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/intel-iommu.h>
  45
  46 #include <asm/processor.h>
  47 #include <asm/io.h>
  48 #include <asm/uaccess.h>
  49 #include <asm/pgtable.h>
  50
  51 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
  52 #include "coalesced_mmio.h"
  53 #endif
  54
  55 MODULE_AUTHOR("Qumranet");
  56 MODULE_LICENSE("GPL");
  57
  58 DEFINE_SPINLOCK(kvm_lock);
  59 LIST_HEAD(vm_list);
  60
  61 static cpumask_t cpus_hardware_enabled;
  62
  63 struct kmem_cache *kvm_vcpu_cache;
  64 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  65
  66 static __read_mostly struct preempt_ops kvm_preempt_ops;
  67
  68 struct dentry *kvm_debugfs_dir;
  69
  70 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  71                            unsigned long arg);
  72
  73 bool kvm_rebooting;
  74
  75 static inline int valid_vcpu(int n)
  76 {
  77         return likely(n >= 0 && n < KVM_MAX_VCPUS);
  78 }
  79
  80 inline int is_mmio_pfn(pfn_t pfn)
  81 {
  82         if (pfn_valid(pfn))
  83                 return PageReserved(pfn_to_page(pfn));
  84
  85         return true;
  86 }
  87
  88 /*
  89  * Switches to specified vcpu, until a matching vcpu_put()
  90  */
  91 void vcpu_load(struct kvm_vcpu *vcpu)
  92 {
  93         int cpu;
  94
  95         mutex_lock(&vcpu->mutex);
  96         cpu = get_cpu();
  97         preempt_notifier_register(&vcpu->preempt_notifier);
  98         kvm_arch_vcpu_load(vcpu, cpu);
  99         put_cpu();
 100 }
 101
 102 void vcpu_put(struct kvm_vcpu *vcpu)
 103 {
 104         preempt_disable();
 105         kvm_arch_vcpu_put(vcpu);
 106         preempt_notifier_unregister(&vcpu->preempt_notifier);
 107         preempt_enable();
 108         mutex_unlock(&vcpu->mutex);
 109 }
 110
 111 static void ack_flush(void *_completed)
 112 {
 113 }
 114
 115 void kvm_flush_remote_tlbs(struct kvm *kvm)
 116 {
 117         int i, cpu, me;
 118         cpumask_t cpus;
 119         struct kvm_vcpu *vcpu;
 120
 121         me = get_cpu();
 122         cpus_clear(cpus);
 123         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 124                 vcpu = kvm->vcpus[i];
 125                 if (!vcpu)
 126                         continue;
 127                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 128                         continue;
 129                 cpu = vcpu->cpu;
 130                 if (cpu != -1 && cpu != me)
 131                         cpu_set(cpu, cpus);
 132         }
 133         if (cpus_empty(cpus))
 134                 goto out;
 135         ++kvm->stat.remote_tlb_flush;
 136         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 137 out:
 138         put_cpu();
 139 }
 140
 141 void kvm_reload_remote_mmus(struct kvm *kvm)
 142 {
 143         int i, cpu, me;
 144         cpumask_t cpus;
 145         struct kvm_vcpu *vcpu;
 146
 147         me = get_cpu();
 148         cpus_clear(cpus);
 149         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 150                 vcpu = kvm->vcpus[i];
 151                 if (!vcpu)
 152                         continue;
 153                 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
 154                         continue;
 155                 cpu = vcpu->cpu;
 156                 if (cpu != -1 && cpu != me)
 157                         cpu_set(cpu, cpus);
 158         }
 159         if (cpus_empty(cpus))
 160                 goto out;
 161         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 162 out:
 163         put_cpu();
 164 }
 165
 166
 167 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 168 {
 169         struct page *page;
 170         int r;
 171
 172         mutex_init(&vcpu->mutex);
 173         vcpu->cpu = -1;
 174         vcpu->kvm = kvm;
 175         vcpu->vcpu_id = id;
 176         init_waitqueue_head(&vcpu->wq);
 177
 178         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 179         if (!page) {
 180                 r = -ENOMEM;
 181                 goto fail;
 182         }
 183         vcpu->run = page_address(page);
 184
 185         r = kvm_arch_vcpu_init(vcpu);
 186         if (r < 0)
 187                 goto fail_free_run;
 188         return 0;
 189
 190 fail_free_run:
 191         free_page((unsigned long)vcpu->run);
 192 fail:
 193         return r;
 194 }
 195 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 196
 197 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 198 {
 199         kvm_arch_vcpu_uninit(vcpu);
 200         free_page((unsigned long)vcpu->run);
 201 }
 202 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 203
 204 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 205 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 206 {
 207         return container_of(mn, struct kvm, mmu_notifier);
 208 }
 209
 210 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 211                                              struct mm_struct *mm,
 212                                              unsigned long address)
 213 {
 214         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 215         int need_tlb_flush;
 216
 217         /*
 218          * When ->invalidate_page runs, the linux pte has been zapped
 219          * already but the page is still allocated until
 220          * ->invalidate_page returns. So if we increase the sequence
 221          * here the kvm page fault will notice if the spte can't be
 222          * established because the page is going to be freed. If
 223          * instead the kvm page fault establishes the spte before
 224          * ->invalidate_page runs, kvm_unmap_hva will release it
 225          * before returning.
 226          *
 227          * The sequence increase only need to be seen at spin_unlock
 228          * time, and not at spin_lock time.
 229          *
 230          * Increasing the sequence after the spin_unlock would be
 231          * unsafe because the kvm page fault could then establish the
 232          * pte after kvm_unmap_hva returned, without noticing the page
 233          * is going to be freed.
 234          */
 235         spin_lock(&kvm->mmu_lock);
 236         kvm->mmu_notifier_seq++;
 237         need_tlb_flush = kvm_unmap_hva(kvm, address);
 238         spin_unlock(&kvm->mmu_lock);
 239
 240         /* we've to flush the tlb before the pages can be freed */
 241         if (need_tlb_flush)
 242                 kvm_flush_remote_tlbs(kvm);
 243
 244 }
 245
 246 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 247                                                     struct mm_struct *mm,
 248                                                     unsigned long start,
 249                                                     unsigned long end)
 250 {
 251         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 252         int need_tlb_flush = 0;
 253
 254         spin_lock(&kvm->mmu_lock);
 255         /*
 256          * The count increase must become visible at unlock time as no
 257          * spte can be established without taking the mmu_lock and
 258          * count is also read inside the mmu_lock critical section.
 259          */
 260         kvm->mmu_notifier_count++;
 261         for (; start < end; start += PAGE_SIZE)
 262                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 263         spin_unlock(&kvm->mmu_lock);
 264
 265         /* we've to flush the tlb before the pages can be freed */
 266         if (need_tlb_flush)
 267                 kvm_flush_remote_tlbs(kvm);
 268 }
 269
 270 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 271                                                   struct mm_struct *mm,
 272                                                   unsigned long start,
 273                                                   unsigned long end)
 274 {
 275         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 276
 277         spin_lock(&kvm->mmu_lock);
 278         /*
 279          * This sequence increase will notify the kvm page fault that
 280          * the page that is going to be mapped in the spte could have
 281          * been freed.
 282          */
 283         kvm->mmu_notifier_seq++;
 284         /*
 285          * The above sequence increase must be visible before the
 286          * below count decrease but both values are read by the kvm
 287          * page fault under mmu_lock spinlock so we don't need to add
 288          * a smb_wmb() here in between the two.
 289          */
 290         kvm->mmu_notifier_count--;
 291         spin_unlock(&kvm->mmu_lock);
 292
 293         BUG_ON(kvm->mmu_notifier_count < 0);
 294 }
 295
 296 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 297                                               struct mm_struct *mm,
 298                                               unsigned long address)
 299 {
 300         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 301         int young;
 302
 303         spin_lock(&kvm->mmu_lock);
 304         young = kvm_age_hva(kvm, address);
 305         spin_unlock(&kvm->mmu_lock);
 306
 307         if (young)
 308                 kvm_flush_remote_tlbs(kvm);
 309
 310         return young;
 311 }
 312
 313 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 314         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 315         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 316         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 317         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 318 };
 319 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 320
 321 static struct kvm *kvm_create_vm(void)
 322 {
 323         struct kvm *kvm = kvm_arch_create_vm();
 324 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 325         struct page *page;
 326 #endif
 327
 328         if (IS_ERR(kvm))
 329                 goto out;
 330
 331 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 332         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 333         if (!page) {
 334                 kfree(kvm);
 335                 return ERR_PTR(-ENOMEM);
 336         }
 337         kvm->coalesced_mmio_ring =
 338                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 339 #endif
 340
 341 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 342         {
 343                 int err;
 344                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 345                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 346                 if (err) {
 347 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 348                         put_page(page);
 349 #endif
 350                         kfree(kvm);
 351                         return ERR_PTR(err);
 352                 }
 353         }
 354 #endif
 355
 356         kvm->mm = current->mm;
 357         atomic_inc(&kvm->mm->mm_count);
 358         spin_lock_init(&kvm->mmu_lock);
 359         kvm_io_bus_init(&kvm->pio_bus);
 360         mutex_init(&kvm->lock);
 361         kvm_io_bus_init(&kvm->mmio_bus);
 362         init_rwsem(&kvm->slots_lock);
 363         atomic_set(&kvm->users_count, 1);
 364         spin_lock(&kvm_lock);
 365         list_add(&kvm->vm_list, &vm_list);
 366         spin_unlock(&kvm_lock);
 367 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 368         kvm_coalesced_mmio_init(kvm);
 369 #endif
 370 out:
 371         return kvm;
 372 }
 373
 374 /*
 375  * Free any memory in @free but not in @dont.
 376  */
 377 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 378                                   struct kvm_memory_slot *dont)
 379 {
 380         if (!dont || free->rmap != dont->rmap)
 381                 vfree(free->rmap);
 382
 383         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 384                 vfree(free->dirty_bitmap);
 385
 386         if (!dont || free->lpage_info != dont->lpage_info)
 387                 vfree(free->lpage_info);
 388
 389         free->npages = 0;
 390         free->dirty_bitmap = NULL;
 391         free->rmap = NULL;
 392         free->lpage_info = NULL;
 393 }
 394
 395 void kvm_free_physmem(struct kvm *kvm)
 396 {
 397         int i;
 398
 399         for (i = 0; i < kvm->nmemslots; ++i)
 400                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 401 }
 402
 403 static void kvm_destroy_vm(struct kvm *kvm)
 404 {
 405         struct mm_struct *mm = kvm->mm;
 406
 407         spin_lock(&kvm_lock);
 408         list_del(&kvm->vm_list);
 409         spin_unlock(&kvm_lock);
 410         kvm_io_bus_destroy(&kvm->pio_bus);
 411         kvm_io_bus_destroy(&kvm->mmio_bus);
 412 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 413         if (kvm->coalesced_mmio_ring != NULL)
 414                 free_page((unsigned long)kvm->coalesced_mmio_ring);
 415 #endif
 416 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 417         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 418 #endif
 419         kvm_arch_destroy_vm(kvm);
 420         mmdrop(mm);
 421 }
 422
 423 void kvm_get_kvm(struct kvm *kvm)
 424 {
 425         atomic_inc(&kvm->users_count);
 426 }
 427 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 428
 429 void kvm_put_kvm(struct kvm *kvm)
 430 {
 431         if (atomic_dec_and_test(&kvm->users_count))
 432                 kvm_destroy_vm(kvm);
 433 }
 434 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 435
 436
 437 static int kvm_vm_release(struct inode *inode, struct file *filp)
 438 {
 439         struct kvm *kvm = filp->private_data;
 440
 441         kvm_put_kvm(kvm);
 442         return 0;
 443 }
 444
 445 /*
 446  * Allocate some memory and give it an address in the guest physical address
 447  * space.
 448  *
 449  * Discontiguous memory is allowed, mostly for framebuffers.
 450  *
 451  * Must be called holding mmap_sem for write.
 452  */
 453 int __kvm_set_memory_region(struct kvm *kvm,
 454                             struct kvm_userspace_memory_region *mem,
 455                             int user_alloc)
 456 {
 457         int r;
 458         gfn_t base_gfn;
 459         unsigned long npages;
 460         unsigned long i;
 461         struct kvm_memory_slot *memslot;
 462         struct kvm_memory_slot old, new;
 463
 464         r = -EINVAL;
 465         /* General sanity checks */
 466         if (mem->memory_size & (PAGE_SIZE - 1))
 467                 goto out;
 468         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 469                 goto out;
 470         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 471                 goto out;
 472         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 473                 goto out;
 474
 475         memslot = &kvm->memslots[mem->slot];
 476         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 477         npages = mem->memory_size >> PAGE_SHIFT;
 478
 479         if (!npages)
 480                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 481
 482         new = old = *memslot;
 483
 484         new.base_gfn = base_gfn;
 485         new.npages = npages;
 486         new.flags = mem->flags;
 487
 488         /* Disallow changing a memory slot's size. */
 489         r = -EINVAL;
 490         if (npages && old.npages && npages != old.npages)
 491                 goto out_free;
 492
 493         /* Check for overlaps */
 494         r = -EEXIST;
 495         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 496                 struct kvm_memory_slot *s = &kvm->memslots[i];
 497
 498                 if (s == memslot)
 499                         continue;
 500                 if (!((base_gfn + npages <= s->base_gfn) ||
 501                       (base_gfn >= s->base_gfn + s->npages)))
 502                         goto out_free;
 503         }
 504
 505         /* Free page dirty bitmap if unneeded */
 506         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 507                 new.dirty_bitmap = NULL;
 508
 509         r = -ENOMEM;
 510
 511         /* Allocate if a slot is being created */
 512 #ifndef CONFIG_S390
 513         if (npages && !new.rmap) {
 514                 new.rmap = vmalloc(npages * sizeof(struct page *));
 515
 516                 if (!new.rmap)
 517                         goto out_free;
 518
 519                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 520
 521                 new.user_alloc = user_alloc;
 522                 /*
 523                  * hva_to_rmmap() serialzies with the mmu_lock and to be
 524                  * safe it has to ignore memslots with !user_alloc &&
 525                  * !userspace_addr.
 526                  */
 527                 if (user_alloc)
 528                         new.userspace_addr = mem->userspace_addr;
 529                 else
 530                         new.userspace_addr = 0;
 531         }
 532         if (npages && !new.lpage_info) {
 533                 int largepages = npages / KVM_PAGES_PER_HPAGE;
 534                 if (npages % KVM_PAGES_PER_HPAGE)
 535                         largepages++;
 536                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 537                         largepages++;
 538
 539                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
 540
 541                 if (!new.lpage_info)
 542                         goto out_free;
 543
 544                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
 545
 546                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 547                         new.lpage_info[0].write_count = 1;
 548                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
 549                         new.lpage_info[largepages-1].write_count = 1;
 550         }
 551
 552         /* Allocate page dirty bitmap if needed */
 553         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 554                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 555
 556                 new.dirty_bitmap = vmalloc(dirty_bytes);
 557                 if (!new.dirty_bitmap)
 558                         goto out_free;
 559                 memset(new.dirty_bitmap, 0, dirty_bytes);
 560         }
 561 #endif /* not defined CONFIG_S390 */
 562
 563         if (!npages)
 564                 kvm_arch_flush_shadow(kvm);
 565
 566         spin_lock(&kvm->mmu_lock);
 567         if (mem->slot >= kvm->nmemslots)
 568                 kvm->nmemslots = mem->slot + 1;
 569
 570         *memslot = new;
 571         spin_unlock(&kvm->mmu_lock);
 572
 573         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
 574         if (r) {
 575                 spin_lock(&kvm->mmu_lock);
 576                 *memslot = old;
 577                 spin_unlock(&kvm->mmu_lock);
 578                 goto out_free;
 579         }
 580
 581         kvm_free_physmem_slot(&old, &new);
 582
 583         /* map the pages in iommu page table */
 584         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
 585         if (r)
 586                 goto out;
 587
 588         return 0;
 589
 590 out_free:
 591         kvm_free_physmem_slot(&new, &old);
 592 out:
 593         return r;
 594
 595 }
 596 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 597
 598 int kvm_set_memory_region(struct kvm *kvm,
 599                           struct kvm_userspace_memory_region *mem,
 600                           int user_alloc)
 601 {
 602         int r;
 603
 604         down_write(&kvm->slots_lock);
 605         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 606         up_write(&kvm->slots_lock);
 607         return r;
 608 }
 609 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 610
 611 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 612                                    struct
 613                                    kvm_userspace_memory_region *mem,
 614                                    int user_alloc)
 615 {
 616         if (mem->slot >= KVM_MEMORY_SLOTS)
 617                 return -EINVAL;
 618         return kvm_set_memory_region(kvm, mem, user_alloc);
 619 }
 620
 621 int kvm_get_dirty_log(struct kvm *kvm,
 622                         struct kvm_dirty_log *log, int *is_dirty)
 623 {
 624         struct kvm_memory_slot *memslot;
 625         int r, i;
 626         int n;
 627         unsigned long any = 0;
 628
 629         r = -EINVAL;
 630         if (log->slot >= KVM_MEMORY_SLOTS)
 631                 goto out;
 632
 633         memslot = &kvm->memslots[log->slot];
 634         r = -ENOENT;
 635         if (!memslot->dirty_bitmap)
 636                 goto out;
 637
 638         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 639
 640         for (i = 0; !any && i < n/sizeof(long); ++i)
 641                 any = memslot->dirty_bitmap[i];
 642
 643         r = -EFAULT;
 644         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 645                 goto out;
 646
 647         if (any)
 648                 *is_dirty = 1;
 649
 650         r = 0;
 651 out:
 652         return r;
 653 }
 654
 655 int is_error_page(struct page *page)
 656 {
 657         return page == bad_page;
 658 }
 659 EXPORT_SYMBOL_GPL(is_error_page);
 660
 661 int is_error_pfn(pfn_t pfn)
 662 {
 663         return pfn == bad_pfn;
 664 }
 665 EXPORT_SYMBOL_GPL(is_error_pfn);
 666
 667 static inline unsigned long bad_hva(void)
 668 {
 669         return PAGE_OFFSET;
 670 }
 671
 672 int kvm_is_error_hva(unsigned long addr)
 673 {
 674         return addr == bad_hva();
 675 }
 676 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 677
 678 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 679 {
 680         int i;
 681
 682         for (i = 0; i < kvm->nmemslots; ++i) {
 683                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 684
 685                 if (gfn >= memslot->base_gfn
 686                     && gfn < memslot->base_gfn + memslot->npages)
 687                         return memslot;
 688         }
 689         return NULL;
 690 }
 691
 692 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 693 {
 694         gfn = unalias_gfn(kvm, gfn);
 695         return __gfn_to_memslot(kvm, gfn);
 696 }
 697
 698 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 699 {
 700         int i;
 701
 702         gfn = unalias_gfn(kvm, gfn);
 703         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 704                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 705
 706                 if (gfn >= memslot->base_gfn
 707                     && gfn < memslot->base_gfn + memslot->npages)
 708                         return 1;
 709         }
 710         return 0;
 711 }
 712 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 713
 714 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 715 {
 716         struct kvm_memory_slot *slot;
 717
 718         gfn = unalias_gfn(kvm, gfn);
 719         slot = __gfn_to_memslot(kvm, gfn);
 720         if (!slot)
 721                 return bad_hva();
 722         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 723 }
 724 EXPORT_SYMBOL_GPL(gfn_to_hva);
 725
 726 /*
 727  * Requires current->mm->mmap_sem to be held
 728  */
 729 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 730 {
 731         struct page *page[1];
 732         unsigned long addr;
 733         int npages;
 734         pfn_t pfn;
 735
 736         might_sleep();
 737
 738         addr = gfn_to_hva(kvm, gfn);
 739         if (kvm_is_error_hva(addr)) {
 740                 get_page(bad_page);
 741                 return page_to_pfn(bad_page);
 742         }
 743
 744         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
 745                                 NULL);
 746
 747         if (unlikely(npages != 1)) {
 748                 struct vm_area_struct *vma;
 749
 750                 vma = find_vma(current->mm, addr);
 751                 if (vma == NULL || addr < vma->vm_start ||
 752                     !(vma->vm_flags & VM_PFNMAP)) {
 753                         get_page(bad_page);
 754                         return page_to_pfn(bad_page);
 755                 }
 756
 757                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 758                 BUG_ON(!is_mmio_pfn(pfn));
 759         } else
 760                 pfn = page_to_pfn(page[0]);
 761
 762         return pfn;
 763 }
 764
 765 EXPORT_SYMBOL_GPL(gfn_to_pfn);
 766
 767 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 768 {
 769         pfn_t pfn;
 770
 771         pfn = gfn_to_pfn(kvm, gfn);
 772         if (!is_mmio_pfn(pfn))
 773                 return pfn_to_page(pfn);
 774
 775         WARN_ON(is_mmio_pfn(pfn));
 776
 777         get_page(bad_page);
 778         return bad_page;
 779 }
 780
 781 EXPORT_SYMBOL_GPL(gfn_to_page);
 782
 783 void kvm_release_page_clean(struct page *page)
 784 {
 785         kvm_release_pfn_clean(page_to_pfn(page));
 786 }
 787 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
 788
 789 void kvm_release_pfn_clean(pfn_t pfn)
 790 {
 791         if (!is_mmio_pfn(pfn))
 792                 put_page(pfn_to_page(pfn));
 793 }
 794 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
 795
 796 void kvm_release_page_dirty(struct page *page)
 797 {
 798         kvm_release_pfn_dirty(page_to_pfn(page));
 799 }
 800 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
 801
 802 void kvm_release_pfn_dirty(pfn_t pfn)
 803 {
 804         kvm_set_pfn_dirty(pfn);
 805         kvm_release_pfn_clean(pfn);
 806 }
 807 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
 808
 809 void kvm_set_page_dirty(struct page *page)
 810 {
 811         kvm_set_pfn_dirty(page_to_pfn(page));
 812 }
 813 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
 814
 815 void kvm_set_pfn_dirty(pfn_t pfn)
 816 {
 817         if (!is_mmio_pfn(pfn)) {
 818                 struct page *page = pfn_to_page(pfn);
 819                 if (!PageReserved(page))
 820                         SetPageDirty(page);
 821         }
 822 }
 823 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
 824
 825 void kvm_set_pfn_accessed(pfn_t pfn)
 826 {
 827         if (!is_mmio_pfn(pfn))
 828                 mark_page_accessed(pfn_to_page(pfn));
 829 }
 830 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
 831
 832 void kvm_get_pfn(pfn_t pfn)
 833 {
 834         if (!is_mmio_pfn(pfn))
 835                 get_page(pfn_to_page(pfn));
 836 }
 837 EXPORT_SYMBOL_GPL(kvm_get_pfn);
 838
 839 static int next_segment(unsigned long len, int offset)
 840 {
 841         if (len > PAGE_SIZE - offset)
 842                 return PAGE_SIZE - offset;
 843         else
 844                 return len;
 845 }
 846
 847 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 848                         int len)
 849 {
 850         int r;
 851         unsigned long addr;
 852
 853         addr = gfn_to_hva(kvm, gfn);
 854         if (kvm_is_error_hva(addr))
 855                 return -EFAULT;
 856         r = copy_from_user(data, (void __user *)addr + offset, len);
 857         if (r)
 858                 return -EFAULT;
 859         return 0;
 860 }
 861 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 862
 863 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 864 {
 865         gfn_t gfn = gpa >> PAGE_SHIFT;
 866         int seg;
 867         int offset = offset_in_page(gpa);
 868         int ret;
 869
 870         while ((seg = next_segment(len, offset)) != 0) {
 871                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 872                 if (ret < 0)
 873                         return ret;
 874                 offset = 0;
 875                 len -= seg;
 876                 data += seg;
 877                 ++gfn;
 878         }
 879         return 0;
 880 }
 881 EXPORT_SYMBOL_GPL(kvm_read_guest);
 882
 883 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
 884                           unsigned long len)
 885 {
 886         int r;
 887         unsigned long addr;
 888         gfn_t gfn = gpa >> PAGE_SHIFT;
 889         int offset = offset_in_page(gpa);
 890
 891         addr = gfn_to_hva(kvm, gfn);
 892         if (kvm_is_error_hva(addr))
 893                 return -EFAULT;
 894         pagefault_disable();
 895         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
 896         pagefault_enable();
 897         if (r)
 898                 return -EFAULT;
 899         return 0;
 900 }
 901 EXPORT_SYMBOL(kvm_read_guest_atomic);
 902
 903 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
 904                          int offset, int len)
 905 {
 906         int r;
 907         unsigned long addr;
 908
 909         addr = gfn_to_hva(kvm, gfn);
 910         if (kvm_is_error_hva(addr))
 911                 return -EFAULT;
 912         r = copy_to_user((void __user *)addr + offset, data, len);
 913         if (r)
 914                 return -EFAULT;
 915         mark_page_dirty(kvm, gfn);
 916         return 0;
 917 }
 918 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
 919
 920 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
 921                     unsigned long len)
 922 {
 923         gfn_t gfn = gpa >> PAGE_SHIFT;
 924         int seg;
 925         int offset = offset_in_page(gpa);
 926         int ret;
 927
 928         while ((seg = next_segment(len, offset)) != 0) {
 929                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
 930                 if (ret < 0)
 931                         return ret;
 932                 offset = 0;
 933                 len -= seg;
 934                 data += seg;
 935                 ++gfn;
 936         }
 937         return 0;
 938 }
 939
 940 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
 941 {
 942         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
 943 }
 944 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
 945
 946 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
 947 {
 948         gfn_t gfn = gpa >> PAGE_SHIFT;
 949         int seg;
 950         int offset = offset_in_page(gpa);
 951         int ret;
 952
 953         while ((seg = next_segment(len, offset)) != 0) {
 954                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
 955                 if (ret < 0)
 956                         return ret;
 957                 offset = 0;
 958                 len -= seg;
 959                 ++gfn;
 960         }
 961         return 0;
 962 }
 963 EXPORT_SYMBOL_GPL(kvm_clear_guest);
 964
 965 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 966 {
 967         struct kvm_memory_slot *memslot;
 968
 969         gfn = unalias_gfn(kvm, gfn);
 970         memslot = __gfn_to_memslot(kvm, gfn);
 971         if (memslot && memslot->dirty_bitmap) {
 972                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 973
 974                 /* avoid RMW */
 975                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 976                         set_bit(rel_gfn, memslot->dirty_bitmap);
 977         }
 978 }
 979
 980 /*
 981  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
 982  */
 983 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 984 {
 985         DEFINE_WAIT(wait);
 986
 987         for (;;) {
 988                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
 989
 990                 if (kvm_cpu_has_interrupt(vcpu) ||
 991                     kvm_cpu_has_pending_timer(vcpu) ||
 992                     kvm_arch_vcpu_runnable(vcpu)) {
 993                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
 994                         break;
 995                 }
 996                 if (signal_pending(current))
 997                         break;
 998
 999                 vcpu_put(vcpu);
1000                 schedule();
1001                 vcpu_load(vcpu);
1002         }
1003
1004         finish_wait(&vcpu->wq, &wait);
1005 }
1006
1007 void kvm_resched(struct kvm_vcpu *vcpu)
1008 {
1009         if (!need_resched())
1010                 return;
1011         cond_resched();
1012 }
1013 EXPORT_SYMBOL_GPL(kvm_resched);
1014
1015 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1016 {
1017         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1018         struct page *page;
1019
1020         if (vmf->pgoff == 0)
1021                 page = virt_to_page(vcpu->run);
1022 #ifdef CONFIG_X86
1023         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1024                 page = virt_to_page(vcpu->arch.pio_data);
1025 #endif
1026 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1027         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1028                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1029 #endif
1030         else
1031                 return VM_FAULT_SIGBUS;
1032         get_page(page);
1033         vmf->page = page;
1034         return 0;
1035 }
1036
1037 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1038         .fault = kvm_vcpu_fault,
1039 };
1040
1041 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1042 {
1043         vma->vm_ops = &kvm_vcpu_vm_ops;
1044         return 0;
1045 }
1046
1047 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1048 {
1049         struct kvm_vcpu *vcpu = filp->private_data;
1050
1051         kvm_put_kvm(vcpu->kvm);
1052         return 0;
1053 }
1054
1055 static const struct file_operations kvm_vcpu_fops = {
1056         .release        = kvm_vcpu_release,
1057         .unlocked_ioctl = kvm_vcpu_ioctl,
1058         .compat_ioctl   = kvm_vcpu_ioctl,
1059         .mmap           = kvm_vcpu_mmap,
1060 };
1061
1062 /*
1063  * Allocates an inode for the vcpu.
1064  */
1065 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1066 {
1067         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1068         if (fd < 0)
1069                 kvm_put_kvm(vcpu->kvm);
1070         return fd;
1071 }
1072
1073 /*
1074  * Creates some virtual cpus.  Good luck creating more than one.
1075  */
1076 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1077 {
1078         int r;
1079         struct kvm_vcpu *vcpu;
1080
1081         if (!valid_vcpu(n))
1082                 return -EINVAL;
1083
1084         vcpu = kvm_arch_vcpu_create(kvm, n);
1085         if (IS_ERR(vcpu))
1086                 return PTR_ERR(vcpu);
1087
1088         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1089
1090         r = kvm_arch_vcpu_setup(vcpu);
1091         if (r)
1092                 goto vcpu_destroy;
1093
1094         mutex_lock(&kvm->lock);
1095         if (kvm->vcpus[n]) {
1096                 r = -EEXIST;
1097                 mutex_unlock(&kvm->lock);
1098                 goto vcpu_destroy;
1099         }
1100         kvm->vcpus[n] = vcpu;
1101         mutex_unlock(&kvm->lock);
1102
1103         /* Now it's all set up, let userspace reach it */
1104         kvm_get_kvm(kvm);
1105         r = create_vcpu_fd(vcpu);
1106         if (r < 0)
1107                 goto unlink;
1108         return r;
1109
1110 unlink:
1111         mutex_lock(&kvm->lock);
1112         kvm->vcpus[n] = NULL;
1113         mutex_unlock(&kvm->lock);
1114 vcpu_destroy:
1115         kvm_arch_vcpu_destroy(vcpu);
1116         return r;
1117 }
1118
1119 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1120 {
1121         if (sigset) {
1122                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1123                 vcpu->sigset_active = 1;
1124                 vcpu->sigset = *sigset;
1125         } else
1126                 vcpu->sigset_active = 0;
1127         return 0;
1128 }
1129
1130 static long kvm_vcpu_ioctl(struct file *filp,
1131                            unsigned int ioctl, unsigned long arg)
1132 {
1133         struct kvm_vcpu *vcpu = filp->private_data;
1134         void __user *argp = (void __user *)arg;
1135         int r;
1136         struct kvm_fpu *fpu = NULL;
1137         struct kvm_sregs *kvm_sregs = NULL;
1138
1139         if (vcpu->kvm->mm != current->mm)
1140                 return -EIO;
1141         switch (ioctl) {
1142         case KVM_RUN:
1143                 r = -EINVAL;
1144                 if (arg)
1145                         goto out;
1146                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1147                 break;
1148         case KVM_GET_REGS: {
1149                 struct kvm_regs *kvm_regs;
1150
1151                 r = -ENOMEM;
1152                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1153                 if (!kvm_regs)
1154                         goto out;
1155                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1156                 if (r)
1157                         goto out_free1;
1158                 r = -EFAULT;
1159                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1160                         goto out_free1;
1161                 r = 0;
1162 out_free1:
1163                 kfree(kvm_regs);
1164                 break;
1165         }
1166         case KVM_SET_REGS: {
1167                 struct kvm_regs *kvm_regs;
1168
1169                 r = -ENOMEM;
1170                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1171                 if (!kvm_regs)
1172                         goto out;
1173                 r = -EFAULT;
1174                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1175                         goto out_free2;
1176                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1177                 if (r)
1178                         goto out_free2;
1179                 r = 0;
1180 out_free2:
1181                 kfree(kvm_regs);
1182                 break;
1183         }
1184         case KVM_GET_SREGS: {
1185                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1186                 r = -ENOMEM;
1187                 if (!kvm_sregs)
1188                         goto out;
1189                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1190                 if (r)
1191                         goto out;
1192                 r = -EFAULT;
1193                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1194                         goto out;
1195                 r = 0;
1196                 break;
1197         }
1198         case KVM_SET_SREGS: {
1199                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1200                 r = -ENOMEM;
1201                 if (!kvm_sregs)
1202                         goto out;
1203                 r = -EFAULT;
1204                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1205                         goto out;
1206                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1207                 if (r)
1208                         goto out;
1209                 r = 0;
1210                 break;
1211         }
1212         case KVM_GET_MP_STATE: {
1213                 struct kvm_mp_state mp_state;
1214
1215                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1216                 if (r)
1217                         goto out;
1218                 r = -EFAULT;
1219                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1220                         goto out;
1221                 r = 0;
1222                 break;
1223         }
1224         case KVM_SET_MP_STATE: {
1225                 struct kvm_mp_state mp_state;
1226
1227                 r = -EFAULT;
1228                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1229                         goto out;
1230                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1231                 if (r)
1232                         goto out;
1233                 r = 0;
1234                 break;
1235         }
1236         case KVM_TRANSLATE: {
1237                 struct kvm_translation tr;
1238
1239                 r = -EFAULT;
1240                 if (copy_from_user(&tr, argp, sizeof tr))
1241                         goto out;
1242                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1243                 if (r)
1244                         goto out;
1245                 r = -EFAULT;
1246                 if (copy_to_user(argp, &tr, sizeof tr))
1247                         goto out;
1248                 r = 0;
1249                 break;
1250         }
1251         case KVM_DEBUG_GUEST: {
1252                 struct kvm_debug_guest dbg;
1253
1254                 r = -EFAULT;
1255                 if (copy_from_user(&dbg, argp, sizeof dbg))
1256                         goto out;
1257                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1258                 if (r)
1259                         goto out;
1260                 r = 0;
1261                 break;
1262         }
1263         case KVM_SET_SIGNAL_MASK: {
1264                 struct kvm_signal_mask __user *sigmask_arg = argp;
1265                 struct kvm_signal_mask kvm_sigmask;
1266                 sigset_t sigset, *p;
1267
1268                 p = NULL;
1269                 if (argp) {
1270                         r = -EFAULT;
1271                         if (copy_from_user(&kvm_sigmask, argp,
1272                                            sizeof kvm_sigmask))
1273                                 goto out;
1274                         r = -EINVAL;
1275                         if (kvm_sigmask.len != sizeof sigset)
1276                                 goto out;
1277                         r = -EFAULT;
1278                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1279                                            sizeof sigset))
1280                                 goto out;
1281                         p = &sigset;
1282                 }
1283                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1284                 break;
1285         }
1286         case KVM_GET_FPU: {
1287                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1288                 r = -ENOMEM;
1289                 if (!fpu)
1290                         goto out;
1291                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1292                 if (r)
1293                         goto out;
1294                 r = -EFAULT;
1295                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1296                         goto out;
1297                 r = 0;
1298                 break;
1299         }
1300         case KVM_SET_FPU: {
1301                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1302                 r = -ENOMEM;
1303                 if (!fpu)
1304                         goto out;
1305                 r = -EFAULT;
1306                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1307                         goto out;
1308                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1309                 if (r)
1310                         goto out;
1311                 r = 0;
1312                 break;
1313         }
1314         default:
1315                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1316         }
1317 out:
1318         kfree(fpu);
1319         kfree(kvm_sregs);
1320         return r;
1321 }
1322
1323 static long kvm_vm_ioctl(struct file *filp,
1324                            unsigned int ioctl, unsigned long arg)
1325 {
1326         struct kvm *kvm = filp->private_data;
1327         void __user *argp = (void __user *)arg;
1328         int r;
1329
1330         if (kvm->mm != current->mm)
1331                 return -EIO;
1332         switch (ioctl) {
1333         case KVM_CREATE_VCPU:
1334                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1335                 if (r < 0)
1336                         goto out;
1337                 break;
1338         case KVM_SET_USER_MEMORY_REGION: {
1339                 struct kvm_userspace_memory_region kvm_userspace_mem;
1340
1341                 r = -EFAULT;
1342                 if (copy_from_user(&kvm_userspace_mem, argp,
1343                                                 sizeof kvm_userspace_mem))
1344                         goto out;
1345
1346                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1347                 if (r)
1348                         goto out;
1349                 break;
1350         }
1351         case KVM_GET_DIRTY_LOG: {
1352                 struct kvm_dirty_log log;
1353
1354                 r = -EFAULT;
1355                 if (copy_from_user(&log, argp, sizeof log))
1356                         goto out;
1357                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1358                 if (r)
1359                         goto out;
1360                 break;
1361         }
1362 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1363         case KVM_REGISTER_COALESCED_MMIO: {
1364                 struct kvm_coalesced_mmio_zone zone;
1365                 r = -EFAULT;
1366                 if (copy_from_user(&zone, argp, sizeof zone))
1367                         goto out;
1368                 r = -ENXIO;
1369                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1370                 if (r)
1371                         goto out;
1372                 r = 0;
1373                 break;
1374         }
1375         case KVM_UNREGISTER_COALESCED_MMIO: {
1376                 struct kvm_coalesced_mmio_zone zone;
1377                 r = -EFAULT;
1378                 if (copy_from_user(&zone, argp, sizeof zone))
1379                         goto out;
1380                 r = -ENXIO;
1381                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1382                 if (r)
1383                         goto out;
1384                 r = 0;
1385                 break;
1386         }
1387 #endif
1388         default:
1389                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1390         }
1391 out:
1392         return r;
1393 }
1394
1395 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1396 {
1397         struct kvm *kvm = vma->vm_file->private_data;
1398         struct page *page;
1399
1400         if (!kvm_is_visible_gfn(kvm, vmf->pgoff))
1401                 return VM_FAULT_SIGBUS;
1402         page = gfn_to_page(kvm, vmf->pgoff);
1403         if (is_error_page(page)) {
1404                 kvm_release_page_clean(page);
1405                 return VM_FAULT_SIGBUS;
1406         }
1407         vmf->page = page;
1408         return 0;
1409 }
1410
1411 static struct vm_operations_struct kvm_vm_vm_ops = {
1412         .fault = kvm_vm_fault,
1413 };
1414
1415 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1416 {
1417         vma->vm_ops = &kvm_vm_vm_ops;
1418         return 0;
1419 }
1420
1421 static const struct file_operations kvm_vm_fops = {
1422         .release        = kvm_vm_release,
1423         .unlocked_ioctl = kvm_vm_ioctl,
1424         .compat_ioctl   = kvm_vm_ioctl,
1425         .mmap           = kvm_vm_mmap,
1426 };
1427
1428 static int kvm_dev_ioctl_create_vm(void)
1429 {
1430         int fd;
1431         struct kvm *kvm;
1432
1433         kvm = kvm_create_vm();
1434         if (IS_ERR(kvm))
1435                 return PTR_ERR(kvm);
1436         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1437         if (fd < 0)
1438                 kvm_put_kvm(kvm);
1439
1440         return fd;
1441 }
1442
1443 static long kvm_dev_ioctl(struct file *filp,
1444                           unsigned int ioctl, unsigned long arg)
1445 {
1446         long r = -EINVAL;
1447
1448         switch (ioctl) {
1449         case KVM_GET_API_VERSION:
1450                 r = -EINVAL;
1451                 if (arg)
1452                         goto out;
1453                 r = KVM_API_VERSION;
1454                 break;
1455         case KVM_CREATE_VM:
1456                 r = -EINVAL;
1457                 if (arg)
1458                         goto out;
1459                 r = kvm_dev_ioctl_create_vm();
1460                 break;
1461         case KVM_CHECK_EXTENSION:
1462                 r = kvm_dev_ioctl_check_extension(arg);
1463                 break;
1464         case KVM_GET_VCPU_MMAP_SIZE:
1465                 r = -EINVAL;
1466                 if (arg)
1467                         goto out;
1468                 r = PAGE_SIZE;     /* struct kvm_run */
1469 #ifdef CONFIG_X86
1470                 r += PAGE_SIZE;    /* pio data page */
1471 #endif
1472 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1473                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1474 #endif
1475                 break;
1476         case KVM_TRACE_ENABLE:
1477         case KVM_TRACE_PAUSE:
1478         case KVM_TRACE_DISABLE:
1479                 r = kvm_trace_ioctl(ioctl, arg);
1480                 break;
1481         default:
1482                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1483         }
1484 out:
1485         return r;
1486 }
1487
1488 static struct file_operations kvm_chardev_ops = {
1489         .unlocked_ioctl = kvm_dev_ioctl,
1490         .compat_ioctl   = kvm_dev_ioctl,
1491 };
1492
1493 static struct miscdevice kvm_dev = {
1494         KVM_MINOR,
1495         "kvm",
1496         &kvm_chardev_ops,
1497 };
1498
1499 static void hardware_enable(void *junk)
1500 {
1501         int cpu = raw_smp_processor_id();
1502
1503         if (cpu_isset(cpu, cpus_hardware_enabled))
1504                 return;
1505         cpu_set(cpu, cpus_hardware_enabled);
1506         kvm_arch_hardware_enable(NULL);
1507 }
1508
1509 static void hardware_disable(void *junk)
1510 {
1511         int cpu = raw_smp_processor_id();
1512
1513         if (!cpu_isset(cpu, cpus_hardware_enabled))
1514                 return;
1515         cpu_clear(cpu, cpus_hardware_enabled);
1516         kvm_arch_hardware_disable(NULL);
1517 }
1518
1519 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1520                            void *v)
1521 {
1522         int cpu = (long)v;
1523
1524         val &= ~CPU_TASKS_FROZEN;
1525         switch (val) {
1526         case CPU_DYING:
1527                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1528                        cpu);
1529                 hardware_disable(NULL);
1530                 break;
1531         case CPU_UP_CANCELED:
1532                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1533                        cpu);
1534                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1535                 break;
1536         case CPU_ONLINE:
1537                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1538                        cpu);
1539                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1540                 break;
1541         }
1542         return NOTIFY_OK;
1543 }
1544
1545
1546 asmlinkage void kvm_handle_fault_on_reboot(void)
1547 {
1548         if (kvm_rebooting)
1549                 /* spin while reset goes on */
1550                 while (true)
1551                         ;
1552         /* Fault while not rebooting.  We want the trace. */
1553         BUG();
1554 }
1555 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1556
1557 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1558                       void *v)
1559 {
1560         if (val == SYS_RESTART) {
1561                 /*
1562                  * Some (well, at least mine) BIOSes hang on reboot if
1563                  * in vmx root mode.
1564                  */
1565                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1566                 kvm_rebooting = true;
1567                 on_each_cpu(hardware_disable, NULL, 1);
1568         }
1569         return NOTIFY_OK;
1570 }
1571
1572 static struct notifier_block kvm_reboot_notifier = {
1573         .notifier_call = kvm_reboot,
1574         .priority = 0,
1575 };
1576
1577 void kvm_io_bus_init(struct kvm_io_bus *bus)
1578 {
1579         memset(bus, 0, sizeof(*bus));
1580 }
1581
1582 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1583 {
1584         int i;
1585
1586         for (i = 0; i < bus->dev_count; i++) {
1587                 struct kvm_io_device *pos = bus->devs[i];
1588
1589                 kvm_iodevice_destructor(pos);
1590         }
1591 }
1592
1593 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
1594                                           gpa_t addr, int len, int is_write)
1595 {
1596         int i;
1597
1598         for (i = 0; i < bus->dev_count; i++) {
1599                 struct kvm_io_device *pos = bus->devs[i];
1600
1601                 if (pos->in_range(pos, addr, len, is_write))
1602                         return pos;
1603         }
1604
1605         return NULL;
1606 }
1607
1608 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1609 {
1610         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1611
1612         bus->devs[bus->dev_count++] = dev;
1613 }
1614
1615 static struct notifier_block kvm_cpu_notifier = {
1616         .notifier_call = kvm_cpu_hotplug,
1617         .priority = 20, /* must be > scheduler priority */
1618 };
1619
1620 static int vm_stat_get(void *_offset, u64 *val)
1621 {
1622         unsigned offset = (long)_offset;
1623         struct kvm *kvm;
1624
1625         *val = 0;
1626         spin_lock(&kvm_lock);
1627         list_for_each_entry(kvm, &vm_list, vm_list)
1628                 *val += *(u32 *)((void *)kvm + offset);
1629         spin_unlock(&kvm_lock);
1630         return 0;
1631 }
1632
1633 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1634
1635 static int vcpu_stat_get(void *_offset, u64 *val)
1636 {
1637         unsigned offset = (long)_offset;
1638         struct kvm *kvm;
1639         struct kvm_vcpu *vcpu;
1640         int i;
1641
1642         *val = 0;
1643         spin_lock(&kvm_lock);
1644         list_for_each_entry(kvm, &vm_list, vm_list)
1645                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1646                         vcpu = kvm->vcpus[i];
1647                         if (vcpu)
1648                                 *val += *(u32 *)((void *)vcpu + offset);
1649                 }
1650         spin_unlock(&kvm_lock);
1651         return 0;
1652 }
1653
1654 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1655
1656 static struct file_operations *stat_fops[] = {
1657         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1658         [KVM_STAT_VM]   = &vm_stat_fops,
1659 };
1660
1661 static void kvm_init_debug(void)
1662 {
1663         struct kvm_stats_debugfs_item *p;
1664
1665         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1666         for (p = debugfs_entries; p->name; ++p)
1667                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1668                                                 (void *)(long)p->offset,
1669                                                 stat_fops[p->kind]);
1670 }
1671
1672 static void kvm_exit_debug(void)
1673 {
1674         struct kvm_stats_debugfs_item *p;
1675
1676         for (p = debugfs_entries; p->name; ++p)
1677                 debugfs_remove(p->dentry);
1678         debugfs_remove(kvm_debugfs_dir);
1679 }
1680
1681 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1682 {
1683         hardware_disable(NULL);
1684         return 0;
1685 }
1686
1687 static int kvm_resume(struct sys_device *dev)
1688 {
1689         hardware_enable(NULL);
1690         return 0;
1691 }
1692
1693 static struct sysdev_class kvm_sysdev_class = {
1694         .name = "kvm",
1695         .suspend = kvm_suspend,
1696         .resume = kvm_resume,
1697 };
1698
1699 static struct sys_device kvm_sysdev = {
1700         .id = 0,
1701         .cls = &kvm_sysdev_class,
1702 };
1703
1704 struct page *bad_page;
1705 pfn_t bad_pfn;
1706
1707 static inline
1708 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1709 {
1710         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1711 }
1712
1713 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1714 {
1715         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1716
1717         kvm_arch_vcpu_load(vcpu, cpu);
1718 }
1719
1720 static void kvm_sched_out(struct preempt_notifier *pn,
1721                           struct task_struct *next)
1722 {
1723         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1724
1725         kvm_arch_vcpu_put(vcpu);
1726 }
1727
1728 int kvm_init(void *opaque, unsigned int vcpu_size,
1729                   struct module *module)
1730 {
1731         int r;
1732         int cpu;
1733
1734         kvm_init_debug();
1735
1736         r = kvm_arch_init(opaque);
1737         if (r)
1738                 goto out_fail;
1739
1740         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1741
1742         if (bad_page == NULL) {
1743                 r = -ENOMEM;
1744                 goto out;
1745         }
1746
1747         bad_pfn = page_to_pfn(bad_page);
1748
1749         r = kvm_arch_hardware_setup();
1750         if (r < 0)
1751                 goto out_free_0;
1752
1753         for_each_online_cpu(cpu) {
1754                 smp_call_function_single(cpu,
1755                                 kvm_arch_check_processor_compat,
1756                                 &r, 1);
1757                 if (r < 0)
1758                         goto out_free_1;
1759         }
1760
1761         on_each_cpu(hardware_enable, NULL, 1);
1762         r = register_cpu_notifier(&kvm_cpu_notifier);
1763         if (r)
1764                 goto out_free_2;
1765         register_reboot_notifier(&kvm_reboot_notifier);
1766
1767         r = sysdev_class_register(&kvm_sysdev_class);
1768         if (r)
1769                 goto out_free_3;
1770
1771         r = sysdev_register(&kvm_sysdev);
1772         if (r)
1773                 goto out_free_4;
1774
1775         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1776         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1777                                            __alignof__(struct kvm_vcpu),
1778                                            0, NULL);
1779         if (!kvm_vcpu_cache) {
1780                 r = -ENOMEM;
1781                 goto out_free_5;
1782         }
1783
1784         kvm_chardev_ops.owner = module;
1785
1786         r = misc_register(&kvm_dev);
1787         if (r) {
1788                 printk(KERN_ERR "kvm: misc device register failed\n");
1789                 goto out_free;
1790         }
1791
1792         kvm_preempt_ops.sched_in = kvm_sched_in;
1793         kvm_preempt_ops.sched_out = kvm_sched_out;
1794
1795         return 0;
1796
1797 out_free:
1798         kmem_cache_destroy(kvm_vcpu_cache);
1799 out_free_5:
1800         sysdev_unregister(&kvm_sysdev);
1801 out_free_4:
1802         sysdev_class_unregister(&kvm_sysdev_class);
1803 out_free_3:
1804         unregister_reboot_notifier(&kvm_reboot_notifier);
1805         unregister_cpu_notifier(&kvm_cpu_notifier);
1806 out_free_2:
1807         on_each_cpu(hardware_disable, NULL, 1);
1808 out_free_1:
1809         kvm_arch_hardware_unsetup();
1810 out_free_0:
1811         __free_page(bad_page);
1812 out:
1813         kvm_arch_exit();
1814         kvm_exit_debug();
1815 out_fail:
1816         return r;
1817 }
1818 EXPORT_SYMBOL_GPL(kvm_init);
1819
1820 void kvm_exit(void)
1821 {
1822         kvm_trace_cleanup();
1823         misc_deregister(&kvm_dev);
1824         kmem_cache_destroy(kvm_vcpu_cache);
1825         sysdev_unregister(&kvm_sysdev);
1826         sysdev_class_unregister(&kvm_sysdev_class);
1827         unregister_reboot_notifier(&kvm_reboot_notifier);
1828         unregister_cpu_notifier(&kvm_cpu_notifier);
1829         on_each_cpu(hardware_disable, NULL, 1);
1830         kvm_arch_hardware_unsetup();
1831         kvm_arch_exit();
1832         kvm_exit_debug();
1833         __free_page(bad_page);
1834 }
1835 EXPORT_SYMBOL_GPL(kvm_exit);