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

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "iodev.h"
  19
  20 #include <linux/kvm_host.h>
  21 #include <linux/kvm.h>
  22 #include <linux/module.h>
  23 #include <linux/errno.h>
  24 #include <linux/percpu.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/sysdev.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46
  47 #include <asm/processor.h>
  48 #include <asm/io.h>
  49 #include <asm/uaccess.h>
  50 #include <asm/pgtable.h>
  51
  52 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
  53 #include "coalesced_mmio.h"
  54 #endif
  55
  56 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
  57 #include <linux/pci.h>
  58 #include <linux/interrupt.h>
  59 #include "irq.h"
  60 #endif
  61
  62 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 static inline int valid_vcpu(int n)
 693 {
 694         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 695 }
 696
 697 inline int kvm_is_mmio_pfn(pfn_t pfn)
 698 {
 699         if (pfn_valid(pfn)) {
 700                 struct page *page = compound_head(pfn_to_page(pfn));
 701                 return PageReserved(page);
 702         }
 703
 704         return true;
 705 }
 706
 707 /*
 708  * Switches to specified vcpu, until a matching vcpu_put()
 709  */
 710 void vcpu_load(struct kvm_vcpu *vcpu)
 711 {
 712         int cpu;
 713
 714         mutex_lock(&vcpu->mutex);
 715         cpu = get_cpu();
 716         preempt_notifier_register(&vcpu->preempt_notifier);
 717         kvm_arch_vcpu_load(vcpu, cpu);
 718         put_cpu();
 719 }
 720
 721 void vcpu_put(struct kvm_vcpu *vcpu)
 722 {
 723         preempt_disable();
 724         kvm_arch_vcpu_put(vcpu);
 725         preempt_notifier_unregister(&vcpu->preempt_notifier);
 726         preempt_enable();
 727         mutex_unlock(&vcpu->mutex);
 728 }
 729
 730 static void ack_flush(void *_completed)
 731 {
 732 }
 733
 734 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 735 {
 736         int i, cpu, me;
 737         cpumask_var_t cpus;
 738         bool called = true;
 739         struct kvm_vcpu *vcpu;
 740
 741         if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
 742                 cpumask_clear(cpus);
 743
 744         me = get_cpu();
 745         spin_lock(&kvm->requests_lock);
 746         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 747                 vcpu = kvm->vcpus[i];
 748                 if (!vcpu)
 749                         continue;
 750                 if (test_and_set_bit(req, &vcpu->requests))
 751                         continue;
 752                 cpu = vcpu->cpu;
 753                 if (cpus != NULL && cpu != -1 && cpu != me)
 754                         cpumask_set_cpu(cpu, cpus);
 755         }
 756         if (unlikely(cpus == NULL))
 757                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 758         else if (!cpumask_empty(cpus))
 759                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 760         else
 761                 called = false;
 762         spin_unlock(&kvm->requests_lock);
 763         put_cpu();
 764         free_cpumask_var(cpus);
 765         return called;
 766 }
 767
 768 void kvm_flush_remote_tlbs(struct kvm *kvm)
 769 {
 770         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 771                 ++kvm->stat.remote_tlb_flush;
 772 }
 773
 774 void kvm_reload_remote_mmus(struct kvm *kvm)
 775 {
 776         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 777 }
 778
 779 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 780 {
 781         struct page *page;
 782         int r;
 783
 784         mutex_init(&vcpu->mutex);
 785         vcpu->cpu = -1;
 786         vcpu->kvm = kvm;
 787         vcpu->vcpu_id = id;
 788         init_waitqueue_head(&vcpu->wq);
 789
 790         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 791         if (!page) {
 792                 r = -ENOMEM;
 793                 goto fail;
 794         }
 795         vcpu->run = page_address(page);
 796
 797         r = kvm_arch_vcpu_init(vcpu);
 798         if (r < 0)
 799                 goto fail_free_run;
 800         return 0;
 801
 802 fail_free_run:
 803         free_page((unsigned long)vcpu->run);
 804 fail:
 805         return r;
 806 }
 807 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 808
 809 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 810 {
 811         kvm_arch_vcpu_uninit(vcpu);
 812         free_page((unsigned long)vcpu->run);
 813 }
 814 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 815
 816 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 817 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 818 {
 819         return container_of(mn, struct kvm, mmu_notifier);
 820 }
 821
 822 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 823                                              struct mm_struct *mm,
 824                                              unsigned long address)
 825 {
 826         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 827         int need_tlb_flush;
 828
 829         /*
 830          * When ->invalidate_page runs, the linux pte has been zapped
 831          * already but the page is still allocated until
 832          * ->invalidate_page returns. So if we increase the sequence
 833          * here the kvm page fault will notice if the spte can't be
 834          * established because the page is going to be freed. If
 835          * instead the kvm page fault establishes the spte before
 836          * ->invalidate_page runs, kvm_unmap_hva will release it
 837          * before returning.
 838          *
 839          * The sequence increase only need to be seen at spin_unlock
 840          * time, and not at spin_lock time.
 841          *
 842          * Increasing the sequence after the spin_unlock would be
 843          * unsafe because the kvm page fault could then establish the
 844          * pte after kvm_unmap_hva returned, without noticing the page
 845          * is going to be freed.
 846          */
 847         spin_lock(&kvm->mmu_lock);
 848         kvm->mmu_notifier_seq++;
 849         need_tlb_flush = kvm_unmap_hva(kvm, address);
 850         spin_unlock(&kvm->mmu_lock);
 851
 852         /* we've to flush the tlb before the pages can be freed */
 853         if (need_tlb_flush)
 854                 kvm_flush_remote_tlbs(kvm);
 855
 856 }
 857
 858 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 859                                                     struct mm_struct *mm,
 860                                                     unsigned long start,
 861                                                     unsigned long end)
 862 {
 863         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 864         int need_tlb_flush = 0;
 865
 866         spin_lock(&kvm->mmu_lock);
 867         /*
 868          * The count increase must become visible at unlock time as no
 869          * spte can be established without taking the mmu_lock and
 870          * count is also read inside the mmu_lock critical section.
 871          */
 872         kvm->mmu_notifier_count++;
 873         for (; start < end; start += PAGE_SIZE)
 874                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 875         spin_unlock(&kvm->mmu_lock);
 876
 877         /* we've to flush the tlb before the pages can be freed */
 878         if (need_tlb_flush)
 879                 kvm_flush_remote_tlbs(kvm);
 880 }
 881
 882 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 883                                                   struct mm_struct *mm,
 884                                                   unsigned long start,
 885                                                   unsigned long end)
 886 {
 887         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 888
 889         spin_lock(&kvm->mmu_lock);
 890         /*
 891          * This sequence increase will notify the kvm page fault that
 892          * the page that is going to be mapped in the spte could have
 893          * been freed.
 894          */
 895         kvm->mmu_notifier_seq++;
 896         /*
 897          * The above sequence increase must be visible before the
 898          * below count decrease but both values are read by the kvm
 899          * page fault under mmu_lock spinlock so we don't need to add
 900          * a smb_wmb() here in between the two.
 901          */
 902         kvm->mmu_notifier_count--;
 903         spin_unlock(&kvm->mmu_lock);
 904
 905         BUG_ON(kvm->mmu_notifier_count < 0);
 906 }
 907
 908 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 909                                               struct mm_struct *mm,
 910                                               unsigned long address)
 911 {
 912         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 913         int young;
 914
 915         spin_lock(&kvm->mmu_lock);
 916         young = kvm_age_hva(kvm, address);
 917         spin_unlock(&kvm->mmu_lock);
 918
 919         if (young)
 920                 kvm_flush_remote_tlbs(kvm);
 921
 922         return young;
 923 }
 924
 925 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 926                                      struct mm_struct *mm)
 927 {
 928         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 929         kvm_arch_flush_shadow(kvm);
 930 }
 931
 932 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 933         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 934         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 935         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 936         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 937         .release                = kvm_mmu_notifier_release,
 938 };
 939 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 940
 941 static struct kvm *kvm_create_vm(void)
 942 {
 943         struct kvm *kvm = kvm_arch_create_vm();
 944 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 945         struct page *page;
 946 #endif
 947
 948         if (IS_ERR(kvm))
 949                 goto out;
 950 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 951         INIT_LIST_HEAD(&kvm->irq_routing);
 952         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 953 #endif
 954
 955 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 956         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 957         if (!page) {
 958                 kfree(kvm);
 959                 return ERR_PTR(-ENOMEM);
 960         }
 961         kvm->coalesced_mmio_ring =
 962                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 963 #endif
 964
 965 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 966         {
 967                 int err;
 968                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 969                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 970                 if (err) {
 971 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 972                         put_page(page);
 973 #endif
 974                         kfree(kvm);
 975                         return ERR_PTR(err);
 976                 }
 977         }
 978 #endif
 979
 980         kvm->mm = current->mm;
 981         atomic_inc(&kvm->mm->mm_count);
 982         spin_lock_init(&kvm->mmu_lock);
 983         spin_lock_init(&kvm->requests_lock);
 984         kvm_io_bus_init(&kvm->pio_bus);
 985         kvm_irqfd_init(kvm);
 986         mutex_init(&kvm->lock);
 987         mutex_init(&kvm->irq_lock);
 988         kvm_io_bus_init(&kvm->mmio_bus);
 989         init_rwsem(&kvm->slots_lock);
 990         atomic_set(&kvm->users_count, 1);
 991         spin_lock(&kvm_lock);
 992         list_add(&kvm->vm_list, &vm_list);
 993         spin_unlock(&kvm_lock);
 994 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 995         kvm_coalesced_mmio_init(kvm);
 996 #endif
 997 out:
 998         return kvm;
 999 }
1000
1001 /*
1002  * Free any memory in @free but not in @dont.
1003  */
1004 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
1005                                   struct kvm_memory_slot *dont)
1006 {
1007         if (!dont || free->rmap != dont->rmap)
1008                 vfree(free->rmap);
1009
1010         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
1011                 vfree(free->dirty_bitmap);
1012
1013         if (!dont || free->lpage_info != dont->lpage_info)
1014                 vfree(free->lpage_info);
1015
1016         free->npages = 0;
1017         free->dirty_bitmap = NULL;
1018         free->rmap = NULL;
1019         free->lpage_info = NULL;
1020 }
1021
1022 void kvm_free_physmem(struct kvm *kvm)
1023 {
1024         int i;
1025
1026         for (i = 0; i < kvm->nmemslots; ++i)
1027                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
1028 }
1029
1030 static void kvm_destroy_vm(struct kvm *kvm)
1031 {
1032         struct mm_struct *mm = kvm->mm;
1033
1034         kvm_arch_sync_events(kvm);
1035         spin_lock(&kvm_lock);
1036         list_del(&kvm->vm_list);
1037         spin_unlock(&kvm_lock);
1038         kvm_free_irq_routing(kvm);
1039         kvm_io_bus_destroy(&kvm->pio_bus);
1040         kvm_io_bus_destroy(&kvm->mmio_bus);
1041 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1042         if (kvm->coalesced_mmio_ring != NULL)
1043                 free_page((unsigned long)kvm->coalesced_mmio_ring);
1044 #endif
1045 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1046         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1047 #else
1048         kvm_arch_flush_shadow(kvm);
1049 #endif
1050         kvm_arch_destroy_vm(kvm);
1051         mmdrop(mm);
1052 }
1053
1054 void kvm_get_kvm(struct kvm *kvm)
1055 {
1056         atomic_inc(&kvm->users_count);
1057 }
1058 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1059
1060 void kvm_put_kvm(struct kvm *kvm)
1061 {
1062         if (atomic_dec_and_test(&kvm->users_count))
1063                 kvm_destroy_vm(kvm);
1064 }
1065 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1066
1067
1068 static int kvm_vm_release(struct inode *inode, struct file *filp)
1069 {
1070         struct kvm *kvm = filp->private_data;
1071
1072         kvm_irqfd_release(kvm);
1073
1074         kvm_put_kvm(kvm);
1075         return 0;
1076 }
1077
1078 /*
1079  * Allocate some memory and give it an address in the guest physical address
1080  * space.
1081  *
1082  * Discontiguous memory is allowed, mostly for framebuffers.
1083  *
1084  * Must be called holding mmap_sem for write.
1085  */
1086 int __kvm_set_memory_region(struct kvm *kvm,
1087                             struct kvm_userspace_memory_region *mem,
1088                             int user_alloc)
1089 {
1090         int r;
1091         gfn_t base_gfn;
1092         unsigned long npages, ugfn;
1093         unsigned long largepages, i;
1094         struct kvm_memory_slot *memslot;
1095         struct kvm_memory_slot old, new;
1096
1097         r = -EINVAL;
1098         /* General sanity checks */
1099         if (mem->memory_size & (PAGE_SIZE - 1))
1100                 goto out;
1101         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1102                 goto out;
1103         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
1104                 goto out;
1105         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
1106                 goto out;
1107         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1108                 goto out;
1109
1110         memslot = &kvm->memslots[mem->slot];
1111         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1112         npages = mem->memory_size >> PAGE_SHIFT;
1113
1114         if (!npages)
1115                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1116
1117         new = old = *memslot;
1118
1119         new.base_gfn = base_gfn;
1120         new.npages = npages;
1121         new.flags = mem->flags;
1122
1123         /* Disallow changing a memory slot's size. */
1124         r = -EINVAL;
1125         if (npages && old.npages && npages != old.npages)
1126                 goto out_free;
1127
1128         /* Check for overlaps */
1129         r = -EEXIST;
1130         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1131                 struct kvm_memory_slot *s = &kvm->memslots[i];
1132
1133                 if (s == memslot || !s->npages)
1134                         continue;
1135                 if (!((base_gfn + npages <= s->base_gfn) ||
1136                       (base_gfn >= s->base_gfn + s->npages)))
1137                         goto out_free;
1138         }
1139
1140         /* Free page dirty bitmap if unneeded */
1141         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1142                 new.dirty_bitmap = NULL;
1143
1144         r = -ENOMEM;
1145
1146         /* Allocate if a slot is being created */
1147 #ifndef CONFIG_S390
1148         if (npages && !new.rmap) {
1149                 new.rmap = vmalloc(npages * sizeof(struct page *));
1150
1151                 if (!new.rmap)
1152                         goto out_free;
1153
1154                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1155
1156                 new.user_alloc = user_alloc;
1157                 /*
1158                  * hva_to_rmmap() serialzies with the mmu_lock and to be
1159                  * safe it has to ignore memslots with !user_alloc &&
1160                  * !userspace_addr.
1161                  */
1162                 if (user_alloc)
1163                         new.userspace_addr = mem->userspace_addr;
1164                 else
1165                         new.userspace_addr = 0;
1166         }
1167         if (npages && !new.lpage_info) {
1168                 largepages = 1 + (base_gfn + npages - 1) / KVM_PAGES_PER_HPAGE;
1169                 largepages -= base_gfn / KVM_PAGES_PER_HPAGE;
1170
1171                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1172
1173                 if (!new.lpage_info)
1174                         goto out_free;
1175
1176                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1177
1178                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1179                         new.lpage_info[0].write_count = 1;
1180                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1181                         new.lpage_info[largepages-1].write_count = 1;
1182                 ugfn = new.userspace_addr >> PAGE_SHIFT;
1183                 /*
1184                  * If the gfn and userspace address are not aligned wrt each
1185                  * other, disable large page support for this slot
1186                  */
1187                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE - 1))
1188                         for (i = 0; i < largepages; ++i)
1189                                 new.lpage_info[i].write_count = 1;
1190         }
1191
1192         /* Allocate page dirty bitmap if needed */
1193         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1194                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1195
1196                 new.dirty_bitmap = vmalloc(dirty_bytes);
1197                 if (!new.dirty_bitmap)
1198                         goto out_free;
1199                 memset(new.dirty_bitmap, 0, dirty_bytes);
1200                 if (old.npages)
1201                         kvm_arch_flush_shadow(kvm);
1202         }
1203 #endif /* not defined CONFIG_S390 */
1204
1205         if (!npages)
1206                 kvm_arch_flush_shadow(kvm);
1207
1208         spin_lock(&kvm->mmu_lock);
1209         if (mem->slot >= kvm->nmemslots)
1210                 kvm->nmemslots = mem->slot + 1;
1211
1212         *memslot = new;
1213         spin_unlock(&kvm->mmu_lock);
1214
1215         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1216         if (r) {
1217                 spin_lock(&kvm->mmu_lock);
1218                 *memslot = old;
1219                 spin_unlock(&kvm->mmu_lock);
1220                 goto out_free;
1221         }
1222
1223         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1224         /* Slot deletion case: we have to update the current slot */
1225         spin_lock(&kvm->mmu_lock);
1226         if (!npages)
1227                 *memslot = old;
1228         spin_unlock(&kvm->mmu_lock);
1229 #ifdef CONFIG_DMAR
1230         /* map the pages in iommu page table */
1231         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1232         if (r)
1233                 goto out;
1234 #endif
1235         return 0;
1236
1237 out_free:
1238         kvm_free_physmem_slot(&new, &old);
1239 out:
1240         return r;
1241
1242 }
1243 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1244
1245 int kvm_set_memory_region(struct kvm *kvm,
1246                           struct kvm_userspace_memory_region *mem,
1247                           int user_alloc)
1248 {
1249         int r;
1250
1251         down_write(&kvm->slots_lock);
1252         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1253         up_write(&kvm->slots_lock);
1254         return r;
1255 }
1256 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1257
1258 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1259                                    struct
1260                                    kvm_userspace_memory_region *mem,
1261                                    int user_alloc)
1262 {
1263         if (mem->slot >= KVM_MEMORY_SLOTS)
1264                 return -EINVAL;
1265         return kvm_set_memory_region(kvm, mem, user_alloc);
1266 }
1267
1268 int kvm_get_dirty_log(struct kvm *kvm,
1269                         struct kvm_dirty_log *log, int *is_dirty)
1270 {
1271         struct kvm_memory_slot *memslot;
1272         int r, i;
1273         int n;
1274         unsigned long any = 0;
1275
1276         r = -EINVAL;
1277         if (log->slot >= KVM_MEMORY_SLOTS)
1278                 goto out;
1279
1280         memslot = &kvm->memslots[log->slot];
1281         r = -ENOENT;
1282         if (!memslot->dirty_bitmap)
1283                 goto out;
1284
1285         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1286
1287         for (i = 0; !any && i < n/sizeof(long); ++i)
1288                 any = memslot->dirty_bitmap[i];
1289
1290         r = -EFAULT;
1291         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1292                 goto out;
1293
1294         if (any)
1295                 *is_dirty = 1;
1296
1297         r = 0;
1298 out:
1299         return r;
1300 }
1301
1302 int is_error_page(struct page *page)
1303 {
1304         return page == bad_page;
1305 }
1306 EXPORT_SYMBOL_GPL(is_error_page);
1307
1308 int is_error_pfn(pfn_t pfn)
1309 {
1310         return pfn == bad_pfn;
1311 }
1312 EXPORT_SYMBOL_GPL(is_error_pfn);
1313
1314 static inline unsigned long bad_hva(void)
1315 {
1316         return PAGE_OFFSET;
1317 }
1318
1319 int kvm_is_error_hva(unsigned long addr)
1320 {
1321         return addr == bad_hva();
1322 }
1323 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1324
1325 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1326 {
1327         int i;
1328
1329         for (i = 0; i < kvm->nmemslots; ++i) {
1330                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1331
1332                 if (gfn >= memslot->base_gfn
1333                     && gfn < memslot->base_gfn + memslot->npages)
1334                         return memslot;
1335         }
1336         return NULL;
1337 }
1338 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1339
1340 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1341 {
1342         gfn = unalias_gfn(kvm, gfn);
1343         return gfn_to_memslot_unaliased(kvm, gfn);
1344 }
1345
1346 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1347 {
1348         int i;
1349
1350         gfn = unalias_gfn(kvm, gfn);
1351         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1352                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1353
1354                 if (gfn >= memslot->base_gfn
1355                     && gfn < memslot->base_gfn + memslot->npages)
1356                         return 1;
1357         }
1358         return 0;
1359 }
1360 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1361
1362 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1363 {
1364         struct kvm_memory_slot *slot;
1365
1366         gfn = unalias_gfn(kvm, gfn);
1367         slot = gfn_to_memslot_unaliased(kvm, gfn);
1368         if (!slot)
1369                 return bad_hva();
1370         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1371 }
1372 EXPORT_SYMBOL_GPL(gfn_to_hva);
1373
1374 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1375 {
1376         struct page *page[1];
1377         unsigned long addr;
1378         int npages;
1379         pfn_t pfn;
1380
1381         might_sleep();
1382
1383         addr = gfn_to_hva(kvm, gfn);
1384         if (kvm_is_error_hva(addr)) {
1385                 get_page(bad_page);
1386                 return page_to_pfn(bad_page);
1387         }
1388
1389         npages = get_user_pages_fast(addr, 1, 1, page);
1390
1391         if (unlikely(npages != 1)) {
1392                 struct vm_area_struct *vma;
1393
1394                 down_read(&current->mm->mmap_sem);
1395                 vma = find_vma(current->mm, addr);
1396
1397                 if (vma == NULL || addr < vma->vm_start ||
1398                     !(vma->vm_flags & VM_PFNMAP)) {
1399                         up_read(&current->mm->mmap_sem);
1400                         get_page(bad_page);
1401                         return page_to_pfn(bad_page);
1402                 }
1403
1404                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1405                 up_read(&current->mm->mmap_sem);
1406                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1407         } else
1408                 pfn = page_to_pfn(page[0]);
1409
1410         return pfn;
1411 }
1412
1413 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1414
1415 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1416 {
1417         pfn_t pfn;
1418
1419         pfn = gfn_to_pfn(kvm, gfn);
1420         if (!kvm_is_mmio_pfn(pfn))
1421                 return pfn_to_page(pfn);
1422
1423         WARN_ON(kvm_is_mmio_pfn(pfn));
1424
1425         get_page(bad_page);
1426         return bad_page;
1427 }
1428
1429 EXPORT_SYMBOL_GPL(gfn_to_page);
1430
1431 void kvm_release_page_clean(struct page *page)
1432 {
1433         kvm_release_pfn_clean(page_to_pfn(page));
1434 }
1435 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1436
1437 void kvm_release_pfn_clean(pfn_t pfn)
1438 {
1439         if (!kvm_is_mmio_pfn(pfn))
1440                 put_page(pfn_to_page(pfn));
1441 }
1442 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1443
1444 void kvm_release_page_dirty(struct page *page)
1445 {
1446         kvm_release_pfn_dirty(page_to_pfn(page));
1447 }
1448 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1449
1450 void kvm_release_pfn_dirty(pfn_t pfn)
1451 {
1452         kvm_set_pfn_dirty(pfn);
1453         kvm_release_pfn_clean(pfn);
1454 }
1455 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1456
1457 void kvm_set_page_dirty(struct page *page)
1458 {
1459         kvm_set_pfn_dirty(page_to_pfn(page));
1460 }
1461 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1462
1463 void kvm_set_pfn_dirty(pfn_t pfn)
1464 {
1465         if (!kvm_is_mmio_pfn(pfn)) {
1466                 struct page *page = pfn_to_page(pfn);
1467                 if (!PageReserved(page))
1468                         SetPageDirty(page);
1469         }
1470 }
1471 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1472
1473 void kvm_set_pfn_accessed(pfn_t pfn)
1474 {
1475         if (!kvm_is_mmio_pfn(pfn))
1476                 mark_page_accessed(pfn_to_page(pfn));
1477 }
1478 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1479
1480 void kvm_get_pfn(pfn_t pfn)
1481 {
1482         if (!kvm_is_mmio_pfn(pfn))
1483                 get_page(pfn_to_page(pfn));
1484 }
1485 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1486
1487 static int next_segment(unsigned long len, int offset)
1488 {
1489         if (len > PAGE_SIZE - offset)
1490                 return PAGE_SIZE - offset;
1491         else
1492                 return len;
1493 }
1494
1495 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1496                         int len)
1497 {
1498         int r;
1499         unsigned long addr;
1500
1501         addr = gfn_to_hva(kvm, gfn);
1502         if (kvm_is_error_hva(addr))
1503                 return -EFAULT;
1504         r = copy_from_user(data, (void __user *)addr + offset, len);
1505         if (r)
1506                 return -EFAULT;
1507         return 0;
1508 }
1509 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1510
1511 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1512 {
1513         gfn_t gfn = gpa >> PAGE_SHIFT;
1514         int seg;
1515         int offset = offset_in_page(gpa);
1516         int ret;
1517
1518         while ((seg = next_segment(len, offset)) != 0) {
1519                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1520                 if (ret < 0)
1521                         return ret;
1522                 offset = 0;
1523                 len -= seg;
1524                 data += seg;
1525                 ++gfn;
1526         }
1527         return 0;
1528 }
1529 EXPORT_SYMBOL_GPL(kvm_read_guest);
1530
1531 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1532                           unsigned long len)
1533 {
1534         int r;
1535         unsigned long addr;
1536         gfn_t gfn = gpa >> PAGE_SHIFT;
1537         int offset = offset_in_page(gpa);
1538
1539         addr = gfn_to_hva(kvm, gfn);
1540         if (kvm_is_error_hva(addr))
1541                 return -EFAULT;
1542         pagefault_disable();
1543         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1544         pagefault_enable();
1545         if (r)
1546                 return -EFAULT;
1547         return 0;
1548 }
1549 EXPORT_SYMBOL(kvm_read_guest_atomic);
1550
1551 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1552                          int offset, int len)
1553 {
1554         int r;
1555         unsigned long addr;
1556
1557         addr = gfn_to_hva(kvm, gfn);
1558         if (kvm_is_error_hva(addr))
1559                 return -EFAULT;
1560         r = copy_to_user((void __user *)addr + offset, data, len);
1561         if (r)
1562                 return -EFAULT;
1563         mark_page_dirty(kvm, gfn);
1564         return 0;
1565 }
1566 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1567
1568 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1569                     unsigned long len)
1570 {
1571         gfn_t gfn = gpa >> PAGE_SHIFT;
1572         int seg;
1573         int offset = offset_in_page(gpa);
1574         int ret;
1575
1576         while ((seg = next_segment(len, offset)) != 0) {
1577                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1578                 if (ret < 0)
1579                         return ret;
1580                 offset = 0;
1581                 len -= seg;
1582                 data += seg;
1583                 ++gfn;
1584         }
1585         return 0;
1586 }
1587
1588 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1589 {
1590         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1591 }
1592 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1593
1594 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1595 {
1596         gfn_t gfn = gpa >> PAGE_SHIFT;
1597         int seg;
1598         int offset = offset_in_page(gpa);
1599         int ret;
1600
1601         while ((seg = next_segment(len, offset)) != 0) {
1602                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1603                 if (ret < 0)
1604                         return ret;
1605                 offset = 0;
1606                 len -= seg;
1607                 ++gfn;
1608         }
1609         return 0;
1610 }
1611 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1612
1613 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1614 {
1615         struct kvm_memory_slot *memslot;
1616
1617         gfn = unalias_gfn(kvm, gfn);
1618         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1619         if (memslot && memslot->dirty_bitmap) {
1620                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1621
1622                 /* avoid RMW */
1623                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1624                         set_bit(rel_gfn, memslot->dirty_bitmap);
1625         }
1626 }
1627
1628 /*
1629  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1630  */
1631 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1632 {
1633         DEFINE_WAIT(wait);
1634
1635         for (;;) {
1636                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1637
1638                 if ((kvm_arch_interrupt_allowed(vcpu) &&
1639                                         kvm_cpu_has_interrupt(vcpu)) ||
1640                                 kvm_arch_vcpu_runnable(vcpu)) {
1641                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1642                         break;
1643                 }
1644                 if (kvm_cpu_has_pending_timer(vcpu))
1645                         break;
1646                 if (signal_pending(current))
1647                         break;
1648
1649                 vcpu_put(vcpu);
1650                 schedule();
1651                 vcpu_load(vcpu);
1652         }
1653
1654         finish_wait(&vcpu->wq, &wait);
1655 }
1656
1657 void kvm_resched(struct kvm_vcpu *vcpu)
1658 {
1659         if (!need_resched())
1660                 return;
1661         cond_resched();
1662 }
1663 EXPORT_SYMBOL_GPL(kvm_resched);
1664
1665 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1666 {
1667         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1668         struct page *page;
1669
1670         if (vmf->pgoff == 0)
1671                 page = virt_to_page(vcpu->run);
1672 #ifdef CONFIG_X86
1673         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1674                 page = virt_to_page(vcpu->arch.pio_data);
1675 #endif
1676 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1677         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1678                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1679 #endif
1680         else
1681                 return VM_FAULT_SIGBUS;
1682         get_page(page);
1683         vmf->page = page;
1684         return 0;
1685 }
1686
1687 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1688         .fault = kvm_vcpu_fault,
1689 };
1690
1691 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1692 {
1693         vma->vm_ops = &kvm_vcpu_vm_ops;
1694         return 0;
1695 }
1696
1697 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1698 {
1699         struct kvm_vcpu *vcpu = filp->private_data;
1700
1701         kvm_put_kvm(vcpu->kvm);
1702         return 0;
1703 }
1704
1705 static struct file_operations kvm_vcpu_fops = {
1706         .release        = kvm_vcpu_release,
1707         .unlocked_ioctl = kvm_vcpu_ioctl,
1708         .compat_ioctl   = kvm_vcpu_ioctl,
1709         .mmap           = kvm_vcpu_mmap,
1710 };
1711
1712 /*
1713  * Allocates an inode for the vcpu.
1714  */
1715 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1716 {
1717         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1718         if (fd < 0)
1719                 kvm_put_kvm(vcpu->kvm);
1720         return fd;
1721 }
1722
1723 /*
1724  * Creates some virtual cpus.  Good luck creating more than one.
1725  */
1726 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1727 {
1728         int r;
1729         struct kvm_vcpu *vcpu;
1730
1731         if (!valid_vcpu(n))
1732                 return -EINVAL;
1733
1734         vcpu = kvm_arch_vcpu_create(kvm, n);
1735         if (IS_ERR(vcpu))
1736                 return PTR_ERR(vcpu);
1737
1738         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1739
1740         r = kvm_arch_vcpu_setup(vcpu);
1741         if (r)
1742                 return r;
1743
1744         mutex_lock(&kvm->lock);
1745         if (kvm->vcpus[n]) {
1746                 r = -EEXIST;
1747                 goto vcpu_destroy;
1748         }
1749         kvm->vcpus[n] = vcpu;
1750         mutex_unlock(&kvm->lock);
1751
1752         /* Now it's all set up, let userspace reach it */
1753         kvm_get_kvm(kvm);
1754         r = create_vcpu_fd(vcpu);
1755         if (r < 0)
1756                 goto unlink;
1757         return r;
1758
1759 unlink:
1760         mutex_lock(&kvm->lock);
1761         kvm->vcpus[n] = NULL;
1762 vcpu_destroy:
1763         mutex_unlock(&kvm->lock);
1764         kvm_arch_vcpu_destroy(vcpu);
1765         return r;
1766 }
1767
1768 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1769 {
1770         if (sigset) {
1771                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1772                 vcpu->sigset_active = 1;
1773                 vcpu->sigset = *sigset;
1774         } else
1775                 vcpu->sigset_active = 0;
1776         return 0;
1777 }
1778
1779 #ifdef __KVM_HAVE_MSIX
1780 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1781                                     struct kvm_assigned_msix_nr *entry_nr)
1782 {
1783         int r = 0;
1784         struct kvm_assigned_dev_kernel *adev;
1785
1786         mutex_lock(&kvm->lock);
1787
1788         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1789                                       entry_nr->assigned_dev_id);
1790         if (!adev) {
1791                 r = -EINVAL;
1792                 goto msix_nr_out;
1793         }
1794
1795         if (adev->entries_nr == 0) {
1796                 adev->entries_nr = entry_nr->entry_nr;
1797                 if (adev->entries_nr == 0 ||
1798                     adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1799                         r = -EINVAL;
1800                         goto msix_nr_out;
1801                 }
1802
1803                 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1804                                                 entry_nr->entry_nr,
1805                                                 GFP_KERNEL);
1806                 if (!adev->host_msix_entries) {
1807                         r = -ENOMEM;
1808                         goto msix_nr_out;
1809                 }
1810                 adev->guest_msix_entries = kzalloc(
1811                                 sizeof(struct kvm_guest_msix_entry) *
1812                                 entry_nr->entry_nr, GFP_KERNEL);
1813                 if (!adev->guest_msix_entries) {
1814                         kfree(adev->host_msix_entries);
1815                         r = -ENOMEM;
1816                         goto msix_nr_out;
1817                 }
1818         } else /* Not allowed set MSI-X number twice */
1819                 r = -EINVAL;
1820 msix_nr_out:
1821         mutex_unlock(&kvm->lock);
1822         return r;
1823 }
1824
1825 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1826                                        struct kvm_assigned_msix_entry *entry)
1827 {
1828         int r = 0, i;
1829         struct kvm_assigned_dev_kernel *adev;
1830
1831         mutex_lock(&kvm->lock);
1832
1833         adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1834                                       entry->assigned_dev_id);
1835
1836         if (!adev) {
1837                 r = -EINVAL;
1838                 goto msix_entry_out;
1839         }
1840
1841         for (i = 0; i < adev->entries_nr; i++)
1842                 if (adev->guest_msix_entries[i].vector == 0 ||
1843                     adev->guest_msix_entries[i].entry == entry->entry) {
1844                         adev->guest_msix_entries[i].entry = entry->entry;
1845                         adev->guest_msix_entries[i].vector = entry->gsi;
1846                         adev->host_msix_entries[i].entry = entry->entry;
1847                         break;
1848                 }
1849         if (i == adev->entries_nr) {
1850                 r = -ENOSPC;
1851                 goto msix_entry_out;
1852         }
1853
1854 msix_entry_out:
1855         mutex_unlock(&kvm->lock);
1856
1857         return r;
1858 }
1859 #endif
1860
1861 static long kvm_vcpu_ioctl(struct file *filp,
1862                            unsigned int ioctl, unsigned long arg)
1863 {
1864         struct kvm_vcpu *vcpu = filp->private_data;
1865         void __user *argp = (void __user *)arg;
1866         int r;
1867         struct kvm_fpu *fpu = NULL;
1868         struct kvm_sregs *kvm_sregs = NULL;
1869
1870         if (vcpu->kvm->mm != current->mm)
1871                 return -EIO;
1872         switch (ioctl) {
1873         case KVM_RUN:
1874                 r = -EINVAL;
1875                 if (arg)
1876                         goto out;
1877                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1878                 break;
1879         case KVM_GET_REGS: {
1880                 struct kvm_regs *kvm_regs;
1881
1882                 r = -ENOMEM;
1883                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1884                 if (!kvm_regs)
1885                         goto out;
1886                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1887                 if (r)
1888                         goto out_free1;
1889                 r = -EFAULT;
1890                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1891                         goto out_free1;
1892                 r = 0;
1893 out_free1:
1894                 kfree(kvm_regs);
1895                 break;
1896         }
1897         case KVM_SET_REGS: {
1898                 struct kvm_regs *kvm_regs;
1899
1900                 r = -ENOMEM;
1901                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1902                 if (!kvm_regs)
1903                         goto out;
1904                 r = -EFAULT;
1905                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1906                         goto out_free2;
1907                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1908                 if (r)
1909                         goto out_free2;
1910                 r = 0;
1911 out_free2:
1912                 kfree(kvm_regs);
1913                 break;
1914         }
1915         case KVM_GET_SREGS: {
1916                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1917                 r = -ENOMEM;
1918                 if (!kvm_sregs)
1919                         goto out;
1920                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1921                 if (r)
1922                         goto out;
1923                 r = -EFAULT;
1924                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1925                         goto out;
1926                 r = 0;
1927                 break;
1928         }
1929         case KVM_SET_SREGS: {
1930                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1931                 r = -ENOMEM;
1932                 if (!kvm_sregs)
1933                         goto out;
1934                 r = -EFAULT;
1935                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1936                         goto out;
1937                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1938                 if (r)
1939                         goto out;
1940                 r = 0;
1941                 break;
1942         }
1943         case KVM_GET_MP_STATE: {
1944                 struct kvm_mp_state mp_state;
1945
1946                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1947                 if (r)
1948                         goto out;
1949                 r = -EFAULT;
1950                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1951                         goto out;
1952                 r = 0;
1953                 break;
1954         }
1955         case KVM_SET_MP_STATE: {
1956                 struct kvm_mp_state mp_state;
1957
1958                 r = -EFAULT;
1959                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1960                         goto out;
1961                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1962                 if (r)
1963                         goto out;
1964                 r = 0;
1965                 break;
1966         }
1967         case KVM_TRANSLATE: {
1968                 struct kvm_translation tr;
1969
1970                 r = -EFAULT;
1971                 if (copy_from_user(&tr, argp, sizeof tr))
1972                         goto out;
1973                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1974                 if (r)
1975                         goto out;
1976                 r = -EFAULT;
1977                 if (copy_to_user(argp, &tr, sizeof tr))
1978                         goto out;
1979                 r = 0;
1980                 break;
1981         }
1982         case KVM_SET_GUEST_DEBUG: {
1983                 struct kvm_guest_debug dbg;
1984
1985                 r = -EFAULT;
1986                 if (copy_from_user(&dbg, argp, sizeof dbg))
1987                         goto out;
1988                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1989                 if (r)
1990                         goto out;
1991                 r = 0;
1992                 break;
1993         }
1994         case KVM_SET_SIGNAL_MASK: {
1995                 struct kvm_signal_mask __user *sigmask_arg = argp;
1996                 struct kvm_signal_mask kvm_sigmask;
1997                 sigset_t sigset, *p;
1998
1999                 p = NULL;
2000                 if (argp) {
2001                         r = -EFAULT;
2002                         if (copy_from_user(&kvm_sigmask, argp,
2003                                            sizeof kvm_sigmask))
2004                                 goto out;
2005                         r = -EINVAL;
2006                         if (kvm_sigmask.len != sizeof sigset)
2007                                 goto out;
2008                         r = -EFAULT;
2009                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2010                                            sizeof sigset))
2011                                 goto out;
2012                         p = &sigset;
2013                 }
2014                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2015                 break;
2016         }
2017         case KVM_GET_FPU: {
2018                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2019                 r = -ENOMEM;
2020                 if (!fpu)
2021                         goto out;
2022                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2023                 if (r)
2024                         goto out;
2025                 r = -EFAULT;
2026                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2027                         goto out;
2028                 r = 0;
2029                 break;
2030         }
2031         case KVM_SET_FPU: {
2032                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2033                 r = -ENOMEM;
2034                 if (!fpu)
2035                         goto out;
2036                 r = -EFAULT;
2037                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
2038                         goto out;
2039                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2040                 if (r)
2041                         goto out;
2042                 r = 0;
2043                 break;
2044         }
2045         default:
2046                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2047         }
2048 out:
2049         kfree(fpu);
2050         kfree(kvm_sregs);
2051         return r;
2052 }
2053
2054 static long kvm_vm_ioctl(struct file *filp,
2055                            unsigned int ioctl, unsigned long arg)
2056 {
2057         struct kvm *kvm = filp->private_data;
2058         void __user *argp = (void __user *)arg;
2059         int r;
2060
2061         if (kvm->mm != current->mm)
2062                 return -EIO;
2063         switch (ioctl) {
2064         case KVM_CREATE_VCPU:
2065                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2066                 if (r < 0)
2067                         goto out;
2068                 break;
2069         case KVM_SET_USER_MEMORY_REGION: {
2070                 struct kvm_userspace_memory_region kvm_userspace_mem;
2071
2072                 r = -EFAULT;
2073                 if (copy_from_user(&kvm_userspace_mem, argp,
2074                                                 sizeof kvm_userspace_mem))
2075                         goto out;
2076
2077                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2078                 if (r)
2079                         goto out;
2080                 break;
2081         }
2082         case KVM_GET_DIRTY_LOG: {
2083                 struct kvm_dirty_log log;
2084
2085                 r = -EFAULT;
2086                 if (copy_from_user(&log, argp, sizeof log))
2087                         goto out;
2088                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2089                 if (r)
2090                         goto out;
2091                 break;
2092         }
2093 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2094         case KVM_REGISTER_COALESCED_MMIO: {
2095                 struct kvm_coalesced_mmio_zone zone;
2096                 r = -EFAULT;
2097                 if (copy_from_user(&zone, argp, sizeof zone))
2098                         goto out;
2099                 r = -ENXIO;
2100                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2101                 if (r)
2102                         goto out;
2103                 r = 0;
2104                 break;
2105         }
2106         case KVM_UNREGISTER_COALESCED_MMIO: {
2107                 struct kvm_coalesced_mmio_zone zone;
2108                 r = -EFAULT;
2109                 if (copy_from_user(&zone, argp, sizeof zone))
2110                         goto out;
2111                 r = -ENXIO;
2112                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2113                 if (r)
2114                         goto out;
2115                 r = 0;
2116                 break;
2117         }
2118 #endif
2119 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
2120         case KVM_ASSIGN_PCI_DEVICE: {
2121                 struct kvm_assigned_pci_dev assigned_dev;
2122
2123                 r = -EFAULT;
2124                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2125                         goto out;
2126                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2127                 if (r)
2128                         goto out;
2129                 break;
2130         }
2131         case KVM_ASSIGN_IRQ: {
2132                 r = -EOPNOTSUPP;
2133                 break;
2134         }
2135 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
2136         case KVM_ASSIGN_DEV_IRQ: {
2137                 struct kvm_assigned_irq assigned_irq;
2138
2139                 r = -EFAULT;
2140                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2141                         goto out;
2142                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2143                 if (r)
2144                         goto out;
2145                 break;
2146         }
2147         case KVM_DEASSIGN_DEV_IRQ: {
2148                 struct kvm_assigned_irq assigned_irq;
2149
2150                 r = -EFAULT;
2151                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2152                         goto out;
2153                 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2154                 if (r)
2155                         goto out;
2156                 break;
2157         }
2158 #endif
2159 #endif
2160 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2161         case KVM_DEASSIGN_PCI_DEVICE: {
2162                 struct kvm_assigned_pci_dev assigned_dev;
2163
2164                 r = -EFAULT;
2165                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2166                         goto out;
2167                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2168                 if (r)
2169                         goto out;
2170                 break;
2171         }
2172 #endif
2173 #ifdef KVM_CAP_IRQ_ROUTING
2174         case KVM_SET_GSI_ROUTING: {
2175                 struct kvm_irq_routing routing;
2176                 struct kvm_irq_routing __user *urouting;
2177                 struct kvm_irq_routing_entry *entries;
2178
2179                 r = -EFAULT;
2180                 if (copy_from_user(&routing, argp, sizeof(routing)))
2181                         goto out;
2182                 r = -EINVAL;
2183                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2184                         goto out;
2185                 if (routing.flags)
2186                         goto out;
2187                 r = -ENOMEM;
2188                 entries = vmalloc(routing.nr * sizeof(*entries));
2189                 if (!entries)
2190                         goto out;
2191                 r = -EFAULT;
2192                 urouting = argp;
2193                 if (copy_from_user(entries, urouting->entries,
2194                                    routing.nr * sizeof(*entries)))
2195                         goto out_free_irq_routing;
2196                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2197                                         routing.flags);
2198         out_free_irq_routing:
2199                 vfree(entries);
2200                 break;
2201         }
2202 #ifdef __KVM_HAVE_MSIX
2203         case KVM_ASSIGN_SET_MSIX_NR: {
2204                 struct kvm_assigned_msix_nr entry_nr;
2205                 r = -EFAULT;
2206                 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2207                         goto out;
2208                 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2209                 if (r)
2210                         goto out;
2211                 break;
2212         }
2213         case KVM_ASSIGN_SET_MSIX_ENTRY: {
2214                 struct kvm_assigned_msix_entry entry;
2215                 r = -EFAULT;
2216                 if (copy_from_user(&entry, argp, sizeof entry))
2217                         goto out;
2218                 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2219                 if (r)
2220                         goto out;
2221                 break;
2222         }
2223 #endif
2224 #endif /* KVM_CAP_IRQ_ROUTING */
2225         case KVM_IRQFD: {
2226                 struct kvm_irqfd data;
2227
2228                 r = -EFAULT;
2229                 if (copy_from_user(&data, argp, sizeof data))
2230                         goto out;
2231                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2232                 break;
2233         }
2234         default:
2235                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2236         }
2237 out:
2238         return r;
2239 }
2240
2241 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2242 {
2243         struct page *page[1];
2244         unsigned long addr;
2245         int npages;
2246         gfn_t gfn = vmf->pgoff;
2247         struct kvm *kvm = vma->vm_file->private_data;
2248
2249         addr = gfn_to_hva(kvm, gfn);
2250         if (kvm_is_error_hva(addr))
2251                 return VM_FAULT_SIGBUS;
2252
2253         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2254                                 NULL);
2255         if (unlikely(npages != 1))
2256                 return VM_FAULT_SIGBUS;
2257
2258         vmf->page = page[0];
2259         return 0;
2260 }
2261
2262 static struct vm_operations_struct kvm_vm_vm_ops = {
2263         .fault = kvm_vm_fault,
2264 };
2265
2266 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2267 {
2268         vma->vm_ops = &kvm_vm_vm_ops;
2269         return 0;
2270 }
2271
2272 static struct file_operations kvm_vm_fops = {
2273         .release        = kvm_vm_release,
2274         .unlocked_ioctl = kvm_vm_ioctl,
2275         .compat_ioctl   = kvm_vm_ioctl,
2276         .mmap           = kvm_vm_mmap,
2277 };
2278
2279 static int kvm_dev_ioctl_create_vm(void)
2280 {
2281         int fd;
2282         struct kvm *kvm;
2283
2284         kvm = kvm_create_vm();
2285         if (IS_ERR(kvm))
2286                 return PTR_ERR(kvm);
2287         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2288         if (fd < 0)
2289                 kvm_put_kvm(kvm);
2290
2291         return fd;
2292 }
2293
2294 static long kvm_dev_ioctl_check_extension_generic(long arg)
2295 {
2296         switch (arg) {
2297         case KVM_CAP_USER_MEMORY:
2298         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2299         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2300                 return 1;
2301 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2302         case KVM_CAP_IRQ_ROUTING:
2303                 return KVM_MAX_IRQ_ROUTES;
2304 #endif
2305         default:
2306                 break;
2307         }
2308         return kvm_dev_ioctl_check_extension(arg);
2309 }
2310
2311 static long kvm_dev_ioctl(struct file *filp,
2312                           unsigned int ioctl, unsigned long arg)
2313 {
2314         long r = -EINVAL;
2315
2316         switch (ioctl) {
2317         case KVM_GET_API_VERSION:
2318                 r = -EINVAL;
2319                 if (arg)
2320                         goto out;
2321                 r = KVM_API_VERSION;
2322                 break;
2323         case KVM_CREATE_VM:
2324                 r = -EINVAL;
2325                 if (arg)
2326                         goto out;
2327                 r = kvm_dev_ioctl_create_vm();
2328                 break;
2329         case KVM_CHECK_EXTENSION:
2330                 r = kvm_dev_ioctl_check_extension_generic(arg);
2331                 break;
2332         case KVM_GET_VCPU_MMAP_SIZE:
2333                 r = -EINVAL;
2334                 if (arg)
2335                         goto out;
2336                 r = PAGE_SIZE;     /* struct kvm_run */
2337 #ifdef CONFIG_X86
2338                 r += PAGE_SIZE;    /* pio data page */
2339 #endif
2340 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2341                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2342 #endif
2343                 break;
2344         case KVM_TRACE_ENABLE:
2345         case KVM_TRACE_PAUSE:
2346         case KVM_TRACE_DISABLE:
2347                 r = kvm_trace_ioctl(ioctl, arg);
2348                 break;
2349         default:
2350                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2351         }
2352 out:
2353         return r;
2354 }
2355
2356 static struct file_operations kvm_chardev_ops = {
2357         .unlocked_ioctl = kvm_dev_ioctl,
2358         .compat_ioctl   = kvm_dev_ioctl,
2359 };
2360
2361 static struct miscdevice kvm_dev = {
2362         KVM_MINOR,
2363         "kvm",
2364         &kvm_chardev_ops,
2365 };
2366
2367 static void hardware_enable(void *junk)
2368 {
2369         int cpu = raw_smp_processor_id();
2370
2371         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2372                 return;
2373         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2374         kvm_arch_hardware_enable(NULL);
2375 }
2376
2377 static void hardware_disable(void *junk)
2378 {
2379         int cpu = raw_smp_processor_id();
2380
2381         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2382                 return;
2383         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2384         kvm_arch_hardware_disable(NULL);
2385 }
2386
2387 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2388                            void *v)
2389 {
2390         int cpu = (long)v;
2391
2392         val &= ~CPU_TASKS_FROZEN;
2393         switch (val) {
2394         case CPU_DYING:
2395                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2396                        cpu);
2397                 hardware_disable(NULL);
2398                 break;
2399         case CPU_UP_CANCELED:
2400                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2401                        cpu);
2402                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2403                 break;
2404         case CPU_ONLINE:
2405                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2406                        cpu);
2407                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2408                 break;
2409         }
2410         return NOTIFY_OK;
2411 }
2412
2413
2414 asmlinkage void kvm_handle_fault_on_reboot(void)
2415 {
2416         if (kvm_rebooting)
2417                 /* spin while reset goes on */
2418                 while (true)
2419                         ;
2420         /* Fault while not rebooting.  We want the trace. */
2421         BUG();
2422 }
2423 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2424
2425 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2426                       void *v)
2427 {
2428         /*
2429          * Some (well, at least mine) BIOSes hang on reboot if
2430          * in vmx root mode.
2431          *
2432          * And Intel TXT required VMX off for all cpu when system shutdown.
2433          */
2434         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2435         kvm_rebooting = true;
2436         on_each_cpu(hardware_disable, NULL, 1);
2437         return NOTIFY_OK;
2438 }
2439
2440 static struct notifier_block kvm_reboot_notifier = {
2441         .notifier_call = kvm_reboot,
2442         .priority = 0,
2443 };
2444
2445 void kvm_io_bus_init(struct kvm_io_bus *bus)
2446 {
2447         memset(bus, 0, sizeof(*bus));
2448 }
2449
2450 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2451 {
2452         int i;
2453
2454         for (i = 0; i < bus->dev_count; i++) {
2455                 struct kvm_io_device *pos = bus->devs[i];
2456
2457                 kvm_iodevice_destructor(pos);
2458         }
2459 }
2460
2461 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2462                                           gpa_t addr, int len, int is_write)
2463 {
2464         int i;
2465
2466         for (i = 0; i < bus->dev_count; i++) {
2467                 struct kvm_io_device *pos = bus->devs[i];
2468
2469                 if (kvm_iodevice_in_range(pos, addr, len, is_write))
2470                         return pos;
2471         }
2472
2473         return NULL;
2474 }
2475
2476 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2477 {
2478         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2479
2480         bus->devs[bus->dev_count++] = dev;
2481 }
2482
2483 static struct notifier_block kvm_cpu_notifier = {
2484         .notifier_call = kvm_cpu_hotplug,
2485         .priority = 20, /* must be > scheduler priority */
2486 };
2487
2488 static int vm_stat_get(void *_offset, u64 *val)
2489 {
2490         unsigned offset = (long)_offset;
2491         struct kvm *kvm;
2492
2493         *val = 0;
2494         spin_lock(&kvm_lock);
2495         list_for_each_entry(kvm, &vm_list, vm_list)
2496                 *val += *(u32 *)((void *)kvm + offset);
2497         spin_unlock(&kvm_lock);
2498         return 0;
2499 }
2500
2501 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2502
2503 static int vcpu_stat_get(void *_offset, u64 *val)
2504 {
2505         unsigned offset = (long)_offset;
2506         struct kvm *kvm;
2507         struct kvm_vcpu *vcpu;
2508         int i;
2509
2510         *val = 0;
2511         spin_lock(&kvm_lock);
2512         list_for_each_entry(kvm, &vm_list, vm_list)
2513                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2514                         vcpu = kvm->vcpus[i];
2515                         if (vcpu)
2516                                 *val += *(u32 *)((void *)vcpu + offset);
2517                 }
2518         spin_unlock(&kvm_lock);
2519         return 0;
2520 }
2521
2522 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2523
2524 static struct file_operations *stat_fops[] = {
2525         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2526         [KVM_STAT_VM]   = &vm_stat_fops,
2527 };
2528
2529 static void kvm_init_debug(void)
2530 {
2531         struct kvm_stats_debugfs_item *p;
2532
2533         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2534         for (p = debugfs_entries; p->name; ++p)
2535                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2536                                                 (void *)(long)p->offset,
2537                                                 stat_fops[p->kind]);
2538 }
2539
2540 static void kvm_exit_debug(void)
2541 {
2542         struct kvm_stats_debugfs_item *p;
2543
2544         for (p = debugfs_entries; p->name; ++p)
2545                 debugfs_remove(p->dentry);
2546         debugfs_remove(kvm_debugfs_dir);
2547 }
2548
2549 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2550 {
2551         hardware_disable(NULL);
2552         return 0;
2553 }
2554
2555 static int kvm_resume(struct sys_device *dev)
2556 {
2557         hardware_enable(NULL);
2558         return 0;
2559 }
2560
2561 static struct sysdev_class kvm_sysdev_class = {
2562         .name = "kvm",
2563         .suspend = kvm_suspend,
2564         .resume = kvm_resume,
2565 };
2566
2567 static struct sys_device kvm_sysdev = {
2568         .id = 0,
2569         .cls = &kvm_sysdev_class,
2570 };
2571
2572 struct page *bad_page;
2573 pfn_t bad_pfn;
2574
2575 static inline
2576 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2577 {
2578         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2579 }
2580
2581 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2582 {
2583         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2584
2585         kvm_arch_vcpu_load(vcpu, cpu);
2586 }
2587
2588 static void kvm_sched_out(struct preempt_notifier *pn,
2589                           struct task_struct *next)
2590 {
2591         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2592
2593         kvm_arch_vcpu_put(vcpu);
2594 }
2595
2596 int kvm_init(void *opaque, unsigned int vcpu_size,
2597                   struct module *module)
2598 {
2599         int r;
2600         int cpu;
2601
2602         kvm_init_debug();
2603
2604         r = kvm_arch_init(opaque);
2605         if (r)
2606                 goto out_fail;
2607
2608         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2609
2610         if (bad_page == NULL) {
2611                 r = -ENOMEM;
2612                 goto out;
2613         }
2614
2615         bad_pfn = page_to_pfn(bad_page);
2616
2617         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2618                 r = -ENOMEM;
2619                 goto out_free_0;
2620         }
2621
2622         r = kvm_arch_hardware_setup();
2623         if (r < 0)
2624                 goto out_free_0a;
2625
2626         for_each_online_cpu(cpu) {
2627                 smp_call_function_single(cpu,
2628                                 kvm_arch_check_processor_compat,
2629                                 &r, 1);
2630                 if (r < 0)
2631                         goto out_free_1;
2632         }
2633
2634         on_each_cpu(hardware_enable, NULL, 1);
2635         r = register_cpu_notifier(&kvm_cpu_notifier);
2636         if (r)
2637                 goto out_free_2;
2638         register_reboot_notifier(&kvm_reboot_notifier);
2639
2640         r = sysdev_class_register(&kvm_sysdev_class);
2641         if (r)
2642                 goto out_free_3;
2643
2644         r = sysdev_register(&kvm_sysdev);
2645         if (r)
2646                 goto out_free_4;
2647
2648         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2649         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2650                                            __alignof__(struct kvm_vcpu),
2651                                            0, NULL);
2652         if (!kvm_vcpu_cache) {
2653                 r = -ENOMEM;
2654                 goto out_free_5;
2655         }
2656
2657         kvm_chardev_ops.owner = module;
2658         kvm_vm_fops.owner = module;
2659         kvm_vcpu_fops.owner = module;
2660
2661         r = misc_register(&kvm_dev);
2662         if (r) {
2663                 printk(KERN_ERR "kvm: misc device register failed\n");
2664                 goto out_free;
2665         }
2666
2667         kvm_preempt_ops.sched_in = kvm_sched_in;
2668         kvm_preempt_ops.sched_out = kvm_sched_out;
2669
2670         return 0;
2671
2672 out_free:
2673         kmem_cache_destroy(kvm_vcpu_cache);
2674 out_free_5:
2675         sysdev_unregister(&kvm_sysdev);
2676 out_free_4:
2677         sysdev_class_unregister(&kvm_sysdev_class);
2678 out_free_3:
2679         unregister_reboot_notifier(&kvm_reboot_notifier);
2680         unregister_cpu_notifier(&kvm_cpu_notifier);
2681 out_free_2:
2682         on_each_cpu(hardware_disable, NULL, 1);
2683 out_free_1:
2684         kvm_arch_hardware_unsetup();
2685 out_free_0a:
2686         free_cpumask_var(cpus_hardware_enabled);
2687 out_free_0:
2688         __free_page(bad_page);
2689 out:
2690         kvm_arch_exit();
2691         kvm_exit_debug();
2692 out_fail:
2693         return r;
2694 }
2695 EXPORT_SYMBOL_GPL(kvm_init);
2696
2697 void kvm_exit(void)
2698 {
2699         kvm_trace_cleanup();
2700         misc_deregister(&kvm_dev);
2701         kmem_cache_destroy(kvm_vcpu_cache);
2702         sysdev_unregister(&kvm_sysdev);
2703         sysdev_class_unregister(&kvm_sysdev_class);
2704         unregister_reboot_notifier(&kvm_reboot_notifier);
2705         unregister_cpu_notifier(&kvm_cpu_notifier);
2706         on_each_cpu(hardware_disable, NULL, 1);
2707         kvm_arch_hardware_unsetup();
2708         kvm_arch_exit();
2709         kvm_exit_debug();
2710         free_cpumask_var(cpus_hardware_enabled);
2711         __free_page(bad_page);
2712 }
2713 EXPORT_SYMBOL_GPL(kvm_exit);