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;
1090         int largepages;
1091         unsigned long i;
1092         struct kvm_memory_slot *memslot;
1093         struct kvm_memory_slot old, new;
1094
1095         r = -EINVAL;
1096         /* General sanity checks */
1097         if (mem->memory_size & (PAGE_SIZE - 1))
1098                 goto out;
1099         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1100                 goto out;
1101         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
1102                 goto out;
1103         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
1104                 goto out;
1105         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1106                 goto out;
1107
1108         memslot = &kvm->memslots[mem->slot];
1109         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1110         npages = mem->memory_size >> PAGE_SHIFT;
1111
1112         if (!npages)
1113                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1114
1115         new = old = *memslot;
1116
1117         new.base_gfn = base_gfn;
1118         new.npages = npages;
1119         new.flags = mem->flags;
1120
1121         /* Disallow changing a memory slot's size. */
1122         r = -EINVAL;
1123         if (npages && old.npages && npages != old.npages)
1124                 goto out_free;
1125
1126         /* Check for overlaps */
1127         r = -EEXIST;
1128         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1129                 struct kvm_memory_slot *s = &kvm->memslots[i];
1130
1131                 if (s == memslot || !s->npages)
1132                         continue;
1133                 if (!((base_gfn + npages <= s->base_gfn) ||
1134                       (base_gfn >= s->base_gfn + s->npages)))
1135                         goto out_free;
1136         }
1137
1138         /* Free page dirty bitmap if unneeded */
1139         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1140                 new.dirty_bitmap = NULL;
1141
1142         r = -ENOMEM;
1143
1144         /* Allocate if a slot is being created */
1145 #ifndef CONFIG_S390
1146         if (npages && !new.rmap) {
1147                 new.rmap = vmalloc(npages * sizeof(struct page *));
1148
1149                 if (!new.rmap)
1150                         goto out_free;
1151
1152                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1153
1154                 new.user_alloc = user_alloc;
1155                 /*
1156                  * hva_to_rmmap() serialzies with the mmu_lock and to be
1157                  * safe it has to ignore memslots with !user_alloc &&
1158                  * !userspace_addr.
1159                  */
1160                 if (user_alloc)
1161                         new.userspace_addr = mem->userspace_addr;
1162                 else
1163                         new.userspace_addr = 0;
1164         }
1165         if (npages && !new.lpage_info) {
1166                 largepages = 1 + (base_gfn + npages - 1) / KVM_PAGES_PER_HPAGE;
1167                 largepages -= base_gfn / KVM_PAGES_PER_HPAGE;
1168
1169                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1170
1171                 if (!new.lpage_info)
1172                         goto out_free;
1173
1174                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1175
1176                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1177                         new.lpage_info[0].write_count = 1;
1178                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1179                         new.lpage_info[largepages-1].write_count = 1;
1180         }
1181
1182         /* Allocate page dirty bitmap if needed */
1183         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1184                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1185
1186                 new.dirty_bitmap = vmalloc(dirty_bytes);
1187                 if (!new.dirty_bitmap)
1188                         goto out_free;
1189                 memset(new.dirty_bitmap, 0, dirty_bytes);
1190         }
1191 #endif /* not defined CONFIG_S390 */
1192
1193         if (!npages)
1194                 kvm_arch_flush_shadow(kvm);
1195
1196         spin_lock(&kvm->mmu_lock);
1197         if (mem->slot >= kvm->nmemslots)
1198                 kvm->nmemslots = mem->slot + 1;
1199
1200         *memslot = new;
1201         spin_unlock(&kvm->mmu_lock);
1202
1203         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1204         if (r) {
1205                 spin_lock(&kvm->mmu_lock);
1206                 *memslot = old;
1207                 spin_unlock(&kvm->mmu_lock);
1208                 goto out_free;
1209         }
1210
1211         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1212         /* Slot deletion case: we have to update the current slot */
1213         if (!npages)
1214                 *memslot = old;
1215 #ifdef CONFIG_DMAR
1216         /* map the pages in iommu page table */
1217         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1218         if (r)
1219                 goto out;
1220 #endif
1221         return 0;
1222
1223 out_free:
1224         kvm_free_physmem_slot(&new, &old);
1225 out:
1226         return r;
1227
1228 }
1229 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1230
1231 int kvm_set_memory_region(struct kvm *kvm,
1232                           struct kvm_userspace_memory_region *mem,
1233                           int user_alloc)
1234 {
1235         int r;
1236
1237         down_write(&kvm->slots_lock);
1238         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1239         up_write(&kvm->slots_lock);
1240         return r;
1241 }
1242 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1243
1244 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1245                                    struct
1246                                    kvm_userspace_memory_region *mem,
1247                                    int user_alloc)
1248 {
1249         if (mem->slot >= KVM_MEMORY_SLOTS)
1250                 return -EINVAL;
1251         return kvm_set_memory_region(kvm, mem, user_alloc);
1252 }
1253
1254 int kvm_get_dirty_log(struct kvm *kvm,
1255                         struct kvm_dirty_log *log, int *is_dirty)
1256 {
1257         struct kvm_memory_slot *memslot;
1258         int r, i;
1259         int n;
1260         unsigned long any = 0;
1261
1262         r = -EINVAL;
1263         if (log->slot >= KVM_MEMORY_SLOTS)
1264                 goto out;
1265
1266         memslot = &kvm->memslots[log->slot];
1267         r = -ENOENT;
1268         if (!memslot->dirty_bitmap)
1269                 goto out;
1270
1271         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1272
1273         for (i = 0; !any && i < n/sizeof(long); ++i)
1274                 any = memslot->dirty_bitmap[i];
1275
1276         r = -EFAULT;
1277         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1278                 goto out;
1279
1280         if (any)
1281                 *is_dirty = 1;
1282
1283         r = 0;
1284 out:
1285         return r;
1286 }
1287
1288 int is_error_page(struct page *page)
1289 {
1290         return page == bad_page;
1291 }
1292 EXPORT_SYMBOL_GPL(is_error_page);
1293
1294 int is_error_pfn(pfn_t pfn)
1295 {
1296         return pfn == bad_pfn;
1297 }
1298 EXPORT_SYMBOL_GPL(is_error_pfn);
1299
1300 static inline unsigned long bad_hva(void)
1301 {
1302         return PAGE_OFFSET;
1303 }
1304
1305 int kvm_is_error_hva(unsigned long addr)
1306 {
1307         return addr == bad_hva();
1308 }
1309 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1310
1311 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1312 {
1313         int i;
1314
1315         for (i = 0; i < kvm->nmemslots; ++i) {
1316                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1317
1318                 if (gfn >= memslot->base_gfn
1319                     && gfn < memslot->base_gfn + memslot->npages)
1320                         return memslot;
1321         }
1322         return NULL;
1323 }
1324 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1325
1326 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1327 {
1328         gfn = unalias_gfn(kvm, gfn);
1329         return gfn_to_memslot_unaliased(kvm, gfn);
1330 }
1331
1332 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1333 {
1334         int i;
1335
1336         gfn = unalias_gfn(kvm, gfn);
1337         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1338                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1339
1340                 if (gfn >= memslot->base_gfn
1341                     && gfn < memslot->base_gfn + memslot->npages)
1342                         return 1;
1343         }
1344         return 0;
1345 }
1346 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1347
1348 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1349 {
1350         struct kvm_memory_slot *slot;
1351
1352         gfn = unalias_gfn(kvm, gfn);
1353         slot = gfn_to_memslot_unaliased(kvm, gfn);
1354         if (!slot)
1355                 return bad_hva();
1356         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1357 }
1358 EXPORT_SYMBOL_GPL(gfn_to_hva);
1359
1360 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1361 {
1362         struct page *page[1];
1363         unsigned long addr;
1364         int npages;
1365         pfn_t pfn;
1366
1367         might_sleep();
1368
1369         addr = gfn_to_hva(kvm, gfn);
1370         if (kvm_is_error_hva(addr)) {
1371                 get_page(bad_page);
1372                 return page_to_pfn(bad_page);
1373         }
1374
1375         npages = get_user_pages_fast(addr, 1, 1, page);
1376
1377         if (unlikely(npages != 1)) {
1378                 struct vm_area_struct *vma;
1379
1380                 down_read(&current->mm->mmap_sem);
1381                 vma = find_vma(current->mm, addr);
1382
1383                 if (vma == NULL || addr < vma->vm_start ||
1384                     !(vma->vm_flags & VM_PFNMAP)) {
1385                         up_read(&current->mm->mmap_sem);
1386                         get_page(bad_page);
1387                         return page_to_pfn(bad_page);
1388                 }
1389
1390                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1391                 up_read(&current->mm->mmap_sem);
1392                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1393         } else
1394                 pfn = page_to_pfn(page[0]);
1395
1396         return pfn;
1397 }
1398
1399 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1400
1401 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1402 {
1403         pfn_t pfn;
1404
1405         pfn = gfn_to_pfn(kvm, gfn);
1406         if (!kvm_is_mmio_pfn(pfn))
1407                 return pfn_to_page(pfn);
1408
1409         WARN_ON(kvm_is_mmio_pfn(pfn));
1410
1411         get_page(bad_page);
1412         return bad_page;
1413 }
1414
1415 EXPORT_SYMBOL_GPL(gfn_to_page);
1416
1417 void kvm_release_page_clean(struct page *page)
1418 {
1419         kvm_release_pfn_clean(page_to_pfn(page));
1420 }
1421 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1422
1423 void kvm_release_pfn_clean(pfn_t pfn)
1424 {
1425         if (!kvm_is_mmio_pfn(pfn))
1426                 put_page(pfn_to_page(pfn));
1427 }
1428 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1429
1430 void kvm_release_page_dirty(struct page *page)
1431 {
1432         kvm_release_pfn_dirty(page_to_pfn(page));
1433 }
1434 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1435
1436 void kvm_release_pfn_dirty(pfn_t pfn)
1437 {
1438         kvm_set_pfn_dirty(pfn);
1439         kvm_release_pfn_clean(pfn);
1440 }
1441 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1442
1443 void kvm_set_page_dirty(struct page *page)
1444 {
1445         kvm_set_pfn_dirty(page_to_pfn(page));
1446 }
1447 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1448
1449 void kvm_set_pfn_dirty(pfn_t pfn)
1450 {
1451         if (!kvm_is_mmio_pfn(pfn)) {
1452                 struct page *page = pfn_to_page(pfn);
1453                 if (!PageReserved(page))
1454                         SetPageDirty(page);
1455         }
1456 }
1457 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1458
1459 void kvm_set_pfn_accessed(pfn_t pfn)
1460 {
1461         if (!kvm_is_mmio_pfn(pfn))
1462                 mark_page_accessed(pfn_to_page(pfn));
1463 }
1464 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1465
1466 void kvm_get_pfn(pfn_t pfn)
1467 {
1468         if (!kvm_is_mmio_pfn(pfn))
1469                 get_page(pfn_to_page(pfn));
1470 }
1471 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1472
1473 static int next_segment(unsigned long len, int offset)
1474 {
1475         if (len > PAGE_SIZE - offset)
1476                 return PAGE_SIZE - offset;
1477         else
1478                 return len;
1479 }
1480
1481 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1482                         int len)
1483 {
1484         int r;
1485         unsigned long addr;
1486
1487         addr = gfn_to_hva(kvm, gfn);
1488         if (kvm_is_error_hva(addr))
1489                 return -EFAULT;
1490         r = copy_from_user(data, (void __user *)addr + offset, len);
1491         if (r)
1492                 return -EFAULT;
1493         return 0;
1494 }
1495 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1496
1497 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1498 {
1499         gfn_t gfn = gpa >> PAGE_SHIFT;
1500         int seg;
1501         int offset = offset_in_page(gpa);
1502         int ret;
1503
1504         while ((seg = next_segment(len, offset)) != 0) {
1505                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1506                 if (ret < 0)
1507                         return ret;
1508                 offset = 0;
1509                 len -= seg;
1510                 data += seg;
1511                 ++gfn;
1512         }
1513         return 0;
1514 }
1515 EXPORT_SYMBOL_GPL(kvm_read_guest);
1516
1517 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1518                           unsigned long len)
1519 {
1520         int r;
1521         unsigned long addr;
1522         gfn_t gfn = gpa >> PAGE_SHIFT;
1523         int offset = offset_in_page(gpa);
1524
1525         addr = gfn_to_hva(kvm, gfn);
1526         if (kvm_is_error_hva(addr))
1527                 return -EFAULT;
1528         pagefault_disable();
1529         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1530         pagefault_enable();
1531         if (r)
1532                 return -EFAULT;
1533         return 0;
1534 }
1535 EXPORT_SYMBOL(kvm_read_guest_atomic);
1536
1537 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1538                          int offset, int len)
1539 {
1540         int r;
1541         unsigned long addr;
1542
1543         addr = gfn_to_hva(kvm, gfn);
1544         if (kvm_is_error_hva(addr))
1545                 return -EFAULT;
1546         r = copy_to_user((void __user *)addr + offset, data, len);
1547         if (r)
1548                 return -EFAULT;
1549         mark_page_dirty(kvm, gfn);
1550         return 0;
1551 }
1552 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1553
1554 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1555                     unsigned long len)
1556 {
1557         gfn_t gfn = gpa >> PAGE_SHIFT;
1558         int seg;
1559         int offset = offset_in_page(gpa);
1560         int ret;
1561
1562         while ((seg = next_segment(len, offset)) != 0) {
1563                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1564                 if (ret < 0)
1565                         return ret;
1566                 offset = 0;
1567                 len -= seg;
1568                 data += seg;
1569                 ++gfn;
1570         }
1571         return 0;
1572 }
1573
1574 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1575 {
1576         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1577 }
1578 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1579
1580 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1581 {
1582         gfn_t gfn = gpa >> PAGE_SHIFT;
1583         int seg;
1584         int offset = offset_in_page(gpa);
1585         int ret;
1586
1587         while ((seg = next_segment(len, offset)) != 0) {
1588                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1589                 if (ret < 0)
1590                         return ret;
1591                 offset = 0;
1592                 len -= seg;
1593                 ++gfn;
1594         }
1595         return 0;
1596 }
1597 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1598
1599 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1600 {
1601         struct kvm_memory_slot *memslot;
1602
1603         gfn = unalias_gfn(kvm, gfn);
1604         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1605         if (memslot && memslot->dirty_bitmap) {
1606                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1607
1608                 /* avoid RMW */
1609                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1610                         set_bit(rel_gfn, memslot->dirty_bitmap);
1611         }
1612 }
1613
1614 /*
1615  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1616  */
1617 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1618 {
1619         DEFINE_WAIT(wait);
1620
1621         for (;;) {
1622                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1623
1624                 if ((kvm_arch_interrupt_allowed(vcpu) &&
1625                                         kvm_cpu_has_interrupt(vcpu)) ||
1626                                 kvm_arch_vcpu_runnable(vcpu)) {
1627                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1628                         break;
1629                 }
1630                 if (kvm_cpu_has_pending_timer(vcpu))
1631                         break;
1632                 if (signal_pending(current))
1633                         break;
1634
1635                 vcpu_put(vcpu);
1636                 schedule();
1637                 vcpu_load(vcpu);
1638         }
1639
1640         finish_wait(&vcpu->wq, &wait);
1641 }
1642
1643 void kvm_resched(struct kvm_vcpu *vcpu)
1644 {
1645         if (!need_resched())
1646                 return;
1647         cond_resched();
1648 }
1649 EXPORT_SYMBOL_GPL(kvm_resched);
1650
1651 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1652 {
1653         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1654         struct page *page;
1655
1656         if (vmf->pgoff == 0)
1657                 page = virt_to_page(vcpu->run);
1658 #ifdef CONFIG_X86
1659         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1660                 page = virt_to_page(vcpu->arch.pio_data);
1661 #endif
1662 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1663         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1664                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1665 #endif
1666         else
1667                 return VM_FAULT_SIGBUS;
1668         get_page(page);
1669         vmf->page = page;
1670         return 0;
1671 }
1672
1673 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1674         .fault = kvm_vcpu_fault,
1675 };
1676
1677 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1678 {
1679         vma->vm_ops = &kvm_vcpu_vm_ops;
1680         return 0;
1681 }
1682
1683 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1684 {
1685         struct kvm_vcpu *vcpu = filp->private_data;
1686
1687         kvm_put_kvm(vcpu->kvm);
1688         return 0;
1689 }
1690
1691 static struct file_operations kvm_vcpu_fops = {
1692         .release        = kvm_vcpu_release,
1693         .unlocked_ioctl = kvm_vcpu_ioctl,
1694         .compat_ioctl   = kvm_vcpu_ioctl,
1695         .mmap           = kvm_vcpu_mmap,
1696 };
1697
1698 /*
1699  * Allocates an inode for the vcpu.
1700  */
1701 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1702 {
1703         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1704         if (fd < 0)
1705                 kvm_put_kvm(vcpu->kvm);
1706         return fd;
1707 }
1708
1709 /*
1710  * Creates some virtual cpus.  Good luck creating more than one.
1711  */
1712 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1713 {
1714         int r;
1715         struct kvm_vcpu *vcpu;
1716
1717         if (!valid_vcpu(n))
1718                 return -EINVAL;
1719
1720         vcpu = kvm_arch_vcpu_create(kvm, n);
1721         if (IS_ERR(vcpu))
1722                 return PTR_ERR(vcpu);
1723
1724         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1725
1726         r = kvm_arch_vcpu_setup(vcpu);
1727         if (r)
1728                 return r;
1729
1730         mutex_lock(&kvm->lock);
1731         if (kvm->vcpus[n]) {
1732                 r = -EEXIST;
1733                 goto vcpu_destroy;
1734         }
1735         kvm->vcpus[n] = vcpu;
1736         mutex_unlock(&kvm->lock);
1737
1738         /* Now it's all set up, let userspace reach it */
1739         kvm_get_kvm(kvm);
1740         r = create_vcpu_fd(vcpu);
1741         if (r < 0)
1742                 goto unlink;
1743         return r;
1744
1745 unlink:
1746         mutex_lock(&kvm->lock);
1747         kvm->vcpus[n] = NULL;
1748 vcpu_destroy:
1749         mutex_unlock(&kvm->lock);
1750         kvm_arch_vcpu_destroy(vcpu);
1751         return r;
1752 }
1753
1754 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1755 {
1756         if (sigset) {
1757                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1758                 vcpu->sigset_active = 1;
1759                 vcpu->sigset = *sigset;
1760         } else
1761                 vcpu->sigset_active = 0;
1762         return 0;
1763 }
1764
1765 #ifdef __KVM_HAVE_MSIX
1766 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1767                                     struct kvm_assigned_msix_nr *entry_nr)
1768 {
1769         int r = 0;
1770         struct kvm_assigned_dev_kernel *adev;
1771
1772         mutex_lock(&kvm->lock);
1773
1774         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1775                                       entry_nr->assigned_dev_id);
1776         if (!adev) {
1777                 r = -EINVAL;
1778                 goto msix_nr_out;
1779         }
1780
1781         if (adev->entries_nr == 0) {
1782                 adev->entries_nr = entry_nr->entry_nr;
1783                 if (adev->entries_nr == 0 ||
1784                     adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1785                         r = -EINVAL;
1786                         goto msix_nr_out;
1787                 }
1788
1789                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1790                                                 entry_nr->entry_nr,
1791                                                 GFP_KERNEL);
1792                 if (!adev->host_msix_entries) {
1793                         r = -ENOMEM;
1794                         goto msix_nr_out;
1795                 }
1796                 adev->guest_msix_entries = kzalloc(
1797                                 sizeof(struct kvm_guest_msix_entry) *
1798                                 entry_nr->entry_nr, GFP_KERNEL);
1799                 if (!adev->guest_msix_entries) {
1800                         kfree(adev->host_msix_entries);
1801                         r = -ENOMEM;
1802                         goto msix_nr_out;
1803                 }
1804         } else /* Not allowed set MSI-X number twice */
1805                 r = -EINVAL;
1806 msix_nr_out:
1807         mutex_unlock(&kvm->lock);
1808         return r;
1809 }
1810
1811 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1812                                        struct kvm_assigned_msix_entry *entry)
1813 {
1814         int r = 0, i;
1815         struct kvm_assigned_dev_kernel *adev;
1816
1817         mutex_lock(&kvm->lock);
1818
1819         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1820                                       entry->assigned_dev_id);
1821
1822         if (!adev) {
1823                 r = -EINVAL;
1824                 goto msix_entry_out;
1825         }
1826
1827         for (i = 0; i < adev->entries_nr; i++)
1828                 if (adev->guest_msix_entries[i].vector == 0 ||
1829                     adev->guest_msix_entries[i].entry == entry->entry) {
1830                         adev->guest_msix_entries[i].entry = entry->entry;
1831                         adev->guest_msix_entries[i].vector = entry->gsi;
1832                         adev->host_msix_entries[i].entry = entry->entry;
1833                         break;
1834                 }
1835         if (i == adev->entries_nr) {
1836                 r = -ENOSPC;
1837                 goto msix_entry_out;
1838         }
1839
1840 msix_entry_out:
1841         mutex_unlock(&kvm->lock);
1842
1843         return r;
1844 }
1845 #endif
1846
1847 static long kvm_vcpu_ioctl(struct file *filp,
1848                            unsigned int ioctl, unsigned long arg)
1849 {
1850         struct kvm_vcpu *vcpu = filp->private_data;
1851         void __user *argp = (void __user *)arg;
1852         int r;
1853         struct kvm_fpu *fpu = NULL;
1854         struct kvm_sregs *kvm_sregs = NULL;
1855
1856         if (vcpu->kvm->mm != current->mm)
1857                 return -EIO;
1858         switch (ioctl) {
1859         case KVM_RUN:
1860                 r = -EINVAL;
1861                 if (arg)
1862                         goto out;
1863                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1864                 break;
1865         case KVM_GET_REGS: {
1866                 struct kvm_regs *kvm_regs;
1867
1868                 r = -ENOMEM;
1869                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1870                 if (!kvm_regs)
1871                         goto out;
1872                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1873                 if (r)
1874                         goto out_free1;
1875                 r = -EFAULT;
1876                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1877                         goto out_free1;
1878                 r = 0;
1879 out_free1:
1880                 kfree(kvm_regs);
1881                 break;
1882         }
1883         case KVM_SET_REGS: {
1884                 struct kvm_regs *kvm_regs;
1885
1886                 r = -ENOMEM;
1887                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1888                 if (!kvm_regs)
1889                         goto out;
1890                 r = -EFAULT;
1891                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1892                         goto out_free2;
1893                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1894                 if (r)
1895                         goto out_free2;
1896                 r = 0;
1897 out_free2:
1898                 kfree(kvm_regs);
1899                 break;
1900         }
1901         case KVM_GET_SREGS: {
1902                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1903                 r = -ENOMEM;
1904                 if (!kvm_sregs)
1905                         goto out;
1906                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1907                 if (r)
1908                         goto out;
1909                 r = -EFAULT;
1910                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1911                         goto out;
1912                 r = 0;
1913                 break;
1914         }
1915         case KVM_SET_SREGS: {
1916                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1917                 r = -ENOMEM;
1918                 if (!kvm_sregs)
1919                         goto out;
1920                 r = -EFAULT;
1921                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1922                         goto out;
1923                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1924                 if (r)
1925                         goto out;
1926                 r = 0;
1927                 break;
1928         }
1929         case KVM_GET_MP_STATE: {
1930                 struct kvm_mp_state mp_state;
1931
1932                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1933                 if (r)
1934                         goto out;
1935                 r = -EFAULT;
1936                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1937                         goto out;
1938                 r = 0;
1939                 break;
1940         }
1941         case KVM_SET_MP_STATE: {
1942                 struct kvm_mp_state mp_state;
1943
1944                 r = -EFAULT;
1945                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1946                         goto out;
1947                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1948                 if (r)
1949                         goto out;
1950                 r = 0;
1951                 break;
1952         }
1953         case KVM_TRANSLATE: {
1954                 struct kvm_translation tr;
1955
1956                 r = -EFAULT;
1957                 if (copy_from_user(&tr, argp, sizeof tr))
1958                         goto out;
1959                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1960                 if (r)
1961                         goto out;
1962                 r = -EFAULT;
1963                 if (copy_to_user(argp, &tr, sizeof tr))
1964                         goto out;
1965                 r = 0;
1966                 break;
1967         }
1968         case KVM_SET_GUEST_DEBUG: {
1969                 struct kvm_guest_debug dbg;
1970
1971                 r = -EFAULT;
1972                 if (copy_from_user(&dbg, argp, sizeof dbg))
1973                         goto out;
1974                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1975                 if (r)
1976                         goto out;
1977                 r = 0;
1978                 break;
1979         }
1980         case KVM_SET_SIGNAL_MASK: {
1981                 struct kvm_signal_mask __user *sigmask_arg = argp;
1982                 struct kvm_signal_mask kvm_sigmask;
1983                 sigset_t sigset, *p;
1984
1985                 p = NULL;
1986                 if (argp) {
1987                         r = -EFAULT;
1988                         if (copy_from_user(&kvm_sigmask, argp,
1989                                            sizeof kvm_sigmask))
1990                                 goto out;
1991                         r = -EINVAL;
1992                         if (kvm_sigmask.len != sizeof sigset)
1993                                 goto out;
1994                         r = -EFAULT;
1995                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1996                                            sizeof sigset))
1997                                 goto out;
1998                         p = &sigset;
1999                 }
2000                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2001                 break;
2002         }
2003         case KVM_GET_FPU: {
2004                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2005                 r = -ENOMEM;
2006                 if (!fpu)
2007                         goto out;
2008                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2009                 if (r)
2010                         goto out;
2011                 r = -EFAULT;
2012                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2013                         goto out;
2014                 r = 0;
2015                 break;
2016         }
2017         case KVM_SET_FPU: {
2018                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2019                 r = -ENOMEM;
2020                 if (!fpu)
2021                         goto out;
2022                 r = -EFAULT;
2023                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2024                         goto out;
2025                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2026                 if (r)
2027                         goto out;
2028                 r = 0;
2029                 break;
2030         }
2031         default:
2032                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2033         }
2034 out:
2035         kfree(fpu);
2036         kfree(kvm_sregs);
2037         return r;
2038 }
2039
2040 static long kvm_vm_ioctl(struct file *filp,
2041                            unsigned int ioctl, unsigned long arg)
2042 {
2043         struct kvm *kvm = filp->private_data;
2044         void __user *argp = (void __user *)arg;
2045         int r;
2046
2047         if (kvm->mm != current->mm)
2048                 return -EIO;
2049         switch (ioctl) {
2050         case KVM_CREATE_VCPU:
2051                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2052                 if (r < 0)
2053                         goto out;
2054                 break;
2055         case KVM_SET_USER_MEMORY_REGION: {
2056                 struct kvm_userspace_memory_region kvm_userspace_mem;
2057
2058                 r = -EFAULT;
2059                 if (copy_from_user(&kvm_userspace_mem, argp,
2060                                                 sizeof kvm_userspace_mem))
2061                         goto out;
2062
2063                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2064                 if (r)
2065                         goto out;
2066                 break;
2067         }
2068         case KVM_GET_DIRTY_LOG: {
2069                 struct kvm_dirty_log log;
2070
2071                 r = -EFAULT;
2072                 if (copy_from_user(&log, argp, sizeof log))
2073                         goto out;
2074                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2075                 if (r)
2076                         goto out;
2077                 break;
2078         }
2079 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2080         case KVM_REGISTER_COALESCED_MMIO: {
2081                 struct kvm_coalesced_mmio_zone zone;
2082                 r = -EFAULT;
2083                 if (copy_from_user(&zone, argp, sizeof zone))
2084                         goto out;
2085                 r = -ENXIO;
2086                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2087                 if (r)
2088                         goto out;
2089                 r = 0;
2090                 break;
2091         }
2092         case KVM_UNREGISTER_COALESCED_MMIO: {
2093                 struct kvm_coalesced_mmio_zone zone;
2094                 r = -EFAULT;
2095                 if (copy_from_user(&zone, argp, sizeof zone))
2096                         goto out;
2097                 r = -ENXIO;
2098                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2099                 if (r)
2100                         goto out;
2101                 r = 0;
2102                 break;
2103         }
2104 #endif
2105 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2106         case KVM_ASSIGN_PCI_DEVICE: {
2107                 struct kvm_assigned_pci_dev assigned_dev;
2108
2109                 r = -EFAULT;
2110                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2111                         goto out;
2112                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2113                 if (r)
2114                         goto out;
2115                 break;
2116         }
2117         case KVM_ASSIGN_IRQ: {
2118                 r = -EOPNOTSUPP;
2119                 break;
2120         }
2121 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2122         case KVM_ASSIGN_DEV_IRQ: {
2123                 struct kvm_assigned_irq assigned_irq;
2124
2125                 r = -EFAULT;
2126                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2127                         goto out;
2128                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2129                 if (r)
2130                         goto out;
2131                 break;
2132         }
2133         case KVM_DEASSIGN_DEV_IRQ: {
2134                 struct kvm_assigned_irq assigned_irq;
2135
2136                 r = -EFAULT;
2137                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2138                         goto out;
2139                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2140                 if (r)
2141                         goto out;
2142                 break;
2143         }
2144 #endif
2145 #endif
2146 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2147         case KVM_DEASSIGN_PCI_DEVICE: {
2148                 struct kvm_assigned_pci_dev assigned_dev;
2149
2150                 r = -EFAULT;
2151                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2152                         goto out;
2153                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2154                 if (r)
2155                         goto out;
2156                 break;
2157         }
2158 #endif
2159 #ifdef KVM_CAP_IRQ_ROUTING
2160         case KVM_SET_GSI_ROUTING: {
2161                 struct kvm_irq_routing routing;
2162                 struct kvm_irq_routing __user *urouting;
2163                 struct kvm_irq_routing_entry *entries;
2164
2165                 r = -EFAULT;
2166                 if (copy_from_user(&routing, argp, sizeof(routing)))
2167                         goto out;
2168                 r = -EINVAL;
2169                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2170                         goto out;
2171                 if (routing.flags)
2172                         goto out;
2173                 r = -ENOMEM;
2174                 entries = vmalloc(routing.nr * sizeof(*entries));
2175                 if (!entries)
2176                         goto out;
2177                 r = -EFAULT;
2178                 urouting = argp;
2179                 if (copy_from_user(entries, urouting->entries,
2180                                    routing.nr * sizeof(*entries)))
2181                         goto out_free_irq_routing;
2182                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2183                                         routing.flags);
2184         out_free_irq_routing:
2185                 vfree(entries);
2186                 break;
2187         }
2188 #ifdef __KVM_HAVE_MSIX
2189         case KVM_ASSIGN_SET_MSIX_NR: {
2190                 struct kvm_assigned_msix_nr entry_nr;
2191                 r = -EFAULT;
2192                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2193                         goto out;
2194                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2195                 if (r)
2196                         goto out;
2197                 break;
2198         }
2199         case KVM_ASSIGN_SET_MSIX_ENTRY: {
2200                 struct kvm_assigned_msix_entry entry;
2201                 r = -EFAULT;
2202                 if (copy_from_user(&entry, argp, sizeof entry))
2203                         goto out;
2204                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2205                 if (r)
2206                         goto out;
2207                 break;
2208         }
2209 #endif
2210 #endif /* KVM_CAP_IRQ_ROUTING */
2211         default:
2212                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2213         }
2214 out:
2215         return r;
2216 }
2217
2218 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2219 {
2220         struct page *page[1];
2221         unsigned long addr;
2222         int npages;
2223         gfn_t gfn = vmf->pgoff;
2224         struct kvm *kvm = vma->vm_file->private_data;
2225
2226         addr = gfn_to_hva(kvm, gfn);
2227         if (kvm_is_error_hva(addr))
2228                 return VM_FAULT_SIGBUS;
2229
2230         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2231                                 NULL);
2232         if (unlikely(npages != 1))
2233                 return VM_FAULT_SIGBUS;
2234
2235         vmf->page = page[0];
2236         return 0;
2237 }
2238
2239 static struct vm_operations_struct kvm_vm_vm_ops = {
2240         .fault = kvm_vm_fault,
2241 };
2242
2243 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2244 {
2245         vma->vm_ops = &kvm_vm_vm_ops;
2246         return 0;
2247 }
2248
2249 static struct file_operations kvm_vm_fops = {
2250         .release        = kvm_vm_release,
2251         .unlocked_ioctl = kvm_vm_ioctl,
2252         .compat_ioctl   = kvm_vm_ioctl,
2253         .mmap           = kvm_vm_mmap,
2254 };
2255
2256 static int kvm_dev_ioctl_create_vm(void)
2257 {
2258         int fd;
2259         struct kvm *kvm;
2260
2261         kvm = kvm_create_vm();
2262         if (IS_ERR(kvm))
2263                 return PTR_ERR(kvm);
2264         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2265         if (fd < 0)
2266                 kvm_put_kvm(kvm);
2267
2268         return fd;
2269 }
2270
2271 static long kvm_dev_ioctl_check_extension_generic(long arg)
2272 {
2273         switch (arg) {
2274         case KVM_CAP_USER_MEMORY:
2275         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2276         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2277                 return 1;
2278 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2279         case KVM_CAP_IRQ_ROUTING:
2280                 return KVM_MAX_IRQ_ROUTES;
2281 #endif
2282         default:
2283                 break;
2284         }
2285         return kvm_dev_ioctl_check_extension(arg);
2286 }
2287
2288 static long kvm_dev_ioctl(struct file *filp,
2289                           unsigned int ioctl, unsigned long arg)
2290 {
2291         long r = -EINVAL;
2292
2293         switch (ioctl) {
2294         case KVM_GET_API_VERSION:
2295                 r = -EINVAL;
2296                 if (arg)
2297                         goto out;
2298                 r = KVM_API_VERSION;
2299                 break;
2300         case KVM_CREATE_VM:
2301                 r = -EINVAL;
2302                 if (arg)
2303                         goto out;
2304                 r = kvm_dev_ioctl_create_vm();
2305                 break;
2306         case KVM_CHECK_EXTENSION:
2307                 r = kvm_dev_ioctl_check_extension_generic(arg);
2308                 break;
2309         case KVM_GET_VCPU_MMAP_SIZE:
2310                 r = -EINVAL;
2311                 if (arg)
2312                         goto out;
2313                 r = PAGE_SIZE;     /* struct kvm_run */
2314 #ifdef CONFIG_X86
2315                 r += PAGE_SIZE;    /* pio data page */
2316 #endif
2317 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2318                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2319 #endif
2320                 break;
2321         case KVM_TRACE_ENABLE:
2322         case KVM_TRACE_PAUSE:
2323         case KVM_TRACE_DISABLE:
2324                 r = kvm_trace_ioctl(ioctl, arg);
2325                 break;
2326         default:
2327                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2328         }
2329 out:
2330         return r;
2331 }
2332
2333 static struct file_operations kvm_chardev_ops = {
2334         .unlocked_ioctl = kvm_dev_ioctl,
2335         .compat_ioctl   = kvm_dev_ioctl,
2336 };
2337
2338 static struct miscdevice kvm_dev = {
2339         KVM_MINOR,
2340         "kvm",
2341         &kvm_chardev_ops,
2342 };
2343
2344 static void hardware_enable(void *junk)
2345 {
2346         int cpu = raw_smp_processor_id();
2347
2348         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2349                 return;
2350         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2351         kvm_arch_hardware_enable(NULL);
2352 }
2353
2354 static void hardware_disable(void *junk)
2355 {
2356         int cpu = raw_smp_processor_id();
2357
2358         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2359                 return;
2360         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2361         kvm_arch_hardware_disable(NULL);
2362 }
2363
2364 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2365                            void *v)
2366 {
2367         int cpu = (long)v;
2368
2369         val &= ~CPU_TASKS_FROZEN;
2370         switch (val) {
2371         case CPU_DYING:
2372                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2373                        cpu);
2374                 hardware_disable(NULL);
2375                 break;
2376         case CPU_UP_CANCELED:
2377                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2378                        cpu);
2379                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2380                 break;
2381         case CPU_ONLINE:
2382                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2383                        cpu);
2384                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2385                 break;
2386         }
2387         return NOTIFY_OK;
2388 }
2389
2390
2391 asmlinkage void kvm_handle_fault_on_reboot(void)
2392 {
2393         if (kvm_rebooting)
2394                 /* spin while reset goes on */
2395                 while (true)
2396                         ;
2397         /* Fault while not rebooting.  We want the trace. */
2398         BUG();
2399 }
2400 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2401
2402 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2403                       void *v)
2404 {
2405         /*
2406          * Some (well, at least mine) BIOSes hang on reboot if
2407          * in vmx root mode.
2408          *
2409          * And Intel TXT required VMX off for all cpu when system shutdown.
2410          */
2411         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2412         kvm_rebooting = true;
2413         on_each_cpu(hardware_disable, NULL, 1);
2414         return NOTIFY_OK;
2415 }
2416
2417 static struct notifier_block kvm_reboot_notifier = {
2418         .notifier_call = kvm_reboot,
2419         .priority = 0,
2420 };
2421
2422 void kvm_io_bus_init(struct kvm_io_bus *bus)
2423 {
2424         memset(bus, 0, sizeof(*bus));
2425 }
2426
2427 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2428 {
2429         int i;
2430
2431         for (i = 0; i < bus->dev_count; i++) {
2432                 struct kvm_io_device *pos = bus->devs[i];
2433
2434                 kvm_iodevice_destructor(pos);
2435         }
2436 }
2437
2438 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2439                                           gpa_t addr, int len, int is_write)
2440 {
2441         int i;
2442
2443         for (i = 0; i < bus->dev_count; i++) {
2444                 struct kvm_io_device *pos = bus->devs[i];
2445
2446                 if (pos->in_range(pos, addr, len, is_write))
2447                         return pos;
2448         }
2449
2450         return NULL;
2451 }
2452
2453 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2454 {
2455         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2456
2457         bus->devs[bus->dev_count++] = dev;
2458 }
2459
2460 static struct notifier_block kvm_cpu_notifier = {
2461         .notifier_call = kvm_cpu_hotplug,
2462         .priority = 20, /* must be > scheduler priority */
2463 };
2464
2465 static int vm_stat_get(void *_offset, u64 *val)
2466 {
2467         unsigned offset = (long)_offset;
2468         struct kvm *kvm;
2469
2470         *val = 0;
2471         spin_lock(&kvm_lock);
2472         list_for_each_entry(kvm, &vm_list, vm_list)
2473                 *val += *(u32 *)((void *)kvm + offset);
2474         spin_unlock(&kvm_lock);
2475         return 0;
2476 }
2477
2478 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2479
2480 static int vcpu_stat_get(void *_offset, u64 *val)
2481 {
2482         unsigned offset = (long)_offset;
2483         struct kvm *kvm;
2484         struct kvm_vcpu *vcpu;
2485         int i;
2486
2487         *val = 0;
2488         spin_lock(&kvm_lock);
2489         list_for_each_entry(kvm, &vm_list, vm_list)
2490                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2491                         vcpu = kvm->vcpus[i];
2492                         if (vcpu)
2493                                 *val += *(u32 *)((void *)vcpu + offset);
2494                 }
2495         spin_unlock(&kvm_lock);
2496         return 0;
2497 }
2498
2499 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2500
2501 static struct file_operations *stat_fops[] = {
2502         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2503         [KVM_STAT_VM]   = &vm_stat_fops,
2504 };
2505
2506 static void kvm_init_debug(void)
2507 {
2508         struct kvm_stats_debugfs_item *p;
2509
2510         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2511         for (p = debugfs_entries; p->name; ++p)
2512                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2513                                                 (void *)(long)p->offset,
2514                                                 stat_fops[p->kind]);
2515 }
2516
2517 static void kvm_exit_debug(void)
2518 {
2519         struct kvm_stats_debugfs_item *p;
2520
2521         for (p = debugfs_entries; p->name; ++p)
2522                 debugfs_remove(p->dentry);
2523         debugfs_remove(kvm_debugfs_dir);
2524 }
2525
2526 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2527 {
2528         hardware_disable(NULL);
2529         return 0;
2530 }
2531
2532 static int kvm_resume(struct sys_device *dev)
2533 {
2534         hardware_enable(NULL);
2535         return 0;
2536 }
2537
2538 static struct sysdev_class kvm_sysdev_class = {
2539         .name = "kvm",
2540         .suspend = kvm_suspend,
2541         .resume = kvm_resume,
2542 };
2543
2544 static struct sys_device kvm_sysdev = {
2545         .id = 0,
2546         .cls = &kvm_sysdev_class,
2547 };
2548
2549 struct page *bad_page;
2550 pfn_t bad_pfn;
2551
2552 static inline
2553 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2554 {
2555         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2556 }
2557
2558 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2559 {
2560         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2561
2562         kvm_arch_vcpu_load(vcpu, cpu);
2563 }
2564
2565 static void kvm_sched_out(struct preempt_notifier *pn,
2566                           struct task_struct *next)
2567 {
2568         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2569
2570         kvm_arch_vcpu_put(vcpu);
2571 }
2572
2573 int kvm_init(void *opaque, unsigned int vcpu_size,
2574                   struct module *module)
2575 {
2576         int r;
2577         int cpu;
2578
2579         kvm_init_debug();
2580
2581         r = kvm_arch_init(opaque);
2582         if (r)
2583                 goto out_fail;
2584
2585         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2586
2587         if (bad_page == NULL) {
2588                 r = -ENOMEM;
2589                 goto out;
2590         }
2591
2592         bad_pfn = page_to_pfn(bad_page);
2593
2594         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2595                 r = -ENOMEM;
2596                 goto out_free_0;
2597         }
2598         cpumask_clear(cpus_hardware_enabled);
2599
2600         r = kvm_arch_hardware_setup();
2601         if (r < 0)
2602                 goto out_free_0a;
2603
2604         for_each_online_cpu(cpu) {
2605                 smp_call_function_single(cpu,
2606                                 kvm_arch_check_processor_compat,
2607                                 &r, 1);
2608                 if (r < 0)
2609                         goto out_free_1;
2610         }
2611
2612         on_each_cpu(hardware_enable, NULL, 1);
2613         r = register_cpu_notifier(&kvm_cpu_notifier);
2614         if (r)
2615                 goto out_free_2;
2616         register_reboot_notifier(&kvm_reboot_notifier);
2617
2618         r = sysdev_class_register(&kvm_sysdev_class);
2619         if (r)
2620                 goto out_free_3;
2621
2622         r = sysdev_register(&kvm_sysdev);
2623         if (r)
2624                 goto out_free_4;
2625
2626         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2627         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2628                                            __alignof__(struct kvm_vcpu),
2629                                            0, NULL);
2630         if (!kvm_vcpu_cache) {
2631                 r = -ENOMEM;
2632                 goto out_free_5;
2633         }
2634
2635         kvm_chardev_ops.owner = module;
2636         kvm_vm_fops.owner = module;
2637         kvm_vcpu_fops.owner = module;
2638
2639         r = misc_register(&kvm_dev);
2640         if (r) {
2641                 printk(KERN_ERR "kvm: misc device register failed\n");
2642                 goto out_free;
2643         }
2644
2645         kvm_preempt_ops.sched_in = kvm_sched_in;
2646         kvm_preempt_ops.sched_out = kvm_sched_out;
2647
2648         return 0;
2649
2650 out_free:
2651         kmem_cache_destroy(kvm_vcpu_cache);
2652 out_free_5:
2653         sysdev_unregister(&kvm_sysdev);
2654 out_free_4:
2655         sysdev_class_unregister(&kvm_sysdev_class);
2656 out_free_3:
2657         unregister_reboot_notifier(&kvm_reboot_notifier);
2658         unregister_cpu_notifier(&kvm_cpu_notifier);
2659 out_free_2:
2660         on_each_cpu(hardware_disable, NULL, 1);
2661 out_free_1:
2662         kvm_arch_hardware_unsetup();
2663 out_free_0a:
2664         free_cpumask_var(cpus_hardware_enabled);
2665 out_free_0:
2666         __free_page(bad_page);
2667 out:
2668         kvm_arch_exit();
2669         kvm_exit_debug();
2670 out_fail:
2671         return r;
2672 }
2673 EXPORT_SYMBOL_GPL(kvm_init);
2674
2675 void kvm_exit(void)
2676 {
2677         kvm_trace_cleanup();
2678         misc_deregister(&kvm_dev);
2679         kmem_cache_destroy(kvm_vcpu_cache);
2680         sysdev_unregister(&kvm_sysdev);
2681         sysdev_class_unregister(&kvm_sysdev_class);
2682         unregister_reboot_notifier(&kvm_reboot_notifier);
2683         unregister_cpu_notifier(&kvm_cpu_notifier);
2684         on_each_cpu(hardware_disable, NULL, 1);
2685         kvm_arch_hardware_unsetup();
2686         kvm_arch_exit();
2687         kvm_exit_debug();
2688         free_cpumask_var(cpus_hardware_enabled);
2689         __free_page(bad_page);
2690 }
2691 EXPORT_SYMBOL_GPL(kvm_exit);