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 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                 }
 147         } else
 148                 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
 149                             assigned_dev->guest_irq, 1);
 150
 151         spin_unlock_irq(&assigned_dev->assigned_dev_lock);
 152         mutex_unlock(&assigned_dev->kvm->lock);
 153 }
 154
 155 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
 156 {
 157         unsigned long flags;
 158         struct kvm_assigned_dev_kernel *assigned_dev =
 159                 (struct kvm_assigned_dev_kernel *) dev_id;
 160
 161         spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
 162         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 163                 int index = find_index_from_host_irq(assigned_dev, irq);
 164                 if (index < 0)
 165                         goto out;
 166                 assigned_dev->guest_msix_entries[index].flags |=
 167                         KVM_ASSIGNED_MSIX_PENDING;
 168         }
 169
 170         schedule_work(&assigned_dev->interrupt_work);
 171
 172         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
 173                 disable_irq_nosync(irq);
 174                 assigned_dev->host_irq_disabled = true;
 175         }
 176
 177 out:
 178         spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
 179         return IRQ_HANDLED;
 180 }
 181
 182 /* Ack the irq line for an assigned device */
 183 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
 184 {
 185         struct kvm_assigned_dev_kernel *dev;
 186         unsigned long flags;
 187
 188         if (kian->gsi == -1)
 189                 return;
 190
 191         dev = container_of(kian, struct kvm_assigned_dev_kernel,
 192                            ack_notifier);
 193
 194         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
 195
 196         /* The guest irq may be shared so this ack may be
 197          * from another device.
 198          */
 199         spin_lock_irqsave(&dev->assigned_dev_lock, flags);
 200         if (dev->host_irq_disabled) {
 201                 enable_irq(dev->host_irq);
 202                 dev->host_irq_disabled = false;
 203         }
 204         spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
 205 }
 206
 207 static void deassign_guest_irq(struct kvm *kvm,
 208                                struct kvm_assigned_dev_kernel *assigned_dev)
 209 {
 210         kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
 211         assigned_dev->ack_notifier.gsi = -1;
 212
 213         if (assigned_dev->irq_source_id != -1)
 214                 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
 215         assigned_dev->irq_source_id = -1;
 216         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
 217 }
 218
 219 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
 220 static void deassign_host_irq(struct kvm *kvm,
 221                               struct kvm_assigned_dev_kernel *assigned_dev)
 222 {
 223         /*
 224          * In kvm_free_device_irq, cancel_work_sync return true if:
 225          * 1. work is scheduled, and then cancelled.
 226          * 2. work callback is executed.
 227          *
 228          * The first one ensured that the irq is disabled and no more events
 229          * would happen. But for the second one, the irq may be enabled (e.g.
 230          * for MSI). So we disable irq here to prevent further events.
 231          *
 232          * Notice this maybe result in nested disable if the interrupt type is
 233          * INTx, but it's OK for we are going to free it.
 234          *
 235          * If this function is a part of VM destroy, please ensure that till
 236          * now, the kvm state is still legal for probably we also have to wait
 237          * interrupt_work done.
 238          */
 239         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 240                 int i;
 241                 for (i = 0; i < assigned_dev->entries_nr; i++)
 242                         disable_irq_nosync(assigned_dev->
 243                                            host_msix_entries[i].vector);
 244
 245                 cancel_work_sync(&assigned_dev->interrupt_work);
 246
 247                 for (i = 0; i < assigned_dev->entries_nr; i++)
 248                         free_irq(assigned_dev->host_msix_entries[i].vector,
 249                                  (void *)assigned_dev);
 250
 251                 assigned_dev->entries_nr = 0;
 252                 kfree(assigned_dev->host_msix_entries);
 253                 kfree(assigned_dev->guest_msix_entries);
 254                 pci_disable_msix(assigned_dev->dev);
 255         } else {
 256                 /* Deal with MSI and INTx */
 257                 disable_irq_nosync(assigned_dev->host_irq);
 258                 cancel_work_sync(&assigned_dev->interrupt_work);
 259
 260                 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
 261
 262                 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
 263                         pci_disable_msi(assigned_dev->dev);
 264         }
 265
 266         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
 267 }
 268
 269 static int kvm_deassign_irq(struct kvm *kvm,
 270                             struct kvm_assigned_dev_kernel *assigned_dev,
 271                             unsigned long irq_requested_type)
 272 {
 273         unsigned long guest_irq_type, host_irq_type;
 274
 275         if (!irqchip_in_kernel(kvm))
 276                 return -EINVAL;
 277         /* no irq assignment to deassign */
 278         if (!assigned_dev->irq_requested_type)
 279                 return -ENXIO;
 280
 281         host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
 282         guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
 283
 284         if (host_irq_type)
 285                 deassign_host_irq(kvm, assigned_dev);
 286         if (guest_irq_type)
 287                 deassign_guest_irq(kvm, assigned_dev);
 288
 289         return 0;
 290 }
 291
 292 static void kvm_free_assigned_irq(struct kvm *kvm,
 293                                   struct kvm_assigned_dev_kernel *assigned_dev)
 294 {
 295         kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
 296 }
 297
 298 static void kvm_free_assigned_device(struct kvm *kvm,
 299                                      struct kvm_assigned_dev_kernel
 300                                      *assigned_dev)
 301 {
 302         kvm_free_assigned_irq(kvm, assigned_dev);
 303
 304         pci_reset_function(assigned_dev->dev);
 305
 306         pci_release_regions(assigned_dev->dev);
 307         pci_disable_device(assigned_dev->dev);
 308         pci_dev_put(assigned_dev->dev);
 309
 310         list_del(&assigned_dev->list);
 311         kfree(assigned_dev);
 312 }
 313
 314 void kvm_free_all_assigned_devices(struct kvm *kvm)
 315 {
 316         struct list_head *ptr, *ptr2;
 317         struct kvm_assigned_dev_kernel *assigned_dev;
 318
 319         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
 320                 assigned_dev = list_entry(ptr,
 321                                           struct kvm_assigned_dev_kernel,
 322                                           list);
 323
 324                 kvm_free_assigned_device(kvm, assigned_dev);
 325         }
 326 }
 327
 328 static int assigned_device_enable_host_intx(struct kvm *kvm,
 329                                             struct kvm_assigned_dev_kernel *dev)
 330 {
 331         dev->host_irq = dev->dev->irq;
 332         /* Even though this is PCI, we don't want to use shared
 333          * interrupts. Sharing host devices with guest-assigned devices
 334          * on the same interrupt line is not a happy situation: there
 335          * are going to be long delays in accepting, acking, etc.
 336          */
 337         if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
 338                         0, "kvm_assigned_intx_device", (void *)dev))
 339                 return -EIO;
 340         return 0;
 341 }
 342
 343 #ifdef __KVM_HAVE_MSI
 344 static int assigned_device_enable_host_msi(struct kvm *kvm,
 345                                            struct kvm_assigned_dev_kernel *dev)
 346 {
 347         int r;
 348
 349         if (!dev->dev->msi_enabled) {
 350                 r = pci_enable_msi(dev->dev);
 351                 if (r)
 352                         return r;
 353         }
 354
 355         dev->host_irq = dev->dev->irq;
 356         if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
 357                         "kvm_assigned_msi_device", (void *)dev)) {
 358                 pci_disable_msi(dev->dev);
 359                 return -EIO;
 360         }
 361
 362         return 0;
 363 }
 364 #endif
 365
 366 #ifdef __KVM_HAVE_MSIX
 367 static int assigned_device_enable_host_msix(struct kvm *kvm,
 368                                             struct kvm_assigned_dev_kernel *dev)
 369 {
 370         int i, r = -EINVAL;
 371
 372         /* host_msix_entries and guest_msix_entries should have been
 373          * initialized */
 374         if (dev->entries_nr == 0)
 375                 return r;
 376
 377         r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
 378         if (r)
 379                 return r;
 380
 381         for (i = 0; i < dev->entries_nr; i++) {
 382                 r = request_irq(dev->host_msix_entries[i].vector,
 383                                 kvm_assigned_dev_intr, 0,
 384                                 "kvm_assigned_msix_device",
 385                                 (void *)dev);
 386                 /* FIXME: free requested_irq's on failure */
 387                 if (r)
 388                         return r;
 389         }
 390
 391         return 0;
 392 }
 393
 394 #endif
 395
 396 static int assigned_device_enable_guest_intx(struct kvm *kvm,
 397                                 struct kvm_assigned_dev_kernel *dev,
 398                                 struct kvm_assigned_irq *irq)
 399 {
 400         dev->guest_irq = irq->guest_irq;
 401         dev->ack_notifier.gsi = irq->guest_irq;
 402         return 0;
 403 }
 404
 405 #ifdef __KVM_HAVE_MSI
 406 static int assigned_device_enable_guest_msi(struct kvm *kvm,
 407                         struct kvm_assigned_dev_kernel *dev,
 408                         struct kvm_assigned_irq *irq)
 409 {
 410         dev->guest_irq = irq->guest_irq;
 411         dev->ack_notifier.gsi = -1;
 412         dev->host_irq_disabled = false;
 413         return 0;
 414 }
 415 #endif
 416 #ifdef __KVM_HAVE_MSIX
 417 static int assigned_device_enable_guest_msix(struct kvm *kvm,
 418                         struct kvm_assigned_dev_kernel *dev,
 419                         struct kvm_assigned_irq *irq)
 420 {
 421         dev->guest_irq = irq->guest_irq;
 422         dev->ack_notifier.gsi = -1;
 423         dev->host_irq_disabled = false;
 424         return 0;
 425 }
 426 #endif
 427
 428 static int assign_host_irq(struct kvm *kvm,
 429                            struct kvm_assigned_dev_kernel *dev,
 430                            __u32 host_irq_type)
 431 {
 432         int r = -EEXIST;
 433
 434         if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
 435                 return r;
 436
 437         switch (host_irq_type) {
 438         case KVM_DEV_IRQ_HOST_INTX:
 439                 r = assigned_device_enable_host_intx(kvm, dev);
 440                 break;
 441 #ifdef __KVM_HAVE_MSI
 442         case KVM_DEV_IRQ_HOST_MSI:
 443                 r = assigned_device_enable_host_msi(kvm, dev);
 444                 break;
 445 #endif
 446 #ifdef __KVM_HAVE_MSIX
 447         case KVM_DEV_IRQ_HOST_MSIX:
 448                 r = assigned_device_enable_host_msix(kvm, dev);
 449                 break;
 450 #endif
 451         default:
 452                 r = -EINVAL;
 453         }
 454
 455         if (!r)
 456                 dev->irq_requested_type |= host_irq_type;
 457
 458         return r;
 459 }
 460
 461 static int assign_guest_irq(struct kvm *kvm,
 462                             struct kvm_assigned_dev_kernel *dev,
 463                             struct kvm_assigned_irq *irq,
 464                             unsigned long guest_irq_type)
 465 {
 466         int id;
 467         int r = -EEXIST;
 468
 469         if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
 470                 return r;
 471
 472         id = kvm_request_irq_source_id(kvm);
 473         if (id < 0)
 474                 return id;
 475
 476         dev->irq_source_id = id;
 477
 478         switch (guest_irq_type) {
 479         case KVM_DEV_IRQ_GUEST_INTX:
 480                 r = assigned_device_enable_guest_intx(kvm, dev, irq);
 481                 break;
 482 #ifdef __KVM_HAVE_MSI
 483         case KVM_DEV_IRQ_GUEST_MSI:
 484                 r = assigned_device_enable_guest_msi(kvm, dev, irq);
 485                 break;
 486 #endif
 487 #ifdef __KVM_HAVE_MSIX
 488         case KVM_DEV_IRQ_GUEST_MSIX:
 489                 r = assigned_device_enable_guest_msix(kvm, dev, irq);
 490                 break;
 491 #endif
 492         default:
 493                 r = -EINVAL;
 494         }
 495
 496         if (!r) {
 497                 dev->irq_requested_type |= guest_irq_type;
 498                 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
 499         } else
 500                 kvm_free_irq_source_id(kvm, dev->irq_source_id);
 501
 502         return r;
 503 }
 504
 505 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
 506 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 507                                    struct kvm_assigned_irq *assigned_irq)
 508 {
 509         int r = -EINVAL;
 510         struct kvm_assigned_dev_kernel *match;
 511         unsigned long host_irq_type, guest_irq_type;
 512
 513         if (!capable(CAP_SYS_RAWIO))
 514                 return -EPERM;
 515
 516         if (!irqchip_in_kernel(kvm))
 517                 return r;
 518
 519         mutex_lock(&kvm->lock);
 520         r = -ENODEV;
 521         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 522                                       assigned_irq->assigned_dev_id);
 523         if (!match)
 524                 goto out;
 525
 526         host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
 527         guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
 528
 529         r = -EINVAL;
 530         /* can only assign one type at a time */
 531         if (hweight_long(host_irq_type) > 1)
 532                 goto out;
 533         if (hweight_long(guest_irq_type) > 1)
 534                 goto out;
 535         if (host_irq_type == 0 && guest_irq_type == 0)
 536                 goto out;
 537
 538         r = 0;
 539         if (host_irq_type)
 540                 r = assign_host_irq(kvm, match, host_irq_type);
 541         if (r)
 542                 goto out;
 543
 544         if (guest_irq_type)
 545                 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
 546 out:
 547         mutex_unlock(&kvm->lock);
 548         return r;
 549 }
 550
 551 static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
 552                                          struct kvm_assigned_irq
 553                                          *assigned_irq)
 554 {
 555         int r = -ENODEV;
 556         struct kvm_assigned_dev_kernel *match;
 557
 558         mutex_lock(&kvm->lock);
 559
 560         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 561                                       assigned_irq->assigned_dev_id);
 562         if (!match)
 563                 goto out;
 564
 565         r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
 566 out:
 567         mutex_unlock(&kvm->lock);
 568         return r;
 569 }
 570
 571 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 572                                       struct kvm_assigned_pci_dev *assigned_dev)
 573 {
 574         int r = 0;
 575         struct kvm_assigned_dev_kernel *match;
 576         struct pci_dev *dev;
 577
 578         down_read(&kvm->slots_lock);
 579         mutex_lock(&kvm->lock);
 580
 581         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 582                                       assigned_dev->assigned_dev_id);
 583         if (match) {
 584                 /* device already assigned */
 585                 r = -EEXIST;
 586                 goto out;
 587         }
 588
 589         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 590         if (match == NULL) {
 591                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 592                        __func__);
 593                 r = -ENOMEM;
 594                 goto out;
 595         }
 596         dev = pci_get_bus_and_slot(assigned_dev->busnr,
 597                                    assigned_dev->devfn);
 598         if (!dev) {
 599                 printk(KERN_INFO "%s: host device not found\n", __func__);
 600                 r = -EINVAL;
 601                 goto out_free;
 602         }
 603         if (pci_enable_device(dev)) {
 604                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 605                 r = -EBUSY;
 606                 goto out_put;
 607         }
 608         r = pci_request_regions(dev, "kvm_assigned_device");
 609         if (r) {
 610                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 611                        __func__);
 612                 goto out_disable;
 613         }
 614
 615         pci_reset_function(dev);
 616
 617         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 618         match->host_busnr = assigned_dev->busnr;
 619         match->host_devfn = assigned_dev->devfn;
 620         match->flags = assigned_dev->flags;
 621         match->dev = dev;
 622         spin_lock_init(&match->assigned_dev_lock);
 623         match->irq_source_id = -1;
 624         match->kvm = kvm;
 625         match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
 626         INIT_WORK(&match->interrupt_work,
 627                   kvm_assigned_dev_interrupt_work_handler);
 628
 629         list_add(&match->list, &kvm->arch.assigned_dev_head);
 630
 631         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
 632                 if (!kvm->arch.iommu_domain) {
 633                         r = kvm_iommu_map_guest(kvm);
 634                         if (r)
 635                                 goto out_list_del;
 636                 }
 637                 r = kvm_assign_device(kvm, match);
 638                 if (r)
 639                         goto out_list_del;
 640         }
 641
 642 out:
 643         mutex_unlock(&kvm->lock);
 644         up_read(&kvm->slots_lock);
 645         return r;
 646 out_list_del:
 647         list_del(&match->list);
 648         pci_release_regions(dev);
 649 out_disable:
 650         pci_disable_device(dev);
 651 out_put:
 652         pci_dev_put(dev);
 653 out_free:
 654         kfree(match);
 655         mutex_unlock(&kvm->lock);
 656         up_read(&kvm->slots_lock);
 657         return r;
 658 }
 659 #endif
 660
 661 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
 662 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
 663                 struct kvm_assigned_pci_dev *assigned_dev)
 664 {
 665         int r = 0;
 666         struct kvm_assigned_dev_kernel *match;
 667
 668         mutex_lock(&kvm->lock);
 669
 670         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 671                                       assigned_dev->assigned_dev_id);
 672         if (!match) {
 673                 printk(KERN_INFO "%s: device hasn't been assigned before, "
 674                   "so cannot be deassigned\n", __func__);
 675                 r = -EINVAL;
 676                 goto out;
 677         }
 678
 679         if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
 680                 kvm_deassign_device(kvm, match);
 681
 682         kvm_free_assigned_device(kvm, match);
 683
 684 out:
 685         mutex_unlock(&kvm->lock);
 686         return r;
 687 }
 688 #endif
 689
 690 static inline int valid_vcpu(int n)
 691 {
 692         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 693 }
 694
 695 inline int kvm_is_mmio_pfn(pfn_t pfn)
 696 {
 697         if (pfn_valid(pfn)) {
 698                 struct page *page = compound_head(pfn_to_page(pfn));
 699                 return PageReserved(page);
 700         }
 701
 702         return true;
 703 }
 704
 705 /*
 706  * Switches to specified vcpu, until a matching vcpu_put()
 707  */
 708 void vcpu_load(struct kvm_vcpu *vcpu)
 709 {
 710         int cpu;
 711
 712         mutex_lock(&vcpu->mutex);
 713         cpu = get_cpu();
 714         preempt_notifier_register(&vcpu->preempt_notifier);
 715         kvm_arch_vcpu_load(vcpu, cpu);
 716         put_cpu();
 717 }
 718
 719 void vcpu_put(struct kvm_vcpu *vcpu)
 720 {
 721         preempt_disable();
 722         kvm_arch_vcpu_put(vcpu);
 723         preempt_notifier_unregister(&vcpu->preempt_notifier);
 724         preempt_enable();
 725         mutex_unlock(&vcpu->mutex);
 726 }
 727
 728 static void ack_flush(void *_completed)
 729 {
 730 }
 731
 732 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 733 {
 734         int i, cpu, me;
 735         cpumask_var_t cpus;
 736         bool called = true;
 737         struct kvm_vcpu *vcpu;
 738
 739         if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
 740                 cpumask_clear(cpus);
 741
 742         me = get_cpu();
 743         spin_lock(&kvm->requests_lock);
 744         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 745                 vcpu = kvm->vcpus[i];
 746                 if (!vcpu)
 747                         continue;
 748                 if (test_and_set_bit(req, &vcpu->requests))
 749                         continue;
 750                 cpu = vcpu->cpu;
 751                 if (cpus != NULL && cpu != -1 && cpu != me)
 752                         cpumask_set_cpu(cpu, cpus);
 753         }
 754         if (unlikely(cpus == NULL))
 755                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 756         else if (!cpumask_empty(cpus))
 757                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 758         else
 759                 called = false;
 760         spin_unlock(&kvm->requests_lock);
 761         put_cpu();
 762         free_cpumask_var(cpus);
 763         return called;
 764 }
 765
 766 void kvm_flush_remote_tlbs(struct kvm *kvm)
 767 {
 768         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 769                 ++kvm->stat.remote_tlb_flush;
 770 }
 771
 772 void kvm_reload_remote_mmus(struct kvm *kvm)
 773 {
 774         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 775 }
 776
 777 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 778 {
 779         struct page *page;
 780         int r;
 781
 782         mutex_init(&vcpu->mutex);
 783         vcpu->cpu = -1;
 784         vcpu->kvm = kvm;
 785         vcpu->vcpu_id = id;
 786         init_waitqueue_head(&vcpu->wq);
 787
 788         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 789         if (!page) {
 790                 r = -ENOMEM;
 791                 goto fail;
 792         }
 793         vcpu->run = page_address(page);
 794
 795         r = kvm_arch_vcpu_init(vcpu);
 796         if (r < 0)
 797                 goto fail_free_run;
 798         return 0;
 799
 800 fail_free_run:
 801         free_page((unsigned long)vcpu->run);
 802 fail:
 803         return r;
 804 }
 805 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 806
 807 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 808 {
 809         kvm_arch_vcpu_uninit(vcpu);
 810         free_page((unsigned long)vcpu->run);
 811 }
 812 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 813
 814 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 815 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 816 {
 817         return container_of(mn, struct kvm, mmu_notifier);
 818 }
 819
 820 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 821                                              struct mm_struct *mm,
 822                                              unsigned long address)
 823 {
 824         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 825         int need_tlb_flush;
 826
 827         /*
 828          * When ->invalidate_page runs, the linux pte has been zapped
 829          * already but the page is still allocated until
 830          * ->invalidate_page returns. So if we increase the sequence
 831          * here the kvm page fault will notice if the spte can't be
 832          * established because the page is going to be freed. If
 833          * instead the kvm page fault establishes the spte before
 834          * ->invalidate_page runs, kvm_unmap_hva will release it
 835          * before returning.
 836          *
 837          * The sequence increase only need to be seen at spin_unlock
 838          * time, and not at spin_lock time.
 839          *
 840          * Increasing the sequence after the spin_unlock would be
 841          * unsafe because the kvm page fault could then establish the
 842          * pte after kvm_unmap_hva returned, without noticing the page
 843          * is going to be freed.
 844          */
 845         spin_lock(&kvm->mmu_lock);
 846         kvm->mmu_notifier_seq++;
 847         need_tlb_flush = kvm_unmap_hva(kvm, address);
 848         spin_unlock(&kvm->mmu_lock);
 849
 850         /* we've to flush the tlb before the pages can be freed */
 851         if (need_tlb_flush)
 852                 kvm_flush_remote_tlbs(kvm);
 853
 854 }
 855
 856 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 857                                                     struct mm_struct *mm,
 858                                                     unsigned long start,
 859                                                     unsigned long end)
 860 {
 861         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 862         int need_tlb_flush = 0;
 863
 864         spin_lock(&kvm->mmu_lock);
 865         /*
 866          * The count increase must become visible at unlock time as no
 867          * spte can be established without taking the mmu_lock and
 868          * count is also read inside the mmu_lock critical section.
 869          */
 870         kvm->mmu_notifier_count++;
 871         for (; start < end; start += PAGE_SIZE)
 872                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 873         spin_unlock(&kvm->mmu_lock);
 874
 875         /* we've to flush the tlb before the pages can be freed */
 876         if (need_tlb_flush)
 877                 kvm_flush_remote_tlbs(kvm);
 878 }
 879
 880 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 881                                                   struct mm_struct *mm,
 882                                                   unsigned long start,
 883                                                   unsigned long end)
 884 {
 885         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 886
 887         spin_lock(&kvm->mmu_lock);
 888         /*
 889          * This sequence increase will notify the kvm page fault that
 890          * the page that is going to be mapped in the spte could have
 891          * been freed.
 892          */
 893         kvm->mmu_notifier_seq++;
 894         /*
 895          * The above sequence increase must be visible before the
 896          * below count decrease but both values are read by the kvm
 897          * page fault under mmu_lock spinlock so we don't need to add
 898          * a smb_wmb() here in between the two.
 899          */
 900         kvm->mmu_notifier_count--;
 901         spin_unlock(&kvm->mmu_lock);
 902
 903         BUG_ON(kvm->mmu_notifier_count < 0);
 904 }
 905
 906 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 907                                               struct mm_struct *mm,
 908                                               unsigned long address)
 909 {
 910         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 911         int young;
 912
 913         spin_lock(&kvm->mmu_lock);
 914         young = kvm_age_hva(kvm, address);
 915         spin_unlock(&kvm->mmu_lock);
 916
 917         if (young)
 918                 kvm_flush_remote_tlbs(kvm);
 919
 920         return young;
 921 }
 922
 923 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 924                                      struct mm_struct *mm)
 925 {
 926         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 927         kvm_arch_flush_shadow(kvm);
 928 }
 929
 930 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 931         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 932         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 933         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 934         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 935         .release                = kvm_mmu_notifier_release,
 936 };
 937 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 938
 939 static struct kvm *kvm_create_vm(void)
 940 {
 941         struct kvm *kvm = kvm_arch_create_vm();
 942 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 943         struct page *page;
 944 #endif
 945
 946         if (IS_ERR(kvm))
 947                 goto out;
 948 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 949         INIT_LIST_HEAD(&kvm->irq_routing);
 950         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 951 #endif
 952
 953 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 954         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 955         if (!page) {
 956                 kfree(kvm);
 957                 return ERR_PTR(-ENOMEM);
 958         }
 959         kvm->coalesced_mmio_ring =
 960                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 961 #endif
 962
 963 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 964         {
 965                 int err;
 966                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 967                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 968                 if (err) {
 969 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 970                         put_page(page);
 971 #endif
 972                         kfree(kvm);
 973                         return ERR_PTR(err);
 974                 }
 975         }
 976 #endif
 977
 978         kvm->mm = current->mm;
 979         atomic_inc(&kvm->mm->mm_count);
 980         spin_lock_init(&kvm->mmu_lock);
 981         spin_lock_init(&kvm->requests_lock);
 982         kvm_io_bus_init(&kvm->pio_bus);
 983         kvm_irqfd_init(kvm);
 984         mutex_init(&kvm->lock);
 985         mutex_init(&kvm->irq_lock);
 986         kvm_io_bus_init(&kvm->mmio_bus);
 987         init_rwsem(&kvm->slots_lock);
 988         atomic_set(&kvm->users_count, 1);
 989         spin_lock(&kvm_lock);
 990         list_add(&kvm->vm_list, &vm_list);
 991         spin_unlock(&kvm_lock);
 992 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 993         kvm_coalesced_mmio_init(kvm);
 994 #endif
 995 out:
 996         return kvm;
 997 }
 998
 999 /*
1000  * Free any memory in @free but not in @dont.
1001  */
1002 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
1003                                   struct kvm_memory_slot *dont)
1004 {
1005         if (!dont || free->rmap != dont->rmap)
1006                 vfree(free->rmap);
1007
1008         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
1009                 vfree(free->dirty_bitmap);
1010
1011         if (!dont || free->lpage_info != dont->lpage_info)
1012                 vfree(free->lpage_info);
1013
1014         free->npages = 0;
1015         free->dirty_bitmap = NULL;
1016         free->rmap = NULL;
1017         free->lpage_info = NULL;
1018 }
1019
1020 void kvm_free_physmem(struct kvm *kvm)
1021 {
1022         int i;
1023
1024         for (i = 0; i < kvm->nmemslots; ++i)
1025                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
1026 }
1027
1028 static void kvm_destroy_vm(struct kvm *kvm)
1029 {
1030         struct mm_struct *mm = kvm->mm;
1031
1032         kvm_arch_sync_events(kvm);
1033         spin_lock(&kvm_lock);
1034         list_del(&kvm->vm_list);
1035         spin_unlock(&kvm_lock);
1036         kvm_free_irq_routing(kvm);
1037         kvm_io_bus_destroy(&kvm->pio_bus);
1038         kvm_io_bus_destroy(&kvm->mmio_bus);
1039 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1040         if (kvm->coalesced_mmio_ring != NULL)
1041                 free_page((unsigned long)kvm->coalesced_mmio_ring);
1042 #endif
1043 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1044         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1045 #else
1046         kvm_arch_flush_shadow(kvm);
1047 #endif
1048         kvm_arch_destroy_vm(kvm);
1049         mmdrop(mm);
1050 }
1051
1052 void kvm_get_kvm(struct kvm *kvm)
1053 {
1054         atomic_inc(&kvm->users_count);
1055 }
1056 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1057
1058 void kvm_put_kvm(struct kvm *kvm)
1059 {
1060         if (atomic_dec_and_test(&kvm->users_count))
1061                 kvm_destroy_vm(kvm);
1062 }
1063 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1064
1065
1066 static int kvm_vm_release(struct inode *inode, struct file *filp)
1067 {
1068         struct kvm *kvm = filp->private_data;
1069
1070         kvm_irqfd_release(kvm);
1071
1072         kvm_put_kvm(kvm);
1073         return 0;
1074 }
1075
1076 /*
1077  * Allocate some memory and give it an address in the guest physical address
1078  * space.
1079  *
1080  * Discontiguous memory is allowed, mostly for framebuffers.
1081  *
1082  * Must be called holding mmap_sem for write.
1083  */
1084 int __kvm_set_memory_region(struct kvm *kvm,
1085                             struct kvm_userspace_memory_region *mem,
1086                             int user_alloc)
1087 {
1088         int r;
1089         gfn_t base_gfn;
1090         unsigned long npages, ugfn;
1091         unsigned long largepages, i;
1092         struct kvm_memory_slot *memslot;
1093         struct kvm_memory_slot old, new;
1094
1095         r = -EINVAL;
1096         /* General sanity checks */
1097         if (mem->memory_size & (PAGE_SIZE - 1))
1098                 goto out;
1099         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1100                 goto out;
1101         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
1102                 goto out;
1103         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
1104                 goto out;
1105         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1106                 goto out;
1107
1108         memslot = &kvm->memslots[mem->slot];
1109         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1110         npages = mem->memory_size >> PAGE_SHIFT;
1111
1112         if (!npages)
1113                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1114
1115         new = old = *memslot;
1116
1117         new.base_gfn = base_gfn;
1118         new.npages = npages;
1119         new.flags = mem->flags;
1120
1121         /* Disallow changing a memory slot's size. */
1122         r = -EINVAL;
1123         if (npages && old.npages && npages != old.npages)
1124                 goto out_free;
1125
1126         /* Check for overlaps */
1127         r = -EEXIST;
1128         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1129                 struct kvm_memory_slot *s = &kvm->memslots[i];
1130
1131                 if (s == memslot || !s->npages)
1132                         continue;
1133                 if (!((base_gfn + npages <= s->base_gfn) ||
1134                       (base_gfn >= s->base_gfn + s->npages)))
1135                         goto out_free;
1136         }
1137
1138         /* Free page dirty bitmap if unneeded */
1139         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1140                 new.dirty_bitmap = NULL;
1141
1142         r = -ENOMEM;
1143
1144         /* Allocate if a slot is being created */
1145 #ifndef CONFIG_S390
1146         if (npages && !new.rmap) {
1147                 new.rmap = vmalloc(npages * sizeof(struct page *));
1148
1149                 if (!new.rmap)
1150                         goto out_free;
1151
1152                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1153
1154                 new.user_alloc = user_alloc;
1155                 /*
1156                  * hva_to_rmmap() serialzies with the mmu_lock and to be
1157                  * safe it has to ignore memslots with !user_alloc &&
1158                  * !userspace_addr.
1159                  */
1160                 if (user_alloc)
1161                         new.userspace_addr = mem->userspace_addr;
1162                 else
1163                         new.userspace_addr = 0;
1164         }
1165         if (npages && !new.lpage_info) {
1166                 largepages = 1 + (base_gfn + npages - 1) / KVM_PAGES_PER_HPAGE;
1167                 largepages -= base_gfn / KVM_PAGES_PER_HPAGE;
1168
1169                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1170
1171                 if (!new.lpage_info)
1172                         goto out_free;
1173
1174                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1175
1176                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1177                         new.lpage_info[0].write_count = 1;
1178                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1179                         new.lpage_info[largepages-1].write_count = 1;
1180                 ugfn = new.userspace_addr >> PAGE_SHIFT;
1181                 /*
1182                  * If the gfn and userspace address are not aligned wrt each
1183                  * other, disable large page support for this slot
1184                  */
1185                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE - 1))
1186                         for (i = 0; i < largepages; ++i)
1187                                 new.lpage_info[i].write_count = 1;
1188         }
1189
1190         /* Allocate page dirty bitmap if needed */
1191         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1192                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1193
1194                 new.dirty_bitmap = vmalloc(dirty_bytes);
1195                 if (!new.dirty_bitmap)
1196                         goto out_free;
1197                 memset(new.dirty_bitmap, 0, dirty_bytes);
1198                 if (old.npages)
1199                         kvm_arch_flush_shadow(kvm);
1200         }
1201 #endif /* not defined CONFIG_S390 */
1202
1203         if (!npages)
1204                 kvm_arch_flush_shadow(kvm);
1205
1206         spin_lock(&kvm->mmu_lock);
1207         if (mem->slot >= kvm->nmemslots)
1208                 kvm->nmemslots = mem->slot + 1;
1209
1210         *memslot = new;
1211         spin_unlock(&kvm->mmu_lock);
1212
1213         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1214         if (r) {
1215                 spin_lock(&kvm->mmu_lock);
1216                 *memslot = old;
1217                 spin_unlock(&kvm->mmu_lock);
1218                 goto out_free;
1219         }
1220
1221         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1222         /* Slot deletion case: we have to update the current slot */
1223         spin_lock(&kvm->mmu_lock);
1224         if (!npages)
1225                 *memslot = old;
1226         spin_unlock(&kvm->mmu_lock);
1227 #ifdef CONFIG_DMAR
1228         /* map the pages in iommu page table */
1229         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1230         if (r)
1231                 goto out;
1232 #endif
1233         return 0;
1234
1235 out_free:
1236         kvm_free_physmem_slot(&new, &old);
1237 out:
1238         return r;
1239
1240 }
1241 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1242
1243 int kvm_set_memory_region(struct kvm *kvm,
1244                           struct kvm_userspace_memory_region *mem,
1245                           int user_alloc)
1246 {
1247         int r;
1248
1249         down_write(&kvm->slots_lock);
1250         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1251         up_write(&kvm->slots_lock);
1252         return r;
1253 }
1254 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1255
1256 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1257                                    struct
1258                                    kvm_userspace_memory_region *mem,
1259                                    int user_alloc)
1260 {
1261         if (mem->slot >= KVM_MEMORY_SLOTS)
1262                 return -EINVAL;
1263         return kvm_set_memory_region(kvm, mem, user_alloc);
1264 }
1265
1266 int kvm_get_dirty_log(struct kvm *kvm,
1267                         struct kvm_dirty_log *log, int *is_dirty)
1268 {
1269         struct kvm_memory_slot *memslot;
1270         int r, i;
1271         int n;
1272         unsigned long any = 0;
1273
1274         r = -EINVAL;
1275         if (log->slot >= KVM_MEMORY_SLOTS)
1276                 goto out;
1277
1278         memslot = &kvm->memslots[log->slot];
1279         r = -ENOENT;
1280         if (!memslot->dirty_bitmap)
1281                 goto out;
1282
1283         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1284
1285         for (i = 0; !any && i < n/sizeof(long); ++i)
1286                 any = memslot->dirty_bitmap[i];
1287
1288         r = -EFAULT;
1289         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1290                 goto out;
1291
1292         if (any)
1293                 *is_dirty = 1;
1294
1295         r = 0;
1296 out:
1297         return r;
1298 }
1299
1300 int is_error_page(struct page *page)
1301 {
1302         return page == bad_page;
1303 }
1304 EXPORT_SYMBOL_GPL(is_error_page);
1305
1306 int is_error_pfn(pfn_t pfn)
1307 {
1308         return pfn == bad_pfn;
1309 }
1310 EXPORT_SYMBOL_GPL(is_error_pfn);
1311
1312 static inline unsigned long bad_hva(void)
1313 {
1314         return PAGE_OFFSET;
1315 }
1316
1317 int kvm_is_error_hva(unsigned long addr)
1318 {
1319         return addr == bad_hva();
1320 }
1321 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1322
1323 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1324 {
1325         int i;
1326
1327         for (i = 0; i < kvm->nmemslots; ++i) {
1328                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1329
1330                 if (gfn >= memslot->base_gfn
1331                     && gfn < memslot->base_gfn + memslot->npages)
1332                         return memslot;
1333         }
1334         return NULL;
1335 }
1336 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1337
1338 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1339 {
1340         gfn = unalias_gfn(kvm, gfn);
1341         return gfn_to_memslot_unaliased(kvm, gfn);
1342 }
1343
1344 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1345 {
1346         int i;
1347
1348         gfn = unalias_gfn(kvm, gfn);
1349         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1350                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1351
1352                 if (gfn >= memslot->base_gfn
1353                     && gfn < memslot->base_gfn + memslot->npages)
1354                         return 1;
1355         }
1356         return 0;
1357 }
1358 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1359
1360 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1361 {
1362         struct kvm_memory_slot *slot;
1363
1364         gfn = unalias_gfn(kvm, gfn);
1365         slot = gfn_to_memslot_unaliased(kvm, gfn);
1366         if (!slot)
1367                 return bad_hva();
1368         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1369 }
1370 EXPORT_SYMBOL_GPL(gfn_to_hva);
1371
1372 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1373 {
1374         struct page *page[1];
1375         unsigned long addr;
1376         int npages;
1377         pfn_t pfn;
1378
1379         might_sleep();
1380
1381         addr = gfn_to_hva(kvm, gfn);
1382         if (kvm_is_error_hva(addr)) {
1383                 get_page(bad_page);
1384                 return page_to_pfn(bad_page);
1385         }
1386
1387         npages = get_user_pages_fast(addr, 1, 1, page);
1388
1389         if (unlikely(npages != 1)) {
1390                 struct vm_area_struct *vma;
1391
1392                 down_read(&current->mm->mmap_sem);
1393                 vma = find_vma(current->mm, addr);
1394
1395                 if (vma == NULL || addr < vma->vm_start ||
1396                     !(vma->vm_flags & VM_PFNMAP)) {
1397                         up_read(&current->mm->mmap_sem);
1398                         get_page(bad_page);
1399                         return page_to_pfn(bad_page);
1400                 }
1401
1402                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1403                 up_read(&current->mm->mmap_sem);
1404                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1405         } else
1406                 pfn = page_to_pfn(page[0]);
1407
1408         return pfn;
1409 }
1410
1411 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1412
1413 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1414 {
1415         pfn_t pfn;
1416
1417         pfn = gfn_to_pfn(kvm, gfn);
1418         if (!kvm_is_mmio_pfn(pfn))
1419                 return pfn_to_page(pfn);
1420
1421         WARN_ON(kvm_is_mmio_pfn(pfn));
1422
1423         get_page(bad_page);
1424         return bad_page;
1425 }
1426
1427 EXPORT_SYMBOL_GPL(gfn_to_page);
1428
1429 void kvm_release_page_clean(struct page *page)
1430 {
1431         kvm_release_pfn_clean(page_to_pfn(page));
1432 }
1433 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1434
1435 void kvm_release_pfn_clean(pfn_t pfn)
1436 {
1437         if (!kvm_is_mmio_pfn(pfn))
1438                 put_page(pfn_to_page(pfn));
1439 }
1440 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1441
1442 void kvm_release_page_dirty(struct page *page)
1443 {
1444         kvm_release_pfn_dirty(page_to_pfn(page));
1445 }
1446 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1447
1448 void kvm_release_pfn_dirty(pfn_t pfn)
1449 {
1450         kvm_set_pfn_dirty(pfn);
1451         kvm_release_pfn_clean(pfn);
1452 }
1453 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1454
1455 void kvm_set_page_dirty(struct page *page)
1456 {
1457         kvm_set_pfn_dirty(page_to_pfn(page));
1458 }
1459 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1460
1461 void kvm_set_pfn_dirty(pfn_t pfn)
1462 {
1463         if (!kvm_is_mmio_pfn(pfn)) {
1464                 struct page *page = pfn_to_page(pfn);
1465                 if (!PageReserved(page))
1466                         SetPageDirty(page);
1467         }
1468 }
1469 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1470
1471 void kvm_set_pfn_accessed(pfn_t pfn)
1472 {
1473         if (!kvm_is_mmio_pfn(pfn))
1474                 mark_page_accessed(pfn_to_page(pfn));
1475 }
1476 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1477
1478 void kvm_get_pfn(pfn_t pfn)
1479 {
1480         if (!kvm_is_mmio_pfn(pfn))
1481                 get_page(pfn_to_page(pfn));
1482 }
1483 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1484
1485 static int next_segment(unsigned long len, int offset)
1486 {
1487         if (len > PAGE_SIZE - offset)
1488                 return PAGE_SIZE - offset;
1489         else
1490                 return len;
1491 }
1492
1493 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1494                         int len)
1495 {
1496         int r;
1497         unsigned long addr;
1498
1499         addr = gfn_to_hva(kvm, gfn);
1500         if (kvm_is_error_hva(addr))
1501                 return -EFAULT;
1502         r = copy_from_user(data, (void __user *)addr + offset, len);
1503         if (r)
1504                 return -EFAULT;
1505         return 0;
1506 }
1507 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1508
1509 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1510 {
1511         gfn_t gfn = gpa >> PAGE_SHIFT;
1512         int seg;
1513         int offset = offset_in_page(gpa);
1514         int ret;
1515
1516         while ((seg = next_segment(len, offset)) != 0) {
1517                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1518                 if (ret < 0)
1519                         return ret;
1520                 offset = 0;
1521                 len -= seg;
1522                 data += seg;
1523                 ++gfn;
1524         }
1525         return 0;
1526 }
1527 EXPORT_SYMBOL_GPL(kvm_read_guest);
1528
1529 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1530                           unsigned long len)
1531 {
1532         int r;
1533         unsigned long addr;
1534         gfn_t gfn = gpa >> PAGE_SHIFT;
1535         int offset = offset_in_page(gpa);
1536
1537         addr = gfn_to_hva(kvm, gfn);
1538         if (kvm_is_error_hva(addr))
1539                 return -EFAULT;
1540         pagefault_disable();
1541         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1542         pagefault_enable();
1543         if (r)
1544                 return -EFAULT;
1545         return 0;
1546 }
1547 EXPORT_SYMBOL(kvm_read_guest_atomic);
1548
1549 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1550                          int offset, int len)
1551 {
1552         int r;
1553         unsigned long addr;
1554
1555         addr = gfn_to_hva(kvm, gfn);
1556         if (kvm_is_error_hva(addr))
1557                 return -EFAULT;
1558         r = copy_to_user((void __user *)addr + offset, data, len);
1559         if (r)
1560                 return -EFAULT;
1561         mark_page_dirty(kvm, gfn);
1562         return 0;
1563 }
1564 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1565
1566 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1567                     unsigned long len)
1568 {
1569         gfn_t gfn = gpa >> PAGE_SHIFT;
1570         int seg;
1571         int offset = offset_in_page(gpa);
1572         int ret;
1573
1574         while ((seg = next_segment(len, offset)) != 0) {
1575                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1576                 if (ret < 0)
1577                         return ret;
1578                 offset = 0;
1579                 len -= seg;
1580                 data += seg;
1581                 ++gfn;
1582         }
1583         return 0;
1584 }
1585
1586 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1587 {
1588         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1589 }
1590 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1591
1592 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1593 {
1594         gfn_t gfn = gpa >> PAGE_SHIFT;
1595         int seg;
1596         int offset = offset_in_page(gpa);
1597         int ret;
1598
1599         while ((seg = next_segment(len, offset)) != 0) {
1600                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1601                 if (ret < 0)
1602                         return ret;
1603                 offset = 0;
1604                 len -= seg;
1605                 ++gfn;
1606         }
1607         return 0;
1608 }
1609 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1610
1611 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1612 {
1613         struct kvm_memory_slot *memslot;
1614
1615         gfn = unalias_gfn(kvm, gfn);
1616         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1617         if (memslot && memslot->dirty_bitmap) {
1618                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1619
1620                 /* avoid RMW */
1621                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1622                         set_bit(rel_gfn, memslot->dirty_bitmap);
1623         }
1624 }
1625
1626 /*
1627  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1628  */
1629 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1630 {
1631         DEFINE_WAIT(wait);
1632
1633         for (;;) {
1634                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1635
1636                 if ((kvm_arch_interrupt_allowed(vcpu) &&
1637                                         kvm_cpu_has_interrupt(vcpu)) ||
1638                                 kvm_arch_vcpu_runnable(vcpu)) {
1639                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1640                         break;
1641                 }
1642                 if (kvm_cpu_has_pending_timer(vcpu))
1643                         break;
1644                 if (signal_pending(current))
1645                         break;
1646
1647                 vcpu_put(vcpu);
1648                 schedule();
1649                 vcpu_load(vcpu);
1650         }
1651
1652         finish_wait(&vcpu->wq, &wait);
1653 }
1654
1655 void kvm_resched(struct kvm_vcpu *vcpu)
1656 {
1657         if (!need_resched())
1658                 return;
1659         cond_resched();
1660 }
1661 EXPORT_SYMBOL_GPL(kvm_resched);
1662
1663 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1664 {
1665         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1666         struct page *page;
1667
1668         if (vmf->pgoff == 0)
1669                 page = virt_to_page(vcpu->run);
1670 #ifdef CONFIG_X86
1671         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1672                 page = virt_to_page(vcpu->arch.pio_data);
1673 #endif
1674 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1675         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1676                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1677 #endif
1678         else
1679                 return VM_FAULT_SIGBUS;
1680         get_page(page);
1681         vmf->page = page;
1682         return 0;
1683 }
1684
1685 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1686         .fault = kvm_vcpu_fault,
1687 };
1688
1689 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1690 {
1691         vma->vm_ops = &kvm_vcpu_vm_ops;
1692         return 0;
1693 }
1694
1695 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1696 {
1697         struct kvm_vcpu *vcpu = filp->private_data;
1698
1699         kvm_put_kvm(vcpu->kvm);
1700         return 0;
1701 }
1702
1703 static struct file_operations kvm_vcpu_fops = {
1704         .release        = kvm_vcpu_release,
1705         .unlocked_ioctl = kvm_vcpu_ioctl,
1706         .compat_ioctl   = kvm_vcpu_ioctl,
1707         .mmap           = kvm_vcpu_mmap,
1708 };
1709
1710 /*
1711  * Allocates an inode for the vcpu.
1712  */
1713 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1714 {
1715         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1716         if (fd < 0)
1717                 kvm_put_kvm(vcpu->kvm);
1718         return fd;
1719 }
1720
1721 /*
1722  * Creates some virtual cpus.  Good luck creating more than one.
1723  */
1724 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1725 {
1726         int r;
1727         struct kvm_vcpu *vcpu;
1728
1729         if (!valid_vcpu(n))
1730                 return -EINVAL;
1731
1732         vcpu = kvm_arch_vcpu_create(kvm, n);
1733         if (IS_ERR(vcpu))
1734                 return PTR_ERR(vcpu);
1735
1736         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1737
1738         r = kvm_arch_vcpu_setup(vcpu);
1739         if (r)
1740                 return r;
1741
1742         mutex_lock(&kvm->lock);
1743         if (kvm->vcpus[n]) {
1744                 r = -EEXIST;
1745                 goto vcpu_destroy;
1746         }
1747         kvm->vcpus[n] = vcpu;
1748         mutex_unlock(&kvm->lock);
1749
1750         /* Now it's all set up, let userspace reach it */
1751         kvm_get_kvm(kvm);
1752         r = create_vcpu_fd(vcpu);
1753         if (r < 0)
1754                 goto unlink;
1755         return r;
1756
1757 unlink:
1758         mutex_lock(&kvm->lock);
1759         kvm->vcpus[n] = NULL;
1760 vcpu_destroy:
1761         mutex_unlock(&kvm->lock);
1762         kvm_arch_vcpu_destroy(vcpu);
1763         return r;
1764 }
1765
1766 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1767 {
1768         if (sigset) {
1769                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1770                 vcpu->sigset_active = 1;
1771                 vcpu->sigset = *sigset;
1772         } else
1773                 vcpu->sigset_active = 0;
1774         return 0;
1775 }
1776
1777 #ifdef __KVM_HAVE_MSIX
1778 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1779                                     struct kvm_assigned_msix_nr *entry_nr)
1780 {
1781         int r = 0;
1782         struct kvm_assigned_dev_kernel *adev;
1783
1784         mutex_lock(&kvm->lock);
1785
1786         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1787                                       entry_nr->assigned_dev_id);
1788         if (!adev) {
1789                 r = -EINVAL;
1790                 goto msix_nr_out;
1791         }
1792
1793         if (adev->entries_nr == 0) {
1794                 adev->entries_nr = entry_nr->entry_nr;
1795                 if (adev->entries_nr == 0 ||
1796                     adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1797                         r = -EINVAL;
1798                         goto msix_nr_out;
1799                 }
1800
1801                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1802                                                 entry_nr->entry_nr,
1803                                                 GFP_KERNEL);
1804                 if (!adev->host_msix_entries) {
1805                         r = -ENOMEM;
1806                         goto msix_nr_out;
1807                 }
1808                 adev->guest_msix_entries = kzalloc(
1809                                 sizeof(struct kvm_guest_msix_entry) *
1810                                 entry_nr->entry_nr, GFP_KERNEL);
1811                 if (!adev->guest_msix_entries) {
1812                         kfree(adev->host_msix_entries);
1813                         r = -ENOMEM;
1814                         goto msix_nr_out;
1815                 }
1816         } else /* Not allowed set MSI-X number twice */
1817                 r = -EINVAL;
1818 msix_nr_out:
1819         mutex_unlock(&kvm->lock);
1820         return r;
1821 }
1822
1823 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1824                                        struct kvm_assigned_msix_entry *entry)
1825 {
1826         int r = 0, i;
1827         struct kvm_assigned_dev_kernel *adev;
1828
1829         mutex_lock(&kvm->lock);
1830
1831         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1832                                       entry->assigned_dev_id);
1833
1834         if (!adev) {
1835                 r = -EINVAL;
1836                 goto msix_entry_out;
1837         }
1838
1839         for (i = 0; i < adev->entries_nr; i++)
1840                 if (adev->guest_msix_entries[i].vector == 0 ||
1841                     adev->guest_msix_entries[i].entry == entry->entry) {
1842                         adev->guest_msix_entries[i].entry = entry->entry;
1843                         adev->guest_msix_entries[i].vector = entry->gsi;
1844                         adev->host_msix_entries[i].entry = entry->entry;
1845                         break;
1846                 }
1847         if (i == adev->entries_nr) {
1848                 r = -ENOSPC;
1849                 goto msix_entry_out;
1850         }
1851
1852 msix_entry_out:
1853         mutex_unlock(&kvm->lock);
1854
1855         return r;
1856 }
1857 #endif
1858
1859 static long kvm_vcpu_ioctl(struct file *filp,
1860                            unsigned int ioctl, unsigned long arg)
1861 {
1862         struct kvm_vcpu *vcpu = filp->private_data;
1863         void __user *argp = (void __user *)arg;
1864         int r;
1865         struct kvm_fpu *fpu = NULL;
1866         struct kvm_sregs *kvm_sregs = NULL;
1867
1868         if (vcpu->kvm->mm != current->mm)
1869                 return -EIO;
1870         switch (ioctl) {
1871         case KVM_RUN:
1872                 r = -EINVAL;
1873                 if (arg)
1874                         goto out;
1875                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1876                 break;
1877         case KVM_GET_REGS: {
1878                 struct kvm_regs *kvm_regs;
1879
1880                 r = -ENOMEM;
1881                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1882                 if (!kvm_regs)
1883                         goto out;
1884                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1885                 if (r)
1886                         goto out_free1;
1887                 r = -EFAULT;
1888                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1889                         goto out_free1;
1890                 r = 0;
1891 out_free1:
1892                 kfree(kvm_regs);
1893                 break;
1894         }
1895         case KVM_SET_REGS: {
1896                 struct kvm_regs *kvm_regs;
1897
1898                 r = -ENOMEM;
1899                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1900                 if (!kvm_regs)
1901                         goto out;
1902                 r = -EFAULT;
1903                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1904                         goto out_free2;
1905                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1906                 if (r)
1907                         goto out_free2;
1908                 r = 0;
1909 out_free2:
1910                 kfree(kvm_regs);
1911                 break;
1912         }
1913         case KVM_GET_SREGS: {
1914                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1915                 r = -ENOMEM;
1916                 if (!kvm_sregs)
1917                         goto out;
1918                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1919                 if (r)
1920                         goto out;
1921                 r = -EFAULT;
1922                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1923                         goto out;
1924                 r = 0;
1925                 break;
1926         }
1927         case KVM_SET_SREGS: {
1928                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1929                 r = -ENOMEM;
1930                 if (!kvm_sregs)
1931                         goto out;
1932                 r = -EFAULT;
1933                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1934                         goto out;
1935                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1936                 if (r)
1937                         goto out;
1938                 r = 0;
1939                 break;
1940         }
1941         case KVM_GET_MP_STATE: {
1942                 struct kvm_mp_state mp_state;
1943
1944                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1945                 if (r)
1946                         goto out;
1947                 r = -EFAULT;
1948                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1949                         goto out;
1950                 r = 0;
1951                 break;
1952         }
1953         case KVM_SET_MP_STATE: {
1954                 struct kvm_mp_state mp_state;
1955
1956                 r = -EFAULT;
1957                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1958                         goto out;
1959                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1960                 if (r)
1961                         goto out;
1962                 r = 0;
1963                 break;
1964         }
1965         case KVM_TRANSLATE: {
1966                 struct kvm_translation tr;
1967
1968                 r = -EFAULT;
1969                 if (copy_from_user(&tr, argp, sizeof tr))
1970                         goto out;
1971                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1972                 if (r)
1973                         goto out;
1974                 r = -EFAULT;
1975                 if (copy_to_user(argp, &tr, sizeof tr))
1976                         goto out;
1977                 r = 0;
1978                 break;
1979         }
1980         case KVM_SET_GUEST_DEBUG: {
1981                 struct kvm_guest_debug dbg;
1982
1983                 r = -EFAULT;
1984                 if (copy_from_user(&dbg, argp, sizeof dbg))
1985                         goto out;
1986                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1987                 if (r)
1988                         goto out;
1989                 r = 0;
1990                 break;
1991         }
1992         case KVM_SET_SIGNAL_MASK: {
1993                 struct kvm_signal_mask __user *sigmask_arg = argp;
1994                 struct kvm_signal_mask kvm_sigmask;
1995                 sigset_t sigset, *p;
1996
1997                 p = NULL;
1998                 if (argp) {
1999                         r = -EFAULT;
2000                         if (copy_from_user(&kvm_sigmask, argp,
2001                                            sizeof kvm_sigmask))
2002                                 goto out;
2003                         r = -EINVAL;
2004                         if (kvm_sigmask.len != sizeof sigset)
2005                                 goto out;
2006                         r = -EFAULT;
2007                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2008                                            sizeof sigset))
2009                                 goto out;
2010                         p = &sigset;
2011                 }
2012                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2013                 break;
2014         }
2015         case KVM_GET_FPU: {
2016                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2017                 r = -ENOMEM;
2018                 if (!fpu)
2019                         goto out;
2020                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2021                 if (r)
2022                         goto out;
2023                 r = -EFAULT;
2024                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2025                         goto out;
2026                 r = 0;
2027                 break;
2028         }
2029         case KVM_SET_FPU: {
2030                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2031                 r = -ENOMEM;
2032                 if (!fpu)
2033                         goto out;
2034                 r = -EFAULT;
2035                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2036                         goto out;
2037                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2038                 if (r)
2039                         goto out;
2040                 r = 0;
2041                 break;
2042         }
2043         default:
2044                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2045         }
2046 out:
2047         kfree(fpu);
2048         kfree(kvm_sregs);
2049         return r;
2050 }
2051
2052 static long kvm_vm_ioctl(struct file *filp,
2053                            unsigned int ioctl, unsigned long arg)
2054 {
2055         struct kvm *kvm = filp->private_data;
2056         void __user *argp = (void __user *)arg;
2057         int r;
2058
2059         if (kvm->mm != current->mm)
2060                 return -EIO;
2061         switch (ioctl) {
2062         case KVM_CREATE_VCPU:
2063                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2064                 if (r < 0)
2065                         goto out;
2066                 break;
2067         case KVM_SET_USER_MEMORY_REGION: {
2068                 struct kvm_userspace_memory_region kvm_userspace_mem;
2069
2070                 r = -EFAULT;
2071                 if (copy_from_user(&kvm_userspace_mem, argp,
2072                                                 sizeof kvm_userspace_mem))
2073                         goto out;
2074
2075                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2076                 if (r)
2077                         goto out;
2078                 break;
2079         }
2080         case KVM_GET_DIRTY_LOG: {
2081                 struct kvm_dirty_log log;
2082
2083                 r = -EFAULT;
2084                 if (copy_from_user(&log, argp, sizeof log))
2085                         goto out;
2086                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2087                 if (r)
2088                         goto out;
2089                 break;
2090         }
2091 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2092         case KVM_REGISTER_COALESCED_MMIO: {
2093                 struct kvm_coalesced_mmio_zone zone;
2094                 r = -EFAULT;
2095                 if (copy_from_user(&zone, argp, sizeof zone))
2096                         goto out;
2097                 r = -ENXIO;
2098                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2099                 if (r)
2100                         goto out;
2101                 r = 0;
2102                 break;
2103         }
2104         case KVM_UNREGISTER_COALESCED_MMIO: {
2105                 struct kvm_coalesced_mmio_zone zone;
2106                 r = -EFAULT;
2107                 if (copy_from_user(&zone, argp, sizeof zone))
2108                         goto out;
2109                 r = -ENXIO;
2110                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2111                 if (r)
2112                         goto out;
2113                 r = 0;
2114                 break;
2115         }
2116 #endif
2117 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2118         case KVM_ASSIGN_PCI_DEVICE: {
2119                 struct kvm_assigned_pci_dev assigned_dev;
2120
2121                 r = -EFAULT;
2122                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2123                         goto out;
2124                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2125                 if (r)
2126                         goto out;
2127                 break;
2128         }
2129         case KVM_ASSIGN_IRQ: {
2130                 r = -EOPNOTSUPP;
2131                 break;
2132         }
2133 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2134         case KVM_ASSIGN_DEV_IRQ: {
2135                 struct kvm_assigned_irq assigned_irq;
2136
2137                 r = -EFAULT;
2138                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2139                         goto out;
2140                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2141                 if (r)
2142                         goto out;
2143                 break;
2144         }
2145         case KVM_DEASSIGN_DEV_IRQ: {
2146                 struct kvm_assigned_irq assigned_irq;
2147
2148                 r = -EFAULT;
2149                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2150                         goto out;
2151                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2152                 if (r)
2153                         goto out;
2154                 break;
2155         }
2156 #endif
2157 #endif
2158 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2159         case KVM_DEASSIGN_PCI_DEVICE: {
2160                 struct kvm_assigned_pci_dev assigned_dev;
2161
2162                 r = -EFAULT;
2163                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2164                         goto out;
2165                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2166                 if (r)
2167                         goto out;
2168                 break;
2169         }
2170 #endif
2171 #ifdef KVM_CAP_IRQ_ROUTING
2172         case KVM_SET_GSI_ROUTING: {
2173                 struct kvm_irq_routing routing;
2174                 struct kvm_irq_routing __user *urouting;
2175                 struct kvm_irq_routing_entry *entries;
2176
2177                 r = -EFAULT;
2178                 if (copy_from_user(&routing, argp, sizeof(routing)))
2179                         goto out;
2180                 r = -EINVAL;
2181                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2182                         goto out;
2183                 if (routing.flags)
2184                         goto out;
2185                 r = -ENOMEM;
2186                 entries = vmalloc(routing.nr * sizeof(*entries));
2187                 if (!entries)
2188                         goto out;
2189                 r = -EFAULT;
2190                 urouting = argp;
2191                 if (copy_from_user(entries, urouting->entries,
2192                                    routing.nr * sizeof(*entries)))
2193                         goto out_free_irq_routing;
2194                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2195                                         routing.flags);
2196         out_free_irq_routing:
2197                 vfree(entries);
2198                 break;
2199         }
2200 #ifdef __KVM_HAVE_MSIX
2201         case KVM_ASSIGN_SET_MSIX_NR: {
2202                 struct kvm_assigned_msix_nr entry_nr;
2203                 r = -EFAULT;
2204                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2205                         goto out;
2206                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2207                 if (r)
2208                         goto out;
2209                 break;
2210         }
2211         case KVM_ASSIGN_SET_MSIX_ENTRY: {
2212                 struct kvm_assigned_msix_entry entry;
2213                 r = -EFAULT;
2214                 if (copy_from_user(&entry, argp, sizeof entry))
2215                         goto out;
2216                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2217                 if (r)
2218                         goto out;
2219                 break;
2220         }
2221 #endif
2222 #endif /* KVM_CAP_IRQ_ROUTING */
2223         case KVM_IRQFD: {
2224                 struct kvm_irqfd data;
2225
2226                 r = -EFAULT;
2227                 if (copy_from_user(&data, argp, sizeof data))
2228                         goto out;
2229                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2230                 break;
2231         }
2232         default:
2233                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2234         }
2235 out:
2236         return r;
2237 }
2238
2239 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2240 {
2241         struct page *page[1];
2242         unsigned long addr;
2243         int npages;
2244         gfn_t gfn = vmf->pgoff;
2245         struct kvm *kvm = vma->vm_file->private_data;
2246
2247         addr = gfn_to_hva(kvm, gfn);
2248         if (kvm_is_error_hva(addr))
2249                 return VM_FAULT_SIGBUS;
2250
2251         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2252                                 NULL);
2253         if (unlikely(npages != 1))
2254                 return VM_FAULT_SIGBUS;
2255
2256         vmf->page = page[0];
2257         return 0;
2258 }
2259
2260 static struct vm_operations_struct kvm_vm_vm_ops = {
2261         .fault = kvm_vm_fault,
2262 };
2263
2264 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2265 {
2266         vma->vm_ops = &kvm_vm_vm_ops;
2267         return 0;
2268 }
2269
2270 static struct file_operations kvm_vm_fops = {
2271         .release        = kvm_vm_release,
2272         .unlocked_ioctl = kvm_vm_ioctl,
2273         .compat_ioctl   = kvm_vm_ioctl,
2274         .mmap           = kvm_vm_mmap,
2275 };
2276
2277 static int kvm_dev_ioctl_create_vm(void)
2278 {
2279         int fd;
2280         struct kvm *kvm;
2281
2282         kvm = kvm_create_vm();
2283         if (IS_ERR(kvm))
2284                 return PTR_ERR(kvm);
2285         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2286         if (fd < 0)
2287                 kvm_put_kvm(kvm);
2288
2289         return fd;
2290 }
2291
2292 static long kvm_dev_ioctl_check_extension_generic(long arg)
2293 {
2294         switch (arg) {
2295         case KVM_CAP_USER_MEMORY:
2296         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2297         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2298                 return 1;
2299 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2300         case KVM_CAP_IRQ_ROUTING:
2301                 return KVM_MAX_IRQ_ROUTES;
2302 #endif
2303         default:
2304                 break;
2305         }
2306         return kvm_dev_ioctl_check_extension(arg);
2307 }
2308
2309 static long kvm_dev_ioctl(struct file *filp,
2310                           unsigned int ioctl, unsigned long arg)
2311 {
2312         long r = -EINVAL;
2313
2314         switch (ioctl) {
2315         case KVM_GET_API_VERSION:
2316                 r = -EINVAL;
2317                 if (arg)
2318                         goto out;
2319                 r = KVM_API_VERSION;
2320                 break;
2321         case KVM_CREATE_VM:
2322                 r = -EINVAL;
2323                 if (arg)
2324                         goto out;
2325                 r = kvm_dev_ioctl_create_vm();
2326                 break;
2327         case KVM_CHECK_EXTENSION:
2328                 r = kvm_dev_ioctl_check_extension_generic(arg);
2329                 break;
2330         case KVM_GET_VCPU_MMAP_SIZE:
2331                 r = -EINVAL;
2332                 if (arg)
2333                         goto out;
2334                 r = PAGE_SIZE;     /* struct kvm_run */
2335 #ifdef CONFIG_X86
2336                 r += PAGE_SIZE;    /* pio data page */
2337 #endif
2338 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2339                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2340 #endif
2341                 break;
2342         case KVM_TRACE_ENABLE:
2343         case KVM_TRACE_PAUSE:
2344         case KVM_TRACE_DISABLE:
2345                 r = kvm_trace_ioctl(ioctl, arg);
2346                 break;
2347         default:
2348                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2349         }
2350 out:
2351         return r;
2352 }
2353
2354 static struct file_operations kvm_chardev_ops = {
2355         .unlocked_ioctl = kvm_dev_ioctl,
2356         .compat_ioctl   = kvm_dev_ioctl,
2357 };
2358
2359 static struct miscdevice kvm_dev = {
2360         KVM_MINOR,
2361         "kvm",
2362         &kvm_chardev_ops,
2363 };
2364
2365 static void hardware_enable(void *junk)
2366 {
2367         int cpu = raw_smp_processor_id();
2368
2369         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2370                 return;
2371         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2372         kvm_arch_hardware_enable(NULL);
2373 }
2374
2375 static void hardware_disable(void *junk)
2376 {
2377         int cpu = raw_smp_processor_id();
2378
2379         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2380                 return;
2381         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2382         kvm_arch_hardware_disable(NULL);
2383 }
2384
2385 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2386                            void *v)
2387 {
2388         int cpu = (long)v;
2389
2390         val &= ~CPU_TASKS_FROZEN;
2391         switch (val) {
2392         case CPU_DYING:
2393                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2394                        cpu);
2395                 hardware_disable(NULL);
2396                 break;
2397         case CPU_UP_CANCELED:
2398                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2399                        cpu);
2400                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2401                 break;
2402         case CPU_ONLINE:
2403                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2404                        cpu);
2405                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2406                 break;
2407         }
2408         return NOTIFY_OK;
2409 }
2410
2411
2412 asmlinkage void kvm_handle_fault_on_reboot(void)
2413 {
2414         if (kvm_rebooting)
2415                 /* spin while reset goes on */
2416                 while (true)
2417                         ;
2418         /* Fault while not rebooting.  We want the trace. */
2419         BUG();
2420 }
2421 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2422
2423 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2424                       void *v)
2425 {
2426         /*
2427          * Some (well, at least mine) BIOSes hang on reboot if
2428          * in vmx root mode.
2429          *
2430          * And Intel TXT required VMX off for all cpu when system shutdown.
2431          */
2432         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2433         kvm_rebooting = true;
2434         on_each_cpu(hardware_disable, NULL, 1);
2435         return NOTIFY_OK;
2436 }
2437
2438 static struct notifier_block kvm_reboot_notifier = {
2439         .notifier_call = kvm_reboot,
2440         .priority = 0,
2441 };
2442
2443 void kvm_io_bus_init(struct kvm_io_bus *bus)
2444 {
2445         memset(bus, 0, sizeof(*bus));
2446 }
2447
2448 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2449 {
2450         int i;
2451
2452         for (i = 0; i < bus->dev_count; i++) {
2453                 struct kvm_io_device *pos = bus->devs[i];
2454
2455                 kvm_iodevice_destructor(pos);
2456         }
2457 }
2458
2459 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2460                                           gpa_t addr, int len, int is_write)
2461 {
2462         int i;
2463
2464         for (i = 0; i < bus->dev_count; i++) {
2465                 struct kvm_io_device *pos = bus->devs[i];
2466
2467                 if (kvm_iodevice_in_range(pos, addr, len, is_write))
2468                         return pos;
2469         }
2470
2471         return NULL;
2472 }
2473
2474 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2475 {
2476         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2477
2478         bus->devs[bus->dev_count++] = dev;
2479 }
2480
2481 static struct notifier_block kvm_cpu_notifier = {
2482         .notifier_call = kvm_cpu_hotplug,
2483         .priority = 20, /* must be > scheduler priority */
2484 };
2485
2486 static int vm_stat_get(void *_offset, u64 *val)
2487 {
2488         unsigned offset = (long)_offset;
2489         struct kvm *kvm;
2490
2491         *val = 0;
2492         spin_lock(&kvm_lock);
2493         list_for_each_entry(kvm, &vm_list, vm_list)
2494                 *val += *(u32 *)((void *)kvm + offset);
2495         spin_unlock(&kvm_lock);
2496         return 0;
2497 }
2498
2499 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2500
2501 static int vcpu_stat_get(void *_offset, u64 *val)
2502 {
2503         unsigned offset = (long)_offset;
2504         struct kvm *kvm;
2505         struct kvm_vcpu *vcpu;
2506         int i;
2507
2508         *val = 0;
2509         spin_lock(&kvm_lock);
2510         list_for_each_entry(kvm, &vm_list, vm_list)
2511                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2512                         vcpu = kvm->vcpus[i];
2513                         if (vcpu)
2514                                 *val += *(u32 *)((void *)vcpu + offset);
2515                 }
2516         spin_unlock(&kvm_lock);
2517         return 0;
2518 }
2519
2520 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2521
2522 static struct file_operations *stat_fops[] = {
2523         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2524         [KVM_STAT_VM]   = &vm_stat_fops,
2525 };
2526
2527 static void kvm_init_debug(void)
2528 {
2529         struct kvm_stats_debugfs_item *p;
2530
2531         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2532         for (p = debugfs_entries; p->name; ++p)
2533                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2534                                                 (void *)(long)p->offset,
2535                                                 stat_fops[p->kind]);
2536 }
2537
2538 static void kvm_exit_debug(void)
2539 {
2540         struct kvm_stats_debugfs_item *p;
2541
2542         for (p = debugfs_entries; p->name; ++p)
2543                 debugfs_remove(p->dentry);
2544         debugfs_remove(kvm_debugfs_dir);
2545 }
2546
2547 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2548 {
2549         hardware_disable(NULL);
2550         return 0;
2551 }
2552
2553 static int kvm_resume(struct sys_device *dev)
2554 {
2555         hardware_enable(NULL);
2556         return 0;
2557 }
2558
2559 static struct sysdev_class kvm_sysdev_class = {
2560         .name = "kvm",
2561         .suspend = kvm_suspend,
2562         .resume = kvm_resume,
2563 };
2564
2565 static struct sys_device kvm_sysdev = {
2566         .id = 0,
2567         .cls = &kvm_sysdev_class,
2568 };
2569
2570 struct page *bad_page;
2571 pfn_t bad_pfn;
2572
2573 static inline
2574 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2575 {
2576         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2577 }
2578
2579 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2580 {
2581         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2582
2583         kvm_arch_vcpu_load(vcpu, cpu);
2584 }
2585
2586 static void kvm_sched_out(struct preempt_notifier *pn,
2587                           struct task_struct *next)
2588 {
2589         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2590
2591         kvm_arch_vcpu_put(vcpu);
2592 }
2593
2594 int kvm_init(void *opaque, unsigned int vcpu_size,
2595                   struct module *module)
2596 {
2597         int r;
2598         int cpu;
2599
2600         kvm_init_debug();
2601
2602         r = kvm_arch_init(opaque);
2603         if (r)
2604                 goto out_fail;
2605
2606         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2607
2608         if (bad_page == NULL) {
2609                 r = -ENOMEM;
2610                 goto out;
2611         }
2612
2613         bad_pfn = page_to_pfn(bad_page);
2614
2615         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2616                 r = -ENOMEM;
2617                 goto out_free_0;
2618         }
2619
2620         r = kvm_arch_hardware_setup();
2621         if (r < 0)
2622                 goto out_free_0a;
2623
2624         for_each_online_cpu(cpu) {
2625                 smp_call_function_single(cpu,
2626                                 kvm_arch_check_processor_compat,
2627                                 &r, 1);
2628                 if (r < 0)
2629                         goto out_free_1;
2630         }
2631
2632         on_each_cpu(hardware_enable, NULL, 1);
2633         r = register_cpu_notifier(&kvm_cpu_notifier);
2634         if (r)
2635                 goto out_free_2;
2636         register_reboot_notifier(&kvm_reboot_notifier);
2637
2638         r = sysdev_class_register(&kvm_sysdev_class);
2639         if (r)
2640                 goto out_free_3;
2641
2642         r = sysdev_register(&kvm_sysdev);
2643         if (r)
2644                 goto out_free_4;
2645
2646         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2647         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2648                                            __alignof__(struct kvm_vcpu),
2649                                            0, NULL);
2650         if (!kvm_vcpu_cache) {
2651                 r = -ENOMEM;
2652                 goto out_free_5;
2653         }
2654
2655         kvm_chardev_ops.owner = module;
2656         kvm_vm_fops.owner = module;
2657         kvm_vcpu_fops.owner = module;
2658
2659         r = misc_register(&kvm_dev);
2660         if (r) {
2661                 printk(KERN_ERR "kvm: misc device register failed\n");
2662                 goto out_free;
2663         }
2664
2665         kvm_preempt_ops.sched_in = kvm_sched_in;
2666         kvm_preempt_ops.sched_out = kvm_sched_out;
2667
2668         return 0;
2669
2670 out_free:
2671         kmem_cache_destroy(kvm_vcpu_cache);
2672 out_free_5:
2673         sysdev_unregister(&kvm_sysdev);
2674 out_free_4:
2675         sysdev_class_unregister(&kvm_sysdev_class);
2676 out_free_3:
2677         unregister_reboot_notifier(&kvm_reboot_notifier);
2678         unregister_cpu_notifier(&kvm_cpu_notifier);
2679 out_free_2:
2680         on_each_cpu(hardware_disable, NULL, 1);
2681 out_free_1:
2682         kvm_arch_hardware_unsetup();
2683 out_free_0a:
2684         free_cpumask_var(cpus_hardware_enabled);
2685 out_free_0:
2686         __free_page(bad_page);
2687 out:
2688         kvm_arch_exit();
2689         kvm_exit_debug();
2690 out_fail:
2691         return r;
2692 }
2693 EXPORT_SYMBOL_GPL(kvm_init);
2694
2695 void kvm_exit(void)
2696 {
2697         kvm_trace_cleanup();
2698         misc_deregister(&kvm_dev);
2699         kmem_cache_destroy(kvm_vcpu_cache);
2700         sysdev_unregister(&kvm_sysdev);
2701         sysdev_class_unregister(&kvm_sysdev_class);
2702         unregister_reboot_notifier(&kvm_reboot_notifier);
2703         unregister_cpu_notifier(&kvm_cpu_notifier);
2704         on_each_cpu(hardware_disable, NULL, 1);
2705         kvm_arch_hardware_unsetup();
2706         kvm_arch_exit();
2707         kvm_exit_debug();
2708         free_cpumask_var(cpus_hardware_enabled);
2709         __free_page(bad_page);
2710 }
2711 EXPORT_SYMBOL_GPL(kvm_exit);