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         if (adev->irq_requested_type & KVM_ASSIGNED_DEV_GUEST_INTX) {
 184                 adev->guest_irq = airq->guest_irq;
 185                 adev->ack_notifier.gsi = airq->guest_irq;
 186                 return 0;
 187         }
 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                 adev->guest_irq = airq->guest_irq;
 198                 adev->ack_notifier.gsi = airq->guest_irq;
 199
 200                 /* Even though this is PCI, we don't want to use shared
 201                  * interrupts. Sharing host devices with guest-assigned devices
 202                  * on the same interrupt line is not a happy situation: there
 203                  * are going to be long delays in accepting, acking, etc.
 204                  */
 205                 if (request_irq(adev->host_irq, kvm_assigned_dev_intr,
 206                                 0, "kvm_assigned_intx_device", (void *)adev))
 207                         return -EIO;
 208         }
 209
 210         adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_INTX |
 211                                    KVM_ASSIGNED_DEV_HOST_INTX;
 212         return 0;
 213 }
 214
 215 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 216                                    struct kvm_assigned_irq
 217                                    *assigned_irq)
 218 {
 219         int r = 0;
 220         struct kvm_assigned_dev_kernel *match;
 221
 222         mutex_lock(&kvm->lock);
 223
 224         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 225                                       assigned_irq->assigned_dev_id);
 226         if (!match) {
 227                 mutex_unlock(&kvm->lock);
 228                 return -EINVAL;
 229         }
 230
 231         if (!match->irq_requested_type) {
 232                 INIT_WORK(&match->interrupt_work,
 233                                 kvm_assigned_dev_interrupt_work_handler);
 234                 if (irqchip_in_kernel(kvm)) {
 235                         /* Register ack nofitier */
 236                         match->ack_notifier.gsi = -1;
 237                         match->ack_notifier.irq_acked =
 238                                         kvm_assigned_dev_ack_irq;
 239                         kvm_register_irq_ack_notifier(kvm,
 240                                         &match->ack_notifier);
 241
 242                         /* Request IRQ source ID */
 243                         r = kvm_request_irq_source_id(kvm);
 244                         if (r < 0)
 245                                 goto out_release;
 246                         else
 247                                 match->irq_source_id = r;
 248                 }
 249         }
 250
 251         r = assigned_device_update_intx(kvm, match, assigned_irq);
 252         if (r)
 253                 goto out_release;
 254
 255         mutex_unlock(&kvm->lock);
 256         return r;
 257 out_release:
 258         mutex_unlock(&kvm->lock);
 259         kvm_free_assigned_device(kvm, match);
 260         return r;
 261 }
 262
 263 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 264                                       struct kvm_assigned_pci_dev *assigned_dev)
 265 {
 266         int r = 0;
 267         struct kvm_assigned_dev_kernel *match;
 268         struct pci_dev *dev;
 269
 270         mutex_lock(&kvm->lock);
 271
 272         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 273                                       assigned_dev->assigned_dev_id);
 274         if (match) {
 275                 /* device already assigned */
 276                 r = -EINVAL;
 277                 goto out;
 278         }
 279
 280         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 281         if (match == NULL) {
 282                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 283                        __func__);
 284                 r = -ENOMEM;
 285                 goto out;
 286         }
 287         dev = pci_get_bus_and_slot(assigned_dev->busnr,
 288                                    assigned_dev->devfn);
 289         if (!dev) {
 290                 printk(KERN_INFO "%s: host device not found\n", __func__);
 291                 r = -EINVAL;
 292                 goto out_free;
 293         }
 294         if (pci_enable_device(dev)) {
 295                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 296                 r = -EBUSY;
 297                 goto out_put;
 298         }
 299         r = pci_request_regions(dev, "kvm_assigned_device");
 300         if (r) {
 301                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 302                        __func__);
 303                 goto out_disable;
 304         }
 305
 306         pci_reset_function(dev);
 307
 308         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 309         match->host_busnr = assigned_dev->busnr;
 310         match->host_devfn = assigned_dev->devfn;
 311         match->dev = dev;
 312
 313         match->kvm = kvm;
 314
 315         list_add(&match->list, &kvm->arch.assigned_dev_head);
 316
 317         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
 318                 r = kvm_iommu_map_guest(kvm, match);
 319                 if (r)
 320                         goto out_list_del;
 321         }
 322
 323 out:
 324         mutex_unlock(&kvm->lock);
 325         return r;
 326 out_list_del:
 327         list_del(&match->list);
 328         pci_release_regions(dev);
 329 out_disable:
 330         pci_disable_device(dev);
 331 out_put:
 332         pci_dev_put(dev);
 333 out_free:
 334         kfree(match);
 335         mutex_unlock(&kvm->lock);
 336         return r;
 337 }
 338 #endif
 339
 340 static inline int valid_vcpu(int n)
 341 {
 342         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 343 }
 344
 345 inline int kvm_is_mmio_pfn(pfn_t pfn)
 346 {
 347         if (pfn_valid(pfn))
 348                 return PageReserved(pfn_to_page(pfn));
 349
 350         return true;
 351 }
 352
 353 /*
 354  * Switches to specified vcpu, until a matching vcpu_put()
 355  */
 356 void vcpu_load(struct kvm_vcpu *vcpu)
 357 {
 358         int cpu;
 359
 360         mutex_lock(&vcpu->mutex);
 361         cpu = get_cpu();
 362         preempt_notifier_register(&vcpu->preempt_notifier);
 363         kvm_arch_vcpu_load(vcpu, cpu);
 364         put_cpu();
 365 }
 366
 367 void vcpu_put(struct kvm_vcpu *vcpu)
 368 {
 369         preempt_disable();
 370         kvm_arch_vcpu_put(vcpu);
 371         preempt_notifier_unregister(&vcpu->preempt_notifier);
 372         preempt_enable();
 373         mutex_unlock(&vcpu->mutex);
 374 }
 375
 376 static void ack_flush(void *_completed)
 377 {
 378 }
 379
 380 void kvm_flush_remote_tlbs(struct kvm *kvm)
 381 {
 382         int i, cpu, me;
 383         cpumask_t cpus;
 384         struct kvm_vcpu *vcpu;
 385
 386         me = get_cpu();
 387         cpus_clear(cpus);
 388         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 389                 vcpu = kvm->vcpus[i];
 390                 if (!vcpu)
 391                         continue;
 392                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 393                         continue;
 394                 cpu = vcpu->cpu;
 395                 if (cpu != -1 && cpu != me)
 396                         cpu_set(cpu, cpus);
 397         }
 398         if (cpus_empty(cpus))
 399                 goto out;
 400         ++kvm->stat.remote_tlb_flush;
 401         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 402 out:
 403         put_cpu();
 404 }
 405
 406 void kvm_reload_remote_mmus(struct kvm *kvm)
 407 {
 408         int i, cpu, me;
 409         cpumask_t cpus;
 410         struct kvm_vcpu *vcpu;
 411
 412         me = get_cpu();
 413         cpus_clear(cpus);
 414         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 415                 vcpu = kvm->vcpus[i];
 416                 if (!vcpu)
 417                         continue;
 418                 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
 419                         continue;
 420                 cpu = vcpu->cpu;
 421                 if (cpu != -1 && cpu != me)
 422                         cpu_set(cpu, cpus);
 423         }
 424         if (cpus_empty(cpus))
 425                 goto out;
 426         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 427 out:
 428         put_cpu();
 429 }
 430
 431
 432 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 433 {
 434         struct page *page;
 435         int r;
 436
 437         mutex_init(&vcpu->mutex);
 438         vcpu->cpu = -1;
 439         vcpu->kvm = kvm;
 440         vcpu->vcpu_id = id;
 441         init_waitqueue_head(&vcpu->wq);
 442
 443         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 444         if (!page) {
 445                 r = -ENOMEM;
 446                 goto fail;
 447         }
 448         vcpu->run = page_address(page);
 449
 450         r = kvm_arch_vcpu_init(vcpu);
 451         if (r < 0)
 452                 goto fail_free_run;
 453         return 0;
 454
 455 fail_free_run:
 456         free_page((unsigned long)vcpu->run);
 457 fail:
 458         return r;
 459 }
 460 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 461
 462 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 463 {
 464         kvm_arch_vcpu_uninit(vcpu);
 465         free_page((unsigned long)vcpu->run);
 466 }
 467 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 468
 469 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 470 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 471 {
 472         return container_of(mn, struct kvm, mmu_notifier);
 473 }
 474
 475 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 476                                              struct mm_struct *mm,
 477                                              unsigned long address)
 478 {
 479         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 480         int need_tlb_flush;
 481
 482         /*
 483          * When ->invalidate_page runs, the linux pte has been zapped
 484          * already but the page is still allocated until
 485          * ->invalidate_page returns. So if we increase the sequence
 486          * here the kvm page fault will notice if the spte can't be
 487          * established because the page is going to be freed. If
 488          * instead the kvm page fault establishes the spte before
 489          * ->invalidate_page runs, kvm_unmap_hva will release it
 490          * before returning.
 491          *
 492          * The sequence increase only need to be seen at spin_unlock
 493          * time, and not at spin_lock time.
 494          *
 495          * Increasing the sequence after the spin_unlock would be
 496          * unsafe because the kvm page fault could then establish the
 497          * pte after kvm_unmap_hva returned, without noticing the page
 498          * is going to be freed.
 499          */
 500         spin_lock(&kvm->mmu_lock);
 501         kvm->mmu_notifier_seq++;
 502         need_tlb_flush = kvm_unmap_hva(kvm, address);
 503         spin_unlock(&kvm->mmu_lock);
 504
 505         /* we've to flush the tlb before the pages can be freed */
 506         if (need_tlb_flush)
 507                 kvm_flush_remote_tlbs(kvm);
 508
 509 }
 510
 511 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 512                                                     struct mm_struct *mm,
 513                                                     unsigned long start,
 514                                                     unsigned long end)
 515 {
 516         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 517         int need_tlb_flush = 0;
 518
 519         spin_lock(&kvm->mmu_lock);
 520         /*
 521          * The count increase must become visible at unlock time as no
 522          * spte can be established without taking the mmu_lock and
 523          * count is also read inside the mmu_lock critical section.
 524          */
 525         kvm->mmu_notifier_count++;
 526         for (; start < end; start += PAGE_SIZE)
 527                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 528         spin_unlock(&kvm->mmu_lock);
 529
 530         /* we've to flush the tlb before the pages can be freed */
 531         if (need_tlb_flush)
 532                 kvm_flush_remote_tlbs(kvm);
 533 }
 534
 535 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 536                                                   struct mm_struct *mm,
 537                                                   unsigned long start,
 538                                                   unsigned long end)
 539 {
 540         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 541
 542         spin_lock(&kvm->mmu_lock);
 543         /*
 544          * This sequence increase will notify the kvm page fault that
 545          * the page that is going to be mapped in the spte could have
 546          * been freed.
 547          */
 548         kvm->mmu_notifier_seq++;
 549         /*
 550          * The above sequence increase must be visible before the
 551          * below count decrease but both values are read by the kvm
 552          * page fault under mmu_lock spinlock so we don't need to add
 553          * a smb_wmb() here in between the two.
 554          */
 555         kvm->mmu_notifier_count--;
 556         spin_unlock(&kvm->mmu_lock);
 557
 558         BUG_ON(kvm->mmu_notifier_count < 0);
 559 }
 560
 561 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 562                                               struct mm_struct *mm,
 563                                               unsigned long address)
 564 {
 565         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 566         int young;
 567
 568         spin_lock(&kvm->mmu_lock);
 569         young = kvm_age_hva(kvm, address);
 570         spin_unlock(&kvm->mmu_lock);
 571
 572         if (young)
 573                 kvm_flush_remote_tlbs(kvm);
 574
 575         return young;
 576 }
 577
 578 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 579         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 580         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 581         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 582         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 583 };
 584 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 585
 586 static struct kvm *kvm_create_vm(void)
 587 {
 588         struct kvm *kvm = kvm_arch_create_vm();
 589 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 590         struct page *page;
 591 #endif
 592
 593         if (IS_ERR(kvm))
 594                 goto out;
 595
 596 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 597         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 598         if (!page) {
 599                 kfree(kvm);
 600                 return ERR_PTR(-ENOMEM);
 601         }
 602         kvm->coalesced_mmio_ring =
 603                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 604 #endif
 605
 606 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 607         {
 608                 int err;
 609                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 610                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 611                 if (err) {
 612 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 613                         put_page(page);
 614 #endif
 615                         kfree(kvm);
 616                         return ERR_PTR(err);
 617                 }
 618         }
 619 #endif
 620
 621         kvm->mm = current->mm;
 622         atomic_inc(&kvm->mm->mm_count);
 623         spin_lock_init(&kvm->mmu_lock);
 624         kvm_io_bus_init(&kvm->pio_bus);
 625         mutex_init(&kvm->lock);
 626         kvm_io_bus_init(&kvm->mmio_bus);
 627         init_rwsem(&kvm->slots_lock);
 628         atomic_set(&kvm->users_count, 1);
 629         spin_lock(&kvm_lock);
 630         list_add(&kvm->vm_list, &vm_list);
 631         spin_unlock(&kvm_lock);
 632 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 633         kvm_coalesced_mmio_init(kvm);
 634 #endif
 635 out:
 636         return kvm;
 637 }
 638
 639 /*
 640  * Free any memory in @free but not in @dont.
 641  */
 642 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 643                                   struct kvm_memory_slot *dont)
 644 {
 645         if (!dont || free->rmap != dont->rmap)
 646                 vfree(free->rmap);
 647
 648         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 649                 vfree(free->dirty_bitmap);
 650
 651         if (!dont || free->lpage_info != dont->lpage_info)
 652                 vfree(free->lpage_info);
 653
 654         free->npages = 0;
 655         free->dirty_bitmap = NULL;
 656         free->rmap = NULL;
 657         free->lpage_info = NULL;
 658 }
 659
 660 void kvm_free_physmem(struct kvm *kvm)
 661 {
 662         int i;
 663
 664         for (i = 0; i < kvm->nmemslots; ++i)
 665                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 666 }
 667
 668 static void kvm_destroy_vm(struct kvm *kvm)
 669 {
 670         struct mm_struct *mm = kvm->mm;
 671
 672         spin_lock(&kvm_lock);
 673         list_del(&kvm->vm_list);
 674         spin_unlock(&kvm_lock);
 675         kvm_io_bus_destroy(&kvm->pio_bus);
 676         kvm_io_bus_destroy(&kvm->mmio_bus);
 677 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 678         if (kvm->coalesced_mmio_ring != NULL)
 679                 free_page((unsigned long)kvm->coalesced_mmio_ring);
 680 #endif
 681 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 682         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 683 #endif
 684         kvm_arch_destroy_vm(kvm);
 685         mmdrop(mm);
 686 }
 687
 688 void kvm_get_kvm(struct kvm *kvm)
 689 {
 690         atomic_inc(&kvm->users_count);
 691 }
 692 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 693
 694 void kvm_put_kvm(struct kvm *kvm)
 695 {
 696         if (atomic_dec_and_test(&kvm->users_count))
 697                 kvm_destroy_vm(kvm);
 698 }
 699 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 700
 701
 702 static int kvm_vm_release(struct inode *inode, struct file *filp)
 703 {
 704         struct kvm *kvm = filp->private_data;
 705
 706         kvm_put_kvm(kvm);
 707         return 0;
 708 }
 709
 710 /*
 711  * Allocate some memory and give it an address in the guest physical address
 712  * space.
 713  *
 714  * Discontiguous memory is allowed, mostly for framebuffers.
 715  *
 716  * Must be called holding mmap_sem for write.
 717  */
 718 int __kvm_set_memory_region(struct kvm *kvm,
 719                             struct kvm_userspace_memory_region *mem,
 720                             int user_alloc)
 721 {
 722         int r;
 723         gfn_t base_gfn;
 724         unsigned long npages;
 725         unsigned long i;
 726         struct kvm_memory_slot *memslot;
 727         struct kvm_memory_slot old, new;
 728
 729         r = -EINVAL;
 730         /* General sanity checks */
 731         if (mem->memory_size & (PAGE_SIZE - 1))
 732                 goto out;
 733         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 734                 goto out;
 735         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 736                 goto out;
 737         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 738                 goto out;
 739         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 740                 goto out;
 741
 742         memslot = &kvm->memslots[mem->slot];
 743         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 744         npages = mem->memory_size >> PAGE_SHIFT;
 745
 746         if (!npages)
 747                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 748
 749         new = old = *memslot;
 750
 751         new.base_gfn = base_gfn;
 752         new.npages = npages;
 753         new.flags = mem->flags;
 754
 755         /* Disallow changing a memory slot's size. */
 756         r = -EINVAL;
 757         if (npages && old.npages && npages != old.npages)
 758                 goto out_free;
 759
 760         /* Check for overlaps */
 761         r = -EEXIST;
 762         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 763                 struct kvm_memory_slot *s = &kvm->memslots[i];
 764
 765                 if (s == memslot)
 766                         continue;
 767                 if (!((base_gfn + npages <= s->base_gfn) ||
 768                       (base_gfn >= s->base_gfn + s->npages)))
 769                         goto out_free;
 770         }
 771
 772         /* Free page dirty bitmap if unneeded */
 773         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 774                 new.dirty_bitmap = NULL;
 775
 776         r = -ENOMEM;
 777
 778         /* Allocate if a slot is being created */
 779 #ifndef CONFIG_S390
 780         if (npages && !new.rmap) {
 781                 new.rmap = vmalloc(npages * sizeof(struct page *));
 782
 783                 if (!new.rmap)
 784                         goto out_free;
 785
 786                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 787
 788                 new.user_alloc = user_alloc;
 789                 /*
 790                  * hva_to_rmmap() serialzies with the mmu_lock and to be
 791                  * safe it has to ignore memslots with !user_alloc &&
 792                  * !userspace_addr.
 793                  */
 794                 if (user_alloc)
 795                         new.userspace_addr = mem->userspace_addr;
 796                 else
 797                         new.userspace_addr = 0;
 798         }
 799         if (npages && !new.lpage_info) {
 800                 int largepages = npages / KVM_PAGES_PER_HPAGE;
 801                 if (npages % KVM_PAGES_PER_HPAGE)
 802                         largepages++;
 803                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 804                         largepages++;
 805
 806                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
 807
 808                 if (!new.lpage_info)
 809                         goto out_free;
 810
 811                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
 812
 813                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 814                         new.lpage_info[0].write_count = 1;
 815                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
 816                         new.lpage_info[largepages-1].write_count = 1;
 817         }
 818
 819         /* Allocate page dirty bitmap if needed */
 820         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 821                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 822
 823                 new.dirty_bitmap = vmalloc(dirty_bytes);
 824                 if (!new.dirty_bitmap)
 825                         goto out_free;
 826                 memset(new.dirty_bitmap, 0, dirty_bytes);
 827         }
 828 #endif /* not defined CONFIG_S390 */
 829
 830         if (!npages)
 831                 kvm_arch_flush_shadow(kvm);
 832
 833         spin_lock(&kvm->mmu_lock);
 834         if (mem->slot >= kvm->nmemslots)
 835                 kvm->nmemslots = mem->slot + 1;
 836
 837         *memslot = new;
 838         spin_unlock(&kvm->mmu_lock);
 839
 840         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
 841         if (r) {
 842                 spin_lock(&kvm->mmu_lock);
 843                 *memslot = old;
 844                 spin_unlock(&kvm->mmu_lock);
 845                 goto out_free;
 846         }
 847
 848         kvm_free_physmem_slot(&old, &new);
 849 #ifdef CONFIG_DMAR
 850         /* map the pages in iommu page table */
 851         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
 852         if (r)
 853                 goto out;
 854 #endif
 855         return 0;
 856
 857 out_free:
 858         kvm_free_physmem_slot(&new, &old);
 859 out:
 860         return r;
 861
 862 }
 863 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 864
 865 int kvm_set_memory_region(struct kvm *kvm,
 866                           struct kvm_userspace_memory_region *mem,
 867                           int user_alloc)
 868 {
 869         int r;
 870
 871         down_write(&kvm->slots_lock);
 872         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 873         up_write(&kvm->slots_lock);
 874         return r;
 875 }
 876 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 877
 878 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 879                                    struct
 880                                    kvm_userspace_memory_region *mem,
 881                                    int user_alloc)
 882 {
 883         if (mem->slot >= KVM_MEMORY_SLOTS)
 884                 return -EINVAL;
 885         return kvm_set_memory_region(kvm, mem, user_alloc);
 886 }
 887
 888 int kvm_get_dirty_log(struct kvm *kvm,
 889                         struct kvm_dirty_log *log, int *is_dirty)
 890 {
 891         struct kvm_memory_slot *memslot;
 892         int r, i;
 893         int n;
 894         unsigned long any = 0;
 895
 896         r = -EINVAL;
 897         if (log->slot >= KVM_MEMORY_SLOTS)
 898                 goto out;
 899
 900         memslot = &kvm->memslots[log->slot];
 901         r = -ENOENT;
 902         if (!memslot->dirty_bitmap)
 903                 goto out;
 904
 905         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 906
 907         for (i = 0; !any && i < n/sizeof(long); ++i)
 908                 any = memslot->dirty_bitmap[i];
 909
 910         r = -EFAULT;
 911         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 912                 goto out;
 913
 914         if (any)
 915                 *is_dirty = 1;
 916
 917         r = 0;
 918 out:
 919         return r;
 920 }
 921
 922 int is_error_page(struct page *page)
 923 {
 924         return page == bad_page;
 925 }
 926 EXPORT_SYMBOL_GPL(is_error_page);
 927
 928 int is_error_pfn(pfn_t pfn)
 929 {
 930         return pfn == bad_pfn;
 931 }
 932 EXPORT_SYMBOL_GPL(is_error_pfn);
 933
 934 static inline unsigned long bad_hva(void)
 935 {
 936         return PAGE_OFFSET;
 937 }
 938
 939 int kvm_is_error_hva(unsigned long addr)
 940 {
 941         return addr == bad_hva();
 942 }
 943 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 944
 945 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
 946 {
 947         int i;
 948
 949         for (i = 0; i < kvm->nmemslots; ++i) {
 950                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 951
 952                 if (gfn >= memslot->base_gfn
 953                     && gfn < memslot->base_gfn + memslot->npages)
 954                         return memslot;
 955         }
 956         return NULL;
 957 }
 958 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
 959
 960 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 961 {
 962         gfn = unalias_gfn(kvm, gfn);
 963         return gfn_to_memslot_unaliased(kvm, gfn);
 964 }
 965
 966 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 967 {
 968         int i;
 969
 970         gfn = unalias_gfn(kvm, gfn);
 971         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 972                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 973
 974                 if (gfn >= memslot->base_gfn
 975                     && gfn < memslot->base_gfn + memslot->npages)
 976                         return 1;
 977         }
 978         return 0;
 979 }
 980 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 981
 982 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 983 {
 984         struct kvm_memory_slot *slot;
 985
 986         gfn = unalias_gfn(kvm, gfn);
 987         slot = gfn_to_memslot_unaliased(kvm, gfn);
 988         if (!slot)
 989                 return bad_hva();
 990         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 991 }
 992 EXPORT_SYMBOL_GPL(gfn_to_hva);
 993
 994 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 995 {
 996         struct page *page[1];
 997         unsigned long addr;
 998         int npages;
 999         pfn_t pfn;
1000
1001         might_sleep();
1002
1003         addr = gfn_to_hva(kvm, gfn);
1004         if (kvm_is_error_hva(addr)) {
1005                 get_page(bad_page);
1006                 return page_to_pfn(bad_page);
1007         }
1008
1009         npages = get_user_pages_fast(addr, 1, 1, page);
1010
1011         if (unlikely(npages != 1)) {
1012                 struct vm_area_struct *vma;
1013
1014                 down_read(&current->mm->mmap_sem);
1015                 vma = find_vma(current->mm, addr);
1016
1017                 if (vma == NULL || addr < vma->vm_start ||
1018                     !(vma->vm_flags & VM_PFNMAP)) {
1019                         up_read(&current->mm->mmap_sem);
1020                         get_page(bad_page);
1021                         return page_to_pfn(bad_page);
1022                 }
1023
1024                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1025                 up_read(&current->mm->mmap_sem);
1026                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1027         } else
1028                 pfn = page_to_pfn(page[0]);
1029
1030         return pfn;
1031 }
1032
1033 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1034
1035 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1036 {
1037         pfn_t pfn;
1038
1039         pfn = gfn_to_pfn(kvm, gfn);
1040         if (!kvm_is_mmio_pfn(pfn))
1041                 return pfn_to_page(pfn);
1042
1043         WARN_ON(kvm_is_mmio_pfn(pfn));
1044
1045         get_page(bad_page);
1046         return bad_page;
1047 }
1048
1049 EXPORT_SYMBOL_GPL(gfn_to_page);
1050
1051 void kvm_release_page_clean(struct page *page)
1052 {
1053         kvm_release_pfn_clean(page_to_pfn(page));
1054 }
1055 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1056
1057 void kvm_release_pfn_clean(pfn_t pfn)
1058 {
1059         if (!kvm_is_mmio_pfn(pfn))
1060                 put_page(pfn_to_page(pfn));
1061 }
1062 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1063
1064 void kvm_release_page_dirty(struct page *page)
1065 {
1066         kvm_release_pfn_dirty(page_to_pfn(page));
1067 }
1068 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1069
1070 void kvm_release_pfn_dirty(pfn_t pfn)
1071 {
1072         kvm_set_pfn_dirty(pfn);
1073         kvm_release_pfn_clean(pfn);
1074 }
1075 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1076
1077 void kvm_set_page_dirty(struct page *page)
1078 {
1079         kvm_set_pfn_dirty(page_to_pfn(page));
1080 }
1081 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1082
1083 void kvm_set_pfn_dirty(pfn_t pfn)
1084 {
1085         if (!kvm_is_mmio_pfn(pfn)) {
1086                 struct page *page = pfn_to_page(pfn);
1087                 if (!PageReserved(page))
1088                         SetPageDirty(page);
1089         }
1090 }
1091 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1092
1093 void kvm_set_pfn_accessed(pfn_t pfn)
1094 {
1095         if (!kvm_is_mmio_pfn(pfn))
1096                 mark_page_accessed(pfn_to_page(pfn));
1097 }
1098 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1099
1100 void kvm_get_pfn(pfn_t pfn)
1101 {
1102         if (!kvm_is_mmio_pfn(pfn))
1103                 get_page(pfn_to_page(pfn));
1104 }
1105 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1106
1107 static int next_segment(unsigned long len, int offset)
1108 {
1109         if (len > PAGE_SIZE - offset)
1110                 return PAGE_SIZE - offset;
1111         else
1112                 return len;
1113 }
1114
1115 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1116                         int len)
1117 {
1118         int r;
1119         unsigned long addr;
1120
1121         addr = gfn_to_hva(kvm, gfn);
1122         if (kvm_is_error_hva(addr))
1123                 return -EFAULT;
1124         r = copy_from_user(data, (void __user *)addr + offset, len);
1125         if (r)
1126                 return -EFAULT;
1127         return 0;
1128 }
1129 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1130
1131 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1132 {
1133         gfn_t gfn = gpa >> PAGE_SHIFT;
1134         int seg;
1135         int offset = offset_in_page(gpa);
1136         int ret;
1137
1138         while ((seg = next_segment(len, offset)) != 0) {
1139                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1140                 if (ret < 0)
1141                         return ret;
1142                 offset = 0;
1143                 len -= seg;
1144                 data += seg;
1145                 ++gfn;
1146         }
1147         return 0;
1148 }
1149 EXPORT_SYMBOL_GPL(kvm_read_guest);
1150
1151 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1152                           unsigned long len)
1153 {
1154         int r;
1155         unsigned long addr;
1156         gfn_t gfn = gpa >> PAGE_SHIFT;
1157         int offset = offset_in_page(gpa);
1158
1159         addr = gfn_to_hva(kvm, gfn);
1160         if (kvm_is_error_hva(addr))
1161                 return -EFAULT;
1162         pagefault_disable();
1163         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1164         pagefault_enable();
1165         if (r)
1166                 return -EFAULT;
1167         return 0;
1168 }
1169 EXPORT_SYMBOL(kvm_read_guest_atomic);
1170
1171 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1172                          int offset, int len)
1173 {
1174         int r;
1175         unsigned long addr;
1176
1177         addr = gfn_to_hva(kvm, gfn);
1178         if (kvm_is_error_hva(addr))
1179                 return -EFAULT;
1180         r = copy_to_user((void __user *)addr + offset, data, len);
1181         if (r)
1182                 return -EFAULT;
1183         mark_page_dirty(kvm, gfn);
1184         return 0;
1185 }
1186 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1187
1188 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1189                     unsigned long len)
1190 {
1191         gfn_t gfn = gpa >> PAGE_SHIFT;
1192         int seg;
1193         int offset = offset_in_page(gpa);
1194         int ret;
1195
1196         while ((seg = next_segment(len, offset)) != 0) {
1197                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1198                 if (ret < 0)
1199                         return ret;
1200                 offset = 0;
1201                 len -= seg;
1202                 data += seg;
1203                 ++gfn;
1204         }
1205         return 0;
1206 }
1207
1208 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1209 {
1210         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1211 }
1212 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1213
1214 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1215 {
1216         gfn_t gfn = gpa >> PAGE_SHIFT;
1217         int seg;
1218         int offset = offset_in_page(gpa);
1219         int ret;
1220
1221         while ((seg = next_segment(len, offset)) != 0) {
1222                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1223                 if (ret < 0)
1224                         return ret;
1225                 offset = 0;
1226                 len -= seg;
1227                 ++gfn;
1228         }
1229         return 0;
1230 }
1231 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1232
1233 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1234 {
1235         struct kvm_memory_slot *memslot;
1236
1237         gfn = unalias_gfn(kvm, gfn);
1238         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1239         if (memslot && memslot->dirty_bitmap) {
1240                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1241
1242                 /* avoid RMW */
1243                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1244                         set_bit(rel_gfn, memslot->dirty_bitmap);
1245         }
1246 }
1247
1248 /*
1249  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1250  */
1251 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1252 {
1253         DEFINE_WAIT(wait);
1254
1255         for (;;) {
1256                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1257
1258                 if (kvm_cpu_has_interrupt(vcpu) ||
1259                     kvm_cpu_has_pending_timer(vcpu) ||
1260                     kvm_arch_vcpu_runnable(vcpu)) {
1261                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1262                         break;
1263                 }
1264                 if (signal_pending(current))
1265                         break;
1266
1267                 vcpu_put(vcpu);
1268                 schedule();
1269                 vcpu_load(vcpu);
1270         }
1271
1272         finish_wait(&vcpu->wq, &wait);
1273 }
1274
1275 void kvm_resched(struct kvm_vcpu *vcpu)
1276 {
1277         if (!need_resched())
1278                 return;
1279         cond_resched();
1280 }
1281 EXPORT_SYMBOL_GPL(kvm_resched);
1282
1283 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1284 {
1285         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1286         struct page *page;
1287
1288         if (vmf->pgoff == 0)
1289                 page = virt_to_page(vcpu->run);
1290 #ifdef CONFIG_X86
1291         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1292                 page = virt_to_page(vcpu->arch.pio_data);
1293 #endif
1294 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1295         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1296                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1297 #endif
1298         else
1299                 return VM_FAULT_SIGBUS;
1300         get_page(page);
1301         vmf->page = page;
1302         return 0;
1303 }
1304
1305 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1306         .fault = kvm_vcpu_fault,
1307 };
1308
1309 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1310 {
1311         vma->vm_ops = &kvm_vcpu_vm_ops;
1312         return 0;
1313 }
1314
1315 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1316 {
1317         struct kvm_vcpu *vcpu = filp->private_data;
1318
1319         kvm_put_kvm(vcpu->kvm);
1320         return 0;
1321 }
1322
1323 static const struct file_operations kvm_vcpu_fops = {
1324         .release        = kvm_vcpu_release,
1325         .unlocked_ioctl = kvm_vcpu_ioctl,
1326         .compat_ioctl   = kvm_vcpu_ioctl,
1327         .mmap           = kvm_vcpu_mmap,
1328 };
1329
1330 /*
1331  * Allocates an inode for the vcpu.
1332  */
1333 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1334 {
1335         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1336         if (fd < 0)
1337                 kvm_put_kvm(vcpu->kvm);
1338         return fd;
1339 }
1340
1341 /*
1342  * Creates some virtual cpus.  Good luck creating more than one.
1343  */
1344 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1345 {
1346         int r;
1347         struct kvm_vcpu *vcpu;
1348
1349         if (!valid_vcpu(n))
1350                 return -EINVAL;
1351
1352         vcpu = kvm_arch_vcpu_create(kvm, n);
1353         if (IS_ERR(vcpu))
1354                 return PTR_ERR(vcpu);
1355
1356         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1357
1358         r = kvm_arch_vcpu_setup(vcpu);
1359         if (r)
1360                 return r;
1361
1362         mutex_lock(&kvm->lock);
1363         if (kvm->vcpus[n]) {
1364                 r = -EEXIST;
1365                 goto vcpu_destroy;
1366         }
1367         kvm->vcpus[n] = vcpu;
1368         mutex_unlock(&kvm->lock);
1369
1370         /* Now it's all set up, let userspace reach it */
1371         kvm_get_kvm(kvm);
1372         r = create_vcpu_fd(vcpu);
1373         if (r < 0)
1374                 goto unlink;
1375         return r;
1376
1377 unlink:
1378         mutex_lock(&kvm->lock);
1379         kvm->vcpus[n] = NULL;
1380 vcpu_destroy:
1381         mutex_unlock(&kvm->lock);
1382         kvm_arch_vcpu_destroy(vcpu);
1383         return r;
1384 }
1385
1386 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1387 {
1388         if (sigset) {
1389                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1390                 vcpu->sigset_active = 1;
1391                 vcpu->sigset = *sigset;
1392         } else
1393                 vcpu->sigset_active = 0;
1394         return 0;
1395 }
1396
1397 static long kvm_vcpu_ioctl(struct file *filp,
1398                            unsigned int ioctl, unsigned long arg)
1399 {
1400         struct kvm_vcpu *vcpu = filp->private_data;
1401         void __user *argp = (void __user *)arg;
1402         int r;
1403         struct kvm_fpu *fpu = NULL;
1404         struct kvm_sregs *kvm_sregs = NULL;
1405
1406         if (vcpu->kvm->mm != current->mm)
1407                 return -EIO;
1408         switch (ioctl) {
1409         case KVM_RUN:
1410                 r = -EINVAL;
1411                 if (arg)
1412                         goto out;
1413                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1414                 break;
1415         case KVM_GET_REGS: {
1416                 struct kvm_regs *kvm_regs;
1417
1418                 r = -ENOMEM;
1419                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1420                 if (!kvm_regs)
1421                         goto out;
1422                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1423                 if (r)
1424                         goto out_free1;
1425                 r = -EFAULT;
1426                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1427                         goto out_free1;
1428                 r = 0;
1429 out_free1:
1430                 kfree(kvm_regs);
1431                 break;
1432         }
1433         case KVM_SET_REGS: {
1434                 struct kvm_regs *kvm_regs;
1435
1436                 r = -ENOMEM;
1437                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1438                 if (!kvm_regs)
1439                         goto out;
1440                 r = -EFAULT;
1441                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1442                         goto out_free2;
1443                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1444                 if (r)
1445                         goto out_free2;
1446                 r = 0;
1447 out_free2:
1448                 kfree(kvm_regs);
1449                 break;
1450         }
1451         case KVM_GET_SREGS: {
1452                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1453                 r = -ENOMEM;
1454                 if (!kvm_sregs)
1455                         goto out;
1456                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1457                 if (r)
1458                         goto out;
1459                 r = -EFAULT;
1460                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1461                         goto out;
1462                 r = 0;
1463                 break;
1464         }
1465         case KVM_SET_SREGS: {
1466                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1467                 r = -ENOMEM;
1468                 if (!kvm_sregs)
1469                         goto out;
1470                 r = -EFAULT;
1471                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1472                         goto out;
1473                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1474                 if (r)
1475                         goto out;
1476                 r = 0;
1477                 break;
1478         }
1479         case KVM_GET_MP_STATE: {
1480                 struct kvm_mp_state mp_state;
1481
1482                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1483                 if (r)
1484                         goto out;
1485                 r = -EFAULT;
1486                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1487                         goto out;
1488                 r = 0;
1489                 break;
1490         }
1491         case KVM_SET_MP_STATE: {
1492                 struct kvm_mp_state mp_state;
1493
1494                 r = -EFAULT;
1495                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1496                         goto out;
1497                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1498                 if (r)
1499                         goto out;
1500                 r = 0;
1501                 break;
1502         }
1503         case KVM_TRANSLATE: {
1504                 struct kvm_translation tr;
1505
1506                 r = -EFAULT;
1507                 if (copy_from_user(&tr, argp, sizeof tr))
1508                         goto out;
1509                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1510                 if (r)
1511                         goto out;
1512                 r = -EFAULT;
1513                 if (copy_to_user(argp, &tr, sizeof tr))
1514                         goto out;
1515                 r = 0;
1516                 break;
1517         }
1518         case KVM_DEBUG_GUEST: {
1519                 struct kvm_debug_guest dbg;
1520
1521                 r = -EFAULT;
1522                 if (copy_from_user(&dbg, argp, sizeof dbg))
1523                         goto out;
1524                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1525                 if (r)
1526                         goto out;
1527                 r = 0;
1528                 break;
1529         }
1530         case KVM_SET_SIGNAL_MASK: {
1531                 struct kvm_signal_mask __user *sigmask_arg = argp;
1532                 struct kvm_signal_mask kvm_sigmask;
1533                 sigset_t sigset, *p;
1534
1535                 p = NULL;
1536                 if (argp) {
1537                         r = -EFAULT;
1538                         if (copy_from_user(&kvm_sigmask, argp,
1539                                            sizeof kvm_sigmask))
1540                                 goto out;
1541                         r = -EINVAL;
1542                         if (kvm_sigmask.len != sizeof sigset)
1543                                 goto out;
1544                         r = -EFAULT;
1545                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1546                                            sizeof sigset))
1547                                 goto out;
1548                         p = &sigset;
1549                 }
1550                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1551                 break;
1552         }
1553         case KVM_GET_FPU: {
1554                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1555                 r = -ENOMEM;
1556                 if (!fpu)
1557                         goto out;
1558                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1559                 if (r)
1560                         goto out;
1561                 r = -EFAULT;
1562                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1563                         goto out;
1564                 r = 0;
1565                 break;
1566         }
1567         case KVM_SET_FPU: {
1568                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1569                 r = -ENOMEM;
1570                 if (!fpu)
1571                         goto out;
1572                 r = -EFAULT;
1573                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1574                         goto out;
1575                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1576                 if (r)
1577                         goto out;
1578                 r = 0;
1579                 break;
1580         }
1581         default:
1582                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1583         }
1584 out:
1585         kfree(fpu);
1586         kfree(kvm_sregs);
1587         return r;
1588 }
1589
1590 static long kvm_vm_ioctl(struct file *filp,
1591                            unsigned int ioctl, unsigned long arg)
1592 {
1593         struct kvm *kvm = filp->private_data;
1594         void __user *argp = (void __user *)arg;
1595         int r;
1596
1597         if (kvm->mm != current->mm)
1598                 return -EIO;
1599         switch (ioctl) {
1600         case KVM_CREATE_VCPU:
1601                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1602                 if (r < 0)
1603                         goto out;
1604                 break;
1605         case KVM_SET_USER_MEMORY_REGION: {
1606                 struct kvm_userspace_memory_region kvm_userspace_mem;
1607
1608                 r = -EFAULT;
1609                 if (copy_from_user(&kvm_userspace_mem, argp,
1610                                                 sizeof kvm_userspace_mem))
1611                         goto out;
1612
1613                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1614                 if (r)
1615                         goto out;
1616                 break;
1617         }
1618         case KVM_GET_DIRTY_LOG: {
1619                 struct kvm_dirty_log log;
1620
1621                 r = -EFAULT;
1622                 if (copy_from_user(&log, argp, sizeof log))
1623                         goto out;
1624                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1625                 if (r)
1626                         goto out;
1627                 break;
1628         }
1629 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1630         case KVM_REGISTER_COALESCED_MMIO: {
1631                 struct kvm_coalesced_mmio_zone zone;
1632                 r = -EFAULT;
1633                 if (copy_from_user(&zone, argp, sizeof zone))
1634                         goto out;
1635                 r = -ENXIO;
1636                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1637                 if (r)
1638                         goto out;
1639                 r = 0;
1640                 break;
1641         }
1642         case KVM_UNREGISTER_COALESCED_MMIO: {
1643                 struct kvm_coalesced_mmio_zone zone;
1644                 r = -EFAULT;
1645                 if (copy_from_user(&zone, argp, sizeof zone))
1646                         goto out;
1647                 r = -ENXIO;
1648                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1649                 if (r)
1650                         goto out;
1651                 r = 0;
1652                 break;
1653         }
1654 #endif
1655 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1656         case KVM_ASSIGN_PCI_DEVICE: {
1657                 struct kvm_assigned_pci_dev assigned_dev;
1658
1659                 r = -EFAULT;
1660                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1661                         goto out;
1662                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1663                 if (r)
1664                         goto out;
1665                 break;
1666         }
1667         case KVM_ASSIGN_IRQ: {
1668                 struct kvm_assigned_irq assigned_irq;
1669
1670                 r = -EFAULT;
1671                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1672                         goto out;
1673                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1674                 if (r)
1675                         goto out;
1676                 break;
1677         }
1678 #endif
1679         default:
1680                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1681         }
1682 out:
1683         return r;
1684 }
1685
1686 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1687 {
1688         struct page *page[1];
1689         unsigned long addr;
1690         int npages;
1691         gfn_t gfn = vmf->pgoff;
1692         struct kvm *kvm = vma->vm_file->private_data;
1693
1694         addr = gfn_to_hva(kvm, gfn);
1695         if (kvm_is_error_hva(addr))
1696                 return VM_FAULT_SIGBUS;
1697
1698         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1699                                 NULL);
1700         if (unlikely(npages != 1))
1701                 return VM_FAULT_SIGBUS;
1702
1703         vmf->page = page[0];
1704         return 0;
1705 }
1706
1707 static struct vm_operations_struct kvm_vm_vm_ops = {
1708         .fault = kvm_vm_fault,
1709 };
1710
1711 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1712 {
1713         vma->vm_ops = &kvm_vm_vm_ops;
1714         return 0;
1715 }
1716
1717 static const struct file_operations kvm_vm_fops = {
1718         .release        = kvm_vm_release,
1719         .unlocked_ioctl = kvm_vm_ioctl,
1720         .compat_ioctl   = kvm_vm_ioctl,
1721         .mmap           = kvm_vm_mmap,
1722 };
1723
1724 static int kvm_dev_ioctl_create_vm(void)
1725 {
1726         int fd;
1727         struct kvm *kvm;
1728
1729         kvm = kvm_create_vm();
1730         if (IS_ERR(kvm))
1731                 return PTR_ERR(kvm);
1732         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1733         if (fd < 0)
1734                 kvm_put_kvm(kvm);
1735
1736         return fd;
1737 }
1738
1739 static long kvm_dev_ioctl(struct file *filp,
1740                           unsigned int ioctl, unsigned long arg)
1741 {
1742         long r = -EINVAL;
1743
1744         switch (ioctl) {
1745         case KVM_GET_API_VERSION:
1746                 r = -EINVAL;
1747                 if (arg)
1748                         goto out;
1749                 r = KVM_API_VERSION;
1750                 break;
1751         case KVM_CREATE_VM:
1752                 r = -EINVAL;
1753                 if (arg)
1754                         goto out;
1755                 r = kvm_dev_ioctl_create_vm();
1756                 break;
1757         case KVM_CHECK_EXTENSION:
1758                 r = kvm_dev_ioctl_check_extension(arg);
1759                 break;
1760         case KVM_GET_VCPU_MMAP_SIZE:
1761                 r = -EINVAL;
1762                 if (arg)
1763                         goto out;
1764                 r = PAGE_SIZE;     /* struct kvm_run */
1765 #ifdef CONFIG_X86
1766                 r += PAGE_SIZE;    /* pio data page */
1767 #endif
1768 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1769                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1770 #endif
1771                 break;
1772         case KVM_TRACE_ENABLE:
1773         case KVM_TRACE_PAUSE:
1774         case KVM_TRACE_DISABLE:
1775                 r = kvm_trace_ioctl(ioctl, arg);
1776                 break;
1777         default:
1778                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1779         }
1780 out:
1781         return r;
1782 }
1783
1784 static struct file_operations kvm_chardev_ops = {
1785         .unlocked_ioctl = kvm_dev_ioctl,
1786         .compat_ioctl   = kvm_dev_ioctl,
1787 };
1788
1789 static struct miscdevice kvm_dev = {
1790         KVM_MINOR,
1791         "kvm",
1792         &kvm_chardev_ops,
1793 };
1794
1795 static void hardware_enable(void *junk)
1796 {
1797         int cpu = raw_smp_processor_id();
1798
1799         if (cpu_isset(cpu, cpus_hardware_enabled))
1800                 return;
1801         cpu_set(cpu, cpus_hardware_enabled);
1802         kvm_arch_hardware_enable(NULL);
1803 }
1804
1805 static void hardware_disable(void *junk)
1806 {
1807         int cpu = raw_smp_processor_id();
1808
1809         if (!cpu_isset(cpu, cpus_hardware_enabled))
1810                 return;
1811         cpu_clear(cpu, cpus_hardware_enabled);
1812         kvm_arch_hardware_disable(NULL);
1813 }
1814
1815 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1816                            void *v)
1817 {
1818         int cpu = (long)v;
1819
1820         val &= ~CPU_TASKS_FROZEN;
1821         switch (val) {
1822         case CPU_DYING:
1823                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1824                        cpu);
1825                 hardware_disable(NULL);
1826                 break;
1827         case CPU_UP_CANCELED:
1828                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1829                        cpu);
1830                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1831                 break;
1832         case CPU_ONLINE:
1833                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1834                        cpu);
1835                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1836                 break;
1837         }
1838         return NOTIFY_OK;
1839 }
1840
1841
1842 asmlinkage void kvm_handle_fault_on_reboot(void)
1843 {
1844         if (kvm_rebooting)
1845                 /* spin while reset goes on */
1846                 while (true)
1847                         ;
1848         /* Fault while not rebooting.  We want the trace. */
1849         BUG();
1850 }
1851 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1852
1853 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1854                       void *v)
1855 {
1856         if (val == SYS_RESTART) {
1857                 /*
1858                  * Some (well, at least mine) BIOSes hang on reboot if
1859                  * in vmx root mode.
1860                  */
1861                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1862                 kvm_rebooting = true;
1863                 on_each_cpu(hardware_disable, NULL, 1);
1864         }
1865         return NOTIFY_OK;
1866 }
1867
1868 static struct notifier_block kvm_reboot_notifier = {
1869         .notifier_call = kvm_reboot,
1870         .priority = 0,
1871 };
1872
1873 void kvm_io_bus_init(struct kvm_io_bus *bus)
1874 {
1875         memset(bus, 0, sizeof(*bus));
1876 }
1877
1878 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1879 {
1880         int i;
1881
1882         for (i = 0; i < bus->dev_count; i++) {
1883                 struct kvm_io_device *pos = bus->devs[i];
1884
1885                 kvm_iodevice_destructor(pos);
1886         }
1887 }
1888
1889 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
1890                                           gpa_t addr, int len, int is_write)
1891 {
1892         int i;
1893
1894         for (i = 0; i < bus->dev_count; i++) {
1895                 struct kvm_io_device *pos = bus->devs[i];
1896
1897                 if (pos->in_range(pos, addr, len, is_write))
1898                         return pos;
1899         }
1900
1901         return NULL;
1902 }
1903
1904 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1905 {
1906         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1907
1908         bus->devs[bus->dev_count++] = dev;
1909 }
1910
1911 static struct notifier_block kvm_cpu_notifier = {
1912         .notifier_call = kvm_cpu_hotplug,
1913         .priority = 20, /* must be > scheduler priority */
1914 };
1915
1916 static int vm_stat_get(void *_offset, u64 *val)
1917 {
1918         unsigned offset = (long)_offset;
1919         struct kvm *kvm;
1920
1921         *val = 0;
1922         spin_lock(&kvm_lock);
1923         list_for_each_entry(kvm, &vm_list, vm_list)
1924                 *val += *(u32 *)((void *)kvm + offset);
1925         spin_unlock(&kvm_lock);
1926         return 0;
1927 }
1928
1929 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1930
1931 static int vcpu_stat_get(void *_offset, u64 *val)
1932 {
1933         unsigned offset = (long)_offset;
1934         struct kvm *kvm;
1935         struct kvm_vcpu *vcpu;
1936         int i;
1937
1938         *val = 0;
1939         spin_lock(&kvm_lock);
1940         list_for_each_entry(kvm, &vm_list, vm_list)
1941                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1942                         vcpu = kvm->vcpus[i];
1943                         if (vcpu)
1944                                 *val += *(u32 *)((void *)vcpu + offset);
1945                 }
1946         spin_unlock(&kvm_lock);
1947         return 0;
1948 }
1949
1950 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1951
1952 static struct file_operations *stat_fops[] = {
1953         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1954         [KVM_STAT_VM]   = &vm_stat_fops,
1955 };
1956
1957 static void kvm_init_debug(void)
1958 {
1959         struct kvm_stats_debugfs_item *p;
1960
1961         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1962         for (p = debugfs_entries; p->name; ++p)
1963                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1964                                                 (void *)(long)p->offset,
1965                                                 stat_fops[p->kind]);
1966 }
1967
1968 static void kvm_exit_debug(void)
1969 {
1970         struct kvm_stats_debugfs_item *p;
1971
1972         for (p = debugfs_entries; p->name; ++p)
1973                 debugfs_remove(p->dentry);
1974         debugfs_remove(kvm_debugfs_dir);
1975 }
1976
1977 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1978 {
1979         hardware_disable(NULL);
1980         return 0;
1981 }
1982
1983 static int kvm_resume(struct sys_device *dev)
1984 {
1985         hardware_enable(NULL);
1986         return 0;
1987 }
1988
1989 static struct sysdev_class kvm_sysdev_class = {
1990         .name = "kvm",
1991         .suspend = kvm_suspend,
1992         .resume = kvm_resume,
1993 };
1994
1995 static struct sys_device kvm_sysdev = {
1996         .id = 0,
1997         .cls = &kvm_sysdev_class,
1998 };
1999
2000 struct page *bad_page;
2001 pfn_t bad_pfn;
2002
2003 static inline
2004 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2005 {
2006         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2007 }
2008
2009 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2010 {
2011         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2012
2013         kvm_arch_vcpu_load(vcpu, cpu);
2014 }
2015
2016 static void kvm_sched_out(struct preempt_notifier *pn,
2017                           struct task_struct *next)
2018 {
2019         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2020
2021         kvm_arch_vcpu_put(vcpu);
2022 }
2023
2024 int kvm_init(void *opaque, unsigned int vcpu_size,
2025                   struct module *module)
2026 {
2027         int r;
2028         int cpu;
2029
2030         kvm_init_debug();
2031
2032         r = kvm_arch_init(opaque);
2033         if (r)
2034                 goto out_fail;
2035
2036         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2037
2038         if (bad_page == NULL) {
2039                 r = -ENOMEM;
2040                 goto out;
2041         }
2042
2043         bad_pfn = page_to_pfn(bad_page);
2044
2045         r = kvm_arch_hardware_setup();
2046         if (r < 0)
2047                 goto out_free_0;
2048
2049         for_each_online_cpu(cpu) {
2050                 smp_call_function_single(cpu,
2051                                 kvm_arch_check_processor_compat,
2052                                 &r, 1);
2053                 if (r < 0)
2054                         goto out_free_1;
2055         }
2056
2057         on_each_cpu(hardware_enable, NULL, 1);
2058         r = register_cpu_notifier(&kvm_cpu_notifier);
2059         if (r)
2060                 goto out_free_2;
2061         register_reboot_notifier(&kvm_reboot_notifier);
2062
2063         r = sysdev_class_register(&kvm_sysdev_class);
2064         if (r)
2065                 goto out_free_3;
2066
2067         r = sysdev_register(&kvm_sysdev);
2068         if (r)
2069                 goto out_free_4;
2070
2071         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2072         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2073                                            __alignof__(struct kvm_vcpu),
2074                                            0, NULL);
2075         if (!kvm_vcpu_cache) {
2076                 r = -ENOMEM;
2077                 goto out_free_5;
2078         }
2079
2080         kvm_chardev_ops.owner = module;
2081
2082         r = misc_register(&kvm_dev);
2083         if (r) {
2084                 printk(KERN_ERR "kvm: misc device register failed\n");
2085                 goto out_free;
2086         }
2087
2088         kvm_preempt_ops.sched_in = kvm_sched_in;
2089         kvm_preempt_ops.sched_out = kvm_sched_out;
2090
2091         return 0;
2092
2093 out_free:
2094         kmem_cache_destroy(kvm_vcpu_cache);
2095 out_free_5:
2096         sysdev_unregister(&kvm_sysdev);
2097 out_free_4:
2098         sysdev_class_unregister(&kvm_sysdev_class);
2099 out_free_3:
2100         unregister_reboot_notifier(&kvm_reboot_notifier);
2101         unregister_cpu_notifier(&kvm_cpu_notifier);
2102 out_free_2:
2103         on_each_cpu(hardware_disable, NULL, 1);
2104 out_free_1:
2105         kvm_arch_hardware_unsetup();
2106 out_free_0:
2107         __free_page(bad_page);
2108 out:
2109         kvm_arch_exit();
2110         kvm_exit_debug();
2111 out_fail:
2112         return r;
2113 }
2114 EXPORT_SYMBOL_GPL(kvm_init);
2115
2116 void kvm_exit(void)
2117 {
2118         kvm_trace_cleanup();
2119         misc_deregister(&kvm_dev);
2120         kmem_cache_destroy(kvm_vcpu_cache);
2121         sysdev_unregister(&kvm_sysdev);
2122         sysdev_class_unregister(&kvm_sysdev_class);
2123         unregister_reboot_notifier(&kvm_reboot_notifier);
2124         unregister_cpu_notifier(&kvm_cpu_notifier);
2125         on_each_cpu(hardware_disable, NULL, 1);
2126         kvm_arch_hardware_unsetup();
2127         kvm_arch_exit();
2128         kvm_exit_debug();
2129         __free_page(bad_page);
2130 }
2131 EXPORT_SYMBOL_GPL(kvm_exit);