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

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "iodev.h"
  19
  20 #include <linux/kvm_host.h>
  21 #include <linux/kvm.h>
  22 #include <linux/module.h>
  23 #include <linux/errno.h>
  24 #include <linux/percpu.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/sysdev.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #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         kvm_io_bus_init(&kvm->mmio_bus);
 986         init_rwsem(&kvm->slots_lock);
 987         atomic_set(&kvm->users_count, 1);
 988         spin_lock(&kvm_lock);
 989         list_add(&kvm->vm_list, &vm_list);
 990         spin_unlock(&kvm_lock);
 991 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 992         kvm_coalesced_mmio_init(kvm);
 993 #endif
 994 out:
 995         return kvm;
 996 }
 997
 998 /*
 999  * Free any memory in @free but not in @dont.
1000  */
1001 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
1002                                   struct kvm_memory_slot *dont)
1003 {
1004         if (!dont || free->rmap != dont->rmap)
1005                 vfree(free->rmap);
1006
1007         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
1008                 vfree(free->dirty_bitmap);
1009
1010         if (!dont || free->lpage_info != dont->lpage_info)
1011                 vfree(free->lpage_info);
1012
1013         free->npages = 0;
1014         free->dirty_bitmap = NULL;
1015         free->rmap = NULL;
1016         free->lpage_info = NULL;
1017 }
1018
1019 void kvm_free_physmem(struct kvm *kvm)
1020 {
1021         int i;
1022
1023         for (i = 0; i < kvm->nmemslots; ++i)
1024                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
1025 }
1026
1027 static void kvm_destroy_vm(struct kvm *kvm)
1028 {
1029         struct mm_struct *mm = kvm->mm;
1030
1031         kvm_arch_sync_events(kvm);
1032         spin_lock(&kvm_lock);
1033         list_del(&kvm->vm_list);
1034         spin_unlock(&kvm_lock);
1035         kvm_free_irq_routing(kvm);
1036         kvm_io_bus_destroy(&kvm->pio_bus);
1037         kvm_io_bus_destroy(&kvm->mmio_bus);
1038 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1039         if (kvm->coalesced_mmio_ring != NULL)
1040                 free_page((unsigned long)kvm->coalesced_mmio_ring);
1041 #endif
1042 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1043         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1044 #else
1045         kvm_arch_flush_shadow(kvm);
1046 #endif
1047         kvm_arch_destroy_vm(kvm);
1048         mmdrop(mm);
1049 }
1050
1051 void kvm_get_kvm(struct kvm *kvm)
1052 {
1053         atomic_inc(&kvm->users_count);
1054 }
1055 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1056
1057 void kvm_put_kvm(struct kvm *kvm)
1058 {
1059         if (atomic_dec_and_test(&kvm->users_count))
1060                 kvm_destroy_vm(kvm);
1061 }
1062 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1063
1064
1065 static int kvm_vm_release(struct inode *inode, struct file *filp)
1066 {
1067         struct kvm *kvm = filp->private_data;
1068
1069         kvm_irqfd_release(kvm);
1070
1071         kvm_put_kvm(kvm);
1072         return 0;
1073 }
1074
1075 /*
1076  * Allocate some memory and give it an address in the guest physical address
1077  * space.
1078  *
1079  * Discontiguous memory is allowed, mostly for framebuffers.
1080  *
1081  * Must be called holding mmap_sem for write.
1082  */
1083 int __kvm_set_memory_region(struct kvm *kvm,
1084                             struct kvm_userspace_memory_region *mem,
1085                             int user_alloc)
1086 {
1087         int r;
1088         gfn_t base_gfn;
1089         unsigned long npages, ugfn;
1090         unsigned long largepages, i;
1091         struct kvm_memory_slot *memslot;
1092         struct kvm_memory_slot old, new;
1093
1094         r = -EINVAL;
1095         /* General sanity checks */
1096         if (mem->memory_size & (PAGE_SIZE - 1))
1097                 goto out;
1098         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1099                 goto out;
1100         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
1101                 goto out;
1102         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
1103                 goto out;
1104         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1105                 goto out;
1106
1107         memslot = &kvm->memslots[mem->slot];
1108         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1109         npages = mem->memory_size >> PAGE_SHIFT;
1110
1111         if (!npages)
1112                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1113
1114         new = old = *memslot;
1115
1116         new.base_gfn = base_gfn;
1117         new.npages = npages;
1118         new.flags = mem->flags;
1119
1120         /* Disallow changing a memory slot's size. */
1121         r = -EINVAL;
1122         if (npages && old.npages && npages != old.npages)
1123                 goto out_free;
1124
1125         /* Check for overlaps */
1126         r = -EEXIST;
1127         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1128                 struct kvm_memory_slot *s = &kvm->memslots[i];
1129
1130                 if (s == memslot || !s->npages)
1131                         continue;
1132                 if (!((base_gfn + npages <= s->base_gfn) ||
1133                       (base_gfn >= s->base_gfn + s->npages)))
1134                         goto out_free;
1135         }
1136
1137         /* Free page dirty bitmap if unneeded */
1138         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1139                 new.dirty_bitmap = NULL;
1140
1141         r = -ENOMEM;
1142
1143         /* Allocate if a slot is being created */
1144 #ifndef CONFIG_S390
1145         if (npages && !new.rmap) {
1146                 new.rmap = vmalloc(npages * sizeof(struct page *));
1147
1148                 if (!new.rmap)
1149                         goto out_free;
1150
1151                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1152
1153                 new.user_alloc = user_alloc;
1154                 /*
1155                  * hva_to_rmmap() serialzies with the mmu_lock and to be
1156                  * safe it has to ignore memslots with !user_alloc &&
1157                  * !userspace_addr.
1158                  */
1159                 if (user_alloc)
1160                         new.userspace_addr = mem->userspace_addr;
1161                 else
1162                         new.userspace_addr = 0;
1163         }
1164         if (npages && !new.lpage_info) {
1165                 largepages = 1 + (base_gfn + npages - 1) / KVM_PAGES_PER_HPAGE;
1166                 largepages -= base_gfn / KVM_PAGES_PER_HPAGE;
1167
1168                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1169
1170                 if (!new.lpage_info)
1171                         goto out_free;
1172
1173                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1174
1175                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1176                         new.lpage_info[0].write_count = 1;
1177                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1178                         new.lpage_info[largepages-1].write_count = 1;
1179                 ugfn = new.userspace_addr >> PAGE_SHIFT;
1180                 /*
1181                  * If the gfn and userspace address are not aligned wrt each
1182                  * other, disable large page support for this slot
1183                  */
1184                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE - 1))
1185                         for (i = 0; i < largepages; ++i)
1186                                 new.lpage_info[i].write_count = 1;
1187         }
1188
1189         /* Allocate page dirty bitmap if needed */
1190         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1191                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1192
1193                 new.dirty_bitmap = vmalloc(dirty_bytes);
1194                 if (!new.dirty_bitmap)
1195                         goto out_free;
1196                 memset(new.dirty_bitmap, 0, dirty_bytes);
1197                 if (old.npages)
1198                         kvm_arch_flush_shadow(kvm);
1199         }
1200 #endif /* not defined CONFIG_S390 */
1201
1202         if (!npages)
1203                 kvm_arch_flush_shadow(kvm);
1204
1205         spin_lock(&kvm->mmu_lock);
1206         if (mem->slot >= kvm->nmemslots)
1207                 kvm->nmemslots = mem->slot + 1;
1208
1209         *memslot = new;
1210         spin_unlock(&kvm->mmu_lock);
1211
1212         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1213         if (r) {
1214                 spin_lock(&kvm->mmu_lock);
1215                 *memslot = old;
1216                 spin_unlock(&kvm->mmu_lock);
1217                 goto out_free;
1218         }
1219
1220         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1221         /* Slot deletion case: we have to update the current slot */
1222         spin_lock(&kvm->mmu_lock);
1223         if (!npages)
1224                 *memslot = old;
1225         spin_unlock(&kvm->mmu_lock);
1226 #ifdef CONFIG_DMAR
1227         /* map the pages in iommu page table */
1228         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1229         if (r)
1230                 goto out;
1231 #endif
1232         return 0;
1233
1234 out_free:
1235         kvm_free_physmem_slot(&new, &old);
1236 out:
1237         return r;
1238
1239 }
1240 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1241
1242 int kvm_set_memory_region(struct kvm *kvm,
1243                           struct kvm_userspace_memory_region *mem,
1244                           int user_alloc)
1245 {
1246         int r;
1247
1248         down_write(&kvm->slots_lock);
1249         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1250         up_write(&kvm->slots_lock);
1251         return r;
1252 }
1253 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1254
1255 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1256                                    struct
1257                                    kvm_userspace_memory_region *mem,
1258                                    int user_alloc)
1259 {
1260         if (mem->slot >= KVM_MEMORY_SLOTS)
1261                 return -EINVAL;
1262         return kvm_set_memory_region(kvm, mem, user_alloc);
1263 }
1264
1265 int kvm_get_dirty_log(struct kvm *kvm,
1266                         struct kvm_dirty_log *log, int *is_dirty)
1267 {
1268         struct kvm_memory_slot *memslot;
1269         int r, i;
1270         int n;
1271         unsigned long any = 0;
1272
1273         r = -EINVAL;
1274         if (log->slot >= KVM_MEMORY_SLOTS)
1275                 goto out;
1276
1277         memslot = &kvm->memslots[log->slot];
1278         r = -ENOENT;
1279         if (!memslot->dirty_bitmap)
1280                 goto out;
1281
1282         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1283
1284         for (i = 0; !any && i < n/sizeof(long); ++i)
1285                 any = memslot->dirty_bitmap[i];
1286
1287         r = -EFAULT;
1288         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1289                 goto out;
1290
1291         if (any)
1292                 *is_dirty = 1;
1293
1294         r = 0;
1295 out:
1296         return r;
1297 }
1298
1299 int is_error_page(struct page *page)
1300 {
1301         return page == bad_page;
1302 }
1303 EXPORT_SYMBOL_GPL(is_error_page);
1304
1305 int is_error_pfn(pfn_t pfn)
1306 {
1307         return pfn == bad_pfn;
1308 }
1309 EXPORT_SYMBOL_GPL(is_error_pfn);
1310
1311 static inline unsigned long bad_hva(void)
1312 {
1313         return PAGE_OFFSET;
1314 }
1315
1316 int kvm_is_error_hva(unsigned long addr)
1317 {
1318         return addr == bad_hva();
1319 }
1320 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1321
1322 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1323 {
1324         int i;
1325
1326         for (i = 0; i < kvm->nmemslots; ++i) {
1327                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1328
1329                 if (gfn >= memslot->base_gfn
1330                     && gfn < memslot->base_gfn + memslot->npages)
1331                         return memslot;
1332         }
1333         return NULL;
1334 }
1335 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1336
1337 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1338 {
1339         gfn = unalias_gfn(kvm, gfn);
1340         return gfn_to_memslot_unaliased(kvm, gfn);
1341 }
1342
1343 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1344 {
1345         int i;
1346
1347         gfn = unalias_gfn(kvm, gfn);
1348         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1349                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1350
1351                 if (gfn >= memslot->base_gfn
1352                     && gfn < memslot->base_gfn + memslot->npages)
1353                         return 1;
1354         }
1355         return 0;
1356 }
1357 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1358
1359 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1360 {
1361         struct kvm_memory_slot *slot;
1362
1363         gfn = unalias_gfn(kvm, gfn);
1364         slot = gfn_to_memslot_unaliased(kvm, gfn);
1365         if (!slot)
1366                 return bad_hva();
1367         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1368 }
1369 EXPORT_SYMBOL_GPL(gfn_to_hva);
1370
1371 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1372 {
1373         struct page *page[1];
1374         unsigned long addr;
1375         int npages;
1376         pfn_t pfn;
1377
1378         might_sleep();
1379
1380         addr = gfn_to_hva(kvm, gfn);
1381         if (kvm_is_error_hva(addr)) {
1382                 get_page(bad_page);
1383                 return page_to_pfn(bad_page);
1384         }
1385
1386         npages = get_user_pages_fast(addr, 1, 1, page);
1387
1388         if (unlikely(npages != 1)) {
1389                 struct vm_area_struct *vma;
1390
1391                 down_read(&current->mm->mmap_sem);
1392                 vma = find_vma(current->mm, addr);
1393
1394                 if (vma == NULL || addr < vma->vm_start ||
1395                     !(vma->vm_flags & VM_PFNMAP)) {
1396                         up_read(&current->mm->mmap_sem);
1397                         get_page(bad_page);
1398                         return page_to_pfn(bad_page);
1399                 }
1400
1401                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1402                 up_read(&current->mm->mmap_sem);
1403                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1404         } else
1405                 pfn = page_to_pfn(page[0]);
1406
1407         return pfn;
1408 }
1409
1410 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1411
1412 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1413 {
1414         pfn_t pfn;
1415
1416         pfn = gfn_to_pfn(kvm, gfn);
1417         if (!kvm_is_mmio_pfn(pfn))
1418                 return pfn_to_page(pfn);
1419
1420         WARN_ON(kvm_is_mmio_pfn(pfn));
1421
1422         get_page(bad_page);
1423         return bad_page;
1424 }
1425
1426 EXPORT_SYMBOL_GPL(gfn_to_page);
1427
1428 void kvm_release_page_clean(struct page *page)
1429 {
1430         kvm_release_pfn_clean(page_to_pfn(page));
1431 }
1432 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1433
1434 void kvm_release_pfn_clean(pfn_t pfn)
1435 {
1436         if (!kvm_is_mmio_pfn(pfn))
1437                 put_page(pfn_to_page(pfn));
1438 }
1439 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1440
1441 void kvm_release_page_dirty(struct page *page)
1442 {
1443         kvm_release_pfn_dirty(page_to_pfn(page));
1444 }
1445 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1446
1447 void kvm_release_pfn_dirty(pfn_t pfn)
1448 {
1449         kvm_set_pfn_dirty(pfn);
1450         kvm_release_pfn_clean(pfn);
1451 }
1452 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1453
1454 void kvm_set_page_dirty(struct page *page)
1455 {
1456         kvm_set_pfn_dirty(page_to_pfn(page));
1457 }
1458 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1459
1460 void kvm_set_pfn_dirty(pfn_t pfn)
1461 {
1462         if (!kvm_is_mmio_pfn(pfn)) {
1463                 struct page *page = pfn_to_page(pfn);
1464                 if (!PageReserved(page))
1465                         SetPageDirty(page);
1466         }
1467 }
1468 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1469
1470 void kvm_set_pfn_accessed(pfn_t pfn)
1471 {
1472         if (!kvm_is_mmio_pfn(pfn))
1473                 mark_page_accessed(pfn_to_page(pfn));
1474 }
1475 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1476
1477 void kvm_get_pfn(pfn_t pfn)
1478 {
1479         if (!kvm_is_mmio_pfn(pfn))
1480                 get_page(pfn_to_page(pfn));
1481 }
1482 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1483
1484 static int next_segment(unsigned long len, int offset)
1485 {
1486         if (len > PAGE_SIZE - offset)
1487                 return PAGE_SIZE - offset;
1488         else
1489                 return len;
1490 }
1491
1492 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1493                         int len)
1494 {
1495         int r;
1496         unsigned long addr;
1497
1498         addr = gfn_to_hva(kvm, gfn);
1499         if (kvm_is_error_hva(addr))
1500                 return -EFAULT;
1501         r = copy_from_user(data, (void __user *)addr + offset, len);
1502         if (r)
1503                 return -EFAULT;
1504         return 0;
1505 }
1506 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1507
1508 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1509 {
1510         gfn_t gfn = gpa >> PAGE_SHIFT;
1511         int seg;
1512         int offset = offset_in_page(gpa);
1513         int ret;
1514
1515         while ((seg = next_segment(len, offset)) != 0) {
1516                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1517                 if (ret < 0)
1518                         return ret;
1519                 offset = 0;
1520                 len -= seg;
1521                 data += seg;
1522                 ++gfn;
1523         }
1524         return 0;
1525 }
1526 EXPORT_SYMBOL_GPL(kvm_read_guest);
1527
1528 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1529                           unsigned long len)
1530 {
1531         int r;
1532         unsigned long addr;
1533         gfn_t gfn = gpa >> PAGE_SHIFT;
1534         int offset = offset_in_page(gpa);
1535
1536         addr = gfn_to_hva(kvm, gfn);
1537         if (kvm_is_error_hva(addr))
1538                 return -EFAULT;
1539         pagefault_disable();
1540         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1541         pagefault_enable();
1542         if (r)
1543                 return -EFAULT;
1544         return 0;
1545 }
1546 EXPORT_SYMBOL(kvm_read_guest_atomic);
1547
1548 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1549                          int offset, int len)
1550 {
1551         int r;
1552         unsigned long addr;
1553
1554         addr = gfn_to_hva(kvm, gfn);
1555         if (kvm_is_error_hva(addr))
1556                 return -EFAULT;
1557         r = copy_to_user((void __user *)addr + offset, data, len);
1558         if (r)
1559                 return -EFAULT;
1560         mark_page_dirty(kvm, gfn);
1561         return 0;
1562 }
1563 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1564
1565 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1566                     unsigned long len)
1567 {
1568         gfn_t gfn = gpa >> PAGE_SHIFT;
1569         int seg;
1570         int offset = offset_in_page(gpa);
1571         int ret;
1572
1573         while ((seg = next_segment(len, offset)) != 0) {
1574                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1575                 if (ret < 0)
1576                         return ret;
1577                 offset = 0;
1578                 len -= seg;
1579                 data += seg;
1580                 ++gfn;
1581         }
1582         return 0;
1583 }
1584
1585 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1586 {
1587         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1588 }
1589 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1590
1591 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1592 {
1593         gfn_t gfn = gpa >> PAGE_SHIFT;
1594         int seg;
1595         int offset = offset_in_page(gpa);
1596         int ret;
1597
1598         while ((seg = next_segment(len, offset)) != 0) {
1599                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1600                 if (ret < 0)
1601                         return ret;
1602                 offset = 0;
1603                 len -= seg;
1604                 ++gfn;
1605         }
1606         return 0;
1607 }
1608 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1609
1610 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1611 {
1612         struct kvm_memory_slot *memslot;
1613
1614         gfn = unalias_gfn(kvm, gfn);
1615         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1616         if (memslot && memslot->dirty_bitmap) {
1617                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1618
1619                 /* avoid RMW */
1620                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1621                         set_bit(rel_gfn, memslot->dirty_bitmap);
1622         }
1623 }
1624
1625 /*
1626  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1627  */
1628 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1629 {
1630         DEFINE_WAIT(wait);
1631
1632         for (;;) {
1633                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1634
1635                 if ((kvm_arch_interrupt_allowed(vcpu) &&
1636                                         kvm_cpu_has_interrupt(vcpu)) ||
1637                                 kvm_arch_vcpu_runnable(vcpu)) {
1638                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1639                         break;
1640                 }
1641                 if (kvm_cpu_has_pending_timer(vcpu))
1642                         break;
1643                 if (signal_pending(current))
1644                         break;
1645
1646                 vcpu_put(vcpu);
1647                 schedule();
1648                 vcpu_load(vcpu);
1649         }
1650
1651         finish_wait(&vcpu->wq, &wait);
1652 }
1653
1654 void kvm_resched(struct kvm_vcpu *vcpu)
1655 {
1656         if (!need_resched())
1657                 return;
1658         cond_resched();
1659 }
1660 EXPORT_SYMBOL_GPL(kvm_resched);
1661
1662 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1663 {
1664         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1665         struct page *page;
1666
1667         if (vmf->pgoff == 0)
1668                 page = virt_to_page(vcpu->run);
1669 #ifdef CONFIG_X86
1670         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1671                 page = virt_to_page(vcpu->arch.pio_data);
1672 #endif
1673 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1674         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1675                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1676 #endif
1677         else
1678                 return VM_FAULT_SIGBUS;
1679         get_page(page);
1680         vmf->page = page;
1681         return 0;
1682 }
1683
1684 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1685         .fault = kvm_vcpu_fault,
1686 };
1687
1688 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1689 {
1690         vma->vm_ops = &kvm_vcpu_vm_ops;
1691         return 0;
1692 }
1693
1694 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1695 {
1696         struct kvm_vcpu *vcpu = filp->private_data;
1697
1698         kvm_put_kvm(vcpu->kvm);
1699         return 0;
1700 }
1701
1702 static struct file_operations kvm_vcpu_fops = {
1703         .release        = kvm_vcpu_release,
1704         .unlocked_ioctl = kvm_vcpu_ioctl,
1705         .compat_ioctl   = kvm_vcpu_ioctl,
1706         .mmap           = kvm_vcpu_mmap,
1707 };
1708
1709 /*
1710  * Allocates an inode for the vcpu.
1711  */
1712 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1713 {
1714         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1715         if (fd < 0)
1716                 kvm_put_kvm(vcpu->kvm);
1717         return fd;
1718 }
1719
1720 /*
1721  * Creates some virtual cpus.  Good luck creating more than one.
1722  */
1723 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1724 {
1725         int r;
1726         struct kvm_vcpu *vcpu;
1727
1728         if (!valid_vcpu(n))
1729                 return -EINVAL;
1730
1731         vcpu = kvm_arch_vcpu_create(kvm, n);
1732         if (IS_ERR(vcpu))
1733                 return PTR_ERR(vcpu);
1734
1735         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1736
1737         r = kvm_arch_vcpu_setup(vcpu);
1738         if (r)
1739                 return r;
1740
1741         mutex_lock(&kvm->lock);
1742         if (kvm->vcpus[n]) {
1743                 r = -EEXIST;
1744                 goto vcpu_destroy;
1745         }
1746         kvm->vcpus[n] = vcpu;
1747         mutex_unlock(&kvm->lock);
1748
1749         /* Now it's all set up, let userspace reach it */
1750         kvm_get_kvm(kvm);
1751         r = create_vcpu_fd(vcpu);
1752         if (r < 0)
1753                 goto unlink;
1754         return r;
1755
1756 unlink:
1757         mutex_lock(&kvm->lock);
1758         kvm->vcpus[n] = NULL;
1759 vcpu_destroy:
1760         mutex_unlock(&kvm->lock);
1761         kvm_arch_vcpu_destroy(vcpu);
1762         return r;
1763 }
1764
1765 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1766 {
1767         if (sigset) {
1768                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1769                 vcpu->sigset_active = 1;
1770                 vcpu->sigset = *sigset;
1771         } else
1772                 vcpu->sigset_active = 0;
1773         return 0;
1774 }
1775
1776 #ifdef __KVM_HAVE_MSIX
1777 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1778                                     struct kvm_assigned_msix_nr *entry_nr)
1779 {
1780         int r = 0;
1781         struct kvm_assigned_dev_kernel *adev;
1782
1783         mutex_lock(&kvm->lock);
1784
1785         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1786                                       entry_nr->assigned_dev_id);
1787         if (!adev) {
1788                 r = -EINVAL;
1789                 goto msix_nr_out;
1790         }
1791
1792         if (adev->entries_nr == 0) {
1793                 adev->entries_nr = entry_nr->entry_nr;
1794                 if (adev->entries_nr == 0 ||
1795                     adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1796                         r = -EINVAL;
1797                         goto msix_nr_out;
1798                 }
1799
1800                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1801                                                 entry_nr->entry_nr,
1802                                                 GFP_KERNEL);
1803                 if (!adev->host_msix_entries) {
1804                         r = -ENOMEM;
1805                         goto msix_nr_out;
1806                 }
1807                 adev->guest_msix_entries = kzalloc(
1808                                 sizeof(struct kvm_guest_msix_entry) *
1809                                 entry_nr->entry_nr, GFP_KERNEL);
1810                 if (!adev->guest_msix_entries) {
1811                         kfree(adev->host_msix_entries);
1812                         r = -ENOMEM;
1813                         goto msix_nr_out;
1814                 }
1815         } else /* Not allowed set MSI-X number twice */
1816                 r = -EINVAL;
1817 msix_nr_out:
1818         mutex_unlock(&kvm->lock);
1819         return r;
1820 }
1821
1822 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1823                                        struct kvm_assigned_msix_entry *entry)
1824 {
1825         int r = 0, i;
1826         struct kvm_assigned_dev_kernel *adev;
1827
1828         mutex_lock(&kvm->lock);
1829
1830         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1831                                       entry->assigned_dev_id);
1832
1833         if (!adev) {
1834                 r = -EINVAL;
1835                 goto msix_entry_out;
1836         }
1837
1838         for (i = 0; i < adev->entries_nr; i++)
1839                 if (adev->guest_msix_entries[i].vector == 0 ||
1840                     adev->guest_msix_entries[i].entry == entry->entry) {
1841                         adev->guest_msix_entries[i].entry = entry->entry;
1842                         adev->guest_msix_entries[i].vector = entry->gsi;
1843                         adev->host_msix_entries[i].entry = entry->entry;
1844                         break;
1845                 }
1846         if (i == adev->entries_nr) {
1847                 r = -ENOSPC;
1848                 goto msix_entry_out;
1849         }
1850
1851 msix_entry_out:
1852         mutex_unlock(&kvm->lock);
1853
1854         return r;
1855 }
1856 #endif
1857
1858 static long kvm_vcpu_ioctl(struct file *filp,
1859                            unsigned int ioctl, unsigned long arg)
1860 {
1861         struct kvm_vcpu *vcpu = filp->private_data;
1862         void __user *argp = (void __user *)arg;
1863         int r;
1864         struct kvm_fpu *fpu = NULL;
1865         struct kvm_sregs *kvm_sregs = NULL;
1866
1867         if (vcpu->kvm->mm != current->mm)
1868                 return -EIO;
1869         switch (ioctl) {
1870         case KVM_RUN:
1871                 r = -EINVAL;
1872                 if (arg)
1873                         goto out;
1874                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1875                 break;
1876         case KVM_GET_REGS: {
1877                 struct kvm_regs *kvm_regs;
1878
1879                 r = -ENOMEM;
1880                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1881                 if (!kvm_regs)
1882                         goto out;
1883                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1884                 if (r)
1885                         goto out_free1;
1886                 r = -EFAULT;
1887                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1888                         goto out_free1;
1889                 r = 0;
1890 out_free1:
1891                 kfree(kvm_regs);
1892                 break;
1893         }
1894         case KVM_SET_REGS: {
1895                 struct kvm_regs *kvm_regs;
1896
1897                 r = -ENOMEM;
1898                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1899                 if (!kvm_regs)
1900                         goto out;
1901                 r = -EFAULT;
1902                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1903                         goto out_free2;
1904                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1905                 if (r)
1906                         goto out_free2;
1907                 r = 0;
1908 out_free2:
1909                 kfree(kvm_regs);
1910                 break;
1911         }
1912         case KVM_GET_SREGS: {
1913                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1914                 r = -ENOMEM;
1915                 if (!kvm_sregs)
1916                         goto out;
1917                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1918                 if (r)
1919                         goto out;
1920                 r = -EFAULT;
1921                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1922                         goto out;
1923                 r = 0;
1924                 break;
1925         }
1926         case KVM_SET_SREGS: {
1927                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1928                 r = -ENOMEM;
1929                 if (!kvm_sregs)
1930                         goto out;
1931                 r = -EFAULT;
1932                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1933                         goto out;
1934                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1935                 if (r)
1936                         goto out;
1937                 r = 0;
1938                 break;
1939         }
1940         case KVM_GET_MP_STATE: {
1941                 struct kvm_mp_state mp_state;
1942
1943                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1944                 if (r)
1945                         goto out;
1946                 r = -EFAULT;
1947                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1948                         goto out;
1949                 r = 0;
1950                 break;
1951         }
1952         case KVM_SET_MP_STATE: {
1953                 struct kvm_mp_state mp_state;
1954
1955                 r = -EFAULT;
1956                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1957                         goto out;
1958                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1959                 if (r)
1960                         goto out;
1961                 r = 0;
1962                 break;
1963         }
1964         case KVM_TRANSLATE: {
1965                 struct kvm_translation tr;
1966
1967                 r = -EFAULT;
1968                 if (copy_from_user(&tr, argp, sizeof tr))
1969                         goto out;
1970                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1971                 if (r)
1972                         goto out;
1973                 r = -EFAULT;
1974                 if (copy_to_user(argp, &tr, sizeof tr))
1975                         goto out;
1976                 r = 0;
1977                 break;
1978         }
1979         case KVM_SET_GUEST_DEBUG: {
1980                 struct kvm_guest_debug dbg;
1981
1982                 r = -EFAULT;
1983                 if (copy_from_user(&dbg, argp, sizeof dbg))
1984                         goto out;
1985                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1986                 if (r)
1987                         goto out;
1988                 r = 0;
1989                 break;
1990         }
1991         case KVM_SET_SIGNAL_MASK: {
1992                 struct kvm_signal_mask __user *sigmask_arg = argp;
1993                 struct kvm_signal_mask kvm_sigmask;
1994                 sigset_t sigset, *p;
1995
1996                 p = NULL;
1997                 if (argp) {
1998                         r = -EFAULT;
1999                         if (copy_from_user(&kvm_sigmask, argp,
2000                                            sizeof kvm_sigmask))
2001                                 goto out;
2002                         r = -EINVAL;
2003                         if (kvm_sigmask.len != sizeof sigset)
2004                                 goto out;
2005                         r = -EFAULT;
2006                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2007                                            sizeof sigset))
2008                                 goto out;
2009                         p = &sigset;
2010                 }
2011                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2012                 break;
2013         }
2014         case KVM_GET_FPU: {
2015                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2016                 r = -ENOMEM;
2017                 if (!fpu)
2018                         goto out;
2019                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2020                 if (r)
2021                         goto out;
2022                 r = -EFAULT;
2023                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2024                         goto out;
2025                 r = 0;
2026                 break;
2027         }
2028         case KVM_SET_FPU: {
2029                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2030                 r = -ENOMEM;
2031                 if (!fpu)
2032                         goto out;
2033                 r = -EFAULT;
2034                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2035                         goto out;
2036                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2037                 if (r)
2038                         goto out;
2039                 r = 0;
2040                 break;
2041         }
2042         default:
2043                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2044         }
2045 out:
2046         kfree(fpu);
2047         kfree(kvm_sregs);
2048         return r;
2049 }
2050
2051 static long kvm_vm_ioctl(struct file *filp,
2052                            unsigned int ioctl, unsigned long arg)
2053 {
2054         struct kvm *kvm = filp->private_data;
2055         void __user *argp = (void __user *)arg;
2056         int r;
2057
2058         if (kvm->mm != current->mm)
2059                 return -EIO;
2060         switch (ioctl) {
2061         case KVM_CREATE_VCPU:
2062                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2063                 if (r < 0)
2064                         goto out;
2065                 break;
2066         case KVM_SET_USER_MEMORY_REGION: {
2067                 struct kvm_userspace_memory_region kvm_userspace_mem;
2068
2069                 r = -EFAULT;
2070                 if (copy_from_user(&kvm_userspace_mem, argp,
2071                                                 sizeof kvm_userspace_mem))
2072                         goto out;
2073
2074                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2075                 if (r)
2076                         goto out;
2077                 break;
2078         }
2079         case KVM_GET_DIRTY_LOG: {
2080                 struct kvm_dirty_log log;
2081
2082                 r = -EFAULT;
2083                 if (copy_from_user(&log, argp, sizeof log))
2084                         goto out;
2085                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2086                 if (r)
2087                         goto out;
2088                 break;
2089         }
2090 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2091         case KVM_REGISTER_COALESCED_MMIO: {
2092                 struct kvm_coalesced_mmio_zone zone;
2093                 r = -EFAULT;
2094                 if (copy_from_user(&zone, argp, sizeof zone))
2095                         goto out;
2096                 r = -ENXIO;
2097                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2098                 if (r)
2099                         goto out;
2100                 r = 0;
2101                 break;
2102         }
2103         case KVM_UNREGISTER_COALESCED_MMIO: {
2104                 struct kvm_coalesced_mmio_zone zone;
2105                 r = -EFAULT;
2106                 if (copy_from_user(&zone, argp, sizeof zone))
2107                         goto out;
2108                 r = -ENXIO;
2109                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2110                 if (r)
2111                         goto out;
2112                 r = 0;
2113                 break;
2114         }
2115 #endif
2116 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2117         case KVM_ASSIGN_PCI_DEVICE: {
2118                 struct kvm_assigned_pci_dev assigned_dev;
2119
2120                 r = -EFAULT;
2121                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2122                         goto out;
2123                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2124                 if (r)
2125                         goto out;
2126                 break;
2127         }
2128         case KVM_ASSIGN_IRQ: {
2129                 r = -EOPNOTSUPP;
2130                 break;
2131         }
2132 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2133         case KVM_ASSIGN_DEV_IRQ: {
2134                 struct kvm_assigned_irq assigned_irq;
2135
2136                 r = -EFAULT;
2137                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2138                         goto out;
2139                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2140                 if (r)
2141                         goto out;
2142                 break;
2143         }
2144         case KVM_DEASSIGN_DEV_IRQ: {
2145                 struct kvm_assigned_irq assigned_irq;
2146
2147                 r = -EFAULT;
2148                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2149                         goto out;
2150                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2151                 if (r)
2152                         goto out;
2153                 break;
2154         }
2155 #endif
2156 #endif
2157 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2158         case KVM_DEASSIGN_PCI_DEVICE: {
2159                 struct kvm_assigned_pci_dev assigned_dev;
2160
2161                 r = -EFAULT;
2162                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2163                         goto out;
2164                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2165                 if (r)
2166                         goto out;
2167                 break;
2168         }
2169 #endif
2170 #ifdef KVM_CAP_IRQ_ROUTING
2171         case KVM_SET_GSI_ROUTING: {
2172                 struct kvm_irq_routing routing;
2173                 struct kvm_irq_routing __user *urouting;
2174                 struct kvm_irq_routing_entry *entries;
2175
2176                 r = -EFAULT;
2177                 if (copy_from_user(&routing, argp, sizeof(routing)))
2178                         goto out;
2179                 r = -EINVAL;
2180                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2181                         goto out;
2182                 if (routing.flags)
2183                         goto out;
2184                 r = -ENOMEM;
2185                 entries = vmalloc(routing.nr * sizeof(*entries));
2186                 if (!entries)
2187                         goto out;
2188                 r = -EFAULT;
2189                 urouting = argp;
2190                 if (copy_from_user(entries, urouting->entries,
2191                                    routing.nr * sizeof(*entries)))
2192                         goto out_free_irq_routing;
2193                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2194                                         routing.flags);
2195         out_free_irq_routing:
2196                 vfree(entries);
2197                 break;
2198         }
2199 #ifdef __KVM_HAVE_MSIX
2200         case KVM_ASSIGN_SET_MSIX_NR: {
2201                 struct kvm_assigned_msix_nr entry_nr;
2202                 r = -EFAULT;
2203                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2204                         goto out;
2205                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2206                 if (r)
2207                         goto out;
2208                 break;
2209         }
2210         case KVM_ASSIGN_SET_MSIX_ENTRY: {
2211                 struct kvm_assigned_msix_entry entry;
2212                 r = -EFAULT;
2213                 if (copy_from_user(&entry, argp, sizeof entry))
2214                         goto out;
2215                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2216                 if (r)
2217                         goto out;
2218                 break;
2219         }
2220 #endif
2221 #endif /* KVM_CAP_IRQ_ROUTING */
2222         case KVM_IRQFD: {
2223                 struct kvm_irqfd data;
2224
2225                 r = -EFAULT;
2226                 if (copy_from_user(&data, argp, sizeof data))
2227                         goto out;
2228                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2229                 break;
2230         }
2231         default:
2232                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2233         }
2234 out:
2235         return r;
2236 }
2237
2238 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2239 {
2240         struct page *page[1];
2241         unsigned long addr;
2242         int npages;
2243         gfn_t gfn = vmf->pgoff;
2244         struct kvm *kvm = vma->vm_file->private_data;
2245
2246         addr = gfn_to_hva(kvm, gfn);
2247         if (kvm_is_error_hva(addr))
2248                 return VM_FAULT_SIGBUS;
2249
2250         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2251                                 NULL);
2252         if (unlikely(npages != 1))
2253                 return VM_FAULT_SIGBUS;
2254
2255         vmf->page = page[0];
2256         return 0;
2257 }
2258
2259 static struct vm_operations_struct kvm_vm_vm_ops = {
2260         .fault = kvm_vm_fault,
2261 };
2262
2263 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2264 {
2265         vma->vm_ops = &kvm_vm_vm_ops;
2266         return 0;
2267 }
2268
2269 static struct file_operations kvm_vm_fops = {
2270         .release        = kvm_vm_release,
2271         .unlocked_ioctl = kvm_vm_ioctl,
2272         .compat_ioctl   = kvm_vm_ioctl,
2273         .mmap           = kvm_vm_mmap,
2274 };
2275
2276 static int kvm_dev_ioctl_create_vm(void)
2277 {
2278         int fd;
2279         struct kvm *kvm;
2280
2281         kvm = kvm_create_vm();
2282         if (IS_ERR(kvm))
2283                 return PTR_ERR(kvm);
2284         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2285         if (fd < 0)
2286                 kvm_put_kvm(kvm);
2287
2288         return fd;
2289 }
2290
2291 static long kvm_dev_ioctl_check_extension_generic(long arg)
2292 {
2293         switch (arg) {
2294         case KVM_CAP_USER_MEMORY:
2295         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2296         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2297                 return 1;
2298 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2299         case KVM_CAP_IRQ_ROUTING:
2300                 return KVM_MAX_IRQ_ROUTES;
2301 #endif
2302         default:
2303                 break;
2304         }
2305         return kvm_dev_ioctl_check_extension(arg);
2306 }
2307
2308 static long kvm_dev_ioctl(struct file *filp,
2309                           unsigned int ioctl, unsigned long arg)
2310 {
2311         long r = -EINVAL;
2312
2313         switch (ioctl) {
2314         case KVM_GET_API_VERSION:
2315                 r = -EINVAL;
2316                 if (arg)
2317                         goto out;
2318                 r = KVM_API_VERSION;
2319                 break;
2320         case KVM_CREATE_VM:
2321                 r = -EINVAL;
2322                 if (arg)
2323                         goto out;
2324                 r = kvm_dev_ioctl_create_vm();
2325                 break;
2326         case KVM_CHECK_EXTENSION:
2327                 r = kvm_dev_ioctl_check_extension_generic(arg);
2328                 break;
2329         case KVM_GET_VCPU_MMAP_SIZE:
2330                 r = -EINVAL;
2331                 if (arg)
2332                         goto out;
2333                 r = PAGE_SIZE;     /* struct kvm_run */
2334 #ifdef CONFIG_X86
2335                 r += PAGE_SIZE;    /* pio data page */
2336 #endif
2337 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2338                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2339 #endif
2340                 break;
2341         case KVM_TRACE_ENABLE:
2342         case KVM_TRACE_PAUSE:
2343         case KVM_TRACE_DISABLE:
2344                 r = kvm_trace_ioctl(ioctl, arg);
2345                 break;
2346         default:
2347                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2348         }
2349 out:
2350         return r;
2351 }
2352
2353 static struct file_operations kvm_chardev_ops = {
2354         .unlocked_ioctl = kvm_dev_ioctl,
2355         .compat_ioctl   = kvm_dev_ioctl,
2356 };
2357
2358 static struct miscdevice kvm_dev = {
2359         KVM_MINOR,
2360         "kvm",
2361         &kvm_chardev_ops,
2362 };
2363
2364 static void hardware_enable(void *junk)
2365 {
2366         int cpu = raw_smp_processor_id();
2367
2368         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2369                 return;
2370         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2371         kvm_arch_hardware_enable(NULL);
2372 }
2373
2374 static void hardware_disable(void *junk)
2375 {
2376         int cpu = raw_smp_processor_id();
2377
2378         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2379                 return;
2380         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2381         kvm_arch_hardware_disable(NULL);
2382 }
2383
2384 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2385                            void *v)
2386 {
2387         int cpu = (long)v;
2388
2389         val &= ~CPU_TASKS_FROZEN;
2390         switch (val) {
2391         case CPU_DYING:
2392                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2393                        cpu);
2394                 hardware_disable(NULL);
2395                 break;
2396         case CPU_UP_CANCELED:
2397                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2398                        cpu);
2399                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2400                 break;
2401         case CPU_ONLINE:
2402                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2403                        cpu);
2404                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2405                 break;
2406         }
2407         return NOTIFY_OK;
2408 }
2409
2410
2411 asmlinkage void kvm_handle_fault_on_reboot(void)
2412 {
2413         if (kvm_rebooting)
2414                 /* spin while reset goes on */
2415                 while (true)
2416                         ;
2417         /* Fault while not rebooting.  We want the trace. */
2418         BUG();
2419 }
2420 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2421
2422 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2423                       void *v)
2424 {
2425         /*
2426          * Some (well, at least mine) BIOSes hang on reboot if
2427          * in vmx root mode.
2428          *
2429          * And Intel TXT required VMX off for all cpu when system shutdown.
2430          */
2431         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2432         kvm_rebooting = true;
2433         on_each_cpu(hardware_disable, NULL, 1);
2434         return NOTIFY_OK;
2435 }
2436
2437 static struct notifier_block kvm_reboot_notifier = {
2438         .notifier_call = kvm_reboot,
2439         .priority = 0,
2440 };
2441
2442 void kvm_io_bus_init(struct kvm_io_bus *bus)
2443 {
2444         memset(bus, 0, sizeof(*bus));
2445 }
2446
2447 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2448 {
2449         int i;
2450
2451         for (i = 0; i < bus->dev_count; i++) {
2452                 struct kvm_io_device *pos = bus->devs[i];
2453
2454                 kvm_iodevice_destructor(pos);
2455         }
2456 }
2457
2458 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2459                                           gpa_t addr, int len, int is_write)
2460 {
2461         int i;
2462
2463         for (i = 0; i < bus->dev_count; i++) {
2464                 struct kvm_io_device *pos = bus->devs[i];
2465
2466                 if (pos->in_range(pos, addr, len, is_write))
2467                         return pos;
2468         }
2469
2470         return NULL;
2471 }
2472
2473 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2474 {
2475         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2476
2477         bus->devs[bus->dev_count++] = dev;
2478 }
2479
2480 static struct notifier_block kvm_cpu_notifier = {
2481         .notifier_call = kvm_cpu_hotplug,
2482         .priority = 20, /* must be > scheduler priority */
2483 };
2484
2485 static int vm_stat_get(void *_offset, u64 *val)
2486 {
2487         unsigned offset = (long)_offset;
2488         struct kvm *kvm;
2489
2490         *val = 0;
2491         spin_lock(&kvm_lock);
2492         list_for_each_entry(kvm, &vm_list, vm_list)
2493                 *val += *(u32 *)((void *)kvm + offset);
2494         spin_unlock(&kvm_lock);
2495         return 0;
2496 }
2497
2498 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2499
2500 static int vcpu_stat_get(void *_offset, u64 *val)
2501 {
2502         unsigned offset = (long)_offset;
2503         struct kvm *kvm;
2504         struct kvm_vcpu *vcpu;
2505         int i;
2506
2507         *val = 0;
2508         spin_lock(&kvm_lock);
2509         list_for_each_entry(kvm, &vm_list, vm_list)
2510                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2511                         vcpu = kvm->vcpus[i];
2512                         if (vcpu)
2513                                 *val += *(u32 *)((void *)vcpu + offset);
2514                 }
2515         spin_unlock(&kvm_lock);
2516         return 0;
2517 }
2518
2519 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2520
2521 static struct file_operations *stat_fops[] = {
2522         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2523         [KVM_STAT_VM]   = &vm_stat_fops,
2524 };
2525
2526 static void kvm_init_debug(void)
2527 {
2528         struct kvm_stats_debugfs_item *p;
2529
2530         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2531         for (p = debugfs_entries; p->name; ++p)
2532                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2533                                                 (void *)(long)p->offset,
2534                                                 stat_fops[p->kind]);
2535 }
2536
2537 static void kvm_exit_debug(void)
2538 {
2539         struct kvm_stats_debugfs_item *p;
2540
2541         for (p = debugfs_entries; p->name; ++p)
2542                 debugfs_remove(p->dentry);
2543         debugfs_remove(kvm_debugfs_dir);
2544 }
2545
2546 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2547 {
2548         hardware_disable(NULL);
2549         return 0;
2550 }
2551
2552 static int kvm_resume(struct sys_device *dev)
2553 {
2554         hardware_enable(NULL);
2555         return 0;
2556 }
2557
2558 static struct sysdev_class kvm_sysdev_class = {
2559         .name = "kvm",
2560         .suspend = kvm_suspend,
2561         .resume = kvm_resume,
2562 };
2563
2564 static struct sys_device kvm_sysdev = {
2565         .id = 0,
2566         .cls = &kvm_sysdev_class,
2567 };
2568
2569 struct page *bad_page;
2570 pfn_t bad_pfn;
2571
2572 static inline
2573 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2574 {
2575         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2576 }
2577
2578 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2579 {
2580         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2581
2582         kvm_arch_vcpu_load(vcpu, cpu);
2583 }
2584
2585 static void kvm_sched_out(struct preempt_notifier *pn,
2586                           struct task_struct *next)
2587 {
2588         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2589
2590         kvm_arch_vcpu_put(vcpu);
2591 }
2592
2593 int kvm_init(void *opaque, unsigned int vcpu_size,
2594                   struct module *module)
2595 {
2596         int r;
2597         int cpu;
2598
2599         kvm_init_debug();
2600
2601         r = kvm_arch_init(opaque);
2602         if (r)
2603                 goto out_fail;
2604
2605         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2606
2607         if (bad_page == NULL) {
2608                 r = -ENOMEM;
2609                 goto out;
2610         }
2611
2612         bad_pfn = page_to_pfn(bad_page);
2613
2614         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2615                 r = -ENOMEM;
2616                 goto out_free_0;
2617         }
2618
2619         r = kvm_arch_hardware_setup();
2620         if (r < 0)
2621                 goto out_free_0a;
2622
2623         for_each_online_cpu(cpu) {
2624                 smp_call_function_single(cpu,
2625                                 kvm_arch_check_processor_compat,
2626                                 &r, 1);
2627                 if (r < 0)
2628                         goto out_free_1;
2629         }
2630
2631         on_each_cpu(hardware_enable, NULL, 1);
2632         r = register_cpu_notifier(&kvm_cpu_notifier);
2633         if (r)
2634                 goto out_free_2;
2635         register_reboot_notifier(&kvm_reboot_notifier);
2636
2637         r = sysdev_class_register(&kvm_sysdev_class);
2638         if (r)
2639                 goto out_free_3;
2640
2641         r = sysdev_register(&kvm_sysdev);
2642         if (r)
2643                 goto out_free_4;
2644
2645         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2646         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2647                                            __alignof__(struct kvm_vcpu),
2648                                            0, NULL);
2649         if (!kvm_vcpu_cache) {
2650                 r = -ENOMEM;
2651                 goto out_free_5;
2652         }
2653
2654         kvm_chardev_ops.owner = module;
2655         kvm_vm_fops.owner = module;
2656         kvm_vcpu_fops.owner = module;
2657
2658         r = misc_register(&kvm_dev);
2659         if (r) {
2660                 printk(KERN_ERR "kvm: misc device register failed\n");
2661                 goto out_free;
2662         }
2663
2664         kvm_preempt_ops.sched_in = kvm_sched_in;
2665         kvm_preempt_ops.sched_out = kvm_sched_out;
2666
2667         return 0;
2668
2669 out_free:
2670         kmem_cache_destroy(kvm_vcpu_cache);
2671 out_free_5:
2672         sysdev_unregister(&kvm_sysdev);
2673 out_free_4:
2674         sysdev_class_unregister(&kvm_sysdev_class);
2675 out_free_3:
2676         unregister_reboot_notifier(&kvm_reboot_notifier);
2677         unregister_cpu_notifier(&kvm_cpu_notifier);
2678 out_free_2:
2679         on_each_cpu(hardware_disable, NULL, 1);
2680 out_free_1:
2681         kvm_arch_hardware_unsetup();
2682 out_free_0a:
2683         free_cpumask_var(cpus_hardware_enabled);
2684 out_free_0:
2685         __free_page(bad_page);
2686 out:
2687         kvm_arch_exit();
2688         kvm_exit_debug();
2689 out_fail:
2690         return r;
2691 }
2692 EXPORT_SYMBOL_GPL(kvm_init);
2693
2694 void kvm_exit(void)
2695 {
2696         kvm_trace_cleanup();
2697         misc_deregister(&kvm_dev);
2698         kmem_cache_destroy(kvm_vcpu_cache);
2699         sysdev_unregister(&kvm_sysdev);
2700         sysdev_class_unregister(&kvm_sysdev_class);
2701         unregister_reboot_notifier(&kvm_reboot_notifier);
2702         unregister_cpu_notifier(&kvm_cpu_notifier);
2703         on_each_cpu(hardware_disable, NULL, 1);
2704         kvm_arch_hardware_unsetup();
2705         kvm_arch_exit();
2706         kvm_exit_debug();
2707         free_cpumask_var(cpus_hardware_enabled);
2708         __free_page(bad_page);
2709 }
2710 EXPORT_SYMBOL_GPL(kvm_exit);