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 CONFIG_X86
  51 #include <asm/msidef.h>
  52 #endif
  53
  54 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
  55 #include "coalesced_mmio.h"
  56 #endif
  57
  58 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
  59 #include <linux/pci.h>
  60 #include <linux/interrupt.h>
  61 #include "irq.h"
  62 #endif
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 static int msi2intx = 1;
  68 module_param(msi2intx, bool, 0);
  69
  70 DEFINE_SPINLOCK(kvm_lock);
  71 LIST_HEAD(vm_list);
  72
  73 static cpumask_var_t cpus_hardware_enabled;
  74
  75 struct kmem_cache *kvm_vcpu_cache;
  76 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  77
  78 static __read_mostly struct preempt_ops kvm_preempt_ops;
  79
  80 struct dentry *kvm_debugfs_dir;
  81
  82 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  83                            unsigned long arg);
  84
  85 static bool kvm_rebooting;
  86
  87 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
  88
  89 #ifdef CONFIG_X86
  90 static void assigned_device_msi_dispatch(struct kvm_assigned_dev_kernel *dev)
  91 {
  92         int vcpu_id;
  93         struct kvm_vcpu *vcpu;
  94         struct kvm_ioapic *ioapic = ioapic_irqchip(dev->kvm);
  95         int dest_id = (dev->guest_msi.address_lo & MSI_ADDR_DEST_ID_MASK)
  96                         >> MSI_ADDR_DEST_ID_SHIFT;
  97         int vector = (dev->guest_msi.data & MSI_DATA_VECTOR_MASK)
  98                         >> MSI_DATA_VECTOR_SHIFT;
  99         int dest_mode = test_bit(MSI_ADDR_DEST_MODE_SHIFT,
 100                                 (unsigned long *)&dev->guest_msi.address_lo);
 101         int trig_mode = test_bit(MSI_DATA_TRIGGER_SHIFT,
 102                                 (unsigned long *)&dev->guest_msi.data);
 103         int delivery_mode = test_bit(MSI_DATA_DELIVERY_MODE_SHIFT,
 104                                 (unsigned long *)&dev->guest_msi.data);
 105         u32 deliver_bitmask;
 106
 107         BUG_ON(!ioapic);
 108
 109         deliver_bitmask = kvm_ioapic_get_delivery_bitmask(ioapic,
 110                                 dest_id, dest_mode);
 111         /* IOAPIC delivery mode value is the same as MSI here */
 112         switch (delivery_mode) {
 113         case IOAPIC_LOWEST_PRIORITY:
 114                 vcpu = kvm_get_lowest_prio_vcpu(ioapic->kvm, vector,
 115                                 deliver_bitmask);
 116                 if (vcpu != NULL)
 117                         kvm_apic_set_irq(vcpu, vector, trig_mode);
 118                 else
 119                         printk(KERN_INFO "kvm: null lowest priority vcpu!\n");
 120                 break;
 121         case IOAPIC_FIXED:
 122                 for (vcpu_id = 0; deliver_bitmask != 0; vcpu_id++) {
 123                         if (!(deliver_bitmask & (1 << vcpu_id)))
 124                                 continue;
 125                         deliver_bitmask &= ~(1 << vcpu_id);
 126                         vcpu = ioapic->kvm->vcpus[vcpu_id];
 127                         if (vcpu)
 128                                 kvm_apic_set_irq(vcpu, vector, trig_mode);
 129                 }
 130                 break;
 131         default:
 132                 printk(KERN_INFO "kvm: unsupported MSI delivery mode\n");
 133         }
 134 }
 135 #else
 136 static void assigned_device_msi_dispatch(struct kvm_assigned_dev_kernel *dev) {}
 137 #endif
 138
 139 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
 140                                                       int assigned_dev_id)
 141 {
 142         struct list_head *ptr;
 143         struct kvm_assigned_dev_kernel *match;
 144
 145         list_for_each(ptr, head) {
 146                 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
 147                 if (match->assigned_dev_id == assigned_dev_id)
 148                         return match;
 149         }
 150         return NULL;
 151 }
 152
 153 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
 154 {
 155         struct kvm_assigned_dev_kernel *assigned_dev;
 156
 157         assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
 158                                     interrupt_work);
 159
 160         /* This is taken to safely inject irq inside the guest. When
 161          * the interrupt injection (or the ioapic code) uses a
 162          * finer-grained lock, update this
 163          */
 164         mutex_lock(&assigned_dev->kvm->lock);
 165         if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_GUEST_INTX)
 166                 kvm_set_irq(assigned_dev->kvm,
 167                             assigned_dev->irq_source_id,
 168                             assigned_dev->guest_irq, 1);
 169         else if (assigned_dev->irq_requested_type &
 170                                 KVM_ASSIGNED_DEV_GUEST_MSI) {
 171                 assigned_device_msi_dispatch(assigned_dev);
 172                 enable_irq(assigned_dev->host_irq);
 173                 assigned_dev->host_irq_disabled = false;
 174         }
 175         mutex_unlock(&assigned_dev->kvm->lock);
 176 }
 177
 178 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
 179 {
 180         struct kvm_assigned_dev_kernel *assigned_dev =
 181                 (struct kvm_assigned_dev_kernel *) dev_id;
 182
 183         schedule_work(&assigned_dev->interrupt_work);
 184
 185         disable_irq_nosync(irq);
 186         assigned_dev->host_irq_disabled = true;
 187
 188         return IRQ_HANDLED;
 189 }
 190
 191 /* Ack the irq line for an assigned device */
 192 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
 193 {
 194         struct kvm_assigned_dev_kernel *dev;
 195
 196         if (kian->gsi == -1)
 197                 return;
 198
 199         dev = container_of(kian, struct kvm_assigned_dev_kernel,
 200                            ack_notifier);
 201
 202         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
 203
 204         /* The guest irq may be shared so this ack may be
 205          * from another device.
 206          */
 207         if (dev->host_irq_disabled) {
 208                 enable_irq(dev->host_irq);
 209                 dev->host_irq_disabled = false;
 210         }
 211 }
 212
 213 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
 214 static void kvm_free_assigned_irq(struct kvm *kvm,
 215                                   struct kvm_assigned_dev_kernel *assigned_dev)
 216 {
 217         if (!irqchip_in_kernel(kvm))
 218                 return;
 219
 220         kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
 221
 222         if (assigned_dev->irq_source_id != -1)
 223                 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
 224         assigned_dev->irq_source_id = -1;
 225
 226         if (!assigned_dev->irq_requested_type)
 227                 return;
 228
 229         /*
 230          * In kvm_free_device_irq, cancel_work_sync return true if:
 231          * 1. work is scheduled, and then cancelled.
 232          * 2. work callback is executed.
 233          *
 234          * The first one ensured that the irq is disabled and no more events
 235          * would happen. But for the second one, the irq may be enabled (e.g.
 236          * for MSI). So we disable irq here to prevent further events.
 237          *
 238          * Notice this maybe result in nested disable if the interrupt type is
 239          * INTx, but it's OK for we are going to free it.
 240          *
 241          * If this function is a part of VM destroy, please ensure that till
 242          * now, the kvm state is still legal for probably we also have to wait
 243          * interrupt_work done.
 244          */
 245         disable_irq_nosync(assigned_dev->host_irq);
 246         cancel_work_sync(&assigned_dev->interrupt_work);
 247
 248         free_irq(assigned_dev->host_irq, (void *)assigned_dev);
 249
 250         if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
 251                 pci_disable_msi(assigned_dev->dev);
 252
 253         assigned_dev->irq_requested_type = 0;
 254 }
 255
 256
 257 static void kvm_free_assigned_device(struct kvm *kvm,
 258                                      struct kvm_assigned_dev_kernel
 259                                      *assigned_dev)
 260 {
 261         kvm_free_assigned_irq(kvm, assigned_dev);
 262
 263         pci_reset_function(assigned_dev->dev);
 264
 265         pci_release_regions(assigned_dev->dev);
 266         pci_disable_device(assigned_dev->dev);
 267         pci_dev_put(assigned_dev->dev);
 268
 269         list_del(&assigned_dev->list);
 270         kfree(assigned_dev);
 271 }
 272
 273 void kvm_free_all_assigned_devices(struct kvm *kvm)
 274 {
 275         struct list_head *ptr, *ptr2;
 276         struct kvm_assigned_dev_kernel *assigned_dev;
 277
 278         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
 279                 assigned_dev = list_entry(ptr,
 280                                           struct kvm_assigned_dev_kernel,
 281                                           list);
 282
 283                 kvm_free_assigned_device(kvm, assigned_dev);
 284         }
 285 }
 286
 287 static int assigned_device_update_intx(struct kvm *kvm,
 288                         struct kvm_assigned_dev_kernel *adev,
 289                         struct kvm_assigned_irq *airq)
 290 {
 291         adev->guest_irq = airq->guest_irq;
 292         adev->ack_notifier.gsi = airq->guest_irq;
 293
 294         if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_INTX)
 295                 return 0;
 296
 297         if (irqchip_in_kernel(kvm)) {
 298                 if (!msi2intx &&
 299                     (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)) {
 300                         free_irq(adev->host_irq, (void *)adev);
 301                         pci_disable_msi(adev->dev);
 302                 }
 303
 304                 if (!capable(CAP_SYS_RAWIO))
 305                         return -EPERM;
 306
 307                 if (airq->host_irq)
 308                         adev->host_irq = airq->host_irq;
 309                 else
 310                         adev->host_irq = adev->dev->irq;
 311
 312                 /* Even though this is PCI, we don't want to use shared
 313                  * interrupts. Sharing host devices with guest-assigned devices
 314                  * on the same interrupt line is not a happy situation: there
 315                  * are going to be long delays in accepting, acking, etc.
 316                  */
 317                 if (request_irq(adev->host_irq, kvm_assigned_dev_intr,
 318                                 0, "kvm_assigned_intx_device", (void *)adev))
 319                         return -EIO;
 320         }
 321
 322         adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_INTX |
 323                                    KVM_ASSIGNED_DEV_HOST_INTX;
 324         return 0;
 325 }
 326
 327 #ifdef CONFIG_X86
 328 static int assigned_device_update_msi(struct kvm *kvm,
 329                         struct kvm_assigned_dev_kernel *adev,
 330                         struct kvm_assigned_irq *airq)
 331 {
 332         int r;
 333
 334         if (airq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI) {
 335                 /* x86 don't care upper address of guest msi message addr */
 336                 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_MSI;
 337                 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_INTX;
 338                 adev->guest_msi.address_lo = airq->guest_msi.addr_lo;
 339                 adev->guest_msi.data = airq->guest_msi.data;
 340                 adev->ack_notifier.gsi = -1;
 341         } else if (msi2intx) {
 342                 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_INTX;
 343                 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_MSI;
 344                 adev->guest_irq = airq->guest_irq;
 345                 adev->ack_notifier.gsi = airq->guest_irq;
 346         } else {
 347                 /*
 348                  * Guest require to disable device MSI, we disable MSI and
 349                  * re-enable INTx by default again. Notice it's only for
 350                  * non-msi2intx.
 351                  */
 352                 assigned_device_update_intx(kvm, adev, airq);
 353                 return 0;
 354         }
 355
 356         if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
 357                 return 0;
 358
 359         if (irqchip_in_kernel(kvm)) {
 360                 if (!msi2intx) {
 361                         if (adev->irq_requested_type &
 362                                         KVM_ASSIGNED_DEV_HOST_INTX)
 363                                 free_irq(adev->host_irq, (void *)adev);
 364
 365                         r = pci_enable_msi(adev->dev);
 366                         if (r)
 367                                 return r;
 368                 }
 369
 370                 adev->host_irq = adev->dev->irq;
 371                 if (request_irq(adev->host_irq, kvm_assigned_dev_intr, 0,
 372                                 "kvm_assigned_msi_device", (void *)adev))
 373                         return -EIO;
 374         }
 375
 376         if (!msi2intx)
 377                 adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_MSI;
 378
 379         adev->irq_requested_type |= KVM_ASSIGNED_DEV_HOST_MSI;
 380         return 0;
 381 }
 382 #endif
 383
 384 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 385                                    struct kvm_assigned_irq
 386                                    *assigned_irq)
 387 {
 388         int r = 0;
 389         struct kvm_assigned_dev_kernel *match;
 390         u32 current_flags = 0, changed_flags;
 391
 392         mutex_lock(&kvm->lock);
 393
 394         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 395                                       assigned_irq->assigned_dev_id);
 396         if (!match) {
 397                 mutex_unlock(&kvm->lock);
 398                 return -EINVAL;
 399         }
 400
 401         if (!match->irq_requested_type) {
 402                 INIT_WORK(&match->interrupt_work,
 403                                 kvm_assigned_dev_interrupt_work_handler);
 404                 if (irqchip_in_kernel(kvm)) {
 405                         /* Register ack nofitier */
 406                         match->ack_notifier.gsi = -1;
 407                         match->ack_notifier.irq_acked =
 408                                         kvm_assigned_dev_ack_irq;
 409                         kvm_register_irq_ack_notifier(kvm,
 410                                         &match->ack_notifier);
 411
 412                         /* Request IRQ source ID */
 413                         r = kvm_request_irq_source_id(kvm);
 414                         if (r < 0)
 415                                 goto out_release;
 416                         else
 417                                 match->irq_source_id = r;
 418
 419 #ifdef CONFIG_X86
 420                         /* Determine host device irq type, we can know the
 421                          * result from dev->msi_enabled */
 422                         if (msi2intx)
 423                                 pci_enable_msi(match->dev);
 424 #endif
 425                 }
 426         }
 427
 428         if ((match->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI) &&
 429                  (match->irq_requested_type & KVM_ASSIGNED_DEV_GUEST_MSI))
 430                 current_flags |= KVM_DEV_IRQ_ASSIGN_ENABLE_MSI;
 431
 432         changed_flags = assigned_irq->flags ^ current_flags;
 433
 434         if ((changed_flags & KVM_DEV_IRQ_ASSIGN_MSI_ACTION) ||
 435             (msi2intx && match->dev->msi_enabled)) {
 436 #ifdef CONFIG_X86
 437                 r = assigned_device_update_msi(kvm, match, assigned_irq);
 438                 if (r) {
 439                         printk(KERN_WARNING "kvm: failed to enable "
 440                                         "MSI device!\n");
 441                         goto out_release;
 442                 }
 443 #else
 444                 r = -ENOTTY;
 445 #endif
 446         } else if (assigned_irq->host_irq == 0 && match->dev->irq == 0) {
 447                 /* Host device IRQ 0 means don't support INTx */
 448                 if (!msi2intx) {
 449                         printk(KERN_WARNING
 450                                "kvm: wait device to enable MSI!\n");
 451                         r = 0;
 452                 } else {
 453                         printk(KERN_WARNING
 454                                "kvm: failed to enable MSI device!\n");
 455                         r = -ENOTTY;
 456                         goto out_release;
 457                 }
 458         } else {
 459                 /* Non-sharing INTx mode */
 460                 r = assigned_device_update_intx(kvm, match, assigned_irq);
 461                 if (r) {
 462                         printk(KERN_WARNING "kvm: failed to enable "
 463                                         "INTx device!\n");
 464                         goto out_release;
 465                 }
 466         }
 467
 468         mutex_unlock(&kvm->lock);
 469         return r;
 470 out_release:
 471         mutex_unlock(&kvm->lock);
 472         kvm_free_assigned_device(kvm, match);
 473         return r;
 474 }
 475
 476 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 477                                       struct kvm_assigned_pci_dev *assigned_dev)
 478 {
 479         int r = 0;
 480         struct kvm_assigned_dev_kernel *match;
 481         struct pci_dev *dev;
 482
 483         down_read(&kvm->slots_lock);
 484         mutex_lock(&kvm->lock);
 485
 486         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 487                                       assigned_dev->assigned_dev_id);
 488         if (match) {
 489                 /* device already assigned */
 490                 r = -EINVAL;
 491                 goto out;
 492         }
 493
 494         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 495         if (match == NULL) {
 496                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 497                        __func__);
 498                 r = -ENOMEM;
 499                 goto out;
 500         }
 501         dev = pci_get_bus_and_slot(assigned_dev->busnr,
 502                                    assigned_dev->devfn);
 503         if (!dev) {
 504                 printk(KERN_INFO "%s: host device not found\n", __func__);
 505                 r = -EINVAL;
 506                 goto out_free;
 507         }
 508         if (pci_enable_device(dev)) {
 509                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 510                 r = -EBUSY;
 511                 goto out_put;
 512         }
 513         r = pci_request_regions(dev, "kvm_assigned_device");
 514         if (r) {
 515                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 516                        __func__);
 517                 goto out_disable;
 518         }
 519
 520         pci_reset_function(dev);
 521
 522         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 523         match->host_busnr = assigned_dev->busnr;
 524         match->host_devfn = assigned_dev->devfn;
 525         match->flags = assigned_dev->flags;
 526         match->dev = dev;
 527         match->irq_source_id = -1;
 528         match->kvm = kvm;
 529
 530         list_add(&match->list, &kvm->arch.assigned_dev_head);
 531
 532         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
 533                 if (!kvm->arch.iommu_domain) {
 534                         r = kvm_iommu_map_guest(kvm);
 535                         if (r)
 536                                 goto out_list_del;
 537                 }
 538                 r = kvm_assign_device(kvm, match);
 539                 if (r)
 540                         goto out_list_del;
 541         }
 542
 543 out:
 544         mutex_unlock(&kvm->lock);
 545         up_read(&kvm->slots_lock);
 546         return r;
 547 out_list_del:
 548         list_del(&match->list);
 549         pci_release_regions(dev);
 550 out_disable:
 551         pci_disable_device(dev);
 552 out_put:
 553         pci_dev_put(dev);
 554 out_free:
 555         kfree(match);
 556         mutex_unlock(&kvm->lock);
 557         up_read(&kvm->slots_lock);
 558         return r;
 559 }
 560 #endif
 561
 562 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
 563 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
 564                 struct kvm_assigned_pci_dev *assigned_dev)
 565 {
 566         int r = 0;
 567         struct kvm_assigned_dev_kernel *match;
 568
 569         mutex_lock(&kvm->lock);
 570
 571         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 572                                       assigned_dev->assigned_dev_id);
 573         if (!match) {
 574                 printk(KERN_INFO "%s: device hasn't been assigned before, "
 575                   "so cannot be deassigned\n", __func__);
 576                 r = -EINVAL;
 577                 goto out;
 578         }
 579
 580         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
 581                 kvm_deassign_device(kvm, match);
 582
 583         kvm_free_assigned_device(kvm, match);
 584
 585 out:
 586         mutex_unlock(&kvm->lock);
 587         return r;
 588 }
 589 #endif
 590
 591 static inline int valid_vcpu(int n)
 592 {
 593         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 594 }
 595
 596 inline int kvm_is_mmio_pfn(pfn_t pfn)
 597 {
 598         if (pfn_valid(pfn))
 599                 return PageReserved(pfn_to_page(pfn));
 600
 601         return true;
 602 }
 603
 604 /*
 605  * Switches to specified vcpu, until a matching vcpu_put()
 606  */
 607 void vcpu_load(struct kvm_vcpu *vcpu)
 608 {
 609         int cpu;
 610
 611         mutex_lock(&vcpu->mutex);
 612         cpu = get_cpu();
 613         preempt_notifier_register(&vcpu->preempt_notifier);
 614         kvm_arch_vcpu_load(vcpu, cpu);
 615         put_cpu();
 616 }
 617
 618 void vcpu_put(struct kvm_vcpu *vcpu)
 619 {
 620         preempt_disable();
 621         kvm_arch_vcpu_put(vcpu);
 622         preempt_notifier_unregister(&vcpu->preempt_notifier);
 623         preempt_enable();
 624         mutex_unlock(&vcpu->mutex);
 625 }
 626
 627 static void ack_flush(void *_completed)
 628 {
 629 }
 630
 631 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 632 {
 633         int i, cpu, me;
 634         cpumask_var_t cpus;
 635         bool called = true;
 636         struct kvm_vcpu *vcpu;
 637
 638         if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
 639                 cpumask_clear(cpus);
 640
 641         me = get_cpu();
 642         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 643                 vcpu = kvm->vcpus[i];
 644                 if (!vcpu)
 645                         continue;
 646                 if (test_and_set_bit(req, &vcpu->requests))
 647                         continue;
 648                 cpu = vcpu->cpu;
 649                 if (cpus != NULL && cpu != -1 && cpu != me)
 650                         cpumask_set_cpu(cpu, cpus);
 651         }
 652         if (unlikely(cpus == NULL))
 653                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 654         else if (!cpumask_empty(cpus))
 655                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 656         else
 657                 called = false;
 658         put_cpu();
 659         free_cpumask_var(cpus);
 660         return called;
 661 }
 662
 663 void kvm_flush_remote_tlbs(struct kvm *kvm)
 664 {
 665         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 666                 ++kvm->stat.remote_tlb_flush;
 667 }
 668
 669 void kvm_reload_remote_mmus(struct kvm *kvm)
 670 {
 671         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 672 }
 673
 674 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 675 {
 676         struct page *page;
 677         int r;
 678
 679         mutex_init(&vcpu->mutex);
 680         vcpu->cpu = -1;
 681         vcpu->kvm = kvm;
 682         vcpu->vcpu_id = id;
 683         init_waitqueue_head(&vcpu->wq);
 684
 685         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 686         if (!page) {
 687                 r = -ENOMEM;
 688                 goto fail;
 689         }
 690         vcpu->run = page_address(page);
 691
 692         r = kvm_arch_vcpu_init(vcpu);
 693         if (r < 0)
 694                 goto fail_free_run;
 695         return 0;
 696
 697 fail_free_run:
 698         free_page((unsigned long)vcpu->run);
 699 fail:
 700         return r;
 701 }
 702 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 703
 704 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 705 {
 706         kvm_arch_vcpu_uninit(vcpu);
 707         free_page((unsigned long)vcpu->run);
 708 }
 709 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 710
 711 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 712 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 713 {
 714         return container_of(mn, struct kvm, mmu_notifier);
 715 }
 716
 717 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 718                                              struct mm_struct *mm,
 719                                              unsigned long address)
 720 {
 721         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 722         int need_tlb_flush;
 723
 724         /*
 725          * When ->invalidate_page runs, the linux pte has been zapped
 726          * already but the page is still allocated until
 727          * ->invalidate_page returns. So if we increase the sequence
 728          * here the kvm page fault will notice if the spte can't be
 729          * established because the page is going to be freed. If
 730          * instead the kvm page fault establishes the spte before
 731          * ->invalidate_page runs, kvm_unmap_hva will release it
 732          * before returning.
 733          *
 734          * The sequence increase only need to be seen at spin_unlock
 735          * time, and not at spin_lock time.
 736          *
 737          * Increasing the sequence after the spin_unlock would be
 738          * unsafe because the kvm page fault could then establish the
 739          * pte after kvm_unmap_hva returned, without noticing the page
 740          * is going to be freed.
 741          */
 742         spin_lock(&kvm->mmu_lock);
 743         kvm->mmu_notifier_seq++;
 744         need_tlb_flush = kvm_unmap_hva(kvm, address);
 745         spin_unlock(&kvm->mmu_lock);
 746
 747         /* we've to flush the tlb before the pages can be freed */
 748         if (need_tlb_flush)
 749                 kvm_flush_remote_tlbs(kvm);
 750
 751 }
 752
 753 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 754                                                     struct mm_struct *mm,
 755                                                     unsigned long start,
 756                                                     unsigned long end)
 757 {
 758         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 759         int need_tlb_flush = 0;
 760
 761         spin_lock(&kvm->mmu_lock);
 762         /*
 763          * The count increase must become visible at unlock time as no
 764          * spte can be established without taking the mmu_lock and
 765          * count is also read inside the mmu_lock critical section.
 766          */
 767         kvm->mmu_notifier_count++;
 768         for (; start < end; start += PAGE_SIZE)
 769                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 770         spin_unlock(&kvm->mmu_lock);
 771
 772         /* we've to flush the tlb before the pages can be freed */
 773         if (need_tlb_flush)
 774                 kvm_flush_remote_tlbs(kvm);
 775 }
 776
 777 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 778                                                   struct mm_struct *mm,
 779                                                   unsigned long start,
 780                                                   unsigned long end)
 781 {
 782         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 783
 784         spin_lock(&kvm->mmu_lock);
 785         /*
 786          * This sequence increase will notify the kvm page fault that
 787          * the page that is going to be mapped in the spte could have
 788          * been freed.
 789          */
 790         kvm->mmu_notifier_seq++;
 791         /*
 792          * The above sequence increase must be visible before the
 793          * below count decrease but both values are read by the kvm
 794          * page fault under mmu_lock spinlock so we don't need to add
 795          * a smb_wmb() here in between the two.
 796          */
 797         kvm->mmu_notifier_count--;
 798         spin_unlock(&kvm->mmu_lock);
 799
 800         BUG_ON(kvm->mmu_notifier_count < 0);
 801 }
 802
 803 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 804                                               struct mm_struct *mm,
 805                                               unsigned long address)
 806 {
 807         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 808         int young;
 809
 810         spin_lock(&kvm->mmu_lock);
 811         young = kvm_age_hva(kvm, address);
 812         spin_unlock(&kvm->mmu_lock);
 813
 814         if (young)
 815                 kvm_flush_remote_tlbs(kvm);
 816
 817         return young;
 818 }
 819
 820 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 821                                      struct mm_struct *mm)
 822 {
 823         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 824         kvm_arch_flush_shadow(kvm);
 825 }
 826
 827 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 828         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 829         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 830         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 831         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 832         .release                = kvm_mmu_notifier_release,
 833 };
 834 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 835
 836 static struct kvm *kvm_create_vm(void)
 837 {
 838         struct kvm *kvm = kvm_arch_create_vm();
 839 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 840         struct page *page;
 841 #endif
 842
 843         if (IS_ERR(kvm))
 844                 goto out;
 845
 846 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 847         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 848         if (!page) {
 849                 kfree(kvm);
 850                 return ERR_PTR(-ENOMEM);
 851         }
 852         kvm->coalesced_mmio_ring =
 853                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 854 #endif
 855
 856 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 857         {
 858                 int err;
 859                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 860                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 861                 if (err) {
 862 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 863                         put_page(page);
 864 #endif
 865                         kfree(kvm);
 866                         return ERR_PTR(err);
 867                 }
 868         }
 869 #endif
 870
 871         kvm->mm = current->mm;
 872         atomic_inc(&kvm->mm->mm_count);
 873         spin_lock_init(&kvm->mmu_lock);
 874         kvm_io_bus_init(&kvm->pio_bus);
 875         mutex_init(&kvm->lock);
 876         kvm_io_bus_init(&kvm->mmio_bus);
 877         init_rwsem(&kvm->slots_lock);
 878         atomic_set(&kvm->users_count, 1);
 879         spin_lock(&kvm_lock);
 880         list_add(&kvm->vm_list, &vm_list);
 881         spin_unlock(&kvm_lock);
 882 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 883         kvm_coalesced_mmio_init(kvm);
 884 #endif
 885 out:
 886         return kvm;
 887 }
 888
 889 /*
 890  * Free any memory in @free but not in @dont.
 891  */
 892 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 893                                   struct kvm_memory_slot *dont)
 894 {
 895         if (!dont || free->rmap != dont->rmap)
 896                 vfree(free->rmap);
 897
 898         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 899                 vfree(free->dirty_bitmap);
 900
 901         if (!dont || free->lpage_info != dont->lpage_info)
 902                 vfree(free->lpage_info);
 903
 904         free->npages = 0;
 905         free->dirty_bitmap = NULL;
 906         free->rmap = NULL;
 907         free->lpage_info = NULL;
 908 }
 909
 910 void kvm_free_physmem(struct kvm *kvm)
 911 {
 912         int i;
 913
 914         for (i = 0; i < kvm->nmemslots; ++i)
 915                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 916 }
 917
 918 static void kvm_destroy_vm(struct kvm *kvm)
 919 {
 920         struct mm_struct *mm = kvm->mm;
 921
 922         kvm_arch_sync_events(kvm);
 923         spin_lock(&kvm_lock);
 924         list_del(&kvm->vm_list);
 925         spin_unlock(&kvm_lock);
 926         kvm_io_bus_destroy(&kvm->pio_bus);
 927         kvm_io_bus_destroy(&kvm->mmio_bus);
 928 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 929         if (kvm->coalesced_mmio_ring != NULL)
 930                 free_page((unsigned long)kvm->coalesced_mmio_ring);
 931 #endif
 932 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 933         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 934 #endif
 935         kvm_arch_destroy_vm(kvm);
 936         mmdrop(mm);
 937 }
 938
 939 void kvm_get_kvm(struct kvm *kvm)
 940 {
 941         atomic_inc(&kvm->users_count);
 942 }
 943 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 944
 945 void kvm_put_kvm(struct kvm *kvm)
 946 {
 947         if (atomic_dec_and_test(&kvm->users_count))
 948                 kvm_destroy_vm(kvm);
 949 }
 950 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 951
 952
 953 static int kvm_vm_release(struct inode *inode, struct file *filp)
 954 {
 955         struct kvm *kvm = filp->private_data;
 956
 957         kvm_put_kvm(kvm);
 958         return 0;
 959 }
 960
 961 /*
 962  * Allocate some memory and give it an address in the guest physical address
 963  * space.
 964  *
 965  * Discontiguous memory is allowed, mostly for framebuffers.
 966  *
 967  * Must be called holding mmap_sem for write.
 968  */
 969 int __kvm_set_memory_region(struct kvm *kvm,
 970                             struct kvm_userspace_memory_region *mem,
 971                             int user_alloc)
 972 {
 973         int r;
 974         gfn_t base_gfn;
 975         unsigned long npages;
 976         unsigned long i;
 977         struct kvm_memory_slot *memslot;
 978         struct kvm_memory_slot old, new;
 979
 980         r = -EINVAL;
 981         /* General sanity checks */
 982         if (mem->memory_size & (PAGE_SIZE - 1))
 983                 goto out;
 984         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 985                 goto out;
 986         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 987                 goto out;
 988         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 989                 goto out;
 990         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 991                 goto out;
 992
 993         memslot = &kvm->memslots[mem->slot];
 994         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 995         npages = mem->memory_size >> PAGE_SHIFT;
 996
 997         if (!npages)
 998                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 999
1000         new = old = *memslot;
1001
1002         new.base_gfn = base_gfn;
1003         new.npages = npages;
1004         new.flags = mem->flags;
1005
1006         /* Disallow changing a memory slot's size. */
1007         r = -EINVAL;
1008         if (npages && old.npages && npages != old.npages)
1009                 goto out_free;
1010
1011         /* Check for overlaps */
1012         r = -EEXIST;
1013         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1014                 struct kvm_memory_slot *s = &kvm->memslots[i];
1015
1016                 if (s == memslot)
1017                         continue;
1018                 if (!((base_gfn + npages <= s->base_gfn) ||
1019                       (base_gfn >= s->base_gfn + s->npages)))
1020                         goto out_free;
1021         }
1022
1023         /* Free page dirty bitmap if unneeded */
1024         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1025                 new.dirty_bitmap = NULL;
1026
1027         r = -ENOMEM;
1028
1029         /* Allocate if a slot is being created */
1030 #ifndef CONFIG_S390
1031         if (npages && !new.rmap) {
1032                 new.rmap = vmalloc(npages * sizeof(struct page *));
1033
1034                 if (!new.rmap)
1035                         goto out_free;
1036
1037                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1038
1039                 new.user_alloc = user_alloc;
1040                 /*
1041                  * hva_to_rmmap() serialzies with the mmu_lock and to be
1042                  * safe it has to ignore memslots with !user_alloc &&
1043                  * !userspace_addr.
1044                  */
1045                 if (user_alloc)
1046                         new.userspace_addr = mem->userspace_addr;
1047                 else
1048                         new.userspace_addr = 0;
1049         }
1050         if (npages && !new.lpage_info) {
1051                 int largepages = npages / KVM_PAGES_PER_HPAGE;
1052                 if (npages % KVM_PAGES_PER_HPAGE)
1053                         largepages++;
1054                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1055                         largepages++;
1056
1057                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1058
1059                 if (!new.lpage_info)
1060                         goto out_free;
1061
1062                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1063
1064                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1065                         new.lpage_info[0].write_count = 1;
1066                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1067                         new.lpage_info[largepages-1].write_count = 1;
1068         }
1069
1070         /* Allocate page dirty bitmap if needed */
1071         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1072                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1073
1074                 new.dirty_bitmap = vmalloc(dirty_bytes);
1075                 if (!new.dirty_bitmap)
1076                         goto out_free;
1077                 memset(new.dirty_bitmap, 0, dirty_bytes);
1078         }
1079 #endif /* not defined CONFIG_S390 */
1080
1081         if (!npages)
1082                 kvm_arch_flush_shadow(kvm);
1083
1084         spin_lock(&kvm->mmu_lock);
1085         if (mem->slot >= kvm->nmemslots)
1086                 kvm->nmemslots = mem->slot + 1;
1087
1088         *memslot = new;
1089         spin_unlock(&kvm->mmu_lock);
1090
1091         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1092         if (r) {
1093                 spin_lock(&kvm->mmu_lock);
1094                 *memslot = old;
1095                 spin_unlock(&kvm->mmu_lock);
1096                 goto out_free;
1097         }
1098
1099         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1100         /* Slot deletion case: we have to update the current slot */
1101         if (!npages)
1102                 *memslot = old;
1103 #ifdef CONFIG_DMAR
1104         /* map the pages in iommu page table */
1105         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1106         if (r)
1107                 goto out;
1108 #endif
1109         return 0;
1110
1111 out_free:
1112         kvm_free_physmem_slot(&new, &old);
1113 out:
1114         return r;
1115
1116 }
1117 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1118
1119 int kvm_set_memory_region(struct kvm *kvm,
1120                           struct kvm_userspace_memory_region *mem,
1121                           int user_alloc)
1122 {
1123         int r;
1124
1125         down_write(&kvm->slots_lock);
1126         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1127         up_write(&kvm->slots_lock);
1128         return r;
1129 }
1130 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1131
1132 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1133                                    struct
1134                                    kvm_userspace_memory_region *mem,
1135                                    int user_alloc)
1136 {
1137         if (mem->slot >= KVM_MEMORY_SLOTS)
1138                 return -EINVAL;
1139         return kvm_set_memory_region(kvm, mem, user_alloc);
1140 }
1141
1142 int kvm_get_dirty_log(struct kvm *kvm,
1143                         struct kvm_dirty_log *log, int *is_dirty)
1144 {
1145         struct kvm_memory_slot *memslot;
1146         int r, i;
1147         int n;
1148         unsigned long any = 0;
1149
1150         r = -EINVAL;
1151         if (log->slot >= KVM_MEMORY_SLOTS)
1152                 goto out;
1153
1154         memslot = &kvm->memslots[log->slot];
1155         r = -ENOENT;
1156         if (!memslot->dirty_bitmap)
1157                 goto out;
1158
1159         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1160
1161         for (i = 0; !any && i < n/sizeof(long); ++i)
1162                 any = memslot->dirty_bitmap[i];
1163
1164         r = -EFAULT;
1165         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1166                 goto out;
1167
1168         if (any)
1169                 *is_dirty = 1;
1170
1171         r = 0;
1172 out:
1173         return r;
1174 }
1175
1176 int is_error_page(struct page *page)
1177 {
1178         return page == bad_page;
1179 }
1180 EXPORT_SYMBOL_GPL(is_error_page);
1181
1182 int is_error_pfn(pfn_t pfn)
1183 {
1184         return pfn == bad_pfn;
1185 }
1186 EXPORT_SYMBOL_GPL(is_error_pfn);
1187
1188 static inline unsigned long bad_hva(void)
1189 {
1190         return PAGE_OFFSET;
1191 }
1192
1193 int kvm_is_error_hva(unsigned long addr)
1194 {
1195         return addr == bad_hva();
1196 }
1197 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1198
1199 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1200 {
1201         int i;
1202
1203         for (i = 0; i < kvm->nmemslots; ++i) {
1204                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1205
1206                 if (gfn >= memslot->base_gfn
1207                     && gfn < memslot->base_gfn + memslot->npages)
1208                         return memslot;
1209         }
1210         return NULL;
1211 }
1212 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1213
1214 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1215 {
1216         gfn = unalias_gfn(kvm, gfn);
1217         return gfn_to_memslot_unaliased(kvm, gfn);
1218 }
1219
1220 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1221 {
1222         int i;
1223
1224         gfn = unalias_gfn(kvm, gfn);
1225         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1226                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1227
1228                 if (gfn >= memslot->base_gfn
1229                     && gfn < memslot->base_gfn + memslot->npages)
1230                         return 1;
1231         }
1232         return 0;
1233 }
1234 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1235
1236 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1237 {
1238         struct kvm_memory_slot *slot;
1239
1240         gfn = unalias_gfn(kvm, gfn);
1241         slot = gfn_to_memslot_unaliased(kvm, gfn);
1242         if (!slot)
1243                 return bad_hva();
1244         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1245 }
1246 EXPORT_SYMBOL_GPL(gfn_to_hva);
1247
1248 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1249 {
1250         struct page *page[1];
1251         unsigned long addr;
1252         int npages;
1253         pfn_t pfn;
1254
1255         might_sleep();
1256
1257         addr = gfn_to_hva(kvm, gfn);
1258         if (kvm_is_error_hva(addr)) {
1259                 get_page(bad_page);
1260                 return page_to_pfn(bad_page);
1261         }
1262
1263         npages = get_user_pages_fast(addr, 1, 1, page);
1264
1265         if (unlikely(npages != 1)) {
1266                 struct vm_area_struct *vma;
1267
1268                 down_read(&current->mm->mmap_sem);
1269                 vma = find_vma(current->mm, addr);
1270
1271                 if (vma == NULL || addr < vma->vm_start ||
1272                     !(vma->vm_flags & VM_PFNMAP)) {
1273                         up_read(&current->mm->mmap_sem);
1274                         get_page(bad_page);
1275                         return page_to_pfn(bad_page);
1276                 }
1277
1278                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1279                 up_read(&current->mm->mmap_sem);
1280                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1281         } else
1282                 pfn = page_to_pfn(page[0]);
1283
1284         return pfn;
1285 }
1286
1287 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1288
1289 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1290 {
1291         pfn_t pfn;
1292
1293         pfn = gfn_to_pfn(kvm, gfn);
1294         if (!kvm_is_mmio_pfn(pfn))
1295                 return pfn_to_page(pfn);
1296
1297         WARN_ON(kvm_is_mmio_pfn(pfn));
1298
1299         get_page(bad_page);
1300         return bad_page;
1301 }
1302
1303 EXPORT_SYMBOL_GPL(gfn_to_page);
1304
1305 void kvm_release_page_clean(struct page *page)
1306 {
1307         kvm_release_pfn_clean(page_to_pfn(page));
1308 }
1309 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1310
1311 void kvm_release_pfn_clean(pfn_t pfn)
1312 {
1313         if (!kvm_is_mmio_pfn(pfn))
1314                 put_page(pfn_to_page(pfn));
1315 }
1316 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1317
1318 void kvm_release_page_dirty(struct page *page)
1319 {
1320         kvm_release_pfn_dirty(page_to_pfn(page));
1321 }
1322 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1323
1324 void kvm_release_pfn_dirty(pfn_t pfn)
1325 {
1326         kvm_set_pfn_dirty(pfn);
1327         kvm_release_pfn_clean(pfn);
1328 }
1329 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1330
1331 void kvm_set_page_dirty(struct page *page)
1332 {
1333         kvm_set_pfn_dirty(page_to_pfn(page));
1334 }
1335 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1336
1337 void kvm_set_pfn_dirty(pfn_t pfn)
1338 {
1339         if (!kvm_is_mmio_pfn(pfn)) {
1340                 struct page *page = pfn_to_page(pfn);
1341                 if (!PageReserved(page))
1342                         SetPageDirty(page);
1343         }
1344 }
1345 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1346
1347 void kvm_set_pfn_accessed(pfn_t pfn)
1348 {
1349         if (!kvm_is_mmio_pfn(pfn))
1350                 mark_page_accessed(pfn_to_page(pfn));
1351 }
1352 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1353
1354 void kvm_get_pfn(pfn_t pfn)
1355 {
1356         if (!kvm_is_mmio_pfn(pfn))
1357                 get_page(pfn_to_page(pfn));
1358 }
1359 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1360
1361 static int next_segment(unsigned long len, int offset)
1362 {
1363         if (len > PAGE_SIZE - offset)
1364                 return PAGE_SIZE - offset;
1365         else
1366                 return len;
1367 }
1368
1369 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1370                         int len)
1371 {
1372         int r;
1373         unsigned long addr;
1374
1375         addr = gfn_to_hva(kvm, gfn);
1376         if (kvm_is_error_hva(addr))
1377                 return -EFAULT;
1378         r = copy_from_user(data, (void __user *)addr + offset, len);
1379         if (r)
1380                 return -EFAULT;
1381         return 0;
1382 }
1383 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1384
1385 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1386 {
1387         gfn_t gfn = gpa >> PAGE_SHIFT;
1388         int seg;
1389         int offset = offset_in_page(gpa);
1390         int ret;
1391
1392         while ((seg = next_segment(len, offset)) != 0) {
1393                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1394                 if (ret < 0)
1395                         return ret;
1396                 offset = 0;
1397                 len -= seg;
1398                 data += seg;
1399                 ++gfn;
1400         }
1401         return 0;
1402 }
1403 EXPORT_SYMBOL_GPL(kvm_read_guest);
1404
1405 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1406                           unsigned long len)
1407 {
1408         int r;
1409         unsigned long addr;
1410         gfn_t gfn = gpa >> PAGE_SHIFT;
1411         int offset = offset_in_page(gpa);
1412
1413         addr = gfn_to_hva(kvm, gfn);
1414         if (kvm_is_error_hva(addr))
1415                 return -EFAULT;
1416         pagefault_disable();
1417         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1418         pagefault_enable();
1419         if (r)
1420                 return -EFAULT;
1421         return 0;
1422 }
1423 EXPORT_SYMBOL(kvm_read_guest_atomic);
1424
1425 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1426                          int offset, int len)
1427 {
1428         int r;
1429         unsigned long addr;
1430
1431         addr = gfn_to_hva(kvm, gfn);
1432         if (kvm_is_error_hva(addr))
1433                 return -EFAULT;
1434         r = copy_to_user((void __user *)addr + offset, data, len);
1435         if (r)
1436                 return -EFAULT;
1437         mark_page_dirty(kvm, gfn);
1438         return 0;
1439 }
1440 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1441
1442 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1443                     unsigned long len)
1444 {
1445         gfn_t gfn = gpa >> PAGE_SHIFT;
1446         int seg;
1447         int offset = offset_in_page(gpa);
1448         int ret;
1449
1450         while ((seg = next_segment(len, offset)) != 0) {
1451                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1452                 if (ret < 0)
1453                         return ret;
1454                 offset = 0;
1455                 len -= seg;
1456                 data += seg;
1457                 ++gfn;
1458         }
1459         return 0;
1460 }
1461
1462 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1463 {
1464         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1465 }
1466 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1467
1468 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1469 {
1470         gfn_t gfn = gpa >> PAGE_SHIFT;
1471         int seg;
1472         int offset = offset_in_page(gpa);
1473         int ret;
1474
1475         while ((seg = next_segment(len, offset)) != 0) {
1476                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1477                 if (ret < 0)
1478                         return ret;
1479                 offset = 0;
1480                 len -= seg;
1481                 ++gfn;
1482         }
1483         return 0;
1484 }
1485 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1486
1487 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1488 {
1489         struct kvm_memory_slot *memslot;
1490
1491         gfn = unalias_gfn(kvm, gfn);
1492         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1493         if (memslot && memslot->dirty_bitmap) {
1494                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1495
1496                 /* avoid RMW */
1497                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1498                         set_bit(rel_gfn, memslot->dirty_bitmap);
1499         }
1500 }
1501
1502 /*
1503  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1504  */
1505 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1506 {
1507         DEFINE_WAIT(wait);
1508
1509         for (;;) {
1510                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1511
1512                 if (kvm_cpu_has_interrupt(vcpu) ||
1513                     kvm_cpu_has_pending_timer(vcpu) ||
1514                     kvm_arch_vcpu_runnable(vcpu)) {
1515                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1516                         break;
1517                 }
1518                 if (signal_pending(current))
1519                         break;
1520
1521                 vcpu_put(vcpu);
1522                 schedule();
1523                 vcpu_load(vcpu);
1524         }
1525
1526         finish_wait(&vcpu->wq, &wait);
1527 }
1528
1529 void kvm_resched(struct kvm_vcpu *vcpu)
1530 {
1531         if (!need_resched())
1532                 return;
1533         cond_resched();
1534 }
1535 EXPORT_SYMBOL_GPL(kvm_resched);
1536
1537 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1538 {
1539         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1540         struct page *page;
1541
1542         if (vmf->pgoff == 0)
1543                 page = virt_to_page(vcpu->run);
1544 #ifdef CONFIG_X86
1545         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1546                 page = virt_to_page(vcpu->arch.pio_data);
1547 #endif
1548 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1549         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1550                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1551 #endif
1552         else
1553                 return VM_FAULT_SIGBUS;
1554         get_page(page);
1555         vmf->page = page;
1556         return 0;
1557 }
1558
1559 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1560         .fault = kvm_vcpu_fault,
1561 };
1562
1563 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1564 {
1565         vma->vm_ops = &kvm_vcpu_vm_ops;
1566         return 0;
1567 }
1568
1569 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1570 {
1571         struct kvm_vcpu *vcpu = filp->private_data;
1572
1573         kvm_put_kvm(vcpu->kvm);
1574         return 0;
1575 }
1576
1577 static struct file_operations kvm_vcpu_fops = {
1578         .release        = kvm_vcpu_release,
1579         .unlocked_ioctl = kvm_vcpu_ioctl,
1580         .compat_ioctl   = kvm_vcpu_ioctl,
1581         .mmap           = kvm_vcpu_mmap,
1582 };
1583
1584 /*
1585  * Allocates an inode for the vcpu.
1586  */
1587 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1588 {
1589         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1590         if (fd < 0)
1591                 kvm_put_kvm(vcpu->kvm);
1592         return fd;
1593 }
1594
1595 /*
1596  * Creates some virtual cpus.  Good luck creating more than one.
1597  */
1598 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1599 {
1600         int r;
1601         struct kvm_vcpu *vcpu;
1602
1603         if (!valid_vcpu(n))
1604                 return -EINVAL;
1605
1606         vcpu = kvm_arch_vcpu_create(kvm, n);
1607         if (IS_ERR(vcpu))
1608                 return PTR_ERR(vcpu);
1609
1610         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1611
1612         r = kvm_arch_vcpu_setup(vcpu);
1613         if (r)
1614                 return r;
1615
1616         mutex_lock(&kvm->lock);
1617         if (kvm->vcpus[n]) {
1618                 r = -EEXIST;
1619                 goto vcpu_destroy;
1620         }
1621         kvm->vcpus[n] = vcpu;
1622         mutex_unlock(&kvm->lock);
1623
1624         /* Now it's all set up, let userspace reach it */
1625         kvm_get_kvm(kvm);
1626         r = create_vcpu_fd(vcpu);
1627         if (r < 0)
1628                 goto unlink;
1629         return r;
1630
1631 unlink:
1632         mutex_lock(&kvm->lock);
1633         kvm->vcpus[n] = NULL;
1634 vcpu_destroy:
1635         mutex_unlock(&kvm->lock);
1636         kvm_arch_vcpu_destroy(vcpu);
1637         return r;
1638 }
1639
1640 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1641 {
1642         if (sigset) {
1643                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1644                 vcpu->sigset_active = 1;
1645                 vcpu->sigset = *sigset;
1646         } else
1647                 vcpu->sigset_active = 0;
1648         return 0;
1649 }
1650
1651 static long kvm_vcpu_ioctl(struct file *filp,
1652                            unsigned int ioctl, unsigned long arg)
1653 {
1654         struct kvm_vcpu *vcpu = filp->private_data;
1655         void __user *argp = (void __user *)arg;
1656         int r;
1657         struct kvm_fpu *fpu = NULL;
1658         struct kvm_sregs *kvm_sregs = NULL;
1659
1660         if (vcpu->kvm->mm != current->mm)
1661                 return -EIO;
1662         switch (ioctl) {
1663         case KVM_RUN:
1664                 r = -EINVAL;
1665                 if (arg)
1666                         goto out;
1667                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1668                 break;
1669         case KVM_GET_REGS: {
1670                 struct kvm_regs *kvm_regs;
1671
1672                 r = -ENOMEM;
1673                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1674                 if (!kvm_regs)
1675                         goto out;
1676                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1677                 if (r)
1678                         goto out_free1;
1679                 r = -EFAULT;
1680                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1681                         goto out_free1;
1682                 r = 0;
1683 out_free1:
1684                 kfree(kvm_regs);
1685                 break;
1686         }
1687         case KVM_SET_REGS: {
1688                 struct kvm_regs *kvm_regs;
1689
1690                 r = -ENOMEM;
1691                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1692                 if (!kvm_regs)
1693                         goto out;
1694                 r = -EFAULT;
1695                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1696                         goto out_free2;
1697                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1698                 if (r)
1699                         goto out_free2;
1700                 r = 0;
1701 out_free2:
1702                 kfree(kvm_regs);
1703                 break;
1704         }
1705         case KVM_GET_SREGS: {
1706                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1707                 r = -ENOMEM;
1708                 if (!kvm_sregs)
1709                         goto out;
1710                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1711                 if (r)
1712                         goto out;
1713                 r = -EFAULT;
1714                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1715                         goto out;
1716                 r = 0;
1717                 break;
1718         }
1719         case KVM_SET_SREGS: {
1720                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1721                 r = -ENOMEM;
1722                 if (!kvm_sregs)
1723                         goto out;
1724                 r = -EFAULT;
1725                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1726                         goto out;
1727                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1728                 if (r)
1729                         goto out;
1730                 r = 0;
1731                 break;
1732         }
1733         case KVM_GET_MP_STATE: {
1734                 struct kvm_mp_state mp_state;
1735
1736                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1737                 if (r)
1738                         goto out;
1739                 r = -EFAULT;
1740                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1741                         goto out;
1742                 r = 0;
1743                 break;
1744         }
1745         case KVM_SET_MP_STATE: {
1746                 struct kvm_mp_state mp_state;
1747
1748                 r = -EFAULT;
1749                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1750                         goto out;
1751                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1752                 if (r)
1753                         goto out;
1754                 r = 0;
1755                 break;
1756         }
1757         case KVM_TRANSLATE: {
1758                 struct kvm_translation tr;
1759
1760                 r = -EFAULT;
1761                 if (copy_from_user(&tr, argp, sizeof tr))
1762                         goto out;
1763                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1764                 if (r)
1765                         goto out;
1766                 r = -EFAULT;
1767                 if (copy_to_user(argp, &tr, sizeof tr))
1768                         goto out;
1769                 r = 0;
1770                 break;
1771         }
1772         case KVM_SET_GUEST_DEBUG: {
1773                 struct kvm_guest_debug dbg;
1774
1775                 r = -EFAULT;
1776                 if (copy_from_user(&dbg, argp, sizeof dbg))
1777                         goto out;
1778                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1779                 if (r)
1780                         goto out;
1781                 r = 0;
1782                 break;
1783         }
1784         case KVM_SET_SIGNAL_MASK: {
1785                 struct kvm_signal_mask __user *sigmask_arg = argp;
1786                 struct kvm_signal_mask kvm_sigmask;
1787                 sigset_t sigset, *p;
1788
1789                 p = NULL;
1790                 if (argp) {
1791                         r = -EFAULT;
1792                         if (copy_from_user(&kvm_sigmask, argp,
1793                                            sizeof kvm_sigmask))
1794                                 goto out;
1795                         r = -EINVAL;
1796                         if (kvm_sigmask.len != sizeof sigset)
1797                                 goto out;
1798                         r = -EFAULT;
1799                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1800                                            sizeof sigset))
1801                                 goto out;
1802                         p = &sigset;
1803                 }
1804                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1805                 break;
1806         }
1807         case KVM_GET_FPU: {
1808                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1809                 r = -ENOMEM;
1810                 if (!fpu)
1811                         goto out;
1812                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1813                 if (r)
1814                         goto out;
1815                 r = -EFAULT;
1816                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1817                         goto out;
1818                 r = 0;
1819                 break;
1820         }
1821         case KVM_SET_FPU: {
1822                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1823                 r = -ENOMEM;
1824                 if (!fpu)
1825                         goto out;
1826                 r = -EFAULT;
1827                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1828                         goto out;
1829                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1830                 if (r)
1831                         goto out;
1832                 r = 0;
1833                 break;
1834         }
1835         default:
1836                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1837         }
1838 out:
1839         kfree(fpu);
1840         kfree(kvm_sregs);
1841         return r;
1842 }
1843
1844 static long kvm_vm_ioctl(struct file *filp,
1845                            unsigned int ioctl, unsigned long arg)
1846 {
1847         struct kvm *kvm = filp->private_data;
1848         void __user *argp = (void __user *)arg;
1849         int r;
1850
1851         if (kvm->mm != current->mm)
1852                 return -EIO;
1853         switch (ioctl) {
1854         case KVM_CREATE_VCPU:
1855                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1856                 if (r < 0)
1857                         goto out;
1858                 break;
1859         case KVM_SET_USER_MEMORY_REGION: {
1860                 struct kvm_userspace_memory_region kvm_userspace_mem;
1861
1862                 r = -EFAULT;
1863                 if (copy_from_user(&kvm_userspace_mem, argp,
1864                                                 sizeof kvm_userspace_mem))
1865                         goto out;
1866
1867                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1868                 if (r)
1869                         goto out;
1870                 break;
1871         }
1872         case KVM_GET_DIRTY_LOG: {
1873                 struct kvm_dirty_log log;
1874
1875                 r = -EFAULT;
1876                 if (copy_from_user(&log, argp, sizeof log))
1877                         goto out;
1878                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1879                 if (r)
1880                         goto out;
1881                 break;
1882         }
1883 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1884         case KVM_REGISTER_COALESCED_MMIO: {
1885                 struct kvm_coalesced_mmio_zone zone;
1886                 r = -EFAULT;
1887                 if (copy_from_user(&zone, argp, sizeof zone))
1888                         goto out;
1889                 r = -ENXIO;
1890                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1891                 if (r)
1892                         goto out;
1893                 r = 0;
1894                 break;
1895         }
1896         case KVM_UNREGISTER_COALESCED_MMIO: {
1897                 struct kvm_coalesced_mmio_zone zone;
1898                 r = -EFAULT;
1899                 if (copy_from_user(&zone, argp, sizeof zone))
1900                         goto out;
1901                 r = -ENXIO;
1902                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1903                 if (r)
1904                         goto out;
1905                 r = 0;
1906                 break;
1907         }
1908 #endif
1909 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1910         case KVM_ASSIGN_PCI_DEVICE: {
1911                 struct kvm_assigned_pci_dev assigned_dev;
1912
1913                 r = -EFAULT;
1914                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1915                         goto out;
1916                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1917                 if (r)
1918                         goto out;
1919                 break;
1920         }
1921         case KVM_ASSIGN_IRQ: {
1922                 struct kvm_assigned_irq assigned_irq;
1923
1924                 r = -EFAULT;
1925                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1926                         goto out;
1927                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1928                 if (r)
1929                         goto out;
1930                 break;
1931         }
1932 #endif
1933 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
1934         case KVM_DEASSIGN_PCI_DEVICE: {
1935                 struct kvm_assigned_pci_dev assigned_dev;
1936
1937                 r = -EFAULT;
1938                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1939                         goto out;
1940                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
1941                 if (r)
1942                         goto out;
1943                 break;
1944         }
1945 #endif
1946         default:
1947                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1948         }
1949 out:
1950         return r;
1951 }
1952
1953 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1954 {
1955         struct page *page[1];
1956         unsigned long addr;
1957         int npages;
1958         gfn_t gfn = vmf->pgoff;
1959         struct kvm *kvm = vma->vm_file->private_data;
1960
1961         addr = gfn_to_hva(kvm, gfn);
1962         if (kvm_is_error_hva(addr))
1963                 return VM_FAULT_SIGBUS;
1964
1965         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1966                                 NULL);
1967         if (unlikely(npages != 1))
1968                 return VM_FAULT_SIGBUS;
1969
1970         vmf->page = page[0];
1971         return 0;
1972 }
1973
1974 static struct vm_operations_struct kvm_vm_vm_ops = {
1975         .fault = kvm_vm_fault,
1976 };
1977
1978 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1979 {
1980         vma->vm_ops = &kvm_vm_vm_ops;
1981         return 0;
1982 }
1983
1984 static struct file_operations kvm_vm_fops = {
1985         .release        = kvm_vm_release,
1986         .unlocked_ioctl = kvm_vm_ioctl,
1987         .compat_ioctl   = kvm_vm_ioctl,
1988         .mmap           = kvm_vm_mmap,
1989 };
1990
1991 static int kvm_dev_ioctl_create_vm(void)
1992 {
1993         int fd;
1994         struct kvm *kvm;
1995
1996         kvm = kvm_create_vm();
1997         if (IS_ERR(kvm))
1998                 return PTR_ERR(kvm);
1999         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2000         if (fd < 0)
2001                 kvm_put_kvm(kvm);
2002
2003         return fd;
2004 }
2005
2006 static long kvm_dev_ioctl_check_extension_generic(long arg)
2007 {
2008         switch (arg) {
2009         case KVM_CAP_USER_MEMORY:
2010         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2011                 return 1;
2012         default:
2013                 break;
2014         }
2015         return kvm_dev_ioctl_check_extension(arg);
2016 }
2017
2018 static long kvm_dev_ioctl(struct file *filp,
2019                           unsigned int ioctl, unsigned long arg)
2020 {
2021         long r = -EINVAL;
2022
2023         switch (ioctl) {
2024         case KVM_GET_API_VERSION:
2025                 r = -EINVAL;
2026                 if (arg)
2027                         goto out;
2028                 r = KVM_API_VERSION;
2029                 break;
2030         case KVM_CREATE_VM:
2031                 r = -EINVAL;
2032                 if (arg)
2033                         goto out;
2034                 r = kvm_dev_ioctl_create_vm();
2035                 break;
2036         case KVM_CHECK_EXTENSION:
2037                 r = kvm_dev_ioctl_check_extension_generic(arg);
2038                 break;
2039         case KVM_GET_VCPU_MMAP_SIZE:
2040                 r = -EINVAL;
2041                 if (arg)
2042                         goto out;
2043                 r = PAGE_SIZE;     /* struct kvm_run */
2044 #ifdef CONFIG_X86
2045                 r += PAGE_SIZE;    /* pio data page */
2046 #endif
2047 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2048                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2049 #endif
2050                 break;
2051         case KVM_TRACE_ENABLE:
2052         case KVM_TRACE_PAUSE:
2053         case KVM_TRACE_DISABLE:
2054                 r = kvm_trace_ioctl(ioctl, arg);
2055                 break;
2056         default:
2057                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2058         }
2059 out:
2060         return r;
2061 }
2062
2063 static struct file_operations kvm_chardev_ops = {
2064         .unlocked_ioctl = kvm_dev_ioctl,
2065         .compat_ioctl   = kvm_dev_ioctl,
2066 };
2067
2068 static struct miscdevice kvm_dev = {
2069         KVM_MINOR,
2070         "kvm",
2071         &kvm_chardev_ops,
2072 };
2073
2074 static void hardware_enable(void *junk)
2075 {
2076         int cpu = raw_smp_processor_id();
2077
2078         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2079                 return;
2080         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2081         kvm_arch_hardware_enable(NULL);
2082 }
2083
2084 static void hardware_disable(void *junk)
2085 {
2086         int cpu = raw_smp_processor_id();
2087
2088         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2089                 return;
2090         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2091         kvm_arch_hardware_disable(NULL);
2092 }
2093
2094 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2095                            void *v)
2096 {
2097         int cpu = (long)v;
2098
2099         val &= ~CPU_TASKS_FROZEN;
2100         switch (val) {
2101         case CPU_DYING:
2102                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2103                        cpu);
2104                 hardware_disable(NULL);
2105                 break;
2106         case CPU_UP_CANCELED:
2107                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2108                        cpu);
2109                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2110                 break;
2111         case CPU_ONLINE:
2112                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2113                        cpu);
2114                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2115                 break;
2116         }
2117         return NOTIFY_OK;
2118 }
2119
2120
2121 asmlinkage void kvm_handle_fault_on_reboot(void)
2122 {
2123         if (kvm_rebooting)
2124                 /* spin while reset goes on */
2125                 while (true)
2126                         ;
2127         /* Fault while not rebooting.  We want the trace. */
2128         BUG();
2129 }
2130 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2131
2132 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2133                       void *v)
2134 {
2135         if (val == SYS_RESTART) {
2136                 /*
2137                  * Some (well, at least mine) BIOSes hang on reboot if
2138                  * in vmx root mode.
2139                  */
2140                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2141                 kvm_rebooting = true;
2142                 on_each_cpu(hardware_disable, NULL, 1);
2143         }
2144         return NOTIFY_OK;
2145 }
2146
2147 static struct notifier_block kvm_reboot_notifier = {
2148         .notifier_call = kvm_reboot,
2149         .priority = 0,
2150 };
2151
2152 void kvm_io_bus_init(struct kvm_io_bus *bus)
2153 {
2154         memset(bus, 0, sizeof(*bus));
2155 }
2156
2157 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2158 {
2159         int i;
2160
2161         for (i = 0; i < bus->dev_count; i++) {
2162                 struct kvm_io_device *pos = bus->devs[i];
2163
2164                 kvm_iodevice_destructor(pos);
2165         }
2166 }
2167
2168 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2169                                           gpa_t addr, int len, int is_write)
2170 {
2171         int i;
2172
2173         for (i = 0; i < bus->dev_count; i++) {
2174                 struct kvm_io_device *pos = bus->devs[i];
2175
2176                 if (pos->in_range(pos, addr, len, is_write))
2177                         return pos;
2178         }
2179
2180         return NULL;
2181 }
2182
2183 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2184 {
2185         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2186
2187         bus->devs[bus->dev_count++] = dev;
2188 }
2189
2190 static struct notifier_block kvm_cpu_notifier = {
2191         .notifier_call = kvm_cpu_hotplug,
2192         .priority = 20, /* must be > scheduler priority */
2193 };
2194
2195 static int vm_stat_get(void *_offset, u64 *val)
2196 {
2197         unsigned offset = (long)_offset;
2198         struct kvm *kvm;
2199
2200         *val = 0;
2201         spin_lock(&kvm_lock);
2202         list_for_each_entry(kvm, &vm_list, vm_list)
2203                 *val += *(u32 *)((void *)kvm + offset);
2204         spin_unlock(&kvm_lock);
2205         return 0;
2206 }
2207
2208 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2209
2210 static int vcpu_stat_get(void *_offset, u64 *val)
2211 {
2212         unsigned offset = (long)_offset;
2213         struct kvm *kvm;
2214         struct kvm_vcpu *vcpu;
2215         int i;
2216
2217         *val = 0;
2218         spin_lock(&kvm_lock);
2219         list_for_each_entry(kvm, &vm_list, vm_list)
2220                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2221                         vcpu = kvm->vcpus[i];
2222                         if (vcpu)
2223                                 *val += *(u32 *)((void *)vcpu + offset);
2224                 }
2225         spin_unlock(&kvm_lock);
2226         return 0;
2227 }
2228
2229 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2230
2231 static struct file_operations *stat_fops[] = {
2232         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2233         [KVM_STAT_VM]   = &vm_stat_fops,
2234 };
2235
2236 static void kvm_init_debug(void)
2237 {
2238         struct kvm_stats_debugfs_item *p;
2239
2240         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2241         for (p = debugfs_entries; p->name; ++p)
2242                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2243                                                 (void *)(long)p->offset,
2244                                                 stat_fops[p->kind]);
2245 }
2246
2247 static void kvm_exit_debug(void)
2248 {
2249         struct kvm_stats_debugfs_item *p;
2250
2251         for (p = debugfs_entries; p->name; ++p)
2252                 debugfs_remove(p->dentry);
2253         debugfs_remove(kvm_debugfs_dir);
2254 }
2255
2256 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2257 {
2258         hardware_disable(NULL);
2259         return 0;
2260 }
2261
2262 static int kvm_resume(struct sys_device *dev)
2263 {
2264         hardware_enable(NULL);
2265         return 0;
2266 }
2267
2268 static struct sysdev_class kvm_sysdev_class = {
2269         .name = "kvm",
2270         .suspend = kvm_suspend,
2271         .resume = kvm_resume,
2272 };
2273
2274 static struct sys_device kvm_sysdev = {
2275         .id = 0,
2276         .cls = &kvm_sysdev_class,
2277 };
2278
2279 struct page *bad_page;
2280 pfn_t bad_pfn;
2281
2282 static inline
2283 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2284 {
2285         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2286 }
2287
2288 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2289 {
2290         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2291
2292         kvm_arch_vcpu_load(vcpu, cpu);
2293 }
2294
2295 static void kvm_sched_out(struct preempt_notifier *pn,
2296                           struct task_struct *next)
2297 {
2298         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2299
2300         kvm_arch_vcpu_put(vcpu);
2301 }
2302
2303 int kvm_init(void *opaque, unsigned int vcpu_size,
2304                   struct module *module)
2305 {
2306         int r;
2307         int cpu;
2308
2309         kvm_init_debug();
2310
2311         r = kvm_arch_init(opaque);
2312         if (r)
2313                 goto out_fail;
2314
2315         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2316
2317         if (bad_page == NULL) {
2318                 r = -ENOMEM;
2319                 goto out;
2320         }
2321
2322         bad_pfn = page_to_pfn(bad_page);
2323
2324         if (!alloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2325                 r = -ENOMEM;
2326                 goto out_free_0;
2327         }
2328
2329         r = kvm_arch_hardware_setup();
2330         if (r < 0)
2331                 goto out_free_0a;
2332
2333         for_each_online_cpu(cpu) {
2334                 smp_call_function_single(cpu,
2335                                 kvm_arch_check_processor_compat,
2336                                 &r, 1);
2337                 if (r < 0)
2338                         goto out_free_1;
2339         }
2340
2341         on_each_cpu(hardware_enable, NULL, 1);
2342         r = register_cpu_notifier(&kvm_cpu_notifier);
2343         if (r)
2344                 goto out_free_2;
2345         register_reboot_notifier(&kvm_reboot_notifier);
2346
2347         r = sysdev_class_register(&kvm_sysdev_class);
2348         if (r)
2349                 goto out_free_3;
2350
2351         r = sysdev_register(&kvm_sysdev);
2352         if (r)
2353                 goto out_free_4;
2354
2355         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2356         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2357                                            __alignof__(struct kvm_vcpu),
2358                                            0, NULL);
2359         if (!kvm_vcpu_cache) {
2360                 r = -ENOMEM;
2361                 goto out_free_5;
2362         }
2363
2364         kvm_chardev_ops.owner = module;
2365         kvm_vm_fops.owner = module;
2366         kvm_vcpu_fops.owner = module;
2367
2368         r = misc_register(&kvm_dev);
2369         if (r) {
2370                 printk(KERN_ERR "kvm: misc device register failed\n");
2371                 goto out_free;
2372         }
2373
2374         kvm_preempt_ops.sched_in = kvm_sched_in;
2375         kvm_preempt_ops.sched_out = kvm_sched_out;
2376 #ifndef CONFIG_X86
2377         msi2intx = 0;
2378 #endif
2379
2380         return 0;
2381
2382 out_free:
2383         kmem_cache_destroy(kvm_vcpu_cache);
2384 out_free_5:
2385         sysdev_unregister(&kvm_sysdev);
2386 out_free_4:
2387         sysdev_class_unregister(&kvm_sysdev_class);
2388 out_free_3:
2389         unregister_reboot_notifier(&kvm_reboot_notifier);
2390         unregister_cpu_notifier(&kvm_cpu_notifier);
2391 out_free_2:
2392         on_each_cpu(hardware_disable, NULL, 1);
2393 out_free_1:
2394         kvm_arch_hardware_unsetup();
2395 out_free_0a:
2396         free_cpumask_var(cpus_hardware_enabled);
2397 out_free_0:
2398         __free_page(bad_page);
2399 out:
2400         kvm_arch_exit();
2401         kvm_exit_debug();
2402 out_fail:
2403         return r;
2404 }
2405 EXPORT_SYMBOL_GPL(kvm_init);
2406
2407 void kvm_exit(void)
2408 {
2409         kvm_trace_cleanup();
2410         misc_deregister(&kvm_dev);
2411         kmem_cache_destroy(kvm_vcpu_cache);
2412         sysdev_unregister(&kvm_sysdev);
2413         sysdev_class_unregister(&kvm_sysdev_class);
2414         unregister_reboot_notifier(&kvm_reboot_notifier);
2415         unregister_cpu_notifier(&kvm_cpu_notifier);
2416         on_each_cpu(hardware_disable, NULL, 1);
2417         kvm_arch_hardware_unsetup();
2418         kvm_arch_exit();
2419         kvm_exit_debug();
2420         free_cpumask_var(cpus_hardware_enabled);
2421         __free_page(bad_page);
2422 }
2423 EXPORT_SYMBOL_GPL(kvm_exit);