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         if (!dont || free->rmap != dont->rmap)
1005                 vfree(free->rmap);
1006
1007         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
1008                 vfree(free->dirty_bitmap);
1009
1010         if (!dont || free->lpage_info != dont->lpage_info)
1011                 vfree(free->lpage_info);
1012
1013         free->npages = 0;
1014         free->dirty_bitmap = NULL;
1015         free->rmap = NULL;
1016         free->lpage_info = NULL;
1017 }
1018
1019 void kvm_free_physmem(struct kvm *kvm)
1020 {
1021         int i;
1022
1023         for (i = 0; i < kvm->nmemslots; ++i)
1024                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
1025 }
1026
1027 static void kvm_destroy_vm(struct kvm *kvm)
1028 {
1029         struct mm_struct *mm = kvm->mm;
1030
1031         kvm_arch_sync_events(kvm);
1032         spin_lock(&kvm_lock);
1033         list_del(&kvm->vm_list);
1034         spin_unlock(&kvm_lock);
1035         kvm_free_irq_routing(kvm);
1036         kvm_io_bus_destroy(&kvm->pio_bus);
1037         kvm_io_bus_destroy(&kvm->mmio_bus);
1038 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1039         if (kvm->coalesced_mmio_ring != NULL)
1040                 free_page((unsigned long)kvm->coalesced_mmio_ring);
1041 #endif
1042 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1043         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1044 #else
1045         kvm_arch_flush_shadow(kvm);
1046 #endif
1047         kvm_arch_destroy_vm(kvm);
1048         mmdrop(mm);
1049 }
1050
1051 void kvm_get_kvm(struct kvm *kvm)
1052 {
1053         atomic_inc(&kvm->users_count);
1054 }
1055 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1056
1057 void kvm_put_kvm(struct kvm *kvm)
1058 {
1059         if (atomic_dec_and_test(&kvm->users_count))
1060                 kvm_destroy_vm(kvm);
1061 }
1062 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1063
1064
1065 static int kvm_vm_release(struct inode *inode, struct file *filp)
1066 {
1067         struct kvm *kvm = filp->private_data;
1068
1069         kvm_irqfd_release(kvm);
1070
1071         kvm_put_kvm(kvm);
1072         return 0;
1073 }
1074
1075 /*
1076  * Allocate some memory and give it an address in the guest physical address
1077  * space.
1078  *
1079  * Discontiguous memory is allowed, mostly for framebuffers.
1080  *
1081  * Must be called holding mmap_sem for write.
1082  */
1083 int __kvm_set_memory_region(struct kvm *kvm,
1084                             struct kvm_userspace_memory_region *mem,
1085                             int user_alloc)
1086 {
1087         int r;
1088         gfn_t base_gfn;
1089         unsigned long npages, ugfn;
1090         unsigned long largepages, i;
1091         struct kvm_memory_slot *memslot;
1092         struct kvm_memory_slot old, new;
1093
1094         r = -EINVAL;
1095         /* General sanity checks */
1096         if (mem->memory_size & (PAGE_SIZE - 1))
1097                 goto out;
1098         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1099                 goto out;
1100         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
1101                 goto out;
1102         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
1103                 goto out;
1104         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1105                 goto out;
1106
1107         memslot = &kvm->memslots[mem->slot];
1108         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1109         npages = mem->memory_size >> PAGE_SHIFT;
1110
1111         if (!npages)
1112                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1113
1114         new = old = *memslot;
1115
1116         new.base_gfn = base_gfn;
1117         new.npages = npages;
1118         new.flags = mem->flags;
1119
1120         /* Disallow changing a memory slot's size. */
1121         r = -EINVAL;
1122         if (npages && old.npages && npages != old.npages)
1123                 goto out_free;
1124
1125         /* Check for overlaps */
1126         r = -EEXIST;
1127         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1128                 struct kvm_memory_slot *s = &kvm->memslots[i];
1129
1130                 if (s == memslot || !s->npages)
1131                         continue;
1132                 if (!((base_gfn + npages <= s->base_gfn) ||
1133                       (base_gfn >= s->base_gfn + s->npages)))
1134                         goto out_free;
1135         }
1136
1137         /* Free page dirty bitmap if unneeded */
1138         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1139                 new.dirty_bitmap = NULL;
1140
1141         r = -ENOMEM;
1142
1143         /* Allocate if a slot is being created */
1144 #ifndef CONFIG_S390
1145         if (npages && !new.rmap) {
1146                 new.rmap = vmalloc(npages * sizeof(struct page *));
1147
1148                 if (!new.rmap)
1149                         goto out_free;
1150
1151                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1152
1153                 new.user_alloc = user_alloc;
1154                 /*
1155                  * hva_to_rmmap() serialzies with the mmu_lock and to be
1156                  * safe it has to ignore memslots with !user_alloc &&
1157                  * !userspace_addr.
1158                  */
1159                 if (user_alloc)
1160                         new.userspace_addr = mem->userspace_addr;
1161                 else
1162                         new.userspace_addr = 0;
1163         }
1164         if (npages && !new.lpage_info) {
1165                 largepages = 1 + (base_gfn + npages - 1) / KVM_PAGES_PER_HPAGE;
1166                 largepages -= base_gfn / KVM_PAGES_PER_HPAGE;
1167
1168                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1169
1170                 if (!new.lpage_info)
1171                         goto out_free;
1172
1173                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1174
1175                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1176                         new.lpage_info[0].write_count = 1;
1177                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1178                         new.lpage_info[largepages-1].write_count = 1;
1179                 ugfn = new.userspace_addr >> PAGE_SHIFT;
1180                 /*
1181                  * If the gfn and userspace address are not aligned wrt each
1182                  * other, or if explicitly asked to, disable large page
1183                  * support for this slot
1184                  */
1185                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE - 1) ||
1186                     !largepages_enabled)
1187                         for (i = 0; i < largepages; ++i)
1188                                 new.lpage_info[i].write_count = 1;
1189         }
1190
1191         /* Allocate page dirty bitmap if needed */
1192         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1193                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1194
1195                 new.dirty_bitmap = vmalloc(dirty_bytes);
1196                 if (!new.dirty_bitmap)
1197                         goto out_free;
1198                 memset(new.dirty_bitmap, 0, dirty_bytes);
1199                 if (old.npages)
1200                         kvm_arch_flush_shadow(kvm);
1201         }
1202 #endif /* not defined CONFIG_S390 */
1203
1204         if (!npages)
1205                 kvm_arch_flush_shadow(kvm);
1206
1207         spin_lock(&kvm->mmu_lock);
1208         if (mem->slot >= kvm->nmemslots)
1209                 kvm->nmemslots = mem->slot + 1;
1210
1211         *memslot = new;
1212         spin_unlock(&kvm->mmu_lock);
1213
1214         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1215         if (r) {
1216                 spin_lock(&kvm->mmu_lock);
1217                 *memslot = old;
1218                 spin_unlock(&kvm->mmu_lock);
1219                 goto out_free;
1220         }
1221
1222         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1223         /* Slot deletion case: we have to update the current slot */
1224         spin_lock(&kvm->mmu_lock);
1225         if (!npages)
1226                 *memslot = old;
1227         spin_unlock(&kvm->mmu_lock);
1228 #ifdef CONFIG_DMAR
1229         /* map the pages in iommu page table */
1230         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1231         if (r)
1232                 goto out;
1233 #endif
1234         return 0;
1235
1236 out_free:
1237         kvm_free_physmem_slot(&new, &old);
1238 out:
1239         return r;
1240
1241 }
1242 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1243
1244 int kvm_set_memory_region(struct kvm *kvm,
1245                           struct kvm_userspace_memory_region *mem,
1246                           int user_alloc)
1247 {
1248         int r;
1249
1250         down_write(&kvm->slots_lock);
1251         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1252         up_write(&kvm->slots_lock);
1253         return r;
1254 }
1255 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1256
1257 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1258                                    struct
1259                                    kvm_userspace_memory_region *mem,
1260                                    int user_alloc)
1261 {
1262         if (mem->slot >= KVM_MEMORY_SLOTS)
1263                 return -EINVAL;
1264         return kvm_set_memory_region(kvm, mem, user_alloc);
1265 }
1266
1267 int kvm_get_dirty_log(struct kvm *kvm,
1268                         struct kvm_dirty_log *log, int *is_dirty)
1269 {
1270         struct kvm_memory_slot *memslot;
1271         int r, i;
1272         int n;
1273         unsigned long any = 0;
1274
1275         r = -EINVAL;
1276         if (log->slot >= KVM_MEMORY_SLOTS)
1277                 goto out;
1278
1279         memslot = &kvm->memslots[log->slot];
1280         r = -ENOENT;
1281         if (!memslot->dirty_bitmap)
1282                 goto out;
1283
1284         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1285
1286         for (i = 0; !any && i < n/sizeof(long); ++i)
1287                 any = memslot->dirty_bitmap[i];
1288
1289         r = -EFAULT;
1290         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1291                 goto out;
1292
1293         if (any)
1294                 *is_dirty = 1;
1295
1296         r = 0;
1297 out:
1298         return r;
1299 }
1300
1301 void kvm_disable_largepages(void)
1302 {
1303         largepages_enabled = false;
1304 }
1305 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1306
1307 int is_error_page(struct page *page)
1308 {
1309         return page == bad_page;
1310 }
1311 EXPORT_SYMBOL_GPL(is_error_page);
1312
1313 int is_error_pfn(pfn_t pfn)
1314 {
1315         return pfn == bad_pfn;
1316 }
1317 EXPORT_SYMBOL_GPL(is_error_pfn);
1318
1319 static inline unsigned long bad_hva(void)
1320 {
1321         return PAGE_OFFSET;
1322 }
1323
1324 int kvm_is_error_hva(unsigned long addr)
1325 {
1326         return addr == bad_hva();
1327 }
1328 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1329
1330 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1331 {
1332         int i;
1333
1334         for (i = 0; i < kvm->nmemslots; ++i) {
1335                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1336
1337                 if (gfn >= memslot->base_gfn
1338                     && gfn < memslot->base_gfn + memslot->npages)
1339                         return memslot;
1340         }
1341         return NULL;
1342 }
1343 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1344
1345 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1346 {
1347         gfn = unalias_gfn(kvm, gfn);
1348         return gfn_to_memslot_unaliased(kvm, gfn);
1349 }
1350
1351 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1352 {
1353         int i;
1354
1355         gfn = unalias_gfn(kvm, gfn);
1356         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1357                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1358
1359                 if (gfn >= memslot->base_gfn
1360                     && gfn < memslot->base_gfn + memslot->npages)
1361                         return 1;
1362         }
1363         return 0;
1364 }
1365 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1366
1367 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1368 {
1369         struct kvm_memory_slot *slot;
1370
1371         gfn = unalias_gfn(kvm, gfn);
1372         slot = gfn_to_memslot_unaliased(kvm, gfn);
1373         if (!slot)
1374                 return bad_hva();
1375         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1376 }
1377 EXPORT_SYMBOL_GPL(gfn_to_hva);
1378
1379 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1380 {
1381         struct page *page[1];
1382         unsigned long addr;
1383         int npages;
1384         pfn_t pfn;
1385
1386         might_sleep();
1387
1388         addr = gfn_to_hva(kvm, gfn);
1389         if (kvm_is_error_hva(addr)) {
1390                 get_page(bad_page);
1391                 return page_to_pfn(bad_page);
1392         }
1393
1394         npages = get_user_pages_fast(addr, 1, 1, page);
1395
1396         if (unlikely(npages != 1)) {
1397                 struct vm_area_struct *vma;
1398
1399                 down_read(&current->mm->mmap_sem);
1400                 vma = find_vma(current->mm, addr);
1401
1402                 if (vma == NULL || addr < vma->vm_start ||
1403                     !(vma->vm_flags & VM_PFNMAP)) {
1404                         up_read(&current->mm->mmap_sem);
1405                         get_page(bad_page);
1406                         return page_to_pfn(bad_page);
1407                 }
1408
1409                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1410                 up_read(&current->mm->mmap_sem);
1411                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1412         } else
1413                 pfn = page_to_pfn(page[0]);
1414
1415         return pfn;
1416 }
1417
1418 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1419
1420 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1421 {
1422         pfn_t pfn;
1423
1424         pfn = gfn_to_pfn(kvm, gfn);
1425         if (!kvm_is_mmio_pfn(pfn))
1426                 return pfn_to_page(pfn);
1427
1428         WARN_ON(kvm_is_mmio_pfn(pfn));
1429
1430         get_page(bad_page);
1431         return bad_page;
1432 }
1433
1434 EXPORT_SYMBOL_GPL(gfn_to_page);
1435
1436 void kvm_release_page_clean(struct page *page)
1437 {
1438         kvm_release_pfn_clean(page_to_pfn(page));
1439 }
1440 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1441
1442 void kvm_release_pfn_clean(pfn_t pfn)
1443 {
1444         if (!kvm_is_mmio_pfn(pfn))
1445                 put_page(pfn_to_page(pfn));
1446 }
1447 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1448
1449 void kvm_release_page_dirty(struct page *page)
1450 {
1451         kvm_release_pfn_dirty(page_to_pfn(page));
1452 }
1453 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1454
1455 void kvm_release_pfn_dirty(pfn_t pfn)
1456 {
1457         kvm_set_pfn_dirty(pfn);
1458         kvm_release_pfn_clean(pfn);
1459 }
1460 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1461
1462 void kvm_set_page_dirty(struct page *page)
1463 {
1464         kvm_set_pfn_dirty(page_to_pfn(page));
1465 }
1466 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1467
1468 void kvm_set_pfn_dirty(pfn_t pfn)
1469 {
1470         if (!kvm_is_mmio_pfn(pfn)) {
1471                 struct page *page = pfn_to_page(pfn);
1472                 if (!PageReserved(page))
1473                         SetPageDirty(page);
1474         }
1475 }
1476 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1477
1478 void kvm_set_pfn_accessed(pfn_t pfn)
1479 {
1480         if (!kvm_is_mmio_pfn(pfn))
1481                 mark_page_accessed(pfn_to_page(pfn));
1482 }
1483 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1484
1485 void kvm_get_pfn(pfn_t pfn)
1486 {
1487         if (!kvm_is_mmio_pfn(pfn))
1488                 get_page(pfn_to_page(pfn));
1489 }
1490 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1491
1492 static int next_segment(unsigned long len, int offset)
1493 {
1494         if (len > PAGE_SIZE - offset)
1495                 return PAGE_SIZE - offset;
1496         else
1497                 return len;
1498 }
1499
1500 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1501                         int len)
1502 {
1503         int r;
1504         unsigned long addr;
1505
1506         addr = gfn_to_hva(kvm, gfn);
1507         if (kvm_is_error_hva(addr))
1508                 return -EFAULT;
1509         r = copy_from_user(data, (void __user *)addr + offset, len);
1510         if (r)
1511                 return -EFAULT;
1512         return 0;
1513 }
1514 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1515
1516 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1517 {
1518         gfn_t gfn = gpa >> PAGE_SHIFT;
1519         int seg;
1520         int offset = offset_in_page(gpa);
1521         int ret;
1522
1523         while ((seg = next_segment(len, offset)) != 0) {
1524                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1525                 if (ret < 0)
1526                         return ret;
1527                 offset = 0;
1528                 len -= seg;
1529                 data += seg;
1530                 ++gfn;
1531         }
1532         return 0;
1533 }
1534 EXPORT_SYMBOL_GPL(kvm_read_guest);
1535
1536 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1537                           unsigned long len)
1538 {
1539         int r;
1540         unsigned long addr;
1541         gfn_t gfn = gpa >> PAGE_SHIFT;
1542         int offset = offset_in_page(gpa);
1543
1544         addr = gfn_to_hva(kvm, gfn);
1545         if (kvm_is_error_hva(addr))
1546                 return -EFAULT;
1547         pagefault_disable();
1548         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1549         pagefault_enable();
1550         if (r)
1551                 return -EFAULT;
1552         return 0;
1553 }
1554 EXPORT_SYMBOL(kvm_read_guest_atomic);
1555
1556 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1557                          int offset, int len)
1558 {
1559         int r;
1560         unsigned long addr;
1561
1562         addr = gfn_to_hva(kvm, gfn);
1563         if (kvm_is_error_hva(addr))
1564                 return -EFAULT;
1565         r = copy_to_user((void __user *)addr + offset, data, len);
1566         if (r)
1567                 return -EFAULT;
1568         mark_page_dirty(kvm, gfn);
1569         return 0;
1570 }
1571 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1572
1573 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1574                     unsigned long len)
1575 {
1576         gfn_t gfn = gpa >> PAGE_SHIFT;
1577         int seg;
1578         int offset = offset_in_page(gpa);
1579         int ret;
1580
1581         while ((seg = next_segment(len, offset)) != 0) {
1582                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1583                 if (ret < 0)
1584                         return ret;
1585                 offset = 0;
1586                 len -= seg;
1587                 data += seg;
1588                 ++gfn;
1589         }
1590         return 0;
1591 }
1592
1593 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1594 {
1595         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1596 }
1597 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1598
1599 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1600 {
1601         gfn_t gfn = gpa >> PAGE_SHIFT;
1602         int seg;
1603         int offset = offset_in_page(gpa);
1604         int ret;
1605
1606         while ((seg = next_segment(len, offset)) != 0) {
1607                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1608                 if (ret < 0)
1609                         return ret;
1610                 offset = 0;
1611                 len -= seg;
1612                 ++gfn;
1613         }
1614         return 0;
1615 }
1616 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1617
1618 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1619 {
1620         struct kvm_memory_slot *memslot;
1621
1622         gfn = unalias_gfn(kvm, gfn);
1623         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1624         if (memslot && memslot->dirty_bitmap) {
1625                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1626
1627                 /* avoid RMW */
1628                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1629                         set_bit(rel_gfn, memslot->dirty_bitmap);
1630         }
1631 }
1632
1633 /*
1634  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1635  */
1636 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1637 {
1638         DEFINE_WAIT(wait);
1639
1640         for (;;) {
1641                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1642
1643                 if ((kvm_arch_interrupt_allowed(vcpu) &&
1644                                         kvm_cpu_has_interrupt(vcpu)) ||
1645                                 kvm_arch_vcpu_runnable(vcpu)) {
1646                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1647                         break;
1648                 }
1649                 if (kvm_cpu_has_pending_timer(vcpu))
1650                         break;
1651                 if (signal_pending(current))
1652                         break;
1653
1654                 vcpu_put(vcpu);
1655                 schedule();
1656                 vcpu_load(vcpu);
1657         }
1658
1659         finish_wait(&vcpu->wq, &wait);
1660 }
1661
1662 void kvm_resched(struct kvm_vcpu *vcpu)
1663 {
1664         if (!need_resched())
1665                 return;
1666         cond_resched();
1667 }
1668 EXPORT_SYMBOL_GPL(kvm_resched);
1669
1670 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1671 {
1672         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1673         struct page *page;
1674
1675         if (vmf->pgoff == 0)
1676                 page = virt_to_page(vcpu->run);
1677 #ifdef CONFIG_X86
1678         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1679                 page = virt_to_page(vcpu->arch.pio_data);
1680 #endif
1681 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1682         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1683                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1684 #endif
1685         else
1686                 return VM_FAULT_SIGBUS;
1687         get_page(page);
1688         vmf->page = page;
1689         return 0;
1690 }
1691
1692 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1693         .fault = kvm_vcpu_fault,
1694 };
1695
1696 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1697 {
1698         vma->vm_ops = &kvm_vcpu_vm_ops;
1699         return 0;
1700 }
1701
1702 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1703 {
1704         struct kvm_vcpu *vcpu = filp->private_data;
1705
1706         kvm_put_kvm(vcpu->kvm);
1707         return 0;
1708 }
1709
1710 static struct file_operations kvm_vcpu_fops = {
1711         .release        = kvm_vcpu_release,
1712         .unlocked_ioctl = kvm_vcpu_ioctl,
1713         .compat_ioctl   = kvm_vcpu_ioctl,
1714         .mmap           = kvm_vcpu_mmap,
1715 };
1716
1717 /*
1718  * Allocates an inode for the vcpu.
1719  */
1720 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1721 {
1722         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1723 }
1724
1725 /*
1726  * Creates some virtual cpus.  Good luck creating more than one.
1727  */
1728 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1729 {
1730         int r;
1731         struct kvm_vcpu *vcpu, *v;
1732
1733         vcpu = kvm_arch_vcpu_create(kvm, id);
1734         if (IS_ERR(vcpu))
1735                 return PTR_ERR(vcpu);
1736
1737         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1738
1739         r = kvm_arch_vcpu_setup(vcpu);
1740         if (r)
1741                 return r;
1742
1743         mutex_lock(&kvm->lock);
1744         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1745                 r = -EINVAL;
1746                 goto vcpu_destroy;
1747         }
1748
1749         kvm_for_each_vcpu(r, v, kvm)
1750                 if (v->vcpu_id == id) {
1751                         r = -EEXIST;
1752                         goto vcpu_destroy;
1753                 }
1754
1755         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1756
1757         /* Now it's all set up, let userspace reach it */
1758         kvm_get_kvm(kvm);
1759         r = create_vcpu_fd(vcpu);
1760         if (r < 0) {
1761                 kvm_put_kvm(kvm);
1762                 goto vcpu_destroy;
1763         }
1764
1765         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1766         smp_wmb();
1767         atomic_inc(&kvm->online_vcpus);
1768
1769 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1770         if (kvm->bsp_vcpu_id == id)
1771                 kvm->bsp_vcpu = vcpu;
1772 #endif
1773         mutex_unlock(&kvm->lock);
1774         return r;
1775
1776 vcpu_destroy:
1777         mutex_unlock(&kvm->lock);
1778         kvm_arch_vcpu_destroy(vcpu);
1779         return r;
1780 }
1781
1782 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1783 {
1784         if (sigset) {
1785                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1786                 vcpu->sigset_active = 1;
1787                 vcpu->sigset = *sigset;
1788         } else
1789                 vcpu->sigset_active = 0;
1790         return 0;
1791 }
1792
1793 #ifdef __KVM_HAVE_MSIX
1794 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1795                                     struct kvm_assigned_msix_nr *entry_nr)
1796 {
1797         int r = 0;
1798         struct kvm_assigned_dev_kernel *adev;
1799
1800         mutex_lock(&kvm->lock);
1801
1802         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1803                                       entry_nr->assigned_dev_id);
1804         if (!adev) {
1805                 r = -EINVAL;
1806                 goto msix_nr_out;
1807         }
1808
1809         if (adev->entries_nr == 0) {
1810                 adev->entries_nr = entry_nr->entry_nr;
1811                 if (adev->entries_nr == 0 ||
1812                     adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1813                         r = -EINVAL;
1814                         goto msix_nr_out;
1815                 }
1816
1817                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1818                                                 entry_nr->entry_nr,
1819                                                 GFP_KERNEL);
1820                 if (!adev->host_msix_entries) {
1821                         r = -ENOMEM;
1822                         goto msix_nr_out;
1823                 }
1824                 adev->guest_msix_entries = kzalloc(
1825                                 sizeof(struct kvm_guest_msix_entry) *
1826                                 entry_nr->entry_nr, GFP_KERNEL);
1827                 if (!adev->guest_msix_entries) {
1828                         kfree(adev->host_msix_entries);
1829                         r = -ENOMEM;
1830                         goto msix_nr_out;
1831                 }
1832         } else /* Not allowed set MSI-X number twice */
1833                 r = -EINVAL;
1834 msix_nr_out:
1835         mutex_unlock(&kvm->lock);
1836         return r;
1837 }
1838
1839 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1840                                        struct kvm_assigned_msix_entry *entry)
1841 {
1842         int r = 0, i;
1843         struct kvm_assigned_dev_kernel *adev;
1844
1845         mutex_lock(&kvm->lock);
1846
1847         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1848                                       entry->assigned_dev_id);
1849
1850         if (!adev) {
1851                 r = -EINVAL;
1852                 goto msix_entry_out;
1853         }
1854
1855         for (i = 0; i < adev->entries_nr; i++)
1856                 if (adev->guest_msix_entries[i].vector == 0 ||
1857                     adev->guest_msix_entries[i].entry == entry->entry) {
1858                         adev->guest_msix_entries[i].entry = entry->entry;
1859                         adev->guest_msix_entries[i].vector = entry->gsi;
1860                         adev->host_msix_entries[i].entry = entry->entry;
1861                         break;
1862                 }
1863         if (i == adev->entries_nr) {
1864                 r = -ENOSPC;
1865                 goto msix_entry_out;
1866         }
1867
1868 msix_entry_out:
1869         mutex_unlock(&kvm->lock);
1870
1871         return r;
1872 }
1873 #endif
1874
1875 static long kvm_vcpu_ioctl(struct file *filp,
1876                            unsigned int ioctl, unsigned long arg)
1877 {
1878         struct kvm_vcpu *vcpu = filp->private_data;
1879         void __user *argp = (void __user *)arg;
1880         int r;
1881         struct kvm_fpu *fpu = NULL;
1882         struct kvm_sregs *kvm_sregs = NULL;
1883
1884         if (vcpu->kvm->mm != current->mm)
1885                 return -EIO;
1886         switch (ioctl) {
1887         case KVM_RUN:
1888                 r = -EINVAL;
1889                 if (arg)
1890                         goto out;
1891                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1892                 break;
1893         case KVM_GET_REGS: {
1894                 struct kvm_regs *kvm_regs;
1895
1896                 r = -ENOMEM;
1897                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1898                 if (!kvm_regs)
1899                         goto out;
1900                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1901                 if (r)
1902                         goto out_free1;
1903                 r = -EFAULT;
1904                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1905                         goto out_free1;
1906                 r = 0;
1907 out_free1:
1908                 kfree(kvm_regs);
1909                 break;
1910         }
1911         case KVM_SET_REGS: {
1912                 struct kvm_regs *kvm_regs;
1913
1914                 r = -ENOMEM;
1915                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1916                 if (!kvm_regs)
1917                         goto out;
1918                 r = -EFAULT;
1919                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1920                         goto out_free2;
1921                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1922                 if (r)
1923                         goto out_free2;
1924                 r = 0;
1925 out_free2:
1926                 kfree(kvm_regs);
1927                 break;
1928         }
1929         case KVM_GET_SREGS: {
1930                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1931                 r = -ENOMEM;
1932                 if (!kvm_sregs)
1933                         goto out;
1934                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1935                 if (r)
1936                         goto out;
1937                 r = -EFAULT;
1938                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1939                         goto out;
1940                 r = 0;
1941                 break;
1942         }
1943         case KVM_SET_SREGS: {
1944                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1945                 r = -ENOMEM;
1946                 if (!kvm_sregs)
1947                         goto out;
1948                 r = -EFAULT;
1949                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1950                         goto out;
1951                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1952                 if (r)
1953                         goto out;
1954                 r = 0;
1955                 break;
1956         }
1957         case KVM_GET_MP_STATE: {
1958                 struct kvm_mp_state mp_state;
1959
1960                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1961                 if (r)
1962                         goto out;
1963                 r = -EFAULT;
1964                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1965                         goto out;
1966                 r = 0;
1967                 break;
1968         }
1969         case KVM_SET_MP_STATE: {
1970                 struct kvm_mp_state mp_state;
1971
1972                 r = -EFAULT;
1973                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1974                         goto out;
1975                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1976                 if (r)
1977                         goto out;
1978                 r = 0;
1979                 break;
1980         }
1981         case KVM_TRANSLATE: {
1982                 struct kvm_translation tr;
1983
1984                 r = -EFAULT;
1985                 if (copy_from_user(&tr, argp, sizeof tr))
1986                         goto out;
1987                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1988                 if (r)
1989                         goto out;
1990                 r = -EFAULT;
1991                 if (copy_to_user(argp, &tr, sizeof tr))
1992                         goto out;
1993                 r = 0;
1994                 break;
1995         }
1996         case KVM_SET_GUEST_DEBUG: {
1997                 struct kvm_guest_debug dbg;
1998
1999                 r = -EFAULT;
2000                 if (copy_from_user(&dbg, argp, sizeof dbg))
2001                         goto out;
2002                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2003                 if (r)
2004                         goto out;
2005                 r = 0;
2006                 break;
2007         }
2008         case KVM_SET_SIGNAL_MASK: {
2009                 struct kvm_signal_mask __user *sigmask_arg = argp;
2010                 struct kvm_signal_mask kvm_sigmask;
2011                 sigset_t sigset, *p;
2012
2013                 p = NULL;
2014                 if (argp) {
2015                         r = -EFAULT;
2016                         if (copy_from_user(&kvm_sigmask, argp,
2017                                            sizeof kvm_sigmask))
2018                                 goto out;
2019                         r = -EINVAL;
2020                         if (kvm_sigmask.len != sizeof sigset)
2021                                 goto out;
2022                         r = -EFAULT;
2023                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2024                                            sizeof sigset))
2025                                 goto out;
2026                         p = &sigset;
2027                 }
2028                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2029                 break;
2030         }
2031         case KVM_GET_FPU: {
2032                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2033                 r = -ENOMEM;
2034                 if (!fpu)
2035                         goto out;
2036                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2037                 if (r)
2038                         goto out;
2039                 r = -EFAULT;
2040                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2041                         goto out;
2042                 r = 0;
2043                 break;
2044         }
2045         case KVM_SET_FPU: {
2046                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2047                 r = -ENOMEM;
2048                 if (!fpu)
2049                         goto out;
2050                 r = -EFAULT;
2051                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2052                         goto out;
2053                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2054                 if (r)
2055                         goto out;
2056                 r = 0;
2057                 break;
2058         }
2059         default:
2060                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2061         }
2062 out:
2063         kfree(fpu);
2064         kfree(kvm_sregs);
2065         return r;
2066 }
2067
2068 static long kvm_vm_ioctl(struct file *filp,
2069                            unsigned int ioctl, unsigned long arg)
2070 {
2071         struct kvm *kvm = filp->private_data;
2072         void __user *argp = (void __user *)arg;
2073         int r;
2074
2075         if (kvm->mm != current->mm)
2076                 return -EIO;
2077         switch (ioctl) {
2078         case KVM_CREATE_VCPU:
2079                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2080                 if (r < 0)
2081                         goto out;
2082                 break;
2083         case KVM_SET_USER_MEMORY_REGION: {
2084                 struct kvm_userspace_memory_region kvm_userspace_mem;
2085
2086                 r = -EFAULT;
2087                 if (copy_from_user(&kvm_userspace_mem, argp,
2088                                                 sizeof kvm_userspace_mem))
2089                         goto out;
2090
2091                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2092                 if (r)
2093                         goto out;
2094                 break;
2095         }
2096         case KVM_GET_DIRTY_LOG: {
2097                 struct kvm_dirty_log log;
2098
2099                 r = -EFAULT;
2100                 if (copy_from_user(&log, argp, sizeof log))
2101                         goto out;
2102                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2103                 if (r)
2104                         goto out;
2105                 break;
2106         }
2107 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2108         case KVM_REGISTER_COALESCED_MMIO: {
2109                 struct kvm_coalesced_mmio_zone zone;
2110                 r = -EFAULT;
2111                 if (copy_from_user(&zone, argp, sizeof zone))
2112                         goto out;
2113                 r = -ENXIO;
2114                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2115                 if (r)
2116                         goto out;
2117                 r = 0;
2118                 break;
2119         }
2120         case KVM_UNREGISTER_COALESCED_MMIO: {
2121                 struct kvm_coalesced_mmio_zone zone;
2122                 r = -EFAULT;
2123                 if (copy_from_user(&zone, argp, sizeof zone))
2124                         goto out;
2125                 r = -ENXIO;
2126                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2127                 if (r)
2128                         goto out;
2129                 r = 0;
2130                 break;
2131         }
2132 #endif
2133 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2134         case KVM_ASSIGN_PCI_DEVICE: {
2135                 struct kvm_assigned_pci_dev assigned_dev;
2136
2137                 r = -EFAULT;
2138                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2139                         goto out;
2140                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2141                 if (r)
2142                         goto out;
2143                 break;
2144         }
2145         case KVM_ASSIGN_IRQ: {
2146                 r = -EOPNOTSUPP;
2147                 break;
2148         }
2149 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2150         case KVM_ASSIGN_DEV_IRQ: {
2151                 struct kvm_assigned_irq assigned_irq;
2152
2153                 r = -EFAULT;
2154                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2155                         goto out;
2156                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2157                 if (r)
2158                         goto out;
2159                 break;
2160         }
2161         case KVM_DEASSIGN_DEV_IRQ: {
2162                 struct kvm_assigned_irq assigned_irq;
2163
2164                 r = -EFAULT;
2165                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2166                         goto out;
2167                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2168                 if (r)
2169                         goto out;
2170                 break;
2171         }
2172 #endif
2173 #endif
2174 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2175         case KVM_DEASSIGN_PCI_DEVICE: {
2176                 struct kvm_assigned_pci_dev assigned_dev;
2177
2178                 r = -EFAULT;
2179                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2180                         goto out;
2181                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2182                 if (r)
2183                         goto out;
2184                 break;
2185         }
2186 #endif
2187 #ifdef KVM_CAP_IRQ_ROUTING
2188         case KVM_SET_GSI_ROUTING: {
2189                 struct kvm_irq_routing routing;
2190                 struct kvm_irq_routing __user *urouting;
2191                 struct kvm_irq_routing_entry *entries;
2192
2193                 r = -EFAULT;
2194                 if (copy_from_user(&routing, argp, sizeof(routing)))
2195                         goto out;
2196                 r = -EINVAL;
2197                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2198                         goto out;
2199                 if (routing.flags)
2200                         goto out;
2201                 r = -ENOMEM;
2202                 entries = vmalloc(routing.nr * sizeof(*entries));
2203                 if (!entries)
2204                         goto out;
2205                 r = -EFAULT;
2206                 urouting = argp;
2207                 if (copy_from_user(entries, urouting->entries,
2208                                    routing.nr * sizeof(*entries)))
2209                         goto out_free_irq_routing;
2210                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2211                                         routing.flags);
2212         out_free_irq_routing:
2213                 vfree(entries);
2214                 break;
2215         }
2216 #ifdef __KVM_HAVE_MSIX
2217         case KVM_ASSIGN_SET_MSIX_NR: {
2218                 struct kvm_assigned_msix_nr entry_nr;
2219                 r = -EFAULT;
2220                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2221                         goto out;
2222                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2223                 if (r)
2224                         goto out;
2225                 break;
2226         }
2227         case KVM_ASSIGN_SET_MSIX_ENTRY: {
2228                 struct kvm_assigned_msix_entry entry;
2229                 r = -EFAULT;
2230                 if (copy_from_user(&entry, argp, sizeof entry))
2231                         goto out;
2232                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2233                 if (r)
2234                         goto out;
2235                 break;
2236         }
2237 #endif
2238 #endif /* KVM_CAP_IRQ_ROUTING */
2239         case KVM_IRQFD: {
2240                 struct kvm_irqfd data;
2241
2242                 r = -EFAULT;
2243                 if (copy_from_user(&data, argp, sizeof data))
2244                         goto out;
2245                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2246                 break;
2247         }
2248 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2249         case KVM_SET_BOOT_CPU_ID:
2250                 r = 0;
2251                 if (atomic_read(&kvm->online_vcpus) != 0)
2252                         r = -EBUSY;
2253                 else
2254                         kvm->bsp_vcpu_id = arg;
2255                 break;
2256 #endif
2257         default:
2258                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2259         }
2260 out:
2261         return r;
2262 }
2263
2264 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2265 {
2266         struct page *page[1];
2267         unsigned long addr;
2268         int npages;
2269         gfn_t gfn = vmf->pgoff;
2270         struct kvm *kvm = vma->vm_file->private_data;
2271
2272         addr = gfn_to_hva(kvm, gfn);
2273         if (kvm_is_error_hva(addr))
2274                 return VM_FAULT_SIGBUS;
2275
2276         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2277                                 NULL);
2278         if (unlikely(npages != 1))
2279                 return VM_FAULT_SIGBUS;
2280
2281         vmf->page = page[0];
2282         return 0;
2283 }
2284
2285 static struct vm_operations_struct kvm_vm_vm_ops = {
2286         .fault = kvm_vm_fault,
2287 };
2288
2289 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2290 {
2291         vma->vm_ops = &kvm_vm_vm_ops;
2292         return 0;
2293 }
2294
2295 static struct file_operations kvm_vm_fops = {
2296         .release        = kvm_vm_release,
2297         .unlocked_ioctl = kvm_vm_ioctl,
2298         .compat_ioctl   = kvm_vm_ioctl,
2299         .mmap           = kvm_vm_mmap,
2300 };
2301
2302 static int kvm_dev_ioctl_create_vm(void)
2303 {
2304         int fd;
2305         struct kvm *kvm;
2306
2307         kvm = kvm_create_vm();
2308         if (IS_ERR(kvm))
2309                 return PTR_ERR(kvm);
2310         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2311         if (fd < 0)
2312                 kvm_put_kvm(kvm);
2313
2314         return fd;
2315 }
2316
2317 static long kvm_dev_ioctl_check_extension_generic(long arg)
2318 {
2319         switch (arg) {
2320         case KVM_CAP_USER_MEMORY:
2321         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2322         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2323 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2324         case KVM_CAP_SET_BOOT_CPU_ID:
2325 #endif
2326                 return 1;
2327 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2328         case KVM_CAP_IRQ_ROUTING:
2329                 return KVM_MAX_IRQ_ROUTES;
2330 #endif
2331         default:
2332                 break;
2333         }
2334         return kvm_dev_ioctl_check_extension(arg);
2335 }
2336
2337 static long kvm_dev_ioctl(struct file *filp,
2338                           unsigned int ioctl, unsigned long arg)
2339 {
2340         long r = -EINVAL;
2341
2342         switch (ioctl) {
2343         case KVM_GET_API_VERSION:
2344                 r = -EINVAL;
2345                 if (arg)
2346                         goto out;
2347                 r = KVM_API_VERSION;
2348                 break;
2349         case KVM_CREATE_VM:
2350                 r = -EINVAL;
2351                 if (arg)
2352                         goto out;
2353                 r = kvm_dev_ioctl_create_vm();
2354                 break;
2355         case KVM_CHECK_EXTENSION:
2356                 r = kvm_dev_ioctl_check_extension_generic(arg);
2357                 break;
2358         case KVM_GET_VCPU_MMAP_SIZE:
2359                 r = -EINVAL;
2360                 if (arg)
2361                         goto out;
2362                 r = PAGE_SIZE;     /* struct kvm_run */
2363 #ifdef CONFIG_X86
2364                 r += PAGE_SIZE;    /* pio data page */
2365 #endif
2366 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2367                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2368 #endif
2369                 break;
2370         case KVM_TRACE_ENABLE:
2371         case KVM_TRACE_PAUSE:
2372         case KVM_TRACE_DISABLE:
2373                 r = kvm_trace_ioctl(ioctl, arg);
2374                 break;
2375         default:
2376                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2377         }
2378 out:
2379         return r;
2380 }
2381
2382 static struct file_operations kvm_chardev_ops = {
2383         .unlocked_ioctl = kvm_dev_ioctl,
2384         .compat_ioctl   = kvm_dev_ioctl,
2385 };
2386
2387 static struct miscdevice kvm_dev = {
2388         KVM_MINOR,
2389         "kvm",
2390         &kvm_chardev_ops,
2391 };
2392
2393 static void hardware_enable(void *junk)
2394 {
2395         int cpu = raw_smp_processor_id();
2396
2397         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2398                 return;
2399         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2400         kvm_arch_hardware_enable(NULL);
2401 }
2402
2403 static void hardware_disable(void *junk)
2404 {
2405         int cpu = raw_smp_processor_id();
2406
2407         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2408                 return;
2409         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2410         kvm_arch_hardware_disable(NULL);
2411 }
2412
2413 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2414                            void *v)
2415 {
2416         int cpu = (long)v;
2417
2418         val &= ~CPU_TASKS_FROZEN;
2419         switch (val) {
2420         case CPU_DYING:
2421                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2422                        cpu);
2423                 hardware_disable(NULL);
2424                 break;
2425         case CPU_UP_CANCELED:
2426                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2427                        cpu);
2428                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2429                 break;
2430         case CPU_ONLINE:
2431                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2432                        cpu);
2433                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2434                 break;
2435         }
2436         return NOTIFY_OK;
2437 }
2438
2439
2440 asmlinkage void kvm_handle_fault_on_reboot(void)
2441 {
2442         if (kvm_rebooting)
2443                 /* spin while reset goes on */
2444                 while (true)
2445                         ;
2446         /* Fault while not rebooting.  We want the trace. */
2447         BUG();
2448 }
2449 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2450
2451 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2452                       void *v)
2453 {
2454         /*
2455          * Some (well, at least mine) BIOSes hang on reboot if
2456          * in vmx root mode.
2457          *
2458          * And Intel TXT required VMX off for all cpu when system shutdown.
2459          */
2460         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2461         kvm_rebooting = true;
2462         on_each_cpu(hardware_disable, NULL, 1);
2463         return NOTIFY_OK;
2464 }
2465
2466 static struct notifier_block kvm_reboot_notifier = {
2467         .notifier_call = kvm_reboot,
2468         .priority = 0,
2469 };
2470
2471 void kvm_io_bus_init(struct kvm_io_bus *bus)
2472 {
2473         memset(bus, 0, sizeof(*bus));
2474 }
2475
2476 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2477 {
2478         int i;
2479
2480         for (i = 0; i < bus->dev_count; i++) {
2481                 struct kvm_io_device *pos = bus->devs[i];
2482
2483                 kvm_iodevice_destructor(pos);
2484         }
2485 }
2486
2487 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2488                                           gpa_t addr, int len, int is_write)
2489 {
2490         int i;
2491
2492         for (i = 0; i < bus->dev_count; i++) {
2493                 struct kvm_io_device *pos = bus->devs[i];
2494
2495                 if (kvm_iodevice_in_range(pos, addr, len, is_write))
2496                         return pos;
2497         }
2498
2499         return NULL;
2500 }
2501
2502 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2503 {
2504         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2505
2506         bus->devs[bus->dev_count++] = dev;
2507 }
2508
2509 static struct notifier_block kvm_cpu_notifier = {
2510         .notifier_call = kvm_cpu_hotplug,
2511         .priority = 20, /* must be > scheduler priority */
2512 };
2513
2514 static int vm_stat_get(void *_offset, u64 *val)
2515 {
2516         unsigned offset = (long)_offset;
2517         struct kvm *kvm;
2518
2519         *val = 0;
2520         spin_lock(&kvm_lock);
2521         list_for_each_entry(kvm, &vm_list, vm_list)
2522                 *val += *(u32 *)((void *)kvm + offset);
2523         spin_unlock(&kvm_lock);
2524         return 0;
2525 }
2526
2527 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2528
2529 static int vcpu_stat_get(void *_offset, u64 *val)
2530 {
2531         unsigned offset = (long)_offset;
2532         struct kvm *kvm;
2533         struct kvm_vcpu *vcpu;
2534         int i;
2535
2536         *val = 0;
2537         spin_lock(&kvm_lock);
2538         list_for_each_entry(kvm, &vm_list, vm_list)
2539                 kvm_for_each_vcpu(i, vcpu, kvm)
2540                         *val += *(u32 *)((void *)vcpu + offset);
2541
2542         spin_unlock(&kvm_lock);
2543         return 0;
2544 }
2545
2546 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2547
2548 static struct file_operations *stat_fops[] = {
2549         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2550         [KVM_STAT_VM]   = &vm_stat_fops,
2551 };
2552
2553 static void kvm_init_debug(void)
2554 {
2555         struct kvm_stats_debugfs_item *p;
2556
2557         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2558         for (p = debugfs_entries; p->name; ++p)
2559                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2560                                                 (void *)(long)p->offset,
2561                                                 stat_fops[p->kind]);
2562 }
2563
2564 static void kvm_exit_debug(void)
2565 {
2566         struct kvm_stats_debugfs_item *p;
2567
2568         for (p = debugfs_entries; p->name; ++p)
2569                 debugfs_remove(p->dentry);
2570         debugfs_remove(kvm_debugfs_dir);
2571 }
2572
2573 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2574 {
2575         hardware_disable(NULL);
2576         return 0;
2577 }
2578
2579 static int kvm_resume(struct sys_device *dev)
2580 {
2581         hardware_enable(NULL);
2582         return 0;
2583 }
2584
2585 static struct sysdev_class kvm_sysdev_class = {
2586         .name = "kvm",
2587         .suspend = kvm_suspend,
2588         .resume = kvm_resume,
2589 };
2590
2591 static struct sys_device kvm_sysdev = {
2592         .id = 0,
2593         .cls = &kvm_sysdev_class,
2594 };
2595
2596 struct page *bad_page;
2597 pfn_t bad_pfn;
2598
2599 static inline
2600 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2601 {
2602         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2603 }
2604
2605 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2606 {
2607         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2608
2609         kvm_arch_vcpu_load(vcpu, cpu);
2610 }
2611
2612 static void kvm_sched_out(struct preempt_notifier *pn,
2613                           struct task_struct *next)
2614 {
2615         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2616
2617         kvm_arch_vcpu_put(vcpu);
2618 }
2619
2620 int kvm_init(void *opaque, unsigned int vcpu_size,
2621                   struct module *module)
2622 {
2623         int r;
2624         int cpu;
2625
2626         kvm_init_debug();
2627
2628         r = kvm_arch_init(opaque);
2629         if (r)
2630                 goto out_fail;
2631
2632         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2633
2634         if (bad_page == NULL) {
2635                 r = -ENOMEM;
2636                 goto out;
2637         }
2638
2639         bad_pfn = page_to_pfn(bad_page);
2640
2641         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2642                 r = -ENOMEM;
2643                 goto out_free_0;
2644         }
2645
2646         r = kvm_arch_hardware_setup();
2647         if (r < 0)
2648                 goto out_free_0a;
2649
2650         for_each_online_cpu(cpu) {
2651                 smp_call_function_single(cpu,
2652                                 kvm_arch_check_processor_compat,
2653                                 &r, 1);
2654                 if (r < 0)
2655                         goto out_free_1;
2656         }
2657
2658         on_each_cpu(hardware_enable, NULL, 1);
2659         r = register_cpu_notifier(&kvm_cpu_notifier);
2660         if (r)
2661                 goto out_free_2;
2662         register_reboot_notifier(&kvm_reboot_notifier);
2663
2664         r = sysdev_class_register(&kvm_sysdev_class);
2665         if (r)
2666                 goto out_free_3;
2667
2668         r = sysdev_register(&kvm_sysdev);
2669         if (r)
2670                 goto out_free_4;
2671
2672         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2673         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2674                                            __alignof__(struct kvm_vcpu),
2675                                            0, NULL);
2676         if (!kvm_vcpu_cache) {
2677                 r = -ENOMEM;
2678                 goto out_free_5;
2679         }
2680
2681         kvm_chardev_ops.owner = module;
2682         kvm_vm_fops.owner = module;
2683         kvm_vcpu_fops.owner = module;
2684
2685         r = misc_register(&kvm_dev);
2686         if (r) {
2687                 printk(KERN_ERR "kvm: misc device register failed\n");
2688                 goto out_free;
2689         }
2690
2691         kvm_preempt_ops.sched_in = kvm_sched_in;
2692         kvm_preempt_ops.sched_out = kvm_sched_out;
2693
2694         return 0;
2695
2696 out_free:
2697         kmem_cache_destroy(kvm_vcpu_cache);
2698 out_free_5:
2699         sysdev_unregister(&kvm_sysdev);
2700 out_free_4:
2701         sysdev_class_unregister(&kvm_sysdev_class);
2702 out_free_3:
2703         unregister_reboot_notifier(&kvm_reboot_notifier);
2704         unregister_cpu_notifier(&kvm_cpu_notifier);
2705 out_free_2:
2706         on_each_cpu(hardware_disable, NULL, 1);
2707 out_free_1:
2708         kvm_arch_hardware_unsetup();
2709 out_free_0a:
2710         free_cpumask_var(cpus_hardware_enabled);
2711 out_free_0:
2712         __free_page(bad_page);
2713 out:
2714         kvm_arch_exit();
2715         kvm_exit_debug();
2716 out_fail:
2717         return r;
2718 }
2719 EXPORT_SYMBOL_GPL(kvm_init);
2720
2721 void kvm_exit(void)
2722 {
2723         kvm_trace_cleanup();
2724         tracepoint_synchronize_unregister();
2725         misc_deregister(&kvm_dev);
2726         kmem_cache_destroy(kvm_vcpu_cache);
2727         sysdev_unregister(&kvm_sysdev);
2728         sysdev_class_unregister(&kvm_sysdev_class);
2729         unregister_reboot_notifier(&kvm_reboot_notifier);
2730         unregister_cpu_notifier(&kvm_cpu_notifier);
2731         on_each_cpu(hardware_disable, NULL, 1);
2732         kvm_arch_hardware_unsetup();
2733         kvm_arch_exit();
2734         kvm_exit_debug();
2735         free_cpumask_var(cpus_hardware_enabled);
2736         __free_page(bad_page);
2737 }
2738 EXPORT_SYMBOL_GPL(kvm_exit);