virt/kvm/kvm_main.c

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