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 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
  55 #include <linux/pci.h>
  56 #include <linux/interrupt.h>
  57 #include "irq.h"
  58 #endif
  59
  60 MODULE_AUTHOR("Qumranet");
  61 MODULE_LICENSE("GPL");
  62
  63 DEFINE_SPINLOCK(kvm_lock);
  64 LIST_HEAD(vm_list);
  65
  66 static cpumask_t cpus_hardware_enabled;
  67
  68 struct kmem_cache *kvm_vcpu_cache;
  69 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  70
  71 static __read_mostly struct preempt_ops kvm_preempt_ops;
  72
  73 struct dentry *kvm_debugfs_dir;
  74
  75 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  76                            unsigned long arg);
  77
  78 bool kvm_rebooting;
  79
  80 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
  81 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
  82                                                       int assigned_dev_id)
  83 {
  84         struct list_head *ptr;
  85         struct kvm_assigned_dev_kernel *match;
  86
  87         list_for_each(ptr, head) {
  88                 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
  89                 if (match->assigned_dev_id == assigned_dev_id)
  90                         return match;
  91         }
  92         return NULL;
  93 }
  94
  95 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
  96 {
  97         struct kvm_assigned_dev_kernel *assigned_dev;
  98
  99         assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
 100                                     interrupt_work);
 101
 102         /* This is taken to safely inject irq inside the guest. When
 103          * the interrupt injection (or the ioapic code) uses a
 104          * finer-grained lock, update this
 105          */
 106         mutex_lock(&assigned_dev->kvm->lock);
 107         kvm_set_irq(assigned_dev->kvm,
 108                     assigned_dev->guest_irq, 1);
 109         mutex_unlock(&assigned_dev->kvm->lock);
 110         kvm_put_kvm(assigned_dev->kvm);
 111 }
 112
 113 /* FIXME: Implement the OR logic needed to make shared interrupts on
 114  * this line behave properly
 115  */
 116 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
 117 {
 118         struct kvm_assigned_dev_kernel *assigned_dev =
 119                 (struct kvm_assigned_dev_kernel *) dev_id;
 120
 121         kvm_get_kvm(assigned_dev->kvm);
 122         schedule_work(&assigned_dev->interrupt_work);
 123         disable_irq_nosync(irq);
 124         return IRQ_HANDLED;
 125 }
 126
 127 /* Ack the irq line for an assigned device */
 128 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
 129 {
 130         struct kvm_assigned_dev_kernel *dev;
 131
 132         if (kian->gsi == -1)
 133                 return;
 134
 135         dev = container_of(kian, struct kvm_assigned_dev_kernel,
 136                            ack_notifier);
 137         kvm_set_irq(dev->kvm, dev->guest_irq, 0);
 138         enable_irq(dev->host_irq);
 139 }
 140
 141 static void kvm_free_assigned_device(struct kvm *kvm,
 142                                      struct kvm_assigned_dev_kernel
 143                                      *assigned_dev)
 144 {
 145         if (irqchip_in_kernel(kvm) && assigned_dev->irq_requested)
 146                 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
 147
 148         kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
 149
 150         if (cancel_work_sync(&assigned_dev->interrupt_work))
 151                 /* We had pending work. That means we will have to take
 152                  * care of kvm_put_kvm.
 153                  */
 154                 kvm_put_kvm(kvm);
 155
 156         pci_release_regions(assigned_dev->dev);
 157         pci_disable_device(assigned_dev->dev);
 158         pci_dev_put(assigned_dev->dev);
 159
 160         list_del(&assigned_dev->list);
 161         kfree(assigned_dev);
 162 }
 163
 164 void kvm_free_all_assigned_devices(struct kvm *kvm)
 165 {
 166         struct list_head *ptr, *ptr2;
 167         struct kvm_assigned_dev_kernel *assigned_dev;
 168
 169         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
 170                 assigned_dev = list_entry(ptr,
 171                                           struct kvm_assigned_dev_kernel,
 172                                           list);
 173
 174                 kvm_free_assigned_device(kvm, assigned_dev);
 175         }
 176 }
 177
 178 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 179                                    struct kvm_assigned_irq
 180                                    *assigned_irq)
 181 {
 182         int r = 0;
 183         struct kvm_assigned_dev_kernel *match;
 184
 185         mutex_lock(&kvm->lock);
 186
 187         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 188                                       assigned_irq->assigned_dev_id);
 189         if (!match) {
 190                 mutex_unlock(&kvm->lock);
 191                 return -EINVAL;
 192         }
 193
 194         if (match->irq_requested) {
 195                 match->guest_irq = assigned_irq->guest_irq;
 196                 match->ack_notifier.gsi = assigned_irq->guest_irq;
 197                 mutex_unlock(&kvm->lock);
 198                 return 0;
 199         }
 200
 201         INIT_WORK(&match->interrupt_work,
 202                   kvm_assigned_dev_interrupt_work_handler);
 203
 204         if (irqchip_in_kernel(kvm)) {
 205                 if (!capable(CAP_SYS_RAWIO)) {
 206                         r = -EPERM;
 207                         goto out_release;
 208                 }
 209
 210                 if (assigned_irq->host_irq)
 211                         match->host_irq = assigned_irq->host_irq;
 212                 else
 213                         match->host_irq = match->dev->irq;
 214                 match->guest_irq = assigned_irq->guest_irq;
 215                 match->ack_notifier.gsi = assigned_irq->guest_irq;
 216                 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
 217                 kvm_register_irq_ack_notifier(kvm, &match->ack_notifier);
 218
 219                 /* Even though this is PCI, we don't want to use shared
 220                  * interrupts. Sharing host devices with guest-assigned devices
 221                  * on the same interrupt line is not a happy situation: there
 222                  * are going to be long delays in accepting, acking, etc.
 223                  */
 224                 if (request_irq(match->host_irq, kvm_assigned_dev_intr, 0,
 225                                 "kvm_assigned_device", (void *)match)) {
 226                         r = -EIO;
 227                         goto out_release;
 228                 }
 229         }
 230
 231         match->irq_requested = true;
 232         mutex_unlock(&kvm->lock);
 233         return r;
 234 out_release:
 235         mutex_unlock(&kvm->lock);
 236         kvm_free_assigned_device(kvm, match);
 237         return r;
 238 }
 239
 240 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 241                                       struct kvm_assigned_pci_dev *assigned_dev)
 242 {
 243         int r = 0;
 244         struct kvm_assigned_dev_kernel *match;
 245         struct pci_dev *dev;
 246
 247         mutex_lock(&kvm->lock);
 248
 249         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 250                                       assigned_dev->assigned_dev_id);
 251         if (match) {
 252                 /* device already assigned */
 253                 r = -EINVAL;
 254                 goto out;
 255         }
 256
 257         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 258         if (match == NULL) {
 259                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 260                        __func__);
 261                 r = -ENOMEM;
 262                 goto out;
 263         }
 264         dev = pci_get_bus_and_slot(assigned_dev->busnr,
 265                                    assigned_dev->devfn);
 266         if (!dev) {
 267                 printk(KERN_INFO "%s: host device not found\n", __func__);
 268                 r = -EINVAL;
 269                 goto out_free;
 270         }
 271         if (pci_enable_device(dev)) {
 272                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 273                 r = -EBUSY;
 274                 goto out_put;
 275         }
 276         r = pci_request_regions(dev, "kvm_assigned_device");
 277         if (r) {
 278                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 279                        __func__);
 280                 goto out_disable;
 281         }
 282         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 283         match->host_busnr = assigned_dev->busnr;
 284         match->host_devfn = assigned_dev->devfn;
 285         match->dev = dev;
 286
 287         match->kvm = kvm;
 288
 289         list_add(&match->list, &kvm->arch.assigned_dev_head);
 290
 291         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
 292                 r = kvm_iommu_map_guest(kvm, match);
 293                 if (r)
 294                         goto out_list_del;
 295         }
 296
 297 out:
 298         mutex_unlock(&kvm->lock);
 299         return r;
 300 out_list_del:
 301         list_del(&match->list);
 302         pci_release_regions(dev);
 303 out_disable:
 304         pci_disable_device(dev);
 305 out_put:
 306         pci_dev_put(dev);
 307 out_free:
 308         kfree(match);
 309         mutex_unlock(&kvm->lock);
 310         return r;
 311 }
 312 #endif
 313
 314 static inline int valid_vcpu(int n)
 315 {
 316         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 317 }
 318
 319 inline int kvm_is_mmio_pfn(pfn_t pfn)
 320 {
 321         if (pfn_valid(pfn))
 322                 return PageReserved(pfn_to_page(pfn));
 323
 324         return true;
 325 }
 326
 327 /*
 328  * Switches to specified vcpu, until a matching vcpu_put()
 329  */
 330 void vcpu_load(struct kvm_vcpu *vcpu)
 331 {
 332         int cpu;
 333
 334         mutex_lock(&vcpu->mutex);
 335         cpu = get_cpu();
 336         preempt_notifier_register(&vcpu->preempt_notifier);
 337         kvm_arch_vcpu_load(vcpu, cpu);
 338         put_cpu();
 339 }
 340
 341 void vcpu_put(struct kvm_vcpu *vcpu)
 342 {
 343         preempt_disable();
 344         kvm_arch_vcpu_put(vcpu);
 345         preempt_notifier_unregister(&vcpu->preempt_notifier);
 346         preempt_enable();
 347         mutex_unlock(&vcpu->mutex);
 348 }
 349
 350 static void ack_flush(void *_completed)
 351 {
 352 }
 353
 354 void kvm_flush_remote_tlbs(struct kvm *kvm)
 355 {
 356         int i, cpu, me;
 357         cpumask_t cpus;
 358         struct kvm_vcpu *vcpu;
 359
 360         me = get_cpu();
 361         cpus_clear(cpus);
 362         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 363                 vcpu = kvm->vcpus[i];
 364                 if (!vcpu)
 365                         continue;
 366                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 367                         continue;
 368                 cpu = vcpu->cpu;
 369                 if (cpu != -1 && cpu != me)
 370                         cpu_set(cpu, cpus);
 371         }
 372         if (cpus_empty(cpus))
 373                 goto out;
 374         ++kvm->stat.remote_tlb_flush;
 375         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 376 out:
 377         put_cpu();
 378 }
 379
 380 void kvm_reload_remote_mmus(struct kvm *kvm)
 381 {
 382         int i, cpu, me;
 383         cpumask_t cpus;
 384         struct kvm_vcpu *vcpu;
 385
 386         me = get_cpu();
 387         cpus_clear(cpus);
 388         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 389                 vcpu = kvm->vcpus[i];
 390                 if (!vcpu)
 391                         continue;
 392                 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
 393                         continue;
 394                 cpu = vcpu->cpu;
 395                 if (cpu != -1 && cpu != me)
 396                         cpu_set(cpu, cpus);
 397         }
 398         if (cpus_empty(cpus))
 399                 goto out;
 400         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 401 out:
 402         put_cpu();
 403 }
 404
 405
 406 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 407 {
 408         struct page *page;
 409         int r;
 410
 411         mutex_init(&vcpu->mutex);
 412         vcpu->cpu = -1;
 413         vcpu->kvm = kvm;
 414         vcpu->vcpu_id = id;
 415         init_waitqueue_head(&vcpu->wq);
 416
 417         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 418         if (!page) {
 419                 r = -ENOMEM;
 420                 goto fail;
 421         }
 422         vcpu->run = page_address(page);
 423
 424         r = kvm_arch_vcpu_init(vcpu);
 425         if (r < 0)
 426                 goto fail_free_run;
 427         return 0;
 428
 429 fail_free_run:
 430         free_page((unsigned long)vcpu->run);
 431 fail:
 432         return r;
 433 }
 434 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 435
 436 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 437 {
 438         kvm_arch_vcpu_uninit(vcpu);
 439         free_page((unsigned long)vcpu->run);
 440 }
 441 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 442
 443 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 444 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 445 {
 446         return container_of(mn, struct kvm, mmu_notifier);
 447 }
 448
 449 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 450                                              struct mm_struct *mm,
 451                                              unsigned long address)
 452 {
 453         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 454         int need_tlb_flush;
 455
 456         /*
 457          * When ->invalidate_page runs, the linux pte has been zapped
 458          * already but the page is still allocated until
 459          * ->invalidate_page returns. So if we increase the sequence
 460          * here the kvm page fault will notice if the spte can't be
 461          * established because the page is going to be freed. If
 462          * instead the kvm page fault establishes the spte before
 463          * ->invalidate_page runs, kvm_unmap_hva will release it
 464          * before returning.
 465          *
 466          * The sequence increase only need to be seen at spin_unlock
 467          * time, and not at spin_lock time.
 468          *
 469          * Increasing the sequence after the spin_unlock would be
 470          * unsafe because the kvm page fault could then establish the
 471          * pte after kvm_unmap_hva returned, without noticing the page
 472          * is going to be freed.
 473          */
 474         spin_lock(&kvm->mmu_lock);
 475         kvm->mmu_notifier_seq++;
 476         need_tlb_flush = kvm_unmap_hva(kvm, address);
 477         spin_unlock(&kvm->mmu_lock);
 478
 479         /* we've to flush the tlb before the pages can be freed */
 480         if (need_tlb_flush)
 481                 kvm_flush_remote_tlbs(kvm);
 482
 483 }
 484
 485 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 486                                                     struct mm_struct *mm,
 487                                                     unsigned long start,
 488                                                     unsigned long end)
 489 {
 490         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 491         int need_tlb_flush = 0;
 492
 493         spin_lock(&kvm->mmu_lock);
 494         /*
 495          * The count increase must become visible at unlock time as no
 496          * spte can be established without taking the mmu_lock and
 497          * count is also read inside the mmu_lock critical section.
 498          */
 499         kvm->mmu_notifier_count++;
 500         for (; start < end; start += PAGE_SIZE)
 501                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 502         spin_unlock(&kvm->mmu_lock);
 503
 504         /* we've to flush the tlb before the pages can be freed */
 505         if (need_tlb_flush)
 506                 kvm_flush_remote_tlbs(kvm);
 507 }
 508
 509 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 510                                                   struct mm_struct *mm,
 511                                                   unsigned long start,
 512                                                   unsigned long end)
 513 {
 514         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 515
 516         spin_lock(&kvm->mmu_lock);
 517         /*
 518          * This sequence increase will notify the kvm page fault that
 519          * the page that is going to be mapped in the spte could have
 520          * been freed.
 521          */
 522         kvm->mmu_notifier_seq++;
 523         /*
 524          * The above sequence increase must be visible before the
 525          * below count decrease but both values are read by the kvm
 526          * page fault under mmu_lock spinlock so we don't need to add
 527          * a smb_wmb() here in between the two.
 528          */
 529         kvm->mmu_notifier_count--;
 530         spin_unlock(&kvm->mmu_lock);
 531
 532         BUG_ON(kvm->mmu_notifier_count < 0);
 533 }
 534
 535 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 536                                               struct mm_struct *mm,
 537                                               unsigned long address)
 538 {
 539         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 540         int young;
 541
 542         spin_lock(&kvm->mmu_lock);
 543         young = kvm_age_hva(kvm, address);
 544         spin_unlock(&kvm->mmu_lock);
 545
 546         if (young)
 547                 kvm_flush_remote_tlbs(kvm);
 548
 549         return young;
 550 }
 551
 552 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 553         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 554         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 555         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 556         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 557 };
 558 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 559
 560 static struct kvm *kvm_create_vm(void)
 561 {
 562         struct kvm *kvm = kvm_arch_create_vm();
 563 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 564         struct page *page;
 565 #endif
 566
 567         if (IS_ERR(kvm))
 568                 goto out;
 569
 570 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 571         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 572         if (!page) {
 573                 kfree(kvm);
 574                 return ERR_PTR(-ENOMEM);
 575         }
 576         kvm->coalesced_mmio_ring =
 577                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 578 #endif
 579
 580 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 581         {
 582                 int err;
 583                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 584                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 585                 if (err) {
 586 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 587                         put_page(page);
 588 #endif
 589                         kfree(kvm);
 590                         return ERR_PTR(err);
 591                 }
 592         }
 593 #endif
 594
 595         kvm->mm = current->mm;
 596         atomic_inc(&kvm->mm->mm_count);
 597         spin_lock_init(&kvm->mmu_lock);
 598         kvm_io_bus_init(&kvm->pio_bus);
 599         mutex_init(&kvm->lock);
 600         kvm_io_bus_init(&kvm->mmio_bus);
 601         init_rwsem(&kvm->slots_lock);
 602         atomic_set(&kvm->users_count, 1);
 603         spin_lock(&kvm_lock);
 604         list_add(&kvm->vm_list, &vm_list);
 605         spin_unlock(&kvm_lock);
 606 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 607         kvm_coalesced_mmio_init(kvm);
 608 #endif
 609 out:
 610         return kvm;
 611 }
 612
 613 /*
 614  * Free any memory in @free but not in @dont.
 615  */
 616 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 617                                   struct kvm_memory_slot *dont)
 618 {
 619         if (!dont || free->rmap != dont->rmap)
 620                 vfree(free->rmap);
 621
 622         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 623                 vfree(free->dirty_bitmap);
 624
 625         if (!dont || free->lpage_info != dont->lpage_info)
 626                 vfree(free->lpage_info);
 627
 628         free->npages = 0;
 629         free->dirty_bitmap = NULL;
 630         free->rmap = NULL;
 631         free->lpage_info = NULL;
 632 }
 633
 634 void kvm_free_physmem(struct kvm *kvm)
 635 {
 636         int i;
 637
 638         for (i = 0; i < kvm->nmemslots; ++i)
 639                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 640 }
 641
 642 static void kvm_destroy_vm(struct kvm *kvm)
 643 {
 644         struct mm_struct *mm = kvm->mm;
 645
 646         spin_lock(&kvm_lock);
 647         list_del(&kvm->vm_list);
 648         spin_unlock(&kvm_lock);
 649         kvm_io_bus_destroy(&kvm->pio_bus);
 650         kvm_io_bus_destroy(&kvm->mmio_bus);
 651 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 652         if (kvm->coalesced_mmio_ring != NULL)
 653                 free_page((unsigned long)kvm->coalesced_mmio_ring);
 654 #endif
 655 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 656         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 657 #endif
 658         kvm_arch_destroy_vm(kvm);
 659         mmdrop(mm);
 660 }
 661
 662 void kvm_get_kvm(struct kvm *kvm)
 663 {
 664         atomic_inc(&kvm->users_count);
 665 }
 666 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 667
 668 void kvm_put_kvm(struct kvm *kvm)
 669 {
 670         if (atomic_dec_and_test(&kvm->users_count))
 671                 kvm_destroy_vm(kvm);
 672 }
 673 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 674
 675
 676 static int kvm_vm_release(struct inode *inode, struct file *filp)
 677 {
 678         struct kvm *kvm = filp->private_data;
 679
 680         kvm_put_kvm(kvm);
 681         return 0;
 682 }
 683
 684 /*
 685  * Allocate some memory and give it an address in the guest physical address
 686  * space.
 687  *
 688  * Discontiguous memory is allowed, mostly for framebuffers.
 689  *
 690  * Must be called holding mmap_sem for write.
 691  */
 692 int __kvm_set_memory_region(struct kvm *kvm,
 693                             struct kvm_userspace_memory_region *mem,
 694                             int user_alloc)
 695 {
 696         int r;
 697         gfn_t base_gfn;
 698         unsigned long npages;
 699         unsigned long i;
 700         struct kvm_memory_slot *memslot;
 701         struct kvm_memory_slot old, new;
 702
 703         r = -EINVAL;
 704         /* General sanity checks */
 705         if (mem->memory_size & (PAGE_SIZE - 1))
 706                 goto out;
 707         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 708                 goto out;
 709         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 710                 goto out;
 711         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 712                 goto out;
 713
 714         memslot = &kvm->memslots[mem->slot];
 715         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 716         npages = mem->memory_size >> PAGE_SHIFT;
 717
 718         if (!npages)
 719                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 720
 721         new = old = *memslot;
 722
 723         new.base_gfn = base_gfn;
 724         new.npages = npages;
 725         new.flags = mem->flags;
 726
 727         /* Disallow changing a memory slot's size. */
 728         r = -EINVAL;
 729         if (npages && old.npages && npages != old.npages)
 730                 goto out_free;
 731
 732         /* Check for overlaps */
 733         r = -EEXIST;
 734         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 735                 struct kvm_memory_slot *s = &kvm->memslots[i];
 736
 737                 if (s == memslot)
 738                         continue;
 739                 if (!((base_gfn + npages <= s->base_gfn) ||
 740                       (base_gfn >= s->base_gfn + s->npages)))
 741                         goto out_free;
 742         }
 743
 744         /* Free page dirty bitmap if unneeded */
 745         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 746                 new.dirty_bitmap = NULL;
 747
 748         r = -ENOMEM;
 749
 750         /* Allocate if a slot is being created */
 751 #ifndef CONFIG_S390
 752         if (npages && !new.rmap) {
 753                 new.rmap = vmalloc(npages * sizeof(struct page *));
 754
 755                 if (!new.rmap)
 756                         goto out_free;
 757
 758                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 759
 760                 new.user_alloc = user_alloc;
 761                 /*
 762                  * hva_to_rmmap() serialzies with the mmu_lock and to be
 763                  * safe it has to ignore memslots with !user_alloc &&
 764                  * !userspace_addr.
 765                  */
 766                 if (user_alloc)
 767                         new.userspace_addr = mem->userspace_addr;
 768                 else
 769                         new.userspace_addr = 0;
 770         }
 771         if (npages && !new.lpage_info) {
 772                 int largepages = npages / KVM_PAGES_PER_HPAGE;
 773                 if (npages % KVM_PAGES_PER_HPAGE)
 774                         largepages++;
 775                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 776                         largepages++;
 777
 778                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
 779
 780                 if (!new.lpage_info)
 781                         goto out_free;
 782
 783                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
 784
 785                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 786                         new.lpage_info[0].write_count = 1;
 787                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
 788                         new.lpage_info[largepages-1].write_count = 1;
 789         }
 790
 791         /* Allocate page dirty bitmap if needed */
 792         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 793                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 794
 795                 new.dirty_bitmap = vmalloc(dirty_bytes);
 796                 if (!new.dirty_bitmap)
 797                         goto out_free;
 798                 memset(new.dirty_bitmap, 0, dirty_bytes);
 799         }
 800 #endif /* not defined CONFIG_S390 */
 801
 802         if (!npages)
 803                 kvm_arch_flush_shadow(kvm);
 804
 805         spin_lock(&kvm->mmu_lock);
 806         if (mem->slot >= kvm->nmemslots)
 807                 kvm->nmemslots = mem->slot + 1;
 808
 809         *memslot = new;
 810         spin_unlock(&kvm->mmu_lock);
 811
 812         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
 813         if (r) {
 814                 spin_lock(&kvm->mmu_lock);
 815                 *memslot = old;
 816                 spin_unlock(&kvm->mmu_lock);
 817                 goto out_free;
 818         }
 819
 820         kvm_free_physmem_slot(&old, &new);
 821 #ifdef CONFIG_DMAR
 822         /* map the pages in iommu page table */
 823         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
 824         if (r)
 825                 goto out;
 826 #endif
 827         return 0;
 828
 829 out_free:
 830         kvm_free_physmem_slot(&new, &old);
 831 out:
 832         return r;
 833
 834 }
 835 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 836
 837 int kvm_set_memory_region(struct kvm *kvm,
 838                           struct kvm_userspace_memory_region *mem,
 839                           int user_alloc)
 840 {
 841         int r;
 842
 843         down_write(&kvm->slots_lock);
 844         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 845         up_write(&kvm->slots_lock);
 846         return r;
 847 }
 848 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 849
 850 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 851                                    struct
 852                                    kvm_userspace_memory_region *mem,
 853                                    int user_alloc)
 854 {
 855         if (mem->slot >= KVM_MEMORY_SLOTS)
 856                 return -EINVAL;
 857         return kvm_set_memory_region(kvm, mem, user_alloc);
 858 }
 859
 860 int kvm_get_dirty_log(struct kvm *kvm,
 861                         struct kvm_dirty_log *log, int *is_dirty)
 862 {
 863         struct kvm_memory_slot *memslot;
 864         int r, i;
 865         int n;
 866         unsigned long any = 0;
 867
 868         r = -EINVAL;
 869         if (log->slot >= KVM_MEMORY_SLOTS)
 870                 goto out;
 871
 872         memslot = &kvm->memslots[log->slot];
 873         r = -ENOENT;
 874         if (!memslot->dirty_bitmap)
 875                 goto out;
 876
 877         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 878
 879         for (i = 0; !any && i < n/sizeof(long); ++i)
 880                 any = memslot->dirty_bitmap[i];
 881
 882         r = -EFAULT;
 883         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 884                 goto out;
 885
 886         if (any)
 887                 *is_dirty = 1;
 888
 889         r = 0;
 890 out:
 891         return r;
 892 }
 893
 894 int is_error_page(struct page *page)
 895 {
 896         return page == bad_page;
 897 }
 898 EXPORT_SYMBOL_GPL(is_error_page);
 899
 900 int is_error_pfn(pfn_t pfn)
 901 {
 902         return pfn == bad_pfn;
 903 }
 904 EXPORT_SYMBOL_GPL(is_error_pfn);
 905
 906 static inline unsigned long bad_hva(void)
 907 {
 908         return PAGE_OFFSET;
 909 }
 910
 911 int kvm_is_error_hva(unsigned long addr)
 912 {
 913         return addr == bad_hva();
 914 }
 915 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 916
 917 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 918 {
 919         int i;
 920
 921         for (i = 0; i < kvm->nmemslots; ++i) {
 922                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 923
 924                 if (gfn >= memslot->base_gfn
 925                     && gfn < memslot->base_gfn + memslot->npages)
 926                         return memslot;
 927         }
 928         return NULL;
 929 }
 930
 931 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 932 {
 933         gfn = unalias_gfn(kvm, gfn);
 934         return __gfn_to_memslot(kvm, gfn);
 935 }
 936
 937 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 938 {
 939         int i;
 940
 941         gfn = unalias_gfn(kvm, gfn);
 942         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 943                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 944
 945                 if (gfn >= memslot->base_gfn
 946                     && gfn < memslot->base_gfn + memslot->npages)
 947                         return 1;
 948         }
 949         return 0;
 950 }
 951 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 952
 953 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 954 {
 955         struct kvm_memory_slot *slot;
 956
 957         gfn = unalias_gfn(kvm, gfn);
 958         slot = __gfn_to_memslot(kvm, gfn);
 959         if (!slot)
 960                 return bad_hva();
 961         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 962 }
 963 EXPORT_SYMBOL_GPL(gfn_to_hva);
 964
 965 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 966 {
 967         struct page *page[1];
 968         unsigned long addr;
 969         int npages;
 970         pfn_t pfn;
 971
 972         might_sleep();
 973
 974         addr = gfn_to_hva(kvm, gfn);
 975         if (kvm_is_error_hva(addr)) {
 976                 get_page(bad_page);
 977                 return page_to_pfn(bad_page);
 978         }
 979
 980         npages = get_user_pages_fast(addr, 1, 1, page);
 981
 982         if (unlikely(npages != 1)) {
 983                 struct vm_area_struct *vma;
 984
 985                 down_read(&current->mm->mmap_sem);
 986                 vma = find_vma(current->mm, addr);
 987
 988                 if (vma == NULL || addr < vma->vm_start ||
 989                     !(vma->vm_flags & VM_PFNMAP)) {
 990                         up_read(&current->mm->mmap_sem);
 991                         get_page(bad_page);
 992                         return page_to_pfn(bad_page);
 993                 }
 994
 995                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 996                 up_read(&current->mm->mmap_sem);
 997                 BUG_ON(!kvm_is_mmio_pfn(pfn));
 998         } else
 999                 pfn = page_to_pfn(page[0]);
1000
1001         return pfn;
1002 }
1003
1004 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1005
1006 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1007 {
1008         pfn_t pfn;
1009
1010         pfn = gfn_to_pfn(kvm, gfn);
1011         if (!kvm_is_mmio_pfn(pfn))
1012                 return pfn_to_page(pfn);
1013
1014         WARN_ON(kvm_is_mmio_pfn(pfn));
1015
1016         get_page(bad_page);
1017         return bad_page;
1018 }
1019
1020 EXPORT_SYMBOL_GPL(gfn_to_page);
1021
1022 void kvm_release_page_clean(struct page *page)
1023 {
1024         kvm_release_pfn_clean(page_to_pfn(page));
1025 }
1026 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1027
1028 void kvm_release_pfn_clean(pfn_t pfn)
1029 {
1030         if (!kvm_is_mmio_pfn(pfn))
1031                 put_page(pfn_to_page(pfn));
1032 }
1033 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1034
1035 void kvm_release_page_dirty(struct page *page)
1036 {
1037         kvm_release_pfn_dirty(page_to_pfn(page));
1038 }
1039 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1040
1041 void kvm_release_pfn_dirty(pfn_t pfn)
1042 {
1043         kvm_set_pfn_dirty(pfn);
1044         kvm_release_pfn_clean(pfn);
1045 }
1046 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1047
1048 void kvm_set_page_dirty(struct page *page)
1049 {
1050         kvm_set_pfn_dirty(page_to_pfn(page));
1051 }
1052 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1053
1054 void kvm_set_pfn_dirty(pfn_t pfn)
1055 {
1056         if (!kvm_is_mmio_pfn(pfn)) {
1057                 struct page *page = pfn_to_page(pfn);
1058                 if (!PageReserved(page))
1059                         SetPageDirty(page);
1060         }
1061 }
1062 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1063
1064 void kvm_set_pfn_accessed(pfn_t pfn)
1065 {
1066         if (!kvm_is_mmio_pfn(pfn))
1067                 mark_page_accessed(pfn_to_page(pfn));
1068 }
1069 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1070
1071 void kvm_get_pfn(pfn_t pfn)
1072 {
1073         if (!kvm_is_mmio_pfn(pfn))
1074                 get_page(pfn_to_page(pfn));
1075 }
1076 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1077
1078 static int next_segment(unsigned long len, int offset)
1079 {
1080         if (len > PAGE_SIZE - offset)
1081                 return PAGE_SIZE - offset;
1082         else
1083                 return len;
1084 }
1085
1086 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1087                         int len)
1088 {
1089         int r;
1090         unsigned long addr;
1091
1092         addr = gfn_to_hva(kvm, gfn);
1093         if (kvm_is_error_hva(addr))
1094                 return -EFAULT;
1095         r = copy_from_user(data, (void __user *)addr + offset, len);
1096         if (r)
1097                 return -EFAULT;
1098         return 0;
1099 }
1100 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1101
1102 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1103 {
1104         gfn_t gfn = gpa >> PAGE_SHIFT;
1105         int seg;
1106         int offset = offset_in_page(gpa);
1107         int ret;
1108
1109         while ((seg = next_segment(len, offset)) != 0) {
1110                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1111                 if (ret < 0)
1112                         return ret;
1113                 offset = 0;
1114                 len -= seg;
1115                 data += seg;
1116                 ++gfn;
1117         }
1118         return 0;
1119 }
1120 EXPORT_SYMBOL_GPL(kvm_read_guest);
1121
1122 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1123                           unsigned long len)
1124 {
1125         int r;
1126         unsigned long addr;
1127         gfn_t gfn = gpa >> PAGE_SHIFT;
1128         int offset = offset_in_page(gpa);
1129
1130         addr = gfn_to_hva(kvm, gfn);
1131         if (kvm_is_error_hva(addr))
1132                 return -EFAULT;
1133         pagefault_disable();
1134         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1135         pagefault_enable();
1136         if (r)
1137                 return -EFAULT;
1138         return 0;
1139 }
1140 EXPORT_SYMBOL(kvm_read_guest_atomic);
1141
1142 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1143                          int offset, int len)
1144 {
1145         int r;
1146         unsigned long addr;
1147
1148         addr = gfn_to_hva(kvm, gfn);
1149         if (kvm_is_error_hva(addr))
1150                 return -EFAULT;
1151         r = copy_to_user((void __user *)addr + offset, data, len);
1152         if (r)
1153                 return -EFAULT;
1154         mark_page_dirty(kvm, gfn);
1155         return 0;
1156 }
1157 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1158
1159 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1160                     unsigned long len)
1161 {
1162         gfn_t gfn = gpa >> PAGE_SHIFT;
1163         int seg;
1164         int offset = offset_in_page(gpa);
1165         int ret;
1166
1167         while ((seg = next_segment(len, offset)) != 0) {
1168                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1169                 if (ret < 0)
1170                         return ret;
1171                 offset = 0;
1172                 len -= seg;
1173                 data += seg;
1174                 ++gfn;
1175         }
1176         return 0;
1177 }
1178
1179 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1180 {
1181         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1182 }
1183 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1184
1185 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1186 {
1187         gfn_t gfn = gpa >> PAGE_SHIFT;
1188         int seg;
1189         int offset = offset_in_page(gpa);
1190         int ret;
1191
1192         while ((seg = next_segment(len, offset)) != 0) {
1193                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1194                 if (ret < 0)
1195                         return ret;
1196                 offset = 0;
1197                 len -= seg;
1198                 ++gfn;
1199         }
1200         return 0;
1201 }
1202 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1203
1204 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1205 {
1206         struct kvm_memory_slot *memslot;
1207
1208         gfn = unalias_gfn(kvm, gfn);
1209         memslot = __gfn_to_memslot(kvm, gfn);
1210         if (memslot && memslot->dirty_bitmap) {
1211                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1212
1213                 /* avoid RMW */
1214                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1215                         set_bit(rel_gfn, memslot->dirty_bitmap);
1216         }
1217 }
1218
1219 /*
1220  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1221  */
1222 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1223 {
1224         DEFINE_WAIT(wait);
1225
1226         for (;;) {
1227                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1228
1229                 if (kvm_cpu_has_interrupt(vcpu) ||
1230                     kvm_cpu_has_pending_timer(vcpu) ||
1231                     kvm_arch_vcpu_runnable(vcpu)) {
1232                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1233                         break;
1234                 }
1235                 if (signal_pending(current))
1236                         break;
1237
1238                 vcpu_put(vcpu);
1239                 schedule();
1240                 vcpu_load(vcpu);
1241         }
1242
1243         finish_wait(&vcpu->wq, &wait);
1244 }
1245
1246 void kvm_resched(struct kvm_vcpu *vcpu)
1247 {
1248         if (!need_resched())
1249                 return;
1250         cond_resched();
1251 }
1252 EXPORT_SYMBOL_GPL(kvm_resched);
1253
1254 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1255 {
1256         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1257         struct page *page;
1258
1259         if (vmf->pgoff == 0)
1260                 page = virt_to_page(vcpu->run);
1261 #ifdef CONFIG_X86
1262         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1263                 page = virt_to_page(vcpu->arch.pio_data);
1264 #endif
1265 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1266         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1267                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1268 #endif
1269         else
1270                 return VM_FAULT_SIGBUS;
1271         get_page(page);
1272         vmf->page = page;
1273         return 0;
1274 }
1275
1276 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1277         .fault = kvm_vcpu_fault,
1278 };
1279
1280 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1281 {
1282         vma->vm_ops = &kvm_vcpu_vm_ops;
1283         return 0;
1284 }
1285
1286 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1287 {
1288         struct kvm_vcpu *vcpu = filp->private_data;
1289
1290         kvm_put_kvm(vcpu->kvm);
1291         return 0;
1292 }
1293
1294 static const struct file_operations kvm_vcpu_fops = {
1295         .release        = kvm_vcpu_release,
1296         .unlocked_ioctl = kvm_vcpu_ioctl,
1297         .compat_ioctl   = kvm_vcpu_ioctl,
1298         .mmap           = kvm_vcpu_mmap,
1299 };
1300
1301 /*
1302  * Allocates an inode for the vcpu.
1303  */
1304 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1305 {
1306         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1307         if (fd < 0)
1308                 kvm_put_kvm(vcpu->kvm);
1309         return fd;
1310 }
1311
1312 /*
1313  * Creates some virtual cpus.  Good luck creating more than one.
1314  */
1315 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1316 {
1317         int r;
1318         struct kvm_vcpu *vcpu;
1319
1320         if (!valid_vcpu(n))
1321                 return -EINVAL;
1322
1323         vcpu = kvm_arch_vcpu_create(kvm, n);
1324         if (IS_ERR(vcpu))
1325                 return PTR_ERR(vcpu);
1326
1327         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1328
1329         r = kvm_arch_vcpu_setup(vcpu);
1330         if (r)
1331                 return r;
1332
1333         mutex_lock(&kvm->lock);
1334         if (kvm->vcpus[n]) {
1335                 r = -EEXIST;
1336                 goto vcpu_destroy;
1337         }
1338         kvm->vcpus[n] = vcpu;
1339         mutex_unlock(&kvm->lock);
1340
1341         /* Now it's all set up, let userspace reach it */
1342         kvm_get_kvm(kvm);
1343         r = create_vcpu_fd(vcpu);
1344         if (r < 0)
1345                 goto unlink;
1346         return r;
1347
1348 unlink:
1349         mutex_lock(&kvm->lock);
1350         kvm->vcpus[n] = NULL;
1351 vcpu_destroy:
1352         mutex_unlock(&kvm->lock);
1353         kvm_arch_vcpu_destroy(vcpu);
1354         return r;
1355 }
1356
1357 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1358 {
1359         if (sigset) {
1360                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1361                 vcpu->sigset_active = 1;
1362                 vcpu->sigset = *sigset;
1363         } else
1364                 vcpu->sigset_active = 0;
1365         return 0;
1366 }
1367
1368 static long kvm_vcpu_ioctl(struct file *filp,
1369                            unsigned int ioctl, unsigned long arg)
1370 {
1371         struct kvm_vcpu *vcpu = filp->private_data;
1372         void __user *argp = (void __user *)arg;
1373         int r;
1374         struct kvm_fpu *fpu = NULL;
1375         struct kvm_sregs *kvm_sregs = NULL;
1376
1377         if (vcpu->kvm->mm != current->mm)
1378                 return -EIO;
1379         switch (ioctl) {
1380         case KVM_RUN:
1381                 r = -EINVAL;
1382                 if (arg)
1383                         goto out;
1384                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1385                 break;
1386         case KVM_GET_REGS: {
1387                 struct kvm_regs *kvm_regs;
1388
1389                 r = -ENOMEM;
1390                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1391                 if (!kvm_regs)
1392                         goto out;
1393                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1394                 if (r)
1395                         goto out_free1;
1396                 r = -EFAULT;
1397                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1398                         goto out_free1;
1399                 r = 0;
1400 out_free1:
1401                 kfree(kvm_regs);
1402                 break;
1403         }
1404         case KVM_SET_REGS: {
1405                 struct kvm_regs *kvm_regs;
1406
1407                 r = -ENOMEM;
1408                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1409                 if (!kvm_regs)
1410                         goto out;
1411                 r = -EFAULT;
1412                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1413                         goto out_free2;
1414                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1415                 if (r)
1416                         goto out_free2;
1417                 r = 0;
1418 out_free2:
1419                 kfree(kvm_regs);
1420                 break;
1421         }
1422         case KVM_GET_SREGS: {
1423                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1424                 r = -ENOMEM;
1425                 if (!kvm_sregs)
1426                         goto out;
1427                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1428                 if (r)
1429                         goto out;
1430                 r = -EFAULT;
1431                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1432                         goto out;
1433                 r = 0;
1434                 break;
1435         }
1436         case KVM_SET_SREGS: {
1437                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1438                 r = -ENOMEM;
1439                 if (!kvm_sregs)
1440                         goto out;
1441                 r = -EFAULT;
1442                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1443                         goto out;
1444                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1445                 if (r)
1446                         goto out;
1447                 r = 0;
1448                 break;
1449         }
1450         case KVM_GET_MP_STATE: {
1451                 struct kvm_mp_state mp_state;
1452
1453                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1454                 if (r)
1455                         goto out;
1456                 r = -EFAULT;
1457                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1458                         goto out;
1459                 r = 0;
1460                 break;
1461         }
1462         case KVM_SET_MP_STATE: {
1463                 struct kvm_mp_state mp_state;
1464
1465                 r = -EFAULT;
1466                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1467                         goto out;
1468                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1469                 if (r)
1470                         goto out;
1471                 r = 0;
1472                 break;
1473         }
1474         case KVM_TRANSLATE: {
1475                 struct kvm_translation tr;
1476
1477                 r = -EFAULT;
1478                 if (copy_from_user(&tr, argp, sizeof tr))
1479                         goto out;
1480                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1481                 if (r)
1482                         goto out;
1483                 r = -EFAULT;
1484                 if (copy_to_user(argp, &tr, sizeof tr))
1485                         goto out;
1486                 r = 0;
1487                 break;
1488         }
1489         case KVM_DEBUG_GUEST: {
1490                 struct kvm_debug_guest dbg;
1491
1492                 r = -EFAULT;
1493                 if (copy_from_user(&dbg, argp, sizeof dbg))
1494                         goto out;
1495                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1496                 if (r)
1497                         goto out;
1498                 r = 0;
1499                 break;
1500         }
1501         case KVM_SET_SIGNAL_MASK: {
1502                 struct kvm_signal_mask __user *sigmask_arg = argp;
1503                 struct kvm_signal_mask kvm_sigmask;
1504                 sigset_t sigset, *p;
1505
1506                 p = NULL;
1507                 if (argp) {
1508                         r = -EFAULT;
1509                         if (copy_from_user(&kvm_sigmask, argp,
1510                                            sizeof kvm_sigmask))
1511                                 goto out;
1512                         r = -EINVAL;
1513                         if (kvm_sigmask.len != sizeof sigset)
1514                                 goto out;
1515                         r = -EFAULT;
1516                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1517                                            sizeof sigset))
1518                                 goto out;
1519                         p = &sigset;
1520                 }
1521                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1522                 break;
1523         }
1524         case KVM_GET_FPU: {
1525                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1526                 r = -ENOMEM;
1527                 if (!fpu)
1528                         goto out;
1529                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1530                 if (r)
1531                         goto out;
1532                 r = -EFAULT;
1533                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1534                         goto out;
1535                 r = 0;
1536                 break;
1537         }
1538         case KVM_SET_FPU: {
1539                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1540                 r = -ENOMEM;
1541                 if (!fpu)
1542                         goto out;
1543                 r = -EFAULT;
1544                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1545                         goto out;
1546                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1547                 if (r)
1548                         goto out;
1549                 r = 0;
1550                 break;
1551         }
1552         default:
1553                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1554         }
1555 out:
1556         kfree(fpu);
1557         kfree(kvm_sregs);
1558         return r;
1559 }
1560
1561 static long kvm_vm_ioctl(struct file *filp,
1562                            unsigned int ioctl, unsigned long arg)
1563 {
1564         struct kvm *kvm = filp->private_data;
1565         void __user *argp = (void __user *)arg;
1566         int r;
1567
1568         if (kvm->mm != current->mm)
1569                 return -EIO;
1570         switch (ioctl) {
1571         case KVM_CREATE_VCPU:
1572                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1573                 if (r < 0)
1574                         goto out;
1575                 break;
1576         case KVM_SET_USER_MEMORY_REGION: {
1577                 struct kvm_userspace_memory_region kvm_userspace_mem;
1578
1579                 r = -EFAULT;
1580                 if (copy_from_user(&kvm_userspace_mem, argp,
1581                                                 sizeof kvm_userspace_mem))
1582                         goto out;
1583
1584                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1585                 if (r)
1586                         goto out;
1587                 break;
1588         }
1589         case KVM_GET_DIRTY_LOG: {
1590                 struct kvm_dirty_log log;
1591
1592                 r = -EFAULT;
1593                 if (copy_from_user(&log, argp, sizeof log))
1594                         goto out;
1595                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1596                 if (r)
1597                         goto out;
1598                 break;
1599         }
1600 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1601         case KVM_REGISTER_COALESCED_MMIO: {
1602                 struct kvm_coalesced_mmio_zone zone;
1603                 r = -EFAULT;
1604                 if (copy_from_user(&zone, argp, sizeof zone))
1605                         goto out;
1606                 r = -ENXIO;
1607                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1608                 if (r)
1609                         goto out;
1610                 r = 0;
1611                 break;
1612         }
1613         case KVM_UNREGISTER_COALESCED_MMIO: {
1614                 struct kvm_coalesced_mmio_zone zone;
1615                 r = -EFAULT;
1616                 if (copy_from_user(&zone, argp, sizeof zone))
1617                         goto out;
1618                 r = -ENXIO;
1619                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1620                 if (r)
1621                         goto out;
1622                 r = 0;
1623                 break;
1624         }
1625 #endif
1626 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1627         case KVM_ASSIGN_PCI_DEVICE: {
1628                 struct kvm_assigned_pci_dev assigned_dev;
1629
1630                 r = -EFAULT;
1631                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1632                         goto out;
1633                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1634                 if (r)
1635                         goto out;
1636                 break;
1637         }
1638         case KVM_ASSIGN_IRQ: {
1639                 struct kvm_assigned_irq assigned_irq;
1640
1641                 r = -EFAULT;
1642                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1643                         goto out;
1644                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1645                 if (r)
1646                         goto out;
1647                 break;
1648         }
1649 #endif
1650         default:
1651                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1652         }
1653 out:
1654         return r;
1655 }
1656
1657 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1658 {
1659         struct page *page[1];
1660         unsigned long addr;
1661         int npages;
1662         gfn_t gfn = vmf->pgoff;
1663         struct kvm *kvm = vma->vm_file->private_data;
1664
1665         addr = gfn_to_hva(kvm, gfn);
1666         if (kvm_is_error_hva(addr))
1667                 return VM_FAULT_SIGBUS;
1668
1669         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1670                                 NULL);
1671         if (unlikely(npages != 1))
1672                 return VM_FAULT_SIGBUS;
1673
1674         vmf->page = page[0];
1675         return 0;
1676 }
1677
1678 static struct vm_operations_struct kvm_vm_vm_ops = {
1679         .fault = kvm_vm_fault,
1680 };
1681
1682 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1683 {
1684         vma->vm_ops = &kvm_vm_vm_ops;
1685         return 0;
1686 }
1687
1688 static const struct file_operations kvm_vm_fops = {
1689         .release        = kvm_vm_release,
1690         .unlocked_ioctl = kvm_vm_ioctl,
1691         .compat_ioctl   = kvm_vm_ioctl,
1692         .mmap           = kvm_vm_mmap,
1693 };
1694
1695 static int kvm_dev_ioctl_create_vm(void)
1696 {
1697         int fd;
1698         struct kvm *kvm;
1699
1700         kvm = kvm_create_vm();
1701         if (IS_ERR(kvm))
1702                 return PTR_ERR(kvm);
1703         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1704         if (fd < 0)
1705                 kvm_put_kvm(kvm);
1706
1707         return fd;
1708 }
1709
1710 static long kvm_dev_ioctl(struct file *filp,
1711                           unsigned int ioctl, unsigned long arg)
1712 {
1713         long r = -EINVAL;
1714
1715         switch (ioctl) {
1716         case KVM_GET_API_VERSION:
1717                 r = -EINVAL;
1718                 if (arg)
1719                         goto out;
1720                 r = KVM_API_VERSION;
1721                 break;
1722         case KVM_CREATE_VM:
1723                 r = -EINVAL;
1724                 if (arg)
1725                         goto out;
1726                 r = kvm_dev_ioctl_create_vm();
1727                 break;
1728         case KVM_CHECK_EXTENSION:
1729                 r = kvm_dev_ioctl_check_extension(arg);
1730                 break;
1731         case KVM_GET_VCPU_MMAP_SIZE:
1732                 r = -EINVAL;
1733                 if (arg)
1734                         goto out;
1735                 r = PAGE_SIZE;     /* struct kvm_run */
1736 #ifdef CONFIG_X86
1737                 r += PAGE_SIZE;    /* pio data page */
1738 #endif
1739 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1740                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1741 #endif
1742                 break;
1743         case KVM_TRACE_ENABLE:
1744         case KVM_TRACE_PAUSE:
1745         case KVM_TRACE_DISABLE:
1746                 r = kvm_trace_ioctl(ioctl, arg);
1747                 break;
1748         default:
1749                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1750         }
1751 out:
1752         return r;
1753 }
1754
1755 static struct file_operations kvm_chardev_ops = {
1756         .unlocked_ioctl = kvm_dev_ioctl,
1757         .compat_ioctl   = kvm_dev_ioctl,
1758 };
1759
1760 static struct miscdevice kvm_dev = {
1761         KVM_MINOR,
1762         "kvm",
1763         &kvm_chardev_ops,
1764 };
1765
1766 static void hardware_enable(void *junk)
1767 {
1768         int cpu = raw_smp_processor_id();
1769
1770         if (cpu_isset(cpu, cpus_hardware_enabled))
1771                 return;
1772         cpu_set(cpu, cpus_hardware_enabled);
1773         kvm_arch_hardware_enable(NULL);
1774 }
1775
1776 static void hardware_disable(void *junk)
1777 {
1778         int cpu = raw_smp_processor_id();
1779
1780         if (!cpu_isset(cpu, cpus_hardware_enabled))
1781                 return;
1782         cpu_clear(cpu, cpus_hardware_enabled);
1783         kvm_arch_hardware_disable(NULL);
1784 }
1785
1786 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1787                            void *v)
1788 {
1789         int cpu = (long)v;
1790
1791         val &= ~CPU_TASKS_FROZEN;
1792         switch (val) {
1793         case CPU_DYING:
1794                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1795                        cpu);
1796                 hardware_disable(NULL);
1797                 break;
1798         case CPU_UP_CANCELED:
1799                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1800                        cpu);
1801                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1802                 break;
1803         case CPU_ONLINE:
1804                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1805                        cpu);
1806                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1807                 break;
1808         }
1809         return NOTIFY_OK;
1810 }
1811
1812
1813 asmlinkage void kvm_handle_fault_on_reboot(void)
1814 {
1815         if (kvm_rebooting)
1816                 /* spin while reset goes on */
1817                 while (true)
1818                         ;
1819         /* Fault while not rebooting.  We want the trace. */
1820         BUG();
1821 }
1822 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1823
1824 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1825                       void *v)
1826 {
1827         if (val == SYS_RESTART) {
1828                 /*
1829                  * Some (well, at least mine) BIOSes hang on reboot if
1830                  * in vmx root mode.
1831                  */
1832                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1833                 kvm_rebooting = true;
1834                 on_each_cpu(hardware_disable, NULL, 1);
1835         }
1836         return NOTIFY_OK;
1837 }
1838
1839 static struct notifier_block kvm_reboot_notifier = {
1840         .notifier_call = kvm_reboot,
1841         .priority = 0,
1842 };
1843
1844 void kvm_io_bus_init(struct kvm_io_bus *bus)
1845 {
1846         memset(bus, 0, sizeof(*bus));
1847 }
1848
1849 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1850 {
1851         int i;
1852
1853         for (i = 0; i < bus->dev_count; i++) {
1854                 struct kvm_io_device *pos = bus->devs[i];
1855
1856                 kvm_iodevice_destructor(pos);
1857         }
1858 }
1859
1860 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
1861                                           gpa_t addr, int len, int is_write)
1862 {
1863         int i;
1864
1865         for (i = 0; i < bus->dev_count; i++) {
1866                 struct kvm_io_device *pos = bus->devs[i];
1867
1868                 if (pos->in_range(pos, addr, len, is_write))
1869                         return pos;
1870         }
1871
1872         return NULL;
1873 }
1874
1875 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1876 {
1877         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1878
1879         bus->devs[bus->dev_count++] = dev;
1880 }
1881
1882 static struct notifier_block kvm_cpu_notifier = {
1883         .notifier_call = kvm_cpu_hotplug,
1884         .priority = 20, /* must be > scheduler priority */
1885 };
1886
1887 static int vm_stat_get(void *_offset, u64 *val)
1888 {
1889         unsigned offset = (long)_offset;
1890         struct kvm *kvm;
1891
1892         *val = 0;
1893         spin_lock(&kvm_lock);
1894         list_for_each_entry(kvm, &vm_list, vm_list)
1895                 *val += *(u32 *)((void *)kvm + offset);
1896         spin_unlock(&kvm_lock);
1897         return 0;
1898 }
1899
1900 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1901
1902 static int vcpu_stat_get(void *_offset, u64 *val)
1903 {
1904         unsigned offset = (long)_offset;
1905         struct kvm *kvm;
1906         struct kvm_vcpu *vcpu;
1907         int i;
1908
1909         *val = 0;
1910         spin_lock(&kvm_lock);
1911         list_for_each_entry(kvm, &vm_list, vm_list)
1912                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1913                         vcpu = kvm->vcpus[i];
1914                         if (vcpu)
1915                                 *val += *(u32 *)((void *)vcpu + offset);
1916                 }
1917         spin_unlock(&kvm_lock);
1918         return 0;
1919 }
1920
1921 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1922
1923 static struct file_operations *stat_fops[] = {
1924         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1925         [KVM_STAT_VM]   = &vm_stat_fops,
1926 };
1927
1928 static void kvm_init_debug(void)
1929 {
1930         struct kvm_stats_debugfs_item *p;
1931
1932         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1933         for (p = debugfs_entries; p->name; ++p)
1934                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1935                                                 (void *)(long)p->offset,
1936                                                 stat_fops[p->kind]);
1937 }
1938
1939 static void kvm_exit_debug(void)
1940 {
1941         struct kvm_stats_debugfs_item *p;
1942
1943         for (p = debugfs_entries; p->name; ++p)
1944                 debugfs_remove(p->dentry);
1945         debugfs_remove(kvm_debugfs_dir);
1946 }
1947
1948 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1949 {
1950         hardware_disable(NULL);
1951         return 0;
1952 }
1953
1954 static int kvm_resume(struct sys_device *dev)
1955 {
1956         hardware_enable(NULL);
1957         return 0;
1958 }
1959
1960 static struct sysdev_class kvm_sysdev_class = {
1961         .name = "kvm",
1962         .suspend = kvm_suspend,
1963         .resume = kvm_resume,
1964 };
1965
1966 static struct sys_device kvm_sysdev = {
1967         .id = 0,
1968         .cls = &kvm_sysdev_class,
1969 };
1970
1971 struct page *bad_page;
1972 pfn_t bad_pfn;
1973
1974 static inline
1975 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1976 {
1977         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1978 }
1979
1980 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1981 {
1982         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1983
1984         kvm_arch_vcpu_load(vcpu, cpu);
1985 }
1986
1987 static void kvm_sched_out(struct preempt_notifier *pn,
1988                           struct task_struct *next)
1989 {
1990         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1991
1992         kvm_arch_vcpu_put(vcpu);
1993 }
1994
1995 int kvm_init(void *opaque, unsigned int vcpu_size,
1996                   struct module *module)
1997 {
1998         int r;
1999         int cpu;
2000
2001         kvm_init_debug();
2002
2003         r = kvm_arch_init(opaque);
2004         if (r)
2005                 goto out_fail;
2006
2007         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2008
2009         if (bad_page == NULL) {
2010                 r = -ENOMEM;
2011                 goto out;
2012         }
2013
2014         bad_pfn = page_to_pfn(bad_page);
2015
2016         r = kvm_arch_hardware_setup();
2017         if (r < 0)
2018                 goto out_free_0;
2019
2020         for_each_online_cpu(cpu) {
2021                 smp_call_function_single(cpu,
2022                                 kvm_arch_check_processor_compat,
2023                                 &r, 1);
2024                 if (r < 0)
2025                         goto out_free_1;
2026         }
2027
2028         on_each_cpu(hardware_enable, NULL, 1);
2029         r = register_cpu_notifier(&kvm_cpu_notifier);
2030         if (r)
2031                 goto out_free_2;
2032         register_reboot_notifier(&kvm_reboot_notifier);
2033
2034         r = sysdev_class_register(&kvm_sysdev_class);
2035         if (r)
2036                 goto out_free_3;
2037
2038         r = sysdev_register(&kvm_sysdev);
2039         if (r)
2040                 goto out_free_4;
2041
2042         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2043         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2044                                            __alignof__(struct kvm_vcpu),
2045                                            0, NULL);
2046         if (!kvm_vcpu_cache) {
2047                 r = -ENOMEM;
2048                 goto out_free_5;
2049         }
2050
2051         kvm_chardev_ops.owner = module;
2052
2053         r = misc_register(&kvm_dev);
2054         if (r) {
2055                 printk(KERN_ERR "kvm: misc device register failed\n");
2056                 goto out_free;
2057         }
2058
2059         kvm_preempt_ops.sched_in = kvm_sched_in;
2060         kvm_preempt_ops.sched_out = kvm_sched_out;
2061
2062         return 0;
2063
2064 out_free:
2065         kmem_cache_destroy(kvm_vcpu_cache);
2066 out_free_5:
2067         sysdev_unregister(&kvm_sysdev);
2068 out_free_4:
2069         sysdev_class_unregister(&kvm_sysdev_class);
2070 out_free_3:
2071         unregister_reboot_notifier(&kvm_reboot_notifier);
2072         unregister_cpu_notifier(&kvm_cpu_notifier);
2073 out_free_2:
2074         on_each_cpu(hardware_disable, NULL, 1);
2075 out_free_1:
2076         kvm_arch_hardware_unsetup();
2077 out_free_0:
2078         __free_page(bad_page);
2079 out:
2080         kvm_arch_exit();
2081         kvm_exit_debug();
2082 out_fail:
2083         return r;
2084 }
2085 EXPORT_SYMBOL_GPL(kvm_init);
2086
2087 void kvm_exit(void)
2088 {
2089         kvm_trace_cleanup();
2090         misc_deregister(&kvm_dev);
2091         kmem_cache_destroy(kvm_vcpu_cache);
2092         sysdev_unregister(&kvm_sysdev);
2093         sysdev_class_unregister(&kvm_sysdev_class);
2094         unregister_reboot_notifier(&kvm_reboot_notifier);
2095         unregister_cpu_notifier(&kvm_cpu_notifier);
2096         on_each_cpu(hardware_disable, NULL, 1);
2097         kvm_arch_hardware_unsetup();
2098         kvm_arch_exit();
2099         kvm_exit_debug();
2100         __free_page(bad_page);
2101 }
2102 EXPORT_SYMBOL_GPL(kvm_exit);