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->slots_lock --> 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         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 742
 743         spin_lock(&kvm->requests_lock);
 744         me = smp_processor_id();
 745         kvm_for_each_vcpu(i, vcpu, kvm) {
 746                 if (test_and_set_bit(req, &vcpu->requests))
 747                         continue;
 748                 cpu = vcpu->cpu;
 749                 if (cpus != NULL && cpu != -1 && cpu != me)
 750                         cpumask_set_cpu(cpu, cpus);
 751         }
 752         if (unlikely(cpus == NULL))
 753                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 754         else if (!cpumask_empty(cpus))
 755                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 756         else
 757                 called = false;
 758         spin_unlock(&kvm->requests_lock);
 759         free_cpumask_var(cpus);
 760         return called;
 761 }
 762
 763 void kvm_flush_remote_tlbs(struct kvm *kvm)
 764 {
 765         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 766                 ++kvm->stat.remote_tlb_flush;
 767 }
 768
 769 void kvm_reload_remote_mmus(struct kvm *kvm)
 770 {
 771         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 772 }
 773
 774 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 775 {
 776         struct page *page;
 777         int r;
 778
 779         mutex_init(&vcpu->mutex);
 780         vcpu->cpu = -1;
 781         vcpu->kvm = kvm;
 782         vcpu->vcpu_id = id;
 783         init_waitqueue_head(&vcpu->wq);
 784
 785         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 786         if (!page) {
 787                 r = -ENOMEM;
 788                 goto fail;
 789         }
 790         vcpu->run = page_address(page);
 791
 792         r = kvm_arch_vcpu_init(vcpu);
 793         if (r < 0)
 794                 goto fail_free_run;
 795         return 0;
 796
 797 fail_free_run:
 798         free_page((unsigned long)vcpu->run);
 799 fail:
 800         return r;
 801 }
 802 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 803
 804 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 805 {
 806         kvm_arch_vcpu_uninit(vcpu);
 807         free_page((unsigned long)vcpu->run);
 808 }
 809 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 810
 811 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 812 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 813 {
 814         return container_of(mn, struct kvm, mmu_notifier);
 815 }
 816
 817 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 818                                              struct mm_struct *mm,
 819                                              unsigned long address)
 820 {
 821         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 822         int need_tlb_flush;
 823
 824         /*
 825          * When ->invalidate_page runs, the linux pte has been zapped
 826          * already but the page is still allocated until
 827          * ->invalidate_page returns. So if we increase the sequence
 828          * here the kvm page fault will notice if the spte can't be
 829          * established because the page is going to be freed. If
 830          * instead the kvm page fault establishes the spte before
 831          * ->invalidate_page runs, kvm_unmap_hva will release it
 832          * before returning.
 833          *
 834          * The sequence increase only need to be seen at spin_unlock
 835          * time, and not at spin_lock time.
 836          *
 837          * Increasing the sequence after the spin_unlock would be
 838          * unsafe because the kvm page fault could then establish the
 839          * pte after kvm_unmap_hva returned, without noticing the page
 840          * is going to be freed.
 841          */
 842         spin_lock(&kvm->mmu_lock);
 843         kvm->mmu_notifier_seq++;
 844         need_tlb_flush = kvm_unmap_hva(kvm, address);
 845         spin_unlock(&kvm->mmu_lock);
 846
 847         /* we've to flush the tlb before the pages can be freed */
 848         if (need_tlb_flush)
 849                 kvm_flush_remote_tlbs(kvm);
 850
 851 }
 852
 853 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 854                                         struct mm_struct *mm,
 855                                         unsigned long address,
 856                                         pte_t pte)
 857 {
 858         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 859
 860         spin_lock(&kvm->mmu_lock);
 861         kvm->mmu_notifier_seq++;
 862         kvm_set_spte_hva(kvm, address, pte);
 863         spin_unlock(&kvm->mmu_lock);
 864 }
 865
 866 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 867                                                     struct mm_struct *mm,
 868                                                     unsigned long start,
 869                                                     unsigned long end)
 870 {
 871         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 872         int need_tlb_flush = 0;
 873
 874         spin_lock(&kvm->mmu_lock);
 875         /*
 876          * The count increase must become visible at unlock time as no
 877          * spte can be established without taking the mmu_lock and
 878          * count is also read inside the mmu_lock critical section.
 879          */
 880         kvm->mmu_notifier_count++;
 881         for (; start < end; start += PAGE_SIZE)
 882                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 883         spin_unlock(&kvm->mmu_lock);
 884
 885         /* we've to flush the tlb before the pages can be freed */
 886         if (need_tlb_flush)
 887                 kvm_flush_remote_tlbs(kvm);
 888 }
 889
 890 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 891                                                   struct mm_struct *mm,
 892                                                   unsigned long start,
 893                                                   unsigned long end)
 894 {
 895         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 896
 897         spin_lock(&kvm->mmu_lock);
 898         /*
 899          * This sequence increase will notify the kvm page fault that
 900          * the page that is going to be mapped in the spte could have
 901          * been freed.
 902          */
 903         kvm->mmu_notifier_seq++;
 904         /*
 905          * The above sequence increase must be visible before the
 906          * below count decrease but both values are read by the kvm
 907          * page fault under mmu_lock spinlock so we don't need to add
 908          * a smb_wmb() here in between the two.
 909          */
 910         kvm->mmu_notifier_count--;
 911         spin_unlock(&kvm->mmu_lock);
 912
 913         BUG_ON(kvm->mmu_notifier_count < 0);
 914 }
 915
 916 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 917                                               struct mm_struct *mm,
 918                                               unsigned long address)
 919 {
 920         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 921         int young;
 922
 923         spin_lock(&kvm->mmu_lock);
 924         young = kvm_age_hva(kvm, address);
 925         spin_unlock(&kvm->mmu_lock);
 926
 927         if (young)
 928                 kvm_flush_remote_tlbs(kvm);
 929
 930         return young;
 931 }
 932
 933 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 934                                      struct mm_struct *mm)
 935 {
 936         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 937         kvm_arch_flush_shadow(kvm);
 938 }
 939
 940 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 941         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 942         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 943         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 944         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 945         .change_pte             = kvm_mmu_notifier_change_pte,
 946         .release                = kvm_mmu_notifier_release,
 947 };
 948 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 949
 950 static struct kvm *kvm_create_vm(void)
 951 {
 952         struct kvm *kvm = kvm_arch_create_vm();
 953 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 954         struct page *page;
 955 #endif
 956
 957         if (IS_ERR(kvm))
 958                 goto out;
 959 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 960         INIT_LIST_HEAD(&kvm->irq_routing);
 961         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 962 #endif
 963
 964 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 965         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 966         if (!page) {
 967                 kfree(kvm);
 968                 return ERR_PTR(-ENOMEM);
 969         }
 970         kvm->coalesced_mmio_ring =
 971                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 972 #endif
 973
 974 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 975         {
 976                 int err;
 977                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 978                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 979                 if (err) {
 980 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 981                         put_page(page);
 982 #endif
 983                         kfree(kvm);
 984                         return ERR_PTR(err);
 985                 }
 986         }
 987 #endif
 988
 989         kvm->mm = current->mm;
 990         atomic_inc(&kvm->mm->mm_count);
 991         spin_lock_init(&kvm->mmu_lock);
 992         spin_lock_init(&kvm->requests_lock);
 993         kvm_io_bus_init(&kvm->pio_bus);
 994         kvm_eventfd_init(kvm);
 995         mutex_init(&kvm->lock);
 996         mutex_init(&kvm->irq_lock);
 997         kvm_io_bus_init(&kvm->mmio_bus);
 998         init_rwsem(&kvm->slots_lock);
 999         atomic_set(&kvm->users_count, 1);
1000         spin_lock(&kvm_lock);
1001         list_add(&kvm->vm_list, &vm_list);
1002         spin_unlock(&kvm_lock);
1003 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1004         kvm_coalesced_mmio_init(kvm);
1005 #endif
1006 out:
1007         return kvm;
1008 }
1009
1010 /*
1011  * Free any memory in @free but not in @dont.
1012  */
1013 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
1014                                   struct kvm_memory_slot *dont)
1015 {
1016         int i;
1017
1018         if (!dont || free->rmap != dont->rmap)
1019                 vfree(free->rmap);
1020
1021         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
1022                 vfree(free->dirty_bitmap);
1023
1024
1025         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
1026                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
1027                         vfree(free->lpage_info[i]);
1028                         free->lpage_info[i] = NULL;
1029                 }
1030         }
1031
1032         free->npages = 0;
1033         free->dirty_bitmap = NULL;
1034         free->rmap = NULL;
1035 }
1036
1037 void kvm_free_physmem(struct kvm *kvm)
1038 {
1039         int i;
1040
1041         for (i = 0; i < kvm->nmemslots; ++i)
1042                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
1043 }
1044
1045 static void kvm_destroy_vm(struct kvm *kvm)
1046 {
1047         struct mm_struct *mm = kvm->mm;
1048
1049         kvm_arch_sync_events(kvm);
1050         spin_lock(&kvm_lock);
1051         list_del(&kvm->vm_list);
1052         spin_unlock(&kvm_lock);
1053         kvm_free_irq_routing(kvm);
1054         kvm_io_bus_destroy(&kvm->pio_bus);
1055         kvm_io_bus_destroy(&kvm->mmio_bus);
1056 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1057         if (kvm->coalesced_mmio_ring != NULL)
1058                 free_page((unsigned long)kvm->coalesced_mmio_ring);
1059 #endif
1060 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1061         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1062 #else
1063         kvm_arch_flush_shadow(kvm);
1064 #endif
1065         kvm_arch_destroy_vm(kvm);
1066         mmdrop(mm);
1067 }
1068
1069 void kvm_get_kvm(struct kvm *kvm)
1070 {
1071         atomic_inc(&kvm->users_count);
1072 }
1073 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1074
1075 void kvm_put_kvm(struct kvm *kvm)
1076 {
1077         if (atomic_dec_and_test(&kvm->users_count))
1078                 kvm_destroy_vm(kvm);
1079 }
1080 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1081
1082
1083 static int kvm_vm_release(struct inode *inode, struct file *filp)
1084 {
1085         struct kvm *kvm = filp->private_data;
1086
1087         kvm_irqfd_release(kvm);
1088
1089         kvm_put_kvm(kvm);
1090         return 0;
1091 }
1092
1093 /*
1094  * Allocate some memory and give it an address in the guest physical address
1095  * space.
1096  *
1097  * Discontiguous memory is allowed, mostly for framebuffers.
1098  *
1099  * Must be called holding mmap_sem for write.
1100  */
1101 int __kvm_set_memory_region(struct kvm *kvm,
1102                             struct kvm_userspace_memory_region *mem,
1103                             int user_alloc)
1104 {
1105         int r;
1106         gfn_t base_gfn;
1107         unsigned long npages;
1108         unsigned long i;
1109         struct kvm_memory_slot *memslot;
1110         struct kvm_memory_slot old, new;
1111
1112         r = -EINVAL;
1113         /* General sanity checks */
1114         if (mem->memory_size & (PAGE_SIZE - 1))
1115                 goto out;
1116         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1117                 goto out;
1118         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
1119                 goto out;
1120         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
1121                 goto out;
1122         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1123                 goto out;
1124
1125         memslot = &kvm->memslots[mem->slot];
1126         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1127         npages = mem->memory_size >> PAGE_SHIFT;
1128
1129         if (!npages)
1130                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1131
1132         new = old = *memslot;
1133
1134         new.base_gfn = base_gfn;
1135         new.npages = npages;
1136         new.flags = mem->flags;
1137
1138         /* Disallow changing a memory slot's size. */
1139         r = -EINVAL;
1140         if (npages && old.npages && npages != old.npages)
1141                 goto out_free;
1142
1143         /* Check for overlaps */
1144         r = -EEXIST;
1145         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1146                 struct kvm_memory_slot *s = &kvm->memslots[i];
1147
1148                 if (s == memslot || !s->npages)
1149                         continue;
1150                 if (!((base_gfn + npages <= s->base_gfn) ||
1151                       (base_gfn >= s->base_gfn + s->npages)))
1152                         goto out_free;
1153         }
1154
1155         /* Free page dirty bitmap if unneeded */
1156         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1157                 new.dirty_bitmap = NULL;
1158
1159         r = -ENOMEM;
1160
1161         /* Allocate if a slot is being created */
1162 #ifndef CONFIG_S390
1163         if (npages && !new.rmap) {
1164                 new.rmap = vmalloc(npages * sizeof(struct page *));
1165
1166                 if (!new.rmap)
1167                         goto out_free;
1168
1169                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1170
1171                 new.user_alloc = user_alloc;
1172                 /*
1173                  * hva_to_rmmap() serialzies with the mmu_lock and to be
1174                  * safe it has to ignore memslots with !user_alloc &&
1175                  * !userspace_addr.
1176                  */
1177                 if (user_alloc)
1178                         new.userspace_addr = mem->userspace_addr;
1179                 else
1180                         new.userspace_addr = 0;
1181         }
1182         if (!npages)
1183                 goto skip_lpage;
1184
1185         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
1186                 unsigned long ugfn;
1187                 unsigned long j;
1188                 int lpages;
1189                 int level = i + 2;
1190
1191                 /* Avoid unused variable warning if no large pages */
1192                 (void)level;
1193
1194                 if (new.lpage_info[i])
1195                         continue;
1196
1197                 lpages = 1 + (base_gfn + npages - 1) /
1198                              KVM_PAGES_PER_HPAGE(level);
1199                 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
1200
1201                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
1202
1203                 if (!new.lpage_info[i])
1204                         goto out_free;
1205
1206                 memset(new.lpage_info[i], 0,
1207                        lpages * sizeof(*new.lpage_info[i]));
1208
1209                 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
1210                         new.lpage_info[i][0].write_count = 1;
1211                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
1212                         new.lpage_info[i][lpages - 1].write_count = 1;
1213                 ugfn = new.userspace_addr >> PAGE_SHIFT;
1214                 /*
1215                  * If the gfn and userspace address are not aligned wrt each
1216                  * other, or if explicitly asked to, disable large page
1217                  * support for this slot
1218                  */
1219                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
1220                     !largepages_enabled)
1221                         for (j = 0; j < lpages; ++j)
1222                                 new.lpage_info[i][j].write_count = 1;
1223         }
1224
1225 skip_lpage:
1226
1227         /* Allocate page dirty bitmap if needed */
1228         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1229                 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(&new);
1230
1231                 new.dirty_bitmap = vmalloc(dirty_bytes);
1232                 if (!new.dirty_bitmap)
1233                         goto out_free;
1234                 memset(new.dirty_bitmap, 0, dirty_bytes);
1235                 if (old.npages)
1236                         kvm_arch_flush_shadow(kvm);
1237         }
1238 #else  /* not defined CONFIG_S390 */
1239         new.user_alloc = user_alloc;
1240         if (user_alloc)
1241                 new.userspace_addr = mem->userspace_addr;
1242 #endif /* not defined CONFIG_S390 */
1243
1244         if (!npages)
1245                 kvm_arch_flush_shadow(kvm);
1246
1247         spin_lock(&kvm->mmu_lock);
1248         if (mem->slot >= kvm->nmemslots)
1249                 kvm->nmemslots = mem->slot + 1;
1250
1251         *memslot = new;
1252         spin_unlock(&kvm->mmu_lock);
1253
1254         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1255         if (r) {
1256                 spin_lock(&kvm->mmu_lock);
1257                 *memslot = old;
1258                 spin_unlock(&kvm->mmu_lock);
1259                 goto out_free;
1260         }
1261
1262         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1263         /* Slot deletion case: we have to update the current slot */
1264         spin_lock(&kvm->mmu_lock);
1265         if (!npages)
1266                 *memslot = old;
1267         spin_unlock(&kvm->mmu_lock);
1268 #ifdef CONFIG_DMAR
1269         /* map the pages in iommu page table */
1270         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1271         if (r)
1272                 goto out;
1273 #endif
1274         return 0;
1275
1276 out_free:
1277         kvm_free_physmem_slot(&new, &old);
1278 out:
1279         return r;
1280
1281 }
1282 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1283
1284 int kvm_set_memory_region(struct kvm *kvm,
1285                           struct kvm_userspace_memory_region *mem,
1286                           int user_alloc)
1287 {
1288         int r;
1289
1290         down_write(&kvm->slots_lock);
1291         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1292         up_write(&kvm->slots_lock);
1293         return r;
1294 }
1295 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1296
1297 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1298                                    struct
1299                                    kvm_userspace_memory_region *mem,
1300                                    int user_alloc)
1301 {
1302         if (mem->slot >= KVM_MEMORY_SLOTS)
1303                 return -EINVAL;
1304         return kvm_set_memory_region(kvm, mem, user_alloc);
1305 }
1306
1307 int kvm_get_dirty_log(struct kvm *kvm,
1308                         struct kvm_dirty_log *log, int *is_dirty)
1309 {
1310         struct kvm_memory_slot *memslot;
1311         int r, i;
1312         unsigned long n;
1313         unsigned long any = 0;
1314
1315         r = -EINVAL;
1316         if (log->slot >= KVM_MEMORY_SLOTS)
1317                 goto out;
1318
1319         memslot = &kvm->memslots[log->slot];
1320         r = -ENOENT;
1321         if (!memslot->dirty_bitmap)
1322                 goto out;
1323
1324         n = kvm_dirty_bitmap_bytes(memslot);
1325
1326         for (i = 0; !any && i < n/sizeof(long); ++i)
1327                 any = memslot->dirty_bitmap[i];
1328
1329         r = -EFAULT;
1330         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1331                 goto out;
1332
1333         if (any)
1334                 *is_dirty = 1;
1335
1336         r = 0;
1337 out:
1338         return r;
1339 }
1340
1341 void kvm_disable_largepages(void)
1342 {
1343         largepages_enabled = false;
1344 }
1345 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1346
1347 int is_error_page(struct page *page)
1348 {
1349         return page == bad_page;
1350 }
1351 EXPORT_SYMBOL_GPL(is_error_page);
1352
1353 int is_error_pfn(pfn_t pfn)
1354 {
1355         return pfn == bad_pfn;
1356 }
1357 EXPORT_SYMBOL_GPL(is_error_pfn);
1358
1359 static inline unsigned long bad_hva(void)
1360 {
1361         return PAGE_OFFSET;
1362 }
1363
1364 int kvm_is_error_hva(unsigned long addr)
1365 {
1366         return addr == bad_hva();
1367 }
1368 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1369
1370 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1371 {
1372         int i;
1373
1374         for (i = 0; i < kvm->nmemslots; ++i) {
1375                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1376
1377                 if (gfn >= memslot->base_gfn
1378                     && gfn < memslot->base_gfn + memslot->npages)
1379                         return memslot;
1380         }
1381         return NULL;
1382 }
1383 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1384
1385 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1386 {
1387         gfn = unalias_gfn(kvm, gfn);
1388         return gfn_to_memslot_unaliased(kvm, gfn);
1389 }
1390
1391 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1392 {
1393         int i;
1394
1395         gfn = unalias_gfn(kvm, gfn);
1396         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1397                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1398
1399                 if (gfn >= memslot->base_gfn
1400                     && gfn < memslot->base_gfn + memslot->npages)
1401                         return 1;
1402         }
1403         return 0;
1404 }
1405 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1406
1407 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1408 {
1409         struct kvm_memory_slot *slot;
1410
1411         gfn = unalias_gfn(kvm, gfn);
1412         slot = gfn_to_memslot_unaliased(kvm, gfn);
1413         if (!slot)
1414                 return bad_hva();
1415         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1416 }
1417 EXPORT_SYMBOL_GPL(gfn_to_hva);
1418
1419 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1420 {
1421         struct page *page[1];
1422         unsigned long addr;
1423         int npages;
1424         pfn_t pfn;
1425
1426         might_sleep();
1427
1428         addr = gfn_to_hva(kvm, gfn);
1429         if (kvm_is_error_hva(addr)) {
1430                 get_page(bad_page);
1431                 return page_to_pfn(bad_page);
1432         }
1433
1434         npages = get_user_pages_fast(addr, 1, 1, page);
1435
1436         if (unlikely(npages != 1)) {
1437                 struct vm_area_struct *vma;
1438
1439                 down_read(&current->mm->mmap_sem);
1440                 vma = find_vma(current->mm, addr);
1441
1442                 if (vma == NULL || addr < vma->vm_start ||
1443                     !(vma->vm_flags & VM_PFNMAP)) {
1444                         up_read(&current->mm->mmap_sem);
1445                         get_page(bad_page);
1446                         return page_to_pfn(bad_page);
1447                 }
1448
1449                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1450                 up_read(&current->mm->mmap_sem);
1451                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1452         } else
1453                 pfn = page_to_pfn(page[0]);
1454
1455         return pfn;
1456 }
1457
1458 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1459
1460 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1461 {
1462         pfn_t pfn;
1463
1464         pfn = gfn_to_pfn(kvm, gfn);
1465         if (!kvm_is_mmio_pfn(pfn))
1466                 return pfn_to_page(pfn);
1467
1468         WARN_ON(kvm_is_mmio_pfn(pfn));
1469
1470         get_page(bad_page);
1471         return bad_page;
1472 }
1473
1474 EXPORT_SYMBOL_GPL(gfn_to_page);
1475
1476 void kvm_release_page_clean(struct page *page)
1477 {
1478         kvm_release_pfn_clean(page_to_pfn(page));
1479 }
1480 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1481
1482 void kvm_release_pfn_clean(pfn_t pfn)
1483 {
1484         if (!kvm_is_mmio_pfn(pfn))
1485                 put_page(pfn_to_page(pfn));
1486 }
1487 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1488
1489 void kvm_release_page_dirty(struct page *page)
1490 {
1491         kvm_release_pfn_dirty(page_to_pfn(page));
1492 }
1493 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1494
1495 void kvm_release_pfn_dirty(pfn_t pfn)
1496 {
1497         kvm_set_pfn_dirty(pfn);
1498         kvm_release_pfn_clean(pfn);
1499 }
1500 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1501
1502 void kvm_set_page_dirty(struct page *page)
1503 {
1504         kvm_set_pfn_dirty(page_to_pfn(page));
1505 }
1506 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1507
1508 void kvm_set_pfn_dirty(pfn_t pfn)
1509 {
1510         if (!kvm_is_mmio_pfn(pfn)) {
1511                 struct page *page = pfn_to_page(pfn);
1512                 if (!PageReserved(page))
1513                         SetPageDirty(page);
1514         }
1515 }
1516 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1517
1518 void kvm_set_pfn_accessed(pfn_t pfn)
1519 {
1520         if (!kvm_is_mmio_pfn(pfn))
1521                 mark_page_accessed(pfn_to_page(pfn));
1522 }
1523 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1524
1525 void kvm_get_pfn(pfn_t pfn)
1526 {
1527         if (!kvm_is_mmio_pfn(pfn))
1528                 get_page(pfn_to_page(pfn));
1529 }
1530 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1531
1532 static int next_segment(unsigned long len, int offset)
1533 {
1534         if (len > PAGE_SIZE - offset)
1535                 return PAGE_SIZE - offset;
1536         else
1537                 return len;
1538 }
1539
1540 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1541                         int len)
1542 {
1543         int r;
1544         unsigned long addr;
1545
1546         addr = gfn_to_hva(kvm, gfn);
1547         if (kvm_is_error_hva(addr))
1548                 return -EFAULT;
1549         r = copy_from_user(data, (void __user *)addr + offset, len);
1550         if (r)
1551                 return -EFAULT;
1552         return 0;
1553 }
1554 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1555
1556 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1557 {
1558         gfn_t gfn = gpa >> PAGE_SHIFT;
1559         int seg;
1560         int offset = offset_in_page(gpa);
1561         int ret;
1562
1563         while ((seg = next_segment(len, offset)) != 0) {
1564                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1565                 if (ret < 0)
1566                         return ret;
1567                 offset = 0;
1568                 len -= seg;
1569                 data += seg;
1570                 ++gfn;
1571         }
1572         return 0;
1573 }
1574 EXPORT_SYMBOL_GPL(kvm_read_guest);
1575
1576 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1577                           unsigned long len)
1578 {
1579         int r;
1580         unsigned long addr;
1581         gfn_t gfn = gpa >> PAGE_SHIFT;
1582         int offset = offset_in_page(gpa);
1583
1584         addr = gfn_to_hva(kvm, gfn);
1585         if (kvm_is_error_hva(addr))
1586                 return -EFAULT;
1587         pagefault_disable();
1588         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1589         pagefault_enable();
1590         if (r)
1591                 return -EFAULT;
1592         return 0;
1593 }
1594 EXPORT_SYMBOL(kvm_read_guest_atomic);
1595
1596 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1597                          int offset, int len)
1598 {
1599         int r;
1600         unsigned long addr;
1601
1602         addr = gfn_to_hva(kvm, gfn);
1603         if (kvm_is_error_hva(addr))
1604                 return -EFAULT;
1605         r = copy_to_user((void __user *)addr + offset, data, len);
1606         if (r)
1607                 return -EFAULT;
1608         mark_page_dirty(kvm, gfn);
1609         return 0;
1610 }
1611 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1612
1613 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1614                     unsigned long len)
1615 {
1616         gfn_t gfn = gpa >> PAGE_SHIFT;
1617         int seg;
1618         int offset = offset_in_page(gpa);
1619         int ret;
1620
1621         while ((seg = next_segment(len, offset)) != 0) {
1622                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1623                 if (ret < 0)
1624                         return ret;
1625                 offset = 0;
1626                 len -= seg;
1627                 data += seg;
1628                 ++gfn;
1629         }
1630         return 0;
1631 }
1632
1633 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1634 {
1635         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1636 }
1637 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1638
1639 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1640 {
1641         gfn_t gfn = gpa >> PAGE_SHIFT;
1642         int seg;
1643         int offset = offset_in_page(gpa);
1644         int ret;
1645
1646         while ((seg = next_segment(len, offset)) != 0) {
1647                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1648                 if (ret < 0)
1649                         return ret;
1650                 offset = 0;
1651                 len -= seg;
1652                 ++gfn;
1653         }
1654         return 0;
1655 }
1656 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1657
1658 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1659 {
1660         struct kvm_memory_slot *memslot;
1661
1662         gfn = unalias_gfn(kvm, gfn);
1663         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1664         if (memslot && memslot->dirty_bitmap) {
1665                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1666                 unsigned long *p = memslot->dirty_bitmap +
1667                                         rel_gfn / BITS_PER_LONG;
1668                 int offset = rel_gfn % BITS_PER_LONG;
1669
1670                 /* avoid RMW */
1671                 if (!test_bit(offset, p))
1672                         set_bit(offset, p);
1673         }
1674 }
1675
1676 /*
1677  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1678  */
1679 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1680 {
1681         DEFINE_WAIT(wait);
1682
1683         for (;;) {
1684                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1685
1686                 if (kvm_arch_vcpu_runnable(vcpu)) {
1687                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1688                         break;
1689                 }
1690                 if (kvm_cpu_has_pending_timer(vcpu))
1691                         break;
1692                 if (signal_pending(current))
1693                         break;
1694
1695                 vcpu_put(vcpu);
1696                 schedule();
1697                 vcpu_load(vcpu);
1698         }
1699
1700         finish_wait(&vcpu->wq, &wait);
1701 }
1702
1703 void kvm_resched(struct kvm_vcpu *vcpu)
1704 {
1705         if (!need_resched())
1706                 return;
1707         cond_resched();
1708 }
1709 EXPORT_SYMBOL_GPL(kvm_resched);
1710
1711 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1712 {
1713         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1714         struct page *page;
1715
1716         if (vmf->pgoff == 0)
1717                 page = virt_to_page(vcpu->run);
1718 #ifdef CONFIG_X86
1719         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1720                 page = virt_to_page(vcpu->arch.pio_data);
1721 #endif
1722 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1723         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1724                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1725 #endif
1726         else
1727                 return VM_FAULT_SIGBUS;
1728         get_page(page);
1729         vmf->page = page;
1730         return 0;
1731 }
1732
1733 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1734         .fault = kvm_vcpu_fault,
1735 };
1736
1737 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1738 {
1739         vma->vm_ops = &kvm_vcpu_vm_ops;
1740         return 0;
1741 }
1742
1743 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1744 {
1745         struct kvm_vcpu *vcpu = filp->private_data;
1746
1747         kvm_put_kvm(vcpu->kvm);
1748         return 0;
1749 }
1750
1751 static struct file_operations kvm_vcpu_fops = {
1752         .release        = kvm_vcpu_release,
1753         .unlocked_ioctl = kvm_vcpu_ioctl,
1754         .compat_ioctl   = kvm_vcpu_ioctl,
1755         .mmap           = kvm_vcpu_mmap,
1756 };
1757
1758 /*
1759  * Allocates an inode for the vcpu.
1760  */
1761 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1762 {
1763         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1764 }
1765
1766 /*
1767  * Creates some virtual cpus.  Good luck creating more than one.
1768  */
1769 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1770 {
1771         int r;
1772         struct kvm_vcpu *vcpu, *v;
1773
1774         vcpu = kvm_arch_vcpu_create(kvm, id);
1775         if (IS_ERR(vcpu))
1776                 return PTR_ERR(vcpu);
1777
1778         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1779
1780         r = kvm_arch_vcpu_setup(vcpu);
1781         if (r)
1782                 return r;
1783
1784         mutex_lock(&kvm->lock);
1785         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1786                 r = -EINVAL;
1787                 goto vcpu_destroy;
1788         }
1789
1790         kvm_for_each_vcpu(r, v, kvm)
1791                 if (v->vcpu_id == id) {
1792                         r = -EEXIST;
1793                         goto vcpu_destroy;
1794                 }
1795
1796         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1797
1798         /* Now it's all set up, let userspace reach it */
1799         kvm_get_kvm(kvm);
1800         r = create_vcpu_fd(vcpu);
1801         if (r < 0) {
1802                 kvm_put_kvm(kvm);
1803                 goto vcpu_destroy;
1804         }
1805
1806         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1807         smp_wmb();
1808         atomic_inc(&kvm->online_vcpus);
1809
1810 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1811         if (kvm->bsp_vcpu_id == id)
1812                 kvm->bsp_vcpu = vcpu;
1813 #endif
1814         mutex_unlock(&kvm->lock);
1815         return r;
1816
1817 vcpu_destroy:
1818         mutex_unlock(&kvm->lock);
1819         kvm_arch_vcpu_destroy(vcpu);
1820         return r;
1821 }
1822
1823 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1824 {
1825         if (sigset) {
1826                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1827                 vcpu->sigset_active = 1;
1828                 vcpu->sigset = *sigset;
1829         } else
1830                 vcpu->sigset_active = 0;
1831         return 0;
1832 }
1833
1834 #ifdef __KVM_HAVE_MSIX
1835 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1836                                     struct kvm_assigned_msix_nr *entry_nr)
1837 {
1838         int r = 0;
1839         struct kvm_assigned_dev_kernel *adev;
1840
1841         mutex_lock(&kvm->lock);
1842
1843         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1844                                       entry_nr->assigned_dev_id);
1845         if (!adev) {
1846                 r = -EINVAL;
1847                 goto msix_nr_out;
1848         }
1849
1850         if (adev->entries_nr == 0) {
1851                 adev->entries_nr = entry_nr->entry_nr;
1852                 if (adev->entries_nr == 0 ||
1853                     adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1854                         r = -EINVAL;
1855                         goto msix_nr_out;
1856                 }
1857
1858                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1859                                                 entry_nr->entry_nr,
1860                                                 GFP_KERNEL);
1861                 if (!adev->host_msix_entries) {
1862                         r = -ENOMEM;
1863                         goto msix_nr_out;
1864                 }
1865                 adev->guest_msix_entries = kzalloc(
1866                                 sizeof(struct kvm_guest_msix_entry) *
1867                                 entry_nr->entry_nr, GFP_KERNEL);
1868                 if (!adev->guest_msix_entries) {
1869                         kfree(adev->host_msix_entries);
1870                         r = -ENOMEM;
1871                         goto msix_nr_out;
1872                 }
1873         } else /* Not allowed set MSI-X number twice */
1874                 r = -EINVAL;
1875 msix_nr_out:
1876         mutex_unlock(&kvm->lock);
1877         return r;
1878 }
1879
1880 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1881                                        struct kvm_assigned_msix_entry *entry)
1882 {
1883         int r = 0, i;
1884         struct kvm_assigned_dev_kernel *adev;
1885
1886         mutex_lock(&kvm->lock);
1887
1888         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1889                                       entry->assigned_dev_id);
1890
1891         if (!adev) {
1892                 r = -EINVAL;
1893                 goto msix_entry_out;
1894         }
1895
1896         for (i = 0; i < adev->entries_nr; i++)
1897                 if (adev->guest_msix_entries[i].vector == 0 ||
1898                     adev->guest_msix_entries[i].entry == entry->entry) {
1899                         adev->guest_msix_entries[i].entry = entry->entry;
1900                         adev->guest_msix_entries[i].vector = entry->gsi;
1901                         adev->host_msix_entries[i].entry = entry->entry;
1902                         break;
1903                 }
1904         if (i == adev->entries_nr) {
1905                 r = -ENOSPC;
1906                 goto msix_entry_out;
1907         }
1908
1909 msix_entry_out:
1910         mutex_unlock(&kvm->lock);
1911
1912         return r;
1913 }
1914 #endif
1915
1916 static long kvm_vcpu_ioctl(struct file *filp,
1917                            unsigned int ioctl, unsigned long arg)
1918 {
1919         struct kvm_vcpu *vcpu = filp->private_data;
1920         void __user *argp = (void __user *)arg;
1921         int r;
1922         struct kvm_fpu *fpu = NULL;
1923         struct kvm_sregs *kvm_sregs = NULL;
1924
1925         if (vcpu->kvm->mm != current->mm)
1926                 return -EIO;
1927         switch (ioctl) {
1928         case KVM_RUN:
1929                 r = -EINVAL;
1930                 if (arg)
1931                         goto out;
1932                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1933                 break;
1934         case KVM_GET_REGS: {
1935                 struct kvm_regs *kvm_regs;
1936
1937                 r = -ENOMEM;
1938                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1939                 if (!kvm_regs)
1940                         goto out;
1941                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1942                 if (r)
1943                         goto out_free1;
1944                 r = -EFAULT;
1945                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1946                         goto out_free1;
1947                 r = 0;
1948 out_free1:
1949                 kfree(kvm_regs);
1950                 break;
1951         }
1952         case KVM_SET_REGS: {
1953                 struct kvm_regs *kvm_regs;
1954
1955                 r = -ENOMEM;
1956                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1957                 if (!kvm_regs)
1958                         goto out;
1959                 r = -EFAULT;
1960                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1961                         goto out_free2;
1962                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1963                 if (r)
1964                         goto out_free2;
1965                 r = 0;
1966 out_free2:
1967                 kfree(kvm_regs);
1968                 break;
1969         }
1970         case KVM_GET_SREGS: {
1971                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1972                 r = -ENOMEM;
1973                 if (!kvm_sregs)
1974                         goto out;
1975                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1976                 if (r)
1977                         goto out;
1978                 r = -EFAULT;
1979                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1980                         goto out;
1981                 r = 0;
1982                 break;
1983         }
1984         case KVM_SET_SREGS: {
1985                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1986                 r = -ENOMEM;
1987                 if (!kvm_sregs)
1988                         goto out;
1989                 r = -EFAULT;
1990                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1991                         goto out;
1992                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1993                 if (r)
1994                         goto out;
1995                 r = 0;
1996                 break;
1997         }
1998         case KVM_GET_MP_STATE: {
1999                 struct kvm_mp_state mp_state;
2000
2001                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2002                 if (r)
2003                         goto out;
2004                 r = -EFAULT;
2005                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2006                         goto out;
2007                 r = 0;
2008                 break;
2009         }
2010         case KVM_SET_MP_STATE: {
2011                 struct kvm_mp_state mp_state;
2012
2013                 r = -EFAULT;
2014                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2015                         goto out;
2016                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2017                 if (r)
2018                         goto out;
2019                 r = 0;
2020                 break;
2021         }
2022         case KVM_TRANSLATE: {
2023                 struct kvm_translation tr;
2024
2025                 r = -EFAULT;
2026                 if (copy_from_user(&tr, argp, sizeof tr))
2027                         goto out;
2028                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2029                 if (r)
2030                         goto out;
2031                 r = -EFAULT;
2032                 if (copy_to_user(argp, &tr, sizeof tr))
2033                         goto out;
2034                 r = 0;
2035                 break;
2036         }
2037         case KVM_SET_GUEST_DEBUG: {
2038                 struct kvm_guest_debug dbg;
2039
2040                 r = -EFAULT;
2041                 if (copy_from_user(&dbg, argp, sizeof dbg))
2042                         goto out;
2043                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2044                 if (r)
2045                         goto out;
2046                 r = 0;
2047                 break;
2048         }
2049         case KVM_SET_SIGNAL_MASK: {
2050                 struct kvm_signal_mask __user *sigmask_arg = argp;
2051                 struct kvm_signal_mask kvm_sigmask;
2052                 sigset_t sigset, *p;
2053
2054                 p = NULL;
2055                 if (argp) {
2056                         r = -EFAULT;
2057                         if (copy_from_user(&kvm_sigmask, argp,
2058                                            sizeof kvm_sigmask))
2059                                 goto out;
2060                         r = -EINVAL;
2061                         if (kvm_sigmask.len != sizeof sigset)
2062                                 goto out;
2063                         r = -EFAULT;
2064                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2065                                            sizeof sigset))
2066                                 goto out;
2067                         p = &sigset;
2068                 }
2069                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2070                 break;
2071         }
2072         case KVM_GET_FPU: {
2073                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2074                 r = -ENOMEM;
2075                 if (!fpu)
2076                         goto out;
2077                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2078                 if (r)
2079                         goto out;
2080                 r = -EFAULT;
2081                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2082                         goto out;
2083                 r = 0;
2084                 break;
2085         }
2086         case KVM_SET_FPU: {
2087                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2088                 r = -ENOMEM;
2089                 if (!fpu)
2090                         goto out;
2091                 r = -EFAULT;
2092                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2093                         goto out;
2094                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2095                 if (r)
2096                         goto out;
2097                 r = 0;
2098                 break;
2099         }
2100         default:
2101                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2102         }
2103 out:
2104         kfree(fpu);
2105         kfree(kvm_sregs);
2106         return r;
2107 }
2108
2109 static long kvm_vm_ioctl(struct file *filp,
2110                            unsigned int ioctl, unsigned long arg)
2111 {
2112         struct kvm *kvm = filp->private_data;
2113         void __user *argp = (void __user *)arg;
2114         int r;
2115
2116         if (kvm->mm != current->mm)
2117                 return -EIO;
2118         switch (ioctl) {
2119         case KVM_CREATE_VCPU:
2120                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2121                 if (r < 0)
2122                         goto out;
2123                 break;
2124         case KVM_SET_USER_MEMORY_REGION: {
2125                 struct kvm_userspace_memory_region kvm_userspace_mem;
2126
2127                 r = -EFAULT;
2128                 if (copy_from_user(&kvm_userspace_mem, argp,
2129                                                 sizeof kvm_userspace_mem))
2130                         goto out;
2131
2132                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2133                 if (r)
2134                         goto out;
2135                 break;
2136         }
2137         case KVM_GET_DIRTY_LOG: {
2138                 struct kvm_dirty_log log;
2139
2140                 r = -EFAULT;
2141                 if (copy_from_user(&log, argp, sizeof log))
2142                         goto out;
2143                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2144                 if (r)
2145                         goto out;
2146                 break;
2147         }
2148 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2149         case KVM_REGISTER_COALESCED_MMIO: {
2150                 struct kvm_coalesced_mmio_zone zone;
2151                 r = -EFAULT;
2152                 if (copy_from_user(&zone, argp, sizeof zone))
2153                         goto out;
2154                 r = -ENXIO;
2155                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2156                 if (r)
2157                         goto out;
2158                 r = 0;
2159                 break;
2160         }
2161         case KVM_UNREGISTER_COALESCED_MMIO: {
2162                 struct kvm_coalesced_mmio_zone zone;
2163                 r = -EFAULT;
2164                 if (copy_from_user(&zone, argp, sizeof zone))
2165                         goto out;
2166                 r = -ENXIO;
2167                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2168                 if (r)
2169                         goto out;
2170                 r = 0;
2171                 break;
2172         }
2173 #endif
2174 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2175         case KVM_ASSIGN_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_assign_device(kvm, &assigned_dev);
2182                 if (r)
2183                         goto out;
2184                 break;
2185         }
2186         case KVM_ASSIGN_IRQ: {
2187                 r = -EOPNOTSUPP;
2188                 break;
2189         }
2190 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2191         case KVM_ASSIGN_DEV_IRQ: {
2192                 struct kvm_assigned_irq assigned_irq;
2193
2194                 r = -EFAULT;
2195                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2196                         goto out;
2197                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2198                 if (r)
2199                         goto out;
2200                 break;
2201         }
2202         case KVM_DEASSIGN_DEV_IRQ: {
2203                 struct kvm_assigned_irq assigned_irq;
2204
2205                 r = -EFAULT;
2206                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2207                         goto out;
2208                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2209                 if (r)
2210                         goto out;
2211                 break;
2212         }
2213 #endif
2214 #endif
2215 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2216         case KVM_DEASSIGN_PCI_DEVICE: {
2217                 struct kvm_assigned_pci_dev assigned_dev;
2218
2219                 r = -EFAULT;
2220                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2221                         goto out;
2222                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2223                 if (r)
2224                         goto out;
2225                 break;
2226         }
2227 #endif
2228 #ifdef KVM_CAP_IRQ_ROUTING
2229         case KVM_SET_GSI_ROUTING: {
2230                 struct kvm_irq_routing routing;
2231                 struct kvm_irq_routing __user *urouting;
2232                 struct kvm_irq_routing_entry *entries;
2233
2234                 r = -EFAULT;
2235                 if (copy_from_user(&routing, argp, sizeof(routing)))
2236                         goto out;
2237                 r = -EINVAL;
2238                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2239                         goto out;
2240                 if (routing.flags)
2241                         goto out;
2242                 r = -ENOMEM;
2243                 entries = vmalloc(routing.nr * sizeof(*entries));
2244                 if (!entries)
2245                         goto out;
2246                 r = -EFAULT;
2247                 urouting = argp;
2248                 if (copy_from_user(entries, urouting->entries,
2249                                    routing.nr * sizeof(*entries)))
2250                         goto out_free_irq_routing;
2251                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2252                                         routing.flags);
2253         out_free_irq_routing:
2254                 vfree(entries);
2255                 break;
2256         }
2257 #endif /* KVM_CAP_IRQ_ROUTING */
2258 #ifdef __KVM_HAVE_MSIX
2259         case KVM_ASSIGN_SET_MSIX_NR: {
2260                 struct kvm_assigned_msix_nr entry_nr;
2261                 r = -EFAULT;
2262                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2263                         goto out;
2264                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2265                 if (r)
2266                         goto out;
2267                 break;
2268         }
2269         case KVM_ASSIGN_SET_MSIX_ENTRY: {
2270                 struct kvm_assigned_msix_entry entry;
2271                 r = -EFAULT;
2272                 if (copy_from_user(&entry, argp, sizeof entry))
2273                         goto out;
2274                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2275                 if (r)
2276                         goto out;
2277                 break;
2278         }
2279 #endif
2280         case KVM_IRQFD: {
2281                 struct kvm_irqfd data;
2282
2283                 r = -EFAULT;
2284                 if (copy_from_user(&data, argp, sizeof data))
2285                         goto out;
2286                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2287                 break;
2288         }
2289         case KVM_IOEVENTFD: {
2290                 struct kvm_ioeventfd data;
2291
2292                 r = -EFAULT;
2293                 if (copy_from_user(&data, argp, sizeof data))
2294                         goto out;
2295                 r = kvm_ioeventfd(kvm, &data);
2296                 break;
2297         }
2298 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2299         case KVM_SET_BOOT_CPU_ID:
2300                 r = 0;
2301                 mutex_lock(&kvm->lock);
2302                 if (atomic_read(&kvm->online_vcpus) != 0)
2303                         r = -EBUSY;
2304                 else
2305                         kvm->bsp_vcpu_id = arg;
2306                 mutex_unlock(&kvm->lock);
2307                 break;
2308 #endif
2309         default:
2310                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2311         }
2312 out:
2313         return r;
2314 }
2315
2316 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2317 {
2318         struct page *page[1];
2319         unsigned long addr;
2320         int npages;
2321         gfn_t gfn = vmf->pgoff;
2322         struct kvm *kvm = vma->vm_file->private_data;
2323
2324         addr = gfn_to_hva(kvm, gfn);
2325         if (kvm_is_error_hva(addr))
2326                 return VM_FAULT_SIGBUS;
2327
2328         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2329                                 NULL);
2330         if (unlikely(npages != 1))
2331                 return VM_FAULT_SIGBUS;
2332
2333         vmf->page = page[0];
2334         return 0;
2335 }
2336
2337 static const struct vm_operations_struct kvm_vm_vm_ops = {
2338         .fault = kvm_vm_fault,
2339 };
2340
2341 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2342 {
2343         vma->vm_ops = &kvm_vm_vm_ops;
2344         return 0;
2345 }
2346
2347 static struct file_operations kvm_vm_fops = {
2348         .release        = kvm_vm_release,
2349         .unlocked_ioctl = kvm_vm_ioctl,
2350         .compat_ioctl   = kvm_vm_ioctl,
2351         .mmap           = kvm_vm_mmap,
2352 };
2353
2354 static int kvm_dev_ioctl_create_vm(void)
2355 {
2356         int fd;
2357         struct kvm *kvm;
2358
2359         kvm = kvm_create_vm();
2360         if (IS_ERR(kvm))
2361                 return PTR_ERR(kvm);
2362         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2363         if (fd < 0)
2364                 kvm_put_kvm(kvm);
2365
2366         return fd;
2367 }
2368
2369 static long kvm_dev_ioctl_check_extension_generic(long arg)
2370 {
2371         switch (arg) {
2372         case KVM_CAP_USER_MEMORY:
2373         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2374         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2375 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2376         case KVM_CAP_SET_BOOT_CPU_ID:
2377 #endif
2378                 return 1;
2379 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2380         case KVM_CAP_IRQ_ROUTING:
2381                 return KVM_MAX_IRQ_ROUTES;
2382 #endif
2383         default:
2384                 break;
2385         }
2386         return kvm_dev_ioctl_check_extension(arg);
2387 }
2388
2389 static long kvm_dev_ioctl(struct file *filp,
2390                           unsigned int ioctl, unsigned long arg)
2391 {
2392         long r = -EINVAL;
2393
2394         switch (ioctl) {
2395         case KVM_GET_API_VERSION:
2396                 r = -EINVAL;
2397                 if (arg)
2398                         goto out;
2399                 r = KVM_API_VERSION;
2400                 break;
2401         case KVM_CREATE_VM:
2402                 r = -EINVAL;
2403                 if (arg)
2404                         goto out;
2405                 r = kvm_dev_ioctl_create_vm();
2406                 break;
2407         case KVM_CHECK_EXTENSION:
2408                 r = kvm_dev_ioctl_check_extension_generic(arg);
2409                 break;
2410         case KVM_GET_VCPU_MMAP_SIZE:
2411                 r = -EINVAL;
2412                 if (arg)
2413                         goto out;
2414                 r = PAGE_SIZE;     /* struct kvm_run */
2415 #ifdef CONFIG_X86
2416                 r += PAGE_SIZE;    /* pio data page */
2417 #endif
2418 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2419                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2420 #endif
2421                 break;
2422         case KVM_TRACE_ENABLE:
2423         case KVM_TRACE_PAUSE:
2424         case KVM_TRACE_DISABLE:
2425                 r = -EOPNOTSUPP;
2426                 break;
2427         default:
2428                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2429         }
2430 out:
2431         return r;
2432 }
2433
2434 static struct file_operations kvm_chardev_ops = {
2435         .unlocked_ioctl = kvm_dev_ioctl,
2436         .compat_ioctl   = kvm_dev_ioctl,
2437 };
2438
2439 static struct miscdevice kvm_dev = {
2440         KVM_MINOR,
2441         "kvm",
2442         &kvm_chardev_ops,
2443 };
2444
2445 static void hardware_enable(void *junk)
2446 {
2447         int cpu = raw_smp_processor_id();
2448
2449         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2450                 return;
2451         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2452         kvm_arch_hardware_enable(NULL);
2453 }
2454
2455 static void hardware_disable(void *junk)
2456 {
2457         int cpu = raw_smp_processor_id();
2458
2459         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2460                 return;
2461         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2462         kvm_arch_hardware_disable(NULL);
2463 }
2464
2465 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2466                            void *v)
2467 {
2468         int cpu = (long)v;
2469
2470         val &= ~CPU_TASKS_FROZEN;
2471         switch (val) {
2472         case CPU_DYING:
2473                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2474                        cpu);
2475                 hardware_disable(NULL);
2476                 break;
2477         case CPU_UP_CANCELED:
2478                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2479                        cpu);
2480                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2481                 break;
2482         case CPU_ONLINE:
2483                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2484                        cpu);
2485                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2486                 break;
2487         }
2488         return NOTIFY_OK;
2489 }
2490
2491
2492 asmlinkage void kvm_handle_fault_on_reboot(void)
2493 {
2494         if (kvm_rebooting)
2495                 /* spin while reset goes on */
2496                 while (true)
2497                         ;
2498         /* Fault while not rebooting.  We want the trace. */
2499         BUG();
2500 }
2501 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2502
2503 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2504                       void *v)
2505 {
2506         /*
2507          * Some (well, at least mine) BIOSes hang on reboot if
2508          * in vmx root mode.
2509          *
2510          * And Intel TXT required VMX off for all cpu when system shutdown.
2511          */
2512         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2513         kvm_rebooting = true;
2514         on_each_cpu(hardware_disable, NULL, 1);
2515         return NOTIFY_OK;
2516 }
2517
2518 static struct notifier_block kvm_reboot_notifier = {
2519         .notifier_call = kvm_reboot,
2520         .priority = 0,
2521 };
2522
2523 void kvm_io_bus_init(struct kvm_io_bus *bus)
2524 {
2525         memset(bus, 0, sizeof(*bus));
2526 }
2527
2528 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2529 {
2530         int i;
2531
2532         for (i = 0; i < bus->dev_count; i++) {
2533                 struct kvm_io_device *pos = bus->devs[i];
2534
2535                 kvm_iodevice_destructor(pos);
2536         }
2537 }
2538
2539 /* kvm_io_bus_write - called under kvm->slots_lock */
2540 int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr,
2541                      int len, const void *val)
2542 {
2543         int i;
2544         for (i = 0; i < bus->dev_count; i++)
2545                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2546                         return 0;
2547         return -EOPNOTSUPP;
2548 }
2549
2550 /* kvm_io_bus_read - called under kvm->slots_lock */
2551 int kvm_io_bus_read(struct kvm_io_bus *bus, gpa_t addr, int len, void *val)
2552 {
2553         int i;
2554         for (i = 0; i < bus->dev_count; i++)
2555                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2556                         return 0;
2557         return -EOPNOTSUPP;
2558 }
2559
2560 int kvm_io_bus_register_dev(struct kvm *kvm, struct kvm_io_bus *bus,
2561                              struct kvm_io_device *dev)
2562 {
2563         int ret;
2564
2565         down_write(&kvm->slots_lock);
2566         ret = __kvm_io_bus_register_dev(bus, dev);
2567         up_write(&kvm->slots_lock);
2568
2569         return ret;
2570 }
2571
2572 /* An unlocked version. Caller must have write lock on slots_lock. */
2573 int __kvm_io_bus_register_dev(struct kvm_io_bus *bus,
2574                               struct kvm_io_device *dev)
2575 {
2576         if (bus->dev_count > NR_IOBUS_DEVS-1)
2577                 return -ENOSPC;
2578
2579         bus->devs[bus->dev_count++] = dev;
2580
2581         return 0;
2582 }
2583
2584 void kvm_io_bus_unregister_dev(struct kvm *kvm,
2585                                struct kvm_io_bus *bus,
2586                                struct kvm_io_device *dev)
2587 {
2588         down_write(&kvm->slots_lock);
2589         __kvm_io_bus_unregister_dev(bus, dev);
2590         up_write(&kvm->slots_lock);
2591 }
2592
2593 /* An unlocked version. Caller must have write lock on slots_lock. */
2594 void __kvm_io_bus_unregister_dev(struct kvm_io_bus *bus,
2595                                  struct kvm_io_device *dev)
2596 {
2597         int i;
2598
2599         for (i = 0; i < bus->dev_count; i++)
2600                 if (bus->devs[i] == dev) {
2601                         bus->devs[i] = bus->devs[--bus->dev_count];
2602                         break;
2603                 }
2604 }
2605
2606 static struct notifier_block kvm_cpu_notifier = {
2607         .notifier_call = kvm_cpu_hotplug,
2608         .priority = 20, /* must be > scheduler priority */
2609 };
2610
2611 static int vm_stat_get(void *_offset, u64 *val)
2612 {
2613         unsigned offset = (long)_offset;
2614         struct kvm *kvm;
2615
2616         *val = 0;
2617         spin_lock(&kvm_lock);
2618         list_for_each_entry(kvm, &vm_list, vm_list)
2619                 *val += *(u32 *)((void *)kvm + offset);
2620         spin_unlock(&kvm_lock);
2621         return 0;
2622 }
2623
2624 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2625
2626 static int vcpu_stat_get(void *_offset, u64 *val)
2627 {
2628         unsigned offset = (long)_offset;
2629         struct kvm *kvm;
2630         struct kvm_vcpu *vcpu;
2631         int i;
2632
2633         *val = 0;
2634         spin_lock(&kvm_lock);
2635         list_for_each_entry(kvm, &vm_list, vm_list)
2636                 kvm_for_each_vcpu(i, vcpu, kvm)
2637                         *val += *(u32 *)((void *)vcpu + offset);
2638
2639         spin_unlock(&kvm_lock);
2640         return 0;
2641 }
2642
2643 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2644
2645 static const struct file_operations *stat_fops[] = {
2646         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2647         [KVM_STAT_VM]   = &vm_stat_fops,
2648 };
2649
2650 static void kvm_init_debug(void)
2651 {
2652         struct kvm_stats_debugfs_item *p;
2653
2654         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2655         for (p = debugfs_entries; p->name; ++p)
2656                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2657                                                 (void *)(long)p->offset,
2658                                                 stat_fops[p->kind]);
2659 }
2660
2661 static void kvm_exit_debug(void)
2662 {
2663         struct kvm_stats_debugfs_item *p;
2664
2665         for (p = debugfs_entries; p->name; ++p)
2666                 debugfs_remove(p->dentry);
2667         debugfs_remove(kvm_debugfs_dir);
2668 }
2669
2670 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2671 {
2672         hardware_disable(NULL);
2673         return 0;
2674 }
2675
2676 static int kvm_resume(struct sys_device *dev)
2677 {
2678         hardware_enable(NULL);
2679         return 0;
2680 }
2681
2682 static struct sysdev_class kvm_sysdev_class = {
2683         .name = "kvm",
2684         .suspend = kvm_suspend,
2685         .resume = kvm_resume,
2686 };
2687
2688 static struct sys_device kvm_sysdev = {
2689         .id = 0,
2690         .cls = &kvm_sysdev_class,
2691 };
2692
2693 struct page *bad_page;
2694 pfn_t bad_pfn;
2695
2696 static inline
2697 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2698 {
2699         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2700 }
2701
2702 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2703 {
2704         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2705
2706         kvm_arch_vcpu_load(vcpu, cpu);
2707 }
2708
2709 static void kvm_sched_out(struct preempt_notifier *pn,
2710                           struct task_struct *next)
2711 {
2712         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2713
2714         kvm_arch_vcpu_put(vcpu);
2715 }
2716
2717 int kvm_init(void *opaque, unsigned int vcpu_size,
2718                   struct module *module)
2719 {
2720         int r;
2721         int cpu;
2722
2723         r = kvm_arch_init(opaque);
2724         if (r)
2725                 goto out_fail;
2726
2727         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2728
2729         if (bad_page == NULL) {
2730                 r = -ENOMEM;
2731                 goto out;
2732         }
2733
2734         bad_pfn = page_to_pfn(bad_page);
2735
2736         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2737                 r = -ENOMEM;
2738                 goto out_free_0;
2739         }
2740
2741         r = kvm_arch_hardware_setup();
2742         if (r < 0)
2743                 goto out_free_0a;
2744
2745         for_each_online_cpu(cpu) {
2746                 smp_call_function_single(cpu,
2747                                 kvm_arch_check_processor_compat,
2748                                 &r, 1);
2749                 if (r < 0)
2750                         goto out_free_1;
2751         }
2752
2753         on_each_cpu(hardware_enable, NULL, 1);
2754         r = register_cpu_notifier(&kvm_cpu_notifier);
2755         if (r)
2756                 goto out_free_2;
2757         register_reboot_notifier(&kvm_reboot_notifier);
2758
2759         r = sysdev_class_register(&kvm_sysdev_class);
2760         if (r)
2761                 goto out_free_3;
2762
2763         r = sysdev_register(&kvm_sysdev);
2764         if (r)
2765                 goto out_free_4;
2766
2767         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2768         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2769                                            __alignof__(struct kvm_vcpu),
2770                                            0, NULL);
2771         if (!kvm_vcpu_cache) {
2772                 r = -ENOMEM;
2773                 goto out_free_5;
2774         }
2775
2776         kvm_chardev_ops.owner = module;
2777         kvm_vm_fops.owner = module;
2778         kvm_vcpu_fops.owner = module;
2779
2780         r = misc_register(&kvm_dev);
2781         if (r) {
2782                 printk(KERN_ERR "kvm: misc device register failed\n");
2783                 goto out_free;
2784         }
2785
2786         kvm_preempt_ops.sched_in = kvm_sched_in;
2787         kvm_preempt_ops.sched_out = kvm_sched_out;
2788
2789         kvm_init_debug();
2790
2791         return 0;
2792
2793 out_free:
2794         kmem_cache_destroy(kvm_vcpu_cache);
2795 out_free_5:
2796         sysdev_unregister(&kvm_sysdev);
2797 out_free_4:
2798         sysdev_class_unregister(&kvm_sysdev_class);
2799 out_free_3:
2800         unregister_reboot_notifier(&kvm_reboot_notifier);
2801         unregister_cpu_notifier(&kvm_cpu_notifier);
2802 out_free_2:
2803         on_each_cpu(hardware_disable, NULL, 1);
2804 out_free_1:
2805         kvm_arch_hardware_unsetup();
2806 out_free_0a:
2807         free_cpumask_var(cpus_hardware_enabled);
2808 out_free_0:
2809         __free_page(bad_page);
2810 out:
2811         kvm_arch_exit();
2812 out_fail:
2813         return r;
2814 }
2815 EXPORT_SYMBOL_GPL(kvm_init);
2816
2817 void kvm_exit(void)
2818 {
2819         tracepoint_synchronize_unregister();
2820         kvm_exit_debug();
2821         misc_deregister(&kvm_dev);
2822         kmem_cache_destroy(kvm_vcpu_cache);
2823         sysdev_unregister(&kvm_sysdev);
2824         sysdev_class_unregister(&kvm_sysdev_class);
2825         unregister_reboot_notifier(&kvm_reboot_notifier);
2826         unregister_cpu_notifier(&kvm_cpu_notifier);
2827         on_each_cpu(hardware_disable, NULL, 1);
2828         kvm_arch_hardware_unsetup();
2829         kvm_arch_exit();
2830         free_cpumask_var(cpus_hardware_enabled);
2831         __free_page(bad_page);
2832 }
2833 EXPORT_SYMBOL_GPL(kvm_exit);