SAFE public projects git trees. - safe/jmp/linux-2.6/blob - virt/kvm/kvm_main.c

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