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
  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->irq_source_id,
 109                     assigned_dev->guest_irq, 1);
 110         mutex_unlock(&assigned_dev->kvm->lock);
 111         kvm_put_kvm(assigned_dev->kvm);
 112 }
 113
 114 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
 115 {
 116         struct kvm_assigned_dev_kernel *assigned_dev =
 117                 (struct kvm_assigned_dev_kernel *) dev_id;
 118
 119         kvm_get_kvm(assigned_dev->kvm);
 120         schedule_work(&assigned_dev->interrupt_work);
 121         disable_irq_nosync(irq);
 122         return IRQ_HANDLED;
 123 }
 124
 125 /* Ack the irq line for an assigned device */
 126 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
 127 {
 128         struct kvm_assigned_dev_kernel *dev;
 129
 130         if (kian->gsi == -1)
 131                 return;
 132
 133         dev = container_of(kian, struct kvm_assigned_dev_kernel,
 134                            ack_notifier);
 135         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
 136         enable_irq(dev->host_irq);
 137 }
 138
 139 static void kvm_free_assigned_device(struct kvm *kvm,
 140                                      struct kvm_assigned_dev_kernel
 141                                      *assigned_dev)
 142 {
 143         if (irqchip_in_kernel(kvm) && assigned_dev->irq_requested_type)
 144                 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
 145
 146         kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
 147         kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
 148
 149         if (cancel_work_sync(&assigned_dev->interrupt_work))
 150                 /* We had pending work. That means we will have to take
 151                  * care of kvm_put_kvm.
 152                  */
 153                 kvm_put_kvm(kvm);
 154
 155         pci_reset_function(assigned_dev->dev);
 156
 157         pci_release_regions(assigned_dev->dev);
 158         pci_disable_device(assigned_dev->dev);
 159         pci_dev_put(assigned_dev->dev);
 160
 161         list_del(&assigned_dev->list);
 162         kfree(assigned_dev);
 163 }
 164
 165 void kvm_free_all_assigned_devices(struct kvm *kvm)
 166 {
 167         struct list_head *ptr, *ptr2;
 168         struct kvm_assigned_dev_kernel *assigned_dev;
 169
 170         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
 171                 assigned_dev = list_entry(ptr,
 172                                           struct kvm_assigned_dev_kernel,
 173                                           list);
 174
 175                 kvm_free_assigned_device(kvm, assigned_dev);
 176         }
 177 }
 178
 179 static int assigned_device_update_intx(struct kvm *kvm,
 180                         struct kvm_assigned_dev_kernel *adev,
 181                         struct kvm_assigned_irq *airq)
 182 {
 183         adev->guest_irq = airq->guest_irq;
 184         adev->ack_notifier.gsi = airq->guest_irq;
 185
 186         if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_INTX)
 187                 return 0;
 188
 189         if (irqchip_in_kernel(kvm)) {
 190                 if (!capable(CAP_SYS_RAWIO))
 191                         return -EPERM;
 192
 193                 if (airq->host_irq)
 194                         adev->host_irq = airq->host_irq;
 195                 else
 196                         adev->host_irq = adev->dev->irq;
 197
 198                 /* Even though this is PCI, we don't want to use shared
 199                  * interrupts. Sharing host devices with guest-assigned devices
 200                  * on the same interrupt line is not a happy situation: there
 201                  * are going to be long delays in accepting, acking, etc.
 202                  */
 203                 if (request_irq(adev->host_irq, kvm_assigned_dev_intr,
 204                                 0, "kvm_assigned_intx_device", (void *)adev))
 205                         return -EIO;
 206         }
 207
 208         adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_INTX |
 209                                    KVM_ASSIGNED_DEV_HOST_INTX;
 210         return 0;
 211 }
 212
 213 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 214                                    struct kvm_assigned_irq
 215                                    *assigned_irq)
 216 {
 217         int r = 0;
 218         struct kvm_assigned_dev_kernel *match;
 219
 220         mutex_lock(&kvm->lock);
 221
 222         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 223                                       assigned_irq->assigned_dev_id);
 224         if (!match) {
 225                 mutex_unlock(&kvm->lock);
 226                 return -EINVAL;
 227         }
 228
 229         if (!match->irq_requested_type) {
 230                 INIT_WORK(&match->interrupt_work,
 231                                 kvm_assigned_dev_interrupt_work_handler);
 232                 if (irqchip_in_kernel(kvm)) {
 233                         /* Register ack nofitier */
 234                         match->ack_notifier.gsi = -1;
 235                         match->ack_notifier.irq_acked =
 236                                         kvm_assigned_dev_ack_irq;
 237                         kvm_register_irq_ack_notifier(kvm,
 238                                         &match->ack_notifier);
 239
 240                         /* Request IRQ source ID */
 241                         r = kvm_request_irq_source_id(kvm);
 242                         if (r < 0)
 243                                 goto out_release;
 244                         else
 245                                 match->irq_source_id = r;
 246                 }
 247         }
 248
 249         r = assigned_device_update_intx(kvm, match, assigned_irq);
 250         if (r)
 251                 goto out_release;
 252
 253         mutex_unlock(&kvm->lock);
 254         return r;
 255 out_release:
 256         mutex_unlock(&kvm->lock);
 257         kvm_free_assigned_device(kvm, match);
 258         return r;
 259 }
 260
 261 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 262                                       struct kvm_assigned_pci_dev *assigned_dev)
 263 {
 264         int r = 0;
 265         struct kvm_assigned_dev_kernel *match;
 266         struct pci_dev *dev;
 267
 268         mutex_lock(&kvm->lock);
 269
 270         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 271                                       assigned_dev->assigned_dev_id);
 272         if (match) {
 273                 /* device already assigned */
 274                 r = -EINVAL;
 275                 goto out;
 276         }
 277
 278         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 279         if (match == NULL) {
 280                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 281                        __func__);
 282                 r = -ENOMEM;
 283                 goto out;
 284         }
 285         dev = pci_get_bus_and_slot(assigned_dev->busnr,
 286                                    assigned_dev->devfn);
 287         if (!dev) {
 288                 printk(KERN_INFO "%s: host device not found\n", __func__);
 289                 r = -EINVAL;
 290                 goto out_free;
 291         }
 292         if (pci_enable_device(dev)) {
 293                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 294                 r = -EBUSY;
 295                 goto out_put;
 296         }
 297         r = pci_request_regions(dev, "kvm_assigned_device");
 298         if (r) {
 299                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 300                        __func__);
 301                 goto out_disable;
 302         }
 303
 304         pci_reset_function(dev);
 305
 306         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 307         match->host_busnr = assigned_dev->busnr;
 308         match->host_devfn = assigned_dev->devfn;
 309         match->dev = dev;
 310
 311         match->kvm = kvm;
 312
 313         list_add(&match->list, &kvm->arch.assigned_dev_head);
 314
 315         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
 316                 r = kvm_iommu_map_guest(kvm, match);
 317                 if (r)
 318                         goto out_list_del;
 319         }
 320
 321 out:
 322         mutex_unlock(&kvm->lock);
 323         return r;
 324 out_list_del:
 325         list_del(&match->list);
 326         pci_release_regions(dev);
 327 out_disable:
 328         pci_disable_device(dev);
 329 out_put:
 330         pci_dev_put(dev);
 331 out_free:
 332         kfree(match);
 333         mutex_unlock(&kvm->lock);
 334         return r;
 335 }
 336 #endif
 337
 338 static inline int valid_vcpu(int n)
 339 {
 340         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 341 }
 342
 343 inline int kvm_is_mmio_pfn(pfn_t pfn)
 344 {
 345         if (pfn_valid(pfn))
 346                 return PageReserved(pfn_to_page(pfn));
 347
 348         return true;
 349 }
 350
 351 /*
 352  * Switches to specified vcpu, until a matching vcpu_put()
 353  */
 354 void vcpu_load(struct kvm_vcpu *vcpu)
 355 {
 356         int cpu;
 357
 358         mutex_lock(&vcpu->mutex);
 359         cpu = get_cpu();
 360         preempt_notifier_register(&vcpu->preempt_notifier);
 361         kvm_arch_vcpu_load(vcpu, cpu);
 362         put_cpu();
 363 }
 364
 365 void vcpu_put(struct kvm_vcpu *vcpu)
 366 {
 367         preempt_disable();
 368         kvm_arch_vcpu_put(vcpu);
 369         preempt_notifier_unregister(&vcpu->preempt_notifier);
 370         preempt_enable();
 371         mutex_unlock(&vcpu->mutex);
 372 }
 373
 374 static void ack_flush(void *_completed)
 375 {
 376 }
 377
 378 void kvm_flush_remote_tlbs(struct kvm *kvm)
 379 {
 380         int i, cpu, me;
 381         cpumask_t cpus;
 382         struct kvm_vcpu *vcpu;
 383
 384         me = get_cpu();
 385         cpus_clear(cpus);
 386         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 387                 vcpu = kvm->vcpus[i];
 388                 if (!vcpu)
 389                         continue;
 390                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 391                         continue;
 392                 cpu = vcpu->cpu;
 393                 if (cpu != -1 && cpu != me)
 394                         cpu_set(cpu, cpus);
 395         }
 396         if (cpus_empty(cpus))
 397                 goto out;
 398         ++kvm->stat.remote_tlb_flush;
 399         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 400 out:
 401         put_cpu();
 402 }
 403
 404 void kvm_reload_remote_mmus(struct kvm *kvm)
 405 {
 406         int i, cpu, me;
 407         cpumask_t cpus;
 408         struct kvm_vcpu *vcpu;
 409
 410         me = get_cpu();
 411         cpus_clear(cpus);
 412         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 413                 vcpu = kvm->vcpus[i];
 414                 if (!vcpu)
 415                         continue;
 416                 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
 417                         continue;
 418                 cpu = vcpu->cpu;
 419                 if (cpu != -1 && cpu != me)
 420                         cpu_set(cpu, cpus);
 421         }
 422         if (cpus_empty(cpus))
 423                 goto out;
 424         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 425 out:
 426         put_cpu();
 427 }
 428
 429
 430 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 431 {
 432         struct page *page;
 433         int r;
 434
 435         mutex_init(&vcpu->mutex);
 436         vcpu->cpu = -1;
 437         vcpu->kvm = kvm;
 438         vcpu->vcpu_id = id;
 439         init_waitqueue_head(&vcpu->wq);
 440
 441         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 442         if (!page) {
 443                 r = -ENOMEM;
 444                 goto fail;
 445         }
 446         vcpu->run = page_address(page);
 447
 448         r = kvm_arch_vcpu_init(vcpu);
 449         if (r < 0)
 450                 goto fail_free_run;
 451         return 0;
 452
 453 fail_free_run:
 454         free_page((unsigned long)vcpu->run);
 455 fail:
 456         return r;
 457 }
 458 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 459
 460 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 461 {
 462         kvm_arch_vcpu_uninit(vcpu);
 463         free_page((unsigned long)vcpu->run);
 464 }
 465 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 466
 467 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 468 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 469 {
 470         return container_of(mn, struct kvm, mmu_notifier);
 471 }
 472
 473 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 474                                              struct mm_struct *mm,
 475                                              unsigned long address)
 476 {
 477         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 478         int need_tlb_flush;
 479
 480         /*
 481          * When ->invalidate_page runs, the linux pte has been zapped
 482          * already but the page is still allocated until
 483          * ->invalidate_page returns. So if we increase the sequence
 484          * here the kvm page fault will notice if the spte can't be
 485          * established because the page is going to be freed. If
 486          * instead the kvm page fault establishes the spte before
 487          * ->invalidate_page runs, kvm_unmap_hva will release it
 488          * before returning.
 489          *
 490          * The sequence increase only need to be seen at spin_unlock
 491          * time, and not at spin_lock time.
 492          *
 493          * Increasing the sequence after the spin_unlock would be
 494          * unsafe because the kvm page fault could then establish the
 495          * pte after kvm_unmap_hva returned, without noticing the page
 496          * is going to be freed.
 497          */
 498         spin_lock(&kvm->mmu_lock);
 499         kvm->mmu_notifier_seq++;
 500         need_tlb_flush = kvm_unmap_hva(kvm, address);
 501         spin_unlock(&kvm->mmu_lock);
 502
 503         /* we've to flush the tlb before the pages can be freed */
 504         if (need_tlb_flush)
 505                 kvm_flush_remote_tlbs(kvm);
 506
 507 }
 508
 509 static void kvm_mmu_notifier_invalidate_range_start(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         int need_tlb_flush = 0;
 516
 517         spin_lock(&kvm->mmu_lock);
 518         /*
 519          * The count increase must become visible at unlock time as no
 520          * spte can be established without taking the mmu_lock and
 521          * count is also read inside the mmu_lock critical section.
 522          */
 523         kvm->mmu_notifier_count++;
 524         for (; start < end; start += PAGE_SIZE)
 525                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 526         spin_unlock(&kvm->mmu_lock);
 527
 528         /* we've to flush the tlb before the pages can be freed */
 529         if (need_tlb_flush)
 530                 kvm_flush_remote_tlbs(kvm);
 531 }
 532
 533 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 534                                                   struct mm_struct *mm,
 535                                                   unsigned long start,
 536                                                   unsigned long end)
 537 {
 538         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 539
 540         spin_lock(&kvm->mmu_lock);
 541         /*
 542          * This sequence increase will notify the kvm page fault that
 543          * the page that is going to be mapped in the spte could have
 544          * been freed.
 545          */
 546         kvm->mmu_notifier_seq++;
 547         /*
 548          * The above sequence increase must be visible before the
 549          * below count decrease but both values are read by the kvm
 550          * page fault under mmu_lock spinlock so we don't need to add
 551          * a smb_wmb() here in between the two.
 552          */
 553         kvm->mmu_notifier_count--;
 554         spin_unlock(&kvm->mmu_lock);
 555
 556         BUG_ON(kvm->mmu_notifier_count < 0);
 557 }
 558
 559 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 560                                               struct mm_struct *mm,
 561                                               unsigned long address)
 562 {
 563         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 564         int young;
 565
 566         spin_lock(&kvm->mmu_lock);
 567         young = kvm_age_hva(kvm, address);
 568         spin_unlock(&kvm->mmu_lock);
 569
 570         if (young)
 571                 kvm_flush_remote_tlbs(kvm);
 572
 573         return young;
 574 }
 575
 576 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 577         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 578         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 579         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 580         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 581 };
 582 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 583
 584 static struct kvm *kvm_create_vm(void)
 585 {
 586         struct kvm *kvm = kvm_arch_create_vm();
 587 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 588         struct page *page;
 589 #endif
 590
 591         if (IS_ERR(kvm))
 592                 goto out;
 593
 594 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 595         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 596         if (!page) {
 597                 kfree(kvm);
 598                 return ERR_PTR(-ENOMEM);
 599         }
 600         kvm->coalesced_mmio_ring =
 601                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 602 #endif
 603
 604 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 605         {
 606                 int err;
 607                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 608                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 609                 if (err) {
 610 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 611                         put_page(page);
 612 #endif
 613                         kfree(kvm);
 614                         return ERR_PTR(err);
 615                 }
 616         }
 617 #endif
 618
 619         kvm->mm = current->mm;
 620         atomic_inc(&kvm->mm->mm_count);
 621         spin_lock_init(&kvm->mmu_lock);
 622         kvm_io_bus_init(&kvm->pio_bus);
 623         mutex_init(&kvm->lock);
 624         kvm_io_bus_init(&kvm->mmio_bus);
 625         init_rwsem(&kvm->slots_lock);
 626         atomic_set(&kvm->users_count, 1);
 627         spin_lock(&kvm_lock);
 628         list_add(&kvm->vm_list, &vm_list);
 629         spin_unlock(&kvm_lock);
 630 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 631         kvm_coalesced_mmio_init(kvm);
 632 #endif
 633 out:
 634         return kvm;
 635 }
 636
 637 /*
 638  * Free any memory in @free but not in @dont.
 639  */
 640 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 641                                   struct kvm_memory_slot *dont)
 642 {
 643         if (!dont || free->rmap != dont->rmap)
 644                 vfree(free->rmap);
 645
 646         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 647                 vfree(free->dirty_bitmap);
 648
 649         if (!dont || free->lpage_info != dont->lpage_info)
 650                 vfree(free->lpage_info);
 651
 652         free->npages = 0;
 653         free->dirty_bitmap = NULL;
 654         free->rmap = NULL;
 655         free->lpage_info = NULL;
 656 }
 657
 658 void kvm_free_physmem(struct kvm *kvm)
 659 {
 660         int i;
 661
 662         for (i = 0; i < kvm->nmemslots; ++i)
 663                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 664 }
 665
 666 static void kvm_destroy_vm(struct kvm *kvm)
 667 {
 668         struct mm_struct *mm = kvm->mm;
 669
 670         spin_lock(&kvm_lock);
 671         list_del(&kvm->vm_list);
 672         spin_unlock(&kvm_lock);
 673         kvm_io_bus_destroy(&kvm->pio_bus);
 674         kvm_io_bus_destroy(&kvm->mmio_bus);
 675 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 676         if (kvm->coalesced_mmio_ring != NULL)
 677                 free_page((unsigned long)kvm->coalesced_mmio_ring);
 678 #endif
 679 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 680         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 681 #endif
 682         kvm_arch_destroy_vm(kvm);
 683         mmdrop(mm);
 684 }
 685
 686 void kvm_get_kvm(struct kvm *kvm)
 687 {
 688         atomic_inc(&kvm->users_count);
 689 }
 690 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 691
 692 void kvm_put_kvm(struct kvm *kvm)
 693 {
 694         if (atomic_dec_and_test(&kvm->users_count))
 695                 kvm_destroy_vm(kvm);
 696 }
 697 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 698
 699
 700 static int kvm_vm_release(struct inode *inode, struct file *filp)
 701 {
 702         struct kvm *kvm = filp->private_data;
 703
 704         kvm_put_kvm(kvm);
 705         return 0;
 706 }
 707
 708 /*
 709  * Allocate some memory and give it an address in the guest physical address
 710  * space.
 711  *
 712  * Discontiguous memory is allowed, mostly for framebuffers.
 713  *
 714  * Must be called holding mmap_sem for write.
 715  */
 716 int __kvm_set_memory_region(struct kvm *kvm,
 717                             struct kvm_userspace_memory_region *mem,
 718                             int user_alloc)
 719 {
 720         int r;
 721         gfn_t base_gfn;
 722         unsigned long npages;
 723         unsigned long i;
 724         struct kvm_memory_slot *memslot;
 725         struct kvm_memory_slot old, new;
 726
 727         r = -EINVAL;
 728         /* General sanity checks */
 729         if (mem->memory_size & (PAGE_SIZE - 1))
 730                 goto out;
 731         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 732                 goto out;
 733         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 734                 goto out;
 735         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 736                 goto out;
 737         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 738                 goto out;
 739
 740         memslot = &kvm->memslots[mem->slot];
 741         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 742         npages = mem->memory_size >> PAGE_SHIFT;
 743
 744         if (!npages)
 745                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 746
 747         new = old = *memslot;
 748
 749         new.base_gfn = base_gfn;
 750         new.npages = npages;
 751         new.flags = mem->flags;
 752
 753         /* Disallow changing a memory slot's size. */
 754         r = -EINVAL;
 755         if (npages && old.npages && npages != old.npages)
 756                 goto out_free;
 757
 758         /* Check for overlaps */
 759         r = -EEXIST;
 760         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 761                 struct kvm_memory_slot *s = &kvm->memslots[i];
 762
 763                 if (s == memslot)
 764                         continue;
 765                 if (!((base_gfn + npages <= s->base_gfn) ||
 766                       (base_gfn >= s->base_gfn + s->npages)))
 767                         goto out_free;
 768         }
 769
 770         /* Free page dirty bitmap if unneeded */
 771         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 772                 new.dirty_bitmap = NULL;
 773
 774         r = -ENOMEM;
 775
 776         /* Allocate if a slot is being created */
 777 #ifndef CONFIG_S390
 778         if (npages && !new.rmap) {
 779                 new.rmap = vmalloc(npages * sizeof(struct page *));
 780
 781                 if (!new.rmap)
 782                         goto out_free;
 783
 784                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 785
 786                 new.user_alloc = user_alloc;
 787                 /*
 788                  * hva_to_rmmap() serialzies with the mmu_lock and to be
 789                  * safe it has to ignore memslots with !user_alloc &&
 790                  * !userspace_addr.
 791                  */
 792                 if (user_alloc)
 793                         new.userspace_addr = mem->userspace_addr;
 794                 else
 795                         new.userspace_addr = 0;
 796         }
 797         if (npages && !new.lpage_info) {
 798                 int largepages = npages / KVM_PAGES_PER_HPAGE;
 799                 if (npages % KVM_PAGES_PER_HPAGE)
 800                         largepages++;
 801                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 802                         largepages++;
 803
 804                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
 805
 806                 if (!new.lpage_info)
 807                         goto out_free;
 808
 809                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
 810
 811                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 812                         new.lpage_info[0].write_count = 1;
 813                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
 814                         new.lpage_info[largepages-1].write_count = 1;
 815         }
 816
 817         /* Allocate page dirty bitmap if needed */
 818         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 819                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 820
 821                 new.dirty_bitmap = vmalloc(dirty_bytes);
 822                 if (!new.dirty_bitmap)
 823                         goto out_free;
 824                 memset(new.dirty_bitmap, 0, dirty_bytes);
 825         }
 826 #endif /* not defined CONFIG_S390 */
 827
 828         if (!npages)
 829                 kvm_arch_flush_shadow(kvm);
 830
 831         spin_lock(&kvm->mmu_lock);
 832         if (mem->slot >= kvm->nmemslots)
 833                 kvm->nmemslots = mem->slot + 1;
 834
 835         *memslot = new;
 836         spin_unlock(&kvm->mmu_lock);
 837
 838         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
 839         if (r) {
 840                 spin_lock(&kvm->mmu_lock);
 841                 *memslot = old;
 842                 spin_unlock(&kvm->mmu_lock);
 843                 goto out_free;
 844         }
 845
 846         kvm_free_physmem_slot(&old, &new);
 847 #ifdef CONFIG_DMAR
 848         /* map the pages in iommu page table */
 849         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
 850         if (r)
 851                 goto out;
 852 #endif
 853         return 0;
 854
 855 out_free:
 856         kvm_free_physmem_slot(&new, &old);
 857 out:
 858         return r;
 859
 860 }
 861 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 862
 863 int kvm_set_memory_region(struct kvm *kvm,
 864                           struct kvm_userspace_memory_region *mem,
 865                           int user_alloc)
 866 {
 867         int r;
 868
 869         down_write(&kvm->slots_lock);
 870         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 871         up_write(&kvm->slots_lock);
 872         return r;
 873 }
 874 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 875
 876 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 877                                    struct
 878                                    kvm_userspace_memory_region *mem,
 879                                    int user_alloc)
 880 {
 881         if (mem->slot >= KVM_MEMORY_SLOTS)
 882                 return -EINVAL;
 883         return kvm_set_memory_region(kvm, mem, user_alloc);
 884 }
 885
 886 int kvm_get_dirty_log(struct kvm *kvm,
 887                         struct kvm_dirty_log *log, int *is_dirty)
 888 {
 889         struct kvm_memory_slot *memslot;
 890         int r, i;
 891         int n;
 892         unsigned long any = 0;
 893
 894         r = -EINVAL;
 895         if (log->slot >= KVM_MEMORY_SLOTS)
 896                 goto out;
 897
 898         memslot = &kvm->memslots[log->slot];
 899         r = -ENOENT;
 900         if (!memslot->dirty_bitmap)
 901                 goto out;
 902
 903         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 904
 905         for (i = 0; !any && i < n/sizeof(long); ++i)
 906                 any = memslot->dirty_bitmap[i];
 907
 908         r = -EFAULT;
 909         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 910                 goto out;
 911
 912         if (any)
 913                 *is_dirty = 1;
 914
 915         r = 0;
 916 out:
 917         return r;
 918 }
 919
 920 int is_error_page(struct page *page)
 921 {
 922         return page == bad_page;
 923 }
 924 EXPORT_SYMBOL_GPL(is_error_page);
 925
 926 int is_error_pfn(pfn_t pfn)
 927 {
 928         return pfn == bad_pfn;
 929 }
 930 EXPORT_SYMBOL_GPL(is_error_pfn);
 931
 932 static inline unsigned long bad_hva(void)
 933 {
 934         return PAGE_OFFSET;
 935 }
 936
 937 int kvm_is_error_hva(unsigned long addr)
 938 {
 939         return addr == bad_hva();
 940 }
 941 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 942
 943 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
 944 {
 945         int i;
 946
 947         for (i = 0; i < kvm->nmemslots; ++i) {
 948                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 949
 950                 if (gfn >= memslot->base_gfn
 951                     && gfn < memslot->base_gfn + memslot->npages)
 952                         return memslot;
 953         }
 954         return NULL;
 955 }
 956 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
 957
 958 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 959 {
 960         gfn = unalias_gfn(kvm, gfn);
 961         return gfn_to_memslot_unaliased(kvm, gfn);
 962 }
 963
 964 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 965 {
 966         int i;
 967
 968         gfn = unalias_gfn(kvm, gfn);
 969         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 970                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 971
 972                 if (gfn >= memslot->base_gfn
 973                     && gfn < memslot->base_gfn + memslot->npages)
 974                         return 1;
 975         }
 976         return 0;
 977 }
 978 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 979
 980 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 981 {
 982         struct kvm_memory_slot *slot;
 983
 984         gfn = unalias_gfn(kvm, gfn);
 985         slot = gfn_to_memslot_unaliased(kvm, gfn);
 986         if (!slot)
 987                 return bad_hva();
 988         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 989 }
 990 EXPORT_SYMBOL_GPL(gfn_to_hva);
 991
 992 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 993 {
 994         struct page *page[1];
 995         unsigned long addr;
 996         int npages;
 997         pfn_t pfn;
 998
 999         might_sleep();
1000
1001         addr = gfn_to_hva(kvm, gfn);
1002         if (kvm_is_error_hva(addr)) {
1003                 get_page(bad_page);
1004                 return page_to_pfn(bad_page);
1005         }
1006
1007         npages = get_user_pages_fast(addr, 1, 1, page);
1008
1009         if (unlikely(npages != 1)) {
1010                 struct vm_area_struct *vma;
1011
1012                 down_read(&current->mm->mmap_sem);
1013                 vma = find_vma(current->mm, addr);
1014
1015                 if (vma == NULL || addr < vma->vm_start ||
1016                     !(vma->vm_flags & VM_PFNMAP)) {
1017                         up_read(&current->mm->mmap_sem);
1018                         get_page(bad_page);
1019                         return page_to_pfn(bad_page);
1020                 }
1021
1022                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1023                 up_read(&current->mm->mmap_sem);
1024                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1025         } else
1026                 pfn = page_to_pfn(page[0]);
1027
1028         return pfn;
1029 }
1030
1031 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1032
1033 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1034 {
1035         pfn_t pfn;
1036
1037         pfn = gfn_to_pfn(kvm, gfn);
1038         if (!kvm_is_mmio_pfn(pfn))
1039                 return pfn_to_page(pfn);
1040
1041         WARN_ON(kvm_is_mmio_pfn(pfn));
1042
1043         get_page(bad_page);
1044         return bad_page;
1045 }
1046
1047 EXPORT_SYMBOL_GPL(gfn_to_page);
1048
1049 void kvm_release_page_clean(struct page *page)
1050 {
1051         kvm_release_pfn_clean(page_to_pfn(page));
1052 }
1053 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1054
1055 void kvm_release_pfn_clean(pfn_t pfn)
1056 {
1057         if (!kvm_is_mmio_pfn(pfn))
1058                 put_page(pfn_to_page(pfn));
1059 }
1060 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1061
1062 void kvm_release_page_dirty(struct page *page)
1063 {
1064         kvm_release_pfn_dirty(page_to_pfn(page));
1065 }
1066 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1067
1068 void kvm_release_pfn_dirty(pfn_t pfn)
1069 {
1070         kvm_set_pfn_dirty(pfn);
1071         kvm_release_pfn_clean(pfn);
1072 }
1073 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1074
1075 void kvm_set_page_dirty(struct page *page)
1076 {
1077         kvm_set_pfn_dirty(page_to_pfn(page));
1078 }
1079 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1080
1081 void kvm_set_pfn_dirty(pfn_t pfn)
1082 {
1083         if (!kvm_is_mmio_pfn(pfn)) {
1084                 struct page *page = pfn_to_page(pfn);
1085                 if (!PageReserved(page))
1086                         SetPageDirty(page);
1087         }
1088 }
1089 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1090
1091 void kvm_set_pfn_accessed(pfn_t pfn)
1092 {
1093         if (!kvm_is_mmio_pfn(pfn))
1094                 mark_page_accessed(pfn_to_page(pfn));
1095 }
1096 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1097
1098 void kvm_get_pfn(pfn_t pfn)
1099 {
1100         if (!kvm_is_mmio_pfn(pfn))
1101                 get_page(pfn_to_page(pfn));
1102 }
1103 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1104
1105 static int next_segment(unsigned long len, int offset)
1106 {
1107         if (len > PAGE_SIZE - offset)
1108                 return PAGE_SIZE - offset;
1109         else
1110                 return len;
1111 }
1112
1113 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1114                         int len)
1115 {
1116         int r;
1117         unsigned long addr;
1118
1119         addr = gfn_to_hva(kvm, gfn);
1120         if (kvm_is_error_hva(addr))
1121                 return -EFAULT;
1122         r = copy_from_user(data, (void __user *)addr + offset, len);
1123         if (r)
1124                 return -EFAULT;
1125         return 0;
1126 }
1127 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1128
1129 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, 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_read_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 EXPORT_SYMBOL_GPL(kvm_read_guest);
1148
1149 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1150                           unsigned long len)
1151 {
1152         int r;
1153         unsigned long addr;
1154         gfn_t gfn = gpa >> PAGE_SHIFT;
1155         int offset = offset_in_page(gpa);
1156
1157         addr = gfn_to_hva(kvm, gfn);
1158         if (kvm_is_error_hva(addr))
1159                 return -EFAULT;
1160         pagefault_disable();
1161         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1162         pagefault_enable();
1163         if (r)
1164                 return -EFAULT;
1165         return 0;
1166 }
1167 EXPORT_SYMBOL(kvm_read_guest_atomic);
1168
1169 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1170                          int offset, int len)
1171 {
1172         int r;
1173         unsigned long addr;
1174
1175         addr = gfn_to_hva(kvm, gfn);
1176         if (kvm_is_error_hva(addr))
1177                 return -EFAULT;
1178         r = copy_to_user((void __user *)addr + offset, data, len);
1179         if (r)
1180                 return -EFAULT;
1181         mark_page_dirty(kvm, gfn);
1182         return 0;
1183 }
1184 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1185
1186 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1187                     unsigned long len)
1188 {
1189         gfn_t gfn = gpa >> PAGE_SHIFT;
1190         int seg;
1191         int offset = offset_in_page(gpa);
1192         int ret;
1193
1194         while ((seg = next_segment(len, offset)) != 0) {
1195                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1196                 if (ret < 0)
1197                         return ret;
1198                 offset = 0;
1199                 len -= seg;
1200                 data += seg;
1201                 ++gfn;
1202         }
1203         return 0;
1204 }
1205
1206 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1207 {
1208         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1209 }
1210 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1211
1212 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1213 {
1214         gfn_t gfn = gpa >> PAGE_SHIFT;
1215         int seg;
1216         int offset = offset_in_page(gpa);
1217         int ret;
1218
1219         while ((seg = next_segment(len, offset)) != 0) {
1220                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1221                 if (ret < 0)
1222                         return ret;
1223                 offset = 0;
1224                 len -= seg;
1225                 ++gfn;
1226         }
1227         return 0;
1228 }
1229 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1230
1231 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1232 {
1233         struct kvm_memory_slot *memslot;
1234
1235         gfn = unalias_gfn(kvm, gfn);
1236         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1237         if (memslot && memslot->dirty_bitmap) {
1238                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1239
1240                 /* avoid RMW */
1241                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1242                         set_bit(rel_gfn, memslot->dirty_bitmap);
1243         }
1244 }
1245
1246 /*
1247  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1248  */
1249 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1250 {
1251         DEFINE_WAIT(wait);
1252
1253         for (;;) {
1254                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1255
1256                 if (kvm_cpu_has_interrupt(vcpu) ||
1257                     kvm_cpu_has_pending_timer(vcpu) ||
1258                     kvm_arch_vcpu_runnable(vcpu)) {
1259                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1260                         break;
1261                 }
1262                 if (signal_pending(current))
1263                         break;
1264
1265                 vcpu_put(vcpu);
1266                 schedule();
1267                 vcpu_load(vcpu);
1268         }
1269
1270         finish_wait(&vcpu->wq, &wait);
1271 }
1272
1273 void kvm_resched(struct kvm_vcpu *vcpu)
1274 {
1275         if (!need_resched())
1276                 return;
1277         cond_resched();
1278 }
1279 EXPORT_SYMBOL_GPL(kvm_resched);
1280
1281 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1282 {
1283         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1284         struct page *page;
1285
1286         if (vmf->pgoff == 0)
1287                 page = virt_to_page(vcpu->run);
1288 #ifdef CONFIG_X86
1289         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1290                 page = virt_to_page(vcpu->arch.pio_data);
1291 #endif
1292 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1293         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1294                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1295 #endif
1296         else
1297                 return VM_FAULT_SIGBUS;
1298         get_page(page);
1299         vmf->page = page;
1300         return 0;
1301 }
1302
1303 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1304         .fault = kvm_vcpu_fault,
1305 };
1306
1307 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1308 {
1309         vma->vm_ops = &kvm_vcpu_vm_ops;
1310         return 0;
1311 }
1312
1313 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1314 {
1315         struct kvm_vcpu *vcpu = filp->private_data;
1316
1317         kvm_put_kvm(vcpu->kvm);
1318         return 0;
1319 }
1320
1321 static const struct file_operations kvm_vcpu_fops = {
1322         .release        = kvm_vcpu_release,
1323         .unlocked_ioctl = kvm_vcpu_ioctl,
1324         .compat_ioctl   = kvm_vcpu_ioctl,
1325         .mmap           = kvm_vcpu_mmap,
1326 };
1327
1328 /*
1329  * Allocates an inode for the vcpu.
1330  */
1331 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1332 {
1333         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1334         if (fd < 0)
1335                 kvm_put_kvm(vcpu->kvm);
1336         return fd;
1337 }
1338
1339 /*
1340  * Creates some virtual cpus.  Good luck creating more than one.
1341  */
1342 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1343 {
1344         int r;
1345         struct kvm_vcpu *vcpu;
1346
1347         if (!valid_vcpu(n))
1348                 return -EINVAL;
1349
1350         vcpu = kvm_arch_vcpu_create(kvm, n);
1351         if (IS_ERR(vcpu))
1352                 return PTR_ERR(vcpu);
1353
1354         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1355
1356         r = kvm_arch_vcpu_setup(vcpu);
1357         if (r)
1358                 return r;
1359
1360         mutex_lock(&kvm->lock);
1361         if (kvm->vcpus[n]) {
1362                 r = -EEXIST;
1363                 goto vcpu_destroy;
1364         }
1365         kvm->vcpus[n] = vcpu;
1366         mutex_unlock(&kvm->lock);
1367
1368         /* Now it's all set up, let userspace reach it */
1369         kvm_get_kvm(kvm);
1370         r = create_vcpu_fd(vcpu);
1371         if (r < 0)
1372                 goto unlink;
1373         return r;
1374
1375 unlink:
1376         mutex_lock(&kvm->lock);
1377         kvm->vcpus[n] = NULL;
1378 vcpu_destroy:
1379         mutex_unlock(&kvm->lock);
1380         kvm_arch_vcpu_destroy(vcpu);
1381         return r;
1382 }
1383
1384 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1385 {
1386         if (sigset) {
1387                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1388                 vcpu->sigset_active = 1;
1389                 vcpu->sigset = *sigset;
1390         } else
1391                 vcpu->sigset_active = 0;
1392         return 0;
1393 }
1394
1395 static long kvm_vcpu_ioctl(struct file *filp,
1396                            unsigned int ioctl, unsigned long arg)
1397 {
1398         struct kvm_vcpu *vcpu = filp->private_data;
1399         void __user *argp = (void __user *)arg;
1400         int r;
1401         struct kvm_fpu *fpu = NULL;
1402         struct kvm_sregs *kvm_sregs = NULL;
1403
1404         if (vcpu->kvm->mm != current->mm)
1405                 return -EIO;
1406         switch (ioctl) {
1407         case KVM_RUN:
1408                 r = -EINVAL;
1409                 if (arg)
1410                         goto out;
1411                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1412                 break;
1413         case KVM_GET_REGS: {
1414                 struct kvm_regs *kvm_regs;
1415
1416                 r = -ENOMEM;
1417                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1418                 if (!kvm_regs)
1419                         goto out;
1420                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1421                 if (r)
1422                         goto out_free1;
1423                 r = -EFAULT;
1424                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1425                         goto out_free1;
1426                 r = 0;
1427 out_free1:
1428                 kfree(kvm_regs);
1429                 break;
1430         }
1431         case KVM_SET_REGS: {
1432                 struct kvm_regs *kvm_regs;
1433
1434                 r = -ENOMEM;
1435                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1436                 if (!kvm_regs)
1437                         goto out;
1438                 r = -EFAULT;
1439                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1440                         goto out_free2;
1441                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1442                 if (r)
1443                         goto out_free2;
1444                 r = 0;
1445 out_free2:
1446                 kfree(kvm_regs);
1447                 break;
1448         }
1449         case KVM_GET_SREGS: {
1450                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1451                 r = -ENOMEM;
1452                 if (!kvm_sregs)
1453                         goto out;
1454                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1455                 if (r)
1456                         goto out;
1457                 r = -EFAULT;
1458                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1459                         goto out;
1460                 r = 0;
1461                 break;
1462         }
1463         case KVM_SET_SREGS: {
1464                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1465                 r = -ENOMEM;
1466                 if (!kvm_sregs)
1467                         goto out;
1468                 r = -EFAULT;
1469                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1470                         goto out;
1471                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1472                 if (r)
1473                         goto out;
1474                 r = 0;
1475                 break;
1476         }
1477         case KVM_GET_MP_STATE: {
1478                 struct kvm_mp_state mp_state;
1479
1480                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1481                 if (r)
1482                         goto out;
1483                 r = -EFAULT;
1484                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1485                         goto out;
1486                 r = 0;
1487                 break;
1488         }
1489         case KVM_SET_MP_STATE: {
1490                 struct kvm_mp_state mp_state;
1491
1492                 r = -EFAULT;
1493                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1494                         goto out;
1495                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1496                 if (r)
1497                         goto out;
1498                 r = 0;
1499                 break;
1500         }
1501         case KVM_TRANSLATE: {
1502                 struct kvm_translation tr;
1503
1504                 r = -EFAULT;
1505                 if (copy_from_user(&tr, argp, sizeof tr))
1506                         goto out;
1507                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1508                 if (r)
1509                         goto out;
1510                 r = -EFAULT;
1511                 if (copy_to_user(argp, &tr, sizeof tr))
1512                         goto out;
1513                 r = 0;
1514                 break;
1515         }
1516         case KVM_DEBUG_GUEST: {
1517                 struct kvm_debug_guest dbg;
1518
1519                 r = -EFAULT;
1520                 if (copy_from_user(&dbg, argp, sizeof dbg))
1521                         goto out;
1522                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1523                 if (r)
1524                         goto out;
1525                 r = 0;
1526                 break;
1527         }
1528         case KVM_SET_SIGNAL_MASK: {
1529                 struct kvm_signal_mask __user *sigmask_arg = argp;
1530                 struct kvm_signal_mask kvm_sigmask;
1531                 sigset_t sigset, *p;
1532
1533                 p = NULL;
1534                 if (argp) {
1535                         r = -EFAULT;
1536                         if (copy_from_user(&kvm_sigmask, argp,
1537                                            sizeof kvm_sigmask))
1538                                 goto out;
1539                         r = -EINVAL;
1540                         if (kvm_sigmask.len != sizeof sigset)
1541                                 goto out;
1542                         r = -EFAULT;
1543                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1544                                            sizeof sigset))
1545                                 goto out;
1546                         p = &sigset;
1547                 }
1548                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1549                 break;
1550         }
1551         case KVM_GET_FPU: {
1552                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1553                 r = -ENOMEM;
1554                 if (!fpu)
1555                         goto out;
1556                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1557                 if (r)
1558                         goto out;
1559                 r = -EFAULT;
1560                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1561                         goto out;
1562                 r = 0;
1563                 break;
1564         }
1565         case KVM_SET_FPU: {
1566                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1567                 r = -ENOMEM;
1568                 if (!fpu)
1569                         goto out;
1570                 r = -EFAULT;
1571                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1572                         goto out;
1573                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1574                 if (r)
1575                         goto out;
1576                 r = 0;
1577                 break;
1578         }
1579         default:
1580                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1581         }
1582 out:
1583         kfree(fpu);
1584         kfree(kvm_sregs);
1585         return r;
1586 }
1587
1588 static long kvm_vm_ioctl(struct file *filp,
1589                            unsigned int ioctl, unsigned long arg)
1590 {
1591         struct kvm *kvm = filp->private_data;
1592         void __user *argp = (void __user *)arg;
1593         int r;
1594
1595         if (kvm->mm != current->mm)
1596                 return -EIO;
1597         switch (ioctl) {
1598         case KVM_CREATE_VCPU:
1599                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1600                 if (r < 0)
1601                         goto out;
1602                 break;
1603         case KVM_SET_USER_MEMORY_REGION: {
1604                 struct kvm_userspace_memory_region kvm_userspace_mem;
1605
1606                 r = -EFAULT;
1607                 if (copy_from_user(&kvm_userspace_mem, argp,
1608                                                 sizeof kvm_userspace_mem))
1609                         goto out;
1610
1611                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1612                 if (r)
1613                         goto out;
1614                 break;
1615         }
1616         case KVM_GET_DIRTY_LOG: {
1617                 struct kvm_dirty_log log;
1618
1619                 r = -EFAULT;
1620                 if (copy_from_user(&log, argp, sizeof log))
1621                         goto out;
1622                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1623                 if (r)
1624                         goto out;
1625                 break;
1626         }
1627 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1628         case KVM_REGISTER_COALESCED_MMIO: {
1629                 struct kvm_coalesced_mmio_zone zone;
1630                 r = -EFAULT;
1631                 if (copy_from_user(&zone, argp, sizeof zone))
1632                         goto out;
1633                 r = -ENXIO;
1634                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1635                 if (r)
1636                         goto out;
1637                 r = 0;
1638                 break;
1639         }
1640         case KVM_UNREGISTER_COALESCED_MMIO: {
1641                 struct kvm_coalesced_mmio_zone zone;
1642                 r = -EFAULT;
1643                 if (copy_from_user(&zone, argp, sizeof zone))
1644                         goto out;
1645                 r = -ENXIO;
1646                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1647                 if (r)
1648                         goto out;
1649                 r = 0;
1650                 break;
1651         }
1652 #endif
1653 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1654         case KVM_ASSIGN_PCI_DEVICE: {
1655                 struct kvm_assigned_pci_dev assigned_dev;
1656
1657                 r = -EFAULT;
1658                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1659                         goto out;
1660                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1661                 if (r)
1662                         goto out;
1663                 break;
1664         }
1665         case KVM_ASSIGN_IRQ: {
1666                 struct kvm_assigned_irq assigned_irq;
1667
1668                 r = -EFAULT;
1669                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1670                         goto out;
1671                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1672                 if (r)
1673                         goto out;
1674                 break;
1675         }
1676 #endif
1677         default:
1678                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1679         }
1680 out:
1681         return r;
1682 }
1683
1684 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1685 {
1686         struct page *page[1];
1687         unsigned long addr;
1688         int npages;
1689         gfn_t gfn = vmf->pgoff;
1690         struct kvm *kvm = vma->vm_file->private_data;
1691
1692         addr = gfn_to_hva(kvm, gfn);
1693         if (kvm_is_error_hva(addr))
1694                 return VM_FAULT_SIGBUS;
1695
1696         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1697                                 NULL);
1698         if (unlikely(npages != 1))
1699                 return VM_FAULT_SIGBUS;
1700
1701         vmf->page = page[0];
1702         return 0;
1703 }
1704
1705 static struct vm_operations_struct kvm_vm_vm_ops = {
1706         .fault = kvm_vm_fault,
1707 };
1708
1709 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1710 {
1711         vma->vm_ops = &kvm_vm_vm_ops;
1712         return 0;
1713 }
1714
1715 static const struct file_operations kvm_vm_fops = {
1716         .release        = kvm_vm_release,
1717         .unlocked_ioctl = kvm_vm_ioctl,
1718         .compat_ioctl   = kvm_vm_ioctl,
1719         .mmap           = kvm_vm_mmap,
1720 };
1721
1722 static int kvm_dev_ioctl_create_vm(void)
1723 {
1724         int fd;
1725         struct kvm *kvm;
1726
1727         kvm = kvm_create_vm();
1728         if (IS_ERR(kvm))
1729                 return PTR_ERR(kvm);
1730         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1731         if (fd < 0)
1732                 kvm_put_kvm(kvm);
1733
1734         return fd;
1735 }
1736
1737 static long kvm_dev_ioctl(struct file *filp,
1738                           unsigned int ioctl, unsigned long arg)
1739 {
1740         long r = -EINVAL;
1741
1742         switch (ioctl) {
1743         case KVM_GET_API_VERSION:
1744                 r = -EINVAL;
1745                 if (arg)
1746                         goto out;
1747                 r = KVM_API_VERSION;
1748                 break;
1749         case KVM_CREATE_VM:
1750                 r = -EINVAL;
1751                 if (arg)
1752                         goto out;
1753                 r = kvm_dev_ioctl_create_vm();
1754                 break;
1755         case KVM_CHECK_EXTENSION:
1756                 r = kvm_dev_ioctl_check_extension(arg);
1757                 break;
1758         case KVM_GET_VCPU_MMAP_SIZE:
1759                 r = -EINVAL;
1760                 if (arg)
1761                         goto out;
1762                 r = PAGE_SIZE;     /* struct kvm_run */
1763 #ifdef CONFIG_X86
1764                 r += PAGE_SIZE;    /* pio data page */
1765 #endif
1766 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1767                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1768 #endif
1769                 break;
1770         case KVM_TRACE_ENABLE:
1771         case KVM_TRACE_PAUSE:
1772         case KVM_TRACE_DISABLE:
1773                 r = kvm_trace_ioctl(ioctl, arg);
1774                 break;
1775         default:
1776                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1777         }
1778 out:
1779         return r;
1780 }
1781
1782 static struct file_operations kvm_chardev_ops = {
1783         .unlocked_ioctl = kvm_dev_ioctl,
1784         .compat_ioctl   = kvm_dev_ioctl,
1785 };
1786
1787 static struct miscdevice kvm_dev = {
1788         KVM_MINOR,
1789         "kvm",
1790         &kvm_chardev_ops,
1791 };
1792
1793 static void hardware_enable(void *junk)
1794 {
1795         int cpu = raw_smp_processor_id();
1796
1797         if (cpu_isset(cpu, cpus_hardware_enabled))
1798                 return;
1799         cpu_set(cpu, cpus_hardware_enabled);
1800         kvm_arch_hardware_enable(NULL);
1801 }
1802
1803 static void hardware_disable(void *junk)
1804 {
1805         int cpu = raw_smp_processor_id();
1806
1807         if (!cpu_isset(cpu, cpus_hardware_enabled))
1808                 return;
1809         cpu_clear(cpu, cpus_hardware_enabled);
1810         kvm_arch_hardware_disable(NULL);
1811 }
1812
1813 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1814                            void *v)
1815 {
1816         int cpu = (long)v;
1817
1818         val &= ~CPU_TASKS_FROZEN;
1819         switch (val) {
1820         case CPU_DYING:
1821                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1822                        cpu);
1823                 hardware_disable(NULL);
1824                 break;
1825         case CPU_UP_CANCELED:
1826                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1827                        cpu);
1828                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1829                 break;
1830         case CPU_ONLINE:
1831                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1832                        cpu);
1833                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1834                 break;
1835         }
1836         return NOTIFY_OK;
1837 }
1838
1839
1840 asmlinkage void kvm_handle_fault_on_reboot(void)
1841 {
1842         if (kvm_rebooting)
1843                 /* spin while reset goes on */
1844                 while (true)
1845                         ;
1846         /* Fault while not rebooting.  We want the trace. */
1847         BUG();
1848 }
1849 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1850
1851 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1852                       void *v)
1853 {
1854         if (val == SYS_RESTART) {
1855                 /*
1856                  * Some (well, at least mine) BIOSes hang on reboot if
1857                  * in vmx root mode.
1858                  */
1859                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1860                 kvm_rebooting = true;
1861                 on_each_cpu(hardware_disable, NULL, 1);
1862         }
1863         return NOTIFY_OK;
1864 }
1865
1866 static struct notifier_block kvm_reboot_notifier = {
1867         .notifier_call = kvm_reboot,
1868         .priority = 0,
1869 };
1870
1871 void kvm_io_bus_init(struct kvm_io_bus *bus)
1872 {
1873         memset(bus, 0, sizeof(*bus));
1874 }
1875
1876 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1877 {
1878         int i;
1879
1880         for (i = 0; i < bus->dev_count; i++) {
1881                 struct kvm_io_device *pos = bus->devs[i];
1882
1883                 kvm_iodevice_destructor(pos);
1884         }
1885 }
1886
1887 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
1888                                           gpa_t addr, int len, int is_write)
1889 {
1890         int i;
1891
1892         for (i = 0; i < bus->dev_count; i++) {
1893                 struct kvm_io_device *pos = bus->devs[i];
1894
1895                 if (pos->in_range(pos, addr, len, is_write))
1896                         return pos;
1897         }
1898
1899         return NULL;
1900 }
1901
1902 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1903 {
1904         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1905
1906         bus->devs[bus->dev_count++] = dev;
1907 }
1908
1909 static struct notifier_block kvm_cpu_notifier = {
1910         .notifier_call = kvm_cpu_hotplug,
1911         .priority = 20, /* must be > scheduler priority */
1912 };
1913
1914 static int vm_stat_get(void *_offset, u64 *val)
1915 {
1916         unsigned offset = (long)_offset;
1917         struct kvm *kvm;
1918
1919         *val = 0;
1920         spin_lock(&kvm_lock);
1921         list_for_each_entry(kvm, &vm_list, vm_list)
1922                 *val += *(u32 *)((void *)kvm + offset);
1923         spin_unlock(&kvm_lock);
1924         return 0;
1925 }
1926
1927 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1928
1929 static int vcpu_stat_get(void *_offset, u64 *val)
1930 {
1931         unsigned offset = (long)_offset;
1932         struct kvm *kvm;
1933         struct kvm_vcpu *vcpu;
1934         int i;
1935
1936         *val = 0;
1937         spin_lock(&kvm_lock);
1938         list_for_each_entry(kvm, &vm_list, vm_list)
1939                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1940                         vcpu = kvm->vcpus[i];
1941                         if (vcpu)
1942                                 *val += *(u32 *)((void *)vcpu + offset);
1943                 }
1944         spin_unlock(&kvm_lock);
1945         return 0;
1946 }
1947
1948 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1949
1950 static struct file_operations *stat_fops[] = {
1951         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1952         [KVM_STAT_VM]   = &vm_stat_fops,
1953 };
1954
1955 static void kvm_init_debug(void)
1956 {
1957         struct kvm_stats_debugfs_item *p;
1958
1959         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1960         for (p = debugfs_entries; p->name; ++p)
1961                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1962                                                 (void *)(long)p->offset,
1963                                                 stat_fops[p->kind]);
1964 }
1965
1966 static void kvm_exit_debug(void)
1967 {
1968         struct kvm_stats_debugfs_item *p;
1969
1970         for (p = debugfs_entries; p->name; ++p)
1971                 debugfs_remove(p->dentry);
1972         debugfs_remove(kvm_debugfs_dir);
1973 }
1974
1975 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1976 {
1977         hardware_disable(NULL);
1978         return 0;
1979 }
1980
1981 static int kvm_resume(struct sys_device *dev)
1982 {
1983         hardware_enable(NULL);
1984         return 0;
1985 }
1986
1987 static struct sysdev_class kvm_sysdev_class = {
1988         .name = "kvm",
1989         .suspend = kvm_suspend,
1990         .resume = kvm_resume,
1991 };
1992
1993 static struct sys_device kvm_sysdev = {
1994         .id = 0,
1995         .cls = &kvm_sysdev_class,
1996 };
1997
1998 struct page *bad_page;
1999 pfn_t bad_pfn;
2000
2001 static inline
2002 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2003 {
2004         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2005 }
2006
2007 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2008 {
2009         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2010
2011         kvm_arch_vcpu_load(vcpu, cpu);
2012 }
2013
2014 static void kvm_sched_out(struct preempt_notifier *pn,
2015                           struct task_struct *next)
2016 {
2017         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2018
2019         kvm_arch_vcpu_put(vcpu);
2020 }
2021
2022 int kvm_init(void *opaque, unsigned int vcpu_size,
2023                   struct module *module)
2024 {
2025         int r;
2026         int cpu;
2027
2028         kvm_init_debug();
2029
2030         r = kvm_arch_init(opaque);
2031         if (r)
2032                 goto out_fail;
2033
2034         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2035
2036         if (bad_page == NULL) {
2037                 r = -ENOMEM;
2038                 goto out;
2039         }
2040
2041         bad_pfn = page_to_pfn(bad_page);
2042
2043         r = kvm_arch_hardware_setup();
2044         if (r < 0)
2045                 goto out_free_0;
2046
2047         for_each_online_cpu(cpu) {
2048                 smp_call_function_single(cpu,
2049                                 kvm_arch_check_processor_compat,
2050                                 &r, 1);
2051                 if (r < 0)
2052                         goto out_free_1;
2053         }
2054
2055         on_each_cpu(hardware_enable, NULL, 1);
2056         r = register_cpu_notifier(&kvm_cpu_notifier);
2057         if (r)
2058                 goto out_free_2;
2059         register_reboot_notifier(&kvm_reboot_notifier);
2060
2061         r = sysdev_class_register(&kvm_sysdev_class);
2062         if (r)
2063                 goto out_free_3;
2064
2065         r = sysdev_register(&kvm_sysdev);
2066         if (r)
2067                 goto out_free_4;
2068
2069         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2070         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2071                                            __alignof__(struct kvm_vcpu),
2072                                            0, NULL);
2073         if (!kvm_vcpu_cache) {
2074                 r = -ENOMEM;
2075                 goto out_free_5;
2076         }
2077
2078         kvm_chardev_ops.owner = module;
2079
2080         r = misc_register(&kvm_dev);
2081         if (r) {
2082                 printk(KERN_ERR "kvm: misc device register failed\n");
2083                 goto out_free;
2084         }
2085
2086         kvm_preempt_ops.sched_in = kvm_sched_in;
2087         kvm_preempt_ops.sched_out = kvm_sched_out;
2088
2089         return 0;
2090
2091 out_free:
2092         kmem_cache_destroy(kvm_vcpu_cache);
2093 out_free_5:
2094         sysdev_unregister(&kvm_sysdev);
2095 out_free_4:
2096         sysdev_class_unregister(&kvm_sysdev_class);
2097 out_free_3:
2098         unregister_reboot_notifier(&kvm_reboot_notifier);
2099         unregister_cpu_notifier(&kvm_cpu_notifier);
2100 out_free_2:
2101         on_each_cpu(hardware_disable, NULL, 1);
2102 out_free_1:
2103         kvm_arch_hardware_unsetup();
2104 out_free_0:
2105         __free_page(bad_page);
2106 out:
2107         kvm_arch_exit();
2108         kvm_exit_debug();
2109 out_fail:
2110         return r;
2111 }
2112 EXPORT_SYMBOL_GPL(kvm_init);
2113
2114 void kvm_exit(void)
2115 {
2116         kvm_trace_cleanup();
2117         misc_deregister(&kvm_dev);
2118         kmem_cache_destroy(kvm_vcpu_cache);
2119         sysdev_unregister(&kvm_sysdev);
2120         sysdev_class_unregister(&kvm_sysdev_class);
2121         unregister_reboot_notifier(&kvm_reboot_notifier);
2122         unregister_cpu_notifier(&kvm_cpu_notifier);
2123         on_each_cpu(hardware_disable, NULL, 1);
2124         kvm_arch_hardware_unsetup();
2125         kvm_arch_exit();
2126         kvm_exit_debug();
2127         __free_page(bad_page);
2128 }
2129 EXPORT_SYMBOL_GPL(kvm_exit);