virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "iodev.h"
  19
  20 #include <linux/kvm_host.h>
  21 #include <linux/kvm.h>
  22 #include <linux/module.h>
  23 #include <linux/errno.h>
  24 #include <linux/percpu.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/sysdev.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46
  47 #include <asm/processor.h>
  48 #include <asm/io.h>
  49 #include <asm/uaccess.h>
  50 #include <asm/pgtable.h>
  51
  52 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
  53 #include "coalesced_mmio.h"
  54 #endif
  55
  56 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
  57 #include <linux/pci.h>
  58 #include <linux/interrupt.h>
  59 #include "irq.h"
  60 #endif
  61
  62 MODULE_AUTHOR("Qumranet");
  63 MODULE_LICENSE("GPL");
  64
  65 DEFINE_SPINLOCK(kvm_lock);
  66 LIST_HEAD(vm_list);
  67
  68 static cpumask_var_t cpus_hardware_enabled;
  69
  70 struct kmem_cache *kvm_vcpu_cache;
  71 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  72
  73 static __read_mostly struct preempt_ops kvm_preempt_ops;
  74
  75 struct dentry *kvm_debugfs_dir;
  76
  77 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  78                            unsigned long arg);
  79
  80 static bool kvm_rebooting;
  81
  82 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
  83 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
  84                                                       int assigned_dev_id)
  85 {
  86         struct list_head *ptr;
  87         struct kvm_assigned_dev_kernel *match;
  88
  89         list_for_each(ptr, head) {
  90                 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
  91                 if (match->assigned_dev_id == assigned_dev_id)
  92                         return match;
  93         }
  94         return NULL;
  95 }
  96
  97 static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
  98                                     *assigned_dev, int irq)
  99 {
 100         int i, index;
 101         struct msix_entry *host_msix_entries;
 102
 103         host_msix_entries = assigned_dev->host_msix_entries;
 104
 105         index = -1;
 106         for (i = 0; i < assigned_dev->entries_nr; i++)
 107                 if (irq == host_msix_entries[i].vector) {
 108                         index = i;
 109                         break;
 110                 }
 111         if (index < 0) {
 112                 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
 113                 return 0;
 114         }
 115
 116         return index;
 117 }
 118
 119 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
 120 {
 121         struct kvm_assigned_dev_kernel *assigned_dev;
 122         struct kvm *kvm;
 123         int irq, i;
 124
 125         assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
 126                                     interrupt_work);
 127         kvm = assigned_dev->kvm;
 128
 129         /* This is taken to safely inject irq inside the guest. When
 130          * the interrupt injection (or the ioapic code) uses a
 131          * finer-grained lock, update this
 132          */
 133         mutex_lock(&kvm->lock);
 134         spin_lock_irq(&assigned_dev->assigned_dev_lock);
 135         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 136                 struct kvm_guest_msix_entry *guest_entries =
 137                         assigned_dev->guest_msix_entries;
 138                 for (i = 0; i < assigned_dev->entries_nr; i++) {
 139                         if (!(guest_entries[i].flags &
 140                                         KVM_ASSIGNED_MSIX_PENDING))
 141                                 continue;
 142                         guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
 143                         kvm_set_irq(assigned_dev->kvm,
 144                                     assigned_dev->irq_source_id,
 145                                     guest_entries[i].vector, 1);
 146                         irq = assigned_dev->host_msix_entries[i].vector;
 147                         if (irq != 0)
 148                                 enable_irq(irq);
 149                         assigned_dev->host_irq_disabled = false;
 150                 }
 151         } else {
 152                 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
 153                             assigned_dev->guest_irq, 1);
 154                 if (assigned_dev->irq_requested_type &
 155                                 KVM_DEV_IRQ_GUEST_MSI) {
 156                         enable_irq(assigned_dev->host_irq);
 157                         assigned_dev->host_irq_disabled = false;
 158                 }
 159         }
 160
 161         spin_unlock_irq(&assigned_dev->assigned_dev_lock);
 162         mutex_unlock(&assigned_dev->kvm->lock);
 163 }
 164
 165 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
 166 {
 167         unsigned long flags;
 168         struct kvm_assigned_dev_kernel *assigned_dev =
 169                 (struct kvm_assigned_dev_kernel *) dev_id;
 170
 171         spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
 172         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 173                 int index = find_index_from_host_irq(assigned_dev, irq);
 174                 if (index < 0)
 175                         goto out;
 176                 assigned_dev->guest_msix_entries[index].flags |=
 177                         KVM_ASSIGNED_MSIX_PENDING;
 178         }
 179
 180         schedule_work(&assigned_dev->interrupt_work);
 181
 182         disable_irq_nosync(irq);
 183         assigned_dev->host_irq_disabled = true;
 184
 185 out:
 186         spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
 187         return IRQ_HANDLED;
 188 }
 189
 190 /* Ack the irq line for an assigned device */
 191 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
 192 {
 193         struct kvm_assigned_dev_kernel *dev;
 194         unsigned long flags;
 195
 196         if (kian->gsi == -1)
 197                 return;
 198
 199         dev = container_of(kian, struct kvm_assigned_dev_kernel,
 200                            ack_notifier);
 201
 202         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
 203
 204         /* The guest irq may be shared so this ack may be
 205          * from another device.
 206          */
 207         spin_lock_irqsave(&dev->assigned_dev_lock, flags);
 208         if (dev->host_irq_disabled) {
 209                 enable_irq(dev->host_irq);
 210                 dev->host_irq_disabled = false;
 211         }
 212         spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
 213 }
 214
 215 static void deassign_guest_irq(struct kvm *kvm,
 216                                struct kvm_assigned_dev_kernel *assigned_dev)
 217 {
 218         kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
 219         assigned_dev->ack_notifier.gsi = -1;
 220
 221         if (assigned_dev->irq_source_id != -1)
 222                 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
 223         assigned_dev->irq_source_id = -1;
 224         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
 225 }
 226
 227 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
 228 static void deassign_host_irq(struct kvm *kvm,
 229                               struct kvm_assigned_dev_kernel *assigned_dev)
 230 {
 231         /*
 232          * In kvm_free_device_irq, cancel_work_sync return true if:
 233          * 1. work is scheduled, and then cancelled.
 234          * 2. work callback is executed.
 235          *
 236          * The first one ensured that the irq is disabled and no more events
 237          * would happen. But for the second one, the irq may be enabled (e.g.
 238          * for MSI). So we disable irq here to prevent further events.
 239          *
 240          * Notice this maybe result in nested disable if the interrupt type is
 241          * INTx, but it's OK for we are going to free it.
 242          *
 243          * If this function is a part of VM destroy, please ensure that till
 244          * now, the kvm state is still legal for probably we also have to wait
 245          * interrupt_work done.
 246          */
 247         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 248                 int i;
 249                 for (i = 0; i < assigned_dev->entries_nr; i++)
 250                         disable_irq_nosync(assigned_dev->
 251                                            host_msix_entries[i].vector);
 252
 253                 cancel_work_sync(&assigned_dev->interrupt_work);
 254
 255                 for (i = 0; i < assigned_dev->entries_nr; i++)
 256                         free_irq(assigned_dev->host_msix_entries[i].vector,
 257                                  (void *)assigned_dev);
 258
 259                 assigned_dev->entries_nr = 0;
 260                 kfree(assigned_dev->host_msix_entries);
 261                 kfree(assigned_dev->guest_msix_entries);
 262                 pci_disable_msix(assigned_dev->dev);
 263         } else {
 264                 /* Deal with MSI and INTx */
 265                 disable_irq_nosync(assigned_dev->host_irq);
 266                 cancel_work_sync(&assigned_dev->interrupt_work);
 267
 268                 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
 269
 270                 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
 271                         pci_disable_msi(assigned_dev->dev);
 272         }
 273
 274         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
 275 }
 276
 277 static int kvm_deassign_irq(struct kvm *kvm,
 278                             struct kvm_assigned_dev_kernel *assigned_dev,
 279                             unsigned long irq_requested_type)
 280 {
 281         unsigned long guest_irq_type, host_irq_type;
 282
 283         if (!irqchip_in_kernel(kvm))
 284                 return -EINVAL;
 285         /* no irq assignment to deassign */
 286         if (!assigned_dev->irq_requested_type)
 287                 return -ENXIO;
 288
 289         host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
 290         guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
 291
 292         if (host_irq_type)
 293                 deassign_host_irq(kvm, assigned_dev);
 294         if (guest_irq_type)
 295                 deassign_guest_irq(kvm, assigned_dev);
 296
 297         return 0;
 298 }
 299
 300 static void kvm_free_assigned_irq(struct kvm *kvm,
 301                                   struct kvm_assigned_dev_kernel *assigned_dev)
 302 {
 303         kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
 304 }
 305
 306 static void kvm_free_assigned_device(struct kvm *kvm,
 307                                      struct kvm_assigned_dev_kernel
 308                                      *assigned_dev)
 309 {
 310         kvm_free_assigned_irq(kvm, assigned_dev);
 311
 312         pci_reset_function(assigned_dev->dev);
 313
 314         pci_release_regions(assigned_dev->dev);
 315         pci_disable_device(assigned_dev->dev);
 316         pci_dev_put(assigned_dev->dev);
 317
 318         list_del(&assigned_dev->list);
 319         kfree(assigned_dev);
 320 }
 321
 322 void kvm_free_all_assigned_devices(struct kvm *kvm)
 323 {
 324         struct list_head *ptr, *ptr2;
 325         struct kvm_assigned_dev_kernel *assigned_dev;
 326
 327         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
 328                 assigned_dev = list_entry(ptr,
 329                                           struct kvm_assigned_dev_kernel,
 330                                           list);
 331
 332                 kvm_free_assigned_device(kvm, assigned_dev);
 333         }
 334 }
 335
 336 static int assigned_device_enable_host_intx(struct kvm *kvm,
 337                                             struct kvm_assigned_dev_kernel *dev)
 338 {
 339         dev->host_irq = dev->dev->irq;
 340         /* Even though this is PCI, we don't want to use shared
 341          * interrupts. Sharing host devices with guest-assigned devices
 342          * on the same interrupt line is not a happy situation: there
 343          * are going to be long delays in accepting, acking, etc.
 344          */
 345         if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
 346                         0, "kvm_assigned_intx_device", (void *)dev))
 347                 return -EIO;
 348         return 0;
 349 }
 350
 351 #ifdef __KVM_HAVE_MSI
 352 static int assigned_device_enable_host_msi(struct kvm *kvm,
 353                                            struct kvm_assigned_dev_kernel *dev)
 354 {
 355         int r;
 356
 357         if (!dev->dev->msi_enabled) {
 358                 r = pci_enable_msi(dev->dev);
 359                 if (r)
 360                         return r;
 361         }
 362
 363         dev->host_irq = dev->dev->irq;
 364         if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
 365                         "kvm_assigned_msi_device", (void *)dev)) {
 366                 pci_disable_msi(dev->dev);
 367                 return -EIO;
 368         }
 369
 370         return 0;
 371 }
 372 #endif
 373
 374 #ifdef __KVM_HAVE_MSIX
 375 static int assigned_device_enable_host_msix(struct kvm *kvm,
 376                                             struct kvm_assigned_dev_kernel *dev)
 377 {
 378         int i, r = -EINVAL;
 379
 380         /* host_msix_entries and guest_msix_entries should have been
 381          * initialized */
 382         if (dev->entries_nr == 0)
 383                 return r;
 384
 385         r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
 386         if (r)
 387                 return r;
 388
 389         for (i = 0; i < dev->entries_nr; i++) {
 390                 r = request_irq(dev->host_msix_entries[i].vector,
 391                                 kvm_assigned_dev_intr, 0,
 392                                 "kvm_assigned_msix_device",
 393                                 (void *)dev);
 394                 /* FIXME: free requested_irq's on failure */
 395                 if (r)
 396                         return r;
 397         }
 398
 399         return 0;
 400 }
 401
 402 #endif
 403
 404 static int assigned_device_enable_guest_intx(struct kvm *kvm,
 405                                 struct kvm_assigned_dev_kernel *dev,
 406                                 struct kvm_assigned_irq *irq)
 407 {
 408         dev->guest_irq = irq->guest_irq;
 409         dev->ack_notifier.gsi = irq->guest_irq;
 410         return 0;
 411 }
 412
 413 #ifdef __KVM_HAVE_MSI
 414 static int assigned_device_enable_guest_msi(struct kvm *kvm,
 415                         struct kvm_assigned_dev_kernel *dev,
 416                         struct kvm_assigned_irq *irq)
 417 {
 418         dev->guest_irq = irq->guest_irq;
 419         dev->ack_notifier.gsi = -1;
 420         return 0;
 421 }
 422 #endif
 423 #ifdef __KVM_HAVE_MSIX
 424 static int assigned_device_enable_guest_msix(struct kvm *kvm,
 425                         struct kvm_assigned_dev_kernel *dev,
 426                         struct kvm_assigned_irq *irq)
 427 {
 428         dev->guest_irq = irq->guest_irq;
 429         dev->ack_notifier.gsi = -1;
 430         return 0;
 431 }
 432 #endif
 433
 434 static int assign_host_irq(struct kvm *kvm,
 435                            struct kvm_assigned_dev_kernel *dev,
 436                            __u32 host_irq_type)
 437 {
 438         int r = -EEXIST;
 439
 440         if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
 441                 return r;
 442
 443         switch (host_irq_type) {
 444         case KVM_DEV_IRQ_HOST_INTX:
 445                 r = assigned_device_enable_host_intx(kvm, dev);
 446                 break;
 447 #ifdef __KVM_HAVE_MSI
 448         case KVM_DEV_IRQ_HOST_MSI:
 449                 r = assigned_device_enable_host_msi(kvm, dev);
 450                 break;
 451 #endif
 452 #ifdef __KVM_HAVE_MSIX
 453         case KVM_DEV_IRQ_HOST_MSIX:
 454                 r = assigned_device_enable_host_msix(kvm, dev);
 455                 break;
 456 #endif
 457         default:
 458                 r = -EINVAL;
 459         }
 460
 461         if (!r)
 462                 dev->irq_requested_type |= host_irq_type;
 463
 464         return r;
 465 }
 466
 467 static int assign_guest_irq(struct kvm *kvm,
 468                             struct kvm_assigned_dev_kernel *dev,
 469                             struct kvm_assigned_irq *irq,
 470                             unsigned long guest_irq_type)
 471 {
 472         int id;
 473         int r = -EEXIST;
 474
 475         if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
 476                 return r;
 477
 478         id = kvm_request_irq_source_id(kvm);
 479         if (id < 0)
 480                 return id;
 481
 482         dev->irq_source_id = id;
 483
 484         switch (guest_irq_type) {
 485         case KVM_DEV_IRQ_GUEST_INTX:
 486                 r = assigned_device_enable_guest_intx(kvm, dev, irq);
 487                 break;
 488 #ifdef __KVM_HAVE_MSI
 489         case KVM_DEV_IRQ_GUEST_MSI:
 490                 r = assigned_device_enable_guest_msi(kvm, dev, irq);
 491                 break;
 492 #endif
 493 #ifdef __KVM_HAVE_MSIX
 494         case KVM_DEV_IRQ_GUEST_MSIX:
 495                 r = assigned_device_enable_guest_msix(kvm, dev, irq);
 496                 break;
 497 #endif
 498         default:
 499                 r = -EINVAL;
 500         }
 501
 502         if (!r) {
 503                 dev->irq_requested_type |= guest_irq_type;
 504                 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
 505         } else
 506                 kvm_free_irq_source_id(kvm, dev->irq_source_id);
 507
 508         return r;
 509 }
 510
 511 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
 512 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 513                                    struct kvm_assigned_irq *assigned_irq)
 514 {
 515         int r = -EINVAL;
 516         struct kvm_assigned_dev_kernel *match;
 517         unsigned long host_irq_type, guest_irq_type;
 518
 519         if (!capable(CAP_SYS_RAWIO))
 520                 return -EPERM;
 521
 522         if (!irqchip_in_kernel(kvm))
 523                 return r;
 524
 525         mutex_lock(&kvm->lock);
 526         r = -ENODEV;
 527         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 528                                       assigned_irq->assigned_dev_id);
 529         if (!match)
 530                 goto out;
 531
 532         host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
 533         guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
 534
 535         r = -EINVAL;
 536         /* can only assign one type at a time */
 537         if (hweight_long(host_irq_type) > 1)
 538                 goto out;
 539         if (hweight_long(guest_irq_type) > 1)
 540                 goto out;
 541         if (host_irq_type == 0 && guest_irq_type == 0)
 542                 goto out;
 543
 544         r = 0;
 545         if (host_irq_type)
 546                 r = assign_host_irq(kvm, match, host_irq_type);
 547         if (r)
 548                 goto out;
 549
 550         if (guest_irq_type)
 551                 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
 552 out:
 553         mutex_unlock(&kvm->lock);
 554         return r;
 555 }
 556
 557 static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
 558                                          struct kvm_assigned_irq
 559                                          *assigned_irq)
 560 {
 561         int r = -ENODEV;
 562         struct kvm_assigned_dev_kernel *match;
 563
 564         mutex_lock(&kvm->lock);
 565
 566         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 567                                       assigned_irq->assigned_dev_id);
 568         if (!match)
 569                 goto out;
 570
 571         r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
 572 out:
 573         mutex_unlock(&kvm->lock);
 574         return r;
 575 }
 576
 577 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 578                                       struct kvm_assigned_pci_dev *assigned_dev)
 579 {
 580         int r = 0;
 581         struct kvm_assigned_dev_kernel *match;
 582         struct pci_dev *dev;
 583
 584         down_read(&kvm->slots_lock);
 585         mutex_lock(&kvm->lock);
 586
 587         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 588                                       assigned_dev->assigned_dev_id);
 589         if (match) {
 590                 /* device already assigned */
 591                 r = -EEXIST;
 592                 goto out;
 593         }
 594
 595         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 596         if (match == NULL) {
 597                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 598                        __func__);
 599                 r = -ENOMEM;
 600                 goto out;
 601         }
 602         dev = pci_get_bus_and_slot(assigned_dev->busnr,
 603                                    assigned_dev->devfn);
 604         if (!dev) {
 605                 printk(KERN_INFO "%s: host device not found\n", __func__);
 606                 r = -EINVAL;
 607                 goto out_free;
 608         }
 609         if (pci_enable_device(dev)) {
 610                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 611                 r = -EBUSY;
 612                 goto out_put;
 613         }
 614         r = pci_request_regions(dev, "kvm_assigned_device");
 615         if (r) {
 616                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 617                        __func__);
 618                 goto out_disable;
 619         }
 620
 621         pci_reset_function(dev);
 622
 623         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 624         match->host_busnr = assigned_dev->busnr;
 625         match->host_devfn = assigned_dev->devfn;
 626         match->flags = assigned_dev->flags;
 627         match->dev = dev;
 628         spin_lock_init(&match->assigned_dev_lock);
 629         match->irq_source_id = -1;
 630         match->kvm = kvm;
 631         match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
 632         INIT_WORK(&match->interrupt_work,
 633                   kvm_assigned_dev_interrupt_work_handler);
 634
 635         list_add(&match->list, &kvm->arch.assigned_dev_head);
 636
 637         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
 638                 if (!kvm->arch.iommu_domain) {
 639                         r = kvm_iommu_map_guest(kvm);
 640                         if (r)
 641                                 goto out_list_del;
 642                 }
 643                 r = kvm_assign_device(kvm, match);
 644                 if (r)
 645                         goto out_list_del;
 646         }
 647
 648 out:
 649         mutex_unlock(&kvm->lock);
 650         up_read(&kvm->slots_lock);
 651         return r;
 652 out_list_del:
 653         list_del(&match->list);
 654         pci_release_regions(dev);
 655 out_disable:
 656         pci_disable_device(dev);
 657 out_put:
 658         pci_dev_put(dev);
 659 out_free:
 660         kfree(match);
 661         mutex_unlock(&kvm->lock);
 662         up_read(&kvm->slots_lock);
 663         return r;
 664 }
 665 #endif
 666
 667 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
 668 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
 669                 struct kvm_assigned_pci_dev *assigned_dev)
 670 {
 671         int r = 0;
 672         struct kvm_assigned_dev_kernel *match;
 673
 674         mutex_lock(&kvm->lock);
 675
 676         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 677                                       assigned_dev->assigned_dev_id);
 678         if (!match) {
 679                 printk(KERN_INFO "%s: device hasn't been assigned before, "
 680                   "so cannot be deassigned\n", __func__);
 681                 r = -EINVAL;
 682                 goto out;
 683         }
 684
 685         if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
 686                 kvm_deassign_device(kvm, match);
 687
 688         kvm_free_assigned_device(kvm, match);
 689
 690 out:
 691         mutex_unlock(&kvm->lock);
 692         return r;
 693 }
 694 #endif
 695
 696 static inline int valid_vcpu(int n)
 697 {
 698         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 699 }
 700
 701 inline int kvm_is_mmio_pfn(pfn_t pfn)
 702 {
 703         if (pfn_valid(pfn)) {
 704                 struct page *page = compound_head(pfn_to_page(pfn));
 705                 return PageReserved(page);
 706         }
 707
 708         return true;
 709 }
 710
 711 /*
 712  * Switches to specified vcpu, until a matching vcpu_put()
 713  */
 714 void vcpu_load(struct kvm_vcpu *vcpu)
 715 {
 716         int cpu;
 717
 718         mutex_lock(&vcpu->mutex);
 719         cpu = get_cpu();
 720         preempt_notifier_register(&vcpu->preempt_notifier);
 721         kvm_arch_vcpu_load(vcpu, cpu);
 722         put_cpu();
 723 }
 724
 725 void vcpu_put(struct kvm_vcpu *vcpu)
 726 {
 727         preempt_disable();
 728         kvm_arch_vcpu_put(vcpu);
 729         preempt_notifier_unregister(&vcpu->preempt_notifier);
 730         preempt_enable();
 731         mutex_unlock(&vcpu->mutex);
 732 }
 733
 734 static void ack_flush(void *_completed)
 735 {
 736 }
 737
 738 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 739 {
 740         int i, cpu, me;
 741         cpumask_var_t cpus;
 742         bool called = true;
 743         struct kvm_vcpu *vcpu;
 744
 745         if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
 746                 cpumask_clear(cpus);
 747
 748         me = get_cpu();
 749         spin_lock(&kvm->requests_lock);
 750         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 751                 vcpu = kvm->vcpus[i];
 752                 if (!vcpu)
 753                         continue;
 754                 if (test_and_set_bit(req, &vcpu->requests))
 755                         continue;
 756                 cpu = vcpu->cpu;
 757                 if (cpus != NULL && cpu != -1 && cpu != me)
 758                         cpumask_set_cpu(cpu, cpus);
 759         }
 760         if (unlikely(cpus == NULL))
 761                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 762         else if (!cpumask_empty(cpus))
 763                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 764         else
 765                 called = false;
 766         spin_unlock(&kvm->requests_lock);
 767         put_cpu();
 768         free_cpumask_var(cpus);
 769         return called;
 770 }
 771
 772 void kvm_flush_remote_tlbs(struct kvm *kvm)
 773 {
 774         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 775                 ++kvm->stat.remote_tlb_flush;
 776 }
 777
 778 void kvm_reload_remote_mmus(struct kvm *kvm)
 779 {
 780         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 781 }
 782
 783 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 784 {
 785         struct page *page;
 786         int r;
 787
 788         mutex_init(&vcpu->mutex);
 789         vcpu->cpu = -1;
 790         vcpu->kvm = kvm;
 791         vcpu->vcpu_id = id;
 792         init_waitqueue_head(&vcpu->wq);
 793
 794         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 795         if (!page) {
 796                 r = -ENOMEM;
 797                 goto fail;
 798         }
 799         vcpu->run = page_address(page);
 800
 801         r = kvm_arch_vcpu_init(vcpu);
 802         if (r < 0)
 803                 goto fail_free_run;
 804         return 0;
 805
 806 fail_free_run:
 807         free_page((unsigned long)vcpu->run);
 808 fail:
 809         return r;
 810 }
 811 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 812
 813 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 814 {
 815         kvm_arch_vcpu_uninit(vcpu);
 816         free_page((unsigned long)vcpu->run);
 817 }
 818 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 819
 820 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 821 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 822 {
 823         return container_of(mn, struct kvm, mmu_notifier);
 824 }
 825
 826 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 827                                              struct mm_struct *mm,
 828                                              unsigned long address)
 829 {
 830         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 831         int need_tlb_flush;
 832
 833         /*
 834          * When ->invalidate_page runs, the linux pte has been zapped
 835          * already but the page is still allocated until
 836          * ->invalidate_page returns. So if we increase the sequence
 837          * here the kvm page fault will notice if the spte can't be
 838          * established because the page is going to be freed. If
 839          * instead the kvm page fault establishes the spte before
 840          * ->invalidate_page runs, kvm_unmap_hva will release it
 841          * before returning.
 842          *
 843          * The sequence increase only need to be seen at spin_unlock
 844          * time, and not at spin_lock time.
 845          *
 846          * Increasing the sequence after the spin_unlock would be
 847          * unsafe because the kvm page fault could then establish the
 848          * pte after kvm_unmap_hva returned, without noticing the page
 849          * is going to be freed.
 850          */
 851         spin_lock(&kvm->mmu_lock);
 852         kvm->mmu_notifier_seq++;
 853         need_tlb_flush = kvm_unmap_hva(kvm, address);
 854         spin_unlock(&kvm->mmu_lock);
 855
 856         /* we've to flush the tlb before the pages can be freed */
 857         if (need_tlb_flush)
 858                 kvm_flush_remote_tlbs(kvm);
 859
 860 }
 861
 862 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 863                                                     struct mm_struct *mm,
 864                                                     unsigned long start,
 865                                                     unsigned long end)
 866 {
 867         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 868         int need_tlb_flush = 0;
 869
 870         spin_lock(&kvm->mmu_lock);
 871         /*
 872          * The count increase must become visible at unlock time as no
 873          * spte can be established without taking the mmu_lock and
 874          * count is also read inside the mmu_lock critical section.
 875          */
 876         kvm->mmu_notifier_count++;
 877         for (; start < end; start += PAGE_SIZE)
 878                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 879         spin_unlock(&kvm->mmu_lock);
 880
 881         /* we've to flush the tlb before the pages can be freed */
 882         if (need_tlb_flush)
 883                 kvm_flush_remote_tlbs(kvm);
 884 }
 885
 886 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 887                                                   struct mm_struct *mm,
 888                                                   unsigned long start,
 889                                                   unsigned long end)
 890 {
 891         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 892
 893         spin_lock(&kvm->mmu_lock);
 894         /*
 895          * This sequence increase will notify the kvm page fault that
 896          * the page that is going to be mapped in the spte could have
 897          * been freed.
 898          */
 899         kvm->mmu_notifier_seq++;
 900         /*
 901          * The above sequence increase must be visible before the
 902          * below count decrease but both values are read by the kvm
 903          * page fault under mmu_lock spinlock so we don't need to add
 904          * a smb_wmb() here in between the two.
 905          */
 906         kvm->mmu_notifier_count--;
 907         spin_unlock(&kvm->mmu_lock);
 908
 909         BUG_ON(kvm->mmu_notifier_count < 0);
 910 }
 911
 912 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 913                                               struct mm_struct *mm,
 914                                               unsigned long address)
 915 {
 916         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 917         int young;
 918
 919         spin_lock(&kvm->mmu_lock);
 920         young = kvm_age_hva(kvm, address);
 921         spin_unlock(&kvm->mmu_lock);
 922
 923         if (young)
 924                 kvm_flush_remote_tlbs(kvm);
 925
 926         return young;
 927 }
 928
 929 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 930                                      struct mm_struct *mm)
 931 {
 932         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 933         kvm_arch_flush_shadow(kvm);
 934 }
 935
 936 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 937         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 938         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 939         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 940         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 941         .release                = kvm_mmu_notifier_release,
 942 };
 943 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 944
 945 static struct kvm *kvm_create_vm(void)
 946 {
 947         struct kvm *kvm = kvm_arch_create_vm();
 948 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 949         struct page *page;
 950 #endif
 951
 952         if (IS_ERR(kvm))
 953                 goto out;
 954 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 955         INIT_LIST_HEAD(&kvm->irq_routing);
 956         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 957 #endif
 958
 959 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 960         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 961         if (!page) {
 962                 kfree(kvm);
 963                 return ERR_PTR(-ENOMEM);
 964         }
 965         kvm->coalesced_mmio_ring =
 966                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 967 #endif
 968
 969 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 970         {
 971                 int err;
 972                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 973                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 974                 if (err) {
 975 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 976                         put_page(page);
 977 #endif
 978                         kfree(kvm);
 979                         return ERR_PTR(err);
 980                 }
 981         }
 982 #endif
 983
 984         kvm->mm = current->mm;
 985         atomic_inc(&kvm->mm->mm_count);
 986         spin_lock_init(&kvm->mmu_lock);
 987         spin_lock_init(&kvm->requests_lock);
 988         kvm_io_bus_init(&kvm->pio_bus);
 989         mutex_init(&kvm->lock);
 990         kvm_io_bus_init(&kvm->mmio_bus);
 991         init_rwsem(&kvm->slots_lock);
 992         atomic_set(&kvm->users_count, 1);
 993         spin_lock(&kvm_lock);
 994         list_add(&kvm->vm_list, &vm_list);
 995         spin_unlock(&kvm_lock);
 996 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 997         kvm_coalesced_mmio_init(kvm);
 998 #endif
 999 out:
1000         return kvm;
1001 }
1002
1003 /*
1004  * Free any memory in @free but not in @dont.
1005  */
1006 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
1007                                   struct kvm_memory_slot *dont)
1008 {
1009         if (!dont || free->rmap != dont->rmap)
1010                 vfree(free->rmap);
1011
1012         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
1013                 vfree(free->dirty_bitmap);
1014
1015         if (!dont || free->lpage_info != dont->lpage_info)
1016                 vfree(free->lpage_info);
1017
1018         free->npages = 0;
1019         free->dirty_bitmap = NULL;
1020         free->rmap = NULL;
1021         free->lpage_info = NULL;
1022 }
1023
1024 void kvm_free_physmem(struct kvm *kvm)
1025 {
1026         int i;
1027
1028         for (i = 0; i < kvm->nmemslots; ++i)
1029                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
1030 }
1031
1032 static void kvm_destroy_vm(struct kvm *kvm)
1033 {
1034         struct mm_struct *mm = kvm->mm;
1035
1036         kvm_arch_sync_events(kvm);
1037         spin_lock(&kvm_lock);
1038         list_del(&kvm->vm_list);
1039         spin_unlock(&kvm_lock);
1040         kvm_free_irq_routing(kvm);
1041         kvm_io_bus_destroy(&kvm->pio_bus);
1042         kvm_io_bus_destroy(&kvm->mmio_bus);
1043 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1044         if (kvm->coalesced_mmio_ring != NULL)
1045                 free_page((unsigned long)kvm->coalesced_mmio_ring);
1046 #endif
1047 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1048         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1049 #else
1050         kvm_arch_flush_shadow(kvm);
1051 #endif
1052         kvm_arch_destroy_vm(kvm);
1053         mmdrop(mm);
1054 }
1055
1056 void kvm_get_kvm(struct kvm *kvm)
1057 {
1058         atomic_inc(&kvm->users_count);
1059 }
1060 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1061
1062 void kvm_put_kvm(struct kvm *kvm)
1063 {
1064         if (atomic_dec_and_test(&kvm->users_count))
1065                 kvm_destroy_vm(kvm);
1066 }
1067 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1068
1069
1070 static int kvm_vm_release(struct inode *inode, struct file *filp)
1071 {
1072         struct kvm *kvm = filp->private_data;
1073
1074         kvm_put_kvm(kvm);
1075         return 0;
1076 }
1077
1078 /*
1079  * Allocate some memory and give it an address in the guest physical address
1080  * space.
1081  *
1082  * Discontiguous memory is allowed, mostly for framebuffers.
1083  *
1084  * Must be called holding mmap_sem for write.
1085  */
1086 int __kvm_set_memory_region(struct kvm *kvm,
1087                             struct kvm_userspace_memory_region *mem,
1088                             int user_alloc)
1089 {
1090         int r;
1091         gfn_t base_gfn;
1092         unsigned long npages, ugfn;
1093         unsigned long largepages, i;
1094         struct kvm_memory_slot *memslot;
1095         struct kvm_memory_slot old, new;
1096
1097         r = -EINVAL;
1098         /* General sanity checks */
1099         if (mem->memory_size & (PAGE_SIZE - 1))
1100                 goto out;
1101         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1102                 goto out;
1103         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
1104                 goto out;
1105         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
1106                 goto out;
1107         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1108                 goto out;
1109
1110         memslot = &kvm->memslots[mem->slot];
1111         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1112         npages = mem->memory_size >> PAGE_SHIFT;
1113
1114         if (!npages)
1115                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1116
1117         new = old = *memslot;
1118
1119         new.base_gfn = base_gfn;
1120         new.npages = npages;
1121         new.flags = mem->flags;
1122
1123         /* Disallow changing a memory slot's size. */
1124         r = -EINVAL;
1125         if (npages && old.npages && npages != old.npages)
1126                 goto out_free;
1127
1128         /* Check for overlaps */
1129         r = -EEXIST;
1130         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1131                 struct kvm_memory_slot *s = &kvm->memslots[i];
1132
1133                 if (s == memslot || !s->npages)
1134                         continue;
1135                 if (!((base_gfn + npages <= s->base_gfn) ||
1136                       (base_gfn >= s->base_gfn + s->npages)))
1137                         goto out_free;
1138         }
1139
1140         /* Free page dirty bitmap if unneeded */
1141         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1142                 new.dirty_bitmap = NULL;
1143
1144         r = -ENOMEM;
1145
1146         /* Allocate if a slot is being created */
1147 #ifndef CONFIG_S390
1148         if (npages && !new.rmap) {
1149                 new.rmap = vmalloc(npages * sizeof(struct page *));
1150
1151                 if (!new.rmap)
1152                         goto out_free;
1153
1154                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1155
1156                 new.user_alloc = user_alloc;
1157                 /*
1158                  * hva_to_rmmap() serialzies with the mmu_lock and to be
1159                  * safe it has to ignore memslots with !user_alloc &&
1160                  * !userspace_addr.
1161                  */
1162                 if (user_alloc)
1163                         new.userspace_addr = mem->userspace_addr;
1164                 else
1165                         new.userspace_addr = 0;
1166         }
1167         if (npages && !new.lpage_info) {
1168                 largepages = 1 + (base_gfn + npages - 1) / KVM_PAGES_PER_HPAGE;
1169                 largepages -= base_gfn / KVM_PAGES_PER_HPAGE;
1170
1171                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1172
1173                 if (!new.lpage_info)
1174                         goto out_free;
1175
1176                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1177
1178                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1179                         new.lpage_info[0].write_count = 1;
1180                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1181                         new.lpage_info[largepages-1].write_count = 1;
1182                 ugfn = new.userspace_addr >> PAGE_SHIFT;
1183                 /*
1184                  * If the gfn and userspace address are not aligned wrt each
1185                  * other, disable large page support for this slot
1186                  */
1187                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE - 1))
1188                         for (i = 0; i < largepages; ++i)
1189                                 new.lpage_info[i].write_count = 1;
1190         }
1191
1192         /* Allocate page dirty bitmap if needed */
1193         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1194                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1195
1196                 new.dirty_bitmap = vmalloc(dirty_bytes);
1197                 if (!new.dirty_bitmap)
1198                         goto out_free;
1199                 memset(new.dirty_bitmap, 0, dirty_bytes);
1200                 if (old.npages)
1201                         kvm_arch_flush_shadow(kvm);
1202         }
1203 #endif /* not defined CONFIG_S390 */
1204
1205         if (!npages)
1206                 kvm_arch_flush_shadow(kvm);
1207
1208         spin_lock(&kvm->mmu_lock);
1209         if (mem->slot >= kvm->nmemslots)
1210                 kvm->nmemslots = mem->slot + 1;
1211
1212         *memslot = new;
1213         spin_unlock(&kvm->mmu_lock);
1214
1215         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1216         if (r) {
1217                 spin_lock(&kvm->mmu_lock);
1218                 *memslot = old;
1219                 spin_unlock(&kvm->mmu_lock);
1220                 goto out_free;
1221         }
1222
1223         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1224         /* Slot deletion case: we have to update the current slot */
1225         spin_lock(&kvm->mmu_lock);
1226         if (!npages)
1227                 *memslot = old;
1228         spin_unlock(&kvm->mmu_lock);
1229 #ifdef CONFIG_DMAR
1230         /* map the pages in iommu page table */
1231         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1232         if (r)
1233                 goto out;
1234 #endif
1235         return 0;
1236
1237 out_free:
1238         kvm_free_physmem_slot(&new, &old);
1239 out:
1240         return r;
1241
1242 }
1243 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1244
1245 int kvm_set_memory_region(struct kvm *kvm,
1246                           struct kvm_userspace_memory_region *mem,
1247                           int user_alloc)
1248 {
1249         int r;
1250
1251         down_write(&kvm->slots_lock);
1252         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1253         up_write(&kvm->slots_lock);
1254         return r;
1255 }
1256 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1257
1258 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1259                                    struct
1260                                    kvm_userspace_memory_region *mem,
1261                                    int user_alloc)
1262 {
1263         if (mem->slot >= KVM_MEMORY_SLOTS)
1264                 return -EINVAL;
1265         return kvm_set_memory_region(kvm, mem, user_alloc);
1266 }
1267
1268 int kvm_get_dirty_log(struct kvm *kvm,
1269                         struct kvm_dirty_log *log, int *is_dirty)
1270 {
1271         struct kvm_memory_slot *memslot;
1272         int r, i;
1273         int n;
1274         unsigned long any = 0;
1275
1276         r = -EINVAL;
1277         if (log->slot >= KVM_MEMORY_SLOTS)
1278                 goto out;
1279
1280         memslot = &kvm->memslots[log->slot];
1281         r = -ENOENT;
1282         if (!memslot->dirty_bitmap)
1283                 goto out;
1284
1285         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1286
1287         for (i = 0; !any && i < n/sizeof(long); ++i)
1288                 any = memslot->dirty_bitmap[i];
1289
1290         r = -EFAULT;
1291         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1292                 goto out;
1293
1294         if (any)
1295                 *is_dirty = 1;
1296
1297         r = 0;
1298 out:
1299         return r;
1300 }
1301
1302 int is_error_page(struct page *page)
1303 {
1304         return page == bad_page;
1305 }
1306 EXPORT_SYMBOL_GPL(is_error_page);
1307
1308 int is_error_pfn(pfn_t pfn)
1309 {
1310         return pfn == bad_pfn;
1311 }
1312 EXPORT_SYMBOL_GPL(is_error_pfn);
1313
1314 static inline unsigned long bad_hva(void)
1315 {
1316         return PAGE_OFFSET;
1317 }
1318
1319 int kvm_is_error_hva(unsigned long addr)
1320 {
1321         return addr == bad_hva();
1322 }
1323 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1324
1325 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1326 {
1327         int i;
1328
1329         for (i = 0; i < kvm->nmemslots; ++i) {
1330                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1331
1332                 if (gfn >= memslot->base_gfn
1333                     && gfn < memslot->base_gfn + memslot->npages)
1334                         return memslot;
1335         }
1336         return NULL;
1337 }
1338 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1339
1340 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1341 {
1342         gfn = unalias_gfn(kvm, gfn);
1343         return gfn_to_memslot_unaliased(kvm, gfn);
1344 }
1345
1346 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1347 {
1348         int i;
1349
1350         gfn = unalias_gfn(kvm, gfn);
1351         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1352                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1353
1354                 if (gfn >= memslot->base_gfn
1355                     && gfn < memslot->base_gfn + memslot->npages)
1356                         return 1;
1357         }
1358         return 0;
1359 }
1360 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1361
1362 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1363 {
1364         struct kvm_memory_slot *slot;
1365
1366         gfn = unalias_gfn(kvm, gfn);
1367         slot = gfn_to_memslot_unaliased(kvm, gfn);
1368         if (!slot)
1369                 return bad_hva();
1370         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1371 }
1372 EXPORT_SYMBOL_GPL(gfn_to_hva);
1373
1374 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1375 {
1376         struct page *page[1];
1377         unsigned long addr;
1378         int npages;
1379         pfn_t pfn;
1380
1381         might_sleep();
1382
1383         addr = gfn_to_hva(kvm, gfn);
1384         if (kvm_is_error_hva(addr)) {
1385                 get_page(bad_page);
1386                 return page_to_pfn(bad_page);
1387         }
1388
1389         npages = get_user_pages_fast(addr, 1, 1, page);
1390
1391         if (unlikely(npages != 1)) {
1392                 struct vm_area_struct *vma;
1393
1394                 down_read(&current->mm->mmap_sem);
1395                 vma = find_vma(current->mm, addr);
1396
1397                 if (vma == NULL || addr < vma->vm_start ||
1398                     !(vma->vm_flags & VM_PFNMAP)) {
1399                         up_read(&current->mm->mmap_sem);
1400                         get_page(bad_page);
1401                         return page_to_pfn(bad_page);
1402                 }
1403
1404                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1405                 up_read(&current->mm->mmap_sem);
1406                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1407         } else
1408                 pfn = page_to_pfn(page[0]);
1409
1410         return pfn;
1411 }
1412
1413 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1414
1415 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1416 {
1417         pfn_t pfn;
1418
1419         pfn = gfn_to_pfn(kvm, gfn);
1420         if (!kvm_is_mmio_pfn(pfn))
1421                 return pfn_to_page(pfn);
1422
1423         WARN_ON(kvm_is_mmio_pfn(pfn));
1424
1425         get_page(bad_page);
1426         return bad_page;
1427 }
1428
1429 EXPORT_SYMBOL_GPL(gfn_to_page);
1430
1431 void kvm_release_page_clean(struct page *page)
1432 {
1433         kvm_release_pfn_clean(page_to_pfn(page));
1434 }
1435 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1436
1437 void kvm_release_pfn_clean(pfn_t pfn)
1438 {
1439         if (!kvm_is_mmio_pfn(pfn))
1440                 put_page(pfn_to_page(pfn));
1441 }
1442 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1443
1444 void kvm_release_page_dirty(struct page *page)
1445 {
1446         kvm_release_pfn_dirty(page_to_pfn(page));
1447 }
1448 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1449
1450 void kvm_release_pfn_dirty(pfn_t pfn)
1451 {
1452         kvm_set_pfn_dirty(pfn);
1453         kvm_release_pfn_clean(pfn);
1454 }
1455 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1456
1457 void kvm_set_page_dirty(struct page *page)
1458 {
1459         kvm_set_pfn_dirty(page_to_pfn(page));
1460 }
1461 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1462
1463 void kvm_set_pfn_dirty(pfn_t pfn)
1464 {
1465         if (!kvm_is_mmio_pfn(pfn)) {
1466                 struct page *page = pfn_to_page(pfn);
1467                 if (!PageReserved(page))
1468                         SetPageDirty(page);
1469         }
1470 }
1471 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1472
1473 void kvm_set_pfn_accessed(pfn_t pfn)
1474 {
1475         if (!kvm_is_mmio_pfn(pfn))
1476                 mark_page_accessed(pfn_to_page(pfn));
1477 }
1478 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1479
1480 void kvm_get_pfn(pfn_t pfn)
1481 {
1482         if (!kvm_is_mmio_pfn(pfn))
1483                 get_page(pfn_to_page(pfn));
1484 }
1485 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1486
1487 static int next_segment(unsigned long len, int offset)
1488 {
1489         if (len > PAGE_SIZE - offset)
1490                 return PAGE_SIZE - offset;
1491         else
1492                 return len;
1493 }
1494
1495 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1496                         int len)
1497 {
1498         int r;
1499         unsigned long addr;
1500
1501         addr = gfn_to_hva(kvm, gfn);
1502         if (kvm_is_error_hva(addr))
1503                 return -EFAULT;
1504         r = copy_from_user(data, (void __user *)addr + offset, len);
1505         if (r)
1506                 return -EFAULT;
1507         return 0;
1508 }
1509 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1510
1511 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1512 {
1513         gfn_t gfn = gpa >> PAGE_SHIFT;
1514         int seg;
1515         int offset = offset_in_page(gpa);
1516         int ret;
1517
1518         while ((seg = next_segment(len, offset)) != 0) {
1519                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1520                 if (ret < 0)
1521                         return ret;
1522                 offset = 0;
1523                 len -= seg;
1524                 data += seg;
1525                 ++gfn;
1526         }
1527         return 0;
1528 }
1529 EXPORT_SYMBOL_GPL(kvm_read_guest);
1530
1531 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1532                           unsigned long len)
1533 {
1534         int r;
1535         unsigned long addr;
1536         gfn_t gfn = gpa >> PAGE_SHIFT;
1537         int offset = offset_in_page(gpa);
1538
1539         addr = gfn_to_hva(kvm, gfn);
1540         if (kvm_is_error_hva(addr))
1541                 return -EFAULT;
1542         pagefault_disable();
1543         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1544         pagefault_enable();
1545         if (r)
1546                 return -EFAULT;
1547         return 0;
1548 }
1549 EXPORT_SYMBOL(kvm_read_guest_atomic);
1550
1551 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1552                          int offset, int len)
1553 {
1554         int r;
1555         unsigned long addr;
1556
1557         addr = gfn_to_hva(kvm, gfn);
1558         if (kvm_is_error_hva(addr))
1559                 return -EFAULT;
1560         r = copy_to_user((void __user *)addr + offset, data, len);
1561         if (r)
1562                 return -EFAULT;
1563         mark_page_dirty(kvm, gfn);
1564         return 0;
1565 }
1566 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1567
1568 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1569                     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_write_guest_page(kvm, gfn, data, offset, seg);
1578                 if (ret < 0)
1579                         return ret;
1580                 offset = 0;
1581                 len -= seg;
1582                 data += seg;
1583                 ++gfn;
1584         }
1585         return 0;
1586 }
1587
1588 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1589 {
1590         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1591 }
1592 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1593
1594 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1595 {
1596         gfn_t gfn = gpa >> PAGE_SHIFT;
1597         int seg;
1598         int offset = offset_in_page(gpa);
1599         int ret;
1600
1601         while ((seg = next_segment(len, offset)) != 0) {
1602                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1603                 if (ret < 0)
1604                         return ret;
1605                 offset = 0;
1606                 len -= seg;
1607                 ++gfn;
1608         }
1609         return 0;
1610 }
1611 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1612
1613 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1614 {
1615         struct kvm_memory_slot *memslot;
1616
1617         gfn = unalias_gfn(kvm, gfn);
1618         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1619         if (memslot && memslot->dirty_bitmap) {
1620                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1621
1622                 /* avoid RMW */
1623                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1624                         set_bit(rel_gfn, memslot->dirty_bitmap);
1625         }
1626 }
1627
1628 /*
1629  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1630  */
1631 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1632 {
1633         DEFINE_WAIT(wait);
1634
1635         for (;;) {
1636                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1637
1638                 if ((kvm_arch_interrupt_allowed(vcpu) &&
1639                                         kvm_cpu_has_interrupt(vcpu)) ||
1640                                 kvm_arch_vcpu_runnable(vcpu)) {
1641                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1642                         break;
1643                 }
1644                 if (kvm_cpu_has_pending_timer(vcpu))
1645                         break;
1646                 if (signal_pending(current))
1647                         break;
1648
1649                 vcpu_put(vcpu);
1650                 schedule();
1651                 vcpu_load(vcpu);
1652         }
1653
1654         finish_wait(&vcpu->wq, &wait);
1655 }
1656
1657 void kvm_resched(struct kvm_vcpu *vcpu)
1658 {
1659         if (!need_resched())
1660                 return;
1661         cond_resched();
1662 }
1663 EXPORT_SYMBOL_GPL(kvm_resched);
1664
1665 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1666 {
1667         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1668         struct page *page;
1669
1670         if (vmf->pgoff == 0)
1671                 page = virt_to_page(vcpu->run);
1672 #ifdef CONFIG_X86
1673         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1674                 page = virt_to_page(vcpu->arch.pio_data);
1675 #endif
1676 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1677         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1678                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1679 #endif
1680         else
1681                 return VM_FAULT_SIGBUS;
1682         get_page(page);
1683         vmf->page = page;
1684         return 0;
1685 }
1686
1687 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1688         .fault = kvm_vcpu_fault,
1689 };
1690
1691 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1692 {
1693         vma->vm_ops = &kvm_vcpu_vm_ops;
1694         return 0;
1695 }
1696
1697 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1698 {
1699         struct kvm_vcpu *vcpu = filp->private_data;
1700
1701         kvm_put_kvm(vcpu->kvm);
1702         return 0;
1703 }
1704
1705 static struct file_operations kvm_vcpu_fops = {
1706         .release        = kvm_vcpu_release,
1707         .unlocked_ioctl = kvm_vcpu_ioctl,
1708         .compat_ioctl   = kvm_vcpu_ioctl,
1709         .mmap           = kvm_vcpu_mmap,
1710 };
1711
1712 /*
1713  * Allocates an inode for the vcpu.
1714  */
1715 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1716 {
1717         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1718         if (fd < 0)
1719                 kvm_put_kvm(vcpu->kvm);
1720         return fd;
1721 }
1722
1723 /*
1724  * Creates some virtual cpus.  Good luck creating more than one.
1725  */
1726 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1727 {
1728         int r;
1729         struct kvm_vcpu *vcpu;
1730
1731         if (!valid_vcpu(n))
1732                 return -EINVAL;
1733
1734         vcpu = kvm_arch_vcpu_create(kvm, n);
1735         if (IS_ERR(vcpu))
1736                 return PTR_ERR(vcpu);
1737
1738         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1739
1740         r = kvm_arch_vcpu_setup(vcpu);
1741         if (r)
1742                 return r;
1743
1744         mutex_lock(&kvm->lock);
1745         if (kvm->vcpus[n]) {
1746                 r = -EEXIST;
1747                 goto vcpu_destroy;
1748         }
1749         kvm->vcpus[n] = vcpu;
1750         mutex_unlock(&kvm->lock);
1751
1752         /* Now it's all set up, let userspace reach it */
1753         kvm_get_kvm(kvm);
1754         r = create_vcpu_fd(vcpu);
1755         if (r < 0)
1756                 goto unlink;
1757         return r;
1758
1759 unlink:
1760         mutex_lock(&kvm->lock);
1761         kvm->vcpus[n] = NULL;
1762 vcpu_destroy:
1763         mutex_unlock(&kvm->lock);
1764         kvm_arch_vcpu_destroy(vcpu);
1765         return r;
1766 }
1767
1768 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1769 {
1770         if (sigset) {
1771                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1772                 vcpu->sigset_active = 1;
1773                 vcpu->sigset = *sigset;
1774         } else
1775                 vcpu->sigset_active = 0;
1776         return 0;
1777 }
1778
1779 #ifdef __KVM_HAVE_MSIX
1780 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1781                                     struct kvm_assigned_msix_nr *entry_nr)
1782 {
1783         int r = 0;
1784         struct kvm_assigned_dev_kernel *adev;
1785
1786         mutex_lock(&kvm->lock);
1787
1788         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1789                                       entry_nr->assigned_dev_id);
1790         if (!adev) {
1791                 r = -EINVAL;
1792                 goto msix_nr_out;
1793         }
1794
1795         if (adev->entries_nr == 0) {
1796                 adev->entries_nr = entry_nr->entry_nr;
1797                 if (adev->entries_nr == 0 ||
1798                     adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1799                         r = -EINVAL;
1800                         goto msix_nr_out;
1801                 }
1802
1803                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1804                                                 entry_nr->entry_nr,
1805                                                 GFP_KERNEL);
1806                 if (!adev->host_msix_entries) {
1807                         r = -ENOMEM;
1808                         goto msix_nr_out;
1809                 }
1810                 adev->guest_msix_entries = kzalloc(
1811                                 sizeof(struct kvm_guest_msix_entry) *
1812                                 entry_nr->entry_nr, GFP_KERNEL);
1813                 if (!adev->guest_msix_entries) {
1814                         kfree(adev->host_msix_entries);
1815                         r = -ENOMEM;
1816                         goto msix_nr_out;
1817                 }
1818         } else /* Not allowed set MSI-X number twice */
1819                 r = -EINVAL;
1820 msix_nr_out:
1821         mutex_unlock(&kvm->lock);
1822         return r;
1823 }
1824
1825 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1826                                        struct kvm_assigned_msix_entry *entry)
1827 {
1828         int r = 0, i;
1829         struct kvm_assigned_dev_kernel *adev;
1830
1831         mutex_lock(&kvm->lock);
1832
1833         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1834                                       entry->assigned_dev_id);
1835
1836         if (!adev) {
1837                 r = -EINVAL;
1838                 goto msix_entry_out;
1839         }
1840
1841         for (i = 0; i < adev->entries_nr; i++)
1842                 if (adev->guest_msix_entries[i].vector == 0 ||
1843                     adev->guest_msix_entries[i].entry == entry->entry) {
1844                         adev->guest_msix_entries[i].entry = entry->entry;
1845                         adev->guest_msix_entries[i].vector = entry->gsi;
1846                         adev->host_msix_entries[i].entry = entry->entry;
1847                         break;
1848                 }
1849         if (i == adev->entries_nr) {
1850                 r = -ENOSPC;
1851                 goto msix_entry_out;
1852         }
1853
1854 msix_entry_out:
1855         mutex_unlock(&kvm->lock);
1856
1857         return r;
1858 }
1859 #endif
1860
1861 static long kvm_vcpu_ioctl(struct file *filp,
1862                            unsigned int ioctl, unsigned long arg)
1863 {
1864         struct kvm_vcpu *vcpu = filp->private_data;
1865         void __user *argp = (void __user *)arg;
1866         int r;
1867         struct kvm_fpu *fpu = NULL;
1868         struct kvm_sregs *kvm_sregs = NULL;
1869
1870         if (vcpu->kvm->mm != current->mm)
1871                 return -EIO;
1872         switch (ioctl) {
1873         case KVM_RUN:
1874                 r = -EINVAL;
1875                 if (arg)
1876                         goto out;
1877                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1878                 break;
1879         case KVM_GET_REGS: {
1880                 struct kvm_regs *kvm_regs;
1881
1882                 r = -ENOMEM;
1883                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1884                 if (!kvm_regs)
1885                         goto out;
1886                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1887                 if (r)
1888                         goto out_free1;
1889                 r = -EFAULT;
1890                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1891                         goto out_free1;
1892                 r = 0;
1893 out_free1:
1894                 kfree(kvm_regs);
1895                 break;
1896         }
1897         case KVM_SET_REGS: {
1898                 struct kvm_regs *kvm_regs;
1899
1900                 r = -ENOMEM;
1901                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1902                 if (!kvm_regs)
1903                         goto out;
1904                 r = -EFAULT;
1905                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1906                         goto out_free2;
1907                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1908                 if (r)
1909                         goto out_free2;
1910                 r = 0;
1911 out_free2:
1912                 kfree(kvm_regs);
1913                 break;
1914         }
1915         case KVM_GET_SREGS: {
1916                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1917                 r = -ENOMEM;
1918                 if (!kvm_sregs)
1919                         goto out;
1920                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1921                 if (r)
1922                         goto out;
1923                 r = -EFAULT;
1924                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1925                         goto out;
1926                 r = 0;
1927                 break;
1928         }
1929         case KVM_SET_SREGS: {
1930                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1931                 r = -ENOMEM;
1932                 if (!kvm_sregs)
1933                         goto out;
1934                 r = -EFAULT;
1935                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1936                         goto out;
1937                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1938                 if (r)
1939                         goto out;
1940                 r = 0;
1941                 break;
1942         }
1943         case KVM_GET_MP_STATE: {
1944                 struct kvm_mp_state mp_state;
1945
1946                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1947                 if (r)
1948                         goto out;
1949                 r = -EFAULT;
1950                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1951                         goto out;
1952                 r = 0;
1953                 break;
1954         }
1955         case KVM_SET_MP_STATE: {
1956                 struct kvm_mp_state mp_state;
1957
1958                 r = -EFAULT;
1959                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1960                         goto out;
1961                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1962                 if (r)
1963                         goto out;
1964                 r = 0;
1965                 break;
1966         }
1967         case KVM_TRANSLATE: {
1968                 struct kvm_translation tr;
1969
1970                 r = -EFAULT;
1971                 if (copy_from_user(&tr, argp, sizeof tr))
1972                         goto out;
1973                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1974                 if (r)
1975                         goto out;
1976                 r = -EFAULT;
1977                 if (copy_to_user(argp, &tr, sizeof tr))
1978                         goto out;
1979                 r = 0;
1980                 break;
1981         }
1982         case KVM_SET_GUEST_DEBUG: {
1983                 struct kvm_guest_debug dbg;
1984
1985                 r = -EFAULT;
1986                 if (copy_from_user(&dbg, argp, sizeof dbg))
1987                         goto out;
1988                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1989                 if (r)
1990                         goto out;
1991                 r = 0;
1992                 break;
1993         }
1994         case KVM_SET_SIGNAL_MASK: {
1995                 struct kvm_signal_mask __user *sigmask_arg = argp;
1996                 struct kvm_signal_mask kvm_sigmask;
1997                 sigset_t sigset, *p;
1998
1999                 p = NULL;
2000                 if (argp) {
2001                         r = -EFAULT;
2002                         if (copy_from_user(&kvm_sigmask, argp,
2003                                            sizeof kvm_sigmask))
2004                                 goto out;
2005                         r = -EINVAL;
2006                         if (kvm_sigmask.len != sizeof sigset)
2007                                 goto out;
2008                         r = -EFAULT;
2009                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2010                                            sizeof sigset))
2011                                 goto out;
2012                         p = &sigset;
2013                 }
2014                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2015                 break;
2016         }
2017         case KVM_GET_FPU: {
2018                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2019                 r = -ENOMEM;
2020                 if (!fpu)
2021                         goto out;
2022                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2023                 if (r)
2024                         goto out;
2025                 r = -EFAULT;
2026                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2027                         goto out;
2028                 r = 0;
2029                 break;
2030         }
2031         case KVM_SET_FPU: {
2032                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2033                 r = -ENOMEM;
2034                 if (!fpu)
2035                         goto out;
2036                 r = -EFAULT;
2037                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2038                         goto out;
2039                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2040                 if (r)
2041                         goto out;
2042                 r = 0;
2043                 break;
2044         }
2045         default:
2046                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2047         }
2048 out:
2049         kfree(fpu);
2050         kfree(kvm_sregs);
2051         return r;
2052 }
2053
2054 static long kvm_vm_ioctl(struct file *filp,
2055                            unsigned int ioctl, unsigned long arg)
2056 {
2057         struct kvm *kvm = filp->private_data;
2058         void __user *argp = (void __user *)arg;
2059         int r;
2060
2061         if (kvm->mm != current->mm)
2062                 return -EIO;
2063         switch (ioctl) {
2064         case KVM_CREATE_VCPU:
2065                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2066                 if (r < 0)
2067                         goto out;
2068                 break;
2069         case KVM_SET_USER_MEMORY_REGION: {
2070                 struct kvm_userspace_memory_region kvm_userspace_mem;
2071
2072                 r = -EFAULT;
2073                 if (copy_from_user(&kvm_userspace_mem, argp,
2074                                                 sizeof kvm_userspace_mem))
2075                         goto out;
2076
2077                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2078                 if (r)
2079                         goto out;
2080                 break;
2081         }
2082         case KVM_GET_DIRTY_LOG: {
2083                 struct kvm_dirty_log log;
2084
2085                 r = -EFAULT;
2086                 if (copy_from_user(&log, argp, sizeof log))
2087                         goto out;
2088                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2089                 if (r)
2090                         goto out;
2091                 break;
2092         }
2093 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2094         case KVM_REGISTER_COALESCED_MMIO: {
2095                 struct kvm_coalesced_mmio_zone zone;
2096                 r = -EFAULT;
2097                 if (copy_from_user(&zone, argp, sizeof zone))
2098                         goto out;
2099                 r = -ENXIO;
2100                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2101                 if (r)
2102                         goto out;
2103                 r = 0;
2104                 break;
2105         }
2106         case KVM_UNREGISTER_COALESCED_MMIO: {
2107                 struct kvm_coalesced_mmio_zone zone;
2108                 r = -EFAULT;
2109                 if (copy_from_user(&zone, argp, sizeof zone))
2110                         goto out;
2111                 r = -ENXIO;
2112                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2113                 if (r)
2114                         goto out;
2115                 r = 0;
2116                 break;
2117         }
2118 #endif
2119 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2120         case KVM_ASSIGN_PCI_DEVICE: {
2121                 struct kvm_assigned_pci_dev assigned_dev;
2122
2123                 r = -EFAULT;
2124                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2125                         goto out;
2126                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2127                 if (r)
2128                         goto out;
2129                 break;
2130         }
2131         case KVM_ASSIGN_IRQ: {
2132                 r = -EOPNOTSUPP;
2133                 break;
2134         }
2135 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2136         case KVM_ASSIGN_DEV_IRQ: {
2137                 struct kvm_assigned_irq assigned_irq;
2138
2139                 r = -EFAULT;
2140                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2141                         goto out;
2142                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2143                 if (r)
2144                         goto out;
2145                 break;
2146         }
2147         case KVM_DEASSIGN_DEV_IRQ: {
2148                 struct kvm_assigned_irq assigned_irq;
2149
2150                 r = -EFAULT;
2151                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2152                         goto out;
2153                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2154                 if (r)
2155                         goto out;
2156                 break;
2157         }
2158 #endif
2159 #endif
2160 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2161         case KVM_DEASSIGN_PCI_DEVICE: {
2162                 struct kvm_assigned_pci_dev assigned_dev;
2163
2164                 r = -EFAULT;
2165                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2166                         goto out;
2167                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2168                 if (r)
2169                         goto out;
2170                 break;
2171         }
2172 #endif
2173 #ifdef KVM_CAP_IRQ_ROUTING
2174         case KVM_SET_GSI_ROUTING: {
2175                 struct kvm_irq_routing routing;
2176                 struct kvm_irq_routing __user *urouting;
2177                 struct kvm_irq_routing_entry *entries;
2178
2179                 r = -EFAULT;
2180                 if (copy_from_user(&routing, argp, sizeof(routing)))
2181                         goto out;
2182                 r = -EINVAL;
2183                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2184                         goto out;
2185                 if (routing.flags)
2186                         goto out;
2187                 r = -ENOMEM;
2188                 entries = vmalloc(routing.nr * sizeof(*entries));
2189                 if (!entries)
2190                         goto out;
2191                 r = -EFAULT;
2192                 urouting = argp;
2193                 if (copy_from_user(entries, urouting->entries,
2194                                    routing.nr * sizeof(*entries)))
2195                         goto out_free_irq_routing;
2196                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2197                                         routing.flags);
2198         out_free_irq_routing:
2199                 vfree(entries);
2200                 break;
2201         }
2202 #ifdef __KVM_HAVE_MSIX
2203         case KVM_ASSIGN_SET_MSIX_NR: {
2204                 struct kvm_assigned_msix_nr entry_nr;
2205                 r = -EFAULT;
2206                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2207                         goto out;
2208                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2209                 if (r)
2210                         goto out;
2211                 break;
2212         }
2213         case KVM_ASSIGN_SET_MSIX_ENTRY: {
2214                 struct kvm_assigned_msix_entry entry;
2215                 r = -EFAULT;
2216                 if (copy_from_user(&entry, argp, sizeof entry))
2217                         goto out;
2218                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2219                 if (r)
2220                         goto out;
2221                 break;
2222         }
2223 #endif
2224 #endif /* KVM_CAP_IRQ_ROUTING */
2225         default:
2226                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2227         }
2228 out:
2229         return r;
2230 }
2231
2232 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2233 {
2234         struct page *page[1];
2235         unsigned long addr;
2236         int npages;
2237         gfn_t gfn = vmf->pgoff;
2238         struct kvm *kvm = vma->vm_file->private_data;
2239
2240         addr = gfn_to_hva(kvm, gfn);
2241         if (kvm_is_error_hva(addr))
2242                 return VM_FAULT_SIGBUS;
2243
2244         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2245                                 NULL);
2246         if (unlikely(npages != 1))
2247                 return VM_FAULT_SIGBUS;
2248
2249         vmf->page = page[0];
2250         return 0;
2251 }
2252
2253 static struct vm_operations_struct kvm_vm_vm_ops = {
2254         .fault = kvm_vm_fault,
2255 };
2256
2257 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2258 {
2259         vma->vm_ops = &kvm_vm_vm_ops;
2260         return 0;
2261 }
2262
2263 static struct file_operations kvm_vm_fops = {
2264         .release        = kvm_vm_release,
2265         .unlocked_ioctl = kvm_vm_ioctl,
2266         .compat_ioctl   = kvm_vm_ioctl,
2267         .mmap           = kvm_vm_mmap,
2268 };
2269
2270 static int kvm_dev_ioctl_create_vm(void)
2271 {
2272         int fd;
2273         struct kvm *kvm;
2274
2275         kvm = kvm_create_vm();
2276         if (IS_ERR(kvm))
2277                 return PTR_ERR(kvm);
2278         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2279         if (fd < 0)
2280                 kvm_put_kvm(kvm);
2281
2282         return fd;
2283 }
2284
2285 static long kvm_dev_ioctl_check_extension_generic(long arg)
2286 {
2287         switch (arg) {
2288         case KVM_CAP_USER_MEMORY:
2289         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2290         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2291                 return 1;
2292 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2293         case KVM_CAP_IRQ_ROUTING:
2294                 return KVM_MAX_IRQ_ROUTES;
2295 #endif
2296         default:
2297                 break;
2298         }
2299         return kvm_dev_ioctl_check_extension(arg);
2300 }
2301
2302 static long kvm_dev_ioctl(struct file *filp,
2303                           unsigned int ioctl, unsigned long arg)
2304 {
2305         long r = -EINVAL;
2306
2307         switch (ioctl) {
2308         case KVM_GET_API_VERSION:
2309                 r = -EINVAL;
2310                 if (arg)
2311                         goto out;
2312                 r = KVM_API_VERSION;
2313                 break;
2314         case KVM_CREATE_VM:
2315                 r = -EINVAL;
2316                 if (arg)
2317                         goto out;
2318                 r = kvm_dev_ioctl_create_vm();
2319                 break;
2320         case KVM_CHECK_EXTENSION:
2321                 r = kvm_dev_ioctl_check_extension_generic(arg);
2322                 break;
2323         case KVM_GET_VCPU_MMAP_SIZE:
2324                 r = -EINVAL;
2325                 if (arg)
2326                         goto out;
2327                 r = PAGE_SIZE;     /* struct kvm_run */
2328 #ifdef CONFIG_X86
2329                 r += PAGE_SIZE;    /* pio data page */
2330 #endif
2331 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2332                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2333 #endif
2334                 break;
2335         case KVM_TRACE_ENABLE:
2336         case KVM_TRACE_PAUSE:
2337         case KVM_TRACE_DISABLE:
2338                 r = kvm_trace_ioctl(ioctl, arg);
2339                 break;
2340         default:
2341                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2342         }
2343 out:
2344         return r;
2345 }
2346
2347 static struct file_operations kvm_chardev_ops = {
2348         .unlocked_ioctl = kvm_dev_ioctl,
2349         .compat_ioctl   = kvm_dev_ioctl,
2350 };
2351
2352 static struct miscdevice kvm_dev = {
2353         KVM_MINOR,
2354         "kvm",
2355         &kvm_chardev_ops,
2356 };
2357
2358 static void hardware_enable(void *junk)
2359 {
2360         int cpu = raw_smp_processor_id();
2361
2362         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2363                 return;
2364         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2365         kvm_arch_hardware_enable(NULL);
2366 }
2367
2368 static void hardware_disable(void *junk)
2369 {
2370         int cpu = raw_smp_processor_id();
2371
2372         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2373                 return;
2374         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2375         kvm_arch_hardware_disable(NULL);
2376 }
2377
2378 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2379                            void *v)
2380 {
2381         int cpu = (long)v;
2382
2383         val &= ~CPU_TASKS_FROZEN;
2384         switch (val) {
2385         case CPU_DYING:
2386                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2387                        cpu);
2388                 hardware_disable(NULL);
2389                 break;
2390         case CPU_UP_CANCELED:
2391                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2392                        cpu);
2393                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2394                 break;
2395         case CPU_ONLINE:
2396                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2397                        cpu);
2398                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2399                 break;
2400         }
2401         return NOTIFY_OK;
2402 }
2403
2404
2405 asmlinkage void kvm_handle_fault_on_reboot(void)
2406 {
2407         if (kvm_rebooting)
2408                 /* spin while reset goes on */
2409                 while (true)
2410                         ;
2411         /* Fault while not rebooting.  We want the trace. */
2412         BUG();
2413 }
2414 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2415
2416 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2417                       void *v)
2418 {
2419         /*
2420          * Some (well, at least mine) BIOSes hang on reboot if
2421          * in vmx root mode.
2422          *
2423          * And Intel TXT required VMX off for all cpu when system shutdown.
2424          */
2425         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2426         kvm_rebooting = true;
2427         on_each_cpu(hardware_disable, NULL, 1);
2428         return NOTIFY_OK;
2429 }
2430
2431 static struct notifier_block kvm_reboot_notifier = {
2432         .notifier_call = kvm_reboot,
2433         .priority = 0,
2434 };
2435
2436 void kvm_io_bus_init(struct kvm_io_bus *bus)
2437 {
2438         memset(bus, 0, sizeof(*bus));
2439 }
2440
2441 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2442 {
2443         int i;
2444
2445         for (i = 0; i < bus->dev_count; i++) {
2446                 struct kvm_io_device *pos = bus->devs[i];
2447
2448                 kvm_iodevice_destructor(pos);
2449         }
2450 }
2451
2452 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2453                                           gpa_t addr, int len, int is_write)
2454 {
2455         int i;
2456
2457         for (i = 0; i < bus->dev_count; i++) {
2458                 struct kvm_io_device *pos = bus->devs[i];
2459
2460                 if (pos->in_range(pos, addr, len, is_write))
2461                         return pos;
2462         }
2463
2464         return NULL;
2465 }
2466
2467 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2468 {
2469         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2470
2471         bus->devs[bus->dev_count++] = dev;
2472 }
2473
2474 static struct notifier_block kvm_cpu_notifier = {
2475         .notifier_call = kvm_cpu_hotplug,
2476         .priority = 20, /* must be > scheduler priority */
2477 };
2478
2479 static int vm_stat_get(void *_offset, u64 *val)
2480 {
2481         unsigned offset = (long)_offset;
2482         struct kvm *kvm;
2483
2484         *val = 0;
2485         spin_lock(&kvm_lock);
2486         list_for_each_entry(kvm, &vm_list, vm_list)
2487                 *val += *(u32 *)((void *)kvm + offset);
2488         spin_unlock(&kvm_lock);
2489         return 0;
2490 }
2491
2492 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2493
2494 static int vcpu_stat_get(void *_offset, u64 *val)
2495 {
2496         unsigned offset = (long)_offset;
2497         struct kvm *kvm;
2498         struct kvm_vcpu *vcpu;
2499         int i;
2500
2501         *val = 0;
2502         spin_lock(&kvm_lock);
2503         list_for_each_entry(kvm, &vm_list, vm_list)
2504                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2505                         vcpu = kvm->vcpus[i];
2506                         if (vcpu)
2507                                 *val += *(u32 *)((void *)vcpu + offset);
2508                 }
2509         spin_unlock(&kvm_lock);
2510         return 0;
2511 }
2512
2513 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2514
2515 static struct file_operations *stat_fops[] = {
2516         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2517         [KVM_STAT_VM]   = &vm_stat_fops,
2518 };
2519
2520 static void kvm_init_debug(void)
2521 {
2522         struct kvm_stats_debugfs_item *p;
2523
2524         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2525         for (p = debugfs_entries; p->name; ++p)
2526                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2527                                                 (void *)(long)p->offset,
2528                                                 stat_fops[p->kind]);
2529 }
2530
2531 static void kvm_exit_debug(void)
2532 {
2533         struct kvm_stats_debugfs_item *p;
2534
2535         for (p = debugfs_entries; p->name; ++p)
2536                 debugfs_remove(p->dentry);
2537         debugfs_remove(kvm_debugfs_dir);
2538 }
2539
2540 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2541 {
2542         hardware_disable(NULL);
2543         return 0;
2544 }
2545
2546 static int kvm_resume(struct sys_device *dev)
2547 {
2548         hardware_enable(NULL);
2549         return 0;
2550 }
2551
2552 static struct sysdev_class kvm_sysdev_class = {
2553         .name = "kvm",
2554         .suspend = kvm_suspend,
2555         .resume = kvm_resume,
2556 };
2557
2558 static struct sys_device kvm_sysdev = {
2559         .id = 0,
2560         .cls = &kvm_sysdev_class,
2561 };
2562
2563 struct page *bad_page;
2564 pfn_t bad_pfn;
2565
2566 static inline
2567 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2568 {
2569         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2570 }
2571
2572 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2573 {
2574         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2575
2576         kvm_arch_vcpu_load(vcpu, cpu);
2577 }
2578
2579 static void kvm_sched_out(struct preempt_notifier *pn,
2580                           struct task_struct *next)
2581 {
2582         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2583
2584         kvm_arch_vcpu_put(vcpu);
2585 }
2586
2587 int kvm_init(void *opaque, unsigned int vcpu_size,
2588                   struct module *module)
2589 {
2590         int r;
2591         int cpu;
2592
2593         kvm_init_debug();
2594
2595         r = kvm_arch_init(opaque);
2596         if (r)
2597                 goto out_fail;
2598
2599         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2600
2601         if (bad_page == NULL) {
2602                 r = -ENOMEM;
2603                 goto out;
2604         }
2605
2606         bad_pfn = page_to_pfn(bad_page);
2607
2608         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2609                 r = -ENOMEM;
2610                 goto out_free_0;
2611         }
2612
2613         r = kvm_arch_hardware_setup();
2614         if (r < 0)
2615                 goto out_free_0a;
2616
2617         for_each_online_cpu(cpu) {
2618                 smp_call_function_single(cpu,
2619                                 kvm_arch_check_processor_compat,
2620                                 &r, 1);
2621                 if (r < 0)
2622                         goto out_free_1;
2623         }
2624
2625         on_each_cpu(hardware_enable, NULL, 1);
2626         r = register_cpu_notifier(&kvm_cpu_notifier);
2627         if (r)
2628                 goto out_free_2;
2629         register_reboot_notifier(&kvm_reboot_notifier);
2630
2631         r = sysdev_class_register(&kvm_sysdev_class);
2632         if (r)
2633                 goto out_free_3;
2634
2635         r = sysdev_register(&kvm_sysdev);
2636         if (r)
2637                 goto out_free_4;
2638
2639         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2640         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2641                                            __alignof__(struct kvm_vcpu),
2642                                            0, NULL);
2643         if (!kvm_vcpu_cache) {
2644                 r = -ENOMEM;
2645                 goto out_free_5;
2646         }
2647
2648         kvm_chardev_ops.owner = module;
2649         kvm_vm_fops.owner = module;
2650         kvm_vcpu_fops.owner = module;
2651
2652         r = misc_register(&kvm_dev);
2653         if (r) {
2654                 printk(KERN_ERR "kvm: misc device register failed\n");
2655                 goto out_free;
2656         }
2657
2658         kvm_preempt_ops.sched_in = kvm_sched_in;
2659         kvm_preempt_ops.sched_out = kvm_sched_out;
2660
2661         return 0;
2662
2663 out_free:
2664         kmem_cache_destroy(kvm_vcpu_cache);
2665 out_free_5:
2666         sysdev_unregister(&kvm_sysdev);
2667 out_free_4:
2668         sysdev_class_unregister(&kvm_sysdev_class);
2669 out_free_3:
2670         unregister_reboot_notifier(&kvm_reboot_notifier);
2671         unregister_cpu_notifier(&kvm_cpu_notifier);
2672 out_free_2:
2673         on_each_cpu(hardware_disable, NULL, 1);
2674 out_free_1:
2675         kvm_arch_hardware_unsetup();
2676 out_free_0a:
2677         free_cpumask_var(cpus_hardware_enabled);
2678 out_free_0:
2679         __free_page(bad_page);
2680 out:
2681         kvm_arch_exit();
2682         kvm_exit_debug();
2683 out_fail:
2684         return r;
2685 }
2686 EXPORT_SYMBOL_GPL(kvm_init);
2687
2688 void kvm_exit(void)
2689 {
2690         kvm_trace_cleanup();
2691         misc_deregister(&kvm_dev);
2692         kmem_cache_destroy(kvm_vcpu_cache);
2693         sysdev_unregister(&kvm_sysdev);
2694         sysdev_class_unregister(&kvm_sysdev_class);
2695         unregister_reboot_notifier(&kvm_reboot_notifier);
2696         unregister_cpu_notifier(&kvm_cpu_notifier);
2697         on_each_cpu(hardware_disable, NULL, 1);
2698         kvm_arch_hardware_unsetup();
2699         kvm_arch_exit();
2700         kvm_exit_debug();
2701         free_cpumask_var(cpus_hardware_enabled);
2702         __free_page(bad_page);
2703 }
2704 EXPORT_SYMBOL_GPL(kvm_exit);