kvm_run */
1786 #ifdef CONFIG_X86
1787                 r += PAGE_SIZE;    /* pio data page */
1788 #endif
1789 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1790                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1791 #endif
1792                 break;
1793         case KVM_TRACE_ENABLE:
1794         case KVM_TRACE_PAUSE:
1795         case KVM_TRACE_DISABLE:
1796                 r = -EOPNOTSUPP;
1797                 break;
1798         default:
1799                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1800         }
1801 out:
1802         return r;
1803 }
1804
1805 static struct file_operations kvm_chardev_ops = {
1806         .unlocked_ioctl = kvm_dev_ioctl,
1807         .compat_ioctl   = kvm_dev_ioctl,
1808 };
1809
1810 static struct miscdevice kvm_dev = {
1811         KVM_MINOR,
1812         "kvm",
1813         &kvm_chardev_ops,
1814 };
1815
1816 static void hardware_enable(void *junk)
1817 {
1818         int cpu = raw_smp_processor_id();
1819         int r;
1820
1821         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1822                 return;
1823
1824         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1825
1826         r = kvm_arch_hardware_enable(NULL);
1827
1828         if (r) {
1829                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1830                 atomic_inc(&hardware_enable_failed);
1831                 printk(KERN_INFO "kvm: enabling virtualization on "
1832                                  "CPU%d failed\n", cpu);
1833         }
1834 }
1835
1836 static void hardware_disable(void *junk)
1837 {
1838         int cpu = raw_smp_processor_id();
1839
1840         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1841                 return;
1842         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1843         kvm_arch_hardware_disable(NULL);
1844 }
1845
1846 static void hardware_disable_all_nolock(void)
1847 {
1848         BUG_ON(!kvm_usage_count);
1849
1850         kvm_usage_count--;
1851         if (!kvm_usage_count)
1852                 on_each_cpu(hardware_disable, NULL, 1);
1853 }
1854
1855 static void hardware_disable_all(void)
1856 {
1857         spin_lock(&kvm_lock);
1858         hardware_disable_all_nolock();
1859         spin_unlock(&kvm_lock);
1860 }
1861
1862 static int hardware_enable_all(void)
1863 {
1864         int r = 0;
1865
1866         spin_lock(&kvm_lock);
1867
1868         kvm_usage_count++;
1869         if (kvm_usage_count == 1) {
1870                 atomic_set(&hardware_enable_failed, 0);
1871                 on_each_cpu(hardware_enable, NULL, 1);
1872
1873                 if (atomic_read(&hardware_enable_failed)) {
1874                         hardware_disable_all_nolock();
1875                         r = -EBUSY;
1876                 }
1877         }
1878
1879         spin_unlock(&kvm_lock);
1880
1881         return r;
1882 }
1883
1884 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1885                            void *v)
1886 {
1887         int cpu = (long)v;
1888
1889         if (!kvm_usage_count)
1890                 return NOTIFY_OK;
1891
1892         val &= ~CPU_TASKS_FROZEN;
1893         switch (val) {
1894         case CPU_DYING:
1895                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1896                        cpu);
1897                 hardware_disable(NULL);
1898                 break;
1899         case CPU_UP_CANCELED:
1900                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1901                        cpu);
1902                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1903                 break;
1904         case CPU_ONLINE:
1905                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1906                        cpu);
1907                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1908                 break;
1909         }
1910         return NOTIFY_OK;
1911 }
1912
1913
1914 asmlinkage void kvm_handle_fault_on_reboot(void)
1915 {
1916         if (kvm_rebooting)
1917                 /* spin while reset goes on */
1918                 while (true)
1919                         ;
1920         /* Fault while not rebooting.  We want the trace. */
1921         BUG();
1922 }
1923 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1924
1925 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1926                       void *v)
1927 {
1928         /*
1929          * Some (well, at least mine) BIOSes hang on reboot if
1930          * in vmx root mode.
1931          *
1932          * And Intel TXT required VMX off for all cpu when system shutdown.
1933          */
1934         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1935         kvm_rebooting = true;
1936         on_each_cpu(hardware_disable, NULL, 1);
1937         return NOTIFY_OK;
1938 }
1939
1940 static struct notifier_block kvm_reboot_notifier = {
1941         .notifier_call = kvm_reboot,
1942         .priority = 0,
1943 };
1944
1945 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1946 {
1947         int i;
1948
1949         for (i = 0; i < bus->dev_count; i++) {
1950                 struct kvm_io_device *pos = bus->devs[i];
1951
1952                 kvm_iodevice_destructor(pos);
1953         }
1954         kfree(bus);
1955 }
1956
1957 /* kvm_io_bus_write - called under kvm->slots_lock */
1958 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
1959                      int len, const void *val)
1960 {
1961         int i;
1962         struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
1963         for (i = 0; i < bus->dev_count; i++)
1964                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
1965                         return 0;
1966         return -EOPNOTSUPP;
1967 }
1968
1969 /* kvm_io_bus_read - called under kvm->slots_lock */
1970 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
1971                     int len, void *val)
1972 {
1973         int i;
1974         struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
1975
1976         for (i = 0; i < bus->dev_count; i++)
1977                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
1978                         return 0;
1979         return -EOPNOTSUPP;
1980 }
1981
1982 /* Caller must hold slots_lock. */
1983 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
1984                             struct kvm_io_device *dev)
1985 {
1986         struct kvm_io_bus *new_bus, *bus;
1987
1988         bus = kvm->buses[bus_idx];
1989         if (bus->dev_count > NR_IOBUS_DEVS-1)
1990                 return -ENOSPC;
1991
1992         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
1993         if (!new_bus)
1994                 return -ENOMEM;
1995         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
1996         new_bus->devs[new_bus->dev_count++] = dev;
1997         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
1998         synchronize_srcu_expedited(&kvm->srcu);
1999         kfree(bus);
2000
2001         return 0;
2002 }
2003
2004 /* Caller must hold slots_lock. */
2005 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2006                               struct kvm_io_device *dev)
2007 {
2008         int i, r;
2009         struct kvm_io_bus *new_bus, *bus;
2010
2011         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2012         if (!new_bus)
2013                 return -ENOMEM;
2014
2015         bus = kvm->buses[bus_idx];
2016         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2017
2018         r = -ENOENT;
2019         for (i = 0; i < new_bus->dev_count; i++)
2020                 if (new_bus->devs[i] == dev) {
2021                         r = 0;
2022                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2023                         break;
2024                 }
2025
2026         if (r) {
2027                 kfree(new_bus);
2028                 return r;
2029         }
2030
2031         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2032         synchronize_srcu_expedited(&kvm->srcu);
2033         kfree(bus);
2034         return r;
2035 }
2036
2037 static struct notifier_block kvm_cpu_notifier = {
2038         .notifier_call = kvm_cpu_hotplug,
2039         .priority = 20, /* must be > scheduler priority */
2040 };
2041
2042 static int vm_stat_get(void *_offset, u64 *val)
2043 {
2044         unsigned offset = (long)_offset;
2045         struct kvm *kvm;
2046
2047         *val = 0;
2048         spin_lock(&kvm_lock);
2049         list_for_each_entry(kvm, &vm_list, vm_list)
2050                 *val += *(u32 *)((void *)kvm + offset);
2051         spin_unlock(&kvm_lock);
2052         return 0;
2053 }
2054
2055 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2056
2057 static int vcpu_stat_get(void *_offset, u64 *val)
2058 {
2059         unsigned offset = (long)_offset;
2060         struct kvm *kvm;
2061         struct kvm_vcpu *vcpu;
2062         int i;
2063
2064         *val = 0;
2065         spin_lock(&kvm_lock);
2066         list_for_each_entry(kvm, &vm_list, vm_list)
2067                 kvm_for_each_vcpu(i, vcpu, kvm)
2068                         *val += *(u32 *)((void *)vcpu + offset);
2069
2070         spin_unlock(&kvm_lock);
2071         return 0;
2072 }
2073
2074 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2075
2076 static const struct file_operations *stat_fops[] = {
2077         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2078         [KVM_STAT_VM]   = &vm_stat_fops,
2079 };
2080
2081 static void kvm_init_debug(void)
2082 {
2083         struct kvm_stats_debugfs_item *p;
2084
2085         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2086         for (p = debugfs_entries; p->name; ++p)
2087                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2088                                                 (void *)(long)p->offset,
2089                                                 stat_fops[p->kind]);
2090 }
2091
2092 static void kvm_exit_debug(void)
2093 {
2094         struct kvm_stats_debugfs_item *p;
2095
2096         for (p = debugfs_entries; p->name; ++p)
2097                 debugfs_remove(p->dentry);
2098         debugfs_remove(kvm_debugfs_dir);
2099 }
2100
2101 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2102 {
2103         if (kvm_usage_count)
2104                 hardware_disable(NULL);
2105         return 0;
2106 }
2107
2108 static int kvm_resume(struct sys_device *dev)
2109 {
2110         if (kvm_usage_count)
2111                 hardware_enable(NULL);
2112         return 0;
2113 }
2114
2115 static struct sysdev_class kvm_sysdev_class = {
2116         .name = "kvm",
2117         .suspend = kvm_suspend,
2118         .resume = kvm_resume,
2119 };
2120
2121 static struct sys_device kvm_sysdev = {
2122         .id = 0,
2123         .cls = &kvm_sysdev_class,
2124 };
2125
2126 struct page *bad_page;
2127 pfn_t bad_pfn;
2128
2129 static inline
2130 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2131 {
2132         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2133 }
2134
2135 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2136 {
2137         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2138
2139         kvm_arch_vcpu_load(vcpu, cpu);
2140 }
2141
2142 static void kvm_sched_out(struct preempt_notifier *pn,
2143                           struct task_struct *next)
2144 {
2145         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2146
2147         kvm_arch_vcpu_put(vcpu);
2148 }
2149
2150 int kvm_init(void *opaque, unsigned int vcpu_size,
2151                   struct module *module)
2152 {
2153         int r;
2154         int cpu;
2155
2156         r = kvm_arch_init(opaque);
2157         if (r)
2158                 goto out_fail;
2159
2160         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2161
2162         if (bad_page == NULL) {
2163                 r = -ENOMEM;
2164                 goto out;
2165         }
2166
2167         bad_pfn = page_to_pfn(bad_page);
2168
2169         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2170                 r = -ENOMEM;
2171                 goto out_free_0;
2172         }
2173
2174         r = kvm_arch_hardware_setup();
2175         if (r < 0)
2176                 goto out_free_0a;
2177
2178         for_each_online_cpu(cpu) {
2179                 smp_call_function_single(cpu,
2180                                 kvm_arch_check_processor_compat,
2181                                 &r, 1);
2182                 if (r < 0)
2183                         goto out_free_1;
2184         }
2185
2186         r = register_cpu_notifier(&kvm_cpu_notifier);
2187         if (r)
2188                 goto out_free_2;
2189         register_reboot_notifier(&kvm_reboot_notifier);
2190
2191         r = sysdev_class_register(&kvm_sysdev_class);
2192         if (r)
2193                 goto out_free_3;
2194
2195         r = sysdev_register(&kvm_sysdev);
2196         if (r)
2197                 goto out_free_4;
2198
2199         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2200         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2201                                            __alignof__(struct kvm_vcpu),
2202                                            0, NULL);
2203         if (!kvm_vcpu_cache) {
2204                 r = -ENOMEM;
2205                 goto out_free_5;
2206         }
2207
2208         kvm_chardev_ops.owner = module;
2209         kvm_vm_fops.owner = module;
2210         kvm_vcpu_fops.owner = module;
2211
2212         r = misc_register(&kvm_dev);
2213         if (r) {
2214                 printk(KERN_ERR "kvm: misc device register failed\n");
2215                 goto out_free;
2216         }
2217
2218         kvm_preempt_ops.sched_in = kvm_sched_in;
2219         kvm_preempt_ops.sched_out = kvm_sched_out;
2220
2221         kvm_init_debug();
2222
2223         return 0;
2224
2225 out_free:
2226         kmem_cache_destroy(kvm_vcpu_cache);
2227 out_free_5:
2228         sysdev_unregister(&kvm_sysdev);
2229 out_free_4:
2230         sysdev_class_unregister(&kvm_sysdev_class);
2231 out_free_3:
2232         unregister_reboot_notifier(&kvm_reboot_notifier);
2233         unregister_cpu_notifier(&kvm_cpu_notifier);
2234 out_free_2:
2235 out_free_1:
2236         kvm_arch_hardware_unsetup();
2237 out_free_0a:
2238         free_cpumask_var(cpus_hardware_enabled);
2239 out_free_0:
2240         __free_page(bad_page);
2241 out:
2242         kvm_arch_exit();
2243 out_fail:
2244         return r;
2245 }
2246 EXPORT_SYMBOL_GPL(kvm_init);
2247
2248 void kvm_exit(void)
2249 {
2250         tracepoint_synchronize_unregister();
2251         kvm_exit_debug();
2252         misc_deregister(&kvm_dev);
2253         kmem_cache_destroy(kvm_vcpu_cache);
2254         sysdev_unregister(&kvm_sysdev);
2255         sysdev_class_unregister(&kvm_sysdev_class);
2256         unregister_reboot_notifier(&kvm_reboot_notifier);
2257         unregister_cpu_notifier(&kvm_cpu_notifier);
2258         on_each_cpu(hardware_disable, NULL, 1);
2259         kvm_arch_hardware_unsetup();
2260         kvm_arch_exit();
2261         free_cpumask_var(cpus_hardware_enabled);
2262         __free_page(bad_page);
2263 }
2264 EXPORT_SYMBOL_GPL(kvm_exit);