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
  45 #include <asm/processor.h>
  46 #include <asm/io.h>
  47 #include <asm/uaccess.h>
  48 #include <asm/pgtable.h>
  49
  50 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
  51 #include "coalesced_mmio.h"
  52 #endif
  53
  54 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
  55 #include <linux/pci.h>
  56 #include <linux/interrupt.h>
  57 #include "irq.h"
  58 #endif
  59
  60 MODULE_AUTHOR("Qumranet");
  61 MODULE_LICENSE("GPL");
  62
  63 static int msi2intx = 1;
  64 module_param(msi2intx, bool, 0);
  65
  66 DEFINE_SPINLOCK(kvm_lock);
  67 LIST_HEAD(vm_list);
  68
  69 static cpumask_var_t cpus_hardware_enabled;
  70
  71 struct kmem_cache *kvm_vcpu_cache;
  72 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  73
  74 static __read_mostly struct preempt_ops kvm_preempt_ops;
  75
  76 struct dentry *kvm_debugfs_dir;
  77
  78 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  79                            unsigned long arg);
  80
  81 static bool kvm_rebooting;
  82
  83 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
  84 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
  85                                                       int assigned_dev_id)
  86 {
  87         struct list_head *ptr;
  88         struct kvm_assigned_dev_kernel *match;
  89
  90         list_for_each(ptr, head) {
  91                 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
  92                 if (match->assigned_dev_id == assigned_dev_id)
  93                         return match;
  94         }
  95         return NULL;
  96 }
  97
  98 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
  99 {
 100         struct kvm_assigned_dev_kernel *assigned_dev;
 101
 102         assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
 103                                     interrupt_work);
 104
 105         /* This is taken to safely inject irq inside the guest. When
 106          * the interrupt injection (or the ioapic code) uses a
 107          * finer-grained lock, update this
 108          */
 109         mutex_lock(&assigned_dev->kvm->lock);
 110         kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
 111                     assigned_dev->guest_irq, 1);
 112
 113         if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_GUEST_MSI) {
 114                 enable_irq(assigned_dev->host_irq);
 115                 assigned_dev->host_irq_disabled = false;
 116         }
 117         mutex_unlock(&assigned_dev->kvm->lock);
 118 }
 119
 120 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
 121 {
 122         struct kvm_assigned_dev_kernel *assigned_dev =
 123                 (struct kvm_assigned_dev_kernel *) dev_id;
 124
 125         schedule_work(&assigned_dev->interrupt_work);
 126
 127         disable_irq_nosync(irq);
 128         assigned_dev->host_irq_disabled = true;
 129
 130         return IRQ_HANDLED;
 131 }
 132
 133 /* Ack the irq line for an assigned device */
 134 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
 135 {
 136         struct kvm_assigned_dev_kernel *dev;
 137
 138         if (kian->gsi == -1)
 139                 return;
 140
 141         dev = container_of(kian, struct kvm_assigned_dev_kernel,
 142                            ack_notifier);
 143
 144         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
 145
 146         /* The guest irq may be shared so this ack may be
 147          * from another device.
 148          */
 149         if (dev->host_irq_disabled) {
 150                 enable_irq(dev->host_irq);
 151                 dev->host_irq_disabled = false;
 152         }
 153 }
 154
 155 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
 156 static void kvm_free_assigned_irq(struct kvm *kvm,
 157                                   struct kvm_assigned_dev_kernel *assigned_dev)
 158 {
 159         if (!irqchip_in_kernel(kvm))
 160                 return;
 161
 162         kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
 163
 164         if (assigned_dev->irq_source_id != -1)
 165                 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
 166         assigned_dev->irq_source_id = -1;
 167
 168         if (!assigned_dev->irq_requested_type)
 169                 return;
 170
 171         /*
 172          * In kvm_free_device_irq, cancel_work_sync return true if:
 173          * 1. work is scheduled, and then cancelled.
 174          * 2. work callback is executed.
 175          *
 176          * The first one ensured that the irq is disabled and no more events
 177          * would happen. But for the second one, the irq may be enabled (e.g.
 178          * for MSI). So we disable irq here to prevent further events.
 179          *
 180          * Notice this maybe result in nested disable if the interrupt type is
 181          * INTx, but it's OK for we are going to free it.
 182          *
 183          * If this function is a part of VM destroy, please ensure that till
 184          * now, the kvm state is still legal for probably we also have to wait
 185          * interrupt_work done.
 186          */
 187         disable_irq_nosync(assigned_dev->host_irq);
 188         cancel_work_sync(&assigned_dev->interrupt_work);
 189
 190         free_irq(assigned_dev->host_irq, (void *)assigned_dev);
 191
 192         if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
 193                 pci_disable_msi(assigned_dev->dev);
 194
 195         assigned_dev->irq_requested_type = 0;
 196 }
 197
 198
 199 static void kvm_free_assigned_device(struct kvm *kvm,
 200                                      struct kvm_assigned_dev_kernel
 201                                      *assigned_dev)
 202 {
 203         kvm_free_assigned_irq(kvm, assigned_dev);
 204
 205         pci_reset_function(assigned_dev->dev);
 206
 207         pci_release_regions(assigned_dev->dev);
 208         pci_disable_device(assigned_dev->dev);
 209         pci_dev_put(assigned_dev->dev);
 210
 211         list_del(&assigned_dev->list);
 212         kfree(assigned_dev);
 213 }
 214
 215 void kvm_free_all_assigned_devices(struct kvm *kvm)
 216 {
 217         struct list_head *ptr, *ptr2;
 218         struct kvm_assigned_dev_kernel *assigned_dev;
 219
 220         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
 221                 assigned_dev = list_entry(ptr,
 222                                           struct kvm_assigned_dev_kernel,
 223                                           list);
 224
 225                 kvm_free_assigned_device(kvm, assigned_dev);
 226         }
 227 }
 228
 229 static int assigned_device_update_intx(struct kvm *kvm,
 230                         struct kvm_assigned_dev_kernel *adev,
 231                         struct kvm_assigned_irq *airq)
 232 {
 233         adev->guest_irq = airq->guest_irq;
 234         adev->ack_notifier.gsi = airq->guest_irq;
 235
 236         if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_INTX)
 237                 return 0;
 238
 239         if (irqchip_in_kernel(kvm)) {
 240                 if (!msi2intx &&
 241                     (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)) {
 242                         free_irq(adev->host_irq, (void *)adev);
 243                         pci_disable_msi(adev->dev);
 244                 }
 245
 246                 if (!capable(CAP_SYS_RAWIO))
 247                         return -EPERM;
 248
 249                 if (airq->host_irq)
 250                         adev->host_irq = airq->host_irq;
 251                 else
 252                         adev->host_irq = adev->dev->irq;
 253
 254                 /* Even though this is PCI, we don't want to use shared
 255                  * interrupts. Sharing host devices with guest-assigned devices
 256                  * on the same interrupt line is not a happy situation: there
 257                  * are going to be long delays in accepting, acking, etc.
 258                  */
 259                 if (request_irq(adev->host_irq, kvm_assigned_dev_intr,
 260                                 0, "kvm_assigned_intx_device", (void *)adev))
 261                         return -EIO;
 262         }
 263
 264         adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_INTX |
 265                                    KVM_ASSIGNED_DEV_HOST_INTX;
 266         return 0;
 267 }
 268
 269 #ifdef CONFIG_X86
 270 static int assigned_device_update_msi(struct kvm *kvm,
 271                         struct kvm_assigned_dev_kernel *adev,
 272                         struct kvm_assigned_irq *airq)
 273 {
 274         int r;
 275
 276         adev->guest_irq = airq->guest_irq;
 277         if (airq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI) {
 278                 /* x86 don't care upper address of guest msi message addr */
 279                 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_MSI;
 280                 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_INTX;
 281                 adev->ack_notifier.gsi = -1;
 282         } else if (msi2intx) {
 283                 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_INTX;
 284                 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_MSI;
 285                 adev->ack_notifier.gsi = airq->guest_irq;
 286         } else {
 287                 /*
 288                  * Guest require to disable device MSI, we disable MSI and
 289                  * re-enable INTx by default again. Notice it's only for
 290                  * non-msi2intx.
 291                  */
 292                 assigned_device_update_intx(kvm, adev, airq);
 293                 return 0;
 294         }
 295
 296         if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
 297                 return 0;
 298
 299         if (irqchip_in_kernel(kvm)) {
 300                 if (!msi2intx) {
 301                         if (adev->irq_requested_type &
 302                                         KVM_ASSIGNED_DEV_HOST_INTX)
 303                                 free_irq(adev->host_irq, (void *)adev);
 304
 305                         r = pci_enable_msi(adev->dev);
 306                         if (r)
 307                                 return r;
 308                 }
 309
 310                 adev->host_irq = adev->dev->irq;
 311                 if (request_irq(adev->host_irq, kvm_assigned_dev_intr, 0,
 312                                 "kvm_assigned_msi_device", (void *)adev))
 313                         return -EIO;
 314         }
 315
 316         if (!msi2intx)
 317                 adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_MSI;
 318
 319         adev->irq_requested_type |= KVM_ASSIGNED_DEV_HOST_MSI;
 320         return 0;
 321 }
 322 #endif
 323
 324 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 325                                    struct kvm_assigned_irq
 326                                    *assigned_irq)
 327 {
 328         int r = 0;
 329         struct kvm_assigned_dev_kernel *match;
 330         u32 current_flags = 0, changed_flags;
 331
 332         mutex_lock(&kvm->lock);
 333
 334         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 335                                       assigned_irq->assigned_dev_id);
 336         if (!match) {
 337                 mutex_unlock(&kvm->lock);
 338                 return -EINVAL;
 339         }
 340
 341         if (!match->irq_requested_type) {
 342                 INIT_WORK(&match->interrupt_work,
 343                                 kvm_assigned_dev_interrupt_work_handler);
 344                 if (irqchip_in_kernel(kvm)) {
 345                         /* Register ack nofitier */
 346                         match->ack_notifier.gsi = -1;
 347                         match->ack_notifier.irq_acked =
 348                                         kvm_assigned_dev_ack_irq;
 349                         kvm_register_irq_ack_notifier(kvm,
 350                                         &match->ack_notifier);
 351
 352                         /* Request IRQ source ID */
 353                         r = kvm_request_irq_source_id(kvm);
 354                         if (r < 0)
 355                                 goto out_release;
 356                         else
 357                                 match->irq_source_id = r;
 358
 359 #ifdef CONFIG_X86
 360                         /* Determine host device irq type, we can know the
 361                          * result from dev->msi_enabled */
 362                         if (msi2intx)
 363                                 pci_enable_msi(match->dev);
 364 #endif
 365                 }
 366         }
 367
 368         if ((match->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI) &&
 369                  (match->irq_requested_type & KVM_ASSIGNED_DEV_GUEST_MSI))
 370                 current_flags |= KVM_DEV_IRQ_ASSIGN_ENABLE_MSI;
 371
 372         changed_flags = assigned_irq->flags ^ current_flags;
 373
 374         if ((changed_flags & KVM_DEV_IRQ_ASSIGN_MSI_ACTION) ||
 375             (msi2intx && match->dev->msi_enabled)) {
 376 #ifdef CONFIG_X86
 377                 r = assigned_device_update_msi(kvm, match, assigned_irq);
 378                 if (r) {
 379                         printk(KERN_WARNING "kvm: failed to enable "
 380                                         "MSI device!\n");
 381                         goto out_release;
 382                 }
 383 #else
 384                 r = -ENOTTY;
 385 #endif
 386         } else if (assigned_irq->host_irq == 0 && match->dev->irq == 0) {
 387                 /* Host device IRQ 0 means don't support INTx */
 388                 if (!msi2intx) {
 389                         printk(KERN_WARNING
 390                                "kvm: wait device to enable MSI!\n");
 391                         r = 0;
 392                 } else {
 393                         printk(KERN_WARNING
 394                                "kvm: failed to enable MSI device!\n");
 395                         r = -ENOTTY;
 396                         goto out_release;
 397                 }
 398         } else {
 399                 /* Non-sharing INTx mode */
 400                 r = assigned_device_update_intx(kvm, match, assigned_irq);
 401                 if (r) {
 402                         printk(KERN_WARNING "kvm: failed to enable "
 403                                         "INTx device!\n");
 404                         goto out_release;
 405                 }
 406         }
 407
 408         mutex_unlock(&kvm->lock);
 409         return r;
 410 out_release:
 411         mutex_unlock(&kvm->lock);
 412         kvm_free_assigned_device(kvm, match);
 413         return r;
 414 }
 415
 416 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 417                                       struct kvm_assigned_pci_dev *assigned_dev)
 418 {
 419         int r = 0;
 420         struct kvm_assigned_dev_kernel *match;
 421         struct pci_dev *dev;
 422
 423         down_read(&kvm->slots_lock);
 424         mutex_lock(&kvm->lock);
 425
 426         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 427                                       assigned_dev->assigned_dev_id);
 428         if (match) {
 429                 /* device already assigned */
 430                 r = -EINVAL;
 431                 goto out;
 432         }
 433
 434         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 435         if (match == NULL) {
 436                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 437                        __func__);
 438                 r = -ENOMEM;
 439                 goto out;
 440         }
 441         dev = pci_get_bus_and_slot(assigned_dev->busnr,
 442                                    assigned_dev->devfn);
 443         if (!dev) {
 444                 printk(KERN_INFO "%s: host device not found\n", __func__);
 445                 r = -EINVAL;
 446                 goto out_free;
 447         }
 448         if (pci_enable_device(dev)) {
 449                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 450                 r = -EBUSY;
 451                 goto out_put;
 452         }
 453         r = pci_request_regions(dev, "kvm_assigned_device");
 454         if (r) {
 455                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 456                        __func__);
 457                 goto out_disable;
 458         }
 459
 460         pci_reset_function(dev);
 461
 462         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 463         match->host_busnr = assigned_dev->busnr;
 464         match->host_devfn = assigned_dev->devfn;
 465         match->flags = assigned_dev->flags;
 466         match->dev = dev;
 467         match->irq_source_id = -1;
 468         match->kvm = kvm;
 469
 470         list_add(&match->list, &kvm->arch.assigned_dev_head);
 471
 472         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
 473                 if (!kvm->arch.iommu_domain) {
 474                         r = kvm_iommu_map_guest(kvm);
 475                         if (r)
 476                                 goto out_list_del;
 477                 }
 478                 r = kvm_assign_device(kvm, match);
 479                 if (r)
 480                         goto out_list_del;
 481         }
 482
 483 out:
 484         mutex_unlock(&kvm->lock);
 485         up_read(&kvm->slots_lock);
 486         return r;
 487 out_list_del:
 488         list_del(&match->list);
 489         pci_release_regions(dev);
 490 out_disable:
 491         pci_disable_device(dev);
 492 out_put:
 493         pci_dev_put(dev);
 494 out_free:
 495         kfree(match);
 496         mutex_unlock(&kvm->lock);
 497         up_read(&kvm->slots_lock);
 498         return r;
 499 }
 500 #endif
 501
 502 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
 503 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
 504                 struct kvm_assigned_pci_dev *assigned_dev)
 505 {
 506         int r = 0;
 507         struct kvm_assigned_dev_kernel *match;
 508
 509         mutex_lock(&kvm->lock);
 510
 511         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 512                                       assigned_dev->assigned_dev_id);
 513         if (!match) {
 514                 printk(KERN_INFO "%s: device hasn't been assigned before, "
 515                   "so cannot be deassigned\n", __func__);
 516                 r = -EINVAL;
 517                 goto out;
 518         }
 519
 520         if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
 521                 kvm_deassign_device(kvm, match);
 522
 523         kvm_free_assigned_device(kvm, match);
 524
 525 out:
 526         mutex_unlock(&kvm->lock);
 527         return r;
 528 }
 529 #endif
 530
 531 static inline int valid_vcpu(int n)
 532 {
 533         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 534 }
 535
 536 inline int kvm_is_mmio_pfn(pfn_t pfn)
 537 {
 538         if (pfn_valid(pfn)) {
 539                 struct page *page = compound_head(pfn_to_page(pfn));
 540                 return PageReserved(page);
 541         }
 542
 543         return true;
 544 }
 545
 546 /*
 547  * Switches to specified vcpu, until a matching vcpu_put()
 548  */
 549 void vcpu_load(struct kvm_vcpu *vcpu)
 550 {
 551         int cpu;
 552
 553         mutex_lock(&vcpu->mutex);
 554         cpu = get_cpu();
 555         preempt_notifier_register(&vcpu->preempt_notifier);
 556         kvm_arch_vcpu_load(vcpu, cpu);
 557         put_cpu();
 558 }
 559
 560 void vcpu_put(struct kvm_vcpu *vcpu)
 561 {
 562         preempt_disable();
 563         kvm_arch_vcpu_put(vcpu);
 564         preempt_notifier_unregister(&vcpu->preempt_notifier);
 565         preempt_enable();
 566         mutex_unlock(&vcpu->mutex);
 567 }
 568
 569 static void ack_flush(void *_completed)
 570 {
 571 }
 572
 573 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 574 {
 575         int i, cpu, me;
 576         cpumask_var_t cpus;
 577         bool called = true;
 578         struct kvm_vcpu *vcpu;
 579
 580         if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
 581                 cpumask_clear(cpus);
 582
 583         me = get_cpu();
 584         spin_lock(&kvm->requests_lock);
 585         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 586                 vcpu = kvm->vcpus[i];
 587                 if (!vcpu)
 588                         continue;
 589                 if (test_and_set_bit(req, &vcpu->requests))
 590                         continue;
 591                 cpu = vcpu->cpu;
 592                 if (cpus != NULL && cpu != -1 && cpu != me)
 593                         cpumask_set_cpu(cpu, cpus);
 594         }
 595         if (unlikely(cpus == NULL))
 596                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 597         else if (!cpumask_empty(cpus))
 598                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 599         else
 600                 called = false;
 601         spin_unlock(&kvm->requests_lock);
 602         put_cpu();
 603         free_cpumask_var(cpus);
 604         return called;
 605 }
 606
 607 void kvm_flush_remote_tlbs(struct kvm *kvm)
 608 {
 609         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 610                 ++kvm->stat.remote_tlb_flush;
 611 }
 612
 613 void kvm_reload_remote_mmus(struct kvm *kvm)
 614 {
 615         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 616 }
 617
 618 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 619 {
 620         struct page *page;
 621         int r;
 622
 623         mutex_init(&vcpu->mutex);
 624         vcpu->cpu = -1;
 625         vcpu->kvm = kvm;
 626         vcpu->vcpu_id = id;
 627         init_waitqueue_head(&vcpu->wq);
 628
 629         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 630         if (!page) {
 631                 r = -ENOMEM;
 632                 goto fail;
 633         }
 634         vcpu->run = page_address(page);
 635
 636         r = kvm_arch_vcpu_init(vcpu);
 637         if (r < 0)
 638                 goto fail_free_run;
 639         return 0;
 640
 641 fail_free_run:
 642         free_page((unsigned long)vcpu->run);
 643 fail:
 644         return r;
 645 }
 646 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 647
 648 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 649 {
 650         kvm_arch_vcpu_uninit(vcpu);
 651         free_page((unsigned long)vcpu->run);
 652 }
 653 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 654
 655 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 656 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 657 {
 658         return container_of(mn, struct kvm, mmu_notifier);
 659 }
 660
 661 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 662                                              struct mm_struct *mm,
 663                                              unsigned long address)
 664 {
 665         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 666         int need_tlb_flush;
 667
 668         /*
 669          * When ->invalidate_page runs, the linux pte has been zapped
 670          * already but the page is still allocated until
 671          * ->invalidate_page returns. So if we increase the sequence
 672          * here the kvm page fault will notice if the spte can't be
 673          * established because the page is going to be freed. If
 674          * instead the kvm page fault establishes the spte before
 675          * ->invalidate_page runs, kvm_unmap_hva will release it
 676          * before returning.
 677          *
 678          * The sequence increase only need to be seen at spin_unlock
 679          * time, and not at spin_lock time.
 680          *
 681          * Increasing the sequence after the spin_unlock would be
 682          * unsafe because the kvm page fault could then establish the
 683          * pte after kvm_unmap_hva returned, without noticing the page
 684          * is going to be freed.
 685          */
 686         spin_lock(&kvm->mmu_lock);
 687         kvm->mmu_notifier_seq++;
 688         need_tlb_flush = kvm_unmap_hva(kvm, address);
 689         spin_unlock(&kvm->mmu_lock);
 690
 691         /* we've to flush the tlb before the pages can be freed */
 692         if (need_tlb_flush)
 693                 kvm_flush_remote_tlbs(kvm);
 694
 695 }
 696
 697 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 698                                                     struct mm_struct *mm,
 699                                                     unsigned long start,
 700                                                     unsigned long end)
 701 {
 702         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 703         int need_tlb_flush = 0;
 704
 705         spin_lock(&kvm->mmu_lock);
 706         /*
 707          * The count increase must become visible at unlock time as no
 708          * spte can be established without taking the mmu_lock and
 709          * count is also read inside the mmu_lock critical section.
 710          */
 711         kvm->mmu_notifier_count++;
 712         for (; start < end; start += PAGE_SIZE)
 713                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 714         spin_unlock(&kvm->mmu_lock);
 715
 716         /* we've to flush the tlb before the pages can be freed */
 717         if (need_tlb_flush)
 718                 kvm_flush_remote_tlbs(kvm);
 719 }
 720
 721 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 722                                                   struct mm_struct *mm,
 723                                                   unsigned long start,
 724                                                   unsigned long end)
 725 {
 726         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 727
 728         spin_lock(&kvm->mmu_lock);
 729         /*
 730          * This sequence increase will notify the kvm page fault that
 731          * the page that is going to be mapped in the spte could have
 732          * been freed.
 733          */
 734         kvm->mmu_notifier_seq++;
 735         /*
 736          * The above sequence increase must be visible before the
 737          * below count decrease but both values are read by the kvm
 738          * page fault under mmu_lock spinlock so we don't need to add
 739          * a smb_wmb() here in between the two.
 740          */
 741         kvm->mmu_notifier_count--;
 742         spin_unlock(&kvm->mmu_lock);
 743
 744         BUG_ON(kvm->mmu_notifier_count < 0);
 745 }
 746
 747 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 748                                               struct mm_struct *mm,
 749                                               unsigned long address)
 750 {
 751         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 752         int young;
 753
 754         spin_lock(&kvm->mmu_lock);
 755         young = kvm_age_hva(kvm, address);
 756         spin_unlock(&kvm->mmu_lock);
 757
 758         if (young)
 759                 kvm_flush_remote_tlbs(kvm);
 760
 761         return young;
 762 }
 763
 764 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 765                                      struct mm_struct *mm)
 766 {
 767         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 768         kvm_arch_flush_shadow(kvm);
 769 }
 770
 771 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 772         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 773         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 774         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 775         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 776         .release                = kvm_mmu_notifier_release,
 777 };
 778 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 779
 780 static struct kvm *kvm_create_vm(void)
 781 {
 782         struct kvm *kvm = kvm_arch_create_vm();
 783 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 784         struct page *page;
 785 #endif
 786
 787         if (IS_ERR(kvm))
 788                 goto out;
 789 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 790         INIT_LIST_HEAD(&kvm->irq_routing);
 791         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 792 #endif
 793
 794 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 795         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 796         if (!page) {
 797                 kfree(kvm);
 798                 return ERR_PTR(-ENOMEM);
 799         }
 800         kvm->coalesced_mmio_ring =
 801                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 802 #endif
 803
 804 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 805         {
 806                 int err;
 807                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 808                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 809                 if (err) {
 810 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 811                         put_page(page);
 812 #endif
 813                         kfree(kvm);
 814                         return ERR_PTR(err);
 815                 }
 816         }
 817 #endif
 818
 819         kvm->mm = current->mm;
 820         atomic_inc(&kvm->mm->mm_count);
 821         spin_lock_init(&kvm->mmu_lock);
 822         spin_lock_init(&kvm->requests_lock);
 823         kvm_io_bus_init(&kvm->pio_bus);
 824         mutex_init(&kvm->lock);
 825         kvm_io_bus_init(&kvm->mmio_bus);
 826         init_rwsem(&kvm->slots_lock);
 827         atomic_set(&kvm->users_count, 1);
 828         spin_lock(&kvm_lock);
 829         list_add(&kvm->vm_list, &vm_list);
 830         spin_unlock(&kvm_lock);
 831 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 832         kvm_coalesced_mmio_init(kvm);
 833 #endif
 834 out:
 835         return kvm;
 836 }
 837
 838 /*
 839  * Free any memory in @free but not in @dont.
 840  */
 841 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 842                                   struct kvm_memory_slot *dont)
 843 {
 844         if (!dont || free->rmap != dont->rmap)
 845                 vfree(free->rmap);
 846
 847         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 848                 vfree(free->dirty_bitmap);
 849
 850         if (!dont || free->lpage_info != dont->lpage_info)
 851                 vfree(free->lpage_info);
 852
 853         free->npages = 0;
 854         free->dirty_bitmap = NULL;
 855         free->rmap = NULL;
 856         free->lpage_info = NULL;
 857 }
 858
 859 void kvm_free_physmem(struct kvm *kvm)
 860 {
 861         int i;
 862
 863         for (i = 0; i < kvm->nmemslots; ++i)
 864                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 865 }
 866
 867 static void kvm_destroy_vm(struct kvm *kvm)
 868 {
 869         struct mm_struct *mm = kvm->mm;
 870
 871         kvm_arch_sync_events(kvm);
 872         spin_lock(&kvm_lock);
 873         list_del(&kvm->vm_list);
 874         spin_unlock(&kvm_lock);
 875         kvm_free_irq_routing(kvm);
 876         kvm_io_bus_destroy(&kvm->pio_bus);
 877         kvm_io_bus_destroy(&kvm->mmio_bus);
 878 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 879         if (kvm->coalesced_mmio_ring != NULL)
 880                 free_page((unsigned long)kvm->coalesced_mmio_ring);
 881 #endif
 882 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 883         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 884 #endif
 885         kvm_arch_destroy_vm(kvm);
 886         mmdrop(mm);
 887 }
 888
 889 void kvm_get_kvm(struct kvm *kvm)
 890 {
 891         atomic_inc(&kvm->users_count);
 892 }
 893 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 894
 895 void kvm_put_kvm(struct kvm *kvm)
 896 {
 897         if (atomic_dec_and_test(&kvm->users_count))
 898                 kvm_destroy_vm(kvm);
 899 }
 900 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 901
 902
 903 static int kvm_vm_release(struct inode *inode, struct file *filp)
 904 {
 905         struct kvm *kvm = filp->private_data;
 906
 907         kvm_put_kvm(kvm);
 908         return 0;
 909 }
 910
 911 /*
 912  * Allocate some memory and give it an address in the guest physical address
 913  * space.
 914  *
 915  * Discontiguous memory is allowed, mostly for framebuffers.
 916  *
 917  * Must be called holding mmap_sem for write.
 918  */
 919 int __kvm_set_memory_region(struct kvm *kvm,
 920                             struct kvm_userspace_memory_region *mem,
 921                             int user_alloc)
 922 {
 923         int r;
 924         gfn_t base_gfn;
 925         unsigned long npages, ugfn;
 926         unsigned long largepages, i;
 927         struct kvm_memory_slot *memslot;
 928         struct kvm_memory_slot old, new;
 929
 930         r = -EINVAL;
 931         /* General sanity checks */
 932         if (mem->memory_size & (PAGE_SIZE - 1))
 933                 goto out;
 934         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 935                 goto out;
 936         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 937                 goto out;
 938         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 939                 goto out;
 940         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 941                 goto out;
 942
 943         memslot = &kvm->memslots[mem->slot];
 944         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 945         npages = mem->memory_size >> PAGE_SHIFT;
 946
 947         if (!npages)
 948                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 949
 950         new = old = *memslot;
 951
 952         new.base_gfn = base_gfn;
 953         new.npages = npages;
 954         new.flags = mem->flags;
 955
 956         /* Disallow changing a memory slot's size. */
 957         r = -EINVAL;
 958         if (npages && old.npages && npages != old.npages)
 959                 goto out_free;
 960
 961         /* Check for overlaps */
 962         r = -EEXIST;
 963         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 964                 struct kvm_memory_slot *s = &kvm->memslots[i];
 965
 966                 if (s == memslot || !s->npages)
 967                         continue;
 968                 if (!((base_gfn + npages <= s->base_gfn) ||
 969                       (base_gfn >= s->base_gfn + s->npages)))
 970                         goto out_free;
 971         }
 972
 973         /* Free page dirty bitmap if unneeded */
 974         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 975                 new.dirty_bitmap = NULL;
 976
 977         r = -ENOMEM;
 978
 979         /* Allocate if a slot is being created */
 980 #ifndef CONFIG_S390
 981         if (npages && !new.rmap) {
 982                 new.rmap = vmalloc(npages * sizeof(struct page *));
 983
 984                 if (!new.rmap)
 985                         goto out_free;
 986
 987                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 988
 989                 new.user_alloc = user_alloc;
 990                 /*
 991                  * hva_to_rmmap() serialzies with the mmu_lock and to be
 992                  * safe it has to ignore memslots with !user_alloc &&
 993                  * !userspace_addr.
 994                  */
 995                 if (user_alloc)
 996                         new.userspace_addr = mem->userspace_addr;
 997                 else
 998                         new.userspace_addr = 0;
 999         }
1000         if (npages && !new.lpage_info) {
1001                 largepages = 1 + (base_gfn + npages - 1) / KVM_PAGES_PER_HPAGE;
1002                 largepages -= base_gfn / KVM_PAGES_PER_HPAGE;
1003
1004                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1005
1006                 if (!new.lpage_info)
1007                         goto out_free;
1008
1009                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1010
1011                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1012                         new.lpage_info[0].write_count = 1;
1013                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1014                         new.lpage_info[largepages-1].write_count = 1;
1015                 ugfn = new.userspace_addr >> PAGE_SHIFT;
1016                 /*
1017                  * If the gfn and userspace address are not aligned wrt each
1018                  * other, disable large page support for this slot
1019                  */
1020                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE - 1))
1021                         for (i = 0; i < largepages; ++i)
1022                                 new.lpage_info[i].write_count = 1;
1023         }
1024
1025         /* Allocate page dirty bitmap if needed */
1026         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1027                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1028
1029                 new.dirty_bitmap = vmalloc(dirty_bytes);
1030                 if (!new.dirty_bitmap)
1031                         goto out_free;
1032                 memset(new.dirty_bitmap, 0, dirty_bytes);
1033                 if (old.npages)
1034                         kvm_arch_flush_shadow(kvm);
1035         }
1036 #endif /* not defined CONFIG_S390 */
1037
1038         if (!npages)
1039                 kvm_arch_flush_shadow(kvm);
1040
1041         spin_lock(&kvm->mmu_lock);
1042         if (mem->slot >= kvm->nmemslots)
1043                 kvm->nmemslots = mem->slot + 1;
1044
1045         *memslot = new;
1046         spin_unlock(&kvm->mmu_lock);
1047
1048         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1049         if (r) {
1050                 spin_lock(&kvm->mmu_lock);
1051                 *memslot = old;
1052                 spin_unlock(&kvm->mmu_lock);
1053                 goto out_free;
1054         }
1055
1056         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1057         /* Slot deletion case: we have to update the current slot */
1058         if (!npages)
1059                 *memslot = old;
1060 #ifdef CONFIG_DMAR
1061         /* map the pages in iommu page table */
1062         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1063         if (r)
1064                 goto out;
1065 #endif
1066         return 0;
1067
1068 out_free:
1069         kvm_free_physmem_slot(&new, &old);
1070 out:
1071         return r;
1072
1073 }
1074 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1075
1076 int kvm_set_memory_region(struct kvm *kvm,
1077                           struct kvm_userspace_memory_region *mem,
1078                           int user_alloc)
1079 {
1080         int r;
1081
1082         down_write(&kvm->slots_lock);
1083         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1084         up_write(&kvm->slots_lock);
1085         return r;
1086 }
1087 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1088
1089 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1090                                    struct
1091                                    kvm_userspace_memory_region *mem,
1092                                    int user_alloc)
1093 {
1094         if (mem->slot >= KVM_MEMORY_SLOTS)
1095                 return -EINVAL;
1096         return kvm_set_memory_region(kvm, mem, user_alloc);
1097 }
1098
1099 int kvm_get_dirty_log(struct kvm *kvm,
1100                         struct kvm_dirty_log *log, int *is_dirty)
1101 {
1102         struct kvm_memory_slot *memslot;
1103         int r, i;
1104         int n;
1105         unsigned long any = 0;
1106
1107         r = -EINVAL;
1108         if (log->slot >= KVM_MEMORY_SLOTS)
1109                 goto out;
1110
1111         memslot = &kvm->memslots[log->slot];
1112         r = -ENOENT;
1113         if (!memslot->dirty_bitmap)
1114                 goto out;
1115
1116         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1117
1118         for (i = 0; !any && i < n/sizeof(long); ++i)
1119                 any = memslot->dirty_bitmap[i];
1120
1121         r = -EFAULT;
1122         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1123                 goto out;
1124
1125         if (any)
1126                 *is_dirty = 1;
1127
1128         r = 0;
1129 out:
1130         return r;
1131 }
1132
1133 int is_error_page(struct page *page)
1134 {
1135         return page == bad_page;
1136 }
1137 EXPORT_SYMBOL_GPL(is_error_page);
1138
1139 int is_error_pfn(pfn_t pfn)
1140 {
1141         return pfn == bad_pfn;
1142 }
1143 EXPORT_SYMBOL_GPL(is_error_pfn);
1144
1145 static inline unsigned long bad_hva(void)
1146 {
1147         return PAGE_OFFSET;
1148 }
1149
1150 int kvm_is_error_hva(unsigned long addr)
1151 {
1152         return addr == bad_hva();
1153 }
1154 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1155
1156 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1157 {
1158         int i;
1159
1160         for (i = 0; i < kvm->nmemslots; ++i) {
1161                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1162
1163                 if (gfn >= memslot->base_gfn
1164                     && gfn < memslot->base_gfn + memslot->npages)
1165                         return memslot;
1166         }
1167         return NULL;
1168 }
1169 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1170
1171 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1172 {
1173         gfn = unalias_gfn(kvm, gfn);
1174         return gfn_to_memslot_unaliased(kvm, gfn);
1175 }
1176
1177 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1178 {
1179         int i;
1180
1181         gfn = unalias_gfn(kvm, gfn);
1182         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1183                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1184
1185                 if (gfn >= memslot->base_gfn
1186                     && gfn < memslot->base_gfn + memslot->npages)
1187                         return 1;
1188         }
1189         return 0;
1190 }
1191 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1192
1193 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1194 {
1195         struct kvm_memory_slot *slot;
1196
1197         gfn = unalias_gfn(kvm, gfn);
1198         slot = gfn_to_memslot_unaliased(kvm, gfn);
1199         if (!slot)
1200                 return bad_hva();
1201         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1202 }
1203 EXPORT_SYMBOL_GPL(gfn_to_hva);
1204
1205 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1206 {
1207         struct page *page[1];
1208         unsigned long addr;
1209         int npages;
1210         pfn_t pfn;
1211
1212         might_sleep();
1213
1214         addr = gfn_to_hva(kvm, gfn);
1215         if (kvm_is_error_hva(addr)) {
1216                 get_page(bad_page);
1217                 return page_to_pfn(bad_page);
1218         }
1219
1220         npages = get_user_pages_fast(addr, 1, 1, page);
1221
1222         if (unlikely(npages != 1)) {
1223                 struct vm_area_struct *vma;
1224
1225                 down_read(&current->mm->mmap_sem);
1226                 vma = find_vma(current->mm, addr);
1227
1228                 if (vma == NULL || addr < vma->vm_start ||
1229                     !(vma->vm_flags & VM_PFNMAP)) {
1230                         up_read(&current->mm->mmap_sem);
1231                         get_page(bad_page);
1232                         return page_to_pfn(bad_page);
1233                 }
1234
1235                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1236                 up_read(&current->mm->mmap_sem);
1237                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1238         } else
1239                 pfn = page_to_pfn(page[0]);
1240
1241         return pfn;
1242 }
1243
1244 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1245
1246 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1247 {
1248         pfn_t pfn;
1249
1250         pfn = gfn_to_pfn(kvm, gfn);
1251         if (!kvm_is_mmio_pfn(pfn))
1252                 return pfn_to_page(pfn);
1253
1254         WARN_ON(kvm_is_mmio_pfn(pfn));
1255
1256         get_page(bad_page);
1257         return bad_page;
1258 }
1259
1260 EXPORT_SYMBOL_GPL(gfn_to_page);
1261
1262 void kvm_release_page_clean(struct page *page)
1263 {
1264         kvm_release_pfn_clean(page_to_pfn(page));
1265 }
1266 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1267
1268 void kvm_release_pfn_clean(pfn_t pfn)
1269 {
1270         if (!kvm_is_mmio_pfn(pfn))
1271                 put_page(pfn_to_page(pfn));
1272 }
1273 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1274
1275 void kvm_release_page_dirty(struct page *page)
1276 {
1277         kvm_release_pfn_dirty(page_to_pfn(page));
1278 }
1279 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1280
1281 void kvm_release_pfn_dirty(pfn_t pfn)
1282 {
1283         kvm_set_pfn_dirty(pfn);
1284         kvm_release_pfn_clean(pfn);
1285 }
1286 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1287
1288 void kvm_set_page_dirty(struct page *page)
1289 {
1290         kvm_set_pfn_dirty(page_to_pfn(page));
1291 }
1292 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1293
1294 void kvm_set_pfn_dirty(pfn_t pfn)
1295 {
1296         if (!kvm_is_mmio_pfn(pfn)) {
1297                 struct page *page = pfn_to_page(pfn);
1298                 if (!PageReserved(page))
1299                         SetPageDirty(page);
1300         }
1301 }
1302 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1303
1304 void kvm_set_pfn_accessed(pfn_t pfn)
1305 {
1306         if (!kvm_is_mmio_pfn(pfn))
1307                 mark_page_accessed(pfn_to_page(pfn));
1308 }
1309 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1310
1311 void kvm_get_pfn(pfn_t pfn)
1312 {
1313         if (!kvm_is_mmio_pfn(pfn))
1314                 get_page(pfn_to_page(pfn));
1315 }
1316 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1317
1318 static int next_segment(unsigned long len, int offset)
1319 {
1320         if (len > PAGE_SIZE - offset)
1321                 return PAGE_SIZE - offset;
1322         else
1323                 return len;
1324 }
1325
1326 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1327                         int len)
1328 {
1329         int r;
1330         unsigned long addr;
1331
1332         addr = gfn_to_hva(kvm, gfn);
1333         if (kvm_is_error_hva(addr))
1334                 return -EFAULT;
1335         r = copy_from_user(data, (void __user *)addr + offset, len);
1336         if (r)
1337                 return -EFAULT;
1338         return 0;
1339 }
1340 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1341
1342 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1343 {
1344         gfn_t gfn = gpa >> PAGE_SHIFT;
1345         int seg;
1346         int offset = offset_in_page(gpa);
1347         int ret;
1348
1349         while ((seg = next_segment(len, offset)) != 0) {
1350                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1351                 if (ret < 0)
1352                         return ret;
1353                 offset = 0;
1354                 len -= seg;
1355                 data += seg;
1356                 ++gfn;
1357         }
1358         return 0;
1359 }
1360 EXPORT_SYMBOL_GPL(kvm_read_guest);
1361
1362 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1363                           unsigned long len)
1364 {
1365         int r;
1366         unsigned long addr;
1367         gfn_t gfn = gpa >> PAGE_SHIFT;
1368         int offset = offset_in_page(gpa);
1369
1370         addr = gfn_to_hva(kvm, gfn);
1371         if (kvm_is_error_hva(addr))
1372                 return -EFAULT;
1373         pagefault_disable();
1374         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1375         pagefault_enable();
1376         if (r)
1377                 return -EFAULT;
1378         return 0;
1379 }
1380 EXPORT_SYMBOL(kvm_read_guest_atomic);
1381
1382 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1383                          int offset, int len)
1384 {
1385         int r;
1386         unsigned long addr;
1387
1388         addr = gfn_to_hva(kvm, gfn);
1389         if (kvm_is_error_hva(addr))
1390                 return -EFAULT;
1391         r = copy_to_user((void __user *)addr + offset, data, len);
1392         if (r)
1393                 return -EFAULT;
1394         mark_page_dirty(kvm, gfn);
1395         return 0;
1396 }
1397 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1398
1399 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1400                     unsigned long len)
1401 {
1402         gfn_t gfn = gpa >> PAGE_SHIFT;
1403         int seg;
1404         int offset = offset_in_page(gpa);
1405         int ret;
1406
1407         while ((seg = next_segment(len, offset)) != 0) {
1408                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1409                 if (ret < 0)
1410                         return ret;
1411                 offset = 0;
1412                 len -= seg;
1413                 data += seg;
1414                 ++gfn;
1415         }
1416         return 0;
1417 }
1418
1419 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1420 {
1421         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1422 }
1423 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1424
1425 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1426 {
1427         gfn_t gfn = gpa >> PAGE_SHIFT;
1428         int seg;
1429         int offset = offset_in_page(gpa);
1430         int ret;
1431
1432         while ((seg = next_segment(len, offset)) != 0) {
1433                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1434                 if (ret < 0)
1435                         return ret;
1436                 offset = 0;
1437                 len -= seg;
1438                 ++gfn;
1439         }
1440         return 0;
1441 }
1442 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1443
1444 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1445 {
1446         struct kvm_memory_slot *memslot;
1447
1448         gfn = unalias_gfn(kvm, gfn);
1449         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1450         if (memslot && memslot->dirty_bitmap) {
1451                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1452
1453                 /* avoid RMW */
1454                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1455                         set_bit(rel_gfn, memslot->dirty_bitmap);
1456         }
1457 }
1458
1459 /*
1460  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1461  */
1462 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1463 {
1464         DEFINE_WAIT(wait);
1465
1466         for (;;) {
1467                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1468
1469                 if (kvm_cpu_has_interrupt(vcpu) ||
1470                     kvm_cpu_has_pending_timer(vcpu) ||
1471                     kvm_arch_vcpu_runnable(vcpu)) {
1472                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1473                         break;
1474                 }
1475                 if (signal_pending(current))
1476                         break;
1477
1478                 vcpu_put(vcpu);
1479                 schedule();
1480                 vcpu_load(vcpu);
1481         }
1482
1483         finish_wait(&vcpu->wq, &wait);
1484 }
1485
1486 void kvm_resched(struct kvm_vcpu *vcpu)
1487 {
1488         if (!need_resched())
1489                 return;
1490         cond_resched();
1491 }
1492 EXPORT_SYMBOL_GPL(kvm_resched);
1493
1494 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1495 {
1496         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1497         struct page *page;
1498
1499         if (vmf->pgoff == 0)
1500                 page = virt_to_page(vcpu->run);
1501 #ifdef CONFIG_X86
1502         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1503                 page = virt_to_page(vcpu->arch.pio_data);
1504 #endif
1505 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1506         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1507                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1508 #endif
1509         else
1510                 return VM_FAULT_SIGBUS;
1511         get_page(page);
1512         vmf->page = page;
1513         return 0;
1514 }
1515
1516 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1517         .fault = kvm_vcpu_fault,
1518 };
1519
1520 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1521 {
1522         vma->vm_ops = &kvm_vcpu_vm_ops;
1523         return 0;
1524 }
1525
1526 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1527 {
1528         struct kvm_vcpu *vcpu = filp->private_data;
1529
1530         kvm_put_kvm(vcpu->kvm);
1531         return 0;
1532 }
1533
1534 static struct file_operations kvm_vcpu_fops = {
1535         .release        = kvm_vcpu_release,
1536         .unlocked_ioctl = kvm_vcpu_ioctl,
1537         .compat_ioctl   = kvm_vcpu_ioctl,
1538         .mmap           = kvm_vcpu_mmap,
1539 };
1540
1541 /*
1542  * Allocates an inode for the vcpu.
1543  */
1544 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1545 {
1546         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1547         if (fd < 0)
1548                 kvm_put_kvm(vcpu->kvm);
1549         return fd;
1550 }
1551
1552 /*
1553  * Creates some virtual cpus.  Good luck creating more than one.
1554  */
1555 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1556 {
1557         int r;
1558         struct kvm_vcpu *vcpu;
1559
1560         if (!valid_vcpu(n))
1561                 return -EINVAL;
1562
1563         vcpu = kvm_arch_vcpu_create(kvm, n);
1564         if (IS_ERR(vcpu))
1565                 return PTR_ERR(vcpu);
1566
1567         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1568
1569         r = kvm_arch_vcpu_setup(vcpu);
1570         if (r)
1571                 return r;
1572
1573         mutex_lock(&kvm->lock);
1574         if (kvm->vcpus[n]) {
1575                 r = -EEXIST;
1576                 goto vcpu_destroy;
1577         }
1578         kvm->vcpus[n] = vcpu;
1579         mutex_unlock(&kvm->lock);
1580
1581         /* Now it's all set up, let userspace reach it */
1582         kvm_get_kvm(kvm);
1583         r = create_vcpu_fd(vcpu);
1584         if (r < 0)
1585                 goto unlink;
1586         return r;
1587
1588 unlink:
1589         mutex_lock(&kvm->lock);
1590         kvm->vcpus[n] = NULL;
1591 vcpu_destroy:
1592         mutex_unlock(&kvm->lock);
1593         kvm_arch_vcpu_destroy(vcpu);
1594         return r;
1595 }
1596
1597 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1598 {
1599         if (sigset) {
1600                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1601                 vcpu->sigset_active = 1;
1602                 vcpu->sigset = *sigset;
1603         } else
1604                 vcpu->sigset_active = 0;
1605         return 0;
1606 }
1607
1608 static long kvm_vcpu_ioctl(struct file *filp,
1609                            unsigned int ioctl, unsigned long arg)
1610 {
1611         struct kvm_vcpu *vcpu = filp->private_data;
1612         void __user *argp = (void __user *)arg;
1613         int r;
1614         struct kvm_fpu *fpu = NULL;
1615         struct kvm_sregs *kvm_sregs = NULL;
1616
1617         if (vcpu->kvm->mm != current->mm)
1618                 return -EIO;
1619         switch (ioctl) {
1620         case KVM_RUN:
1621                 r = -EINVAL;
1622                 if (arg)
1623                         goto out;
1624                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1625                 break;
1626         case KVM_GET_REGS: {
1627                 struct kvm_regs *kvm_regs;
1628
1629                 r = -ENOMEM;
1630                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1631                 if (!kvm_regs)
1632                         goto out;
1633                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1634                 if (r)
1635                         goto out_free1;
1636                 r = -EFAULT;
1637                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1638                         goto out_free1;
1639                 r = 0;
1640 out_free1:
1641                 kfree(kvm_regs);
1642                 break;
1643         }
1644         case KVM_SET_REGS: {
1645                 struct kvm_regs *kvm_regs;
1646
1647                 r = -ENOMEM;
1648                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1649                 if (!kvm_regs)
1650                         goto out;
1651                 r = -EFAULT;
1652                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1653                         goto out_free2;
1654                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1655                 if (r)
1656                         goto out_free2;
1657                 r = 0;
1658 out_free2:
1659                 kfree(kvm_regs);
1660                 break;
1661         }
1662         case KVM_GET_SREGS: {
1663                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1664                 r = -ENOMEM;
1665                 if (!kvm_sregs)
1666                         goto out;
1667                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1668                 if (r)
1669                         goto out;
1670                 r = -EFAULT;
1671                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1672                         goto out;
1673                 r = 0;
1674                 break;
1675         }
1676         case KVM_SET_SREGS: {
1677                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1678                 r = -ENOMEM;
1679                 if (!kvm_sregs)
1680                         goto out;
1681                 r = -EFAULT;
1682                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1683                         goto out;
1684                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1685                 if (r)
1686                         goto out;
1687                 r = 0;
1688                 break;
1689         }
1690         case KVM_GET_MP_STATE: {
1691                 struct kvm_mp_state mp_state;
1692
1693                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1694                 if (r)
1695                         goto out;
1696                 r = -EFAULT;
1697                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1698                         goto out;
1699                 r = 0;
1700                 break;
1701         }
1702         case KVM_SET_MP_STATE: {
1703                 struct kvm_mp_state mp_state;
1704
1705                 r = -EFAULT;
1706                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1707                         goto out;
1708                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1709                 if (r)
1710                         goto out;
1711                 r = 0;
1712                 break;
1713         }
1714         case KVM_TRANSLATE: {
1715                 struct kvm_translation tr;
1716
1717                 r = -EFAULT;
1718                 if (copy_from_user(&tr, argp, sizeof tr))
1719                         goto out;
1720                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1721                 if (r)
1722                         goto out;
1723                 r = -EFAULT;
1724                 if (copy_to_user(argp, &tr, sizeof tr))
1725                         goto out;
1726                 r = 0;
1727                 break;
1728         }
1729         case KVM_SET_GUEST_DEBUG: {
1730                 struct kvm_guest_debug dbg;
1731
1732                 r = -EFAULT;
1733                 if (copy_from_user(&dbg, argp, sizeof dbg))
1734                         goto out;
1735                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1736                 if (r)
1737                         goto out;
1738                 r = 0;
1739                 break;
1740         }
1741         case KVM_SET_SIGNAL_MASK: {
1742                 struct kvm_signal_mask __user *sigmask_arg = argp;
1743                 struct kvm_signal_mask kvm_sigmask;
1744                 sigset_t sigset, *p;
1745
1746                 p = NULL;
1747                 if (argp) {
1748                         r = -EFAULT;
1749                         if (copy_from_user(&kvm_sigmask, argp,
1750                                            sizeof kvm_sigmask))
1751                                 goto out;
1752                         r = -EINVAL;
1753                         if (kvm_sigmask.len != sizeof sigset)
1754                                 goto out;
1755                         r = -EFAULT;
1756                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1757                                            sizeof sigset))
1758                                 goto out;
1759                         p = &sigset;
1760                 }
1761                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1762                 break;
1763         }
1764         case KVM_GET_FPU: {
1765                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1766                 r = -ENOMEM;
1767                 if (!fpu)
1768                         goto out;
1769                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1770                 if (r)
1771                         goto out;
1772                 r = -EFAULT;
1773                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1774                         goto out;
1775                 r = 0;
1776                 break;
1777         }
1778         case KVM_SET_FPU: {
1779                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1780                 r = -ENOMEM;
1781                 if (!fpu)
1782                         goto out;
1783                 r = -EFAULT;
1784                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1785                         goto out;
1786                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1787                 if (r)
1788                         goto out;
1789                 r = 0;
1790                 break;
1791         }
1792         default:
1793                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1794         }
1795 out:
1796         kfree(fpu);
1797         kfree(kvm_sregs);
1798         return r;
1799 }
1800
1801 static long kvm_vm_ioctl(struct file *filp,
1802                            unsigned int ioctl, unsigned long arg)
1803 {
1804         struct kvm *kvm = filp->private_data;
1805         void __user *argp = (void __user *)arg;
1806         int r;
1807
1808         if (kvm->mm != current->mm)
1809                 return -EIO;
1810         switch (ioctl) {
1811         case KVM_CREATE_VCPU:
1812                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1813                 if (r < 0)
1814                         goto out;
1815                 break;
1816         case KVM_SET_USER_MEMORY_REGION: {
1817                 struct kvm_userspace_memory_region kvm_userspace_mem;
1818
1819                 r = -EFAULT;
1820                 if (copy_from_user(&kvm_userspace_mem, argp,
1821                                                 sizeof kvm_userspace_mem))
1822                         goto out;
1823
1824                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1825                 if (r)
1826                         goto out;
1827                 break;
1828         }
1829         case KVM_GET_DIRTY_LOG: {
1830                 struct kvm_dirty_log log;
1831
1832                 r = -EFAULT;
1833                 if (copy_from_user(&log, argp, sizeof log))
1834                         goto out;
1835                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1836                 if (r)
1837                         goto out;
1838                 break;
1839         }
1840 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1841         case KVM_REGISTER_COALESCED_MMIO: {
1842                 struct kvm_coalesced_mmio_zone zone;
1843                 r = -EFAULT;
1844                 if (copy_from_user(&zone, argp, sizeof zone))
1845                         goto out;
1846                 r = -ENXIO;
1847                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1848                 if (r)
1849                         goto out;
1850                 r = 0;
1851                 break;
1852         }
1853         case KVM_UNREGISTER_COALESCED_MMIO: {
1854                 struct kvm_coalesced_mmio_zone zone;
1855                 r = -EFAULT;
1856                 if (copy_from_user(&zone, argp, sizeof zone))
1857                         goto out;
1858                 r = -ENXIO;
1859                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1860                 if (r)
1861                         goto out;
1862                 r = 0;
1863                 break;
1864         }
1865 #endif
1866 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1867         case KVM_ASSIGN_PCI_DEVICE: {
1868                 struct kvm_assigned_pci_dev assigned_dev;
1869
1870                 r = -EFAULT;
1871                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1872                         goto out;
1873                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1874                 if (r)
1875                         goto out;
1876                 break;
1877         }
1878         case KVM_ASSIGN_IRQ: {
1879                 struct kvm_assigned_irq assigned_irq;
1880
1881                 r = -EFAULT;
1882                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1883                         goto out;
1884                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1885                 if (r)
1886                         goto out;
1887                 break;
1888         }
1889 #endif
1890 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
1891         case KVM_DEASSIGN_PCI_DEVICE: {
1892                 struct kvm_assigned_pci_dev assigned_dev;
1893
1894                 r = -EFAULT;
1895                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1896                         goto out;
1897                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
1898                 if (r)
1899                         goto out;
1900                 break;
1901         }
1902 #endif
1903 #ifdef KVM_CAP_IRQ_ROUTING
1904         case KVM_SET_GSI_ROUTING: {
1905                 struct kvm_irq_routing routing;
1906                 struct kvm_irq_routing __user *urouting;
1907                 struct kvm_irq_routing_entry *entries;
1908
1909                 r = -EFAULT;
1910                 if (copy_from_user(&routing, argp, sizeof(routing)))
1911                         goto out;
1912                 r = -EINVAL;
1913                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
1914                         goto out;
1915                 if (routing.flags)
1916                         goto out;
1917                 r = -ENOMEM;
1918                 entries = vmalloc(routing.nr * sizeof(*entries));
1919                 if (!entries)
1920                         goto out;
1921                 r = -EFAULT;
1922                 urouting = argp;
1923                 if (copy_from_user(entries, urouting->entries,
1924                                    routing.nr * sizeof(*entries)))
1925                         goto out_free_irq_routing;
1926                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
1927                                         routing.flags);
1928         out_free_irq_routing:
1929                 vfree(entries);
1930                 break;
1931         }
1932 #endif
1933         default:
1934                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1935         }
1936 out:
1937         return r;
1938 }
1939
1940 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1941 {
1942         struct page *page[1];
1943         unsigned long addr;
1944         int npages;
1945         gfn_t gfn = vmf->pgoff;
1946         struct kvm *kvm = vma->vm_file->private_data;
1947
1948         addr = gfn_to_hva(kvm, gfn);
1949         if (kvm_is_error_hva(addr))
1950                 return VM_FAULT_SIGBUS;
1951
1952         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1953                                 NULL);
1954         if (unlikely(npages != 1))
1955                 return VM_FAULT_SIGBUS;
1956
1957         vmf->page = page[0];
1958         return 0;
1959 }
1960
1961 static struct vm_operations_struct kvm_vm_vm_ops = {
1962         .fault = kvm_vm_fault,
1963 };
1964
1965 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1966 {
1967         vma->vm_ops = &kvm_vm_vm_ops;
1968         return 0;
1969 }
1970
1971 static struct file_operations kvm_vm_fops = {
1972         .release        = kvm_vm_release,
1973         .unlocked_ioctl = kvm_vm_ioctl,
1974         .compat_ioctl   = kvm_vm_ioctl,
1975         .mmap           = kvm_vm_mmap,
1976 };
1977
1978 static int kvm_dev_ioctl_create_vm(void)
1979 {
1980         int fd;
1981         struct kvm *kvm;
1982
1983         kvm = kvm_create_vm();
1984         if (IS_ERR(kvm))
1985                 return PTR_ERR(kvm);
1986         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1987         if (fd < 0)
1988                 kvm_put_kvm(kvm);
1989
1990         return fd;
1991 }
1992
1993 static long kvm_dev_ioctl_check_extension_generic(long arg)
1994 {
1995         switch (arg) {
1996         case KVM_CAP_USER_MEMORY:
1997         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1998         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1999                 return 1;
2000 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2001         case KVM_CAP_IRQ_ROUTING:
2002                 return KVM_MAX_IRQ_ROUTES;
2003 #endif
2004         default:
2005                 break;
2006         }
2007         return kvm_dev_ioctl_check_extension(arg);
2008 }
2009
2010 static long kvm_dev_ioctl(struct file *filp,
2011                           unsigned int ioctl, unsigned long arg)
2012 {
2013         long r = -EINVAL;
2014
2015         switch (ioctl) {
2016         case KVM_GET_API_VERSION:
2017                 r = -EINVAL;
2018                 if (arg)
2019                         goto out;
2020                 r = KVM_API_VERSION;
2021                 break;
2022         case KVM_CREATE_VM:
2023                 r = -EINVAL;
2024                 if (arg)
2025                         goto out;
2026                 r = kvm_dev_ioctl_create_vm();
2027                 break;
2028         case KVM_CHECK_EXTENSION:
2029                 r = kvm_dev_ioctl_check_extension_generic(arg);
2030                 break;
2031         case KVM_GET_VCPU_MMAP_SIZE:
2032                 r = -EINVAL;
2033                 if (arg)
2034                         goto out;
2035                 r = PAGE_SIZE;     /* struct kvm_run */
2036 #ifdef CONFIG_X86
2037                 r += PAGE_SIZE;    /* pio data page */
2038 #endif
2039 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2040                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2041 #endif
2042                 break;
2043         case KVM_TRACE_ENABLE:
2044         case KVM_TRACE_PAUSE:
2045         case KVM_TRACE_DISABLE:
2046                 r = kvm_trace_ioctl(ioctl, arg);
2047                 break;
2048         default:
2049                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2050         }
2051 out:
2052         return r;
2053 }
2054
2055 static struct file_operations kvm_chardev_ops = {
2056         .unlocked_ioctl = kvm_dev_ioctl,
2057         .compat_ioctl   = kvm_dev_ioctl,
2058 };
2059
2060 static struct miscdevice kvm_dev = {
2061         KVM_MINOR,
2062         "kvm",
2063         &kvm_chardev_ops,
2064 };
2065
2066 static void hardware_enable(void *junk)
2067 {
2068         int cpu = raw_smp_processor_id();
2069
2070         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2071                 return;
2072         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2073         kvm_arch_hardware_enable(NULL);
2074 }
2075
2076 static void hardware_disable(void *junk)
2077 {
2078         int cpu = raw_smp_processor_id();
2079
2080         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2081                 return;
2082         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2083         kvm_arch_hardware_disable(NULL);
2084 }
2085
2086 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2087                            void *v)
2088 {
2089         int cpu = (long)v;
2090
2091         val &= ~CPU_TASKS_FROZEN;
2092         switch (val) {
2093         case CPU_DYING:
2094                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2095                        cpu);
2096                 hardware_disable(NULL);
2097                 break;
2098         case CPU_UP_CANCELED:
2099                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2100                        cpu);
2101                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2102                 break;
2103         case CPU_ONLINE:
2104                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2105                        cpu);
2106                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2107                 break;
2108         }
2109         return NOTIFY_OK;
2110 }
2111
2112
2113 asmlinkage void kvm_handle_fault_on_reboot(void)
2114 {
2115         if (kvm_rebooting)
2116                 /* spin while reset goes on */
2117                 while (true)
2118                         ;
2119         /* Fault while not rebooting.  We want the trace. */
2120         BUG();
2121 }
2122 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2123
2124 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2125                       void *v)
2126 {
2127         if (val == SYS_RESTART) {
2128                 /*
2129                  * Some (well, at least mine) BIOSes hang on reboot if
2130                  * in vmx root mode.
2131                  */
2132                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2133                 kvm_rebooting = true;
2134                 on_each_cpu(hardware_disable, NULL, 1);
2135         }
2136         return NOTIFY_OK;
2137 }
2138
2139 static struct notifier_block kvm_reboot_notifier = {
2140         .notifier_call = kvm_reboot,
2141         .priority = 0,
2142 };
2143
2144 void kvm_io_bus_init(struct kvm_io_bus *bus)
2145 {
2146         memset(bus, 0, sizeof(*bus));
2147 }
2148
2149 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2150 {
2151         int i;
2152
2153         for (i = 0; i < bus->dev_count; i++) {
2154                 struct kvm_io_device *pos = bus->devs[i];
2155
2156                 kvm_iodevice_destructor(pos);
2157         }
2158 }
2159
2160 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2161                                           gpa_t addr, int len, int is_write)
2162 {
2163         int i;
2164
2165         for (i = 0; i < bus->dev_count; i++) {
2166                 struct kvm_io_device *pos = bus->devs[i];
2167
2168                 if (pos->in_range(pos, addr, len, is_write))
2169                         return pos;
2170         }
2171
2172         return NULL;
2173 }
2174
2175 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2176 {
2177         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2178
2179         bus->devs[bus->dev_count++] = dev;
2180 }
2181
2182 static struct notifier_block kvm_cpu_notifier = {
2183         .notifier_call = kvm_cpu_hotplug,
2184         .priority = 20, /* must be > scheduler priority */
2185 };
2186
2187 static int vm_stat_get(void *_offset, u64 *val)
2188 {
2189         unsigned offset = (long)_offset;
2190         struct kvm *kvm;
2191
2192         *val = 0;
2193         spin_lock(&kvm_lock);
2194         list_for_each_entry(kvm, &vm_list, vm_list)
2195                 *val += *(u32 *)((void *)kvm + offset);
2196         spin_unlock(&kvm_lock);
2197         return 0;
2198 }
2199
2200 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2201
2202 static int vcpu_stat_get(void *_offset, u64 *val)
2203 {
2204         unsigned offset = (long)_offset;
2205         struct kvm *kvm;
2206         struct kvm_vcpu *vcpu;
2207         int i;
2208
2209         *val = 0;
2210         spin_lock(&kvm_lock);
2211         list_for_each_entry(kvm, &vm_list, vm_list)
2212                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2213                         vcpu = kvm->vcpus[i];
2214                         if (vcpu)
2215                                 *val += *(u32 *)((void *)vcpu + offset);
2216                 }
2217         spin_unlock(&kvm_lock);
2218         return 0;
2219 }
2220
2221 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2222
2223 static struct file_operations *stat_fops[] = {
2224         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2225         [KVM_STAT_VM]   = &vm_stat_fops,
2226 };
2227
2228 static void kvm_init_debug(void)
2229 {
2230         struct kvm_stats_debugfs_item *p;
2231
2232         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2233         for (p = debugfs_entries; p->name; ++p)
2234                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2235                                                 (void *)(long)p->offset,
2236                                                 stat_fops[p->kind]);
2237 }
2238
2239 static void kvm_exit_debug(void)
2240 {
2241         struct kvm_stats_debugfs_item *p;
2242
2243         for (p = debugfs_entries; p->name; ++p)
2244                 debugfs_remove(p->dentry);
2245         debugfs_remove(kvm_debugfs_dir);
2246 }
2247
2248 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2249 {
2250         hardware_disable(NULL);
2251         return 0;
2252 }
2253
2254 static int kvm_resume(struct sys_device *dev)
2255 {
2256         hardware_enable(NULL);
2257         return 0;
2258 }
2259
2260 static struct sysdev_class kvm_sysdev_class = {
2261         .name = "kvm",
2262         .suspend = kvm_suspend,
2263         .resume = kvm_resume,
2264 };
2265
2266 static struct sys_device kvm_sysdev = {
2267         .id = 0,
2268         .cls = &kvm_sysdev_class,
2269 };
2270
2271 struct page *bad_page;
2272 pfn_t bad_pfn;
2273
2274 static inline
2275 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2276 {
2277         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2278 }
2279
2280 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2281 {
2282         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2283
2284         kvm_arch_vcpu_load(vcpu, cpu);
2285 }
2286
2287 static void kvm_sched_out(struct preempt_notifier *pn,
2288                           struct task_struct *next)
2289 {
2290         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2291
2292         kvm_arch_vcpu_put(vcpu);
2293 }
2294
2295 int kvm_init(void *opaque, unsigned int vcpu_size,
2296                   struct module *module)
2297 {
2298         int r;
2299         int cpu;
2300
2301         kvm_init_debug();
2302
2303         r = kvm_arch_init(opaque);
2304         if (r)
2305                 goto out_fail;
2306
2307         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2308
2309         if (bad_page == NULL) {
2310                 r = -ENOMEM;
2311                 goto out;
2312         }
2313
2314         bad_pfn = page_to_pfn(bad_page);
2315
2316         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2317                 r = -ENOMEM;
2318                 goto out_free_0;
2319         }
2320         cpumask_clear(cpus_hardware_enabled);
2321
2322         r = kvm_arch_hardware_setup();
2323         if (r < 0)
2324                 goto out_free_0a;
2325
2326         for_each_online_cpu(cpu) {
2327                 smp_call_function_single(cpu,
2328                                 kvm_arch_check_processor_compat,
2329                                 &r, 1);
2330                 if (r < 0)
2331                         goto out_free_1;
2332         }
2333
2334         on_each_cpu(hardware_enable, NULL, 1);
2335         r = register_cpu_notifier(&kvm_cpu_notifier);
2336         if (r)
2337                 goto out_free_2;
2338         register_reboot_notifier(&kvm_reboot_notifier);
2339
2340         r = sysdev_class_register(&kvm_sysdev_class);
2341         if (r)
2342                 goto out_free_3;
2343
2344         r = sysdev_register(&kvm_sysdev);
2345         if (r)
2346                 goto out_free_4;
2347
2348         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2349         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2350                                            __alignof__(struct kvm_vcpu),
2351                                            0, NULL);
2352         if (!kvm_vcpu_cache) {
2353                 r = -ENOMEM;
2354                 goto out_free_5;
2355         }
2356
2357         kvm_chardev_ops.owner = module;
2358         kvm_vm_fops.owner = module;
2359         kvm_vcpu_fops.owner = module;
2360
2361         r = misc_register(&kvm_dev);
2362         if (r) {
2363                 printk(KERN_ERR "kvm: misc device register failed\n");
2364                 goto out_free;
2365         }
2366
2367         kvm_preempt_ops.sched_in = kvm_sched_in;
2368         kvm_preempt_ops.sched_out = kvm_sched_out;
2369 #ifndef CONFIG_X86
2370         msi2intx = 0;
2371 #endif
2372
2373         return 0;
2374
2375 out_free:
2376         kmem_cache_destroy(kvm_vcpu_cache);
2377 out_free_5:
2378         sysdev_unregister(&kvm_sysdev);
2379 out_free_4:
2380         sysdev_class_unregister(&kvm_sysdev_class);
2381 out_free_3:
2382         unregister_reboot_notifier(&kvm_reboot_notifier);
2383         unregister_cpu_notifier(&kvm_cpu_notifier);
2384 out_free_2:
2385         on_each_cpu(hardware_disable, NULL, 1);
2386 out_free_1:
2387         kvm_arch_hardware_unsetup();
2388 out_free_0a:
2389         free_cpumask_var(cpus_hardware_enabled);
2390 out_free_0:
2391         __free_page(bad_page);
2392 out:
2393         kvm_arch_exit();
2394         kvm_exit_debug();
2395 out_fail:
2396         return r;
2397 }
2398 EXPORT_SYMBOL_GPL(kvm_init);
2399
2400 void kvm_exit(void)
2401 {
2402         kvm_trace_cleanup();
2403         misc_deregister(&kvm_dev);
2404         kmem_cache_destroy(kvm_vcpu_cache);
2405         sysdev_unregister(&kvm_sysdev);
2406         sysdev_class_unregister(&kvm_sysdev_class);
2407         unregister_reboot_notifier(&kvm_reboot_notifier);
2408         unregister_cpu_notifier(&kvm_cpu_notifier);
2409         on_each_cpu(hardware_disable, NULL, 1);
2410         kvm_arch_hardware_unsetup();
2411         kvm_arch_exit();
2412         kvm_exit_debug();
2413         free_cpumask_var(cpus_hardware_enabled);
2414         __free_page(bad_page);
2415 }
2416 EXPORT_SYMBOL_GPL(kvm_exit);