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

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "iodev.h"
  19
  20 #include <linux/kvm_host.h>
  21 #include <linux/kvm.h>
  22 #include <linux/module.h>
  23 #include <linux/errno.h>
  24 #include <linux/percpu.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/sysdev.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46
  47 #include <asm/processor.h>
  48 #include <asm/io.h>
  49 #include <asm/uaccess.h>
  50 #include <asm/pgtable.h>
  51
  52 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
  53 #include "coalesced_mmio.h"
  54 #endif
  55
  56 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
  57 #include <linux/pci.h>
  58 #include <linux/interrupt.h>
  59 #include "irq.h"
  60 #endif
  61
  62 #define CREATE_TRACE_POINTS
  63 #include <trace/events/kvm.h>
  64
  65 MODULE_AUTHOR("Qumranet");
  66 MODULE_LICENSE("GPL");
  67
  68 /*
  69  * Ordering of locks:
  70  *
  71  *              kvm->lock --> kvm->irq_lock
  72  */
  73
  74 DEFINE_SPINLOCK(kvm_lock);
  75 LIST_HEAD(vm_list);
  76
  77 static cpumask_var_t cpus_hardware_enabled;
  78
  79 struct kmem_cache *kvm_vcpu_cache;
  80 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  81
  82 static __read_mostly struct preempt_ops kvm_preempt_ops;
  83
  84 struct dentry *kvm_debugfs_dir;
  85
  86 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  87                            unsigned long arg);
  88
  89 static bool kvm_rebooting;
  90
  91 static bool largepages_enabled = true;
  92
  93 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
  94 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
  95                                                       int assigned_dev_id)
  96 {
  97         struct list_head *ptr;
  98         struct kvm_assigned_dev_kernel *match;
  99
 100         list_for_each(ptr, head) {
 101                 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
 102                 if (match->assigned_dev_id == assigned_dev_id)
 103                         return match;
 104         }
 105         return NULL;
 106 }
 107
 108 static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
 109                                     *assigned_dev, int irq)
 110 {
 111         int i, index;
 112         struct msix_entry *host_msix_entries;
 113
 114         host_msix_entries = assigned_dev->host_msix_entries;
 115
 116         index = -1;
 117         for (i = 0; i < assigned_dev->entries_nr; i++)
 118                 if (irq == host_msix_entries[i].vector) {
 119                         index = i;
 120                         break;
 121                 }
 122         if (index < 0) {
 123                 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
 124                 return 0;
 125         }
 126
 127         return index;
 128 }
 129
 130 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
 131 {
 132         struct kvm_assigned_dev_kernel *assigned_dev;
 133         struct kvm *kvm;
 134         int i;
 135
 136         assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
 137                                     interrupt_work);
 138         kvm = assigned_dev->kvm;
 139
 140         mutex_lock(&kvm->irq_lock);
 141         spin_lock_irq(&assigned_dev->assigned_dev_lock);
 142         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 143                 struct kvm_guest_msix_entry *guest_entries =
 144                         assigned_dev->guest_msix_entries;
 145                 for (i = 0; i < assigned_dev->entries_nr; i++) {
 146                         if (!(guest_entries[i].flags &
 147                                         KVM_ASSIGNED_MSIX_PENDING))
 148                                 continue;
 149                         guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
 150                         kvm_set_irq(assigned_dev->kvm,
 151                                     assigned_dev->irq_source_id,
 152                                     guest_entries[i].vector, 1);
 153                 }
 154         } else
 155                 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
 156                             assigned_dev->guest_irq, 1);
 157
 158         spin_unlock_irq(&assigned_dev->assigned_dev_lock);
 159         mutex_unlock(&assigned_dev->kvm->irq_lock);
 160 }
 161
 162 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
 163 {
 164         unsigned long flags;
 165         struct kvm_assigned_dev_kernel *assigned_dev =
 166                 (struct kvm_assigned_dev_kernel *) dev_id;
 167
 168         spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
 169         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 170                 int index = find_index_from_host_irq(assigned_dev, irq);
 171                 if (index < 0)
 172                         goto out;
 173                 assigned_dev->guest_msix_entries[index].flags |=
 174                         KVM_ASSIGNED_MSIX_PENDING;
 175         }
 176
 177         schedule_work(&assigned_dev->interrupt_work);
 178
 179         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
 180                 disable_irq_nosync(irq);
 181                 assigned_dev->host_irq_disabled = true;
 182         }
 183
 184 out:
 185         spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
 186         return IRQ_HANDLED;
 187 }
 188
 189 /* Ack the irq line for an assigned device */
 190 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
 191 {
 192         struct kvm_assigned_dev_kernel *dev;
 193         unsigned long flags;
 194
 195         if (kian->gsi == -1)
 196                 return;
 197
 198         dev = container_of(kian, struct kvm_assigned_dev_kernel,
 199                            ack_notifier);
 200
 201         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
 202
 203         /* The guest irq may be shared so this ack may be
 204          * from another device.
 205          */
 206         spin_lock_irqsave(&dev->assigned_dev_lock, flags);
 207         if (dev->host_irq_disabled) {
 208                 enable_irq(dev->host_irq);
 209                 dev->host_irq_disabled = false;
 210         }
 211         spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
 212 }
 213
 214 static void deassign_guest_irq(struct kvm *kvm,
 215                                struct kvm_assigned_dev_kernel *assigned_dev)
 216 {
 217         kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
 218         assigned_dev->ack_notifier.gsi = -1;
 219
 220         if (assigned_dev->irq_source_id != -1)
 221                 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
 222         assigned_dev->irq_source_id = -1;
 223         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
 224 }
 225
 226 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
 227 static void deassign_host_irq(struct kvm *kvm,
 228                               struct kvm_assigned_dev_kernel *assigned_dev)
 229 {
 230         /*
 231          * In kvm_free_device_irq, cancel_work_sync return true if:
 232          * 1. work is scheduled, and then cancelled.
 233          * 2. work callback is executed.
 234          *
 235          * The first one ensured that the irq is disabled and no more events
 236          * would happen. But for the second one, the irq may be enabled (e.g.
 237          * for MSI). So we disable irq here to prevent further events.
 238          *
 239          * Notice this maybe result in nested disable if the interrupt type is
 240          * INTx, but it's OK for we are going to free it.
 241          *
 242          * If this function is a part of VM destroy, please ensure that till
 243          * now, the kvm state is still legal for probably we also have to wait
 244          * interrupt_work done.
 245          */
 246         if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
 247                 int i;
 248                 for (i = 0; i < assigned_dev->entries_nr; i++)
 249                         disable_irq_nosync(assigned_dev->
 250                                            host_msix_entries[i].vector);
 251
 252                 cancel_work_sync(&assigned_dev->interrupt_work);
 253
 254                 for (i = 0; i < assigned_dev->entries_nr; i++)
 255                         free_irq(assigned_dev->host_msix_entries[i].vector,
 256                                  (void *)assigned_dev);
 257
 258                 assigned_dev->entries_nr = 0;
 259                 kfree(assigned_dev->host_msix_entries);
 260                 kfree(assigned_dev->guest_msix_entries);
 261                 pci_disable_msix(assigned_dev->dev);
 262         } else {
 263                 /* Deal with MSI and INTx */
 264                 disable_irq_nosync(assigned_dev->host_irq);
 265                 cancel_work_sync(&assigned_dev->interrupt_work);
 266
 267                 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
 268
 269                 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
 270                         pci_disable_msi(assigned_dev->dev);
 271         }
 272
 273         assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
 274 }
 275
 276 static int kvm_deassign_irq(struct kvm *kvm,
 277                             struct kvm_assigned_dev_kernel *assigned_dev,
 278                             unsigned long irq_requested_type)
 279 {
 280         unsigned long guest_irq_type, host_irq_type;
 281
 282         if (!irqchip_in_kernel(kvm))
 283                 return -EINVAL;
 284         /* no irq assignment to deassign */
 285         if (!assigned_dev->irq_requested_type)
 286                 return -ENXIO;
 287
 288         host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
 289         guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
 290
 291         if (host_irq_type)
 292                 deassign_host_irq(kvm, assigned_dev);
 293         if (guest_irq_type)
 294                 deassign_guest_irq(kvm, assigned_dev);
 295
 296         return 0;
 297 }
 298
 299 static void kvm_free_assigned_irq(struct kvm *kvm,
 300                                   struct kvm_assigned_dev_kernel *assigned_dev)
 301 {
 302         kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
 303 }
 304
 305 static void kvm_free_assigned_device(struct kvm *kvm,
 306                                      struct kvm_assigned_dev_kernel
 307                                      *assigned_dev)
 308 {
 309         kvm_free_assigned_irq(kvm, assigned_dev);
 310
 311         pci_reset_function(assigned_dev->dev);
 312
 313         pci_release_regions(assigned_dev->dev);
 314         pci_disable_device(assigned_dev->dev);
 315         pci_dev_put(assigned_dev->dev);
 316
 317         list_del(&assigned_dev->list);
 318         kfree(assigned_dev);
 319 }
 320
 321 void kvm_free_all_assigned_devices(struct kvm *kvm)
 322 {
 323         struct list_head *ptr, *ptr2;
 324         struct kvm_assigned_dev_kernel *assigned_dev;
 325
 326         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
 327                 assigned_dev = list_entry(ptr,
 328                                           struct kvm_assigned_dev_kernel,
 329                                           list);
 330
 331                 kvm_free_assigned_device(kvm, assigned_dev);
 332         }
 333 }
 334
 335 static int assigned_device_enable_host_intx(struct kvm *kvm,
 336                                             struct kvm_assigned_dev_kernel *dev)
 337 {
 338         dev->host_irq = dev->dev->irq;
 339         /* Even though this is PCI, we don't want to use shared
 340          * interrupts. Sharing host devices with guest-assigned devices
 341          * on the same interrupt line is not a happy situation: there
 342          * are going to be long delays in accepting, acking, etc.
 343          */
 344         if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
 345                         0, "kvm_assigned_intx_device", (void *)dev))
 346                 return -EIO;
 347         return 0;
 348 }
 349
 350 #ifdef __KVM_HAVE_MSI
 351 static int assigned_device_enable_host_msi(struct kvm *kvm,
 352                                            struct kvm_assigned_dev_kernel *dev)
 353 {
 354         int r;
 355
 356         if (!dev->dev->msi_enabled) {
 357                 r = pci_enable_msi(dev->dev);
 358                 if (r)
 359                         return r;
 360         }
 361
 362         dev->host_irq = dev->dev->irq;
 363         if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
 364                         "kvm_assigned_msi_device", (void *)dev)) {
 365                 pci_disable_msi(dev->dev);
 366                 return -EIO;
 367         }
 368
 369         return 0;
 370 }
 371 #endif
 372
 373 #ifdef __KVM_HAVE_MSIX
 374 static int assigned_device_enable_host_msix(struct kvm *kvm,
 375                                             struct kvm_assigned_dev_kernel *dev)
 376 {
 377         int i, r = -EINVAL;
 378
 379         /* host_msix_entries and guest_msix_entries should have been
 380          * initialized */
 381         if (dev->entries_nr == 0)
 382                 return r;
 383
 384         r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
 385         if (r)
 386                 return r;
 387
 388         for (i = 0; i < dev->entries_nr; i++) {
 389                 r = request_irq(dev->host_msix_entries[i].vector,
 390                                 kvm_assigned_dev_intr, 0,
 391                                 "kvm_assigned_msix_device",
 392                                 (void *)dev);
 393                 /* FIXME: free requested_irq's on failure */
 394                 if (r)
 395                         return r;
 396         }
 397
 398         return 0;
 399 }
 400
 401 #endif
 402
 403 static int assigned_device_enable_guest_intx(struct kvm *kvm,
 404                                 struct kvm_assigned_dev_kernel *dev,
 405                                 struct kvm_assigned_irq *irq)
 406 {
 407         dev->guest_irq = irq->guest_irq;
 408         dev->ack_notifier.gsi = irq->guest_irq;
 409         return 0;
 410 }
 411
 412 #ifdef __KVM_HAVE_MSI
 413 static int assigned_device_enable_guest_msi(struct kvm *kvm,
 414                         struct kvm_assigned_dev_kernel *dev,
 415                         struct kvm_assigned_irq *irq)
 416 {
 417         dev->guest_irq = irq->guest_irq;
 418         dev->ack_notifier.gsi = -1;
 419         dev->host_irq_disabled = false;
 420         return 0;
 421 }
 422 #endif
 423 #ifdef __KVM_HAVE_MSIX
 424 static int assigned_device_enable_guest_msix(struct kvm *kvm,
 425                         struct kvm_assigned_dev_kernel *dev,
 426                         struct kvm_assigned_irq *irq)
 427 {
 428         dev->guest_irq = irq->guest_irq;
 429         dev->ack_notifier.gsi = -1;
 430         dev->host_irq_disabled = false;
 431         return 0;
 432 }
 433 #endif
 434
 435 static int assign_host_irq(struct kvm *kvm,
 436                            struct kvm_assigned_dev_kernel *dev,
 437                            __u32 host_irq_type)
 438 {
 439         int r = -EEXIST;
 440
 441         if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
 442                 return r;
 443
 444         switch (host_irq_type) {
 445         case KVM_DEV_IRQ_HOST_INTX:
 446                 r = assigned_device_enable_host_intx(kvm, dev);
 447                 break;
 448 #ifdef __KVM_HAVE_MSI
 449         case KVM_DEV_IRQ_HOST_MSI:
 450                 r = assigned_device_enable_host_msi(kvm, dev);
 451                 break;
 452 #endif
 453 #ifdef __KVM_HAVE_MSIX
 454         case KVM_DEV_IRQ_HOST_MSIX:
 455                 r = assigned_device_enable_host_msix(kvm, dev);
 456                 break;
 457 #endif
 458         default:
 459                 r = -EINVAL;
 460         }
 461
 462         if (!r)
 463                 dev->irq_requested_type |= host_irq_type;
 464
 465         return r;
 466 }
 467
 468 static int assign_guest_irq(struct kvm *kvm,
 469                             struct kvm_assigned_dev_kernel *dev,
 470                             struct kvm_assigned_irq *irq,
 471                             unsigned long guest_irq_type)
 472 {
 473         int id;
 474         int r = -EEXIST;
 475
 476         if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
 477                 return r;
 478
 479         id = kvm_request_irq_source_id(kvm);
 480         if (id < 0)
 481                 return id;
 482
 483         dev->irq_source_id = id;
 484
 485         switch (guest_irq_type) {
 486         case KVM_DEV_IRQ_GUEST_INTX:
 487                 r = assigned_device_enable_guest_intx(kvm, dev, irq);
 488                 break;
 489 #ifdef __KVM_HAVE_MSI
 490         case KVM_DEV_IRQ_GUEST_MSI:
 491                 r = assigned_device_enable_guest_msi(kvm, dev, irq);
 492                 break;
 493 #endif
 494 #ifdef __KVM_HAVE_MSIX
 495         case KVM_DEV_IRQ_GUEST_MSIX:
 496                 r = assigned_device_enable_guest_msix(kvm, dev, irq);
 497                 break;
 498 #endif
 499         default:
 500                 r = -EINVAL;
 501         }
 502
 503         if (!r) {
 504                 dev->irq_requested_type |= guest_irq_type;
 505                 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
 506         } else
 507                 kvm_free_irq_source_id(kvm, dev->irq_source_id);
 508
 509         return r;
 510 }
 511
 512 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
 513 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 514                                    struct kvm_assigned_irq *assigned_irq)
 515 {
 516         int r = -EINVAL;
 517         struct kvm_assigned_dev_kernel *match;
 518         unsigned long host_irq_type, guest_irq_type;
 519
 520         if (!capable(CAP_SYS_RAWIO))
 521                 return -EPERM;
 522
 523         if (!irqchip_in_kernel(kvm))
 524                 return r;
 525
 526         mutex_lock(&kvm->lock);
 527         r = -ENODEV;
 528         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 529                                       assigned_irq->assigned_dev_id);
 530         if (!match)
 531                 goto out;
 532
 533         host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
 534         guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
 535
 536         r = -EINVAL;
 537         /* can only assign one type at a time */
 538         if (hweight_long(host_irq_type) > 1)
 539                 goto out;
 540         if (hweight_long(guest_irq_type) > 1)
 541                 goto out;
 542         if (host_irq_type == 0 && guest_irq_type == 0)
 543                 goto out;
 544
 545         r = 0;
 546         if (host_irq_type)
 547                 r = assign_host_irq(kvm, match, host_irq_type);
 548         if (r)
 549                 goto out;
 550
 551         if (guest_irq_type)
 552                 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
 553 out:
 554         mutex_unlock(&kvm->lock);
 555         return r;
 556 }
 557
 558 static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
 559                                          struct kvm_assigned_irq
 560                                          *assigned_irq)
 561 {
 562         int r = -ENODEV;
 563         struct kvm_assigned_dev_kernel *match;
 564
 565         mutex_lock(&kvm->lock);
 566
 567         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 568                                       assigned_irq->assigned_dev_id);
 569         if (!match)
 570                 goto out;
 571
 572         r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
 573 out:
 574         mutex_unlock(&kvm->lock);
 575         return r;
 576 }
 577
 578 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 579                                       struct kvm_assigned_pci_dev *assigned_dev)
 580 {
 581         int r = 0;
 582         struct kvm_assigned_dev_kernel *match;
 583         struct pci_dev *dev;
 584
 585         down_read(&kvm->slots_lock);
 586         mutex_lock(&kvm->lock);
 587
 588         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 589                                       assigned_dev->assigned_dev_id);
 590         if (match) {
 591                 /* device already assigned */
 592                 r = -EEXIST;
 593                 goto out;
 594         }
 595
 596         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 597         if (match == NULL) {
 598                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 599                        __func__);
 600                 r = -ENOMEM;
 601                 goto out;
 602         }
 603         dev = pci_get_bus_and_slot(assigned_dev->busnr,
 604                                    assigned_dev->devfn);
 605         if (!dev) {
 606                 printk(KERN_INFO "%s: host device not found\n", __func__);
 607                 r = -EINVAL;
 608                 goto out_free;
 609         }
 610         if (pci_enable_device(dev)) {
 611                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 612                 r = -EBUSY;
 613                 goto out_put;
 614         }
 615         r = pci_request_regions(dev, "kvm_assigned_device");
 616         if (r) {
 617                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 618                        __func__);
 619                 goto out_disable;
 620         }
 621
 622         pci_reset_function(dev);
 623
 624         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 625         match->host_busnr = assigned_dev->busnr;
 626         match->host_devfn = assigned_dev->devfn;
 627         match->flags = assigned_dev->flags;
 628         match->dev = dev;
 629         spin_lock_init(&match->assigned_dev_lock);
 630         match->irq_source_id = -1;
 631         match->kvm = kvm;
 632         match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
 633         INIT_WORK(&match->interrupt_work,
 634                   kvm_assigned_dev_interrupt_work_handler);
 635
 636         list_add(&match->list, &kvm->arch.assigned_dev_head);
 637
 638         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
 639                 if (!kvm->arch.iommu_domain) {
 640                         r = kvm_iommu_map_guest(kvm);
 641                         if (r)
 642                                 goto out_list_del;
 643                 }
 644                 r = kvm_assign_device(kvm, match);
 645                 if (r)
 646                         goto out_list_del;
 647         }
 648
 649 out:
 650         mutex_unlock(&kvm->lock);
 651         up_read(&kvm->slots_lock);
 652         return r;
 653 out_list_del:
 654         list_del(&match->list);
 655         pci_release_regions(dev);
 656 out_disable:
 657         pci_disable_device(dev);
 658 out_put:
 659         pci_dev_put(dev);
 660 out_free:
 661         kfree(match);
 662         mutex_unlock(&kvm->lock);
 663         up_read(&kvm->slots_lock);
 664         return r;
 665 }
 666 #endif
 667
 668 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
 669 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
 670                 struct kvm_assigned_pci_dev *assigned_dev)
 671 {
 672         int r = 0;
 673         struct kvm_assigned_dev_kernel *match;
 674
 675         mutex_lock(&kvm->lock);
 676
 677         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 678                                       assigned_dev->assigned_dev_id);
 679         if (!match) {
 680                 printk(KERN_INFO "%s: device hasn't been assigned before, "
 681                   "so cannot be deassigned\n", __func__);
 682                 r = -EINVAL;
 683                 goto out;
 684         }
 685
 686         if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
 687                 kvm_deassign_device(kvm, match);
 688
 689         kvm_free_assigned_device(kvm, match);
 690
 691 out:
 692         mutex_unlock(&kvm->lock);
 693         return r;
 694 }
 695 #endif
 696
 697 inline int kvm_is_mmio_pfn(pfn_t pfn)
 698 {
 699         if (pfn_valid(pfn)) {
 700                 struct page *page = compound_head(pfn_to_page(pfn));
 701                 return PageReserved(page);
 702         }
 703
 704         return true;
 705 }
 706
 707 /*
 708  * Switches to specified vcpu, until a matching vcpu_put()
 709  */
 710 void vcpu_load(struct kvm_vcpu *vcpu)
 711 {
 712         int cpu;
 713
 714         mutex_lock(&vcpu->mutex);
 715         cpu = get_cpu();
 716         preempt_notifier_register(&vcpu->preempt_notifier);
 717         kvm_arch_vcpu_load(vcpu, cpu);
 718         put_cpu();
 719 }
 720
 721 void vcpu_put(struct kvm_vcpu *vcpu)
 722 {
 723         preempt_disable();
 724         kvm_arch_vcpu_put(vcpu);
 725         preempt_notifier_unregister(&vcpu->preempt_notifier);
 726         preempt_enable();
 727         mutex_unlock(&vcpu->mutex);
 728 }
 729
 730 static void ack_flush(void *_completed)
 731 {
 732 }
 733
 734 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 735 {
 736         int i, cpu, me;
 737         cpumask_var_t cpus;
 738         bool called = true;
 739         struct kvm_vcpu *vcpu;
 740
 741         if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
 742                 cpumask_clear(cpus);
 743
 744         me = get_cpu();
 745         spin_lock(&kvm->requests_lock);
 746         kvm_for_each_vcpu(i, vcpu, kvm) {
 747                 if (test_and_set_bit(req, &vcpu->requests))
 748                         continue;
 749                 cpu = vcpu->cpu;
 750                 if (cpus != NULL && cpu != -1 && cpu != me)
 751                         cpumask_set_cpu(cpu, cpus);
 752         }
 753         if (unlikely(cpus == NULL))
 754                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 755         else if (!cpumask_empty(cpus))
 756                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 757         else
 758                 called = false;
 759         spin_unlock(&kvm->requests_lock);
 760         put_cpu();
 761         free_cpumask_var(cpus);
 762         return called;
 763 }
 764
 765 void kvm_flush_remote_tlbs(struct kvm *kvm)
 766 {
 767         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 768                 ++kvm->stat.remote_tlb_flush;
 769 }
 770
 771 void kvm_reload_remote_mmus(struct kvm *kvm)
 772 {
 773         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 774 }
 775
 776 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 777 {
 778         struct page *page;
 779         int r;
 780
 781         mutex_init(&vcpu->mutex);
 782         vcpu->cpu = -1;
 783         vcpu->kvm = kvm;
 784         vcpu->vcpu_id = id;
 785         init_waitqueue_head(&vcpu->wq);
 786
 787         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 788         if (!page) {
 789                 r = -ENOMEM;
 790                 goto fail;
 791         }
 792         vcpu->run = page_address(page);
 793
 794         r = kvm_arch_vcpu_init(vcpu);
 795         if (r < 0)
 796                 goto fail_free_run;
 797         return 0;
 798
 799 fail_free_run:
 800         free_page((unsigned long)vcpu->run);
 801 fail:
 802         return r;
 803 }
 804 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 805
 806 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 807 {
 808         kvm_arch_vcpu_uninit(vcpu);
 809         free_page((unsigned long)vcpu->run);
 810 }
 811 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 812
 813 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 814 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 815 {
 816         return container_of(mn, struct kvm, mmu_notifier);
 817 }
 818
 819 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 820                                              struct mm_struct *mm,
 821                                              unsigned long address)
 822 {
 823         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 824         int need_tlb_flush;
 825
 826         /*
 827          * When ->invalidate_page runs, the linux pte has been zapped
 828          * already but the page is still allocated until
 829          * ->invalidate_page returns. So if we increase the sequence
 830          * here the kvm page fault will notice if the spte can't be
 831          * established because the page is going to be freed. If
 832          * instead the kvm page fault establishes the spte before
 833          * ->invalidate_page runs, kvm_unmap_hva will release it
 834          * before returning.
 835          *
 836          * The sequence increase only need to be seen at spin_unlock
 837          * time, and not at spin_lock time.
 838          *
 839          * Increasing the sequence after the spin_unlock would be
 840          * unsafe because the kvm page fault could then establish the
 841          * pte after kvm_unmap_hva returned, without noticing the page
 842          * is going to be freed.
 843          */
 844         spin_lock(&kvm->mmu_lock);
 845         kvm->mmu_notifier_seq++;
 846         need_tlb_flush = kvm_unmap_hva(kvm, address);
 847         spin_unlock(&kvm->mmu_lock);
 848
 849         /* we've to flush the tlb before the pages can be freed */
 850         if (need_tlb_flush)
 851                 kvm_flush_remote_tlbs(kvm);
 852
 853 }
 854
 855 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 856                                                     struct mm_struct *mm,
 857                                                     unsigned long start,
 858                                                     unsigned long end)
 859 {
 860         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 861         int need_tlb_flush = 0;
 862
 863         spin_lock(&kvm->mmu_lock);
 864         /*
 865          * The count increase must become visible at unlock time as no
 866          * spte can be established without taking the mmu_lock and
 867          * count is also read inside the mmu_lock critical section.
 868          */
 869         kvm->mmu_notifier_count++;
 870         for (; start < end; start += PAGE_SIZE)
 871                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 872         spin_unlock(&kvm->mmu_lock);
 873
 874         /* we've to flush the tlb before the pages can be freed */
 875         if (need_tlb_flush)
 876                 kvm_flush_remote_tlbs(kvm);
 877 }
 878
 879 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 880                                                   struct mm_struct *mm,
 881                                                   unsigned long start,
 882                                                   unsigned long end)
 883 {
 884         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 885
 886         spin_lock(&kvm->mmu_lock);
 887         /*
 888          * This sequence increase will notify the kvm page fault that
 889          * the page that is going to be mapped in the spte could have
 890          * been freed.
 891          */
 892         kvm->mmu_notifier_seq++;
 893         /*
 894          * The above sequence increase must be visible before the
 895          * below count decrease but both values are read by the kvm
 896          * page fault under mmu_lock spinlock so we don't need to add
 897          * a smb_wmb() here in between the two.
 898          */
 899         kvm->mmu_notifier_count--;
 900         spin_unlock(&kvm->mmu_lock);
 901
 902         BUG_ON(kvm->mmu_notifier_count < 0);
 903 }
 904
 905 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 906                                               struct mm_struct *mm,
 907                                               unsigned long address)
 908 {
 909         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 910         int young;
 911
 912         spin_lock(&kvm->mmu_lock);
 913         young = kvm_age_hva(kvm, address);
 914         spin_unlock(&kvm->mmu_lock);
 915
 916         if (young)
 917                 kvm_flush_remote_tlbs(kvm);
 918
 919         return young;
 920 }
 921
 922 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 923                                      struct mm_struct *mm)
 924 {
 925         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 926         kvm_arch_flush_shadow(kvm);
 927 }
 928
 929 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 930         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 931         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 932         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 933         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 934         .release                = kvm_mmu_notifier_release,
 935 };
 936 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 937
 938 static struct kvm *kvm_create_vm(void)
 939 {
 940         struct kvm *kvm = kvm_arch_create_vm();
 941 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 942         struct page *page;
 943 #endif
 944
 945         if (IS_ERR(kvm))
 946                 goto out;
 947 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 948         INIT_LIST_HEAD(&kvm->irq_routing);
 949         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 950 #endif
 951
 952 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 953         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 954         if (!page) {
 955                 kfree(kvm);
 956                 return ERR_PTR(-ENOMEM);
 957         }
 958         kvm->coalesced_mmio_ring =
 959                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 960 #endif
 961
 962 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 963         {
 964                 int err;
 965                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 966                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 967                 if (err) {
 968 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 969                         put_page(page);
 970 #endif
 971                         kfree(kvm);
 972                         return ERR_PTR(err);
 973                 }
 974         }
 975 #endif
 976
 977         kvm->mm = current->mm;
 978         atomic_inc(&kvm->mm->mm_count);
 979         spin_lock_init(&kvm->mmu_lock);
 980         spin_lock_init(&kvm->requests_lock);
 981         kvm_io_bus_init(&kvm->pio_bus);
 982         kvm_irqfd_init(kvm);
 983         mutex_init(&kvm->lock);
 984         mutex_init(&kvm->irq_lock);
 985         kvm_io_bus_init(&kvm->mmio_bus);
 986         init_rwsem(&kvm->slots_lock);
 987         atomic_set(&kvm->users_count, 1);
 988         spin_lock(&kvm_lock);
 989         list_add(&kvm->vm_list, &vm_list);
 990         spin_unlock(&kvm_lock);
 991 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 992         kvm_coalesced_mmio_init(kvm);
 993 #endif
 994 out:
 995         return kvm;
 996 }
 997
 998 /*
 999  * Free any memory in @free but not in @dont.
1000  */
1001 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
1002                                   struct kvm_memory_slot *dont)
1003 {
1004         int i;
1005
1006         if (!dont || free->rmap != dont->rmap)
1007                 vfree(free->rmap);
1008
1009         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
1010                 vfree(free->dirty_bitmap);
1011
1012
1013         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
1014                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
1015                         vfree(free->lpage_info[i]);
1016                         free->lpage_info[i] = NULL;
1017                 }
1018         }
1019
1020         free->npages = 0;
1021         free->dirty_bitmap = NULL;
1022         free->rmap = NULL;
1023 }
1024
1025 void kvm_free_physmem(struct kvm *kvm)
1026 {
1027         int i;
1028
1029         for (i = 0; i < kvm->nmemslots; ++i)
1030                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
1031 }
1032
1033 static void kvm_destroy_vm(struct kvm *kvm)
1034 {
1035         struct mm_struct *mm = kvm->mm;
1036
1037         kvm_arch_sync_events(kvm);
1038         spin_lock(&kvm_lock);
1039         list_del(&kvm->vm_list);
1040         spin_unlock(&kvm_lock);
1041         kvm_free_irq_routing(kvm);
1042         kvm_io_bus_destroy(&kvm->pio_bus);
1043         kvm_io_bus_destroy(&kvm->mmio_bus);
1044 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1045         if (kvm->coalesced_mmio_ring != NULL)
1046                 free_page((unsigned long)kvm->coalesced_mmio_ring);
1047 #endif
1048 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1049         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1050 #else
1051         kvm_arch_flush_shadow(kvm);
1052 #endif
1053         kvm_arch_destroy_vm(kvm);
1054         mmdrop(mm);
1055 }
1056
1057 void kvm_get_kvm(struct kvm *kvm)
1058 {
1059         atomic_inc(&kvm->users_count);
1060 }
1061 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1062
1063 void kvm_put_kvm(struct kvm *kvm)
1064 {
1065         if (atomic_dec_and_test(&kvm->users_count))
1066                 kvm_destroy_vm(kvm);
1067 }
1068 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1069
1070
1071 static int kvm_vm_release(struct inode *inode, struct file *filp)
1072 {
1073         struct kvm *kvm = filp->private_data;
1074
1075         kvm_irqfd_release(kvm);
1076
1077         kvm_put_kvm(kvm);
1078         return 0;
1079 }
1080
1081 /*
1082  * Allocate some memory and give it an address in the guest physical address
1083  * space.
1084  *
1085  * Discontiguous memory is allowed, mostly for framebuffers.
1086  *
1087  * Must be called holding mmap_sem for write.
1088  */
1089 int __kvm_set_memory_region(struct kvm *kvm,
1090                             struct kvm_userspace_memory_region *mem,
1091                             int user_alloc)
1092 {
1093         int r;
1094         gfn_t base_gfn;
1095         unsigned long npages, ugfn;
1096         int lpages;
1097         unsigned long i, j;
1098         struct kvm_memory_slot *memslot;
1099         struct kvm_memory_slot old, new;
1100
1101         r = -EINVAL;
1102         /* General sanity checks */
1103         if (mem->memory_size & (PAGE_SIZE - 1))
1104                 goto out;
1105         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1106                 goto out;
1107         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
1108                 goto out;
1109         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
1110                 goto out;
1111         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1112                 goto out;
1113
1114         memslot = &kvm->memslots[mem->slot];
1115         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1116         npages = mem->memory_size >> PAGE_SHIFT;
1117
1118         if (!npages)
1119                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1120
1121         new = old = *memslot;
1122
1123         new.base_gfn = base_gfn;
1124         new.npages = npages;
1125         new.flags = mem->flags;
1126
1127         /* Disallow changing a memory slot's size. */
1128         r = -EINVAL;
1129         if (npages && old.npages && npages != old.npages)
1130                 goto out_free;
1131
1132         /* Check for overlaps */
1133         r = -EEXIST;
1134         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1135                 struct kvm_memory_slot *s = &kvm->memslots[i];
1136
1137                 if (s == memslot || !s->npages)
1138                         continue;
1139                 if (!((base_gfn + npages <= s->base_gfn) ||
1140                       (base_gfn >= s->base_gfn + s->npages)))
1141                         goto out_free;
1142         }
1143
1144         /* Free page dirty bitmap if unneeded */
1145         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1146                 new.dirty_bitmap = NULL;
1147
1148         r = -ENOMEM;
1149
1150         /* Allocate if a slot is being created */
1151 #ifndef CONFIG_S390
1152         if (npages && !new.rmap) {
1153                 new.rmap = vmalloc(npages * sizeof(struct page *));
1154
1155                 if (!new.rmap)
1156                         goto out_free;
1157
1158                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1159
1160                 new.user_alloc = user_alloc;
1161                 /*
1162                  * hva_to_rmmap() serialzies with the mmu_lock and to be
1163                  * safe it has to ignore memslots with !user_alloc &&
1164                  * !userspace_addr.
1165                  */
1166                 if (user_alloc)
1167                         new.userspace_addr = mem->userspace_addr;
1168                 else
1169                         new.userspace_addr = 0;
1170         }
1171         if (!npages)
1172                 goto skip_lpage;
1173
1174         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
1175                 int level = i + 2;
1176
1177                 /* Avoid unused variable warning if no large pages */
1178                 (void)level;
1179
1180                 if (new.lpage_info[i])
1181                         continue;
1182
1183                 lpages = 1 + (base_gfn + npages - 1) /
1184                              KVM_PAGES_PER_HPAGE(level);
1185                 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
1186
1187                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
1188
1189                 if (!new.lpage_info[i])
1190                         goto out_free;
1191
1192                 memset(new.lpage_info[i], 0,
1193                        lpages * sizeof(*new.lpage_info[i]));
1194
1195                 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
1196                         new.lpage_info[i][0].write_count = 1;
1197                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
1198                         new.lpage_info[i][lpages - 1].write_count = 1;
1199                 ugfn = new.userspace_addr >> PAGE_SHIFT;
1200                 /*
1201                  * If the gfn and userspace address are not aligned wrt each
1202                  * other, or if explicitly asked to, disable large page
1203                  * support for this slot
1204                  */
1205                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
1206                     !largepages_enabled)
1207                         for (j = 0; j < lpages; ++j)
1208                                 new.lpage_info[i][j].write_count = 1;
1209         }
1210
1211 skip_lpage:
1212
1213         /* Allocate page dirty bitmap if needed */
1214         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1215                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1216
1217                 new.dirty_bitmap = vmalloc(dirty_bytes);
1218                 if (!new.dirty_bitmap)
1219                         goto out_free;
1220                 memset(new.dirty_bitmap, 0, dirty_bytes);
1221                 if (old.npages)
1222                         kvm_arch_flush_shadow(kvm);
1223         }
1224 #else  /* not defined CONFIG_S390 */
1225         new.user_alloc = user_alloc;
1226         if (user_alloc)
1227                 new.userspace_addr = mem->userspace_addr;
1228 #endif /* not defined CONFIG_S390 */
1229
1230         if (!npages)
1231                 kvm_arch_flush_shadow(kvm);
1232
1233         spin_lock(&kvm->mmu_lock);
1234         if (mem->slot >= kvm->nmemslots)
1235                 kvm->nmemslots = mem->slot + 1;
1236
1237         *memslot = new;
1238         spin_unlock(&kvm->mmu_lock);
1239
1240         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1241         if (r) {
1242                 spin_lock(&kvm->mmu_lock);
1243                 *memslot = old;
1244                 spin_unlock(&kvm->mmu_lock);
1245                 goto out_free;
1246         }
1247
1248         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1249         /* Slot deletion case: we have to update the current slot */
1250         spin_lock(&kvm->mmu_lock);
1251         if (!npages)
1252                 *memslot = old;
1253         spin_unlock(&kvm->mmu_lock);
1254 #ifdef CONFIG_DMAR
1255         /* map the pages in iommu page table */
1256         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1257         if (r)
1258                 goto out;
1259 #endif
1260         return 0;
1261
1262 out_free:
1263         kvm_free_physmem_slot(&new, &old);
1264 out:
1265         return r;
1266
1267 }
1268 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1269
1270 int kvm_set_memory_region(struct kvm *kvm,
1271                           struct kvm_userspace_memory_region *mem,
1272                           int user_alloc)
1273 {
1274         int r;
1275
1276         down_write(&kvm->slots_lock);
1277         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1278         up_write(&kvm->slots_lock);
1279         return r;
1280 }
1281 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1282
1283 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1284                                    struct
1285                                    kvm_userspace_memory_region *mem,
1286                                    int user_alloc)
1287 {
1288         if (mem->slot >= KVM_MEMORY_SLOTS)
1289                 return -EINVAL;
1290         return kvm_set_memory_region(kvm, mem, user_alloc);
1291 }
1292
1293 int kvm_get_dirty_log(struct kvm *kvm,
1294                         struct kvm_dirty_log *log, int *is_dirty)
1295 {
1296         struct kvm_memory_slot *memslot;
1297         int r, i;
1298         int n;
1299         unsigned long any = 0;
1300
1301         r = -EINVAL;
1302         if (log->slot >= KVM_MEMORY_SLOTS)
1303                 goto out;
1304
1305         memslot = &kvm->memslots[log->slot];
1306         r = -ENOENT;
1307         if (!memslot->dirty_bitmap)
1308                 goto out;
1309
1310         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1311
1312         for (i = 0; !any && i < n/sizeof(long); ++i)
1313                 any = memslot->dirty_bitmap[i];
1314
1315         r = -EFAULT;
1316         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1317                 goto out;
1318
1319         if (any)
1320                 *is_dirty = 1;
1321
1322         r = 0;
1323 out:
1324         return r;
1325 }
1326
1327 void kvm_disable_largepages(void)
1328 {
1329         largepages_enabled = false;
1330 }
1331 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1332
1333 int is_error_page(struct page *page)
1334 {
1335         return page == bad_page;
1336 }
1337 EXPORT_SYMBOL_GPL(is_error_page);
1338
1339 int is_error_pfn(pfn_t pfn)
1340 {
1341         return pfn == bad_pfn;
1342 }
1343 EXPORT_SYMBOL_GPL(is_error_pfn);
1344
1345 static inline unsigned long bad_hva(void)
1346 {
1347         return PAGE_OFFSET;
1348 }
1349
1350 int kvm_is_error_hva(unsigned long addr)
1351 {
1352         return addr == bad_hva();
1353 }
1354 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1355
1356 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1357 {
1358         int i;
1359
1360         for (i = 0; i < kvm->nmemslots; ++i) {
1361                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1362
1363                 if (gfn >= memslot->base_gfn
1364                     && gfn < memslot->base_gfn + memslot->npages)
1365                         return memslot;
1366         }
1367         return NULL;
1368 }
1369 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1370
1371 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1372 {
1373         gfn = unalias_gfn(kvm, gfn);
1374         return gfn_to_memslot_unaliased(kvm, gfn);
1375 }
1376
1377 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1378 {
1379         int i;
1380
1381         gfn = unalias_gfn(kvm, gfn);
1382         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1383                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1384
1385                 if (gfn >= memslot->base_gfn
1386                     && gfn < memslot->base_gfn + memslot->npages)
1387                         return 1;
1388         }
1389         return 0;
1390 }
1391 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1392
1393 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1394 {
1395         struct kvm_memory_slot *slot;
1396
1397         gfn = unalias_gfn(kvm, gfn);
1398         slot = gfn_to_memslot_unaliased(kvm, gfn);
1399         if (!slot)
1400                 return bad_hva();
1401         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1402 }
1403 EXPORT_SYMBOL_GPL(gfn_to_hva);
1404
1405 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1406 {
1407         struct page *page[1];
1408         unsigned long addr;
1409         int npages;
1410         pfn_t pfn;
1411
1412         might_sleep();
1413
1414         addr = gfn_to_hva(kvm, gfn);
1415         if (kvm_is_error_hva(addr)) {
1416                 get_page(bad_page);
1417                 return page_to_pfn(bad_page);
1418         }
1419
1420         npages = get_user_pages_fast(addr, 1, 1, page);
1421
1422         if (unlikely(npages != 1)) {
1423                 struct vm_area_struct *vma;
1424
1425                 down_read(&current->mm->mmap_sem);
1426                 vma = find_vma(current->mm, addr);
1427
1428                 if (vma == NULL || addr < vma->vm_start ||
1429                     !(vma->vm_flags & VM_PFNMAP)) {
1430                         up_read(&current->mm->mmap_sem);
1431                         get_page(bad_page);
1432                         return page_to_pfn(bad_page);
1433                 }
1434
1435                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1436                 up_read(&current->mm->mmap_sem);
1437                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1438         } else
1439                 pfn = page_to_pfn(page[0]);
1440
1441         return pfn;
1442 }
1443
1444 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1445
1446 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1447 {
1448         pfn_t pfn;
1449
1450         pfn = gfn_to_pfn(kvm, gfn);
1451         if (!kvm_is_mmio_pfn(pfn))
1452                 return pfn_to_page(pfn);
1453
1454         WARN_ON(kvm_is_mmio_pfn(pfn));
1455
1456         get_page(bad_page);
1457         return bad_page;
1458 }
1459
1460 EXPORT_SYMBOL_GPL(gfn_to_page);
1461
1462 void kvm_release_page_clean(struct page *page)
1463 {
1464         kvm_release_pfn_clean(page_to_pfn(page));
1465 }
1466 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1467
1468 void kvm_release_pfn_clean(pfn_t pfn)
1469 {
1470         if (!kvm_is_mmio_pfn(pfn))
1471                 put_page(pfn_to_page(pfn));
1472 }
1473 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1474
1475 void kvm_release_page_dirty(struct page *page)
1476 {
1477         kvm_release_pfn_dirty(page_to_pfn(page));
1478 }
1479 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1480
1481 void kvm_release_pfn_dirty(pfn_t pfn)
1482 {
1483         kvm_set_pfn_dirty(pfn);
1484         kvm_release_pfn_clean(pfn);
1485 }
1486 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1487
1488 void kvm_set_page_dirty(struct page *page)
1489 {
1490         kvm_set_pfn_dirty(page_to_pfn(page));
1491 }
1492 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1493
1494 void kvm_set_pfn_dirty(pfn_t pfn)
1495 {
1496         if (!kvm_is_mmio_pfn(pfn)) {
1497                 struct page *page = pfn_to_page(pfn);
1498                 if (!PageReserved(page))
1499                         SetPageDirty(page);
1500         }
1501 }
1502 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1503
1504 void kvm_set_pfn_accessed(pfn_t pfn)
1505 {
1506         if (!kvm_is_mmio_pfn(pfn))
1507                 mark_page_accessed(pfn_to_page(pfn));
1508 }
1509 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1510
1511 void kvm_get_pfn(pfn_t pfn)
1512 {
1513         if (!kvm_is_mmio_pfn(pfn))
1514                 get_page(pfn_to_page(pfn));
1515 }
1516 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1517
1518 static int next_segment(unsigned long len, int offset)
1519 {
1520         if (len > PAGE_SIZE - offset)
1521                 return PAGE_SIZE - offset;
1522         else
1523                 return len;
1524 }
1525
1526 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1527                         int len)
1528 {
1529         int r;
1530         unsigned long addr;
1531
1532         addr = gfn_to_hva(kvm, gfn);
1533         if (kvm_is_error_hva(addr))
1534                 return -EFAULT;
1535         r = copy_from_user(data, (void __user *)addr + offset, len);
1536         if (r)
1537                 return -EFAULT;
1538         return 0;
1539 }
1540 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1541
1542 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1543 {
1544         gfn_t gfn = gpa >> PAGE_SHIFT;
1545         int seg;
1546         int offset = offset_in_page(gpa);
1547         int ret;
1548
1549         while ((seg = next_segment(len, offset)) != 0) {
1550                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1551                 if (ret < 0)
1552                         return ret;
1553                 offset = 0;
1554                 len -= seg;
1555                 data += seg;
1556                 ++gfn;
1557         }
1558         return 0;
1559 }
1560 EXPORT_SYMBOL_GPL(kvm_read_guest);
1561
1562 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1563                           unsigned long len)
1564 {
1565         int r;
1566         unsigned long addr;
1567         gfn_t gfn = gpa >> PAGE_SHIFT;
1568         int offset = offset_in_page(gpa);
1569
1570         addr = gfn_to_hva(kvm, gfn);
1571         if (kvm_is_error_hva(addr))
1572                 return -EFAULT;
1573         pagefault_disable();
1574         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1575         pagefault_enable();
1576         if (r)
1577                 return -EFAULT;
1578         return 0;
1579 }
1580 EXPORT_SYMBOL(kvm_read_guest_atomic);
1581
1582 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1583                          int offset, int len)
1584 {
1585         int r;
1586         unsigned long addr;
1587
1588         addr = gfn_to_hva(kvm, gfn);
1589         if (kvm_is_error_hva(addr))
1590                 return -EFAULT;
1591         r = copy_to_user((void __user *)addr + offset, data, len);
1592         if (r)
1593                 return -EFAULT;
1594         mark_page_dirty(kvm, gfn);
1595         return 0;
1596 }
1597 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1598
1599 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1600                     unsigned long len)
1601 {
1602         gfn_t gfn = gpa >> PAGE_SHIFT;
1603         int seg;
1604         int offset = offset_in_page(gpa);
1605         int ret;
1606
1607         while ((seg = next_segment(len, offset)) != 0) {
1608                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1609                 if (ret < 0)
1610                         return ret;
1611                 offset = 0;
1612                 len -= seg;
1613                 data += seg;
1614                 ++gfn;
1615         }
1616         return 0;
1617 }
1618
1619 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1620 {
1621         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1622 }
1623 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1624
1625 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1626 {
1627         gfn_t gfn = gpa >> PAGE_SHIFT;
1628         int seg;
1629         int offset = offset_in_page(gpa);
1630         int ret;
1631
1632         while ((seg = next_segment(len, offset)) != 0) {
1633                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1634                 if (ret < 0)
1635                         return ret;
1636                 offset = 0;
1637                 len -= seg;
1638                 ++gfn;
1639         }
1640         return 0;
1641 }
1642 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1643
1644 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1645 {
1646         struct kvm_memory_slot *memslot;
1647
1648         gfn = unalias_gfn(kvm, gfn);
1649         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1650         if (memslot && memslot->dirty_bitmap) {
1651                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1652
1653                 /* avoid RMW */
1654                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1655                         set_bit(rel_gfn, memslot->dirty_bitmap);
1656         }
1657 }
1658
1659 /*
1660  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1661  */
1662 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1663 {
1664         DEFINE_WAIT(wait);
1665
1666         for (;;) {
1667                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1668
1669                 if ((kvm_arch_interrupt_allowed(vcpu) &&
1670                                         kvm_cpu_has_interrupt(vcpu)) ||
1671                                 kvm_arch_vcpu_runnable(vcpu)) {
1672                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1673                         break;
1674                 }
1675                 if (kvm_cpu_has_pending_timer(vcpu))
1676                         break;
1677                 if (signal_pending(current))
1678                         break;
1679
1680                 vcpu_put(vcpu);
1681                 schedule();
1682                 vcpu_load(vcpu);
1683         }
1684
1685         finish_wait(&vcpu->wq, &wait);
1686 }
1687
1688 void kvm_resched(struct kvm_vcpu *vcpu)
1689 {
1690         if (!need_resched())
1691                 return;
1692         cond_resched();
1693 }
1694 EXPORT_SYMBOL_GPL(kvm_resched);
1695
1696 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1697 {
1698         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1699         struct page *page;
1700
1701         if (vmf->pgoff == 0)
1702                 page = virt_to_page(vcpu->run);
1703 #ifdef CONFIG_X86
1704         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1705                 page = virt_to_page(vcpu->arch.pio_data);
1706 #endif
1707 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1708         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1709                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1710 #endif
1711         else
1712                 return VM_FAULT_SIGBUS;
1713         get_page(page);
1714         vmf->page = page;
1715         return 0;
1716 }
1717
1718 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1719         .fault = kvm_vcpu_fault,
1720 };
1721
1722 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1723 {
1724         vma->vm_ops = &kvm_vcpu_vm_ops;
1725         return 0;
1726 }
1727
1728 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1729 {
1730         struct kvm_vcpu *vcpu = filp->private_data;
1731
1732         kvm_put_kvm(vcpu->kvm);
1733         return 0;
1734 }
1735
1736 static struct file_operations kvm_vcpu_fops = {
1737         .release        = kvm_vcpu_release,
1738         .unlocked_ioctl = kvm_vcpu_ioctl,
1739         .compat_ioctl   = kvm_vcpu_ioctl,
1740         .mmap           = kvm_vcpu_mmap,
1741 };
1742
1743 /*
1744  * Allocates an inode for the vcpu.
1745  */
1746 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1747 {
1748         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1749 }
1750
1751 /*
1752  * Creates some virtual cpus.  Good luck creating more than one.
1753  */
1754 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1755 {
1756         int r;
1757         struct kvm_vcpu *vcpu, *v;
1758
1759         vcpu = kvm_arch_vcpu_create(kvm, id);
1760         if (IS_ERR(vcpu))
1761                 return PTR_ERR(vcpu);
1762
1763         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1764
1765         r = kvm_arch_vcpu_setup(vcpu);
1766         if (r)
1767                 return r;
1768
1769         mutex_lock(&kvm->lock);
1770         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1771                 r = -EINVAL;
1772                 goto vcpu_destroy;
1773         }
1774
1775         kvm_for_each_vcpu(r, v, kvm)
1776                 if (v->vcpu_id == id) {
1777                         r = -EEXIST;
1778                         goto vcpu_destroy;
1779                 }
1780
1781         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1782
1783         /* Now it's all set up, let userspace reach it */
1784         kvm_get_kvm(kvm);
1785         r = create_vcpu_fd(vcpu);
1786         if (r < 0) {
1787                 kvm_put_kvm(kvm);
1788                 goto vcpu_destroy;
1789         }
1790
1791         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1792         smp_wmb();
1793         atomic_inc(&kvm->online_vcpus);
1794
1795 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1796         if (kvm->bsp_vcpu_id == id)
1797                 kvm->bsp_vcpu = vcpu;
1798 #endif
1799         mutex_unlock(&kvm->lock);
1800         return r;
1801
1802 vcpu_destroy:
1803         mutex_unlock(&kvm->lock);
1804         kvm_arch_vcpu_destroy(vcpu);
1805         return r;
1806 }
1807
1808 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1809 {
1810         if (sigset) {
1811                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1812                 vcpu->sigset_active = 1;
1813                 vcpu->sigset = *sigset;
1814         } else
1815                 vcpu->sigset_active = 0;
1816         return 0;
1817 }
1818
1819 #ifdef __KVM_HAVE_MSIX
1820 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1821                                     struct kvm_assigned_msix_nr *entry_nr)
1822 {
1823         int r = 0;
1824         struct kvm_assigned_dev_kernel *adev;
1825
1826         mutex_lock(&kvm->lock);
1827
1828         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1829                                       entry_nr->assigned_dev_id);
1830         if (!adev) {
1831                 r = -EINVAL;
1832                 goto msix_nr_out;
1833         }
1834
1835         if (adev->entries_nr == 0) {
1836                 adev->entries_nr = entry_nr->entry_nr;
1837                 if (adev->entries_nr == 0 ||
1838                     adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1839                         r = -EINVAL;
1840                         goto msix_nr_out;
1841                 }
1842
1843                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1844                                                 entry_nr->entry_nr,
1845                                                 GFP_KERNEL);
1846                 if (!adev->host_msix_entries) {
1847                         r = -ENOMEM;
1848                         goto msix_nr_out;
1849                 }
1850                 adev->guest_msix_entries = kzalloc(
1851                                 sizeof(struct kvm_guest_msix_entry) *
1852                                 entry_nr->entry_nr, GFP_KERNEL);
1853                 if (!adev->guest_msix_entries) {
1854                         kfree(adev->host_msix_entries);
1855                         r = -ENOMEM;
1856                         goto msix_nr_out;
1857                 }
1858         } else /* Not allowed set MSI-X number twice */
1859                 r = -EINVAL;
1860 msix_nr_out:
1861         mutex_unlock(&kvm->lock);
1862         return r;
1863 }
1864
1865 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1866                                        struct kvm_assigned_msix_entry *entry)
1867 {
1868         int r = 0, i;
1869         struct kvm_assigned_dev_kernel *adev;
1870
1871         mutex_lock(&kvm->lock);
1872
1873         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1874                                       entry->assigned_dev_id);
1875
1876         if (!adev) {
1877                 r = -EINVAL;
1878                 goto msix_entry_out;
1879         }
1880
1881         for (i = 0; i < adev->entries_nr; i++)
1882                 if (adev->guest_msix_entries[i].vector == 0 ||
1883                     adev->guest_msix_entries[i].entry == entry->entry) {
1884                         adev->guest_msix_entries[i].entry = entry->entry;
1885                         adev->guest_msix_entries[i].vector = entry->gsi;
1886                         adev->host_msix_entries[i].entry = entry->entry;
1887                         break;
1888                 }
1889         if (i == adev->entries_nr) {
1890                 r = -ENOSPC;
1891                 goto msix_entry_out;
1892         }
1893
1894 msix_entry_out:
1895         mutex_unlock(&kvm->lock);
1896
1897         return r;
1898 }
1899 #endif
1900
1901 static long kvm_vcpu_ioctl(struct file *filp,
1902                            unsigned int ioctl, unsigned long arg)
1903 {
1904         struct kvm_vcpu *vcpu = filp->private_data;
1905         void __user *argp = (void __user *)arg;
1906         int r;
1907         struct kvm_fpu *fpu = NULL;
1908         struct kvm_sregs *kvm_sregs = NULL;
1909
1910         if (vcpu->kvm->mm != current->mm)
1911                 return -EIO;
1912         switch (ioctl) {
1913         case KVM_RUN:
1914                 r = -EINVAL;
1915                 if (arg)
1916                         goto out;
1917                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1918                 break;
1919         case KVM_GET_REGS: {
1920                 struct kvm_regs *kvm_regs;
1921
1922                 r = -ENOMEM;
1923                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1924                 if (!kvm_regs)
1925                         goto out;
1926                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1927                 if (r)
1928                         goto out_free1;
1929                 r = -EFAULT;
1930                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1931                         goto out_free1;
1932                 r = 0;
1933 out_free1:
1934                 kfree(kvm_regs);
1935                 break;
1936         }
1937         case KVM_SET_REGS: {
1938                 struct kvm_regs *kvm_regs;
1939
1940                 r = -ENOMEM;
1941                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1942                 if (!kvm_regs)
1943                         goto out;
1944                 r = -EFAULT;
1945                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1946                         goto out_free2;
1947                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1948                 if (r)
1949                         goto out_free2;
1950                 r = 0;
1951 out_free2:
1952                 kfree(kvm_regs);
1953                 break;
1954         }
1955         case KVM_GET_SREGS: {
1956                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1957                 r = -ENOMEM;
1958                 if (!kvm_sregs)
1959                         goto out;
1960                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1961                 if (r)
1962                         goto out;
1963                 r = -EFAULT;
1964                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1965                         goto out;
1966                 r = 0;
1967                 break;
1968         }
1969         case KVM_SET_SREGS: {
1970                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1971                 r = -ENOMEM;
1972                 if (!kvm_sregs)
1973                         goto out;
1974                 r = -EFAULT;
1975                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1976                         goto out;
1977                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1978                 if (r)
1979                         goto out;
1980                 r = 0;
1981                 break;
1982         }
1983         case KVM_GET_MP_STATE: {
1984                 struct kvm_mp_state mp_state;
1985
1986                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1987                 if (r)
1988                         goto out;
1989                 r = -EFAULT;
1990                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1991                         goto out;
1992                 r = 0;
1993                 break;
1994         }
1995         case KVM_SET_MP_STATE: {
1996                 struct kvm_mp_state mp_state;
1997
1998                 r = -EFAULT;
1999                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2000                         goto out;
2001                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2002                 if (r)
2003                         goto out;
2004                 r = 0;
2005                 break;
2006         }
2007         case KVM_TRANSLATE: {
2008                 struct kvm_translation tr;
2009
2010                 r = -EFAULT;
2011                 if (copy_from_user(&tr, argp, sizeof tr))
2012                         goto out;
2013                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2014                 if (r)
2015                         goto out;
2016                 r = -EFAULT;
2017                 if (copy_to_user(argp, &tr, sizeof tr))
2018                         goto out;
2019                 r = 0;
2020                 break;
2021         }
2022         case KVM_SET_GUEST_DEBUG: {
2023                 struct kvm_guest_debug dbg;
2024
2025                 r = -EFAULT;
2026                 if (copy_from_user(&dbg, argp, sizeof dbg))
2027                         goto out;
2028                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2029                 if (r)
2030                         goto out;
2031                 r = 0;
2032                 break;
2033         }
2034         case KVM_SET_SIGNAL_MASK: {
2035                 struct kvm_signal_mask __user *sigmask_arg = argp;
2036                 struct kvm_signal_mask kvm_sigmask;
2037                 sigset_t sigset, *p;
2038
2039                 p = NULL;
2040                 if (argp) {
2041                         r = -EFAULT;
2042                         if (copy_from_user(&kvm_sigmask, argp,
2043                                            sizeof kvm_sigmask))
2044                                 goto out;
2045                         r = -EINVAL;
2046                         if (kvm_sigmask.len != sizeof sigset)
2047                                 goto out;
2048                         r = -EFAULT;
2049                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2050                                            sizeof sigset))
2051                                 goto out;
2052                         p = &sigset;
2053                 }
2054                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2055                 break;
2056         }
2057         case KVM_GET_FPU: {
2058                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2059                 r = -ENOMEM;
2060                 if (!fpu)
2061                         goto out;
2062                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2063                 if (r)
2064                         goto out;
2065                 r = -EFAULT;
2066                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2067                         goto out;
2068                 r = 0;
2069                 break;
2070         }
2071         case KVM_SET_FPU: {
2072                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2073                 r = -ENOMEM;
2074                 if (!fpu)
2075                         goto out;
2076                 r = -EFAULT;
2077                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2078                         goto out;
2079                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2080                 if (r)
2081                         goto out;
2082                 r = 0;
2083                 break;
2084         }
2085         default:
2086                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2087         }
2088 out:
2089         kfree(fpu);
2090         kfree(kvm_sregs);
2091         return r;
2092 }
2093
2094 static long kvm_vm_ioctl(struct file *filp,
2095                            unsigned int ioctl, unsigned long arg)
2096 {
2097         struct kvm *kvm = filp->private_data;
2098         void __user *argp = (void __user *)arg;
2099         int r;
2100
2101         if (kvm->mm != current->mm)
2102                 return -EIO;
2103         switch (ioctl) {
2104         case KVM_CREATE_VCPU:
2105                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2106                 if (r < 0)
2107                         goto out;
2108                 break;
2109         case KVM_SET_USER_MEMORY_REGION: {
2110                 struct kvm_userspace_memory_region kvm_userspace_mem;
2111
2112                 r = -EFAULT;
2113                 if (copy_from_user(&kvm_userspace_mem, argp,
2114                                                 sizeof kvm_userspace_mem))
2115                         goto out;
2116
2117                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2118                 if (r)
2119                         goto out;
2120                 break;
2121         }
2122         case KVM_GET_DIRTY_LOG: {
2123                 struct kvm_dirty_log log;
2124
2125                 r = -EFAULT;
2126                 if (copy_from_user(&log, argp, sizeof log))
2127                         goto out;
2128                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2129                 if (r)
2130                         goto out;
2131                 break;
2132         }
2133 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2134         case KVM_REGISTER_COALESCED_MMIO: {
2135                 struct kvm_coalesced_mmio_zone zone;
2136                 r = -EFAULT;
2137                 if (copy_from_user(&zone, argp, sizeof zone))
2138                         goto out;
2139                 r = -ENXIO;
2140                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2141                 if (r)
2142                         goto out;
2143                 r = 0;
2144                 break;
2145         }
2146         case KVM_UNREGISTER_COALESCED_MMIO: {
2147                 struct kvm_coalesced_mmio_zone zone;
2148                 r = -EFAULT;
2149                 if (copy_from_user(&zone, argp, sizeof zone))
2150                         goto out;
2151                 r = -ENXIO;
2152                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2153                 if (r)
2154                         goto out;
2155                 r = 0;
2156                 break;
2157         }
2158 #endif
2159 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2160         case KVM_ASSIGN_PCI_DEVICE: {
2161                 struct kvm_assigned_pci_dev assigned_dev;
2162
2163                 r = -EFAULT;
2164                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2165                         goto out;
2166                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2167                 if (r)
2168                         goto out;
2169                 break;
2170         }
2171         case KVM_ASSIGN_IRQ: {
2172                 r = -EOPNOTSUPP;
2173                 break;
2174         }
2175 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2176         case KVM_ASSIGN_DEV_IRQ: {
2177                 struct kvm_assigned_irq assigned_irq;
2178
2179                 r = -EFAULT;
2180                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2181                         goto out;
2182                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2183                 if (r)
2184                         goto out;
2185                 break;
2186         }
2187         case KVM_DEASSIGN_DEV_IRQ: {
2188                 struct kvm_assigned_irq assigned_irq;
2189
2190                 r = -EFAULT;
2191                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2192                         goto out;
2193                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2194                 if (r)
2195                         goto out;
2196                 break;
2197         }
2198 #endif
2199 #endif
2200 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2201         case KVM_DEASSIGN_PCI_DEVICE: {
2202                 struct kvm_assigned_pci_dev assigned_dev;
2203
2204                 r = -EFAULT;
2205                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2206                         goto out;
2207                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2208                 if (r)
2209                         goto out;
2210                 break;
2211         }
2212 #endif
2213 #ifdef KVM_CAP_IRQ_ROUTING
2214         case KVM_SET_GSI_ROUTING: {
2215                 struct kvm_irq_routing routing;
2216                 struct kvm_irq_routing __user *urouting;
2217                 struct kvm_irq_routing_entry *entries;
2218
2219                 r = -EFAULT;
2220                 if (copy_from_user(&routing, argp, sizeof(routing)))
2221                         goto out;
2222                 r = -EINVAL;
2223                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2224                         goto out;
2225                 if (routing.flags)
2226                         goto out;
2227                 r = -ENOMEM;
2228                 entries = vmalloc(routing.nr * sizeof(*entries));
2229                 if (!entries)
2230                         goto out;
2231                 r = -EFAULT;
2232                 urouting = argp;
2233                 if (copy_from_user(entries, urouting->entries,
2234                                    routing.nr * sizeof(*entries)))
2235                         goto out_free_irq_routing;
2236                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2237                                         routing.flags);
2238         out_free_irq_routing:
2239                 vfree(entries);
2240                 break;
2241         }
2242 #ifdef __KVM_HAVE_MSIX
2243         case KVM_ASSIGN_SET_MSIX_NR: {
2244                 struct kvm_assigned_msix_nr entry_nr;
2245                 r = -EFAULT;
2246                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2247                         goto out;
2248                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2249                 if (r)
2250                         goto out;
2251                 break;
2252         }
2253         case KVM_ASSIGN_SET_MSIX_ENTRY: {
2254                 struct kvm_assigned_msix_entry entry;
2255                 r = -EFAULT;
2256                 if (copy_from_user(&entry, argp, sizeof entry))
2257                         goto out;
2258                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2259                 if (r)
2260                         goto out;
2261                 break;
2262         }
2263 #endif
2264 #endif /* KVM_CAP_IRQ_ROUTING */
2265         case KVM_IRQFD: {
2266                 struct kvm_irqfd data;
2267
2268                 r = -EFAULT;
2269                 if (copy_from_user(&data, argp, sizeof data))
2270                         goto out;
2271                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2272                 break;
2273         }
2274 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2275         case KVM_SET_BOOT_CPU_ID:
2276                 r = 0;
2277                 mutex_lock(&kvm->lock);
2278                 if (atomic_read(&kvm->online_vcpus) != 0)
2279                         r = -EBUSY;
2280                 else
2281                         kvm->bsp_vcpu_id = arg;
2282                 mutex_unlock(&kvm->lock);
2283                 break;
2284 #endif
2285         default:
2286                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2287         }
2288 out:
2289         return r;
2290 }
2291
2292 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2293 {
2294         struct page *page[1];
2295         unsigned long addr;
2296         int npages;
2297         gfn_t gfn = vmf->pgoff;
2298         struct kvm *kvm = vma->vm_file->private_data;
2299
2300         addr = gfn_to_hva(kvm, gfn);
2301         if (kvm_is_error_hva(addr))
2302                 return VM_FAULT_SIGBUS;
2303
2304         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2305                                 NULL);
2306         if (unlikely(npages != 1))
2307                 return VM_FAULT_SIGBUS;
2308
2309         vmf->page = page[0];
2310         return 0;
2311 }
2312
2313 static struct vm_operations_struct kvm_vm_vm_ops = {
2314         .fault = kvm_vm_fault,
2315 };
2316
2317 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2318 {
2319         vma->vm_ops = &kvm_vm_vm_ops;
2320         return 0;
2321 }
2322
2323 static struct file_operations kvm_vm_fops = {
2324         .release        = kvm_vm_release,
2325         .unlocked_ioctl = kvm_vm_ioctl,
2326         .compat_ioctl   = kvm_vm_ioctl,
2327         .mmap           = kvm_vm_mmap,
2328 };
2329
2330 static int kvm_dev_ioctl_create_vm(void)
2331 {
2332         int fd;
2333         struct kvm *kvm;
2334
2335         kvm = kvm_create_vm();
2336         if (IS_ERR(kvm))
2337                 return PTR_ERR(kvm);
2338         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2339         if (fd < 0)
2340                 kvm_put_kvm(kvm);
2341
2342         return fd;
2343 }
2344
2345 static long kvm_dev_ioctl_check_extension_generic(long arg)
2346 {
2347         switch (arg) {
2348         case KVM_CAP_USER_MEMORY:
2349         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2350         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2351 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2352         case KVM_CAP_SET_BOOT_CPU_ID:
2353 #endif
2354                 return 1;
2355 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2356         case KVM_CAP_IRQ_ROUTING:
2357                 return KVM_MAX_IRQ_ROUTES;
2358 #endif
2359         default:
2360                 break;
2361         }
2362         return kvm_dev_ioctl_check_extension(arg);
2363 }
2364
2365 static long kvm_dev_ioctl(struct file *filp,
2366                           unsigned int ioctl, unsigned long arg)
2367 {
2368         long r = -EINVAL;
2369
2370         switch (ioctl) {
2371         case KVM_GET_API_VERSION:
2372                 r = -EINVAL;
2373                 if (arg)
2374                         goto out;
2375                 r = KVM_API_VERSION;
2376                 break;
2377         case KVM_CREATE_VM:
2378                 r = -EINVAL;
2379                 if (arg)
2380                         goto out;
2381                 r = kvm_dev_ioctl_create_vm();
2382                 break;
2383         case KVM_CHECK_EXTENSION:
2384                 r = kvm_dev_ioctl_check_extension_generic(arg);
2385                 break;
2386         case KVM_GET_VCPU_MMAP_SIZE:
2387                 r = -EINVAL;
2388                 if (arg)
2389                         goto out;
2390                 r = PAGE_SIZE;     /* struct kvm_run */
2391 #ifdef CONFIG_X86
2392                 r += PAGE_SIZE;    /* pio data page */
2393 #endif
2394 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2395                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2396 #endif
2397                 break;
2398         case KVM_TRACE_ENABLE:
2399         case KVM_TRACE_PAUSE:
2400         case KVM_TRACE_DISABLE:
2401                 r = -EOPNOTSUPP;
2402                 break;
2403         default:
2404                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2405         }
2406 out:
2407         return r;
2408 }
2409
2410 static struct file_operations kvm_chardev_ops = {
2411         .unlocked_ioctl = kvm_dev_ioctl,
2412         .compat_ioctl   = kvm_dev_ioctl,
2413 };
2414
2415 static struct miscdevice kvm_dev = {
2416         KVM_MINOR,
2417         "kvm",
2418         &kvm_chardev_ops,
2419 };
2420
2421 static void hardware_enable(void *junk)
2422 {
2423         int cpu = raw_smp_processor_id();
2424
2425         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2426                 return;
2427         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2428         kvm_arch_hardware_enable(NULL);
2429 }
2430
2431 static void hardware_disable(void *junk)
2432 {
2433         int cpu = raw_smp_processor_id();
2434
2435         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2436                 return;
2437         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2438         kvm_arch_hardware_disable(NULL);
2439 }
2440
2441 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2442                            void *v)
2443 {
2444         int cpu = (long)v;
2445
2446         val &= ~CPU_TASKS_FROZEN;
2447         switch (val) {
2448         case CPU_DYING:
2449                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2450                        cpu);
2451                 hardware_disable(NULL);
2452                 break;
2453         case CPU_UP_CANCELED:
2454                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2455                        cpu);
2456                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2457                 break;
2458         case CPU_ONLINE:
2459                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2460                        cpu);
2461                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2462                 break;
2463         }
2464         return NOTIFY_OK;
2465 }
2466
2467
2468 asmlinkage void kvm_handle_fault_on_reboot(void)
2469 {
2470         if (kvm_rebooting)
2471                 /* spin while reset goes on */
2472                 while (true)
2473                         ;
2474         /* Fault while not rebooting.  We want the trace. */
2475         BUG();
2476 }
2477 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2478
2479 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2480                       void *v)
2481 {
2482         /*
2483          * Some (well, at least mine) BIOSes hang on reboot if
2484          * in vmx root mode.
2485          *
2486          * And Intel TXT required VMX off for all cpu when system shutdown.
2487          */
2488         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2489         kvm_rebooting = true;
2490         on_each_cpu(hardware_disable, NULL, 1);
2491         return NOTIFY_OK;
2492 }
2493
2494 static struct notifier_block kvm_reboot_notifier = {
2495         .notifier_call = kvm_reboot,
2496         .priority = 0,
2497 };
2498
2499 void kvm_io_bus_init(struct kvm_io_bus *bus)
2500 {
2501         memset(bus, 0, sizeof(*bus));
2502 }
2503
2504 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2505 {
2506         int i;
2507
2508         for (i = 0; i < bus->dev_count; i++) {
2509                 struct kvm_io_device *pos = bus->devs[i];
2510
2511                 kvm_iodevice_destructor(pos);
2512         }
2513 }
2514
2515 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2516                                           gpa_t addr, int len, int is_write)
2517 {
2518         int i;
2519
2520         for (i = 0; i < bus->dev_count; i++) {
2521                 struct kvm_io_device *pos = bus->devs[i];
2522
2523                 if (kvm_iodevice_in_range(pos, addr, len, is_write))
2524                         return pos;
2525         }
2526
2527         return NULL;
2528 }
2529
2530 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2531 {
2532         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2533
2534         bus->devs[bus->dev_count++] = dev;
2535 }
2536
2537 static struct notifier_block kvm_cpu_notifier = {
2538         .notifier_call = kvm_cpu_hotplug,
2539         .priority = 20, /* must be > scheduler priority */
2540 };
2541
2542 static int vm_stat_get(void *_offset, u64 *val)
2543 {
2544         unsigned offset = (long)_offset;
2545         struct kvm *kvm;
2546
2547         *val = 0;
2548         spin_lock(&kvm_lock);
2549         list_for_each_entry(kvm, &vm_list, vm_list)
2550                 *val += *(u32 *)((void *)kvm + offset);
2551         spin_unlock(&kvm_lock);
2552         return 0;
2553 }
2554
2555 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2556
2557 static int vcpu_stat_get(void *_offset, u64 *val)
2558 {
2559         unsigned offset = (long)_offset;
2560         struct kvm *kvm;
2561         struct kvm_vcpu *vcpu;
2562         int i;
2563
2564         *val = 0;
2565         spin_lock(&kvm_lock);
2566         list_for_each_entry(kvm, &vm_list, vm_list)
2567                 kvm_for_each_vcpu(i, vcpu, kvm)
2568                         *val += *(u32 *)((void *)vcpu + offset);
2569
2570         spin_unlock(&kvm_lock);
2571         return 0;
2572 }
2573
2574 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2575
2576 static struct file_operations *stat_fops[] = {
2577         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2578         [KVM_STAT_VM]   = &vm_stat_fops,
2579 };
2580
2581 static void kvm_init_debug(void)
2582 {
2583         struct kvm_stats_debugfs_item *p;
2584
2585         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2586         for (p = debugfs_entries; p->name; ++p)
2587                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2588                                                 (void *)(long)p->offset,
2589                                                 stat_fops[p->kind]);
2590 }
2591
2592 static void kvm_exit_debug(void)
2593 {
2594         struct kvm_stats_debugfs_item *p;
2595
2596         for (p = debugfs_entries; p->name; ++p)
2597                 debugfs_remove(p->dentry);
2598         debugfs_remove(kvm_debugfs_dir);
2599 }
2600
2601 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2602 {
2603         hardware_disable(NULL);
2604         return 0;
2605 }
2606
2607 static int kvm_resume(struct sys_device *dev)
2608 {
2609         hardware_enable(NULL);
2610         return 0;
2611 }
2612
2613 static struct sysdev_class kvm_sysdev_class = {
2614         .name = "kvm",
2615         .suspend = kvm_suspend,
2616         .resume = kvm_resume,
2617 };
2618
2619 static struct sys_device kvm_sysdev = {
2620         .id = 0,
2621         .cls = &kvm_sysdev_class,
2622 };
2623
2624 struct page *bad_page;
2625 pfn_t bad_pfn;
2626
2627 static inline
2628 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2629 {
2630         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2631 }
2632
2633 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2634 {
2635         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2636
2637         kvm_arch_vcpu_load(vcpu, cpu);
2638 }
2639
2640 static void kvm_sched_out(struct preempt_notifier *pn,
2641                           struct task_struct *next)
2642 {
2643         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2644
2645         kvm_arch_vcpu_put(vcpu);
2646 }
2647
2648 int kvm_init(void *opaque, unsigned int vcpu_size,
2649                   struct module *module)
2650 {
2651         int r;
2652         int cpu;
2653
2654         kvm_init_debug();
2655
2656         r = kvm_arch_init(opaque);
2657         if (r)
2658                 goto out_fail;
2659
2660         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2661
2662         if (bad_page == NULL) {
2663                 r = -ENOMEM;
2664                 goto out;
2665         }
2666
2667         bad_pfn = page_to_pfn(bad_page);
2668
2669         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2670                 r = -ENOMEM;
2671                 goto out_free_0;
2672         }
2673
2674         r = kvm_arch_hardware_setup();
2675         if (r < 0)
2676                 goto out_free_0a;
2677
2678         for_each_online_cpu(cpu) {
2679                 smp_call_function_single(cpu,
2680                                 kvm_arch_check_processor_compat,
2681                                 &r, 1);
2682                 if (r < 0)
2683                         goto out_free_1;
2684         }
2685
2686         on_each_cpu(hardware_enable, NULL, 1);
2687         r = register_cpu_notifier(&kvm_cpu_notifier);
2688         if (r)
2689                 goto out_free_2;
2690         register_reboot_notifier(&kvm_reboot_notifier);
2691
2692         r = sysdev_class_register(&kvm_sysdev_class);
2693         if (r)
2694                 goto out_free_3;
2695
2696         r = sysdev_register(&kvm_sysdev);
2697         if (r)
2698                 goto out_free_4;
2699
2700         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2701         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2702                                            __alignof__(struct kvm_vcpu),
2703                                            0, NULL);
2704         if (!kvm_vcpu_cache) {
2705                 r = -ENOMEM;
2706                 goto out_free_5;
2707         }
2708
2709         kvm_chardev_ops.owner = module;
2710         kvm_vm_fops.owner = module;
2711         kvm_vcpu_fops.owner = module;
2712
2713         r = misc_register(&kvm_dev);
2714         if (r) {
2715                 printk(KERN_ERR "kvm: misc device register failed\n");
2716                 goto out_free;
2717         }
2718
2719         kvm_preempt_ops.sched_in = kvm_sched_in;
2720         kvm_preempt_ops.sched_out = kvm_sched_out;
2721
2722         return 0;
2723
2724 out_free:
2725         kmem_cache_destroy(kvm_vcpu_cache);
2726 out_free_5:
2727         sysdev_unregister(&kvm_sysdev);
2728 out_free_4:
2729         sysdev_class_unregister(&kvm_sysdev_class);
2730 out_free_3:
2731         unregister_reboot_notifier(&kvm_reboot_notifier);
2732         unregister_cpu_notifier(&kvm_cpu_notifier);
2733 out_free_2:
2734         on_each_cpu(hardware_disable, NULL, 1);
2735 out_free_1:
2736         kvm_arch_hardware_unsetup();
2737 out_free_0a:
2738         free_cpumask_var(cpus_hardware_enabled);
2739 out_free_0:
2740         __free_page(bad_page);
2741 out:
2742         kvm_arch_exit();
2743         kvm_exit_debug();
2744 out_fail:
2745         return r;
2746 }
2747 EXPORT_SYMBOL_GPL(kvm_init);
2748
2749 void kvm_exit(void)
2750 {
2751         tracepoint_synchronize_unregister();
2752         misc_deregister(&kvm_dev);
2753         kmem_cache_destroy(kvm_vcpu_cache);
2754         sysdev_unregister(&kvm_sysdev);
2755         sysdev_class_unregister(&kvm_sysdev_class);
2756         unregister_reboot_notifier(&kvm_reboot_notifier);
2757         unregister_cpu_notifier(&kvm_cpu_notifier);
2758         on_each_cpu(hardware_disable, NULL, 1);
2759         kvm_arch_hardware_unsetup();
2760         kvm_arch_exit();
2761         kvm_exit_debug();
2762         free_cpumask_var(cpus_hardware_enabled);
2763         __free_page(bad_page);
2764 }
2765 EXPORT_SYMBOL_GPL(kvm_exit);