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