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)
 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 kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 180                                    struct kvm_assigned_irq
 181                                    *assigned_irq)
 182 {
 183         int r = 0;
 184         struct kvm_assigned_dev_kernel *match;
 185
 186         mutex_lock(&kvm->lock);
 187
 188         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 189                                       assigned_irq->assigned_dev_id);
 190         if (!match) {
 191                 mutex_unlock(&kvm->lock);
 192                 return -EINVAL;
 193         }
 194
 195         if (match->irq_requested) {
 196                 match->guest_irq = assigned_irq->guest_irq;
 197                 match->ack_notifier.gsi = assigned_irq->guest_irq;
 198                 mutex_unlock(&kvm->lock);
 199                 return 0;
 200         }
 201
 202         INIT_WORK(&match->interrupt_work,
 203                   kvm_assigned_dev_interrupt_work_handler);
 204
 205         if (irqchip_in_kernel(kvm)) {
 206                 if (!capable(CAP_SYS_RAWIO)) {
 207                         r = -EPERM;
 208                         goto out_release;
 209                 }
 210
 211                 if (assigned_irq->host_irq)
 212                         match->host_irq = assigned_irq->host_irq;
 213                 else
 214                         match->host_irq = match->dev->irq;
 215                 match->guest_irq = assigned_irq->guest_irq;
 216                 match->ack_notifier.gsi = assigned_irq->guest_irq;
 217                 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
 218                 kvm_register_irq_ack_notifier(kvm, &match->ack_notifier);
 219                 r = kvm_request_irq_source_id(kvm);
 220                 if (r < 0)
 221                         goto out_release;
 222                 else
 223                         match->irq_source_id = r;
 224
 225                 /* Even though this is PCI, we don't want to use shared
 226                  * interrupts. Sharing host devices with guest-assigned devices
 227                  * on the same interrupt line is not a happy situation: there
 228                  * are going to be long delays in accepting, acking, etc.
 229                  */
 230                 if (request_irq(match->host_irq, kvm_assigned_dev_intr, 0,
 231                                 "kvm_assigned_device", (void *)match)) {
 232                         r = -EIO;
 233                         goto out_release;
 234                 }
 235         }
 236
 237         match->irq_requested = true;
 238         mutex_unlock(&kvm->lock);
 239         return r;
 240 out_release:
 241         mutex_unlock(&kvm->lock);
 242         kvm_free_assigned_device(kvm, match);
 243         return r;
 244 }
 245
 246 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 247                                       struct kvm_assigned_pci_dev *assigned_dev)
 248 {
 249         int r = 0;
 250         struct kvm_assigned_dev_kernel *match;
 251         struct pci_dev *dev;
 252
 253         mutex_lock(&kvm->lock);
 254
 255         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 256                                       assigned_dev->assigned_dev_id);
 257         if (match) {
 258                 /* device already assigned */
 259                 r = -EINVAL;
 260                 goto out;
 261         }
 262
 263         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 264         if (match == NULL) {
 265                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 266                        __func__);
 267                 r = -ENOMEM;
 268                 goto out;
 269         }
 270         dev = pci_get_bus_and_slot(assigned_dev->busnr,
 271                                    assigned_dev->devfn);
 272         if (!dev) {
 273                 printk(KERN_INFO "%s: host device not found\n", __func__);
 274                 r = -EINVAL;
 275                 goto out_free;
 276         }
 277         if (pci_enable_device(dev)) {
 278                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 279                 r = -EBUSY;
 280                 goto out_put;
 281         }
 282         r = pci_request_regions(dev, "kvm_assigned_device");
 283         if (r) {
 284                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 285                        __func__);
 286                 goto out_disable;
 287         }
 288
 289         pci_reset_function(dev);
 290
 291         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 292         match->host_busnr = assigned_dev->busnr;
 293         match->host_devfn = assigned_dev->devfn;
 294         match->dev = dev;
 295
 296         match->kvm = kvm;
 297
 298         list_add(&match->list, &kvm->arch.assigned_dev_head);
 299
 300         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
 301                 r = kvm_iommu_map_guest(kvm, match);
 302                 if (r)
 303                         goto out_list_del;
 304         }
 305
 306 out:
 307         mutex_unlock(&kvm->lock);
 308         return r;
 309 out_list_del:
 310         list_del(&match->list);
 311         pci_release_regions(dev);
 312 out_disable:
 313         pci_disable_device(dev);
 314 out_put:
 315         pci_dev_put(dev);
 316 out_free:
 317         kfree(match);
 318         mutex_unlock(&kvm->lock);
 319         return r;
 320 }
 321 #endif
 322
 323 static inline int valid_vcpu(int n)
 324 {
 325         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 326 }
 327
 328 inline int kvm_is_mmio_pfn(pfn_t pfn)
 329 {
 330         if (pfn_valid(pfn))
 331                 return PageReserved(pfn_to_page(pfn));
 332
 333         return true;
 334 }
 335
 336 /*
 337  * Switches to specified vcpu, until a matching vcpu_put()
 338  */
 339 void vcpu_load(struct kvm_vcpu *vcpu)
 340 {
 341         int cpu;
 342
 343         mutex_lock(&vcpu->mutex);
 344         cpu = get_cpu();
 345         preempt_notifier_register(&vcpu->preempt_notifier);
 346         kvm_arch_vcpu_load(vcpu, cpu);
 347         put_cpu();
 348 }
 349
 350 void vcpu_put(struct kvm_vcpu *vcpu)
 351 {
 352         preempt_disable();
 353         kvm_arch_vcpu_put(vcpu);
 354         preempt_notifier_unregister(&vcpu->preempt_notifier);
 355         preempt_enable();
 356         mutex_unlock(&vcpu->mutex);
 357 }
 358
 359 static void ack_flush(void *_completed)
 360 {
 361 }
 362
 363 void kvm_flush_remote_tlbs(struct kvm *kvm)
 364 {
 365         int i, cpu, me;
 366         cpumask_t cpus;
 367         struct kvm_vcpu *vcpu;
 368
 369         me = get_cpu();
 370         cpus_clear(cpus);
 371         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 372                 vcpu = kvm->vcpus[i];
 373                 if (!vcpu)
 374                         continue;
 375                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 376                         continue;
 377                 cpu = vcpu->cpu;
 378                 if (cpu != -1 && cpu != me)
 379                         cpu_set(cpu, cpus);
 380         }
 381         if (cpus_empty(cpus))
 382                 goto out;
 383         ++kvm->stat.remote_tlb_flush;
 384         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 385 out:
 386         put_cpu();
 387 }
 388
 389 void kvm_reload_remote_mmus(struct kvm *kvm)
 390 {
 391         int i, cpu, me;
 392         cpumask_t cpus;
 393         struct kvm_vcpu *vcpu;
 394
 395         me = get_cpu();
 396         cpus_clear(cpus);
 397         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 398                 vcpu = kvm->vcpus[i];
 399                 if (!vcpu)
 400                         continue;
 401                 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
 402                         continue;
 403                 cpu = vcpu->cpu;
 404                 if (cpu != -1 && cpu != me)
 405                         cpu_set(cpu, cpus);
 406         }
 407         if (cpus_empty(cpus))
 408                 goto out;
 409         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 410 out:
 411         put_cpu();
 412 }
 413
 414
 415 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 416 {
 417         struct page *page;
 418         int r;
 419
 420         mutex_init(&vcpu->mutex);
 421         vcpu->cpu = -1;
 422         vcpu->kvm = kvm;
 423         vcpu->vcpu_id = id;
 424         init_waitqueue_head(&vcpu->wq);
 425
 426         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 427         if (!page) {
 428                 r = -ENOMEM;
 429                 goto fail;
 430         }
 431         vcpu->run = page_address(page);
 432
 433         r = kvm_arch_vcpu_init(vcpu);
 434         if (r < 0)
 435                 goto fail_free_run;
 436         return 0;
 437
 438 fail_free_run:
 439         free_page((unsigned long)vcpu->run);
 440 fail:
 441         return r;
 442 }
 443 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 444
 445 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 446 {
 447         kvm_arch_vcpu_uninit(vcpu);
 448         free_page((unsigned long)vcpu->run);
 449 }
 450 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 451
 452 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 453 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 454 {
 455         return container_of(mn, struct kvm, mmu_notifier);
 456 }
 457
 458 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 459                                              struct mm_struct *mm,
 460                                              unsigned long address)
 461 {
 462         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 463         int need_tlb_flush;
 464
 465         /*
 466          * When ->invalidate_page runs, the linux pte has been zapped
 467          * already but the page is still allocated until
 468          * ->invalidate_page returns. So if we increase the sequence
 469          * here the kvm page fault will notice if the spte can't be
 470          * established because the page is going to be freed. If
 471          * instead the kvm page fault establishes the spte before
 472          * ->invalidate_page runs, kvm_unmap_hva will release it
 473          * before returning.
 474          *
 475          * The sequence increase only need to be seen at spin_unlock
 476          * time, and not at spin_lock time.
 477          *
 478          * Increasing the sequence after the spin_unlock would be
 479          * unsafe because the kvm page fault could then establish the
 480          * pte after kvm_unmap_hva returned, without noticing the page
 481          * is going to be freed.
 482          */
 483         spin_lock(&kvm->mmu_lock);
 484         kvm->mmu_notifier_seq++;
 485         need_tlb_flush = kvm_unmap_hva(kvm, address);
 486         spin_unlock(&kvm->mmu_lock);
 487
 488         /* we've to flush the tlb before the pages can be freed */
 489         if (need_tlb_flush)
 490                 kvm_flush_remote_tlbs(kvm);
 491
 492 }
 493
 494 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 495                                                     struct mm_struct *mm,
 496                                                     unsigned long start,
 497                                                     unsigned long end)
 498 {
 499         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 500         int need_tlb_flush = 0;
 501
 502         spin_lock(&kvm->mmu_lock);
 503         /*
 504          * The count increase must become visible at unlock time as no
 505          * spte can be established without taking the mmu_lock and
 506          * count is also read inside the mmu_lock critical section.
 507          */
 508         kvm->mmu_notifier_count++;
 509         for (; start < end; start += PAGE_SIZE)
 510                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 511         spin_unlock(&kvm->mmu_lock);
 512
 513         /* we've to flush the tlb before the pages can be freed */
 514         if (need_tlb_flush)
 515                 kvm_flush_remote_tlbs(kvm);
 516 }
 517
 518 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 519                                                   struct mm_struct *mm,
 520                                                   unsigned long start,
 521                                                   unsigned long end)
 522 {
 523         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 524
 525         spin_lock(&kvm->mmu_lock);
 526         /*
 527          * This sequence increase will notify the kvm page fault that
 528          * the page that is going to be mapped in the spte could have
 529          * been freed.
 530          */
 531         kvm->mmu_notifier_seq++;
 532         /*
 533          * The above sequence increase must be visible before the
 534          * below count decrease but both values are read by the kvm
 535          * page fault under mmu_lock spinlock so we don't need to add
 536          * a smb_wmb() here in between the two.
 537          */
 538         kvm->mmu_notifier_count--;
 539         spin_unlock(&kvm->mmu_lock);
 540
 541         BUG_ON(kvm->mmu_notifier_count < 0);
 542 }
 543
 544 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 545                                               struct mm_struct *mm,
 546                                               unsigned long address)
 547 {
 548         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 549         int young;
 550
 551         spin_lock(&kvm->mmu_lock);
 552         young = kvm_age_hva(kvm, address);
 553         spin_unlock(&kvm->mmu_lock);
 554
 555         if (young)
 556                 kvm_flush_remote_tlbs(kvm);
 557
 558         return young;
 559 }
 560
 561 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 562         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 563         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 564         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 565         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 566 };
 567 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 568
 569 static struct kvm *kvm_create_vm(void)
 570 {
 571         struct kvm *kvm = kvm_arch_create_vm();
 572 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 573         struct page *page;
 574 #endif
 575
 576         if (IS_ERR(kvm))
 577                 goto out;
 578
 579 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 580         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 581         if (!page) {
 582                 kfree(kvm);
 583                 return ERR_PTR(-ENOMEM);
 584         }
 585         kvm->coalesced_mmio_ring =
 586                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 587 #endif
 588
 589 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 590         {
 591                 int err;
 592                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 593                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 594                 if (err) {
 595 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 596                         put_page(page);
 597 #endif
 598                         kfree(kvm);
 599                         return ERR_PTR(err);
 600                 }
 601         }
 602 #endif
 603
 604         kvm->mm = current->mm;
 605         atomic_inc(&kvm->mm->mm_count);
 606         spin_lock_init(&kvm->mmu_lock);
 607         kvm_io_bus_init(&kvm->pio_bus);
 608         mutex_init(&kvm->lock);
 609         kvm_io_bus_init(&kvm->mmio_bus);
 610         init_rwsem(&kvm->slots_lock);
 611         atomic_set(&kvm->users_count, 1);
 612         spin_lock(&kvm_lock);
 613         list_add(&kvm->vm_list, &vm_list);
 614         spin_unlock(&kvm_lock);
 615 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 616         kvm_coalesced_mmio_init(kvm);
 617 #endif
 618 out:
 619         return kvm;
 620 }
 621
 622 /*
 623  * Free any memory in @free but not in @dont.
 624  */
 625 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 626                                   struct kvm_memory_slot *dont)
 627 {
 628         if (!dont || free->rmap != dont->rmap)
 629                 vfree(free->rmap);
 630
 631         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 632                 vfree(free->dirty_bitmap);
 633
 634         if (!dont || free->lpage_info != dont->lpage_info)
 635                 vfree(free->lpage_info);
 636
 637         free->npages = 0;
 638         free->dirty_bitmap = NULL;
 639         free->rmap = NULL;
 640         free->lpage_info = NULL;
 641 }
 642
 643 void kvm_free_physmem(struct kvm *kvm)
 644 {
 645         int i;
 646
 647         for (i = 0; i < kvm->nmemslots; ++i)
 648                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 649 }
 650
 651 static void kvm_destroy_vm(struct kvm *kvm)
 652 {
 653         struct mm_struct *mm = kvm->mm;
 654
 655         spin_lock(&kvm_lock);
 656         list_del(&kvm->vm_list);
 657         spin_unlock(&kvm_lock);
 658         kvm_io_bus_destroy(&kvm->pio_bus);
 659         kvm_io_bus_destroy(&kvm->mmio_bus);
 660 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 661         if (kvm->coalesced_mmio_ring != NULL)
 662                 free_page((unsigned long)kvm->coalesced_mmio_ring);
 663 #endif
 664 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 665         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 666 #endif
 667         kvm_arch_destroy_vm(kvm);
 668         mmdrop(mm);
 669 }
 670
 671 void kvm_get_kvm(struct kvm *kvm)
 672 {
 673         atomic_inc(&kvm->users_count);
 674 }
 675 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 676
 677 void kvm_put_kvm(struct kvm *kvm)
 678 {
 679         if (atomic_dec_and_test(&kvm->users_count))
 680                 kvm_destroy_vm(kvm);
 681 }
 682 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 683
 684
 685 static int kvm_vm_release(struct inode *inode, struct file *filp)
 686 {
 687         struct kvm *kvm = filp->private_data;
 688
 689         kvm_put_kvm(kvm);
 690         return 0;
 691 }
 692
 693 /*
 694  * Allocate some memory and give it an address in the guest physical address
 695  * space.
 696  *
 697  * Discontiguous memory is allowed, mostly for framebuffers.
 698  *
 699  * Must be called holding mmap_sem for write.
 700  */
 701 int __kvm_set_memory_region(struct kvm *kvm,
 702                             struct kvm_userspace_memory_region *mem,
 703                             int user_alloc)
 704 {
 705         int r;
 706         gfn_t base_gfn;
 707         unsigned long npages;
 708         unsigned long i;
 709         struct kvm_memory_slot *memslot;
 710         struct kvm_memory_slot old, new;
 711
 712         r = -EINVAL;
 713         /* General sanity checks */
 714         if (mem->memory_size & (PAGE_SIZE - 1))
 715                 goto out;
 716         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 717                 goto out;
 718         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 719                 goto out;
 720         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 721                 goto out;
 722
 723         memslot = &kvm->memslots[mem->slot];
 724         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 725         npages = mem->memory_size >> PAGE_SHIFT;
 726
 727         if (!npages)
 728                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 729
 730         new = old = *memslot;
 731
 732         new.base_gfn = base_gfn;
 733         new.npages = npages;
 734         new.flags = mem->flags;
 735
 736         /* Disallow changing a memory slot's size. */
 737         r = -EINVAL;
 738         if (npages && old.npages && npages != old.npages)
 739                 goto out_free;
 740
 741         /* Check for overlaps */
 742         r = -EEXIST;
 743         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 744                 struct kvm_memory_slot *s = &kvm->memslots[i];
 745
 746                 if (s == memslot)
 747                         continue;
 748                 if (!((base_gfn + npages <= s->base_gfn) ||
 749                       (base_gfn >= s->base_gfn + s->npages)))
 750                         goto out_free;
 751         }
 752
 753         /* Free page dirty bitmap if unneeded */
 754         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 755                 new.dirty_bitmap = NULL;
 756
 757         r = -ENOMEM;
 758
 759         /* Allocate if a slot is being created */
 760 #ifndef CONFIG_S390
 761         if (npages && !new.rmap) {
 762                 new.rmap = vmalloc(npages * sizeof(struct page *));
 763
 764                 if (!new.rmap)
 765                         goto out_free;
 766
 767                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 768
 769                 new.user_alloc = user_alloc;
 770                 /*
 771                  * hva_to_rmmap() serialzies with the mmu_lock and to be
 772                  * safe it has to ignore memslots with !user_alloc &&
 773                  * !userspace_addr.
 774                  */
 775                 if (user_alloc)
 776                         new.userspace_addr = mem->userspace_addr;
 777                 else
 778                         new.userspace_addr = 0;
 779         }
 780         if (npages && !new.lpage_info) {
 781                 int largepages = npages / KVM_PAGES_PER_HPAGE;
 782                 if (npages % KVM_PAGES_PER_HPAGE)
 783                         largepages++;
 784                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 785                         largepages++;
 786
 787                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
 788
 789                 if (!new.lpage_info)
 790                         goto out_free;
 791
 792                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
 793
 794                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 795                         new.lpage_info[0].write_count = 1;
 796                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
 797                         new.lpage_info[largepages-1].write_count = 1;
 798         }
 799
 800         /* Allocate page dirty bitmap if needed */
 801         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 802                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 803
 804                 new.dirty_bitmap = vmalloc(dirty_bytes);
 805                 if (!new.dirty_bitmap)
 806                         goto out_free;
 807                 memset(new.dirty_bitmap, 0, dirty_bytes);
 808         }
 809 #endif /* not defined CONFIG_S390 */
 810
 811         if (!npages)
 812                 kvm_arch_flush_shadow(kvm);
 813
 814         spin_lock(&kvm->mmu_lock);
 815         if (mem->slot >= kvm->nmemslots)
 816                 kvm->nmemslots = mem->slot + 1;
 817
 818         *memslot = new;
 819         spin_unlock(&kvm->mmu_lock);
 820
 821         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
 822         if (r) {
 823                 spin_lock(&kvm->mmu_lock);
 824                 *memslot = old;
 825                 spin_unlock(&kvm->mmu_lock);
 826                 goto out_free;
 827         }
 828
 829         kvm_free_physmem_slot(&old, &new);
 830 #ifdef CONFIG_DMAR
 831         /* map the pages in iommu page table */
 832         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
 833         if (r)
 834                 goto out;
 835 #endif
 836         return 0;
 837
 838 out_free:
 839         kvm_free_physmem_slot(&new, &old);
 840 out:
 841         return r;
 842
 843 }
 844 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 845
 846 int kvm_set_memory_region(struct kvm *kvm,
 847                           struct kvm_userspace_memory_region *mem,
 848                           int user_alloc)
 849 {
 850         int r;
 851
 852         down_write(&kvm->slots_lock);
 853         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 854         up_write(&kvm->slots_lock);
 855         return r;
 856 }
 857 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 858
 859 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 860                                    struct
 861                                    kvm_userspace_memory_region *mem,
 862                                    int user_alloc)
 863 {
 864         if (mem->slot >= KVM_MEMORY_SLOTS)
 865                 return -EINVAL;
 866         return kvm_set_memory_region(kvm, mem, user_alloc);
 867 }
 868
 869 int kvm_get_dirty_log(struct kvm *kvm,
 870                         struct kvm_dirty_log *log, int *is_dirty)
 871 {
 872         struct kvm_memory_slot *memslot;
 873         int r, i;
 874         int n;
 875         unsigned long any = 0;
 876
 877         r = -EINVAL;
 878         if (log->slot >= KVM_MEMORY_SLOTS)
 879                 goto out;
 880
 881         memslot = &kvm->memslots[log->slot];
 882         r = -ENOENT;
 883         if (!memslot->dirty_bitmap)
 884                 goto out;
 885
 886         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 887
 888         for (i = 0; !any && i < n/sizeof(long); ++i)
 889                 any = memslot->dirty_bitmap[i];
 890
 891         r = -EFAULT;
 892         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 893                 goto out;
 894
 895         if (any)
 896                 *is_dirty = 1;
 897
 898         r = 0;
 899 out:
 900         return r;
 901 }
 902
 903 int is_error_page(struct page *page)
 904 {
 905         return page == bad_page;
 906 }
 907 EXPORT_SYMBOL_GPL(is_error_page);
 908
 909 int is_error_pfn(pfn_t pfn)
 910 {
 911         return pfn == bad_pfn;
 912 }
 913 EXPORT_SYMBOL_GPL(is_error_pfn);
 914
 915 static inline unsigned long bad_hva(void)
 916 {
 917         return PAGE_OFFSET;
 918 }
 919
 920 int kvm_is_error_hva(unsigned long addr)
 921 {
 922         return addr == bad_hva();
 923 }
 924 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 925
 926 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
 927 {
 928         int i;
 929
 930         for (i = 0; i < kvm->nmemslots; ++i) {
 931                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 932
 933                 if (gfn >= memslot->base_gfn
 934                     && gfn < memslot->base_gfn + memslot->npages)
 935                         return memslot;
 936         }
 937         return NULL;
 938 }
 939 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
 940
 941 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 942 {
 943         gfn = unalias_gfn(kvm, gfn);
 944         return gfn_to_memslot_unaliased(kvm, gfn);
 945 }
 946
 947 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 948 {
 949         int i;
 950
 951         gfn = unalias_gfn(kvm, gfn);
 952         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 953                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 954
 955                 if (gfn >= memslot->base_gfn
 956                     && gfn < memslot->base_gfn + memslot->npages)
 957                         return 1;
 958         }
 959         return 0;
 960 }
 961 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 962
 963 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 964 {
 965         struct kvm_memory_slot *slot;
 966
 967         gfn = unalias_gfn(kvm, gfn);
 968         slot = gfn_to_memslot_unaliased(kvm, gfn);
 969         if (!slot)
 970                 return bad_hva();
 971         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 972 }
 973 EXPORT_SYMBOL_GPL(gfn_to_hva);
 974
 975 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 976 {
 977         struct page *page[1];
 978         unsigned long addr;
 979         int npages;
 980         pfn_t pfn;
 981
 982         might_sleep();
 983
 984         addr = gfn_to_hva(kvm, gfn);
 985         if (kvm_is_error_hva(addr)) {
 986                 get_page(bad_page);
 987                 return page_to_pfn(bad_page);
 988         }
 989
 990         npages = get_user_pages_fast(addr, 1, 1, page);
 991
 992         if (unlikely(npages != 1)) {
 993                 struct vm_area_struct *vma;
 994
 995                 down_read(&current->mm->mmap_sem);
 996                 vma = find_vma(current->mm, addr);
 997
 998                 if (vma == NULL || addr < vma->vm_start ||
 999                     !(vma->vm_flags & VM_PFNMAP)) {
1000                         up_read(&current->mm->mmap_sem);
1001                         get_page(bad_page);
1002                         return page_to_pfn(bad_page);
1003                 }
1004
1005                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1006                 up_read(&current->mm->mmap_sem);
1007                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1008         } else
1009                 pfn = page_to_pfn(page[0]);
1010
1011         return pfn;
1012 }
1013
1014 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1015
1016 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1017 {
1018         pfn_t pfn;
1019
1020         pfn = gfn_to_pfn(kvm, gfn);
1021         if (!kvm_is_mmio_pfn(pfn))
1022                 return pfn_to_page(pfn);
1023
1024         WARN_ON(kvm_is_mmio_pfn(pfn));
1025
1026         get_page(bad_page);
1027         return bad_page;
1028 }
1029
1030 EXPORT_SYMBOL_GPL(gfn_to_page);
1031
1032 void kvm_release_page_clean(struct page *page)
1033 {
1034         kvm_release_pfn_clean(page_to_pfn(page));
1035 }
1036 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1037
1038 void kvm_release_pfn_clean(pfn_t pfn)
1039 {
1040         if (!kvm_is_mmio_pfn(pfn))
1041                 put_page(pfn_to_page(pfn));
1042 }
1043 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1044
1045 void kvm_release_page_dirty(struct page *page)
1046 {
1047         kvm_release_pfn_dirty(page_to_pfn(page));
1048 }
1049 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1050
1051 void kvm_release_pfn_dirty(pfn_t pfn)
1052 {
1053         kvm_set_pfn_dirty(pfn);
1054         kvm_release_pfn_clean(pfn);
1055 }
1056 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1057
1058 void kvm_set_page_dirty(struct page *page)
1059 {
1060         kvm_set_pfn_dirty(page_to_pfn(page));
1061 }
1062 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1063
1064 void kvm_set_pfn_dirty(pfn_t pfn)
1065 {
1066         if (!kvm_is_mmio_pfn(pfn)) {
1067                 struct page *page = pfn_to_page(pfn);
1068                 if (!PageReserved(page))
1069                         SetPageDirty(page);
1070         }
1071 }
1072 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1073
1074 void kvm_set_pfn_accessed(pfn_t pfn)
1075 {
1076         if (!kvm_is_mmio_pfn(pfn))
1077                 mark_page_accessed(pfn_to_page(pfn));
1078 }
1079 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1080
1081 void kvm_get_pfn(pfn_t pfn)
1082 {
1083         if (!kvm_is_mmio_pfn(pfn))
1084                 get_page(pfn_to_page(pfn));
1085 }
1086 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1087
1088 static int next_segment(unsigned long len, int offset)
1089 {
1090         if (len > PAGE_SIZE - offset)
1091                 return PAGE_SIZE - offset;
1092         else
1093                 return len;
1094 }
1095
1096 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1097                         int len)
1098 {
1099         int r;
1100         unsigned long addr;
1101
1102         addr = gfn_to_hva(kvm, gfn);
1103         if (kvm_is_error_hva(addr))
1104                 return -EFAULT;
1105         r = copy_from_user(data, (void __user *)addr + offset, len);
1106         if (r)
1107                 return -EFAULT;
1108         return 0;
1109 }
1110 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1111
1112 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1113 {
1114         gfn_t gfn = gpa >> PAGE_SHIFT;
1115         int seg;
1116         int offset = offset_in_page(gpa);
1117         int ret;
1118
1119         while ((seg = next_segment(len, offset)) != 0) {
1120                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1121                 if (ret < 0)
1122                         return ret;
1123                 offset = 0;
1124                 len -= seg;
1125                 data += seg;
1126                 ++gfn;
1127         }
1128         return 0;
1129 }
1130 EXPORT_SYMBOL_GPL(kvm_read_guest);
1131
1132 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1133                           unsigned long len)
1134 {
1135         int r;
1136         unsigned long addr;
1137         gfn_t gfn = gpa >> PAGE_SHIFT;
1138         int offset = offset_in_page(gpa);
1139
1140         addr = gfn_to_hva(kvm, gfn);
1141         if (kvm_is_error_hva(addr))
1142                 return -EFAULT;
1143         pagefault_disable();
1144         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1145         pagefault_enable();
1146         if (r)
1147                 return -EFAULT;
1148         return 0;
1149 }
1150 EXPORT_SYMBOL(kvm_read_guest_atomic);
1151
1152 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1153                          int offset, int len)
1154 {
1155         int r;
1156         unsigned long addr;
1157
1158         addr = gfn_to_hva(kvm, gfn);
1159         if (kvm_is_error_hva(addr))
1160                 return -EFAULT;
1161         r = copy_to_user((void __user *)addr + offset, data, len);
1162         if (r)
1163                 return -EFAULT;
1164         mark_page_dirty(kvm, gfn);
1165         return 0;
1166 }
1167 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1168
1169 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1170                     unsigned long len)
1171 {
1172         gfn_t gfn = gpa >> PAGE_SHIFT;
1173         int seg;
1174         int offset = offset_in_page(gpa);
1175         int ret;
1176
1177         while ((seg = next_segment(len, offset)) != 0) {
1178                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1179                 if (ret < 0)
1180                         return ret;
1181                 offset = 0;
1182                 len -= seg;
1183                 data += seg;
1184                 ++gfn;
1185         }
1186         return 0;
1187 }
1188
1189 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1190 {
1191         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1192 }
1193 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1194
1195 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1196 {
1197         gfn_t gfn = gpa >> PAGE_SHIFT;
1198         int seg;
1199         int offset = offset_in_page(gpa);
1200         int ret;
1201
1202         while ((seg = next_segment(len, offset)) != 0) {
1203                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1204                 if (ret < 0)
1205                         return ret;
1206                 offset = 0;
1207                 len -= seg;
1208                 ++gfn;
1209         }
1210         return 0;
1211 }
1212 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1213
1214 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1215 {
1216         struct kvm_memory_slot *memslot;
1217
1218         gfn = unalias_gfn(kvm, gfn);
1219         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1220         if (memslot && memslot->dirty_bitmap) {
1221                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1222
1223                 /* avoid RMW */
1224                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1225                         set_bit(rel_gfn, memslot->dirty_bitmap);
1226         }
1227 }
1228
1229 /*
1230  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1231  */
1232 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1233 {
1234         DEFINE_WAIT(wait);
1235
1236         for (;;) {
1237                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1238
1239                 if (kvm_cpu_has_interrupt(vcpu) ||
1240                     kvm_cpu_has_pending_timer(vcpu) ||
1241                     kvm_arch_vcpu_runnable(vcpu)) {
1242                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1243                         break;
1244                 }
1245                 if (signal_pending(current))
1246                         break;
1247
1248                 vcpu_put(vcpu);
1249                 schedule();
1250                 vcpu_load(vcpu);
1251         }
1252
1253         finish_wait(&vcpu->wq, &wait);
1254 }
1255
1256 void kvm_resched(struct kvm_vcpu *vcpu)
1257 {
1258         if (!need_resched())
1259                 return;
1260         cond_resched();
1261 }
1262 EXPORT_SYMBOL_GPL(kvm_resched);
1263
1264 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1265 {
1266         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1267         struct page *page;
1268
1269         if (vmf->pgoff == 0)
1270                 page = virt_to_page(vcpu->run);
1271 #ifdef CONFIG_X86
1272         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1273                 page = virt_to_page(vcpu->arch.pio_data);
1274 #endif
1275 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1276         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1277                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1278 #endif
1279         else
1280                 return VM_FAULT_SIGBUS;
1281         get_page(page);
1282         vmf->page = page;
1283         return 0;
1284 }
1285
1286 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1287         .fault = kvm_vcpu_fault,
1288 };
1289
1290 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1291 {
1292         vma->vm_ops = &kvm_vcpu_vm_ops;
1293         return 0;
1294 }
1295
1296 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1297 {
1298         struct kvm_vcpu *vcpu = filp->private_data;
1299
1300         kvm_put_kvm(vcpu->kvm);
1301         return 0;
1302 }
1303
1304 static const struct file_operations kvm_vcpu_fops = {
1305         .release        = kvm_vcpu_release,
1306         .unlocked_ioctl = kvm_vcpu_ioctl,
1307         .compat_ioctl   = kvm_vcpu_ioctl,
1308         .mmap           = kvm_vcpu_mmap,
1309 };
1310
1311 /*
1312  * Allocates an inode for the vcpu.
1313  */
1314 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1315 {
1316         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1317         if (fd < 0)
1318                 kvm_put_kvm(vcpu->kvm);
1319         return fd;
1320 }
1321
1322 /*
1323  * Creates some virtual cpus.  Good luck creating more than one.
1324  */
1325 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1326 {
1327         int r;
1328         struct kvm_vcpu *vcpu;
1329
1330         if (!valid_vcpu(n))
1331                 return -EINVAL;
1332
1333         vcpu = kvm_arch_vcpu_create(kvm, n);
1334         if (IS_ERR(vcpu))
1335                 return PTR_ERR(vcpu);
1336
1337         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1338
1339         r = kvm_arch_vcpu_setup(vcpu);
1340         if (r)
1341                 return r;
1342
1343         mutex_lock(&kvm->lock);
1344         if (kvm->vcpus[n]) {
1345                 r = -EEXIST;
1346                 goto vcpu_destroy;
1347         }
1348         kvm->vcpus[n] = vcpu;
1349         mutex_unlock(&kvm->lock);
1350
1351         /* Now it's all set up, let userspace reach it */
1352         kvm_get_kvm(kvm);
1353         r = create_vcpu_fd(vcpu);
1354         if (r < 0)
1355                 goto unlink;
1356         return r;
1357
1358 unlink:
1359         mutex_lock(&kvm->lock);
1360         kvm->vcpus[n] = NULL;
1361 vcpu_destroy:
1362         mutex_unlock(&kvm->lock);
1363         kvm_arch_vcpu_destroy(vcpu);
1364         return r;
1365 }
1366
1367 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1368 {
1369         if (sigset) {
1370                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1371                 vcpu->sigset_active = 1;
1372                 vcpu->sigset = *sigset;
1373         } else
1374                 vcpu->sigset_active = 0;
1375         return 0;
1376 }
1377
1378 static long kvm_vcpu_ioctl(struct file *filp,
1379                            unsigned int ioctl, unsigned long arg)
1380 {
1381         struct kvm_vcpu *vcpu = filp->private_data;
1382         void __user *argp = (void __user *)arg;
1383         int r;
1384         struct kvm_fpu *fpu = NULL;
1385         struct kvm_sregs *kvm_sregs = NULL;
1386
1387         if (vcpu->kvm->mm != current->mm)
1388                 return -EIO;
1389         switch (ioctl) {
1390         case KVM_RUN:
1391                 r = -EINVAL;
1392                 if (arg)
1393                         goto out;
1394                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1395                 break;
1396         case KVM_GET_REGS: {
1397                 struct kvm_regs *kvm_regs;
1398
1399                 r = -ENOMEM;
1400                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1401                 if (!kvm_regs)
1402                         goto out;
1403                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1404                 if (r)
1405                         goto out_free1;
1406                 r = -EFAULT;
1407                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1408                         goto out_free1;
1409                 r = 0;
1410 out_free1:
1411                 kfree(kvm_regs);
1412                 break;
1413         }
1414         case KVM_SET_REGS: {
1415                 struct kvm_regs *kvm_regs;
1416
1417                 r = -ENOMEM;
1418                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1419                 if (!kvm_regs)
1420                         goto out;
1421                 r = -EFAULT;
1422                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1423                         goto out_free2;
1424                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1425                 if (r)
1426                         goto out_free2;
1427                 r = 0;
1428 out_free2:
1429                 kfree(kvm_regs);
1430                 break;
1431         }
1432         case KVM_GET_SREGS: {
1433                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1434                 r = -ENOMEM;
1435                 if (!kvm_sregs)
1436                         goto out;
1437                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1438                 if (r)
1439                         goto out;
1440                 r = -EFAULT;
1441                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1442                         goto out;
1443                 r = 0;
1444                 break;
1445         }
1446         case KVM_SET_SREGS: {
1447                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1448                 r = -ENOMEM;
1449                 if (!kvm_sregs)
1450                         goto out;
1451                 r = -EFAULT;
1452                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1453                         goto out;
1454                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1455                 if (r)
1456                         goto out;
1457                 r = 0;
1458                 break;
1459         }
1460         case KVM_GET_MP_STATE: {
1461                 struct kvm_mp_state mp_state;
1462
1463                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1464                 if (r)
1465                         goto out;
1466                 r = -EFAULT;
1467                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1468                         goto out;
1469                 r = 0;
1470                 break;
1471         }
1472         case KVM_SET_MP_STATE: {
1473                 struct kvm_mp_state mp_state;
1474
1475                 r = -EFAULT;
1476                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1477                         goto out;
1478                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1479                 if (r)
1480                         goto out;
1481                 r = 0;
1482                 break;
1483         }
1484         case KVM_TRANSLATE: {
1485                 struct kvm_translation tr;
1486
1487                 r = -EFAULT;
1488                 if (copy_from_user(&tr, argp, sizeof tr))
1489                         goto out;
1490                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1491                 if (r)
1492                         goto out;
1493                 r = -EFAULT;
1494                 if (copy_to_user(argp, &tr, sizeof tr))
1495                         goto out;
1496                 r = 0;
1497                 break;
1498         }
1499         case KVM_DEBUG_GUEST: {
1500                 struct kvm_debug_guest dbg;
1501
1502                 r = -EFAULT;
1503                 if (copy_from_user(&dbg, argp, sizeof dbg))
1504                         goto out;
1505                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1506                 if (r)
1507                         goto out;
1508                 r = 0;
1509                 break;
1510         }
1511         case KVM_SET_SIGNAL_MASK: {
1512                 struct kvm_signal_mask __user *sigmask_arg = argp;
1513                 struct kvm_signal_mask kvm_sigmask;
1514                 sigset_t sigset, *p;
1515
1516                 p = NULL;
1517                 if (argp) {
1518                         r = -EFAULT;
1519                         if (copy_from_user(&kvm_sigmask, argp,
1520                                            sizeof kvm_sigmask))
1521                                 goto out;
1522                         r = -EINVAL;
1523                         if (kvm_sigmask.len != sizeof sigset)
1524                                 goto out;
1525                         r = -EFAULT;
1526                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1527                                            sizeof sigset))
1528                                 goto out;
1529                         p = &sigset;
1530                 }
1531                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1532                 break;
1533         }
1534         case KVM_GET_FPU: {
1535                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1536                 r = -ENOMEM;
1537                 if (!fpu)
1538                         goto out;
1539                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1540                 if (r)
1541                         goto out;
1542                 r = -EFAULT;
1543                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1544                         goto out;
1545                 r = 0;
1546                 break;
1547         }
1548         case KVM_SET_FPU: {
1549                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1550                 r = -ENOMEM;
1551                 if (!fpu)
1552                         goto out;
1553                 r = -EFAULT;
1554                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1555                         goto out;
1556                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1557                 if (r)
1558                         goto out;
1559                 r = 0;
1560                 break;
1561         }
1562         default:
1563                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1564         }
1565 out:
1566         kfree(fpu);
1567         kfree(kvm_sregs);
1568         return r;
1569 }
1570
1571 static long kvm_vm_ioctl(struct file *filp,
1572                            unsigned int ioctl, unsigned long arg)
1573 {
1574         struct kvm *kvm = filp->private_data;
1575         void __user *argp = (void __user *)arg;
1576         int r;
1577
1578         if (kvm->mm != current->mm)
1579                 return -EIO;
1580         switch (ioctl) {
1581         case KVM_CREATE_VCPU:
1582                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1583                 if (r < 0)
1584                         goto out;
1585                 break;
1586         case KVM_SET_USER_MEMORY_REGION: {
1587                 struct kvm_userspace_memory_region kvm_userspace_mem;
1588
1589                 r = -EFAULT;
1590                 if (copy_from_user(&kvm_userspace_mem, argp,
1591                                                 sizeof kvm_userspace_mem))
1592                         goto out;
1593
1594                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1595                 if (r)
1596                         goto out;
1597                 break;
1598         }
1599         case KVM_GET_DIRTY_LOG: {
1600                 struct kvm_dirty_log log;
1601
1602                 r = -EFAULT;
1603                 if (copy_from_user(&log, argp, sizeof log))
1604                         goto out;
1605                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1606                 if (r)
1607                         goto out;
1608                 break;
1609         }
1610 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1611         case KVM_REGISTER_COALESCED_MMIO: {
1612                 struct kvm_coalesced_mmio_zone zone;
1613                 r = -EFAULT;
1614                 if (copy_from_user(&zone, argp, sizeof zone))
1615                         goto out;
1616                 r = -ENXIO;
1617                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1618                 if (r)
1619                         goto out;
1620                 r = 0;
1621                 break;
1622         }
1623         case KVM_UNREGISTER_COALESCED_MMIO: {
1624                 struct kvm_coalesced_mmio_zone zone;
1625                 r = -EFAULT;
1626                 if (copy_from_user(&zone, argp, sizeof zone))
1627                         goto out;
1628                 r = -ENXIO;
1629                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1630                 if (r)
1631                         goto out;
1632                 r = 0;
1633                 break;
1634         }
1635 #endif
1636 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1637         case KVM_ASSIGN_PCI_DEVICE: {
1638                 struct kvm_assigned_pci_dev assigned_dev;
1639
1640                 r = -EFAULT;
1641                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1642                         goto out;
1643                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1644                 if (r)
1645                         goto out;
1646                 break;
1647         }
1648         case KVM_ASSIGN_IRQ: {
1649                 struct kvm_assigned_irq assigned_irq;
1650
1651                 r = -EFAULT;
1652                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1653                         goto out;
1654                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1655                 if (r)
1656                         goto out;
1657                 break;
1658         }
1659 #endif
1660         default:
1661                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1662         }
1663 out:
1664         return r;
1665 }
1666
1667 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1668 {
1669         struct page *page[1];
1670         unsigned long addr;
1671         int npages;
1672         gfn_t gfn = vmf->pgoff;
1673         struct kvm *kvm = vma->vm_file->private_data;
1674
1675         addr = gfn_to_hva(kvm, gfn);
1676         if (kvm_is_error_hva(addr))
1677                 return VM_FAULT_SIGBUS;
1678
1679         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1680                                 NULL);
1681         if (unlikely(npages != 1))
1682                 return VM_FAULT_SIGBUS;
1683
1684         vmf->page = page[0];
1685         return 0;
1686 }
1687
1688 static struct vm_operations_struct kvm_vm_vm_ops = {
1689         .fault = kvm_vm_fault,
1690 };
1691
1692 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1693 {
1694         vma->vm_ops = &kvm_vm_vm_ops;
1695         return 0;
1696 }
1697
1698 static const struct file_operations kvm_vm_fops = {
1699         .release        = kvm_vm_release,
1700         .unlocked_ioctl = kvm_vm_ioctl,
1701         .compat_ioctl   = kvm_vm_ioctl,
1702         .mmap           = kvm_vm_mmap,
1703 };
1704
1705 static int kvm_dev_ioctl_create_vm(void)
1706 {
1707         int fd;
1708         struct kvm *kvm;
1709
1710         kvm = kvm_create_vm();
1711         if (IS_ERR(kvm))
1712                 return PTR_ERR(kvm);
1713         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1714         if (fd < 0)
1715                 kvm_put_kvm(kvm);
1716
1717         return fd;
1718 }
1719
1720 static long kvm_dev_ioctl(struct file *filp,
1721                           unsigned int ioctl, unsigned long arg)
1722 {
1723         long r = -EINVAL;
1724
1725         switch (ioctl) {
1726         case KVM_GET_API_VERSION:
1727                 r = -EINVAL;
1728                 if (arg)
1729                         goto out;
1730                 r = KVM_API_VERSION;
1731                 break;
1732         case KVM_CREATE_VM:
1733                 r = -EINVAL;
1734                 if (arg)
1735                         goto out;
1736                 r = kvm_dev_ioctl_create_vm();
1737                 break;
1738         case KVM_CHECK_EXTENSION:
1739                 r = kvm_dev_ioctl_check_extension(arg);
1740                 break;
1741         case KVM_GET_VCPU_MMAP_SIZE:
1742                 r = -EINVAL;
1743                 if (arg)
1744                         goto out;
1745                 r = PAGE_SIZE;     /* struct kvm_run */
1746 #ifdef CONFIG_X86
1747                 r += PAGE_SIZE;    /* pio data page */
1748 #endif
1749 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1750                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1751 #endif
1752                 break;
1753         case KVM_TRACE_ENABLE:
1754         case KVM_TRACE_PAUSE:
1755         case KVM_TRACE_DISABLE:
1756                 r = kvm_trace_ioctl(ioctl, arg);
1757                 break;
1758         default:
1759                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1760         }
1761 out:
1762         return r;
1763 }
1764
1765 static struct file_operations kvm_chardev_ops = {
1766         .unlocked_ioctl = kvm_dev_ioctl,
1767         .compat_ioctl   = kvm_dev_ioctl,
1768 };
1769
1770 static struct miscdevice kvm_dev = {
1771         KVM_MINOR,
1772         "kvm",
1773         &kvm_chardev_ops,
1774 };
1775
1776 static void hardware_enable(void *junk)
1777 {
1778         int cpu = raw_smp_processor_id();
1779
1780         if (cpu_isset(cpu, cpus_hardware_enabled))
1781                 return;
1782         cpu_set(cpu, cpus_hardware_enabled);
1783         kvm_arch_hardware_enable(NULL);
1784 }
1785
1786 static void hardware_disable(void *junk)
1787 {
1788         int cpu = raw_smp_processor_id();
1789
1790         if (!cpu_isset(cpu, cpus_hardware_enabled))
1791                 return;
1792         cpu_clear(cpu, cpus_hardware_enabled);
1793         kvm_arch_hardware_disable(NULL);
1794 }
1795
1796 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1797                            void *v)
1798 {
1799         int cpu = (long)v;
1800
1801         val &= ~CPU_TASKS_FROZEN;
1802         switch (val) {
1803         case CPU_DYING:
1804                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1805                        cpu);
1806                 hardware_disable(NULL);
1807                 break;
1808         case CPU_UP_CANCELED:
1809                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1810                        cpu);
1811                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1812                 break;
1813         case CPU_ONLINE:
1814                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1815                        cpu);
1816                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1817                 break;
1818         }
1819         return NOTIFY_OK;
1820 }
1821
1822
1823 asmlinkage void kvm_handle_fault_on_reboot(void)
1824 {
1825         if (kvm_rebooting)
1826                 /* spin while reset goes on */
1827                 while (true)
1828                         ;
1829         /* Fault while not rebooting.  We want the trace. */
1830         BUG();
1831 }
1832 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1833
1834 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1835                       void *v)
1836 {
1837         if (val == SYS_RESTART) {
1838                 /*
1839                  * Some (well, at least mine) BIOSes hang on reboot if
1840                  * in vmx root mode.
1841                  */
1842                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1843                 kvm_rebooting = true;
1844                 on_each_cpu(hardware_disable, NULL, 1);
1845         }
1846         return NOTIFY_OK;
1847 }
1848
1849 static struct notifier_block kvm_reboot_notifier = {
1850         .notifier_call = kvm_reboot,
1851         .priority = 0,
1852 };
1853
1854 void kvm_io_bus_init(struct kvm_io_bus *bus)
1855 {
1856         memset(bus, 0, sizeof(*bus));
1857 }
1858
1859 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1860 {
1861         int i;
1862
1863         for (i = 0; i < bus->dev_count; i++) {
1864                 struct kvm_io_device *pos = bus->devs[i];
1865
1866                 kvm_iodevice_destructor(pos);
1867         }
1868 }
1869
1870 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
1871                                           gpa_t addr, int len, int is_write)
1872 {
1873         int i;
1874
1875         for (i = 0; i < bus->dev_count; i++) {
1876                 struct kvm_io_device *pos = bus->devs[i];
1877
1878                 if (pos->in_range(pos, addr, len, is_write))
1879                         return pos;
1880         }
1881
1882         return NULL;
1883 }
1884
1885 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1886 {
1887         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1888
1889         bus->devs[bus->dev_count++] = dev;
1890 }
1891
1892 static struct notifier_block kvm_cpu_notifier = {
1893         .notifier_call = kvm_cpu_hotplug,
1894         .priority = 20, /* must be > scheduler priority */
1895 };
1896
1897 static int vm_stat_get(void *_offset, u64 *val)
1898 {
1899         unsigned offset = (long)_offset;
1900         struct kvm *kvm;
1901
1902         *val = 0;
1903         spin_lock(&kvm_lock);
1904         list_for_each_entry(kvm, &vm_list, vm_list)
1905                 *val += *(u32 *)((void *)kvm + offset);
1906         spin_unlock(&kvm_lock);
1907         return 0;
1908 }
1909
1910 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1911
1912 static int vcpu_stat_get(void *_offset, u64 *val)
1913 {
1914         unsigned offset = (long)_offset;
1915         struct kvm *kvm;
1916         struct kvm_vcpu *vcpu;
1917         int i;
1918
1919         *val = 0;
1920         spin_lock(&kvm_lock);
1921         list_for_each_entry(kvm, &vm_list, vm_list)
1922                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1923                         vcpu = kvm->vcpus[i];
1924                         if (vcpu)
1925                                 *val += *(u32 *)((void *)vcpu + offset);
1926                 }
1927         spin_unlock(&kvm_lock);
1928         return 0;
1929 }
1930
1931 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1932
1933 static struct file_operations *stat_fops[] = {
1934         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1935         [KVM_STAT_VM]   = &vm_stat_fops,
1936 };
1937
1938 static void kvm_init_debug(void)
1939 {
1940         struct kvm_stats_debugfs_item *p;
1941
1942         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1943         for (p = debugfs_entries; p->name; ++p)
1944                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1945                                                 (void *)(long)p->offset,
1946                                                 stat_fops[p->kind]);
1947 }
1948
1949 static void kvm_exit_debug(void)
1950 {
1951         struct kvm_stats_debugfs_item *p;
1952
1953         for (p = debugfs_entries; p->name; ++p)
1954                 debugfs_remove(p->dentry);
1955         debugfs_remove(kvm_debugfs_dir);
1956 }
1957
1958 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1959 {
1960         hardware_disable(NULL);
1961         return 0;
1962 }
1963
1964 static int kvm_resume(struct sys_device *dev)
1965 {
1966         hardware_enable(NULL);
1967         return 0;
1968 }
1969
1970 static struct sysdev_class kvm_sysdev_class = {
1971         .name = "kvm",
1972         .suspend = kvm_suspend,
1973         .resume = kvm_resume,
1974 };
1975
1976 static struct sys_device kvm_sysdev = {
1977         .id = 0,
1978         .cls = &kvm_sysdev_class,
1979 };
1980
1981 struct page *bad_page;
1982 pfn_t bad_pfn;
1983
1984 static inline
1985 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1986 {
1987         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1988 }
1989
1990 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1991 {
1992         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1993
1994         kvm_arch_vcpu_load(vcpu, cpu);
1995 }
1996
1997 static void kvm_sched_out(struct preempt_notifier *pn,
1998                           struct task_struct *next)
1999 {
2000         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2001
2002         kvm_arch_vcpu_put(vcpu);
2003 }
2004
2005 int kvm_init(void *opaque, unsigned int vcpu_size,
2006                   struct module *module)
2007 {
2008         int r;
2009         int cpu;
2010
2011         kvm_init_debug();
2012
2013         r = kvm_arch_init(opaque);
2014         if (r)
2015                 goto out_fail;
2016
2017         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2018
2019         if (bad_page == NULL) {
2020                 r = -ENOMEM;
2021                 goto out;
2022         }
2023
2024         bad_pfn = page_to_pfn(bad_page);
2025
2026         r = kvm_arch_hardware_setup();
2027         if (r < 0)
2028                 goto out_free_0;
2029
2030         for_each_online_cpu(cpu) {
2031                 smp_call_function_single(cpu,
2032                                 kvm_arch_check_processor_compat,
2033                                 &r, 1);
2034                 if (r < 0)
2035                         goto out_free_1;
2036         }
2037
2038         on_each_cpu(hardware_enable, NULL, 1);
2039         r = register_cpu_notifier(&kvm_cpu_notifier);
2040         if (r)
2041                 goto out_free_2;
2042         register_reboot_notifier(&kvm_reboot_notifier);
2043
2044         r = sysdev_class_register(&kvm_sysdev_class);
2045         if (r)
2046                 goto out_free_3;
2047
2048         r = sysdev_register(&kvm_sysdev);
2049         if (r)
2050                 goto out_free_4;
2051
2052         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2053         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2054                                            __alignof__(struct kvm_vcpu),
2055                                            0, NULL);
2056         if (!kvm_vcpu_cache) {
2057                 r = -ENOMEM;
2058                 goto out_free_5;
2059         }
2060
2061         kvm_chardev_ops.owner = module;
2062
2063         r = misc_register(&kvm_dev);
2064         if (r) {
2065                 printk(KERN_ERR "kvm: misc device register failed\n");
2066                 goto out_free;
2067         }
2068
2069         kvm_preempt_ops.sched_in = kvm_sched_in;
2070         kvm_preempt_ops.sched_out = kvm_sched_out;
2071
2072         return 0;
2073
2074 out_free:
2075         kmem_cache_destroy(kvm_vcpu_cache);
2076 out_free_5:
2077         sysdev_unregister(&kvm_sysdev);
2078 out_free_4:
2079         sysdev_class_unregister(&kvm_sysdev_class);
2080 out_free_3:
2081         unregister_reboot_notifier(&kvm_reboot_notifier);
2082         unregister_cpu_notifier(&kvm_cpu_notifier);
2083 out_free_2:
2084         on_each_cpu(hardware_disable, NULL, 1);
2085 out_free_1:
2086         kvm_arch_hardware_unsetup();
2087 out_free_0:
2088         __free_page(bad_page);
2089 out:
2090         kvm_arch_exit();
2091         kvm_exit_debug();
2092 out_fail:
2093         return r;
2094 }
2095 EXPORT_SYMBOL_GPL(kvm_init);
2096
2097 void kvm_exit(void)
2098 {
2099         kvm_trace_cleanup();
2100         misc_deregister(&kvm_dev);
2101         kmem_cache_destroy(kvm_vcpu_cache);
2102         sysdev_unregister(&kvm_sysdev);
2103         sysdev_class_unregister(&kvm_sysdev_class);
2104         unregister_reboot_notifier(&kvm_reboot_notifier);
2105         unregister_cpu_notifier(&kvm_cpu_notifier);
2106         on_each_cpu(hardware_disable, NULL, 1);
2107         kvm_arch_hardware_unsetup();
2108         kvm_arch_exit();
2109         kvm_exit_debug();
2110         __free_page(bad_page);
2111 }
2112 EXPORT_SYMBOL_GPL(kvm_exit);