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

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "iodev.h"
  19
  20 #include <linux/kvm_host.h>
  21 #include <linux/kvm.h>
  22 #include <linux/module.h>
  23 #include <linux/errno.h>
  24 #include <linux/percpu.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/sysdev.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48
  49 #include <asm/processor.h>
  50 #include <asm/io.h>
  51 #include <asm/uaccess.h>
  52 #include <asm/pgtable.h>
  53 #include <asm-generic/bitops/le.h>
  54
  55 #include "coalesced_mmio.h"
  56
  57 #define CREATE_TRACE_POINTS
  58 #include <trace/events/kvm.h>
  59
  60 MODULE_AUTHOR("Qumranet");
  61 MODULE_LICENSE("GPL");
  62
  63 /*
  64  * Ordering of locks:
  65  *
  66  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  67  */
  68
  69 DEFINE_SPINLOCK(kvm_lock);
  70 LIST_HEAD(vm_list);
  71
  72 static cpumask_var_t cpus_hardware_enabled;
  73 static int kvm_usage_count = 0;
  74 static atomic_t hardware_enable_failed;
  75
  76 struct kmem_cache *kvm_vcpu_cache;
  77 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  78
  79 static __read_mostly struct preempt_ops kvm_preempt_ops;
  80
  81 struct dentry *kvm_debugfs_dir;
  82
  83 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  84                            unsigned long arg);
  85 static int hardware_enable_all(void);
  86 static void hardware_disable_all(void);
  87
  88 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  89
  90 static bool kvm_rebooting;
  91
  92 static bool largepages_enabled = true;
  93
  94 inline int kvm_is_mmio_pfn(pfn_t pfn)
  95 {
  96         if (pfn_valid(pfn)) {
  97                 struct page *page = compound_head(pfn_to_page(pfn));
  98                 return PageReserved(page);
  99         }
 100
 101         return true;
 102 }
 103
 104 /*
 105  * Switches to specified vcpu, until a matching vcpu_put()
 106  */
 107 void vcpu_load(struct kvm_vcpu *vcpu)
 108 {
 109         int cpu;
 110
 111         mutex_lock(&vcpu->mutex);
 112         cpu = get_cpu();
 113         preempt_notifier_register(&vcpu->preempt_notifier);
 114         kvm_arch_vcpu_load(vcpu, cpu);
 115         put_cpu();
 116 }
 117
 118 void vcpu_put(struct kvm_vcpu *vcpu)
 119 {
 120         preempt_disable();
 121         kvm_arch_vcpu_put(vcpu);
 122         preempt_notifier_unregister(&vcpu->preempt_notifier);
 123         preempt_enable();
 124         mutex_unlock(&vcpu->mutex);
 125 }
 126
 127 static void ack_flush(void *_completed)
 128 {
 129 }
 130
 131 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 132 {
 133         int i, cpu, me;
 134         cpumask_var_t cpus;
 135         bool called = true;
 136         struct kvm_vcpu *vcpu;
 137
 138         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 139
 140         spin_lock(&kvm->requests_lock);
 141         me = smp_processor_id();
 142         kvm_for_each_vcpu(i, vcpu, kvm) {
 143                 if (test_and_set_bit(req, &vcpu->requests))
 144                         continue;
 145                 cpu = vcpu->cpu;
 146                 if (cpus != NULL && cpu != -1 && cpu != me)
 147                         cpumask_set_cpu(cpu, cpus);
 148         }
 149         if (unlikely(cpus == NULL))
 150                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 151         else if (!cpumask_empty(cpus))
 152                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 153         else
 154                 called = false;
 155         spin_unlock(&kvm->requests_lock);
 156         free_cpumask_var(cpus);
 157         return called;
 158 }
 159
 160 void kvm_flush_remote_tlbs(struct kvm *kvm)
 161 {
 162         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 163                 ++kvm->stat.remote_tlb_flush;
 164 }
 165
 166 void kvm_reload_remote_mmus(struct kvm *kvm)
 167 {
 168         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 169 }
 170
 171 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 172 {
 173         struct page *page;
 174         int r;
 175
 176         mutex_init(&vcpu->mutex);
 177         vcpu->cpu = -1;
 178         vcpu->kvm = kvm;
 179         vcpu->vcpu_id = id;
 180         init_waitqueue_head(&vcpu->wq);
 181
 182         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 183         if (!page) {
 184                 r = -ENOMEM;
 185                 goto fail;
 186         }
 187         vcpu->run = page_address(page);
 188
 189         r = kvm_arch_vcpu_init(vcpu);
 190         if (r < 0)
 191                 goto fail_free_run;
 192         return 0;
 193
 194 fail_free_run:
 195         free_page((unsigned long)vcpu->run);
 196 fail:
 197         return r;
 198 }
 199 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 200
 201 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 202 {
 203         kvm_arch_vcpu_uninit(vcpu);
 204         free_page((unsigned long)vcpu->run);
 205 }
 206 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 207
 208 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 209 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 210 {
 211         return container_of(mn, struct kvm, mmu_notifier);
 212 }
 213
 214 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 215                                              struct mm_struct *mm,
 216                                              unsigned long address)
 217 {
 218         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 219         int need_tlb_flush, idx;
 220
 221         /*
 222          * When ->invalidate_page runs, the linux pte has been zapped
 223          * already but the page is still allocated until
 224          * ->invalidate_page returns. So if we increase the sequence
 225          * here the kvm page fault will notice if the spte can't be
 226          * established because the page is going to be freed. If
 227          * instead the kvm page fault establishes the spte before
 228          * ->invalidate_page runs, kvm_unmap_hva will release it
 229          * before returning.
 230          *
 231          * The sequence increase only need to be seen at spin_unlock
 232          * time, and not at spin_lock time.
 233          *
 234          * Increasing the sequence after the spin_unlock would be
 235          * unsafe because the kvm page fault could then establish the
 236          * pte after kvm_unmap_hva returned, without noticing the page
 237          * is going to be freed.
 238          */
 239         idx = srcu_read_lock(&kvm->srcu);
 240         spin_lock(&kvm->mmu_lock);
 241         kvm->mmu_notifier_seq++;
 242         need_tlb_flush = kvm_unmap_hva(kvm, address);
 243         spin_unlock(&kvm->mmu_lock);
 244         srcu_read_unlock(&kvm->srcu, idx);
 245
 246         /* we've to flush the tlb before the pages can be freed */
 247         if (need_tlb_flush)
 248                 kvm_flush_remote_tlbs(kvm);
 249
 250 }
 251
 252 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 253                                         struct mm_struct *mm,
 254                                         unsigned long address,
 255                                         pte_t pte)
 256 {
 257         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 258         int idx;
 259
 260         idx = srcu_read_lock(&kvm->srcu);
 261         spin_lock(&kvm->mmu_lock);
 262         kvm->mmu_notifier_seq++;
 263         kvm_set_spte_hva(kvm, address, pte);
 264         spin_unlock(&kvm->mmu_lock);
 265         srcu_read_unlock(&kvm->srcu, idx);
 266 }
 267
 268 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 269                                                     struct mm_struct *mm,
 270                                                     unsigned long start,
 271                                                     unsigned long end)
 272 {
 273         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 274         int need_tlb_flush = 0, idx;
 275
 276         idx = srcu_read_lock(&kvm->srcu);
 277         spin_lock(&kvm->mmu_lock);
 278         /*
 279          * The count increase must become visible at unlock time as no
 280          * spte can be established without taking the mmu_lock and
 281          * count is also read inside the mmu_lock critical section.
 282          */
 283         kvm->mmu_notifier_count++;
 284         for (; start < end; start += PAGE_SIZE)
 285                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 286         spin_unlock(&kvm->mmu_lock);
 287         srcu_read_unlock(&kvm->srcu, idx);
 288
 289         /* we've to flush the tlb before the pages can be freed */
 290         if (need_tlb_flush)
 291                 kvm_flush_remote_tlbs(kvm);
 292 }
 293
 294 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 295                                                   struct mm_struct *mm,
 296                                                   unsigned long start,
 297                                                   unsigned long end)
 298 {
 299         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 300
 301         spin_lock(&kvm->mmu_lock);
 302         /*
 303          * This sequence increase will notify the kvm page fault that
 304          * the page that is going to be mapped in the spte could have
 305          * been freed.
 306          */
 307         kvm->mmu_notifier_seq++;
 308         /*
 309          * The above sequence increase must be visible before the
 310          * below count decrease but both values are read by the kvm
 311          * page fault under mmu_lock spinlock so we don't need to add
 312          * a smb_wmb() here in between the two.
 313          */
 314         kvm->mmu_notifier_count--;
 315         spin_unlock(&kvm->mmu_lock);
 316
 317         BUG_ON(kvm->mmu_notifier_count < 0);
 318 }
 319
 320 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 321                                               struct mm_struct *mm,
 322                                               unsigned long address)
 323 {
 324         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 325         int young, idx;
 326
 327         idx = srcu_read_lock(&kvm->srcu);
 328         spin_lock(&kvm->mmu_lock);
 329         young = kvm_age_hva(kvm, address);
 330         spin_unlock(&kvm->mmu_lock);
 331         srcu_read_unlock(&kvm->srcu, idx);
 332
 333         if (young)
 334                 kvm_flush_remote_tlbs(kvm);
 335
 336         return young;
 337 }
 338
 339 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 340                                      struct mm_struct *mm)
 341 {
 342         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 343         kvm_arch_flush_shadow(kvm);
 344 }
 345
 346 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 347         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 348         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 349         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 350         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 351         .change_pte             = kvm_mmu_notifier_change_pte,
 352         .release                = kvm_mmu_notifier_release,
 353 };
 354
 355 static int kvm_init_mmu_notifier(struct kvm *kvm)
 356 {
 357         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 358         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 359 }
 360
 361 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 362
 363 static int kvm_init_mmu_notifier(struct kvm *kvm)
 364 {
 365         return 0;
 366 }
 367
 368 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 369
 370 static struct kvm *kvm_create_vm(void)
 371 {
 372         int r = 0, i;
 373         struct kvm *kvm = kvm_arch_create_vm();
 374 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 375         struct page *page;
 376 #endif
 377
 378         if (IS_ERR(kvm))
 379                 goto out;
 380
 381         r = hardware_enable_all();
 382         if (r)
 383                 goto out_err_nodisable;
 384
 385 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 386         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 387         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 388 #endif
 389
 390         r = -ENOMEM;
 391         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 392         if (!kvm->memslots)
 393                 goto out_err;
 394         if (init_srcu_struct(&kvm->srcu))
 395                 goto out_err;
 396         for (i = 0; i < KVM_NR_BUSES; i++) {
 397                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 398                                         GFP_KERNEL);
 399                 if (!kvm->buses[i]) {
 400                         cleanup_srcu_struct(&kvm->srcu);
 401                         goto out_err;
 402                 }
 403         }
 404
 405 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 406         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 407         if (!page) {
 408                 cleanup_srcu_struct(&kvm->srcu);
 409                 goto out_err;
 410         }
 411
 412         kvm->coalesced_mmio_ring =
 413                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 414 #endif
 415
 416         r = kvm_init_mmu_notifier(kvm);
 417         if (r) {
 418                 cleanup_srcu_struct(&kvm->srcu);
 419 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 420                 put_page(page);
 421 #endif
 422                 goto out_err;
 423         }
 424
 425         kvm->mm = current->mm;
 426         atomic_inc(&kvm->mm->mm_count);
 427         spin_lock_init(&kvm->mmu_lock);
 428         spin_lock_init(&kvm->requests_lock);
 429         kvm_eventfd_init(kvm);
 430         mutex_init(&kvm->lock);
 431         mutex_init(&kvm->irq_lock);
 432         mutex_init(&kvm->slots_lock);
 433         atomic_set(&kvm->users_count, 1);
 434         spin_lock(&kvm_lock);
 435         list_add(&kvm->vm_list, &vm_list);
 436         spin_unlock(&kvm_lock);
 437 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 438         kvm_coalesced_mmio_init(kvm);
 439 #endif
 440 out:
 441         return kvm;
 442
 443 out_err:
 444         hardware_disable_all();
 445 out_err_nodisable:
 446         for (i = 0; i < KVM_NR_BUSES; i++)
 447                 kfree(kvm->buses[i]);
 448         kfree(kvm->memslots);
 449         kfree(kvm);
 450         return ERR_PTR(r);
 451 }
 452
 453 /*
 454  * Free any memory in @free but not in @dont.
 455  */
 456 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 457                                   struct kvm_memory_slot *dont)
 458 {
 459         int i;
 460
 461         if (!dont || free->rmap != dont->rmap)
 462                 vfree(free->rmap);
 463
 464         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 465                 vfree(free->dirty_bitmap);
 466
 467
 468         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 469                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 470                         vfree(free->lpage_info[i]);
 471                         free->lpage_info[i] = NULL;
 472                 }
 473         }
 474
 475         free->npages = 0;
 476         free->dirty_bitmap = NULL;
 477         free->rmap = NULL;
 478 }
 479
 480 void kvm_free_physmem(struct kvm *kvm)
 481 {
 482         int i;
 483         struct kvm_memslots *slots = kvm->memslots;
 484
 485         for (i = 0; i < slots->nmemslots; ++i)
 486                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 487
 488         kfree(kvm->memslots);
 489 }
 490
 491 static void kvm_destroy_vm(struct kvm *kvm)
 492 {
 493         int i;
 494         struct mm_struct *mm = kvm->mm;
 495
 496         kvm_arch_sync_events(kvm);
 497         spin_lock(&kvm_lock);
 498         list_del(&kvm->vm_list);
 499         spin_unlock(&kvm_lock);
 500         kvm_free_irq_routing(kvm);
 501         for (i = 0; i < KVM_NR_BUSES; i++)
 502                 kvm_io_bus_destroy(kvm->buses[i]);
 503         kvm_coalesced_mmio_free(kvm);
 504 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 505         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 506 #else
 507         kvm_arch_flush_shadow(kvm);
 508 #endif
 509         kvm_arch_destroy_vm(kvm);
 510         hardware_disable_all();
 511         mmdrop(mm);
 512 }
 513
 514 void kvm_get_kvm(struct kvm *kvm)
 515 {
 516         atomic_inc(&kvm->users_count);
 517 }
 518 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 519
 520 void kvm_put_kvm(struct kvm *kvm)
 521 {
 522         if (atomic_dec_and_test(&kvm->users_count))
 523                 kvm_destroy_vm(kvm);
 524 }
 525 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 526
 527
 528 static int kvm_vm_release(struct inode *inode, struct file *filp)
 529 {
 530         struct kvm *kvm = filp->private_data;
 531
 532         kvm_irqfd_release(kvm);
 533
 534         kvm_put_kvm(kvm);
 535         return 0;
 536 }
 537
 538 /*
 539  * Allocate some memory and give it an address in the guest physical address
 540  * space.
 541  *
 542  * Discontiguous memory is allowed, mostly for framebuffers.
 543  *
 544  * Must be called holding mmap_sem for write.
 545  */
 546 int __kvm_set_memory_region(struct kvm *kvm,
 547                             struct kvm_userspace_memory_region *mem,
 548                             int user_alloc)
 549 {
 550         int r, flush_shadow = 0;
 551         gfn_t base_gfn;
 552         unsigned long npages;
 553         unsigned long i;
 554         struct kvm_memory_slot *memslot;
 555         struct kvm_memory_slot old, new;
 556         struct kvm_memslots *slots, *old_memslots;
 557
 558         r = -EINVAL;
 559         /* General sanity checks */
 560         if (mem->memory_size & (PAGE_SIZE - 1))
 561                 goto out;
 562         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 563                 goto out;
 564         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 565                 goto out;
 566         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 567                 goto out;
 568         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 569                 goto out;
 570
 571         memslot = &kvm->memslots->memslots[mem->slot];
 572         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 573         npages = mem->memory_size >> PAGE_SHIFT;
 574
 575         if (!npages)
 576                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 577
 578         new = old = *memslot;
 579
 580         new.base_gfn = base_gfn;
 581         new.npages = npages;
 582         new.flags = mem->flags;
 583
 584         /* Disallow changing a memory slot's size. */
 585         r = -EINVAL;
 586         if (npages && old.npages && npages != old.npages)
 587                 goto out_free;
 588
 589         /* Check for overlaps */
 590         r = -EEXIST;
 591         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 592                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 593
 594                 if (s == memslot || !s->npages)
 595                         continue;
 596                 if (!((base_gfn + npages <= s->base_gfn) ||
 597                       (base_gfn >= s->base_gfn + s->npages)))
 598                         goto out_free;
 599         }
 600
 601         /* Free page dirty bitmap if unneeded */
 602         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 603                 new.dirty_bitmap = NULL;
 604
 605         r = -ENOMEM;
 606
 607         /* Allocate if a slot is being created */
 608 #ifndef CONFIG_S390
 609         if (npages && !new.rmap) {
 610                 new.rmap = vmalloc(npages * sizeof(struct page *));
 611
 612                 if (!new.rmap)
 613                         goto out_free;
 614
 615                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 616
 617                 new.user_alloc = user_alloc;
 618                 new.userspace_addr = mem->userspace_addr;
 619         }
 620         if (!npages)
 621                 goto skip_lpage;
 622
 623         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 624                 unsigned long ugfn;
 625                 unsigned long j;
 626                 int lpages;
 627                 int level = i + 2;
 628
 629                 /* Avoid unused variable warning if no large pages */
 630                 (void)level;
 631
 632                 if (new.lpage_info[i])
 633                         continue;
 634
 635                 lpages = 1 + (base_gfn + npages - 1) /
 636                              KVM_PAGES_PER_HPAGE(level);
 637                 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
 638
 639                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
 640
 641                 if (!new.lpage_info[i])
 642                         goto out_free;
 643
 644                 memset(new.lpage_info[i], 0,
 645                        lpages * sizeof(*new.lpage_info[i]));
 646
 647                 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
 648                         new.lpage_info[i][0].write_count = 1;
 649                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
 650                         new.lpage_info[i][lpages - 1].write_count = 1;
 651                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 652                 /*
 653                  * If the gfn and userspace address are not aligned wrt each
 654                  * other, or if explicitly asked to, disable large page
 655                  * support for this slot
 656                  */
 657                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 658                     !largepages_enabled)
 659                         for (j = 0; j < lpages; ++j)
 660                                 new.lpage_info[i][j].write_count = 1;
 661         }
 662
 663 skip_lpage:
 664
 665         /* Allocate page dirty bitmap if needed */
 666         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 667                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 668
 669                 new.dirty_bitmap = vmalloc(dirty_bytes);
 670                 if (!new.dirty_bitmap)
 671                         goto out_free;
 672                 memset(new.dirty_bitmap, 0, dirty_bytes);
 673                 /* destroy any largepage mappings for dirty tracking */
 674                 if (old.npages)
 675                         flush_shadow = 1;
 676         }
 677 #else  /* not defined CONFIG_S390 */
 678         new.user_alloc = user_alloc;
 679         if (user_alloc)
 680                 new.userspace_addr = mem->userspace_addr;
 681 #endif /* not defined CONFIG_S390 */
 682
 683         if (!npages) {
 684                 r = -ENOMEM;
 685                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 686                 if (!slots)
 687                         goto out_free;
 688                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 689                 if (mem->slot >= slots->nmemslots)
 690                         slots->nmemslots = mem->slot + 1;
 691                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 692
 693                 old_memslots = kvm->memslots;
 694                 rcu_assign_pointer(kvm->memslots, slots);
 695                 synchronize_srcu_expedited(&kvm->srcu);
 696                 /* From this point no new shadow pages pointing to a deleted
 697                  * memslot will be created.
 698                  *
 699                  * validation of sp->gfn happens in:
 700                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 701                  *      - kvm_is_visible_gfn (mmu_check_roots)
 702                  */
 703                 kvm_arch_flush_shadow(kvm);
 704                 kfree(old_memslots);
 705         }
 706
 707         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 708         if (r)
 709                 goto out_free;
 710
 711 #ifdef CONFIG_DMAR
 712         /* map the pages in iommu page table */
 713         if (npages) {
 714                 r = kvm_iommu_map_pages(kvm, &new);
 715                 if (r)
 716                         goto out_free;
 717         }
 718 #endif
 719
 720         r = -ENOMEM;
 721         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 722         if (!slots)
 723                 goto out_free;
 724         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 725         if (mem->slot >= slots->nmemslots)
 726                 slots->nmemslots = mem->slot + 1;
 727
 728         /* actual memory is freed via old in kvm_free_physmem_slot below */
 729         if (!npages) {
 730                 new.rmap = NULL;
 731                 new.dirty_bitmap = NULL;
 732                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 733                         new.lpage_info[i] = NULL;
 734         }
 735
 736         slots->memslots[mem->slot] = new;
 737         old_memslots = kvm->memslots;
 738         rcu_assign_pointer(kvm->memslots, slots);
 739         synchronize_srcu_expedited(&kvm->srcu);
 740
 741         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 742
 743         kvm_free_physmem_slot(&old, &new);
 744         kfree(old_memslots);
 745
 746         if (flush_shadow)
 747                 kvm_arch_flush_shadow(kvm);
 748
 749         return 0;
 750
 751 out_free:
 752         kvm_free_physmem_slot(&new, &old);
 753 out:
 754         return r;
 755
 756 }
 757 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 758
 759 int kvm_set_memory_region(struct kvm *kvm,
 760                           struct kvm_userspace_memory_region *mem,
 761                           int user_alloc)
 762 {
 763         int r;
 764
 765         mutex_lock(&kvm->slots_lock);
 766         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 767         mutex_unlock(&kvm->slots_lock);
 768         return r;
 769 }
 770 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 771
 772 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 773                                    struct
 774                                    kvm_userspace_memory_region *mem,
 775                                    int user_alloc)
 776 {
 777         if (mem->slot >= KVM_MEMORY_SLOTS)
 778                 return -EINVAL;
 779         return kvm_set_memory_region(kvm, mem, user_alloc);
 780 }
 781
 782 int kvm_get_dirty_log(struct kvm *kvm,
 783                         struct kvm_dirty_log *log, int *is_dirty)
 784 {
 785         struct kvm_memory_slot *memslot;
 786         int r, i;
 787         int n;
 788         unsigned long any = 0;
 789
 790         r = -EINVAL;
 791         if (log->slot >= KVM_MEMORY_SLOTS)
 792                 goto out;
 793
 794         memslot = &kvm->memslots->memslots[log->slot];
 795         r = -ENOENT;
 796         if (!memslot->dirty_bitmap)
 797                 goto out;
 798
 799         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 800
 801         for (i = 0; !any && i < n/sizeof(long); ++i)
 802                 any = memslot->dirty_bitmap[i];
 803
 804         r = -EFAULT;
 805         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 806                 goto out;
 807
 808         if (any)
 809                 *is_dirty = 1;
 810
 811         r = 0;
 812 out:
 813         return r;
 814 }
 815
 816 void kvm_disable_largepages(void)
 817 {
 818         largepages_enabled = false;
 819 }
 820 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 821
 822 int is_error_page(struct page *page)
 823 {
 824         return page == bad_page;
 825 }
 826 EXPORT_SYMBOL_GPL(is_error_page);
 827
 828 int is_error_pfn(pfn_t pfn)
 829 {
 830         return pfn == bad_pfn;
 831 }
 832 EXPORT_SYMBOL_GPL(is_error_pfn);
 833
 834 static inline unsigned long bad_hva(void)
 835 {
 836         return PAGE_OFFSET;
 837 }
 838
 839 int kvm_is_error_hva(unsigned long addr)
 840 {
 841         return addr == bad_hva();
 842 }
 843 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 844
 845 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
 846 {
 847         int i;
 848         struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
 849
 850         for (i = 0; i < slots->nmemslots; ++i) {
 851                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 852
 853                 if (gfn >= memslot->base_gfn
 854                     && gfn < memslot->base_gfn + memslot->npages)
 855                         return memslot;
 856         }
 857         return NULL;
 858 }
 859 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
 860
 861 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 862 {
 863         gfn = unalias_gfn(kvm, gfn);
 864         return gfn_to_memslot_unaliased(kvm, gfn);
 865 }
 866
 867 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 868 {
 869         int i;
 870         struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
 871
 872         gfn = unalias_gfn_instantiation(kvm, gfn);
 873         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 874                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 875
 876                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 877                         continue;
 878
 879                 if (gfn >= memslot->base_gfn
 880                     && gfn < memslot->base_gfn + memslot->npages)
 881                         return 1;
 882         }
 883         return 0;
 884 }
 885 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 886
 887 int memslot_id(struct kvm *kvm, gfn_t gfn)
 888 {
 889         int i;
 890         struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
 891         struct kvm_memory_slot *memslot = NULL;
 892
 893         gfn = unalias_gfn(kvm, gfn);
 894         for (i = 0; i < slots->nmemslots; ++i) {
 895                 memslot = &slots->memslots[i];
 896
 897                 if (gfn >= memslot->base_gfn
 898                     && gfn < memslot->base_gfn + memslot->npages)
 899                         break;
 900         }
 901
 902         return memslot - slots->memslots;
 903 }
 904
 905 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 906 {
 907         struct kvm_memory_slot *slot;
 908
 909         gfn = unalias_gfn_instantiation(kvm, gfn);
 910         slot = gfn_to_memslot_unaliased(kvm, gfn);
 911         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 912                 return bad_hva();
 913         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 914 }
 915 EXPORT_SYMBOL_GPL(gfn_to_hva);
 916
 917 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr)
 918 {
 919         struct page *page[1];
 920         int npages;
 921         pfn_t pfn;
 922
 923         might_sleep();
 924
 925         npages = get_user_pages_fast(addr, 1, 1, page);
 926
 927         if (unlikely(npages != 1)) {
 928                 struct vm_area_struct *vma;
 929
 930                 down_read(&current->mm->mmap_sem);
 931                 vma = find_vma(current->mm, addr);
 932
 933                 if (vma == NULL || addr < vma->vm_start ||
 934                     !(vma->vm_flags & VM_PFNMAP)) {
 935                         up_read(&current->mm->mmap_sem);
 936                         get_page(bad_page);
 937                         return page_to_pfn(bad_page);
 938                 }
 939
 940                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 941                 up_read(&current->mm->mmap_sem);
 942                 BUG_ON(!kvm_is_mmio_pfn(pfn));
 943         } else
 944                 pfn = page_to_pfn(page[0]);
 945
 946         return pfn;
 947 }
 948
 949 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 950 {
 951         unsigned long addr;
 952
 953         addr = gfn_to_hva(kvm, gfn);
 954         if (kvm_is_error_hva(addr)) {
 955                 get_page(bad_page);
 956                 return page_to_pfn(bad_page);
 957         }
 958
 959         return hva_to_pfn(kvm, addr);
 960 }
 961 EXPORT_SYMBOL_GPL(gfn_to_pfn);
 962
 963 static unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
 964 {
 965         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 966 }
 967
 968 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
 969                          struct kvm_memory_slot *slot, gfn_t gfn)
 970 {
 971         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
 972         return hva_to_pfn(kvm, addr);
 973 }
 974
 975 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 976 {
 977         pfn_t pfn;
 978
 979         pfn = gfn_to_pfn(kvm, gfn);
 980         if (!kvm_is_mmio_pfn(pfn))
 981                 return pfn_to_page(pfn);
 982
 983         WARN_ON(kvm_is_mmio_pfn(pfn));
 984
 985         get_page(bad_page);
 986         return bad_page;
 987 }
 988
 989 EXPORT_SYMBOL_GPL(gfn_to_page);
 990
 991 void kvm_release_page_clean(struct page *page)
 992 {
 993         kvm_release_pfn_clean(page_to_pfn(page));
 994 }
 995 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
 996
 997 void kvm_release_pfn_clean(pfn_t pfn)
 998 {
 999         if (!kvm_is_mmio_pfn(pfn))
1000                 put_page(pfn_to_page(pfn));
1001 }
1002 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1003
1004 void kvm_release_page_dirty(struct page *page)
1005 {
1006         kvm_release_pfn_dirty(page_to_pfn(page));
1007 }
1008 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1009
1010 void kvm_release_pfn_dirty(pfn_t pfn)
1011 {
1012         kvm_set_pfn_dirty(pfn);
1013         kvm_release_pfn_clean(pfn);
1014 }
1015 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1016
1017 void kvm_set_page_dirty(struct page *page)
1018 {
1019         kvm_set_pfn_dirty(page_to_pfn(page));
1020 }
1021 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1022
1023 void kvm_set_pfn_dirty(pfn_t pfn)
1024 {
1025         if (!kvm_is_mmio_pfn(pfn)) {
1026                 struct page *page = pfn_to_page(pfn);
1027                 if (!PageReserved(page))
1028                         SetPageDirty(page);
1029         }
1030 }
1031 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1032
1033 void kvm_set_pfn_accessed(pfn_t pfn)
1034 {
1035         if (!kvm_is_mmio_pfn(pfn))
1036                 mark_page_accessed(pfn_to_page(pfn));
1037 }
1038 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1039
1040 void kvm_get_pfn(pfn_t pfn)
1041 {
1042         if (!kvm_is_mmio_pfn(pfn))
1043                 get_page(pfn_to_page(pfn));
1044 }
1045 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1046
1047 static int next_segment(unsigned long len, int offset)
1048 {
1049         if (len > PAGE_SIZE - offset)
1050                 return PAGE_SIZE - offset;
1051         else
1052                 return len;
1053 }
1054
1055 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1056                         int len)
1057 {
1058         int r;
1059         unsigned long addr;
1060
1061         addr = gfn_to_hva(kvm, gfn);
1062         if (kvm_is_error_hva(addr))
1063                 return -EFAULT;
1064         r = copy_from_user(data, (void __user *)addr + offset, len);
1065         if (r)
1066                 return -EFAULT;
1067         return 0;
1068 }
1069 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1070
1071 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1072 {
1073         gfn_t gfn = gpa >> PAGE_SHIFT;
1074         int seg;
1075         int offset = offset_in_page(gpa);
1076         int ret;
1077
1078         while ((seg = next_segment(len, offset)) != 0) {
1079                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1080                 if (ret < 0)
1081                         return ret;
1082                 offset = 0;
1083                 len -= seg;
1084                 data += seg;
1085                 ++gfn;
1086         }
1087         return 0;
1088 }
1089 EXPORT_SYMBOL_GPL(kvm_read_guest);
1090
1091 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1092                           unsigned long len)
1093 {
1094         int r;
1095         unsigned long addr;
1096         gfn_t gfn = gpa >> PAGE_SHIFT;
1097         int offset = offset_in_page(gpa);
1098
1099         addr = gfn_to_hva(kvm, gfn);
1100         if (kvm_is_error_hva(addr))
1101                 return -EFAULT;
1102         pagefault_disable();
1103         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1104         pagefault_enable();
1105         if (r)
1106                 return -EFAULT;
1107         return 0;
1108 }
1109 EXPORT_SYMBOL(kvm_read_guest_atomic);
1110
1111 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1112                          int offset, int len)
1113 {
1114         int r;
1115         unsigned long addr;
1116
1117         addr = gfn_to_hva(kvm, gfn);
1118         if (kvm_is_error_hva(addr))
1119                 return -EFAULT;
1120         r = copy_to_user((void __user *)addr + offset, data, len);
1121         if (r)
1122                 return -EFAULT;
1123         mark_page_dirty(kvm, gfn);
1124         return 0;
1125 }
1126 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1127
1128 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1129                     unsigned long len)
1130 {
1131         gfn_t gfn = gpa >> PAGE_SHIFT;
1132         int seg;
1133         int offset = offset_in_page(gpa);
1134         int ret;
1135
1136         while ((seg = next_segment(len, offset)) != 0) {
1137                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1138                 if (ret < 0)
1139                         return ret;
1140                 offset = 0;
1141                 len -= seg;
1142                 data += seg;
1143                 ++gfn;
1144         }
1145         return 0;
1146 }
1147
1148 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1149 {
1150         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1151 }
1152 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1153
1154 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1155 {
1156         gfn_t gfn = gpa >> PAGE_SHIFT;
1157         int seg;
1158         int offset = offset_in_page(gpa);
1159         int ret;
1160
1161         while ((seg = next_segment(len, offset)) != 0) {
1162                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1163                 if (ret < 0)
1164                         return ret;
1165                 offset = 0;
1166                 len -= seg;
1167                 ++gfn;
1168         }
1169         return 0;
1170 }
1171 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1172
1173 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1174 {
1175         struct kvm_memory_slot *memslot;
1176
1177         gfn = unalias_gfn(kvm, gfn);
1178         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1179         if (memslot && memslot->dirty_bitmap) {
1180                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1181
1182                 /* avoid RMW */
1183                 if (!generic_test_le_bit(rel_gfn, memslot->dirty_bitmap))
1184                         generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1185         }
1186 }
1187
1188 /*
1189  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1190  */
1191 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1192 {
1193         DEFINE_WAIT(wait);
1194
1195         for (;;) {
1196                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1197
1198                 if (kvm_arch_vcpu_runnable(vcpu)) {
1199                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1200                         break;
1201                 }
1202                 if (kvm_cpu_has_pending_timer(vcpu))
1203                         break;
1204                 if (signal_pending(current))
1205                         break;
1206
1207                 schedule();
1208         }
1209
1210         finish_wait(&vcpu->wq, &wait);
1211 }
1212
1213 void kvm_resched(struct kvm_vcpu *vcpu)
1214 {
1215         if (!need_resched())
1216                 return;
1217         cond_resched();
1218 }
1219 EXPORT_SYMBOL_GPL(kvm_resched);
1220
1221 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1222 {
1223         ktime_t expires;
1224         DEFINE_WAIT(wait);
1225
1226         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1227
1228         /* Sleep for 100 us, and hope lock-holder got scheduled */
1229         expires = ktime_add_ns(ktime_get(), 100000UL);
1230         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1231
1232         finish_wait(&vcpu->wq, &wait);
1233 }
1234 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1235
1236 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1237 {
1238         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1239         struct page *page;
1240
1241         if (vmf->pgoff == 0)
1242                 page = virt_to_page(vcpu->run);
1243 #ifdef CONFIG_X86
1244         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1245                 page = virt_to_page(vcpu->arch.pio_data);
1246 #endif
1247 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1248         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1249                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1250 #endif
1251         else
1252                 return VM_FAULT_SIGBUS;
1253         get_page(page);
1254         vmf->page = page;
1255         return 0;
1256 }
1257
1258 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1259         .fault = kvm_vcpu_fault,
1260 };
1261
1262 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1263 {
1264         vma->vm_ops = &kvm_vcpu_vm_ops;
1265         return 0;
1266 }
1267
1268 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1269 {
1270         struct kvm_vcpu *vcpu = filp->private_data;
1271
1272         kvm_put_kvm(vcpu->kvm);
1273         return 0;
1274 }
1275
1276 static struct file_operations kvm_vcpu_fops = {
1277         .release        = kvm_vcpu_release,
1278         .unlocked_ioctl = kvm_vcpu_ioctl,
1279         .compat_ioctl   = kvm_vcpu_ioctl,
1280         .mmap           = kvm_vcpu_mmap,
1281 };
1282
1283 /*
1284  * Allocates an inode for the vcpu.
1285  */
1286 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1287 {
1288         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1289 }
1290
1291 /*
1292  * Creates some virtual cpus.  Good luck creating more than one.
1293  */
1294 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1295 {
1296         int r;
1297         struct kvm_vcpu *vcpu, *v;
1298
1299         vcpu = kvm_arch_vcpu_create(kvm, id);
1300         if (IS_ERR(vcpu))
1301                 return PTR_ERR(vcpu);
1302
1303         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1304
1305         r = kvm_arch_vcpu_setup(vcpu);
1306         if (r)
1307                 return r;
1308
1309         mutex_lock(&kvm->lock);
1310         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1311                 r = -EINVAL;
1312                 goto vcpu_destroy;
1313         }
1314
1315         kvm_for_each_vcpu(r, v, kvm)
1316                 if (v->vcpu_id == id) {
1317                         r = -EEXIST;
1318                         goto vcpu_destroy;
1319                 }
1320
1321         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1322
1323         /* Now it's all set up, let userspace reach it */
1324         kvm_get_kvm(kvm);
1325         r = create_vcpu_fd(vcpu);
1326         if (r < 0) {
1327                 kvm_put_kvm(kvm);
1328                 goto vcpu_destroy;
1329         }
1330
1331         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1332         smp_wmb();
1333         atomic_inc(&kvm->online_vcpus);
1334
1335 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1336         if (kvm->bsp_vcpu_id == id)
1337                 kvm->bsp_vcpu = vcpu;
1338 #endif
1339         mutex_unlock(&kvm->lock);
1340         return r;
1341
1342 vcpu_destroy:
1343         mutex_unlock(&kvm->lock);
1344         kvm_arch_vcpu_destroy(vcpu);
1345         return r;
1346 }
1347
1348 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1349 {
1350         if (sigset) {
1351                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1352                 vcpu->sigset_active = 1;
1353                 vcpu->sigset = *sigset;
1354         } else
1355                 vcpu->sigset_active = 0;
1356         return 0;
1357 }
1358
1359 static long kvm_vcpu_ioctl(struct file *filp,
1360                            unsigned int ioctl, unsigned long arg)
1361 {
1362         struct kvm_vcpu *vcpu = filp->private_data;
1363         void __user *argp = (void __user *)arg;
1364         int r;
1365         struct kvm_fpu *fpu = NULL;
1366         struct kvm_sregs *kvm_sregs = NULL;
1367
1368         if (vcpu->kvm->mm != current->mm)
1369                 return -EIO;
1370         switch (ioctl) {
1371         case KVM_RUN:
1372                 r = -EINVAL;
1373                 if (arg)
1374                         goto out;
1375                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1376                 break;
1377         case KVM_GET_REGS: {
1378                 struct kvm_regs *kvm_regs;
1379
1380                 r = -ENOMEM;
1381                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1382                 if (!kvm_regs)
1383                         goto out;
1384                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1385                 if (r)
1386                         goto out_free1;
1387                 r = -EFAULT;
1388                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1389                         goto out_free1;
1390                 r = 0;
1391 out_free1:
1392                 kfree(kvm_regs);
1393                 break;
1394         }
1395         case KVM_SET_REGS: {
1396                 struct kvm_regs *kvm_regs;
1397
1398                 r = -ENOMEM;
1399                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1400                 if (!kvm_regs)
1401                         goto out;
1402                 r = -EFAULT;
1403                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1404                         goto out_free2;
1405                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1406                 if (r)
1407                         goto out_free2;
1408                 r = 0;
1409 out_free2:
1410                 kfree(kvm_regs);
1411                 break;
1412         }
1413         case KVM_GET_SREGS: {
1414                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1415                 r = -ENOMEM;
1416                 if (!kvm_sregs)
1417                         goto out;
1418                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1419                 if (r)
1420                         goto out;
1421                 r = -EFAULT;
1422                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1423                         goto out;
1424                 r = 0;
1425                 break;
1426         }
1427         case KVM_SET_SREGS: {
1428                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1429                 r = -ENOMEM;
1430                 if (!kvm_sregs)
1431                         goto out;
1432                 r = -EFAULT;
1433                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1434                         goto out;
1435                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1436                 if (r)
1437                         goto out;
1438                 r = 0;
1439                 break;
1440         }
1441         case KVM_GET_MP_STATE: {
1442                 struct kvm_mp_state mp_state;
1443
1444                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1445                 if (r)
1446                         goto out;
1447                 r = -EFAULT;
1448                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1449                         goto out;
1450                 r = 0;
1451                 break;
1452         }
1453         case KVM_SET_MP_STATE: {
1454                 struct kvm_mp_state mp_state;
1455
1456                 r = -EFAULT;
1457                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1458                         goto out;
1459                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1460                 if (r)
1461                         goto out;
1462                 r = 0;
1463                 break;
1464         }
1465         case KVM_TRANSLATE: {
1466                 struct kvm_translation tr;
1467
1468                 r = -EFAULT;
1469                 if (copy_from_user(&tr, argp, sizeof tr))
1470                         goto out;
1471                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1472                 if (r)
1473                         goto out;
1474                 r = -EFAULT;
1475                 if (copy_to_user(argp, &tr, sizeof tr))
1476                         goto out;
1477                 r = 0;
1478                 break;
1479         }
1480         case KVM_SET_GUEST_DEBUG: {
1481                 struct kvm_guest_debug dbg;
1482
1483                 r = -EFAULT;
1484                 if (copy_from_user(&dbg, argp, sizeof dbg))
1485                         goto out;
1486                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1487                 if (r)
1488                         goto out;
1489                 r = 0;
1490                 break;
1491         }
1492         case KVM_SET_SIGNAL_MASK: {
1493                 struct kvm_signal_mask __user *sigmask_arg = argp;
1494                 struct kvm_signal_mask kvm_sigmask;
1495                 sigset_t sigset, *p;
1496
1497                 p = NULL;
1498                 if (argp) {
1499                         r = -EFAULT;
1500                         if (copy_from_user(&kvm_sigmask, argp,
1501                                            sizeof kvm_sigmask))
1502                                 goto out;
1503                         r = -EINVAL;
1504                         if (kvm_sigmask.len != sizeof sigset)
1505                                 goto out;
1506                         r = -EFAULT;
1507                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1508                                            sizeof sigset))
1509                                 goto out;
1510                         p = &sigset;
1511                 }
1512                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1513                 break;
1514         }
1515         case KVM_GET_FPU: {
1516                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1517                 r = -ENOMEM;
1518                 if (!fpu)
1519                         goto out;
1520                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1521                 if (r)
1522                         goto out;
1523                 r = -EFAULT;
1524                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1525                         goto out;
1526                 r = 0;
1527                 break;
1528         }
1529         case KVM_SET_FPU: {
1530                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1531                 r = -ENOMEM;
1532                 if (!fpu)
1533                         goto out;
1534                 r = -EFAULT;
1535                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1536                         goto out;
1537                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1538                 if (r)
1539                         goto out;
1540                 r = 0;
1541                 break;
1542         }
1543         default:
1544                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1545         }
1546 out:
1547         kfree(fpu);
1548         kfree(kvm_sregs);
1549         return r;
1550 }
1551
1552 static long kvm_vm_ioctl(struct file *filp,
1553                            unsigned int ioctl, unsigned long arg)
1554 {
1555         struct kvm *kvm = filp->private_data;
1556         void __user *argp = (void __user *)arg;
1557         int r;
1558
1559         if (kvm->mm != current->mm)
1560                 return -EIO;
1561         switch (ioctl) {
1562         case KVM_CREATE_VCPU:
1563                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1564                 if (r < 0)
1565                         goto out;
1566                 break;
1567         case KVM_SET_USER_MEMORY_REGION: {
1568                 struct kvm_userspace_memory_region kvm_userspace_mem;
1569
1570                 r = -EFAULT;
1571                 if (copy_from_user(&kvm_userspace_mem, argp,
1572                                                 sizeof kvm_userspace_mem))
1573                         goto out;
1574
1575                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1576                 if (r)
1577                         goto out;
1578                 break;
1579         }
1580         case KVM_GET_DIRTY_LOG: {
1581                 struct kvm_dirty_log log;
1582
1583                 r = -EFAULT;
1584                 if (copy_from_user(&log, argp, sizeof log))
1585                         goto out;
1586                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1587                 if (r)
1588                         goto out;
1589                 break;
1590         }
1591 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1592         case KVM_REGISTER_COALESCED_MMIO: {
1593                 struct kvm_coalesced_mmio_zone zone;
1594                 r = -EFAULT;
1595                 if (copy_from_user(&zone, argp, sizeof zone))
1596                         goto out;
1597                 r = -ENXIO;
1598                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1599                 if (r)
1600                         goto out;
1601                 r = 0;
1602                 break;
1603         }
1604         case KVM_UNREGISTER_COALESCED_MMIO: {
1605                 struct kvm_coalesced_mmio_zone zone;
1606                 r = -EFAULT;
1607                 if (copy_from_user(&zone, argp, sizeof zone))
1608                         goto out;
1609                 r = -ENXIO;
1610                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1611                 if (r)
1612                         goto out;
1613                 r = 0;
1614                 break;
1615         }
1616 #endif
1617         case KVM_IRQFD: {
1618                 struct kvm_irqfd data;
1619
1620                 r = -EFAULT;
1621                 if (copy_from_user(&data, argp, sizeof data))
1622                         goto out;
1623                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1624                 break;
1625         }
1626         case KVM_IOEVENTFD: {
1627                 struct kvm_ioeventfd data;
1628
1629                 r = -EFAULT;
1630                 if (copy_from_user(&data, argp, sizeof data))
1631                         goto out;
1632                 r = kvm_ioeventfd(kvm, &data);
1633                 break;
1634         }
1635 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1636         case KVM_SET_BOOT_CPU_ID:
1637                 r = 0;
1638                 mutex_lock(&kvm->lock);
1639                 if (atomic_read(&kvm->online_vcpus) != 0)
1640                         r = -EBUSY;
1641                 else
1642                         kvm->bsp_vcpu_id = arg;
1643                 mutex_unlock(&kvm->lock);
1644                 break;
1645 #endif
1646         default:
1647                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1648                 if (r == -ENOTTY)
1649                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1650         }
1651 out:
1652         return r;
1653 }
1654
1655 #ifdef CONFIG_COMPAT
1656 struct compat_kvm_dirty_log {
1657         __u32 slot;
1658         __u32 padding1;
1659         union {
1660                 compat_uptr_t dirty_bitmap; /* one bit per page */
1661                 __u64 padding2;
1662         };
1663 };
1664
1665 static long kvm_vm_compat_ioctl(struct file *filp,
1666                            unsigned int ioctl, unsigned long arg)
1667 {
1668         struct kvm *kvm = filp->private_data;
1669         int r;
1670
1671         if (kvm->mm != current->mm)
1672                 return -EIO;
1673         switch (ioctl) {
1674         case KVM_GET_DIRTY_LOG: {
1675                 struct compat_kvm_dirty_log compat_log;
1676                 struct kvm_dirty_log log;
1677
1678                 r = -EFAULT;
1679                 if (copy_from_user(&compat_log, (void __user *)arg,
1680                                    sizeof(compat_log)))
1681                         goto out;
1682                 log.slot         = compat_log.slot;
1683                 log.padding1     = compat_log.padding1;
1684                 log.padding2     = compat_log.padding2;
1685                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1686
1687                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1688                 if (r)
1689                         goto out;
1690                 break;
1691         }
1692         default:
1693                 r = kvm_vm_ioctl(filp, ioctl, arg);
1694         }
1695
1696 out:
1697         return r;
1698 }
1699 #endif
1700
1701 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1702 {
1703         struct page *page[1];
1704         unsigned long addr;
1705         int npages;
1706         gfn_t gfn = vmf->pgoff;
1707         struct kvm *kvm = vma->vm_file->private_data;
1708
1709         addr = gfn_to_hva(kvm, gfn);
1710         if (kvm_is_error_hva(addr))
1711                 return VM_FAULT_SIGBUS;
1712
1713         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1714                                 NULL);
1715         if (unlikely(npages != 1))
1716                 return VM_FAULT_SIGBUS;
1717
1718         vmf->page = page[0];
1719         return 0;
1720 }
1721
1722 static const struct vm_operations_struct kvm_vm_vm_ops = {
1723         .fault = kvm_vm_fault,
1724 };
1725
1726 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1727 {
1728         vma->vm_ops = &kvm_vm_vm_ops;
1729         return 0;
1730 }
1731
1732 static struct file_operations kvm_vm_fops = {
1733         .release        = kvm_vm_release,
1734         .unlocked_ioctl = kvm_vm_ioctl,
1735 #ifdef CONFIG_COMPAT
1736         .compat_ioctl   = kvm_vm_compat_ioctl,
1737 #endif
1738         .mmap           = kvm_vm_mmap,
1739 };
1740
1741 static int kvm_dev_ioctl_create_vm(void)
1742 {
1743         int fd;
1744         struct kvm *kvm;
1745
1746         kvm = kvm_create_vm();
1747         if (IS_ERR(kvm))
1748                 return PTR_ERR(kvm);
1749         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1750         if (fd < 0)
1751                 kvm_put_kvm(kvm);
1752
1753         return fd;
1754 }
1755
1756 static long kvm_dev_ioctl_check_extension_generic(long arg)
1757 {
1758         switch (arg) {
1759         case KVM_CAP_USER_MEMORY:
1760         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1761         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1762 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1763         case KVM_CAP_SET_BOOT_CPU_ID:
1764 #endif
1765         case KVM_CAP_INTERNAL_ERROR_DATA:
1766                 return 1;
1767 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1768         case KVM_CAP_IRQ_ROUTING:
1769                 return KVM_MAX_IRQ_ROUTES;
1770 #endif
1771         default:
1772                 break;
1773         }
1774         return kvm_dev_ioctl_check_extension(arg);
1775 }
1776
1777 static long kvm_dev_ioctl(struct file *filp,
1778                           unsigned int ioctl, unsigned long arg)
1779 {
1780         long r = -EINVAL;
1781
1782         switch (ioctl) {
1783         case KVM_GET_API_VERSION:
1784                 r = -EINVAL;
1785                 if (arg)
1786                         goto out;
1787                 r = KVM_API_VERSION;
1788                 break;
1789         case KVM_CREATE_VM:
1790                 r = -EINVAL;
1791                 if (arg)
1792                         goto out;
1793                 r = kvm_dev_ioctl_create_vm();
1794                 break;
1795         case KVM_CHECK_EXTENSION:
1796                 r = kvm_dev_ioctl_check_extension_generic(arg);
1797                 break;
1798         case KVM_GET_VCPU_MMAP_SIZE:
1799                 r = -EINVAL;
1800                 if (arg)
1801                         goto out;
1802                 r = PAGE_SIZE;     /* struct kvm_run */
1803 #ifdef CONFIG_X86
1804                 r += PAGE_SIZE;    /* pio data page */
1805 #endif
1806 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1807                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1808 #endif
1809                 break;
1810         case KVM_TRACE_ENABLE:
1811         case KVM_TRACE_PAUSE:
1812         case KVM_TRACE_DISABLE:
1813                 r = -EOPNOTSUPP;
1814                 break;
1815         default:
1816                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1817         }
1818 out:
1819         return r;
1820 }
1821
1822 static struct file_operations kvm_chardev_ops = {
1823         .unlocked_ioctl = kvm_dev_ioctl,
1824         .compat_ioctl   = kvm_dev_ioctl,
1825 };
1826
1827 static struct miscdevice kvm_dev = {
1828         KVM_MINOR,
1829         "kvm",
1830         &kvm_chardev_ops,
1831 };
1832
1833 static void hardware_enable(void *junk)
1834 {
1835         int cpu = raw_smp_processor_id();
1836         int r;
1837
1838         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1839                 return;
1840
1841         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1842
1843         r = kvm_arch_hardware_enable(NULL);
1844
1845         if (r) {
1846                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1847                 atomic_inc(&hardware_enable_failed);
1848                 printk(KERN_INFO "kvm: enabling virtualization on "
1849                                  "CPU%d failed\n", cpu);
1850         }
1851 }
1852
1853 static void hardware_disable(void *junk)
1854 {
1855         int cpu = raw_smp_processor_id();
1856
1857         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1858                 return;
1859         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1860         kvm_arch_hardware_disable(NULL);
1861 }
1862
1863 static void hardware_disable_all_nolock(void)
1864 {
1865         BUG_ON(!kvm_usage_count);
1866
1867         kvm_usage_count--;
1868         if (!kvm_usage_count)
1869                 on_each_cpu(hardware_disable, NULL, 1);
1870 }
1871
1872 static void hardware_disable_all(void)
1873 {
1874         spin_lock(&kvm_lock);
1875         hardware_disable_all_nolock();
1876         spin_unlock(&kvm_lock);
1877 }
1878
1879 static int hardware_enable_all(void)
1880 {
1881         int r = 0;
1882
1883         spin_lock(&kvm_lock);
1884
1885         kvm_usage_count++;
1886         if (kvm_usage_count == 1) {
1887                 atomic_set(&hardware_enable_failed, 0);
1888                 on_each_cpu(hardware_enable, NULL, 1);
1889
1890                 if (atomic_read(&hardware_enable_failed)) {
1891                         hardware_disable_all_nolock();
1892                         r = -EBUSY;
1893                 }
1894         }
1895
1896         spin_unlock(&kvm_lock);
1897
1898         return r;
1899 }
1900
1901 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1902                            void *v)
1903 {
1904         int cpu = (long)v;
1905
1906         if (!kvm_usage_count)
1907                 return NOTIFY_OK;
1908
1909         val &= ~CPU_TASKS_FROZEN;
1910         switch (val) {
1911         case CPU_DYING:
1912                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1913                        cpu);
1914                 hardware_disable(NULL);
1915                 break;
1916         case CPU_UP_CANCELED:
1917                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1918                        cpu);
1919                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1920                 break;
1921         case CPU_ONLINE:
1922                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1923                        cpu);
1924                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1925                 break;
1926         }
1927         return NOTIFY_OK;
1928 }
1929
1930
1931 asmlinkage void kvm_handle_fault_on_reboot(void)
1932 {
1933         if (kvm_rebooting)
1934                 /* spin while reset goes on */
1935                 while (true)
1936                         ;
1937         /* Fault while not rebooting.  We want the trace. */
1938         BUG();
1939 }
1940 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1941
1942 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1943                       void *v)
1944 {
1945         /*
1946          * Some (well, at least mine) BIOSes hang on reboot if
1947          * in vmx root mode.
1948          *
1949          * And Intel TXT required VMX off for all cpu when system shutdown.
1950          */
1951         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1952         kvm_rebooting = true;
1953         on_each_cpu(hardware_disable, NULL, 1);
1954         return NOTIFY_OK;
1955 }
1956
1957 static struct notifier_block kvm_reboot_notifier = {
1958         .notifier_call = kvm_reboot,
1959         .priority = 0,
1960 };
1961
1962 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1963 {
1964         int i;
1965
1966         for (i = 0; i < bus->dev_count; i++) {
1967                 struct kvm_io_device *pos = bus->devs[i];
1968
1969                 kvm_iodevice_destructor(pos);
1970         }
1971         kfree(bus);
1972 }
1973
1974 /* kvm_io_bus_write - called under kvm->slots_lock */
1975 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
1976                      int len, const void *val)
1977 {
1978         int i;
1979         struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
1980         for (i = 0; i < bus->dev_count; i++)
1981                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
1982                         return 0;
1983         return -EOPNOTSUPP;
1984 }
1985
1986 /* kvm_io_bus_read - called under kvm->slots_lock */
1987 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
1988                     int len, void *val)
1989 {
1990         int i;
1991         struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
1992
1993         for (i = 0; i < bus->dev_count; i++)
1994                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
1995                         return 0;
1996         return -EOPNOTSUPP;
1997 }
1998
1999 /* Caller must hold slots_lock. */
2000 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2001                             struct kvm_io_device *dev)
2002 {
2003         struct kvm_io_bus *new_bus, *bus;
2004
2005         bus = kvm->buses[bus_idx];
2006         if (bus->dev_count > NR_IOBUS_DEVS-1)
2007                 return -ENOSPC;
2008
2009         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2010         if (!new_bus)
2011                 return -ENOMEM;
2012         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2013         new_bus->devs[new_bus->dev_count++] = dev;
2014         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2015         synchronize_srcu_expedited(&kvm->srcu);
2016         kfree(bus);
2017
2018         return 0;
2019 }
2020
2021 /* Caller must hold slots_lock. */
2022 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2023                               struct kvm_io_device *dev)
2024 {
2025         int i, r;
2026         struct kvm_io_bus *new_bus, *bus;
2027
2028         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2029         if (!new_bus)
2030                 return -ENOMEM;
2031
2032         bus = kvm->buses[bus_idx];
2033         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2034
2035         r = -ENOENT;
2036         for (i = 0; i < new_bus->dev_count; i++)
2037                 if (new_bus->devs[i] == dev) {
2038                         r = 0;
2039                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2040                         break;
2041                 }
2042
2043         if (r) {
2044                 kfree(new_bus);
2045                 return r;
2046         }
2047
2048         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2049         synchronize_srcu_expedited(&kvm->srcu);
2050         kfree(bus);
2051         return r;
2052 }
2053
2054 static struct notifier_block kvm_cpu_notifier = {
2055         .notifier_call = kvm_cpu_hotplug,
2056         .priority = 20, /* must be > scheduler priority */
2057 };
2058
2059 static int vm_stat_get(void *_offset, u64 *val)
2060 {
2061         unsigned offset = (long)_offset;
2062         struct kvm *kvm;
2063
2064         *val = 0;
2065         spin_lock(&kvm_lock);
2066         list_for_each_entry(kvm, &vm_list, vm_list)
2067                 *val += *(u32 *)((void *)kvm + offset);
2068         spin_unlock(&kvm_lock);
2069         return 0;
2070 }
2071
2072 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2073
2074 static int vcpu_stat_get(void *_offset, u64 *val)
2075 {
2076         unsigned offset = (long)_offset;
2077         struct kvm *kvm;
2078         struct kvm_vcpu *vcpu;
2079         int i;
2080
2081         *val = 0;
2082         spin_lock(&kvm_lock);
2083         list_for_each_entry(kvm, &vm_list, vm_list)
2084                 kvm_for_each_vcpu(i, vcpu, kvm)
2085                         *val += *(u32 *)((void *)vcpu + offset);
2086
2087         spin_unlock(&kvm_lock);
2088         return 0;
2089 }
2090
2091 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2092
2093 static const struct file_operations *stat_fops[] = {
2094         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2095         [KVM_STAT_VM]   = &vm_stat_fops,
2096 };
2097
2098 static void kvm_init_debug(void)
2099 {
2100         struct kvm_stats_debugfs_item *p;
2101
2102         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2103         for (p = debugfs_entries; p->name; ++p)
2104                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2105                                                 (void *)(long)p->offset,
2106                                                 stat_fops[p->kind]);
2107 }
2108
2109 static void kvm_exit_debug(void)
2110 {
2111         struct kvm_stats_debugfs_item *p;
2112
2113         for (p = debugfs_entries; p->name; ++p)
2114                 debugfs_remove(p->dentry);
2115         debugfs_remove(kvm_debugfs_dir);
2116 }
2117
2118 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2119 {
2120         if (kvm_usage_count)
2121                 hardware_disable(NULL);
2122         return 0;
2123 }
2124
2125 static int kvm_resume(struct sys_device *dev)
2126 {
2127         if (kvm_usage_count)
2128                 hardware_enable(NULL);
2129         return 0;
2130 }
2131
2132 static struct sysdev_class kvm_sysdev_class = {
2133         .name = "kvm",
2134         .suspend = kvm_suspend,
2135         .resume = kvm_resume,
2136 };
2137
2138 static struct sys_device kvm_sysdev = {
2139         .id = 0,
2140         .cls = &kvm_sysdev_class,
2141 };
2142
2143 struct page *bad_page;
2144 pfn_t bad_pfn;
2145
2146 static inline
2147 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2148 {
2149         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2150 }
2151
2152 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2153 {
2154         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2155
2156         kvm_arch_vcpu_load(vcpu, cpu);
2157 }
2158
2159 static void kvm_sched_out(struct preempt_notifier *pn,
2160                           struct task_struct *next)
2161 {
2162         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2163
2164         kvm_arch_vcpu_put(vcpu);
2165 }
2166
2167 int kvm_init(void *opaque, unsigned int vcpu_size,
2168                   struct module *module)
2169 {
2170         int r;
2171         int cpu;
2172
2173         r = kvm_arch_init(opaque);
2174         if (r)
2175                 goto out_fail;
2176
2177         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2178
2179         if (bad_page == NULL) {
2180                 r = -ENOMEM;
2181                 goto out;
2182         }
2183
2184         bad_pfn = page_to_pfn(bad_page);
2185
2186         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2187                 r = -ENOMEM;
2188                 goto out_free_0;
2189         }
2190
2191         r = kvm_arch_hardware_setup();
2192         if (r < 0)
2193                 goto out_free_0a;
2194
2195         for_each_online_cpu(cpu) {
2196                 smp_call_function_single(cpu,
2197                                 kvm_arch_check_processor_compat,
2198                                 &r, 1);
2199                 if (r < 0)
2200                         goto out_free_1;
2201         }
2202
2203         r = register_cpu_notifier(&kvm_cpu_notifier);
2204         if (r)
2205                 goto out_free_2;
2206         register_reboot_notifier(&kvm_reboot_notifier);
2207
2208         r = sysdev_class_register(&kvm_sysdev_class);
2209         if (r)
2210                 goto out_free_3;
2211
2212         r = sysdev_register(&kvm_sysdev);
2213         if (r)
2214                 goto out_free_4;
2215
2216         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2217         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2218                                            __alignof__(struct kvm_vcpu),
2219                                            0, NULL);
2220         if (!kvm_vcpu_cache) {
2221                 r = -ENOMEM;
2222                 goto out_free_5;
2223         }
2224
2225         kvm_chardev_ops.owner = module;
2226         kvm_vm_fops.owner = module;
2227         kvm_vcpu_fops.owner = module;
2228
2229         r = misc_register(&kvm_dev);
2230         if (r) {
2231                 printk(KERN_ERR "kvm: misc device register failed\n");
2232                 goto out_free;
2233         }
2234
2235         kvm_preempt_ops.sched_in = kvm_sched_in;
2236         kvm_preempt_ops.sched_out = kvm_sched_out;
2237
2238         kvm_init_debug();
2239
2240         return 0;
2241
2242 out_free:
2243         kmem_cache_destroy(kvm_vcpu_cache);
2244 out_free_5:
2245         sysdev_unregister(&kvm_sysdev);
2246 out_free_4:
2247         sysdev_class_unregister(&kvm_sysdev_class);
2248 out_free_3:
2249         unregister_reboot_notifier(&kvm_reboot_notifier);
2250         unregister_cpu_notifier(&kvm_cpu_notifier);
2251 out_free_2:
2252 out_free_1:
2253         kvm_arch_hardware_unsetup();
2254 out_free_0a:
2255         free_cpumask_var(cpus_hardware_enabled);
2256 out_free_0:
2257         __free_page(bad_page);
2258 out:
2259         kvm_arch_exit();
2260 out_fail:
2261         return r;
2262 }
2263 EXPORT_SYMBOL_GPL(kvm_init);
2264
2265 void kvm_exit(void)
2266 {
2267         tracepoint_synchronize_unregister();
2268         kvm_exit_debug();
2269         misc_deregister(&kvm_dev);
2270         kmem_cache_destroy(kvm_vcpu_cache);
2271         sysdev_unregister(&kvm_sysdev);
2272         sysdev_class_unregister(&kvm_sysdev_class);
2273         unregister_reboot_notifier(&kvm_reboot_notifier);
2274         unregister_cpu_notifier(&kvm_cpu_notifier);
2275         on_each_cpu(hardware_disable, NULL, 1);
2276         kvm_arch_hardware_unsetup();
2277         kvm_arch_exit();
2278         free_cpumask_var(cpus_hardware_enabled);
2279         __free_page(bad_page);
2280 }
2281 EXPORT_SYMBOL_GPL(kvm_exit);