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 /*
  66  * Ordering of locks:
  67  *
  68  *              kvm->lock --> kvm->irq_lock
  69  */
  70
  71 DEFINE_SPINLOCK(kvm_lock);
  72 LIST_HEAD(vm_list);
  73
  74 static cpumask_var_t cpus_hardware_enabled;
  75
  76 struct kmem_cache *kvm_vcpu_cache;
  77 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  78
  79 static __read_mostly struct preempt_ops kvm_preempt_ops;
  80
  81 struct dentry *kvm_debugfs_dir;
  82
  83 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  84                            unsigned long arg);
  85
  86 static bool kvm_rebooting;
  87
  88 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
  89 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
  90                                                       int assigned_dev_id)
  91 {
  92         struct list_head *ptr;
  93         struct kvm_assigned_dev_kernel *match;
  94
  95         list_for_each(ptr, head) {
  96                 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
  97                 if (match->assigned_dev_id == assigned_dev_id)
  98                         return match;
  99         }
 100         return NULL;
 101 }
 102
 103 static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
 104                                     *assigned_dev, int irq)
 105 {
 106         int i, index;
 107         struct msix_entry *host_msix_entries;
 108
 109         host_msix_entries = assigned_dev->host_msix_entries;
 110
 111         index = -1;
 112         for (i = 0; i < assigned_dev->entries_nr; i++)
 113                 if (irq == host_msix_entries[i].vector) {
 114                         index = i;
 115                         break;
 116                 }
 117         if (index < 0) {
 118                 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
 119                 return 0;
 120         }
 121
 122         return index;
 123 }
 124
 125 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
 126 {
 127         struct kvm_assigned_dev_kernel *assigned_dev;
 128         struct kvm *kvm;
 129         int i;
 130
 131         assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
 132                                     interrupt_work);
 133         kvm = assigned_dev->kvm;
 134
 135         mutex_lock(&kvm->irq_lock);
 136         spin_lock_irq(&assigned_dev->assigned_dev_lock);
 137         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 138                 struct kvm_guest_msix_entry *guest_entries =
 139                         assigned_dev->guest_msix_entries;
 140                 for (i = 0; i < assigned_dev->entries_nr; i++) {
 141                         if (!(guest_entries[i].flags &
 142                                         KVM_ASSIGNED_MSIX_PENDING))
 143                                 continue;
 144                         guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
 145                         kvm_set_irq(assigned_dev->kvm,
 146                                     assigned_dev->irq_source_id,
 147                                     guest_entries[i].vector, 1);
 148                 }
 149         } else
 150                 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
 151                             assigned_dev->guest_irq, 1);
 152
 153         spin_unlock_irq(&assigned_dev->assigned_dev_lock);
 154         mutex_unlock(&assigned_dev->kvm->irq_lock);
 155 }
 156
 157 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
 158 {
 159         unsigned long flags;
 160         struct kvm_assigned_dev_kernel *assigned_dev =
 161                 (struct kvm_assigned_dev_kernel *) dev_id;
 162
 163         spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
 164         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 165                 int index = find_index_from_host_irq(assigned_dev, irq);
 166                 if (index < 0)
 167                         goto out;
 168                 assigned_dev->guest_msix_entries[index].flags |=
 169                         KVM_ASSIGNED_MSIX_PENDING;
 170         }
 171
 172         schedule_work(&assigned_dev->interrupt_work);
 173
 174         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
 175                 disable_irq_nosync(irq);
 176                 assigned_dev->host_irq_disabled = true;
 177         }
 178
 179 out:
 180         spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
 181         return IRQ_HANDLED;
 182 }
 183
 184 /* Ack the irq line for an assigned device */
 185 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
 186 {
 187         struct kvm_assigned_dev_kernel *dev;
 188         unsigned long flags;
 189
 190         if (kian->gsi == -1)
 191                 return;
 192
 193         dev = container_of(kian, struct kvm_assigned_dev_kernel,
 194                            ack_notifier);
 195
 196         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
 197
 198         /* The guest irq may be shared so this ack may be
 199          * from another device.
 200          */
 201         spin_lock_irqsave(&dev->assigned_dev_lock, flags);
 202         if (dev->host_irq_disabled) {
 203                 enable_irq(dev->host_irq);
 204                 dev->host_irq_disabled = false;
 205         }
 206         spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
 207 }
 208
 209 static void deassign_guest_irq(struct kvm *kvm,
 210                                struct kvm_assigned_dev_kernel *assigned_dev)
 211 {
 212         kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
 213         assigned_dev->ack_notifier.gsi = -1;
 214
 215         if (assigned_dev->irq_source_id != -1)
 216                 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
 217         assigned_dev->irq_source_id = -1;
 218         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
 219 }
 220
 221 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
 222 static void deassign_host_irq(struct kvm *kvm,
 223                               struct kvm_assigned_dev_kernel *assigned_dev)
 224 {
 225         /*
 226          * In kvm_free_device_irq, cancel_work_sync return true if:
 227          * 1. work is scheduled, and then cancelled.
 228          * 2. work callback is executed.
 229          *
 230          * The first one ensured that the irq is disabled and no more events
 231          * would happen. But for the second one, the irq may be enabled (e.g.
 232          * for MSI). So we disable irq here to prevent further events.
 233          *
 234          * Notice this maybe result in nested disable if the interrupt type is
 235          * INTx, but it's OK for we are going to free it.
 236          *
 237          * If this function is a part of VM destroy, please ensure that till
 238          * now, the kvm state is still legal for probably we also have to wait
 239          * interrupt_work done.
 240          */
 241         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 242                 int i;
 243                 for (i = 0; i < assigned_dev->entries_nr; i++)
 244                         disable_irq_nosync(assigned_dev->
 245                                            host_msix_entries[i].vector);
 246
 247                 cancel_work_sync(&assigned_dev->interrupt_work);
 248
 249                 for (i = 0; i < assigned_dev->entries_nr; i++)
 250                         free_irq(assigned_dev->host_msix_entries[i].vector,
 251                                  (void *)assigned_dev);
 252
 253                 assigned_dev->entries_nr = 0;
 254                 kfree(assigned_dev->host_msix_entries);
 255                 kfree(assigned_dev->guest_msix_entries);
 256                 pci_disable_msix(assigned_dev->dev);
 257         } else {
 258                 /* Deal with MSI and INTx */
 259                 disable_irq_nosync(assigned_dev->host_irq);
 260                 cancel_work_sync(&assigned_dev->interrupt_work);
 261
 262                 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
 263
 264                 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
 265                         pci_disable_msi(assigned_dev->dev);
 266         }
 267
 268         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
 269 }
 270
 271 static int kvm_deassign_irq(struct kvm *kvm,
 272                             struct kvm_assigned_dev_kernel *assigned_dev,
 273                             unsigned long irq_requested_type)
 274 {
 275         unsigned long guest_irq_type, host_irq_type;
 276
 277         if (!irqchip_in_kernel(kvm))
 278                 return -EINVAL;
 279         /* no irq assignment to deassign */
 280         if (!assigned_dev->irq_requested_type)
 281                 return -ENXIO;
 282
 283         host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
 284         guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
 285
 286         if (host_irq_type)
 287                 deassign_host_irq(kvm, assigned_dev);
 288         if (guest_irq_type)
 289                 deassign_guest_irq(kvm, assigned_dev);
 290
 291         return 0;
 292 }
 293
 294 static void kvm_free_assigned_irq(struct kvm *kvm,
 295                                   struct kvm_assigned_dev_kernel *assigned_dev)
 296 {
 297         kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
 298 }
 299
 300 static void kvm_free_assigned_device(struct kvm *kvm,
 301                                      struct kvm_assigned_dev_kernel
 302                                      *assigned_dev)
 303 {
 304         kvm_free_assigned_irq(kvm, assigned_dev);
 305
 306         pci_reset_function(assigned_dev->dev);
 307
 308         pci_release_regions(assigned_dev->dev);
 309         pci_disable_device(assigned_dev->dev);
 310         pci_dev_put(assigned_dev->dev);
 311
 312         list_del(&assigned_dev->list);
 313         kfree(assigned_dev);
 314 }
 315
 316 void kvm_free_all_assigned_devices(struct kvm *kvm)
 317 {
 318         struct list_head *ptr, *ptr2;
 319         struct kvm_assigned_dev_kernel *assigned_dev;
 320
 321         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
 322                 assigned_dev = list_entry(ptr,
 323                                           struct kvm_assigned_dev_kernel,
 324                                           list);
 325
 326                 kvm_free_assigned_device(kvm, assigned_dev);
 327         }
 328 }
 329
 330 static int assigned_device_enable_host_intx(struct kvm *kvm,
 331                                             struct kvm_assigned_dev_kernel *dev)
 332 {
 333         dev->host_irq = dev->dev->irq;
 334         /* Even though this is PCI, we don't want to use shared
 335          * interrupts. Sharing host devices with guest-assigned devices
 336          * on the same interrupt line is not a happy situation: there
 337          * are going to be long delays in accepting, acking, etc.
 338          */
 339         if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
 340                         0, "kvm_assigned_intx_device", (void *)dev))
 341                 return -EIO;
 342         return 0;
 343 }
 344
 345 #ifdef __KVM_HAVE_MSI
 346 static int assigned_device_enable_host_msi(struct kvm *kvm,
 347                                            struct kvm_assigned_dev_kernel *dev)
 348 {
 349         int r;
 350
 351         if (!dev->dev->msi_enabled) {
 352                 r = pci_enable_msi(dev->dev);
 353                 if (r)
 354                         return r;
 355         }
 356
 357         dev->host_irq = dev->dev->irq;
 358         if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
 359                         "kvm_assigned_msi_device", (void *)dev)) {
 360                 pci_disable_msi(dev->dev);
 361                 return -EIO;
 362         }
 363
 364         return 0;
 365 }
 366 #endif
 367
 368 #ifdef __KVM_HAVE_MSIX
 369 static int assigned_device_enable_host_msix(struct kvm *kvm,
 370                                             struct kvm_assigned_dev_kernel *dev)
 371 {
 372         int i, r = -EINVAL;
 373
 374         /* host_msix_entries and guest_msix_entries should have been
 375          * initialized */
 376         if (dev->entries_nr == 0)
 377                 return r;
 378
 379         r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
 380         if (r)
 381                 return r;
 382
 383         for (i = 0; i < dev->entries_nr; i++) {
 384                 r = request_irq(dev->host_msix_entries[i].vector,
 385                                 kvm_assigned_dev_intr, 0,
 386                                 "kvm_assigned_msix_device",
 387                                 (void *)dev);
 388                 /* FIXME: free requested_irq's on failure */
 389                 if (r)
 390                         return r;
 391         }
 392
 393         return 0;
 394 }
 395
 396 #endif
 397
 398 static int assigned_device_enable_guest_intx(struct kvm *kvm,
 399                                 struct kvm_assigned_dev_kernel *dev,
 400                                 struct kvm_assigned_irq *irq)
 401 {
 402         dev->guest_irq = irq->guest_irq;
 403         dev->ack_notifier.gsi = irq->guest_irq;
 404         return 0;
 405 }
 406
 407 #ifdef __KVM_HAVE_MSI
 408 static int assigned_device_enable_guest_msi(struct kvm *kvm,
 409                         struct kvm_assigned_dev_kernel *dev,
 410                         struct kvm_assigned_irq *irq)
 411 {
 412         dev->guest_irq = irq->guest_irq;
 413         dev->ack_notifier.gsi = -1;
 414         dev->host_irq_disabled = false;
 415         return 0;
 416 }
 417 #endif
 418 #ifdef __KVM_HAVE_MSIX
 419 static int assigned_device_enable_guest_msix(struct kvm *kvm,
 420                         struct kvm_assigned_dev_kernel *dev,
 421                         struct kvm_assigned_irq *irq)
 422 {
 423         dev->guest_irq = irq->guest_irq;
 424         dev->ack_notifier.gsi = -1;
 425         dev->host_irq_disabled = false;
 426         return 0;
 427 }
 428 #endif
 429
 430 static int assign_host_irq(struct kvm *kvm,
 431                            struct kvm_assigned_dev_kernel *dev,
 432                            __u32 host_irq_type)
 433 {
 434         int r = -EEXIST;
 435
 436         if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
 437                 return r;
 438
 439         switch (host_irq_type) {
 440         case KVM_DEV_IRQ_HOST_INTX:
 441                 r = assigned_device_enable_host_intx(kvm, dev);
 442                 break;
 443 #ifdef __KVM_HAVE_MSI
 444         case KVM_DEV_IRQ_HOST_MSI:
 445                 r = assigned_device_enable_host_msi(kvm, dev);
 446                 break;
 447 #endif
 448 #ifdef __KVM_HAVE_MSIX
 449         case KVM_DEV_IRQ_HOST_MSIX:
 450                 r = assigned_device_enable_host_msix(kvm, dev);
 451                 break;
 452 #endif
 453         default:
 454                 r = -EINVAL;
 455         }
 456
 457         if (!r)
 458                 dev->irq_requested_type |= host_irq_type;
 459
 460         return r;
 461 }
 462
 463 static int assign_guest_irq(struct kvm *kvm,
 464                             struct kvm_assigned_dev_kernel *dev,
 465                             struct kvm_assigned_irq *irq,
 466                             unsigned long guest_irq_type)
 467 {
 468         int id;
 469         int r = -EEXIST;
 470
 471         if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
 472                 return r;
 473
 474         id = kvm_request_irq_source_id(kvm);
 475         if (id < 0)
 476                 return id;
 477
 478         dev->irq_source_id = id;
 479
 480         switch (guest_irq_type) {
 481         case KVM_DEV_IRQ_GUEST_INTX:
 482                 r = assigned_device_enable_guest_intx(kvm, dev, irq);
 483                 break;
 484 #ifdef __KVM_HAVE_MSI
 485         case KVM_DEV_IRQ_GUEST_MSI:
 486                 r = assigned_device_enable_guest_msi(kvm, dev, irq);
 487                 break;
 488 #endif
 489 #ifdef __KVM_HAVE_MSIX
 490         case KVM_DEV_IRQ_GUEST_MSIX:
 491                 r = assigned_device_enable_guest_msix(kvm, dev, irq);
 492                 break;
 493 #endif
 494         default:
 495                 r = -EINVAL;
 496         }
 497
 498         if (!r) {
 499                 dev->irq_requested_type |= guest_irq_type;
 500                 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
 501         } else
 502                 kvm_free_irq_source_id(kvm, dev->irq_source_id);
 503
 504         return r;
 505 }
 506
 507 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
 508 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 509                                    struct kvm_assigned_irq *assigned_irq)
 510 {
 511         int r = -EINVAL;
 512         struct kvm_assigned_dev_kernel *match;
 513         unsigned long host_irq_type, guest_irq_type;
 514
 515         if (!capable(CAP_SYS_RAWIO))
 516                 return -EPERM;
 517
 518         if (!irqchip_in_kernel(kvm))
 519                 return r;
 520
 521         mutex_lock(&kvm->lock);
 522         r = -ENODEV;
 523         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 524                                       assigned_irq->assigned_dev_id);
 525         if (!match)
 526                 goto out;
 527
 528         host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
 529         guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
 530
 531         r = -EINVAL;
 532         /* can only assign one type at a time */
 533         if (hweight_long(host_irq_type) > 1)
 534                 goto out;
 535         if (hweight_long(guest_irq_type) > 1)
 536                 goto out;
 537         if (host_irq_type == 0 && guest_irq_type == 0)
 538                 goto out;
 539
 540         r = 0;
 541         if (host_irq_type)
 542                 r = assign_host_irq(kvm, match, host_irq_type);
 543         if (r)
 544                 goto out;
 545
 546         if (guest_irq_type)
 547                 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
 548 out:
 549         mutex_unlock(&kvm->lock);
 550         return r;
 551 }
 552
 553 static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
 554                                          struct kvm_assigned_irq
 555                                          *assigned_irq)
 556 {
 557         int r = -ENODEV;
 558         struct kvm_assigned_dev_kernel *match;
 559
 560         mutex_lock(&kvm->lock);
 561
 562         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 563                                       assigned_irq->assigned_dev_id);
 564         if (!match)
 565                 goto out;
 566
 567         r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
 568 out:
 569         mutex_unlock(&kvm->lock);
 570         return r;
 571 }
 572
 573 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 574                                       struct kvm_assigned_pci_dev *assigned_dev)
 575 {
 576         int r = 0;
 577         struct kvm_assigned_dev_kernel *match;
 578         struct pci_dev *dev;
 579
 580         down_read(&kvm->slots_lock);
 581         mutex_lock(&kvm->lock);
 582
 583         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 584                                       assigned_dev->assigned_dev_id);
 585         if (match) {
 586                 /* device already assigned */
 587                 r = -EEXIST;
 588                 goto out;
 589         }
 590
 591         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 592         if (match == NULL) {
 593                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 594                        __func__);
 595                 r = -ENOMEM;
 596                 goto out;
 597         }
 598         dev = pci_get_bus_and_slot(assigned_dev->busnr,
 599                                    assigned_dev->devfn);
 600         if (!dev) {
 601                 printk(KERN_INFO "%s: host device not found\n", __func__);
 602                 r = -EINVAL;
 603                 goto out_free;
 604         }
 605         if (pci_enable_device(dev)) {
 606                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 607                 r = -EBUSY;
 608                 goto out_put;
 609         }
 610         r = pci_request_regions(dev, "kvm_assigned_device");
 611         if (r) {
 612                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 613                        __func__);
 614                 goto out_disable;
 615         }
 616
 617         pci_reset_function(dev);
 618
 619         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 620         match->host_busnr = assigned_dev->busnr;
 621         match->host_devfn = assigned_dev->devfn;
 622         match->flags = assigned_dev->flags;
 623         match->dev = dev;
 624         spin_lock_init(&match->assigned_dev_lock);
 625         match->irq_source_id = -1;
 626         match->kvm = kvm;
 627         match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
 628         INIT_WORK(&match->interrupt_work,
 629                   kvm_assigned_dev_interrupt_work_handler);
 630
 631         list_add(&match->list, &kvm->arch.assigned_dev_head);
 632
 633         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
 634                 if (!kvm->arch.iommu_domain) {
 635                         r = kvm_iommu_map_guest(kvm);
 636                         if (r)
 637                                 goto out_list_del;
 638                 }
 639                 r = kvm_assign_device(kvm, match);
 640                 if (r)
 641                         goto out_list_del;
 642         }
 643
 644 out:
 645         mutex_unlock(&kvm->lock);
 646         up_read(&kvm->slots_lock);
 647         return r;
 648 out_list_del:
 649         list_del(&match->list);
 650         pci_release_regions(dev);
 651 out_disable:
 652         pci_disable_device(dev);
 653 out_put:
 654         pci_dev_put(dev);
 655 out_free:
 656         kfree(match);
 657         mutex_unlock(&kvm->lock);
 658         up_read(&kvm->slots_lock);
 659         return r;
 660 }
 661 #endif
 662
 663 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
 664 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
 665                 struct kvm_assigned_pci_dev *assigned_dev)
 666 {
 667         int r = 0;
 668         struct kvm_assigned_dev_kernel *match;
 669
 670         mutex_lock(&kvm->lock);
 671
 672         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 673                                       assigned_dev->assigned_dev_id);
 674         if (!match) {
 675                 printk(KERN_INFO "%s: device hasn't been assigned before, "
 676                   "so cannot be deassigned\n", __func__);
 677                 r = -EINVAL;
 678                 goto out;
 679         }
 680
 681         if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
 682                 kvm_deassign_device(kvm, match);
 683
 684         kvm_free_assigned_device(kvm, match);
 685
 686 out:
 687         mutex_unlock(&kvm->lock);
 688         return r;
 689 }
 690 #endif
 691
 692 inline int kvm_is_mmio_pfn(pfn_t pfn)
 693 {
 694         if (pfn_valid(pfn)) {
 695                 struct page *page = compound_head(pfn_to_page(pfn));
 696                 return PageReserved(page);
 697         }
 698
 699         return true;
 700 }
 701
 702 /*
 703  * Switches to specified vcpu, until a matching vcpu_put()
 704  */
 705 void vcpu_load(struct kvm_vcpu *vcpu)
 706 {
 707         int cpu;
 708
 709         mutex_lock(&vcpu->mutex);
 710         cpu = get_cpu();
 711         preempt_notifier_register(&vcpu->preempt_notifier);
 712         kvm_arch_vcpu_load(vcpu, cpu);
 713         put_cpu();
 714 }
 715
 716 void vcpu_put(struct kvm_vcpu *vcpu)
 717 {
 718         preempt_disable();
 719         kvm_arch_vcpu_put(vcpu);
 720         preempt_notifier_unregister(&vcpu->preempt_notifier);
 721         preempt_enable();
 722         mutex_unlock(&vcpu->mutex);
 723 }
 724
 725 static void ack_flush(void *_completed)
 726 {
 727 }
 728
 729 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 730 {
 731         int i, cpu, me;
 732         cpumask_var_t cpus;
 733         bool called = true;
 734         struct kvm_vcpu *vcpu;
 735
 736         if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
 737                 cpumask_clear(cpus);
 738
 739         me = get_cpu();
 740         spin_lock(&kvm->requests_lock);
 741         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 742                 vcpu = kvm->vcpus[i];
 743                 if (!vcpu)
 744                         continue;
 745                 if (test_and_set_bit(req, &vcpu->requests))
 746                         continue;
 747                 cpu = vcpu->cpu;
 748                 if (cpus != NULL && cpu != -1 && cpu != me)
 749                         cpumask_set_cpu(cpu, cpus);
 750         }
 751         if (unlikely(cpus == NULL))
 752                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 753         else if (!cpumask_empty(cpus))
 754                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 755         else
 756                 called = false;
 757         spin_unlock(&kvm->requests_lock);
 758         put_cpu();
 759         free_cpumask_var(cpus);
 760         return called;
 761 }
 762
 763 void kvm_flush_remote_tlbs(struct kvm *kvm)
 764 {
 765         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 766                 ++kvm->stat.remote_tlb_flush;
 767 }
 768
 769 void kvm_reload_remote_mmus(struct kvm *kvm)
 770 {
 771         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 772 }
 773
 774 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 775 {
 776         struct page *page;
 777         int r;
 778
 779         mutex_init(&vcpu->mutex);
 780         vcpu->cpu = -1;
 781         vcpu->kvm = kvm;
 782         vcpu->vcpu_id = id;
 783         init_waitqueue_head(&vcpu->wq);
 784
 785         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 786         if (!page) {
 787                 r = -ENOMEM;
 788                 goto fail;
 789         }
 790         vcpu->run = page_address(page);
 791
 792         r = kvm_arch_vcpu_init(vcpu);
 793         if (r < 0)
 794                 goto fail_free_run;
 795         return 0;
 796
 797 fail_free_run:
 798         free_page((unsigned long)vcpu->run);
 799 fail:
 800         return r;
 801 }
 802 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 803
 804 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 805 {
 806         kvm_arch_vcpu_uninit(vcpu);
 807         free_page((unsigned long)vcpu->run);
 808 }
 809 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 810
 811 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 812 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 813 {
 814         return container_of(mn, struct kvm, mmu_notifier);
 815 }
 816
 817 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 818                                              struct mm_struct *mm,
 819                                              unsigned long address)
 820 {
 821         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 822         int need_tlb_flush;
 823
 824         /*
 825          * When ->invalidate_page runs, the linux pte has been zapped
 826          * already but the page is still allocated until
 827          * ->invalidate_page returns. So if we increase the sequence
 828          * here the kvm page fault will notice if the spte can't be
 829          * established because the page is going to be freed. If
 830          * instead the kvm page fault establishes the spte before
 831          * ->invalidate_page runs, kvm_unmap_hva will release it
 832          * before returning.
 833          *
 834          * The sequence increase only need to be seen at spin_unlock
 835          * time, and not at spin_lock time.
 836          *
 837          * Increasing the sequence after the spin_unlock would be
 838          * unsafe because the kvm page fault could then establish the
 839          * pte after kvm_unmap_hva returned, without noticing the page
 840          * is going to be freed.
 841          */
 842         spin_lock(&kvm->mmu_lock);
 843         kvm->mmu_notifier_seq++;
 844         need_tlb_flush = kvm_unmap_hva(kvm, address);
 845         spin_unlock(&kvm->mmu_lock);
 846
 847         /* we've to flush the tlb before the pages can be freed */
 848         if (need_tlb_flush)
 849                 kvm_flush_remote_tlbs(kvm);
 850
 851 }
 852
 853 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 854                                                     struct mm_struct *mm,
 855                                                     unsigned long start,
 856                                                     unsigned long end)
 857 {
 858         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 859         int need_tlb_flush = 0;
 860
 861         spin_lock(&kvm->mmu_lock);
 862         /*
 863          * The count increase must become visible at unlock time as no
 864          * spte can be established without taking the mmu_lock and
 865          * count is also read inside the mmu_lock critical section.
 866          */
 867         kvm->mmu_notifier_count++;
 868         for (; start < end; start += PAGE_SIZE)
 869                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 870         spin_unlock(&kvm->mmu_lock);
 871
 872         /* we've to flush the tlb before the pages can be freed */
 873         if (need_tlb_flush)
 874                 kvm_flush_remote_tlbs(kvm);
 875 }
 876
 877 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 878                                                   struct mm_struct *mm,
 879                                                   unsigned long start,
 880                                                   unsigned long end)
 881 {
 882         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 883
 884         spin_lock(&kvm->mmu_lock);
 885         /*
 886          * This sequence increase will notify the kvm page fault that
 887          * the page that is going to be mapped in the spte could have
 888          * been freed.
 889          */
 890         kvm->mmu_notifier_seq++;
 891         /*
 892          * The above sequence increase must be visible before the
 893          * below count decrease but both values are read by the kvm
 894          * page fault under mmu_lock spinlock so we don't need to add
 895          * a smb_wmb() here in between the two.
 896          */
 897         kvm->mmu_notifier_count--;
 898         spin_unlock(&kvm->mmu_lock);
 899
 900         BUG_ON(kvm->mmu_notifier_count < 0);
 901 }
 902
 903 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 904                                               struct mm_struct *mm,
 905                                               unsigned long address)
 906 {
 907         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 908         int young;
 909
 910         spin_lock(&kvm->mmu_lock);
 911         young = kvm_age_hva(kvm, address);
 912         spin_unlock(&kvm->mmu_lock);
 913
 914         if (young)
 915                 kvm_flush_remote_tlbs(kvm);
 916
 917         return young;
 918 }
 919
 920 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 921                                      struct mm_struct *mm)
 922 {
 923         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 924         kvm_arch_flush_shadow(kvm);
 925 }
 926
 927 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 928         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 929         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 930         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 931         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 932         .release                = kvm_mmu_notifier_release,
 933 };
 934 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 935
 936 static struct kvm *kvm_create_vm(void)
 937 {
 938         struct kvm *kvm = kvm_arch_create_vm();
 939 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 940         struct page *page;
 941 #endif
 942
 943         if (IS_ERR(kvm))
 944                 goto out;
 945 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 946         INIT_LIST_HEAD(&kvm->irq_routing);
 947         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 948 #endif
 949
 950 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 951         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 952         if (!page) {
 953                 kfree(kvm);
 954                 return ERR_PTR(-ENOMEM);
 955         }
 956         kvm->coalesced_mmio_ring =
 957                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 958 #endif
 959
 960 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 961         {
 962                 int err;
 963                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 964                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 965                 if (err) {
 966 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 967                         put_page(page);
 968 #endif
 969                         kfree(kvm);
 970                         return ERR_PTR(err);
 971                 }
 972         }
 973 #endif
 974
 975         kvm->mm = current->mm;
 976         atomic_inc(&kvm->mm->mm_count);
 977         spin_lock_init(&kvm->mmu_lock);
 978         spin_lock_init(&kvm->requests_lock);
 979         kvm_io_bus_init(&kvm->pio_bus);
 980         kvm_irqfd_init(kvm);
 981         mutex_init(&kvm->lock);
 982         mutex_init(&kvm->irq_lock);
 983         kvm_io_bus_init(&kvm->mmio_bus);
 984         init_rwsem(&kvm->slots_lock);
 985         atomic_set(&kvm->users_count, 1);
 986         spin_lock(&kvm_lock);
 987         list_add(&kvm->vm_list, &vm_list);
 988         spin_unlock(&kvm_lock);
 989 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 990         kvm_coalesced_mmio_init(kvm);
 991 #endif
 992 out:
 993         return kvm;
 994 }
 995
 996 /*
 997  * Free any memory in @free but not in @dont.
 998  */
 999 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
1000                                   struct kvm_memory_slot *dont)
1001 {
1002         if (!dont || free->rmap != dont->rmap)
1003                 vfree(free->rmap);
1004
1005         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
1006                 vfree(free->dirty_bitmap);
1007
1008         if (!dont || free->lpage_info != dont->lpage_info)
1009                 vfree(free->lpage_info);
1010
1011         free->npages = 0;
1012         free->dirty_bitmap = NULL;
1013         free->rmap = NULL;
1014         free->lpage_info = NULL;
1015 }
1016
1017 void kvm_free_physmem(struct kvm *kvm)
1018 {
1019         int i;
1020
1021         for (i = 0; i < kvm->nmemslots; ++i)
1022                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
1023 }
1024
1025 static void kvm_destroy_vm(struct kvm *kvm)
1026 {
1027         struct mm_struct *mm = kvm->mm;
1028
1029         kvm_arch_sync_events(kvm);
1030         spin_lock(&kvm_lock);
1031         list_del(&kvm->vm_list);
1032         spin_unlock(&kvm_lock);
1033         kvm_free_irq_routing(kvm);
1034         kvm_io_bus_destroy(&kvm->pio_bus);
1035         kvm_io_bus_destroy(&kvm->mmio_bus);
1036 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1037         if (kvm->coalesced_mmio_ring != NULL)
1038                 free_page((unsigned long)kvm->coalesced_mmio_ring);
1039 #endif
1040 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1041         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1042 #else
1043         kvm_arch_flush_shadow(kvm);
1044 #endif
1045         kvm_arch_destroy_vm(kvm);
1046         mmdrop(mm);
1047 }
1048
1049 void kvm_get_kvm(struct kvm *kvm)
1050 {
1051         atomic_inc(&kvm->users_count);
1052 }
1053 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1054
1055 void kvm_put_kvm(struct kvm *kvm)
1056 {
1057         if (atomic_dec_and_test(&kvm->users_count))
1058                 kvm_destroy_vm(kvm);
1059 }
1060 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1061
1062
1063 static int kvm_vm_release(struct inode *inode, struct file *filp)
1064 {
1065         struct kvm *kvm = filp->private_data;
1066
1067         kvm_irqfd_release(kvm);
1068
1069         kvm_put_kvm(kvm);
1070         return 0;
1071 }
1072
1073 /*
1074  * Allocate some memory and give it an address in the guest physical address
1075  * space.
1076  *
1077  * Discontiguous memory is allowed, mostly for framebuffers.
1078  *
1079  * Must be called holding mmap_sem for write.
1080  */
1081 int __kvm_set_memory_region(struct kvm *kvm,
1082                             struct kvm_userspace_memory_region *mem,
1083                             int user_alloc)
1084 {
1085         int r;
1086         gfn_t base_gfn;
1087         unsigned long npages, ugfn;
1088         unsigned long largepages, i;
1089         struct kvm_memory_slot *memslot;
1090         struct kvm_memory_slot old, new;
1091
1092         r = -EINVAL;
1093         /* General sanity checks */
1094         if (mem->memory_size & (PAGE_SIZE - 1))
1095                 goto out;
1096         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1097                 goto out;
1098         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
1099                 goto out;
1100         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
1101                 goto out;
1102         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1103                 goto out;
1104
1105         memslot = &kvm->memslots[mem->slot];
1106         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1107         npages = mem->memory_size >> PAGE_SHIFT;
1108
1109         if (!npages)
1110                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1111
1112         new = old = *memslot;
1113
1114         new.base_gfn = base_gfn;
1115         new.npages = npages;
1116         new.flags = mem->flags;
1117
1118         /* Disallow changing a memory slot's size. */
1119         r = -EINVAL;
1120         if (npages && old.npages && npages != old.npages)
1121                 goto out_free;
1122
1123         /* Check for overlaps */
1124         r = -EEXIST;
1125         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1126                 struct kvm_memory_slot *s = &kvm->memslots[i];
1127
1128                 if (s == memslot || !s->npages)
1129                         continue;
1130                 if (!((base_gfn + npages <= s->base_gfn) ||
1131                       (base_gfn >= s->base_gfn + s->npages)))
1132                         goto out_free;
1133         }
1134
1135         /* Free page dirty bitmap if unneeded */
1136         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1137                 new.dirty_bitmap = NULL;
1138
1139         r = -ENOMEM;
1140
1141         /* Allocate if a slot is being created */
1142 #ifndef CONFIG_S390
1143         if (npages && !new.rmap) {
1144                 new.rmap = vmalloc(npages * sizeof(struct page *));
1145
1146                 if (!new.rmap)
1147                         goto out_free;
1148
1149                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1150
1151                 new.user_alloc = user_alloc;
1152                 /*
1153                  * hva_to_rmmap() serialzies with the mmu_lock and to be
1154                  * safe it has to ignore memslots with !user_alloc &&
1155                  * !userspace_addr.
1156                  */
1157                 if (user_alloc)
1158                         new.userspace_addr = mem->userspace_addr;
1159                 else
1160                         new.userspace_addr = 0;
1161         }
1162         if (npages && !new.lpage_info) {
1163                 largepages = 1 + (base_gfn + npages - 1) / KVM_PAGES_PER_HPAGE;
1164                 largepages -= base_gfn / KVM_PAGES_PER_HPAGE;
1165
1166                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1167
1168                 if (!new.lpage_info)
1169                         goto out_free;
1170
1171                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1172
1173                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1174                         new.lpage_info[0].write_count = 1;
1175                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1176                         new.lpage_info[largepages-1].write_count = 1;
1177                 ugfn = new.userspace_addr >> PAGE_SHIFT;
1178                 /*
1179                  * If the gfn and userspace address are not aligned wrt each
1180                  * other, disable large page support for this slot
1181                  */
1182                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE - 1))
1183                         for (i = 0; i < largepages; ++i)
1184                                 new.lpage_info[i].write_count = 1;
1185         }
1186
1187         /* Allocate page dirty bitmap if needed */
1188         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1189                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1190
1191                 new.dirty_bitmap = vmalloc(dirty_bytes);
1192                 if (!new.dirty_bitmap)
1193                         goto out_free;
1194                 memset(new.dirty_bitmap, 0, dirty_bytes);
1195                 if (old.npages)
1196                         kvm_arch_flush_shadow(kvm);
1197         }
1198 #endif /* not defined CONFIG_S390 */
1199
1200         if (!npages)
1201                 kvm_arch_flush_shadow(kvm);
1202
1203         spin_lock(&kvm->mmu_lock);
1204         if (mem->slot >= kvm->nmemslots)
1205                 kvm->nmemslots = mem->slot + 1;
1206
1207         *memslot = new;
1208         spin_unlock(&kvm->mmu_lock);
1209
1210         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1211         if (r) {
1212                 spin_lock(&kvm->mmu_lock);
1213                 *memslot = old;
1214                 spin_unlock(&kvm->mmu_lock);
1215                 goto out_free;
1216         }
1217
1218         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1219         /* Slot deletion case: we have to update the current slot */
1220         spin_lock(&kvm->mmu_lock);
1221         if (!npages)
1222                 *memslot = old;
1223         spin_unlock(&kvm->mmu_lock);
1224 #ifdef CONFIG_DMAR
1225         /* map the pages in iommu page table */
1226         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1227         if (r)
1228                 goto out;
1229 #endif
1230         return 0;
1231
1232 out_free:
1233         kvm_free_physmem_slot(&new, &old);
1234 out:
1235         return r;
1236
1237 }
1238 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1239
1240 int kvm_set_memory_region(struct kvm *kvm,
1241                           struct kvm_userspace_memory_region *mem,
1242                           int user_alloc)
1243 {
1244         int r;
1245
1246         down_write(&kvm->slots_lock);
1247         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1248         up_write(&kvm->slots_lock);
1249         return r;
1250 }
1251 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1252
1253 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1254                                    struct
1255                                    kvm_userspace_memory_region *mem,
1256                                    int user_alloc)
1257 {
1258         if (mem->slot >= KVM_MEMORY_SLOTS)
1259                 return -EINVAL;
1260         return kvm_set_memory_region(kvm, mem, user_alloc);
1261 }
1262
1263 int kvm_get_dirty_log(struct kvm *kvm,
1264                         struct kvm_dirty_log *log, int *is_dirty)
1265 {
1266         struct kvm_memory_slot *memslot;
1267         int r, i;
1268         int n;
1269         unsigned long any = 0;
1270
1271         r = -EINVAL;
1272         if (log->slot >= KVM_MEMORY_SLOTS)
1273                 goto out;
1274
1275         memslot = &kvm->memslots[log->slot];
1276         r = -ENOENT;
1277         if (!memslot->dirty_bitmap)
1278                 goto out;
1279
1280         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1281
1282         for (i = 0; !any && i < n/sizeof(long); ++i)
1283                 any = memslot->dirty_bitmap[i];
1284
1285         r = -EFAULT;
1286         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1287                 goto out;
1288
1289         if (any)
1290                 *is_dirty = 1;
1291
1292         r = 0;
1293 out:
1294         return r;
1295 }
1296
1297 int is_error_page(struct page *page)
1298 {
1299         return page == bad_page;
1300 }
1301 EXPORT_SYMBOL_GPL(is_error_page);
1302
1303 int is_error_pfn(pfn_t pfn)
1304 {
1305         return pfn == bad_pfn;
1306 }
1307 EXPORT_SYMBOL_GPL(is_error_pfn);
1308
1309 static inline unsigned long bad_hva(void)
1310 {
1311         return PAGE_OFFSET;
1312 }
1313
1314 int kvm_is_error_hva(unsigned long addr)
1315 {
1316         return addr == bad_hva();
1317 }
1318 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1319
1320 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1321 {
1322         int i;
1323
1324         for (i = 0; i < kvm->nmemslots; ++i) {
1325                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1326
1327                 if (gfn >= memslot->base_gfn
1328                     && gfn < memslot->base_gfn + memslot->npages)
1329                         return memslot;
1330         }
1331         return NULL;
1332 }
1333 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1334
1335 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1336 {
1337         gfn = unalias_gfn(kvm, gfn);
1338         return gfn_to_memslot_unaliased(kvm, gfn);
1339 }
1340
1341 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1342 {
1343         int i;
1344
1345         gfn = unalias_gfn(kvm, gfn);
1346         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1347                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1348
1349                 if (gfn >= memslot->base_gfn
1350                     && gfn < memslot->base_gfn + memslot->npages)
1351                         return 1;
1352         }
1353         return 0;
1354 }
1355 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1356
1357 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1358 {
1359         struct kvm_memory_slot *slot;
1360
1361         gfn = unalias_gfn(kvm, gfn);
1362         slot = gfn_to_memslot_unaliased(kvm, gfn);
1363         if (!slot)
1364                 return bad_hva();
1365         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1366 }
1367 EXPORT_SYMBOL_GPL(gfn_to_hva);
1368
1369 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1370 {
1371         struct page *page[1];
1372         unsigned long addr;
1373         int npages;
1374         pfn_t pfn;
1375
1376         might_sleep();
1377
1378         addr = gfn_to_hva(kvm, gfn);
1379         if (kvm_is_error_hva(addr)) {
1380                 get_page(bad_page);
1381                 return page_to_pfn(bad_page);
1382         }
1383
1384         npages = get_user_pages_fast(addr, 1, 1, page);
1385
1386         if (unlikely(npages != 1)) {
1387                 struct vm_area_struct *vma;
1388
1389                 down_read(&current->mm->mmap_sem);
1390                 vma = find_vma(current->mm, addr);
1391
1392                 if (vma == NULL || addr < vma->vm_start ||
1393                     !(vma->vm_flags & VM_PFNMAP)) {
1394                         up_read(&current->mm->mmap_sem);
1395                         get_page(bad_page);
1396                         return page_to_pfn(bad_page);
1397                 }
1398
1399                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1400                 up_read(&current->mm->mmap_sem);
1401                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1402         } else
1403                 pfn = page_to_pfn(page[0]);
1404
1405         return pfn;
1406 }
1407
1408 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1409
1410 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1411 {
1412         pfn_t pfn;
1413
1414         pfn = gfn_to_pfn(kvm, gfn);
1415         if (!kvm_is_mmio_pfn(pfn))
1416                 return pfn_to_page(pfn);
1417
1418         WARN_ON(kvm_is_mmio_pfn(pfn));
1419
1420         get_page(bad_page);
1421         return bad_page;
1422 }
1423
1424 EXPORT_SYMBOL_GPL(gfn_to_page);
1425
1426 void kvm_release_page_clean(struct page *page)
1427 {
1428         kvm_release_pfn_clean(page_to_pfn(page));
1429 }
1430 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1431
1432 void kvm_release_pfn_clean(pfn_t pfn)
1433 {
1434         if (!kvm_is_mmio_pfn(pfn))
1435                 put_page(pfn_to_page(pfn));
1436 }
1437 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1438
1439 void kvm_release_page_dirty(struct page *page)
1440 {
1441         kvm_release_pfn_dirty(page_to_pfn(page));
1442 }
1443 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1444
1445 void kvm_release_pfn_dirty(pfn_t pfn)
1446 {
1447         kvm_set_pfn_dirty(pfn);
1448         kvm_release_pfn_clean(pfn);
1449 }
1450 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1451
1452 void kvm_set_page_dirty(struct page *page)
1453 {
1454         kvm_set_pfn_dirty(page_to_pfn(page));
1455 }
1456 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1457
1458 void kvm_set_pfn_dirty(pfn_t pfn)
1459 {
1460         if (!kvm_is_mmio_pfn(pfn)) {
1461                 struct page *page = pfn_to_page(pfn);
1462                 if (!PageReserved(page))
1463                         SetPageDirty(page);
1464         }
1465 }
1466 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1467
1468 void kvm_set_pfn_accessed(pfn_t pfn)
1469 {
1470         if (!kvm_is_mmio_pfn(pfn))
1471                 mark_page_accessed(pfn_to_page(pfn));
1472 }
1473 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1474
1475 void kvm_get_pfn(pfn_t pfn)
1476 {
1477         if (!kvm_is_mmio_pfn(pfn))
1478                 get_page(pfn_to_page(pfn));
1479 }
1480 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1481
1482 static int next_segment(unsigned long len, int offset)
1483 {
1484         if (len > PAGE_SIZE - offset)
1485                 return PAGE_SIZE - offset;
1486         else
1487                 return len;
1488 }
1489
1490 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1491                         int len)
1492 {
1493         int r;
1494         unsigned long addr;
1495
1496         addr = gfn_to_hva(kvm, gfn);
1497         if (kvm_is_error_hva(addr))
1498                 return -EFAULT;
1499         r = copy_from_user(data, (void __user *)addr + offset, len);
1500         if (r)
1501                 return -EFAULT;
1502         return 0;
1503 }
1504 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1505
1506 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1507 {
1508         gfn_t gfn = gpa >> PAGE_SHIFT;
1509         int seg;
1510         int offset = offset_in_page(gpa);
1511         int ret;
1512
1513         while ((seg = next_segment(len, offset)) != 0) {
1514                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1515                 if (ret < 0)
1516                         return ret;
1517                 offset = 0;
1518                 len -= seg;
1519                 data += seg;
1520                 ++gfn;
1521         }
1522         return 0;
1523 }
1524 EXPORT_SYMBOL_GPL(kvm_read_guest);
1525
1526 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1527                           unsigned long len)
1528 {
1529         int r;
1530         unsigned long addr;
1531         gfn_t gfn = gpa >> PAGE_SHIFT;
1532         int offset = offset_in_page(gpa);
1533
1534         addr = gfn_to_hva(kvm, gfn);
1535         if (kvm_is_error_hva(addr))
1536                 return -EFAULT;
1537         pagefault_disable();
1538         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1539         pagefault_enable();
1540         if (r)
1541                 return -EFAULT;
1542         return 0;
1543 }
1544 EXPORT_SYMBOL(kvm_read_guest_atomic);
1545
1546 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1547                          int offset, int len)
1548 {
1549         int r;
1550         unsigned long addr;
1551
1552         addr = gfn_to_hva(kvm, gfn);
1553         if (kvm_is_error_hva(addr))
1554                 return -EFAULT;
1555         r = copy_to_user((void __user *)addr + offset, data, len);
1556         if (r)
1557                 return -EFAULT;
1558         mark_page_dirty(kvm, gfn);
1559         return 0;
1560 }
1561 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1562
1563 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1564                     unsigned long len)
1565 {
1566         gfn_t gfn = gpa >> PAGE_SHIFT;
1567         int seg;
1568         int offset = offset_in_page(gpa);
1569         int ret;
1570
1571         while ((seg = next_segment(len, offset)) != 0) {
1572                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1573                 if (ret < 0)
1574                         return ret;
1575                 offset = 0;
1576                 len -= seg;
1577                 data += seg;
1578                 ++gfn;
1579         }
1580         return 0;
1581 }
1582
1583 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1584 {
1585         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1586 }
1587 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1588
1589 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1590 {
1591         gfn_t gfn = gpa >> PAGE_SHIFT;
1592         int seg;
1593         int offset = offset_in_page(gpa);
1594         int ret;
1595
1596         while ((seg = next_segment(len, offset)) != 0) {
1597                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1598                 if (ret < 0)
1599                         return ret;
1600                 offset = 0;
1601                 len -= seg;
1602                 ++gfn;
1603         }
1604         return 0;
1605 }
1606 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1607
1608 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1609 {
1610         struct kvm_memory_slot *memslot;
1611
1612         gfn = unalias_gfn(kvm, gfn);
1613         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1614         if (memslot && memslot->dirty_bitmap) {
1615                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1616
1617                 /* avoid RMW */
1618                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1619                         set_bit(rel_gfn, memslot->dirty_bitmap);
1620         }
1621 }
1622
1623 /*
1624  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1625  */
1626 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1627 {
1628         DEFINE_WAIT(wait);
1629
1630         for (;;) {
1631                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1632
1633                 if ((kvm_arch_interrupt_allowed(vcpu) &&
1634                                         kvm_cpu_has_interrupt(vcpu)) ||
1635                                 kvm_arch_vcpu_runnable(vcpu)) {
1636                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1637                         break;
1638                 }
1639                 if (kvm_cpu_has_pending_timer(vcpu))
1640                         break;
1641                 if (signal_pending(current))
1642                         break;
1643
1644                 vcpu_put(vcpu);
1645                 schedule();
1646                 vcpu_load(vcpu);
1647         }
1648
1649         finish_wait(&vcpu->wq, &wait);
1650 }
1651
1652 void kvm_resched(struct kvm_vcpu *vcpu)
1653 {
1654         if (!need_resched())
1655                 return;
1656         cond_resched();
1657 }
1658 EXPORT_SYMBOL_GPL(kvm_resched);
1659
1660 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1661 {
1662         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1663         struct page *page;
1664
1665         if (vmf->pgoff == 0)
1666                 page = virt_to_page(vcpu->run);
1667 #ifdef CONFIG_X86
1668         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1669                 page = virt_to_page(vcpu->arch.pio_data);
1670 #endif
1671 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1672         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1673                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1674 #endif
1675         else
1676                 return VM_FAULT_SIGBUS;
1677         get_page(page);
1678         vmf->page = page;
1679         return 0;
1680 }
1681
1682 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1683         .fault = kvm_vcpu_fault,
1684 };
1685
1686 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1687 {
1688         vma->vm_ops = &kvm_vcpu_vm_ops;
1689         return 0;
1690 }
1691
1692 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1693 {
1694         struct kvm_vcpu *vcpu = filp->private_data;
1695
1696         kvm_put_kvm(vcpu->kvm);
1697         return 0;
1698 }
1699
1700 static struct file_operations kvm_vcpu_fops = {
1701         .release        = kvm_vcpu_release,
1702         .unlocked_ioctl = kvm_vcpu_ioctl,
1703         .compat_ioctl   = kvm_vcpu_ioctl,
1704         .mmap           = kvm_vcpu_mmap,
1705 };
1706
1707 /*
1708  * Allocates an inode for the vcpu.
1709  */
1710 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1711 {
1712         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1713 }
1714
1715 /*
1716  * Creates some virtual cpus.  Good luck creating more than one.
1717  */
1718 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1719 {
1720         int r;
1721         struct kvm_vcpu *vcpu;
1722
1723         vcpu = kvm_arch_vcpu_create(kvm, id);
1724         if (IS_ERR(vcpu))
1725                 return PTR_ERR(vcpu);
1726
1727         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1728
1729         r = kvm_arch_vcpu_setup(vcpu);
1730         if (r)
1731                 return r;
1732
1733         mutex_lock(&kvm->lock);
1734         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1735                 r = -EINVAL;
1736                 goto vcpu_destroy;
1737         }
1738
1739         for (r = 0; r < atomic_read(&kvm->online_vcpus); r++)
1740                 if (kvm->vcpus[r]->vcpu_id == id) {
1741                         r = -EEXIST;
1742                         goto vcpu_destroy;
1743                 }
1744
1745         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1746
1747         /* Now it's all set up, let userspace reach it */
1748         kvm_get_kvm(kvm);
1749         r = create_vcpu_fd(vcpu);
1750         if (r < 0) {
1751                 kvm_put_kvm(kvm);
1752                 goto vcpu_destroy;
1753         }
1754
1755         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1756         smp_wmb();
1757         atomic_inc(&kvm->online_vcpus);
1758
1759 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1760         if (kvm->bsp_vcpu_id == id)
1761                 kvm->bsp_vcpu = vcpu;
1762 #endif
1763         mutex_unlock(&kvm->lock);
1764         return r;
1765
1766 vcpu_destroy:
1767         mutex_unlock(&kvm->lock);
1768         kvm_arch_vcpu_destroy(vcpu);
1769         return r;
1770 }
1771
1772 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1773 {
1774         if (sigset) {
1775                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1776                 vcpu->sigset_active = 1;
1777                 vcpu->sigset = *sigset;
1778         } else
1779                 vcpu->sigset_active = 0;
1780         return 0;
1781 }
1782
1783 #ifdef __KVM_HAVE_MSIX
1784 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1785                                     struct kvm_assigned_msix_nr *entry_nr)
1786 {
1787         int r = 0;
1788         struct kvm_assigned_dev_kernel *adev;
1789
1790         mutex_lock(&kvm->lock);
1791
1792         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1793                                       entry_nr->assigned_dev_id);
1794         if (!adev) {
1795                 r = -EINVAL;
1796                 goto msix_nr_out;
1797         }
1798
1799         if (adev->entries_nr == 0) {
1800                 adev->entries_nr = entry_nr->entry_nr;
1801                 if (adev->entries_nr == 0 ||
1802                     adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1803                         r = -EINVAL;
1804                         goto msix_nr_out;
1805                 }
1806
1807                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1808                                                 entry_nr->entry_nr,
1809                                                 GFP_KERNEL);
1810                 if (!adev->host_msix_entries) {
1811                         r = -ENOMEM;
1812                         goto msix_nr_out;
1813                 }
1814                 adev->guest_msix_entries = kzalloc(
1815                                 sizeof(struct kvm_guest_msix_entry) *
1816                                 entry_nr->entry_nr, GFP_KERNEL);
1817                 if (!adev->guest_msix_entries) {
1818                         kfree(adev->host_msix_entries);
1819                         r = -ENOMEM;
1820                         goto msix_nr_out;
1821                 }
1822         } else /* Not allowed set MSI-X number twice */
1823                 r = -EINVAL;
1824 msix_nr_out:
1825         mutex_unlock(&kvm->lock);
1826         return r;
1827 }
1828
1829 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1830                                        struct kvm_assigned_msix_entry *entry)
1831 {
1832         int r = 0, i;
1833         struct kvm_assigned_dev_kernel *adev;
1834
1835         mutex_lock(&kvm->lock);
1836
1837         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1838                                       entry->assigned_dev_id);
1839
1840         if (!adev) {
1841                 r = -EINVAL;
1842                 goto msix_entry_out;
1843         }
1844
1845         for (i = 0; i < adev->entries_nr; i++)
1846                 if (adev->guest_msix_entries[i].vector == 0 ||
1847                     adev->guest_msix_entries[i].entry == entry->entry) {
1848                         adev->guest_msix_entries[i].entry = entry->entry;
1849                         adev->guest_msix_entries[i].vector = entry->gsi;
1850                         adev->host_msix_entries[i].entry = entry->entry;
1851                         break;
1852                 }
1853         if (i == adev->entries_nr) {
1854                 r = -ENOSPC;
1855                 goto msix_entry_out;
1856         }
1857
1858 msix_entry_out:
1859         mutex_unlock(&kvm->lock);
1860
1861         return r;
1862 }
1863 #endif
1864
1865 static long kvm_vcpu_ioctl(struct file *filp,
1866                            unsigned int ioctl, unsigned long arg)
1867 {
1868         struct kvm_vcpu *vcpu = filp->private_data;
1869         void __user *argp = (void __user *)arg;
1870         int r;
1871         struct kvm_fpu *fpu = NULL;
1872         struct kvm_sregs *kvm_sregs = NULL;
1873
1874         if (vcpu->kvm->mm != current->mm)
1875                 return -EIO;
1876         switch (ioctl) {
1877         case KVM_RUN:
1878                 r = -EINVAL;
1879                 if (arg)
1880                         goto out;
1881                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1882                 break;
1883         case KVM_GET_REGS: {
1884                 struct kvm_regs *kvm_regs;
1885
1886                 r = -ENOMEM;
1887                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1888                 if (!kvm_regs)
1889                         goto out;
1890                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1891                 if (r)
1892                         goto out_free1;
1893                 r = -EFAULT;
1894                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1895                         goto out_free1;
1896                 r = 0;
1897 out_free1:
1898                 kfree(kvm_regs);
1899                 break;
1900         }
1901         case KVM_SET_REGS: {
1902                 struct kvm_regs *kvm_regs;
1903
1904                 r = -ENOMEM;
1905                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1906                 if (!kvm_regs)
1907                         goto out;
1908                 r = -EFAULT;
1909                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1910                         goto out_free2;
1911                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1912                 if (r)
1913                         goto out_free2;
1914                 r = 0;
1915 out_free2:
1916                 kfree(kvm_regs);
1917                 break;
1918         }
1919         case KVM_GET_SREGS: {
1920                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1921                 r = -ENOMEM;
1922                 if (!kvm_sregs)
1923                         goto out;
1924                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1925                 if (r)
1926                         goto out;
1927                 r = -EFAULT;
1928                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1929                         goto out;
1930                 r = 0;
1931                 break;
1932         }
1933         case KVM_SET_SREGS: {
1934                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1935                 r = -ENOMEM;
1936                 if (!kvm_sregs)
1937                         goto out;
1938                 r = -EFAULT;
1939                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1940                         goto out;
1941                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1942                 if (r)
1943                         goto out;
1944                 r = 0;
1945                 break;
1946         }
1947         case KVM_GET_MP_STATE: {
1948                 struct kvm_mp_state mp_state;
1949
1950                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1951                 if (r)
1952                         goto out;
1953                 r = -EFAULT;
1954                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1955                         goto out;
1956                 r = 0;
1957                 break;
1958         }
1959         case KVM_SET_MP_STATE: {
1960                 struct kvm_mp_state mp_state;
1961
1962                 r = -EFAULT;
1963                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1964                         goto out;
1965                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1966                 if (r)
1967                         goto out;
1968                 r = 0;
1969                 break;
1970         }
1971         case KVM_TRANSLATE: {
1972                 struct kvm_translation tr;
1973
1974                 r = -EFAULT;
1975                 if (copy_from_user(&tr, argp, sizeof tr))
1976                         goto out;
1977                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1978                 if (r)
1979                         goto out;
1980                 r = -EFAULT;
1981                 if (copy_to_user(argp, &tr, sizeof tr))
1982                         goto out;
1983                 r = 0;
1984                 break;
1985         }
1986         case KVM_SET_GUEST_DEBUG: {
1987                 struct kvm_guest_debug dbg;
1988
1989                 r = -EFAULT;
1990                 if (copy_from_user(&dbg, argp, sizeof dbg))
1991                         goto out;
1992                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1993                 if (r)
1994                         goto out;
1995                 r = 0;
1996                 break;
1997         }
1998         case KVM_SET_SIGNAL_MASK: {
1999                 struct kvm_signal_mask __user *sigmask_arg = argp;
2000                 struct kvm_signal_mask kvm_sigmask;
2001                 sigset_t sigset, *p;
2002
2003                 p = NULL;
2004                 if (argp) {
2005                         r = -EFAULT;
2006                         if (copy_from_user(&kvm_sigmask, argp,
2007                                            sizeof kvm_sigmask))
2008                                 goto out;
2009                         r = -EINVAL;
2010                         if (kvm_sigmask.len != sizeof sigset)
2011                                 goto out;
2012                         r = -EFAULT;
2013                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2014                                            sizeof sigset))
2015                                 goto out;
2016                         p = &sigset;
2017                 }
2018                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2019                 break;
2020         }
2021         case KVM_GET_FPU: {
2022                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2023                 r = -ENOMEM;
2024                 if (!fpu)
2025                         goto out;
2026                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2027                 if (r)
2028                         goto out;
2029                 r = -EFAULT;
2030                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2031                         goto out;
2032                 r = 0;
2033                 break;
2034         }
2035         case KVM_SET_FPU: {
2036                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2037                 r = -ENOMEM;
2038                 if (!fpu)
2039                         goto out;
2040                 r = -EFAULT;
2041                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2042                         goto out;
2043                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2044                 if (r)
2045                         goto out;
2046                 r = 0;
2047                 break;
2048         }
2049         default:
2050                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2051         }
2052 out:
2053         kfree(fpu);
2054         kfree(kvm_sregs);
2055         return r;
2056 }
2057
2058 static long kvm_vm_ioctl(struct file *filp,
2059                            unsigned int ioctl, unsigned long arg)
2060 {
2061         struct kvm *kvm = filp->private_data;
2062         void __user *argp = (void __user *)arg;
2063         int r;
2064
2065         if (kvm->mm != current->mm)
2066                 return -EIO;
2067         switch (ioctl) {
2068         case KVM_CREATE_VCPU:
2069                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2070                 if (r < 0)
2071                         goto out;
2072                 break;
2073         case KVM_SET_USER_MEMORY_REGION: {
2074                 struct kvm_userspace_memory_region kvm_userspace_mem;
2075
2076                 r = -EFAULT;
2077                 if (copy_from_user(&kvm_userspace_mem, argp,
2078                                                 sizeof kvm_userspace_mem))
2079                         goto out;
2080
2081                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2082                 if (r)
2083                         goto out;
2084                 break;
2085         }
2086         case KVM_GET_DIRTY_LOG: {
2087                 struct kvm_dirty_log log;
2088
2089                 r = -EFAULT;
2090                 if (copy_from_user(&log, argp, sizeof log))
2091                         goto out;
2092                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2093                 if (r)
2094                         goto out;
2095                 break;
2096         }
2097 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2098         case KVM_REGISTER_COALESCED_MMIO: {
2099                 struct kvm_coalesced_mmio_zone zone;
2100                 r = -EFAULT;
2101                 if (copy_from_user(&zone, argp, sizeof zone))
2102                         goto out;
2103                 r = -ENXIO;
2104                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2105                 if (r)
2106                         goto out;
2107                 r = 0;
2108                 break;
2109         }
2110         case KVM_UNREGISTER_COALESCED_MMIO: {
2111                 struct kvm_coalesced_mmio_zone zone;
2112                 r = -EFAULT;
2113                 if (copy_from_user(&zone, argp, sizeof zone))
2114                         goto out;
2115                 r = -ENXIO;
2116                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2117                 if (r)
2118                         goto out;
2119                 r = 0;
2120                 break;
2121         }
2122 #endif
2123 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2124         case KVM_ASSIGN_PCI_DEVICE: {
2125                 struct kvm_assigned_pci_dev assigned_dev;
2126
2127                 r = -EFAULT;
2128                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2129                         goto out;
2130                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2131                 if (r)
2132                         goto out;
2133                 break;
2134         }
2135         case KVM_ASSIGN_IRQ: {
2136                 r = -EOPNOTSUPP;
2137                 break;
2138         }
2139 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2140         case KVM_ASSIGN_DEV_IRQ: {
2141                 struct kvm_assigned_irq assigned_irq;
2142
2143                 r = -EFAULT;
2144                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2145                         goto out;
2146                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2147                 if (r)
2148                         goto out;
2149                 break;
2150         }
2151         case KVM_DEASSIGN_DEV_IRQ: {
2152                 struct kvm_assigned_irq assigned_irq;
2153
2154                 r = -EFAULT;
2155                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2156                         goto out;
2157                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2158                 if (r)
2159                         goto out;
2160                 break;
2161         }
2162 #endif
2163 #endif
2164 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2165         case KVM_DEASSIGN_PCI_DEVICE: {
2166                 struct kvm_assigned_pci_dev assigned_dev;
2167
2168                 r = -EFAULT;
2169                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2170                         goto out;
2171                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2172                 if (r)
2173                         goto out;
2174                 break;
2175         }
2176 #endif
2177 #ifdef KVM_CAP_IRQ_ROUTING
2178         case KVM_SET_GSI_ROUTING: {
2179                 struct kvm_irq_routing routing;
2180                 struct kvm_irq_routing __user *urouting;
2181                 struct kvm_irq_routing_entry *entries;
2182
2183                 r = -EFAULT;
2184                 if (copy_from_user(&routing, argp, sizeof(routing)))
2185                         goto out;
2186                 r = -EINVAL;
2187                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2188                         goto out;
2189                 if (routing.flags)
2190                         goto out;
2191                 r = -ENOMEM;
2192                 entries = vmalloc(routing.nr * sizeof(*entries));
2193                 if (!entries)
2194                         goto out;
2195                 r = -EFAULT;
2196                 urouting = argp;
2197                 if (copy_from_user(entries, urouting->entries,
2198                                    routing.nr * sizeof(*entries)))
2199                         goto out_free_irq_routing;
2200                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2201                                         routing.flags);
2202         out_free_irq_routing:
2203                 vfree(entries);
2204                 break;
2205         }
2206 #ifdef __KVM_HAVE_MSIX
2207         case KVM_ASSIGN_SET_MSIX_NR: {
2208                 struct kvm_assigned_msix_nr entry_nr;
2209                 r = -EFAULT;
2210                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2211                         goto out;
2212                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2213                 if (r)
2214                         goto out;
2215                 break;
2216         }
2217         case KVM_ASSIGN_SET_MSIX_ENTRY: {
2218                 struct kvm_assigned_msix_entry entry;
2219                 r = -EFAULT;
2220                 if (copy_from_user(&entry, argp, sizeof entry))
2221                         goto out;
2222                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2223                 if (r)
2224                         goto out;
2225                 break;
2226         }
2227 #endif
2228 #endif /* KVM_CAP_IRQ_ROUTING */
2229         case KVM_IRQFD: {
2230                 struct kvm_irqfd data;
2231
2232                 r = -EFAULT;
2233                 if (copy_from_user(&data, argp, sizeof data))
2234                         goto out;
2235                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2236                 break;
2237         }
2238 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2239         case KVM_SET_BOOT_CPU_ID:
2240                 r = 0;
2241                 if (atomic_read(&kvm->online_vcpus) != 0)
2242                         r = -EBUSY;
2243                 else
2244                         kvm->bsp_vcpu_id = arg;
2245                 break;
2246 #endif
2247         default:
2248                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2249         }
2250 out:
2251         return r;
2252 }
2253
2254 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2255 {
2256         struct page *page[1];
2257         unsigned long addr;
2258         int npages;
2259         gfn_t gfn = vmf->pgoff;
2260         struct kvm *kvm = vma->vm_file->private_data;
2261
2262         addr = gfn_to_hva(kvm, gfn);
2263         if (kvm_is_error_hva(addr))
2264                 return VM_FAULT_SIGBUS;
2265
2266         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2267                                 NULL);
2268         if (unlikely(npages != 1))
2269                 return VM_FAULT_SIGBUS;
2270
2271         vmf->page = page[0];
2272         return 0;
2273 }
2274
2275 static struct vm_operations_struct kvm_vm_vm_ops = {
2276         .fault = kvm_vm_fault,
2277 };
2278
2279 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2280 {
2281         vma->vm_ops = &kvm_vm_vm_ops;
2282         return 0;
2283 }
2284
2285 static struct file_operations kvm_vm_fops = {
2286         .release        = kvm_vm_release,
2287         .unlocked_ioctl = kvm_vm_ioctl,
2288         .compat_ioctl   = kvm_vm_ioctl,
2289         .mmap           = kvm_vm_mmap,
2290 };
2291
2292 static int kvm_dev_ioctl_create_vm(void)
2293 {
2294         int fd;
2295         struct kvm *kvm;
2296
2297         kvm = kvm_create_vm();
2298         if (IS_ERR(kvm))
2299                 return PTR_ERR(kvm);
2300         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2301         if (fd < 0)
2302                 kvm_put_kvm(kvm);
2303
2304         return fd;
2305 }
2306
2307 static long kvm_dev_ioctl_check_extension_generic(long arg)
2308 {
2309         switch (arg) {
2310         case KVM_CAP_USER_MEMORY:
2311         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2312         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2313 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2314         case KVM_CAP_SET_BOOT_CPU_ID:
2315 #endif
2316                 return 1;
2317 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2318         case KVM_CAP_IRQ_ROUTING:
2319                 return KVM_MAX_IRQ_ROUTES;
2320 #endif
2321         default:
2322                 break;
2323         }
2324         return kvm_dev_ioctl_check_extension(arg);
2325 }
2326
2327 static long kvm_dev_ioctl(struct file *filp,
2328                           unsigned int ioctl, unsigned long arg)
2329 {
2330         long r = -EINVAL;
2331
2332         switch (ioctl) {
2333         case KVM_GET_API_VERSION:
2334                 r = -EINVAL;
2335                 if (arg)
2336                         goto out;
2337                 r = KVM_API_VERSION;
2338                 break;
2339         case KVM_CREATE_VM:
2340                 r = -EINVAL;
2341                 if (arg)
2342                         goto out;
2343                 r = kvm_dev_ioctl_create_vm();
2344                 break;
2345         case KVM_CHECK_EXTENSION:
2346                 r = kvm_dev_ioctl_check_extension_generic(arg);
2347                 break;
2348         case KVM_GET_VCPU_MMAP_SIZE:
2349                 r = -EINVAL;
2350                 if (arg)
2351                         goto out;
2352                 r = PAGE_SIZE;     /* struct kvm_run */
2353 #ifdef CONFIG_X86
2354                 r += PAGE_SIZE;    /* pio data page */
2355 #endif
2356 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2357                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2358 #endif
2359                 break;
2360         case KVM_TRACE_ENABLE:
2361         case KVM_TRACE_PAUSE:
2362         case KVM_TRACE_DISABLE:
2363                 r = kvm_trace_ioctl(ioctl, arg);
2364                 break;
2365         default:
2366                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2367         }
2368 out:
2369         return r;
2370 }
2371
2372 static struct file_operations kvm_chardev_ops = {
2373         .unlocked_ioctl = kvm_dev_ioctl,
2374         .compat_ioctl   = kvm_dev_ioctl,
2375 };
2376
2377 static struct miscdevice kvm_dev = {
2378         KVM_MINOR,
2379         "kvm",
2380         &kvm_chardev_ops,
2381 };
2382
2383 static void hardware_enable(void *junk)
2384 {
2385         int cpu = raw_smp_processor_id();
2386
2387         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2388                 return;
2389         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2390         kvm_arch_hardware_enable(NULL);
2391 }
2392
2393 static void hardware_disable(void *junk)
2394 {
2395         int cpu = raw_smp_processor_id();
2396
2397         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2398                 return;
2399         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2400         kvm_arch_hardware_disable(NULL);
2401 }
2402
2403 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2404                            void *v)
2405 {
2406         int cpu = (long)v;
2407
2408         val &= ~CPU_TASKS_FROZEN;
2409         switch (val) {
2410         case CPU_DYING:
2411                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2412                        cpu);
2413                 hardware_disable(NULL);
2414                 break;
2415         case CPU_UP_CANCELED:
2416                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2417                        cpu);
2418                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2419                 break;
2420         case CPU_ONLINE:
2421                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2422                        cpu);
2423                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2424                 break;
2425         }
2426         return NOTIFY_OK;
2427 }
2428
2429
2430 asmlinkage void kvm_handle_fault_on_reboot(void)
2431 {
2432         if (kvm_rebooting)
2433                 /* spin while reset goes on */
2434                 while (true)
2435                         ;
2436         /* Fault while not rebooting.  We want the trace. */
2437         BUG();
2438 }
2439 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2440
2441 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2442                       void *v)
2443 {
2444         /*
2445          * Some (well, at least mine) BIOSes hang on reboot if
2446          * in vmx root mode.
2447          *
2448          * And Intel TXT required VMX off for all cpu when system shutdown.
2449          */
2450         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2451         kvm_rebooting = true;
2452         on_each_cpu(hardware_disable, NULL, 1);
2453         return NOTIFY_OK;
2454 }
2455
2456 static struct notifier_block kvm_reboot_notifier = {
2457         .notifier_call = kvm_reboot,
2458         .priority = 0,
2459 };
2460
2461 void kvm_io_bus_init(struct kvm_io_bus *bus)
2462 {
2463         memset(bus, 0, sizeof(*bus));
2464 }
2465
2466 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2467 {
2468         int i;
2469
2470         for (i = 0; i < bus->dev_count; i++) {
2471                 struct kvm_io_device *pos = bus->devs[i];
2472
2473                 kvm_iodevice_destructor(pos);
2474         }
2475 }
2476
2477 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2478                                           gpa_t addr, int len, int is_write)
2479 {
2480         int i;
2481
2482         for (i = 0; i < bus->dev_count; i++) {
2483                 struct kvm_io_device *pos = bus->devs[i];
2484
2485                 if (kvm_iodevice_in_range(pos, addr, len, is_write))
2486                         return pos;
2487         }
2488
2489         return NULL;
2490 }
2491
2492 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2493 {
2494         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2495
2496         bus->devs[bus->dev_count++] = dev;
2497 }
2498
2499 static struct notifier_block kvm_cpu_notifier = {
2500         .notifier_call = kvm_cpu_hotplug,
2501         .priority = 20, /* must be > scheduler priority */
2502 };
2503
2504 static int vm_stat_get(void *_offset, u64 *val)
2505 {
2506         unsigned offset = (long)_offset;
2507         struct kvm *kvm;
2508
2509         *val = 0;
2510         spin_lock(&kvm_lock);
2511         list_for_each_entry(kvm, &vm_list, vm_list)
2512                 *val += *(u32 *)((void *)kvm + offset);
2513         spin_unlock(&kvm_lock);
2514         return 0;
2515 }
2516
2517 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2518
2519 static int vcpu_stat_get(void *_offset, u64 *val)
2520 {
2521         unsigned offset = (long)_offset;
2522         struct kvm *kvm;
2523         struct kvm_vcpu *vcpu;
2524         int i;
2525
2526         *val = 0;
2527         spin_lock(&kvm_lock);
2528         list_for_each_entry(kvm, &vm_list, vm_list)
2529                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2530                         vcpu = kvm->vcpus[i];
2531                         if (vcpu)
2532                                 *val += *(u32 *)((void *)vcpu + offset);
2533                 }
2534         spin_unlock(&kvm_lock);
2535         return 0;
2536 }
2537
2538 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2539
2540 static struct file_operations *stat_fops[] = {
2541         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2542         [KVM_STAT_VM]   = &vm_stat_fops,
2543 };
2544
2545 static void kvm_init_debug(void)
2546 {
2547         struct kvm_stats_debugfs_item *p;
2548
2549         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2550         for (p = debugfs_entries; p->name; ++p)
2551                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2552                                                 (void *)(long)p->offset,
2553                                                 stat_fops[p->kind]);
2554 }
2555
2556 static void kvm_exit_debug(void)
2557 {
2558         struct kvm_stats_debugfs_item *p;
2559
2560         for (p = debugfs_entries; p->name; ++p)
2561                 debugfs_remove(p->dentry);
2562         debugfs_remove(kvm_debugfs_dir);
2563 }
2564
2565 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2566 {
2567         hardware_disable(NULL);
2568         return 0;
2569 }
2570
2571 static int kvm_resume(struct sys_device *dev)
2572 {
2573         hardware_enable(NULL);
2574         return 0;
2575 }
2576
2577 static struct sysdev_class kvm_sysdev_class = {
2578         .name = "kvm",
2579         .suspend = kvm_suspend,
2580         .resume = kvm_resume,
2581 };
2582
2583 static struct sys_device kvm_sysdev = {
2584         .id = 0,
2585         .cls = &kvm_sysdev_class,
2586 };
2587
2588 struct page *bad_page;
2589 pfn_t bad_pfn;
2590
2591 static inline
2592 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2593 {
2594         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2595 }
2596
2597 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2598 {
2599         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2600
2601         kvm_arch_vcpu_load(vcpu, cpu);
2602 }
2603
2604 static void kvm_sched_out(struct preempt_notifier *pn,
2605                           struct task_struct *next)
2606 {
2607         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2608
2609         kvm_arch_vcpu_put(vcpu);
2610 }
2611
2612 int kvm_init(void *opaque, unsigned int vcpu_size,
2613                   struct module *module)
2614 {
2615         int r;
2616         int cpu;
2617
2618         kvm_init_debug();
2619
2620         r = kvm_arch_init(opaque);
2621         if (r)
2622                 goto out_fail;
2623
2624         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2625
2626         if (bad_page == NULL) {
2627                 r = -ENOMEM;
2628                 goto out;
2629         }
2630
2631         bad_pfn = page_to_pfn(bad_page);
2632
2633         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2634                 r = -ENOMEM;
2635                 goto out_free_0;
2636         }
2637
2638         r = kvm_arch_hardware_setup();
2639         if (r < 0)
2640                 goto out_free_0a;
2641
2642         for_each_online_cpu(cpu) {
2643                 smp_call_function_single(cpu,
2644                                 kvm_arch_check_processor_compat,
2645                                 &r, 1);
2646                 if (r < 0)
2647                         goto out_free_1;
2648         }
2649
2650         on_each_cpu(hardware_enable, NULL, 1);
2651         r = register_cpu_notifier(&kvm_cpu_notifier);
2652         if (r)
2653                 goto out_free_2;
2654         register_reboot_notifier(&kvm_reboot_notifier);
2655
2656         r = sysdev_class_register(&kvm_sysdev_class);
2657         if (r)
2658                 goto out_free_3;
2659
2660         r = sysdev_register(&kvm_sysdev);
2661         if (r)
2662                 goto out_free_4;
2663
2664         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2665         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2666                                            __alignof__(struct kvm_vcpu),
2667                                            0, NULL);
2668         if (!kvm_vcpu_cache) {
2669                 r = -ENOMEM;
2670                 goto out_free_5;
2671         }
2672
2673         kvm_chardev_ops.owner = module;
2674         kvm_vm_fops.owner = module;
2675         kvm_vcpu_fops.owner = module;
2676
2677         r = misc_register(&kvm_dev);
2678         if (r) {
2679                 printk(KERN_ERR "kvm: misc device register failed\n");
2680                 goto out_free;
2681         }
2682
2683         kvm_preempt_ops.sched_in = kvm_sched_in;
2684         kvm_preempt_ops.sched_out = kvm_sched_out;
2685
2686         return 0;
2687
2688 out_free:
2689         kmem_cache_destroy(kvm_vcpu_cache);
2690 out_free_5:
2691         sysdev_unregister(&kvm_sysdev);
2692 out_free_4:
2693         sysdev_class_unregister(&kvm_sysdev_class);
2694 out_free_3:
2695         unregister_reboot_notifier(&kvm_reboot_notifier);
2696         unregister_cpu_notifier(&kvm_cpu_notifier);
2697 out_free_2:
2698         on_each_cpu(hardware_disable, NULL, 1);
2699 out_free_1:
2700         kvm_arch_hardware_unsetup();
2701 out_free_0a:
2702         free_cpumask_var(cpus_hardware_enabled);
2703 out_free_0:
2704         __free_page(bad_page);
2705 out:
2706         kvm_arch_exit();
2707         kvm_exit_debug();
2708 out_fail:
2709         return r;
2710 }
2711 EXPORT_SYMBOL_GPL(kvm_init);
2712
2713 void kvm_exit(void)
2714 {
2715         kvm_trace_cleanup();
2716         misc_deregister(&kvm_dev);
2717         kmem_cache_destroy(kvm_vcpu_cache);
2718         sysdev_unregister(&kvm_sysdev);
2719         sysdev_class_unregister(&kvm_sysdev_class);
2720         unregister_reboot_notifier(&kvm_reboot_notifier);
2721         unregister_cpu_notifier(&kvm_cpu_notifier);
2722         on_each_cpu(hardware_disable, NULL, 1);
2723         kvm_arch_hardware_unsetup();
2724         kvm_arch_exit();
2725         kvm_exit_debug();
2726         free_cpumask_var(cpus_hardware_enabled);
2727         __free_page(bad_page);
2728 }
2729 EXPORT_SYMBOL_GPL(kvm_exit);