virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "iodev.h"
  19
  20 #include <linux/kvm_host.h>
  21 #include <linux/kvm.h>
  22 #include <linux/module.h>
  23 #include <linux/errno.h>
  24 #include <linux/percpu.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/sysdev.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46
  47 #include <asm/processor.h>
  48 #include <asm/io.h>
  49 #include <asm/uaccess.h>
  50 #include <asm/pgtable.h>
  51
  52 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
  53 #include "coalesced_mmio.h"
  54 #endif
  55
  56 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
  57 #include <linux/pci.h>
  58 #include <linux/interrupt.h>
  59 #include "irq.h"
  60 #endif
  61
  62 MODULE_AUTHOR("Qumranet");
  63 MODULE_LICENSE("GPL");
  64
  65 DEFINE_SPINLOCK(kvm_lock);
  66 LIST_HEAD(vm_list);
  67
  68 static cpumask_var_t cpus_hardware_enabled;
  69
  70 struct kmem_cache *kvm_vcpu_cache;
  71 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  72
  73 static __read_mostly struct preempt_ops kvm_preempt_ops;
  74
  75 struct dentry *kvm_debugfs_dir;
  76
  77 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  78                            unsigned long arg);
  79
  80 static bool kvm_rebooting;
  81
  82 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
  83 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
  84                                                       int assigned_dev_id)
  85 {
  86         struct list_head *ptr;
  87         struct kvm_assigned_dev_kernel *match;
  88
  89         list_for_each(ptr, head) {
  90                 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
  91                 if (match->assigned_dev_id == assigned_dev_id)
  92                         return match;
  93         }
  94         return NULL;
  95 }
  96
  97 static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
  98                                     *assigned_dev, int irq)
  99 {
 100         int i, index;
 101         struct msix_entry *host_msix_entries;
 102
 103         host_msix_entries = assigned_dev->host_msix_entries;
 104
 105         index = -1;
 106         for (i = 0; i < assigned_dev->entries_nr; i++)
 107                 if (irq == host_msix_entries[i].vector) {
 108                         index = i;
 109                         break;
 110                 }
 111         if (index < 0) {
 112                 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
 113                 return 0;
 114         }
 115
 116         return index;
 117 }
 118
 119 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
 120 {
 121         struct kvm_assigned_dev_kernel *assigned_dev;
 122         struct kvm *kvm;
 123         int irq, i;
 124
 125         assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
 126                                     interrupt_work);
 127         kvm = assigned_dev->kvm;
 128
 129         /* This is taken to safely inject irq inside the guest. When
 130          * the interrupt injection (or the ioapic code) uses a
 131          * finer-grained lock, update this
 132          */
 133         mutex_lock(&kvm->lock);
 134         spin_lock_irq(&assigned_dev->assigned_dev_lock);
 135         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 136                 struct kvm_guest_msix_entry *guest_entries =
 137                         assigned_dev->guest_msix_entries;
 138                 for (i = 0; i < assigned_dev->entries_nr; i++) {
 139                         if (!(guest_entries[i].flags &
 140                                         KVM_ASSIGNED_MSIX_PENDING))
 141                                 continue;
 142                         guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
 143                         kvm_set_irq(assigned_dev->kvm,
 144                                     assigned_dev->irq_source_id,
 145                                     guest_entries[i].vector, 1);
 146                         irq = assigned_dev->host_msix_entries[i].vector;
 147                         if (irq != 0)
 148                                 enable_irq(irq);
 149                         assigned_dev->host_irq_disabled = false;
 150                 }
 151         } else {
 152                 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
 153                             assigned_dev->guest_irq, 1);
 154                 if (assigned_dev->irq_requested_type &
 155                                 KVM_DEV_IRQ_GUEST_MSI) {
 156                         enable_irq(assigned_dev->host_irq);
 157                         assigned_dev->host_irq_disabled = false;
 158                 }
 159         }
 160
 161         spin_unlock_irq(&assigned_dev->assigned_dev_lock);
 162         mutex_unlock(&assigned_dev->kvm->lock);
 163 }
 164
 165 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
 166 {
 167         unsigned long flags;
 168         struct kvm_assigned_dev_kernel *assigned_dev =
 169                 (struct kvm_assigned_dev_kernel *) dev_id;
 170
 171         spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
 172         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 173                 int index = find_index_from_host_irq(assigned_dev, irq);
 174                 if (index < 0)
 175                         goto out;
 176                 assigned_dev->guest_msix_entries[index].flags |=
 177                         KVM_ASSIGNED_MSIX_PENDING;
 178         }
 179
 180         schedule_work(&assigned_dev->interrupt_work);
 181
 182         disable_irq_nosync(irq);
 183         assigned_dev->host_irq_disabled = true;
 184
 185 out:
 186         spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
 187         return IRQ_HANDLED;
 188 }
 189
 190 /* Ack the irq line for an assigned device */
 191 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
 192 {
 193         struct kvm_assigned_dev_kernel *dev;
 194         unsigned long flags;
 195
 196         if (kian->gsi == -1)
 197                 return;
 198
 199         dev = container_of(kian, struct kvm_assigned_dev_kernel,
 200                            ack_notifier);
 201
 202         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
 203
 204         /* The guest irq may be shared so this ack may be
 205          * from another device.
 206          */
 207         spin_lock_irqsave(&dev->assigned_dev_lock, flags);
 208         if (dev->host_irq_disabled) {
 209                 enable_irq(dev->host_irq);
 210                 dev->host_irq_disabled = false;
 211         }
 212         spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
 213 }
 214
 215 static void deassign_guest_irq(struct kvm *kvm,
 216                                struct kvm_assigned_dev_kernel *assigned_dev)
 217 {
 218         kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
 219         assigned_dev->ack_notifier.gsi = -1;
 220
 221         if (assigned_dev->irq_source_id != -1)
 222                 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
 223         assigned_dev->irq_source_id = -1;
 224         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
 225 }
 226
 227 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
 228 static void deassign_host_irq(struct kvm *kvm,
 229                               struct kvm_assigned_dev_kernel *assigned_dev)
 230 {
 231         /*
 232          * In kvm_free_device_irq, cancel_work_sync return true if:
 233          * 1. work is scheduled, and then cancelled.
 234          * 2. work callback is executed.
 235          *
 236          * The first one ensured that the irq is disabled and no more events
 237          * would happen. But for the second one, the irq may be enabled (e.g.
 238          * for MSI). So we disable irq here to prevent further events.
 239          *
 240          * Notice this maybe result in nested disable if the interrupt type is
 241          * INTx, but it's OK for we are going to free it.
 242          *
 243          * If this function is a part of VM destroy, please ensure that till
 244          * now, the kvm state is still legal for probably we also have to wait
 245          * interrupt_work done.
 246          */
 247         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 248                 int i;
 249                 for (i = 0; i < assigned_dev->entries_nr; i++)
 250                         disable_irq_nosync(assigned_dev->
 251                                            host_msix_entries[i].vector);
 252
 253                 cancel_work_sync(&assigned_dev->interrupt_work);
 254
 255                 for (i = 0; i < assigned_dev->entries_nr; i++)
 256                         free_irq(assigned_dev->host_msix_entries[i].vector,
 257                                  (void *)assigned_dev);
 258
 259                 assigned_dev->entries_nr = 0;
 260                 kfree(assigned_dev->host_msix_entries);
 261                 kfree(assigned_dev->guest_msix_entries);
 262                 pci_disable_msix(assigned_dev->dev);
 263         } else {
 264                 /* Deal with MSI and INTx */
 265                 disable_irq_nosync(assigned_dev->host_irq);
 266                 cancel_work_sync(&assigned_dev->interrupt_work);
 267
 268                 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
 269
 270                 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
 271                         pci_disable_msi(assigned_dev->dev);
 272         }
 273
 274         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
 275 }
 276
 277 static int kvm_deassign_irq(struct kvm *kvm,
 278                             struct kvm_assigned_dev_kernel *assigned_dev,
 279                             unsigned long irq_requested_type)
 280 {
 281         unsigned long guest_irq_type, host_irq_type;
 282
 283         if (!irqchip_in_kernel(kvm))
 284                 return -EINVAL;
 285         /* no irq assignment to deassign */
 286         if (!assigned_dev->irq_requested_type)
 287                 return -ENXIO;
 288
 289         host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
 290         guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
 291
 292         if (host_irq_type)
 293                 deassign_host_irq(kvm, assigned_dev);
 294         if (guest_irq_type)
 295                 deassign_guest_irq(kvm, assigned_dev);
 296
 297         return 0;
 298 }
 299
 300 static void kvm_free_assigned_irq(struct kvm *kvm,
 301                                   struct kvm_assigned_dev_kernel *assigned_dev)
 302 {
 303         kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
 304 }
 305
 306 static void kvm_free_assigned_device(struct kvm *kvm,
 307                                      struct kvm_assigned_dev_kernel
 308                                      *assigned_dev)
 309 {
 310         kvm_free_assigned_irq(kvm, assigned_dev);
 311
 312         pci_reset_function(assigned_dev->dev);
 313
 314         pci_release_regions(assigned_dev->dev);
 315         pci_disable_device(assigned_dev->dev);
 316         pci_dev_put(assigned_dev->dev);
 317
 318         list_del(&assigned_dev->list);
 319         kfree(assigned_dev);
 320 }
 321
 322 void kvm_free_all_assigned_devices(struct kvm *kvm)
 323 {
 324         struct list_head *ptr, *ptr2;
 325         struct kvm_assigned_dev_kernel *assigned_dev;
 326
 327         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
 328                 assigned_dev = list_entry(ptr,
 329                                           struct kvm_assigned_dev_kernel,
 330                                           list);
 331
 332                 kvm_free_assigned_device(kvm, assigned_dev);
 333         }
 334 }
 335
 336 static int assigned_device_enable_host_intx(struct kvm *kvm,
 337                                             struct kvm_assigned_dev_kernel *dev)
 338 {
 339         dev->host_irq = dev->dev->irq;
 340         /* Even though this is PCI, we don't want to use shared
 341          * interrupts. Sharing host devices with guest-assigned devices
 342          * on the same interrupt line is not a happy situation: there
 343          * are going to be long delays in accepting, acking, etc.
 344          */
 345         if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
 346                         0, "kvm_assigned_intx_device", (void *)dev))
 347                 return -EIO;
 348         return 0;
 349 }
 350
 351 #ifdef __KVM_HAVE_MSI
 352 static int assigned_device_enable_host_msi(struct kvm *kvm,
 353                                            struct kvm_assigned_dev_kernel *dev)
 354 {
 355         int r;
 356
 357         if (!dev->dev->msi_enabled) {
 358                 r = pci_enable_msi(dev->dev);
 359                 if (r)
 360                         return r;
 361         }
 362
 363         dev->host_irq = dev->dev->irq;
 364         if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
 365                         "kvm_assigned_msi_device", (void *)dev)) {
 366                 pci_disable_msi(dev->dev);
 367                 return -EIO;
 368         }
 369
 370         return 0;
 371 }
 372 #endif
 373
 374 #ifdef __KVM_HAVE_MSIX
 375 static int assigned_device_enable_host_msix(struct kvm *kvm,
 376                                             struct kvm_assigned_dev_kernel *dev)
 377 {
 378         int i, r = -EINVAL;
 379
 380         /* host_msix_entries and guest_msix_entries should have been
 381          * initialized */
 382         if (dev->entries_nr == 0)
 383                 return r;
 384
 385         r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
 386         if (r)
 387                 return r;
 388
 389         for (i = 0; i < dev->entries_nr; i++) {
 390                 r = request_irq(dev->host_msix_entries[i].vector,
 391                                 kvm_assigned_dev_intr, 0,
 392                                 "kvm_assigned_msix_device",
 393                                 (void *)dev);
 394                 /* FIXME: free requested_irq's on failure */
 395                 if (r)
 396                         return r;
 397         }
 398
 399         return 0;
 400 }
 401
 402 #endif
 403
 404 static int assigned_device_enable_guest_intx(struct kvm *kvm,
 405                                 struct kvm_assigned_dev_kernel *dev,
 406                                 struct kvm_assigned_irq *irq)
 407 {
 408         dev->guest_irq = irq->guest_irq;
 409         dev->ack_notifier.gsi = irq->guest_irq;
 410         return 0;
 411 }
 412
 413 #ifdef __KVM_HAVE_MSI
 414 static int assigned_device_enable_guest_msi(struct kvm *kvm,
 415                         struct kvm_assigned_dev_kernel *dev,
 416                         struct kvm_assigned_irq *irq)
 417 {
 418         dev->guest_irq = irq->guest_irq;
 419         dev->ack_notifier.gsi = -1;
 420         return 0;
 421 }
 422 #endif
 423 #ifdef __KVM_HAVE_MSIX
 424 static int assigned_device_enable_guest_msix(struct kvm *kvm,
 425                         struct kvm_assigned_dev_kernel *dev,
 426                         struct kvm_assigned_irq *irq)
 427 {
 428         dev->guest_irq = irq->guest_irq;
 429         dev->ack_notifier.gsi = -1;
 430         return 0;
 431 }
 432 #endif
 433
 434 static int assign_host_irq(struct kvm *kvm,
 435                            struct kvm_assigned_dev_kernel *dev,
 436                            __u32 host_irq_type)
 437 {
 438         int r = -EEXIST;
 439
 440         if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
 441                 return r;
 442
 443         switch (host_irq_type) {
 444         case KVM_DEV_IRQ_HOST_INTX:
 445                 r = assigned_device_enable_host_intx(kvm, dev);
 446                 break;
 447 #ifdef __KVM_HAVE_MSI
 448         case KVM_DEV_IRQ_HOST_MSI:
 449                 r = assigned_device_enable_host_msi(kvm, dev);
 450                 break;
 451 #endif
 452 #ifdef __KVM_HAVE_MSIX
 453         case KVM_DEV_IRQ_HOST_MSIX:
 454                 r = assigned_device_enable_host_msix(kvm, dev);
 455                 break;
 456 #endif
 457         default:
 458                 r = -EINVAL;
 459         }
 460
 461         if (!r)
 462                 dev->irq_requested_type |= host_irq_type;
 463
 464         return r;
 465 }
 466
 467 static int assign_guest_irq(struct kvm *kvm,
 468                             struct kvm_assigned_dev_kernel *dev,
 469                             struct kvm_assigned_irq *irq,
 470                             unsigned long guest_irq_type)
 471 {
 472         int id;
 473         int r = -EEXIST;
 474
 475         if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
 476                 return r;
 477
 478         id = kvm_request_irq_source_id(kvm);
 479         if (id < 0)
 480                 return id;
 481
 482         dev->irq_source_id = id;
 483
 484         switch (guest_irq_type) {
 485         case KVM_DEV_IRQ_GUEST_INTX:
 486                 r = assigned_device_enable_guest_intx(kvm, dev, irq);
 487                 break;
 488 #ifdef __KVM_HAVE_MSI
 489         case KVM_DEV_IRQ_GUEST_MSI:
 490                 r = assigned_device_enable_guest_msi(kvm, dev, irq);
 491                 break;
 492 #endif
 493 #ifdef __KVM_HAVE_MSIX
 494         case KVM_DEV_IRQ_GUEST_MSIX:
 495                 r = assigned_device_enable_guest_msix(kvm, dev, irq);
 496                 break;
 497 #endif
 498         default:
 499                 r = -EINVAL;
 500         }
 501
 502         if (!r) {
 503                 dev->irq_requested_type |= guest_irq_type;
 504                 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
 505         } else
 506                 kvm_free_irq_source_id(kvm, dev->irq_source_id);
 507
 508         return r;
 509 }
 510
 511 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
 512 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 513                                    struct kvm_assigned_irq *assigned_irq)
 514 {
 515         int r = -EINVAL;
 516         struct kvm_assigned_dev_kernel *match;
 517         unsigned long host_irq_type, guest_irq_type;
 518
 519         if (!capable(CAP_SYS_RAWIO))
 520                 return -EPERM;
 521
 522         if (!irqchip_in_kernel(kvm))
 523                 return r;
 524
 525         mutex_lock(&kvm->lock);
 526         r = -ENODEV;
 527         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 528                                       assigned_irq->assigned_dev_id);
 529         if (!match)
 530                 goto out;
 531
 532         host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
 533         guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
 534
 535         r = -EINVAL;
 536         /* can only assign one type at a time */
 537         if (hweight_long(host_irq_type) > 1)
 538                 goto out;
 539         if (hweight_long(guest_irq_type) > 1)
 540                 goto out;
 541         if (host_irq_type == 0 && guest_irq_type == 0)
 542                 goto out;
 543
 544         r = 0;
 545         if (host_irq_type)
 546                 r = assign_host_irq(kvm, match, host_irq_type);
 547         if (r)
 548                 goto out;
 549
 550         if (guest_irq_type)
 551                 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
 552 out:
 553         mutex_unlock(&kvm->lock);
 554         return r;
 555 }
 556
 557 static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
 558                                          struct kvm_assigned_irq
 559                                          *assigned_irq)
 560 {
 561         int r = -ENODEV;
 562         struct kvm_assigned_dev_kernel *match;
 563
 564         mutex_lock(&kvm->lock);
 565
 566         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 567                                       assigned_irq->assigned_dev_id);
 568         if (!match)
 569                 goto out;
 570
 571         r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
 572 out:
 573         mutex_unlock(&kvm->lock);
 574         return r;
 575 }
 576
 577 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 578                                       struct kvm_assigned_pci_dev *assigned_dev)
 579 {
 580         int r = 0;
 581         struct kvm_assigned_dev_kernel *match;
 582         struct pci_dev *dev;
 583
 584         down_read(&kvm->slots_lock);
 585         mutex_lock(&kvm->lock);
 586
 587         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 588                                       assigned_dev->assigned_dev_id);
 589         if (match) {
 590                 /* device already assigned */
 591                 r = -EEXIST;
 592                 goto out;
 593         }
 594
 595         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 596         if (match == NULL) {
 597                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 598                        __func__);
 599                 r = -ENOMEM;
 600                 goto out;
 601         }
 602         dev = pci_get_bus_and_slot(assigned_dev->busnr,
 603                                    assigned_dev->devfn);
 604         if (!dev) {
 605                 printk(KERN_INFO "%s: host device not found\n", __func__);
 606                 r = -EINVAL;
 607                 goto out_free;
 608         }
 609         if (pci_enable_device(dev)) {
 610                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 611                 r = -EBUSY;
 612                 goto out_put;
 613         }
 614         r = pci_request_regions(dev, "kvm_assigned_device");
 615         if (r) {
 616                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 617                        __func__);
 618                 goto out_disable;
 619         }
 620
 621         pci_reset_function(dev);
 622
 623         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 624         match->host_busnr = assigned_dev->busnr;
 625         match->host_devfn = assigned_dev->devfn;
 626         match->flags = assigned_dev->flags;
 627         match->dev = dev;
 628         spin_lock_init(&match->assigned_dev_lock);
 629         match->irq_source_id = -1;
 630         match->kvm = kvm;
 631         match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
 632         INIT_WORK(&match->interrupt_work,
 633                   kvm_assigned_dev_interrupt_work_handler);
 634
 635         list_add(&match->list, &kvm->arch.assigned_dev_head);
 636
 637         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
 638                 if (!kvm->arch.iommu_domain) {
 639                         r = kvm_iommu_map_guest(kvm);
 640                         if (r)
 641                                 goto out_list_del;
 642                 }
 643                 r = kvm_assign_device(kvm, match);
 644                 if (r)
 645                         goto out_list_del;
 646         }
 647
 648 out:
 649         mutex_unlock(&kvm->lock);
 650         up_read(&kvm->slots_lock);
 651         return r;
 652 out_list_del:
 653         list_del(&match->list);
 654         pci_release_regions(dev);
 655 out_disable:
 656         pci_disable_device(dev);
 657 out_put:
 658         pci_dev_put(dev);
 659 out_free:
 660         kfree(match);
 661         mutex_unlock(&kvm->lock);
 662         up_read(&kvm->slots_lock);
 663         return r;
 664 }
 665 #endif
 666
 667 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
 668 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
 669                 struct kvm_assigned_pci_dev *assigned_dev)
 670 {
 671         int r = 0;
 672         struct kvm_assigned_dev_kernel *match;
 673
 674         mutex_lock(&kvm->lock);
 675
 676         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 677                                       assigned_dev->assigned_dev_id);
 678         if (!match) {
 679                 printk(KERN_INFO "%s: device hasn't been assigned before, "
 680                   "so cannot be deassigned\n", __func__);
 681                 r = -EINVAL;
 682                 goto out;
 683         }
 684
 685         if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
 686                 kvm_deassign_device(kvm, match);
 687
 688         kvm_free_assigned_device(kvm, match);
 689
 690 out:
 691         mutex_unlock(&kvm->lock);
 692         return r;
 693 }
 694 #endif
 695
 696 static inline int valid_vcpu(int n)
 697 {
 698         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 699 }
 700
 701 inline int kvm_is_mmio_pfn(pfn_t pfn)
 702 {
 703         if (pfn_valid(pfn)) {
 704                 struct page *page = compound_head(pfn_to_page(pfn));
 705                 return PageReserved(page);
 706         }
 707
 708         return true;
 709 }
 710
 711 /*
 712  * Switches to specified vcpu, until a matching vcpu_put()
 713  */
 714 void vcpu_load(struct kvm_vcpu *vcpu)
 715 {
 716         int cpu;
 717
 718         mutex_lock(&vcpu->mutex);
 719         cpu = get_cpu();
 720         preempt_notifier_register(&vcpu->preempt_notifier);
 721         kvm_arch_vcpu_load(vcpu, cpu);
 722         put_cpu();
 723 }
 724
 725 void vcpu_put(struct kvm_vcpu *vcpu)
 726 {
 727         preempt_disable();
 728         kvm_arch_vcpu_put(vcpu);
 729         preempt_notifier_unregister(&vcpu->preempt_notifier);
 730         preempt_enable();
 731         mutex_unlock(&vcpu->mutex);
 732 }
 733
 734 static void ack_flush(void *_completed)
 735 {
 736 }
 737
 738 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 739 {
 740         int i, cpu, me;
 741         cpumask_var_t cpus;
 742         bool called = true;
 743         struct kvm_vcpu *vcpu;
 744
 745         if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
 746                 cpumask_clear(cpus);
 747
 748         me = get_cpu();
 749         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 750                 vcpu = kvm->vcpus[i];
 751                 if (!vcpu)
 752                         continue;
 753                 if (test_and_set_bit(req, &vcpu->requests))
 754                         continue;
 755                 cpu = vcpu->cpu;
 756                 if (cpus != NULL && cpu != -1 && cpu != me)
 757                         cpumask_set_cpu(cpu, cpus);
 758         }
 759         if (unlikely(cpus == NULL))
 760                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 761         else if (!cpumask_empty(cpus))
 762                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 763         else
 764                 called = false;
 765         put_cpu();
 766         free_cpumask_var(cpus);
 767         return called;
 768 }
 769
 770 void kvm_flush_remote_tlbs(struct kvm *kvm)
 771 {
 772         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 773                 ++kvm->stat.remote_tlb_flush;
 774 }
 775
 776 void kvm_reload_remote_mmus(struct kvm *kvm)
 777 {
 778         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 779 }
 780
 781 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 782 {
 783         struct page *page;
 784         int r;
 785
 786         mutex_init(&vcpu->mutex);
 787         vcpu->cpu = -1;
 788         vcpu->kvm = kvm;
 789         vcpu->vcpu_id = id;
 790         init_waitqueue_head(&vcpu->wq);
 791
 792         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 793         if (!page) {
 794                 r = -ENOMEM;
 795                 goto fail;
 796         }
 797         vcpu->run = page_address(page);
 798
 799         r = kvm_arch_vcpu_init(vcpu);
 800         if (r < 0)
 801                 goto fail_free_run;
 802         return 0;
 803
 804 fail_free_run:
 805         free_page((unsigned long)vcpu->run);
 806 fail:
 807         return r;
 808 }
 809 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 810
 811 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 812 {
 813         kvm_arch_vcpu_uninit(vcpu);
 814         free_page((unsigned long)vcpu->run);
 815 }
 816 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 817
 818 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 819 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 820 {
 821         return container_of(mn, struct kvm, mmu_notifier);
 822 }
 823
 824 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 825                                              struct mm_struct *mm,
 826                                              unsigned long address)
 827 {
 828         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 829         int need_tlb_flush;
 830
 831         /*
 832          * When ->invalidate_page runs, the linux pte has been zapped
 833          * already but the page is still allocated until
 834          * ->invalidate_page returns. So if we increase the sequence
 835          * here the kvm page fault will notice if the spte can't be
 836          * established because the page is going to be freed. If
 837          * instead the kvm page fault establishes the spte before
 838          * ->invalidate_page runs, kvm_unmap_hva will release it
 839          * before returning.
 840          *
 841          * The sequence increase only need to be seen at spin_unlock
 842          * time, and not at spin_lock time.
 843          *
 844          * Increasing the sequence after the spin_unlock would be
 845          * unsafe because the kvm page fault could then establish the
 846          * pte after kvm_unmap_hva returned, without noticing the page
 847          * is going to be freed.
 848          */
 849         spin_lock(&kvm->mmu_lock);
 850         kvm->mmu_notifier_seq++;
 851         need_tlb_flush = kvm_unmap_hva(kvm, address);
 852         spin_unlock(&kvm->mmu_lock);
 853
 854         /* we've to flush the tlb before the pages can be freed */
 855         if (need_tlb_flush)
 856                 kvm_flush_remote_tlbs(kvm);
 857
 858 }
 859
 860 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 861                                                     struct mm_struct *mm,
 862                                                     unsigned long start,
 863                                                     unsigned long end)
 864 {
 865         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 866         int need_tlb_flush = 0;
 867
 868         spin_lock(&kvm->mmu_lock);
 869         /*
 870          * The count increase must become visible at unlock time as no
 871          * spte can be established without taking the mmu_lock and
 872          * count is also read inside the mmu_lock critical section.
 873          */
 874         kvm->mmu_notifier_count++;
 875         for (; start < end; start += PAGE_SIZE)
 876                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 877         spin_unlock(&kvm->mmu_lock);
 878
 879         /* we've to flush the tlb before the pages can be freed */
 880         if (need_tlb_flush)
 881                 kvm_flush_remote_tlbs(kvm);
 882 }
 883
 884 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 885                                                   struct mm_struct *mm,
 886                                                   unsigned long start,
 887                                                   unsigned long end)
 888 {
 889         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 890
 891         spin_lock(&kvm->mmu_lock);
 892         /*
 893          * This sequence increase will notify the kvm page fault that
 894          * the page that is going to be mapped in the spte could have
 895          * been freed.
 896          */
 897         kvm->mmu_notifier_seq++;
 898         /*
 899          * The above sequence increase must be visible before the
 900          * below count decrease but both values are read by the kvm
 901          * page fault under mmu_lock spinlock so we don't need to add
 902          * a smb_wmb() here in between the two.
 903          */
 904         kvm->mmu_notifier_count--;
 905         spin_unlock(&kvm->mmu_lock);
 906
 907         BUG_ON(kvm->mmu_notifier_count < 0);
 908 }
 909
 910 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 911                                               struct mm_struct *mm,
 912                                               unsigned long address)
 913 {
 914         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 915         int young;
 916
 917         spin_lock(&kvm->mmu_lock);
 918         young = kvm_age_hva(kvm, address);
 919         spin_unlock(&kvm->mmu_lock);
 920
 921         if (young)
 922                 kvm_flush_remote_tlbs(kvm);
 923
 924         return young;
 925 }
 926
 927 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 928                                      struct mm_struct *mm)
 929 {
 930         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 931         kvm_arch_flush_shadow(kvm);
 932 }
 933
 934 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 935         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 936         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 937         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 938         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 939         .release                = kvm_mmu_notifier_release,
 940 };
 941 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 942
 943 static struct kvm *kvm_create_vm(void)
 944 {
 945         struct kvm *kvm = kvm_arch_create_vm();
 946 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 947         struct page *page;
 948 #endif
 949
 950         if (IS_ERR(kvm))
 951                 goto out;
 952 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 953         INIT_LIST_HEAD(&kvm->irq_routing);
 954         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 955 #endif
 956
 957 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 958         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 959         if (!page) {
 960                 kfree(kvm);
 961                 return ERR_PTR(-ENOMEM);
 962         }
 963         kvm->coalesced_mmio_ring =
 964                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 965 #endif
 966
 967 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 968         {
 969                 int err;
 970                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 971                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 972                 if (err) {
 973 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 974                         put_page(page);
 975 #endif
 976                         kfree(kvm);
 977                         return ERR_PTR(err);
 978                 }
 979         }
 980 #endif
 981
 982         kvm->mm = current->mm;
 983         atomic_inc(&kvm->mm->mm_count);
 984         spin_lock_init(&kvm->mmu_lock);
 985         kvm_io_bus_init(&kvm->pio_bus);
 986         mutex_init(&kvm->lock);
 987         kvm_io_bus_init(&kvm->mmio_bus);
 988         init_rwsem(&kvm->slots_lock);
 989         atomic_set(&kvm->users_count, 1);
 990         spin_lock(&kvm_lock);
 991         list_add(&kvm->vm_list, &vm_list);
 992         spin_unlock(&kvm_lock);
 993 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 994         kvm_coalesced_mmio_init(kvm);
 995 #endif
 996 out:
 997         return kvm;
 998 }
 999
1000 /*
1001  * Free any memory in @free but not in @dont.
1002  */
1003 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
1004                                   struct kvm_memory_slot *dont)
1005 {
1006         if (!dont || free->rmap != dont->rmap)
1007                 vfree(free->rmap);
1008
1009         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
1010                 vfree(free->dirty_bitmap);
1011
1012         if (!dont || free->lpage_info != dont->lpage_info)
1013                 vfree(free->lpage_info);
1014
1015         free->npages = 0;
1016         free->dirty_bitmap = NULL;
1017         free->rmap = NULL;
1018         free->lpage_info = NULL;
1019 }
1020
1021 void kvm_free_physmem(struct kvm *kvm)
1022 {
1023         int i;
1024
1025         for (i = 0; i < kvm->nmemslots; ++i)
1026                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
1027 }
1028
1029 static void kvm_destroy_vm(struct kvm *kvm)
1030 {
1031         struct mm_struct *mm = kvm->mm;
1032
1033         kvm_arch_sync_events(kvm);
1034         spin_lock(&kvm_lock);
1035         list_del(&kvm->vm_list);
1036         spin_unlock(&kvm_lock);
1037         kvm_free_irq_routing(kvm);
1038         kvm_io_bus_destroy(&kvm->pio_bus);
1039         kvm_io_bus_destroy(&kvm->mmio_bus);
1040 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1041         if (kvm->coalesced_mmio_ring != NULL)
1042                 free_page((unsigned long)kvm->coalesced_mmio_ring);
1043 #endif
1044 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1045         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1046 #else
1047         kvm_arch_flush_shadow(kvm);
1048 #endif
1049         kvm_arch_destroy_vm(kvm);
1050         mmdrop(mm);
1051 }
1052
1053 void kvm_get_kvm(struct kvm *kvm)
1054 {
1055         atomic_inc(&kvm->users_count);
1056 }
1057 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1058
1059 void kvm_put_kvm(struct kvm *kvm)
1060 {
1061         if (atomic_dec_and_test(&kvm->users_count))
1062                 kvm_destroy_vm(kvm);
1063 }
1064 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1065
1066
1067 static int kvm_vm_release(struct inode *inode, struct file *filp)
1068 {
1069         struct kvm *kvm = filp->private_data;
1070
1071         kvm_put_kvm(kvm);
1072         return 0;
1073 }
1074
1075 /*
1076  * Allocate some memory and give it an address in the guest physical address
1077  * space.
1078  *
1079  * Discontiguous memory is allowed, mostly for framebuffers.
1080  *
1081  * Must be called holding mmap_sem for write.
1082  */
1083 int __kvm_set_memory_region(struct kvm *kvm,
1084                             struct kvm_userspace_memory_region *mem,
1085                             int user_alloc)
1086 {
1087         int r;
1088         gfn_t base_gfn;
1089         unsigned long npages, ugfn;
1090         unsigned long largepages, i;
1091         struct kvm_memory_slot *memslot;
1092         struct kvm_memory_slot old, new;
1093
1094         r = -EINVAL;
1095         /* General sanity checks */
1096         if (mem->memory_size & (PAGE_SIZE - 1))
1097                 goto out;
1098         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1099                 goto out;
1100         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
1101                 goto out;
1102         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
1103                 goto out;
1104         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1105                 goto out;
1106
1107         memslot = &kvm->memslots[mem->slot];
1108         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1109         npages = mem->memory_size >> PAGE_SHIFT;
1110
1111         if (!npages)
1112                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1113
1114         new = old = *memslot;
1115
1116         new.base_gfn = base_gfn;
1117         new.npages = npages;
1118         new.flags = mem->flags;
1119
1120         /* Disallow changing a memory slot's size. */
1121         r = -EINVAL;
1122         if (npages && old.npages && npages != old.npages)
1123                 goto out_free;
1124
1125         /* Check for overlaps */
1126         r = -EEXIST;
1127         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1128                 struct kvm_memory_slot *s = &kvm->memslots[i];
1129
1130                 if (s == memslot || !s->npages)
1131                         continue;
1132                 if (!((base_gfn + npages <= s->base_gfn) ||
1133                       (base_gfn >= s->base_gfn + s->npages)))
1134                         goto out_free;
1135         }
1136
1137         /* Free page dirty bitmap if unneeded */
1138         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1139                 new.dirty_bitmap = NULL;
1140
1141         r = -ENOMEM;
1142
1143         /* Allocate if a slot is being created */
1144 #ifndef CONFIG_S390
1145         if (npages && !new.rmap) {
1146                 new.rmap = vmalloc(npages * sizeof(struct page *));
1147
1148                 if (!new.rmap)
1149                         goto out_free;
1150
1151                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1152
1153                 new.user_alloc = user_alloc;
1154                 /*
1155                  * hva_to_rmmap() serialzies with the mmu_lock and to be
1156                  * safe it has to ignore memslots with !user_alloc &&
1157                  * !userspace_addr.
1158                  */
1159                 if (user_alloc)
1160                         new.userspace_addr = mem->userspace_addr;
1161                 else
1162                         new.userspace_addr = 0;
1163         }
1164         if (npages && !new.lpage_info) {
1165                 largepages = 1 + (base_gfn + npages - 1) / KVM_PAGES_PER_HPAGE;
1166                 largepages -= base_gfn / KVM_PAGES_PER_HPAGE;
1167
1168                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1169
1170                 if (!new.lpage_info)
1171                         goto out_free;
1172
1173                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1174
1175                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1176                         new.lpage_info[0].write_count = 1;
1177                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1178                         new.lpage_info[largepages-1].write_count = 1;
1179                 ugfn = new.userspace_addr >> PAGE_SHIFT;
1180                 /*
1181                  * If the gfn and userspace address are not aligned wrt each
1182                  * other, disable large page support for this slot
1183                  */
1184                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE - 1))
1185                         for (i = 0; i < largepages; ++i)
1186                                 new.lpage_info[i].write_count = 1;
1187         }
1188
1189         /* Allocate page dirty bitmap if needed */
1190         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1191                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1192
1193                 new.dirty_bitmap = vmalloc(dirty_bytes);
1194                 if (!new.dirty_bitmap)
1195                         goto out_free;
1196                 memset(new.dirty_bitmap, 0, dirty_bytes);
1197         }
1198 #endif /* not defined CONFIG_S390 */
1199
1200         if (!npages)
1201                 kvm_arch_flush_shadow(kvm);
1202
1203         spin_lock(&kvm->mmu_lock);
1204         if (mem->slot >= kvm->nmemslots)
1205                 kvm->nmemslots = mem->slot + 1;
1206
1207         *memslot = new;
1208         spin_unlock(&kvm->mmu_lock);
1209
1210         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1211         if (r) {
1212                 spin_lock(&kvm->mmu_lock);
1213                 *memslot = old;
1214                 spin_unlock(&kvm->mmu_lock);
1215                 goto out_free;
1216         }
1217
1218         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1219         /* Slot deletion case: we have to update the current slot */
1220         spin_lock(&kvm->mmu_lock);
1221         if (!npages)
1222                 *memslot = old;
1223         spin_unlock(&kvm->mmu_lock);
1224 #ifdef CONFIG_DMAR
1225         /* map the pages in iommu page table */
1226         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1227         if (r)
1228                 goto out;
1229 #endif
1230         return 0;
1231
1232 out_free:
1233         kvm_free_physmem_slot(&new, &old);
1234 out:
1235         return r;
1236
1237 }
1238 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1239
1240 int kvm_set_memory_region(struct kvm *kvm,
1241                           struct kvm_userspace_memory_region *mem,
1242                           int user_alloc)
1243 {
1244         int r;
1245
1246         down_write(&kvm->slots_lock);
1247         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1248         up_write(&kvm->slots_lock);
1249         return r;
1250 }
1251 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1252
1253 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1254                                    struct
1255                                    kvm_userspace_memory_region *mem,
1256                                    int user_alloc)
1257 {
1258         if (mem->slot >= KVM_MEMORY_SLOTS)
1259                 return -EINVAL;
1260         return kvm_set_memory_region(kvm, mem, user_alloc);
1261 }
1262
1263 int kvm_get_dirty_log(struct kvm *kvm,
1264                         struct kvm_dirty_log *log, int *is_dirty)
1265 {
1266         struct kvm_memory_slot *memslot;
1267         int r, i;
1268         int n;
1269         unsigned long any = 0;
1270
1271         r = -EINVAL;
1272         if (log->slot >= KVM_MEMORY_SLOTS)
1273                 goto out;
1274
1275         memslot = &kvm->memslots[log->slot];
1276         r = -ENOENT;
1277         if (!memslot->dirty_bitmap)
1278                 goto out;
1279
1280         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1281
1282         for (i = 0; !any && i < n/sizeof(long); ++i)
1283                 any = memslot->dirty_bitmap[i];
1284
1285         r = -EFAULT;
1286         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1287                 goto out;
1288
1289         if (any)
1290                 *is_dirty = 1;
1291
1292         r = 0;
1293 out:
1294         return r;
1295 }
1296
1297 int is_error_page(struct page *page)
1298 {
1299         return page == bad_page;
1300 }
1301 EXPORT_SYMBOL_GPL(is_error_page);
1302
1303 int is_error_pfn(pfn_t pfn)
1304 {
1305         return pfn == bad_pfn;
1306 }
1307 EXPORT_SYMBOL_GPL(is_error_pfn);
1308
1309 static inline unsigned long bad_hva(void)
1310 {
1311         return PAGE_OFFSET;
1312 }
1313
1314 int kvm_is_error_hva(unsigned long addr)
1315 {
1316         return addr == bad_hva();
1317 }
1318 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1319
1320 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1321 {
1322         int i;
1323
1324         for (i = 0; i < kvm->nmemslots; ++i) {
1325                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1326
1327                 if (gfn >= memslot->base_gfn
1328                     && gfn < memslot->base_gfn + memslot->npages)
1329                         return memslot;
1330         }
1331         return NULL;
1332 }
1333 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1334
1335 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1336 {
1337         gfn = unalias_gfn(kvm, gfn);
1338         return gfn_to_memslot_unaliased(kvm, gfn);
1339 }
1340
1341 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1342 {
1343         int i;
1344
1345         gfn = unalias_gfn(kvm, gfn);
1346         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1347                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1348
1349                 if (gfn >= memslot->base_gfn
1350                     && gfn < memslot->base_gfn + memslot->npages)
1351                         return 1;
1352         }
1353         return 0;
1354 }
1355 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1356
1357 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1358 {
1359         struct kvm_memory_slot *slot;
1360
1361         gfn = unalias_gfn(kvm, gfn);
1362         slot = gfn_to_memslot_unaliased(kvm, gfn);
1363         if (!slot)
1364                 return bad_hva();
1365         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1366 }
1367 EXPORT_SYMBOL_GPL(gfn_to_hva);
1368
1369 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1370 {
1371         struct page *page[1];
1372         unsigned long addr;
1373         int npages;
1374         pfn_t pfn;
1375
1376         might_sleep();
1377
1378         addr = gfn_to_hva(kvm, gfn);
1379         if (kvm_is_error_hva(addr)) {
1380                 get_page(bad_page);
1381                 return page_to_pfn(bad_page);
1382         }
1383
1384         npages = get_user_pages_fast(addr, 1, 1, page);
1385
1386         if (unlikely(npages != 1)) {
1387                 struct vm_area_struct *vma;
1388
1389                 down_read(&current->mm->mmap_sem);
1390                 vma = find_vma(current->mm, addr);
1391
1392                 if (vma == NULL || addr < vma->vm_start ||
1393                     !(vma->vm_flags & VM_PFNMAP)) {
1394                         up_read(&current->mm->mmap_sem);
1395                         get_page(bad_page);
1396                         return page_to_pfn(bad_page);
1397                 }
1398
1399                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1400                 up_read(&current->mm->mmap_sem);
1401                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1402         } else
1403                 pfn = page_to_pfn(page[0]);
1404
1405         return pfn;
1406 }
1407
1408 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1409
1410 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1411 {
1412         pfn_t pfn;
1413
1414         pfn = gfn_to_pfn(kvm, gfn);
1415         if (!kvm_is_mmio_pfn(pfn))
1416                 return pfn_to_page(pfn);
1417
1418         WARN_ON(kvm_is_mmio_pfn(pfn));
1419
1420         get_page(bad_page);
1421         return bad_page;
1422 }
1423
1424 EXPORT_SYMBOL_GPL(gfn_to_page);
1425
1426 void kvm_release_page_clean(struct page *page)
1427 {
1428         kvm_release_pfn_clean(page_to_pfn(page));
1429 }
1430 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1431
1432 void kvm_release_pfn_clean(pfn_t pfn)
1433 {
1434         if (!kvm_is_mmio_pfn(pfn))
1435                 put_page(pfn_to_page(pfn));
1436 }
1437 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1438
1439 void kvm_release_page_dirty(struct page *page)
1440 {
1441         kvm_release_pfn_dirty(page_to_pfn(page));
1442 }
1443 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1444
1445 void kvm_release_pfn_dirty(pfn_t pfn)
1446 {
1447         kvm_set_pfn_dirty(pfn);
1448         kvm_release_pfn_clean(pfn);
1449 }
1450 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1451
1452 void kvm_set_page_dirty(struct page *page)
1453 {
1454         kvm_set_pfn_dirty(page_to_pfn(page));
1455 }
1456 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1457
1458 void kvm_set_pfn_dirty(pfn_t pfn)
1459 {
1460         if (!kvm_is_mmio_pfn(pfn)) {
1461                 struct page *page = pfn_to_page(pfn);
1462                 if (!PageReserved(page))
1463                         SetPageDirty(page);
1464         }
1465 }
1466 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1467
1468 void kvm_set_pfn_accessed(pfn_t pfn)
1469 {
1470         if (!kvm_is_mmio_pfn(pfn))
1471                 mark_page_accessed(pfn_to_page(pfn));
1472 }
1473 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1474
1475 void kvm_get_pfn(pfn_t pfn)
1476 {
1477         if (!kvm_is_mmio_pfn(pfn))
1478                 get_page(pfn_to_page(pfn));
1479 }
1480 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1481
1482 static int next_segment(unsigned long len, int offset)
1483 {
1484         if (len > PAGE_SIZE - offset)
1485                 return PAGE_SIZE - offset;
1486         else
1487                 return len;
1488 }
1489
1490 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1491                         int len)
1492 {
1493         int r;
1494         unsigned long addr;
1495
1496         addr = gfn_to_hva(kvm, gfn);
1497         if (kvm_is_error_hva(addr))
1498                 return -EFAULT;
1499         r = copy_from_user(data, (void __user *)addr + offset, len);
1500         if (r)
1501                 return -EFAULT;
1502         return 0;
1503 }
1504 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1505
1506 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1507 {
1508         gfn_t gfn = gpa >> PAGE_SHIFT;
1509         int seg;
1510         int offset = offset_in_page(gpa);
1511         int ret;
1512
1513         while ((seg = next_segment(len, offset)) != 0) {
1514                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1515                 if (ret < 0)
1516                         return ret;
1517                 offset = 0;
1518                 len -= seg;
1519                 data += seg;
1520                 ++gfn;
1521         }
1522         return 0;
1523 }
1524 EXPORT_SYMBOL_GPL(kvm_read_guest);
1525
1526 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1527                           unsigned long len)
1528 {
1529         int r;
1530         unsigned long addr;
1531         gfn_t gfn = gpa >> PAGE_SHIFT;
1532         int offset = offset_in_page(gpa);
1533
1534         addr = gfn_to_hva(kvm, gfn);
1535         if (kvm_is_error_hva(addr))
1536                 return -EFAULT;
1537         pagefault_disable();
1538         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1539         pagefault_enable();
1540         if (r)
1541                 return -EFAULT;
1542         return 0;
1543 }
1544 EXPORT_SYMBOL(kvm_read_guest_atomic);
1545
1546 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1547                          int offset, int len)
1548 {
1549         int r;
1550         unsigned long addr;
1551
1552         addr = gfn_to_hva(kvm, gfn);
1553         if (kvm_is_error_hva(addr))
1554                 return -EFAULT;
1555         r = copy_to_user((void __user *)addr + offset, data, len);
1556         if (r)
1557                 return -EFAULT;
1558         mark_page_dirty(kvm, gfn);
1559         return 0;
1560 }
1561 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1562
1563 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1564                     unsigned long len)
1565 {
1566         gfn_t gfn = gpa >> PAGE_SHIFT;
1567         int seg;
1568         int offset = offset_in_page(gpa);
1569         int ret;
1570
1571         while ((seg = next_segment(len, offset)) != 0) {
1572                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1573                 if (ret < 0)
1574                         return ret;
1575                 offset = 0;
1576                 len -= seg;
1577                 data += seg;
1578                 ++gfn;
1579         }
1580         return 0;
1581 }
1582
1583 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1584 {
1585         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1586 }
1587 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1588
1589 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1590 {
1591         gfn_t gfn = gpa >> PAGE_SHIFT;
1592         int seg;
1593         int offset = offset_in_page(gpa);
1594         int ret;
1595
1596         while ((seg = next_segment(len, offset)) != 0) {
1597                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1598                 if (ret < 0)
1599                         return ret;
1600                 offset = 0;
1601                 len -= seg;
1602                 ++gfn;
1603         }
1604         return 0;
1605 }
1606 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1607
1608 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1609 {
1610         struct kvm_memory_slot *memslot;
1611
1612         gfn = unalias_gfn(kvm, gfn);
1613         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1614         if (memslot && memslot->dirty_bitmap) {
1615                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1616
1617                 /* avoid RMW */
1618                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1619                         set_bit(rel_gfn, memslot->dirty_bitmap);
1620         }
1621 }
1622
1623 /*
1624  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1625  */
1626 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1627 {
1628         DEFINE_WAIT(wait);
1629
1630         for (;;) {
1631                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1632
1633                 if ((kvm_arch_interrupt_allowed(vcpu) &&
1634                                         kvm_cpu_has_interrupt(vcpu)) ||
1635                                 kvm_arch_vcpu_runnable(vcpu)) {
1636                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1637                         break;
1638                 }
1639                 if (kvm_cpu_has_pending_timer(vcpu))
1640                         break;
1641                 if (signal_pending(current))
1642                         break;
1643
1644                 vcpu_put(vcpu);
1645                 schedule();
1646                 vcpu_load(vcpu);
1647         }
1648
1649         finish_wait(&vcpu->wq, &wait);
1650 }
1651
1652 void kvm_resched(struct kvm_vcpu *vcpu)
1653 {
1654         if (!need_resched())
1655                 return;
1656         cond_resched();
1657 }
1658 EXPORT_SYMBOL_GPL(kvm_resched);
1659
1660 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1661 {
1662         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1663         struct page *page;
1664
1665         if (vmf->pgoff == 0)
1666                 page = virt_to_page(vcpu->run);
1667 #ifdef CONFIG_X86
1668         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1669                 page = virt_to_page(vcpu->arch.pio_data);
1670 #endif
1671 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1672         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1673                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1674 #endif
1675         else
1676                 return VM_FAULT_SIGBUS;
1677         get_page(page);
1678         vmf->page = page;
1679         return 0;
1680 }
1681
1682 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1683         .fault = kvm_vcpu_fault,
1684 };
1685
1686 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1687 {
1688         vma->vm_ops = &kvm_vcpu_vm_ops;
1689         return 0;
1690 }
1691
1692 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1693 {
1694         struct kvm_vcpu *vcpu = filp->private_data;
1695
1696         kvm_put_kvm(vcpu->kvm);
1697         return 0;
1698 }
1699
1700 static struct file_operations kvm_vcpu_fops = {
1701         .release        = kvm_vcpu_release,
1702         .unlocked_ioctl = kvm_vcpu_ioctl,
1703         .compat_ioctl   = kvm_vcpu_ioctl,
1704         .mmap           = kvm_vcpu_mmap,
1705 };
1706
1707 /*
1708  * Allocates an inode for the vcpu.
1709  */
1710 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1711 {
1712         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1713         if (fd < 0)
1714                 kvm_put_kvm(vcpu->kvm);
1715         return fd;
1716 }
1717
1718 /*
1719  * Creates some virtual cpus.  Good luck creating more than one.
1720  */
1721 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1722 {
1723         int r;
1724         struct kvm_vcpu *vcpu;
1725
1726         if (!valid_vcpu(n))
1727                 return -EINVAL;
1728
1729         vcpu = kvm_arch_vcpu_create(kvm, n);
1730         if (IS_ERR(vcpu))
1731                 return PTR_ERR(vcpu);
1732
1733         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1734
1735         r = kvm_arch_vcpu_setup(vcpu);
1736         if (r)
1737                 return r;
1738
1739         mutex_lock(&kvm->lock);
1740         if (kvm->vcpus[n]) {
1741                 r = -EEXIST;
1742                 goto vcpu_destroy;
1743         }
1744         kvm->vcpus[n] = vcpu;
1745         mutex_unlock(&kvm->lock);
1746
1747         /* Now it's all set up, let userspace reach it */
1748         kvm_get_kvm(kvm);
1749         r = create_vcpu_fd(vcpu);
1750         if (r < 0)
1751                 goto unlink;
1752         return r;
1753
1754 unlink:
1755         mutex_lock(&kvm->lock);
1756         kvm->vcpus[n] = NULL;
1757 vcpu_destroy:
1758         mutex_unlock(&kvm->lock);
1759         kvm_arch_vcpu_destroy(vcpu);
1760         return r;
1761 }
1762
1763 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1764 {
1765         if (sigset) {
1766                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1767                 vcpu->sigset_active = 1;
1768                 vcpu->sigset = *sigset;
1769         } else
1770                 vcpu->sigset_active = 0;
1771         return 0;
1772 }
1773
1774 #ifdef __KVM_HAVE_MSIX
1775 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1776                                     struct kvm_assigned_msix_nr *entry_nr)
1777 {
1778         int r = 0;
1779         struct kvm_assigned_dev_kernel *adev;
1780
1781         mutex_lock(&kvm->lock);
1782
1783         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1784                                       entry_nr->assigned_dev_id);
1785         if (!adev) {
1786                 r = -EINVAL;
1787                 goto msix_nr_out;
1788         }
1789
1790         if (adev->entries_nr == 0) {
1791                 adev->entries_nr = entry_nr->entry_nr;
1792                 if (adev->entries_nr == 0 ||
1793                     adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1794                         r = -EINVAL;
1795                         goto msix_nr_out;
1796                 }
1797
1798                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1799                                                 entry_nr->entry_nr,
1800                                                 GFP_KERNEL);
1801                 if (!adev->host_msix_entries) {
1802                         r = -ENOMEM;
1803                         goto msix_nr_out;
1804                 }
1805                 adev->guest_msix_entries = kzalloc(
1806                                 sizeof(struct kvm_guest_msix_entry) *
1807                                 entry_nr->entry_nr, GFP_KERNEL);
1808                 if (!adev->guest_msix_entries) {
1809                         kfree(adev->host_msix_entries);
1810                         r = -ENOMEM;
1811                         goto msix_nr_out;
1812                 }
1813         } else /* Not allowed set MSI-X number twice */
1814                 r = -EINVAL;
1815 msix_nr_out:
1816         mutex_unlock(&kvm->lock);
1817         return r;
1818 }
1819
1820 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1821                                        struct kvm_assigned_msix_entry *entry)
1822 {
1823         int r = 0, i;
1824         struct kvm_assigned_dev_kernel *adev;
1825
1826         mutex_lock(&kvm->lock);
1827
1828         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1829                                       entry->assigned_dev_id);
1830
1831         if (!adev) {
1832                 r = -EINVAL;
1833                 goto msix_entry_out;
1834         }
1835
1836         for (i = 0; i < adev->entries_nr; i++)
1837                 if (adev->guest_msix_entries[i].vector == 0 ||
1838                     adev->guest_msix_entries[i].entry == entry->entry) {
1839                         adev->guest_msix_entries[i].entry = entry->entry;
1840                         adev->guest_msix_entries[i].vector = entry->gsi;
1841                         adev->host_msix_entries[i].entry = entry->entry;
1842                         break;
1843                 }
1844         if (i == adev->entries_nr) {
1845                 r = -ENOSPC;
1846                 goto msix_entry_out;
1847         }
1848
1849 msix_entry_out:
1850         mutex_unlock(&kvm->lock);
1851
1852         return r;
1853 }
1854 #endif
1855
1856 static long kvm_vcpu_ioctl(struct file *filp,
1857                            unsigned int ioctl, unsigned long arg)
1858 {
1859         struct kvm_vcpu *vcpu = filp->private_data;
1860         void __user *argp = (void __user *)arg;
1861         int r;
1862         struct kvm_fpu *fpu = NULL;
1863         struct kvm_sregs *kvm_sregs = NULL;
1864
1865         if (vcpu->kvm->mm != current->mm)
1866                 return -EIO;
1867         switch (ioctl) {
1868         case KVM_RUN:
1869                 r = -EINVAL;
1870                 if (arg)
1871                         goto out;
1872                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1873                 break;
1874         case KVM_GET_REGS: {
1875                 struct kvm_regs *kvm_regs;
1876
1877                 r = -ENOMEM;
1878                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1879                 if (!kvm_regs)
1880                         goto out;
1881                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1882                 if (r)
1883                         goto out_free1;
1884                 r = -EFAULT;
1885                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1886                         goto out_free1;
1887                 r = 0;
1888 out_free1:
1889                 kfree(kvm_regs);
1890                 break;
1891         }
1892         case KVM_SET_REGS: {
1893                 struct kvm_regs *kvm_regs;
1894
1895                 r = -ENOMEM;
1896                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1897                 if (!kvm_regs)
1898                         goto out;
1899                 r = -EFAULT;
1900                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1901                         goto out_free2;
1902                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1903                 if (r)
1904                         goto out_free2;
1905                 r = 0;
1906 out_free2:
1907                 kfree(kvm_regs);
1908                 break;
1909         }
1910         case KVM_GET_SREGS: {
1911                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1912                 r = -ENOMEM;
1913                 if (!kvm_sregs)
1914                         goto out;
1915                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1916                 if (r)
1917                         goto out;
1918                 r = -EFAULT;
1919                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1920                         goto out;
1921                 r = 0;
1922                 break;
1923         }
1924         case KVM_SET_SREGS: {
1925                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1926                 r = -ENOMEM;
1927                 if (!kvm_sregs)
1928                         goto out;
1929                 r = -EFAULT;
1930                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1931                         goto out;
1932                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1933                 if (r)
1934                         goto out;
1935                 r = 0;
1936                 break;
1937         }
1938         case KVM_GET_MP_STATE: {
1939                 struct kvm_mp_state mp_state;
1940
1941                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1942                 if (r)
1943                         goto out;
1944                 r = -EFAULT;
1945                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1946                         goto out;
1947                 r = 0;
1948                 break;
1949         }
1950         case KVM_SET_MP_STATE: {
1951                 struct kvm_mp_state mp_state;
1952
1953                 r = -EFAULT;
1954                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1955                         goto out;
1956                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1957                 if (r)
1958                         goto out;
1959                 r = 0;
1960                 break;
1961         }
1962         case KVM_TRANSLATE: {
1963                 struct kvm_translation tr;
1964
1965                 r = -EFAULT;
1966                 if (copy_from_user(&tr, argp, sizeof tr))
1967                         goto out;
1968                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1969                 if (r)
1970                         goto out;
1971                 r = -EFAULT;
1972                 if (copy_to_user(argp, &tr, sizeof tr))
1973                         goto out;
1974                 r = 0;
1975                 break;
1976         }
1977         case KVM_SET_GUEST_DEBUG: {
1978                 struct kvm_guest_debug dbg;
1979
1980                 r = -EFAULT;
1981                 if (copy_from_user(&dbg, argp, sizeof dbg))
1982                         goto out;
1983                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1984                 if (r)
1985                         goto out;
1986                 r = 0;
1987                 break;
1988         }
1989         case KVM_SET_SIGNAL_MASK: {
1990                 struct kvm_signal_mask __user *sigmask_arg = argp;
1991                 struct kvm_signal_mask kvm_sigmask;
1992                 sigset_t sigset, *p;
1993
1994                 p = NULL;
1995                 if (argp) {
1996                         r = -EFAULT;
1997                         if (copy_from_user(&kvm_sigmask, argp,
1998                                            sizeof kvm_sigmask))
1999                                 goto out;
2000                         r = -EINVAL;
2001                         if (kvm_sigmask.len != sizeof sigset)
2002                                 goto out;
2003                         r = -EFAULT;
2004                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2005                                            sizeof sigset))
2006                                 goto out;
2007                         p = &sigset;
2008                 }
2009                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2010                 break;
2011         }
2012         case KVM_GET_FPU: {
2013                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2014                 r = -ENOMEM;
2015                 if (!fpu)
2016                         goto out;
2017                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2018                 if (r)
2019                         goto out;
2020                 r = -EFAULT;
2021                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2022                         goto out;
2023                 r = 0;
2024                 break;
2025         }
2026         case KVM_SET_FPU: {
2027                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2028                 r = -ENOMEM;
2029                 if (!fpu)
2030                         goto out;
2031                 r = -EFAULT;
2032                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2033                         goto out;
2034                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2035                 if (r)
2036                         goto out;
2037                 r = 0;
2038                 break;
2039         }
2040         default:
2041                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2042         }
2043 out:
2044         kfree(fpu);
2045         kfree(kvm_sregs);
2046         return r;
2047 }
2048
2049 static long kvm_vm_ioctl(struct file *filp,
2050                            unsigned int ioctl, unsigned long arg)
2051 {
2052         struct kvm *kvm = filp->private_data;
2053         void __user *argp = (void __user *)arg;
2054         int r;
2055
2056         if (kvm->mm != current->mm)
2057                 return -EIO;
2058         switch (ioctl) {
2059         case KVM_CREATE_VCPU:
2060                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2061                 if (r < 0)
2062                         goto out;
2063                 break;
2064         case KVM_SET_USER_MEMORY_REGION: {
2065                 struct kvm_userspace_memory_region kvm_userspace_mem;
2066
2067                 r = -EFAULT;
2068                 if (copy_from_user(&kvm_userspace_mem, argp,
2069                                                 sizeof kvm_userspace_mem))
2070                         goto out;
2071
2072                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2073                 if (r)
2074                         goto out;
2075                 break;
2076         }
2077         case KVM_GET_DIRTY_LOG: {
2078                 struct kvm_dirty_log log;
2079
2080                 r = -EFAULT;
2081                 if (copy_from_user(&log, argp, sizeof log))
2082                         goto out;
2083                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2084                 if (r)
2085                         goto out;
2086                 break;
2087         }
2088 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2089         case KVM_REGISTER_COALESCED_MMIO: {
2090                 struct kvm_coalesced_mmio_zone zone;
2091                 r = -EFAULT;
2092                 if (copy_from_user(&zone, argp, sizeof zone))
2093                         goto out;
2094                 r = -ENXIO;
2095                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2096                 if (r)
2097                         goto out;
2098                 r = 0;
2099                 break;
2100         }
2101         case KVM_UNREGISTER_COALESCED_MMIO: {
2102                 struct kvm_coalesced_mmio_zone zone;
2103                 r = -EFAULT;
2104                 if (copy_from_user(&zone, argp, sizeof zone))
2105                         goto out;
2106                 r = -ENXIO;
2107                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2108                 if (r)
2109                         goto out;
2110                 r = 0;
2111                 break;
2112         }
2113 #endif
2114 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2115         case KVM_ASSIGN_PCI_DEVICE: {
2116                 struct kvm_assigned_pci_dev assigned_dev;
2117
2118                 r = -EFAULT;
2119                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2120                         goto out;
2121                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2122                 if (r)
2123                         goto out;
2124                 break;
2125         }
2126         case KVM_ASSIGN_IRQ: {
2127                 r = -EOPNOTSUPP;
2128                 break;
2129         }
2130 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2131         case KVM_ASSIGN_DEV_IRQ: {
2132                 struct kvm_assigned_irq assigned_irq;
2133
2134                 r = -EFAULT;
2135                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2136                         goto out;
2137                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2138                 if (r)
2139                         goto out;
2140                 break;
2141         }
2142         case KVM_DEASSIGN_DEV_IRQ: {
2143                 struct kvm_assigned_irq assigned_irq;
2144
2145                 r = -EFAULT;
2146                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2147                         goto out;
2148                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2149                 if (r)
2150                         goto out;
2151                 break;
2152         }
2153 #endif
2154 #endif
2155 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2156         case KVM_DEASSIGN_PCI_DEVICE: {
2157                 struct kvm_assigned_pci_dev assigned_dev;
2158
2159                 r = -EFAULT;
2160                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2161                         goto out;
2162                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2163                 if (r)
2164                         goto out;
2165                 break;
2166         }
2167 #endif
2168 #ifdef KVM_CAP_IRQ_ROUTING
2169         case KVM_SET_GSI_ROUTING: {
2170                 struct kvm_irq_routing routing;
2171                 struct kvm_irq_routing __user *urouting;
2172                 struct kvm_irq_routing_entry *entries;
2173
2174                 r = -EFAULT;
2175                 if (copy_from_user(&routing, argp, sizeof(routing)))
2176                         goto out;
2177                 r = -EINVAL;
2178                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2179                         goto out;
2180                 if (routing.flags)
2181                         goto out;
2182                 r = -ENOMEM;
2183                 entries = vmalloc(routing.nr * sizeof(*entries));
2184                 if (!entries)
2185                         goto out;
2186                 r = -EFAULT;
2187                 urouting = argp;
2188                 if (copy_from_user(entries, urouting->entries,
2189                                    routing.nr * sizeof(*entries)))
2190                         goto out_free_irq_routing;
2191                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2192                                         routing.flags);
2193         out_free_irq_routing:
2194                 vfree(entries);
2195                 break;
2196         }
2197 #ifdef __KVM_HAVE_MSIX
2198         case KVM_ASSIGN_SET_MSIX_NR: {
2199                 struct kvm_assigned_msix_nr entry_nr;
2200                 r = -EFAULT;
2201                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2202                         goto out;
2203                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2204                 if (r)
2205                         goto out;
2206                 break;
2207         }
2208         case KVM_ASSIGN_SET_MSIX_ENTRY: {
2209                 struct kvm_assigned_msix_entry entry;
2210                 r = -EFAULT;
2211                 if (copy_from_user(&entry, argp, sizeof entry))
2212                         goto out;
2213                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2214                 if (r)
2215                         goto out;
2216                 break;
2217         }
2218 #endif
2219 #endif /* KVM_CAP_IRQ_ROUTING */
2220         default:
2221                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2222         }
2223 out:
2224         return r;
2225 }
2226
2227 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2228 {
2229         struct page *page[1];
2230         unsigned long addr;
2231         int npages;
2232         gfn_t gfn = vmf->pgoff;
2233         struct kvm *kvm = vma->vm_file->private_data;
2234
2235         addr = gfn_to_hva(kvm, gfn);
2236         if (kvm_is_error_hva(addr))
2237                 return VM_FAULT_SIGBUS;
2238
2239         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2240                                 NULL);
2241         if (unlikely(npages != 1))
2242                 return VM_FAULT_SIGBUS;
2243
2244         vmf->page = page[0];
2245         return 0;
2246 }
2247
2248 static struct vm_operations_struct kvm_vm_vm_ops = {
2249         .fault = kvm_vm_fault,
2250 };
2251
2252 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2253 {
2254         vma->vm_ops = &kvm_vm_vm_ops;
2255         return 0;
2256 }
2257
2258 static struct file_operations kvm_vm_fops = {
2259         .release        = kvm_vm_release,
2260         .unlocked_ioctl = kvm_vm_ioctl,
2261         .compat_ioctl   = kvm_vm_ioctl,
2262         .mmap           = kvm_vm_mmap,
2263 };
2264
2265 static int kvm_dev_ioctl_create_vm(void)
2266 {
2267         int fd;
2268         struct kvm *kvm;
2269
2270         kvm = kvm_create_vm();
2271         if (IS_ERR(kvm))
2272                 return PTR_ERR(kvm);
2273         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2274         if (fd < 0)
2275                 kvm_put_kvm(kvm);
2276
2277         return fd;
2278 }
2279
2280 static long kvm_dev_ioctl_check_extension_generic(long arg)
2281 {
2282         switch (arg) {
2283         case KVM_CAP_USER_MEMORY:
2284         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2285         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2286                 return 1;
2287 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2288         case KVM_CAP_IRQ_ROUTING:
2289                 return KVM_MAX_IRQ_ROUTES;
2290 #endif
2291         default:
2292                 break;
2293         }
2294         return kvm_dev_ioctl_check_extension(arg);
2295 }
2296
2297 static long kvm_dev_ioctl(struct file *filp,
2298                           unsigned int ioctl, unsigned long arg)
2299 {
2300         long r = -EINVAL;
2301
2302         switch (ioctl) {
2303         case KVM_GET_API_VERSION:
2304                 r = -EINVAL;
2305                 if (arg)
2306                         goto out;
2307                 r = KVM_API_VERSION;
2308                 break;
2309         case KVM_CREATE_VM:
2310                 r = -EINVAL;
2311                 if (arg)
2312                         goto out;
2313                 r = kvm_dev_ioctl_create_vm();
2314                 break;
2315         case KVM_CHECK_EXTENSION:
2316                 r = kvm_dev_ioctl_check_extension_generic(arg);
2317                 break;
2318         case KVM_GET_VCPU_MMAP_SIZE:
2319                 r = -EINVAL;
2320                 if (arg)
2321                         goto out;
2322                 r = PAGE_SIZE;     /* struct kvm_run */
2323 #ifdef CONFIG_X86
2324                 r += PAGE_SIZE;    /* pio data page */
2325 #endif
2326 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2327                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2328 #endif
2329                 break;
2330         case KVM_TRACE_ENABLE:
2331         case KVM_TRACE_PAUSE:
2332         case KVM_TRACE_DISABLE:
2333                 r = kvm_trace_ioctl(ioctl, arg);
2334                 break;
2335         default:
2336                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2337         }
2338 out:
2339         return r;
2340 }
2341
2342 static struct file_operations kvm_chardev_ops = {
2343         .unlocked_ioctl = kvm_dev_ioctl,
2344         .compat_ioctl   = kvm_dev_ioctl,
2345 };
2346
2347 static struct miscdevice kvm_dev = {
2348         KVM_MINOR,
2349         "kvm",
2350         &kvm_chardev_ops,
2351 };
2352
2353 static void hardware_enable(void *junk)
2354 {
2355         int cpu = raw_smp_processor_id();
2356
2357         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2358                 return;
2359         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2360         kvm_arch_hardware_enable(NULL);
2361 }
2362
2363 static void hardware_disable(void *junk)
2364 {
2365         int cpu = raw_smp_processor_id();
2366
2367         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2368                 return;
2369         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2370         kvm_arch_hardware_disable(NULL);
2371 }
2372
2373 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2374                            void *v)
2375 {
2376         int cpu = (long)v;
2377
2378         val &= ~CPU_TASKS_FROZEN;
2379         switch (val) {
2380         case CPU_DYING:
2381                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2382                        cpu);
2383                 hardware_disable(NULL);
2384                 break;
2385         case CPU_UP_CANCELED:
2386                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2387                        cpu);
2388                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2389                 break;
2390         case CPU_ONLINE:
2391                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2392                        cpu);
2393                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2394                 break;
2395         }
2396         return NOTIFY_OK;
2397 }
2398
2399
2400 asmlinkage void kvm_handle_fault_on_reboot(void)
2401 {
2402         if (kvm_rebooting)
2403                 /* spin while reset goes on */
2404                 while (true)
2405                         ;
2406         /* Fault while not rebooting.  We want the trace. */
2407         BUG();
2408 }
2409 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2410
2411 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2412                       void *v)
2413 {
2414         /*
2415          * Some (well, at least mine) BIOSes hang on reboot if
2416          * in vmx root mode.
2417          *
2418          * And Intel TXT required VMX off for all cpu when system shutdown.
2419          */
2420         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2421         kvm_rebooting = true;
2422         on_each_cpu(hardware_disable, NULL, 1);
2423         return NOTIFY_OK;
2424 }
2425
2426 static struct notifier_block kvm_reboot_notifier = {
2427         .notifier_call = kvm_reboot,
2428         .priority = 0,
2429 };
2430
2431 void kvm_io_bus_init(struct kvm_io_bus *bus)
2432 {
2433         memset(bus, 0, sizeof(*bus));
2434 }
2435
2436 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2437 {
2438         int i;
2439
2440         for (i = 0; i < bus->dev_count; i++) {
2441                 struct kvm_io_device *pos = bus->devs[i];
2442
2443                 kvm_iodevice_destructor(pos);
2444         }
2445 }
2446
2447 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2448                                           gpa_t addr, int len, int is_write)
2449 {
2450         int i;
2451
2452         for (i = 0; i < bus->dev_count; i++) {
2453                 struct kvm_io_device *pos = bus->devs[i];
2454
2455                 if (pos->in_range(pos, addr, len, is_write))
2456                         return pos;
2457         }
2458
2459         return NULL;
2460 }
2461
2462 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2463 {
2464         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2465
2466         bus->devs[bus->dev_count++] = dev;
2467 }
2468
2469 static struct notifier_block kvm_cpu_notifier = {
2470         .notifier_call = kvm_cpu_hotplug,
2471         .priority = 20, /* must be > scheduler priority */
2472 };
2473
2474 static int vm_stat_get(void *_offset, u64 *val)
2475 {
2476         unsigned offset = (long)_offset;
2477         struct kvm *kvm;
2478
2479         *val = 0;
2480         spin_lock(&kvm_lock);
2481         list_for_each_entry(kvm, &vm_list, vm_list)
2482                 *val += *(u32 *)((void *)kvm + offset);
2483         spin_unlock(&kvm_lock);
2484         return 0;
2485 }
2486
2487 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2488
2489 static int vcpu_stat_get(void *_offset, u64 *val)
2490 {
2491         unsigned offset = (long)_offset;
2492         struct kvm *kvm;
2493         struct kvm_vcpu *vcpu;
2494         int i;
2495
2496         *val = 0;
2497         spin_lock(&kvm_lock);
2498         list_for_each_entry(kvm, &vm_list, vm_list)
2499                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2500                         vcpu = kvm->vcpus[i];
2501                         if (vcpu)
2502                                 *val += *(u32 *)((void *)vcpu + offset);
2503                 }
2504         spin_unlock(&kvm_lock);
2505         return 0;
2506 }
2507
2508 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2509
2510 static struct file_operations *stat_fops[] = {
2511         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2512         [KVM_STAT_VM]   = &vm_stat_fops,
2513 };
2514
2515 static void kvm_init_debug(void)
2516 {
2517         struct kvm_stats_debugfs_item *p;
2518
2519         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2520         for (p = debugfs_entries; p->name; ++p)
2521                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2522                                                 (void *)(long)p->offset,
2523                                                 stat_fops[p->kind]);
2524 }
2525
2526 static void kvm_exit_debug(void)
2527 {
2528         struct kvm_stats_debugfs_item *p;
2529
2530         for (p = debugfs_entries; p->name; ++p)
2531                 debugfs_remove(p->dentry);
2532         debugfs_remove(kvm_debugfs_dir);
2533 }
2534
2535 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2536 {
2537         hardware_disable(NULL);
2538         return 0;
2539 }
2540
2541 static int kvm_resume(struct sys_device *dev)
2542 {
2543         hardware_enable(NULL);
2544         return 0;
2545 }
2546
2547 static struct sysdev_class kvm_sysdev_class = {
2548         .name = "kvm",
2549         .suspend = kvm_suspend,
2550         .resume = kvm_resume,
2551 };
2552
2553 static struct sys_device kvm_sysdev = {
2554         .id = 0,
2555         .cls = &kvm_sysdev_class,
2556 };
2557
2558 struct page *bad_page;
2559 pfn_t bad_pfn;
2560
2561 static inline
2562 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2563 {
2564         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2565 }
2566
2567 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2568 {
2569         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2570
2571         kvm_arch_vcpu_load(vcpu, cpu);
2572 }
2573
2574 static void kvm_sched_out(struct preempt_notifier *pn,
2575                           struct task_struct *next)
2576 {
2577         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2578
2579         kvm_arch_vcpu_put(vcpu);
2580 }
2581
2582 int kvm_init(void *opaque, unsigned int vcpu_size,
2583                   struct module *module)
2584 {
2585         int r;
2586         int cpu;
2587
2588         kvm_init_debug();
2589
2590         r = kvm_arch_init(opaque);
2591         if (r)
2592                 goto out_fail;
2593
2594         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2595
2596         if (bad_page == NULL) {
2597                 r = -ENOMEM;
2598                 goto out;
2599         }
2600
2601         bad_pfn = page_to_pfn(bad_page);
2602
2603         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2604                 r = -ENOMEM;
2605                 goto out_free_0;
2606         }
2607
2608         r = kvm_arch_hardware_setup();
2609         if (r < 0)
2610                 goto out_free_0a;
2611
2612         for_each_online_cpu(cpu) {
2613                 smp_call_function_single(cpu,
2614                                 kvm_arch_check_processor_compat,
2615                                 &r, 1);
2616                 if (r < 0)
2617                         goto out_free_1;
2618         }
2619
2620         on_each_cpu(hardware_enable, NULL, 1);
2621         r = register_cpu_notifier(&kvm_cpu_notifier);
2622         if (r)
2623                 goto out_free_2;
2624         register_reboot_notifier(&kvm_reboot_notifier);
2625
2626         r = sysdev_class_register(&kvm_sysdev_class);
2627         if (r)
2628                 goto out_free_3;
2629
2630         r = sysdev_register(&kvm_sysdev);
2631         if (r)
2632                 goto out_free_4;
2633
2634         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2635         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2636                                            __alignof__(struct kvm_vcpu),
2637                                            0, NULL);
2638         if (!kvm_vcpu_cache) {
2639                 r = -ENOMEM;
2640                 goto out_free_5;
2641         }
2642
2643         kvm_chardev_ops.owner = module;
2644         kvm_vm_fops.owner = module;
2645         kvm_vcpu_fops.owner = module;
2646
2647         r = misc_register(&kvm_dev);
2648         if (r) {
2649                 printk(KERN_ERR "kvm: misc device register failed\n");
2650                 goto out_free;
2651         }
2652
2653         kvm_preempt_ops.sched_in = kvm_sched_in;
2654         kvm_preempt_ops.sched_out = kvm_sched_out;
2655
2656         return 0;
2657
2658 out_free:
2659         kmem_cache_destroy(kvm_vcpu_cache);
2660 out_free_5:
2661         sysdev_unregister(&kvm_sysdev);
2662 out_free_4:
2663         sysdev_class_unregister(&kvm_sysdev_class);
2664 out_free_3:
2665         unregister_reboot_notifier(&kvm_reboot_notifier);
2666         unregister_cpu_notifier(&kvm_cpu_notifier);
2667 out_free_2:
2668         on_each_cpu(hardware_disable, NULL, 1);
2669 out_free_1:
2670         kvm_arch_hardware_unsetup();
2671 out_free_0a:
2672         free_cpumask_var(cpus_hardware_enabled);
2673 out_free_0:
2674         __free_page(bad_page);
2675 out:
2676         kvm_arch_exit();
2677         kvm_exit_debug();
2678 out_fail:
2679         return r;
2680 }
2681 EXPORT_SYMBOL_GPL(kvm_init);
2682
2683 void kvm_exit(void)
2684 {
2685         kvm_trace_cleanup();
2686         misc_deregister(&kvm_dev);
2687         kmem_cache_destroy(kvm_vcpu_cache);
2688         sysdev_unregister(&kvm_sysdev);
2689         sysdev_class_unregister(&kvm_sysdev_class);
2690         unregister_reboot_notifier(&kvm_reboot_notifier);
2691         unregister_cpu_notifier(&kvm_cpu_notifier);
2692         on_each_cpu(hardware_disable, NULL, 1);
2693         kvm_arch_hardware_unsetup();
2694         kvm_arch_exit();
2695         kvm_exit_debug();
2696         free_cpumask_var(cpus_hardware_enabled);
2697         __free_page(bad_page);
2698 }
2699 EXPORT_SYMBOL_GPL(kvm_exit);