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

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "iodev.h"
  19
  20 #include <linux/kvm_host.h>
  21 #include <linux/kvm.h>
  22 #include <linux/module.h>
  23 #include <linux/errno.h>
  24 #include <linux/percpu.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/sysdev.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44
  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 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 846         INIT_LIST_HEAD(&kvm->irq_routing);
 847         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 848 #endif
 849
 850 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 851         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 852         if (!page) {
 853                 kfree(kvm);
 854                 return ERR_PTR(-ENOMEM);
 855         }
 856         kvm->coalesced_mmio_ring =
 857                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 858 #endif
 859
 860 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 861         {
 862                 int err;
 863                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 864                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 865                 if (err) {
 866 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 867                         put_page(page);
 868 #endif
 869                         kfree(kvm);
 870                         return ERR_PTR(err);
 871                 }
 872         }
 873 #endif
 874
 875         kvm->mm = current->mm;
 876         atomic_inc(&kvm->mm->mm_count);
 877         spin_lock_init(&kvm->mmu_lock);
 878         kvm_io_bus_init(&kvm->pio_bus);
 879         mutex_init(&kvm->lock);
 880         kvm_io_bus_init(&kvm->mmio_bus);
 881         init_rwsem(&kvm->slots_lock);
 882         atomic_set(&kvm->users_count, 1);
 883         spin_lock(&kvm_lock);
 884         list_add(&kvm->vm_list, &vm_list);
 885         spin_unlock(&kvm_lock);
 886 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 887         kvm_coalesced_mmio_init(kvm);
 888 #endif
 889 out:
 890         return kvm;
 891 }
 892
 893 /*
 894  * Free any memory in @free but not in @dont.
 895  */
 896 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 897                                   struct kvm_memory_slot *dont)
 898 {
 899         if (!dont || free->rmap != dont->rmap)
 900                 vfree(free->rmap);
 901
 902         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 903                 vfree(free->dirty_bitmap);
 904
 905         if (!dont || free->lpage_info != dont->lpage_info)
 906                 vfree(free->lpage_info);
 907
 908         free->npages = 0;
 909         free->dirty_bitmap = NULL;
 910         free->rmap = NULL;
 911         free->lpage_info = NULL;
 912 }
 913
 914 void kvm_free_physmem(struct kvm *kvm)
 915 {
 916         int i;
 917
 918         for (i = 0; i < kvm->nmemslots; ++i)
 919                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 920 }
 921
 922 static void kvm_destroy_vm(struct kvm *kvm)
 923 {
 924         struct mm_struct *mm = kvm->mm;
 925
 926         kvm_arch_sync_events(kvm);
 927         spin_lock(&kvm_lock);
 928         list_del(&kvm->vm_list);
 929         spin_unlock(&kvm_lock);
 930         kvm_free_irq_routing(kvm);
 931         kvm_io_bus_destroy(&kvm->pio_bus);
 932         kvm_io_bus_destroy(&kvm->mmio_bus);
 933 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 934         if (kvm->coalesced_mmio_ring != NULL)
 935                 free_page((unsigned long)kvm->coalesced_mmio_ring);
 936 #endif
 937 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 938         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 939 #endif
 940         kvm_arch_destroy_vm(kvm);
 941         mmdrop(mm);
 942 }
 943
 944 void kvm_get_kvm(struct kvm *kvm)
 945 {
 946         atomic_inc(&kvm->users_count);
 947 }
 948 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 949
 950 void kvm_put_kvm(struct kvm *kvm)
 951 {
 952         if (atomic_dec_and_test(&kvm->users_count))
 953                 kvm_destroy_vm(kvm);
 954 }
 955 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 956
 957
 958 static int kvm_vm_release(struct inode *inode, struct file *filp)
 959 {
 960         struct kvm *kvm = filp->private_data;
 961
 962         kvm_put_kvm(kvm);
 963         return 0;
 964 }
 965
 966 /*
 967  * Allocate some memory and give it an address in the guest physical address
 968  * space.
 969  *
 970  * Discontiguous memory is allowed, mostly for framebuffers.
 971  *
 972  * Must be called holding mmap_sem for write.
 973  */
 974 int __kvm_set_memory_region(struct kvm *kvm,
 975                             struct kvm_userspace_memory_region *mem,
 976                             int user_alloc)
 977 {
 978         int r;
 979         gfn_t base_gfn;
 980         unsigned long npages;
 981         unsigned long i;
 982         struct kvm_memory_slot *memslot;
 983         struct kvm_memory_slot old, new;
 984
 985         r = -EINVAL;
 986         /* General sanity checks */
 987         if (mem->memory_size & (PAGE_SIZE - 1))
 988                 goto out;
 989         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 990                 goto out;
 991         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 992                 goto out;
 993         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 994                 goto out;
 995         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 996                 goto out;
 997
 998         memslot = &kvm->memslots[mem->slot];
 999         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1000         npages = mem->memory_size >> PAGE_SHIFT;
1001
1002         if (!npages)
1003                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1004
1005         new = old = *memslot;
1006
1007         new.base_gfn = base_gfn;
1008         new.npages = npages;
1009         new.flags = mem->flags;
1010
1011         /* Disallow changing a memory slot's size. */
1012         r = -EINVAL;
1013         if (npages && old.npages && npages != old.npages)
1014                 goto out_free;
1015
1016         /* Check for overlaps */
1017         r = -EEXIST;
1018         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1019                 struct kvm_memory_slot *s = &kvm->memslots[i];
1020
1021                 if (s == memslot)
1022                         continue;
1023                 if (!((base_gfn + npages <= s->base_gfn) ||
1024                       (base_gfn >= s->base_gfn + s->npages)))
1025                         goto out_free;
1026         }
1027
1028         /* Free page dirty bitmap if unneeded */
1029         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1030                 new.dirty_bitmap = NULL;
1031
1032         r = -ENOMEM;
1033
1034         /* Allocate if a slot is being created */
1035 #ifndef CONFIG_S390
1036         if (npages && !new.rmap) {
1037                 new.rmap = vmalloc(npages * sizeof(struct page *));
1038
1039                 if (!new.rmap)
1040                         goto out_free;
1041
1042                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1043
1044                 new.user_alloc = user_alloc;
1045                 /*
1046                  * hva_to_rmmap() serialzies with the mmu_lock and to be
1047                  * safe it has to ignore memslots with !user_alloc &&
1048                  * !userspace_addr.
1049                  */
1050                 if (user_alloc)
1051                         new.userspace_addr = mem->userspace_addr;
1052                 else
1053                         new.userspace_addr = 0;
1054         }
1055         if (npages && !new.lpage_info) {
1056                 int largepages = npages / KVM_PAGES_PER_HPAGE;
1057                 if (npages % KVM_PAGES_PER_HPAGE)
1058                         largepages++;
1059                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1060                         largepages++;
1061
1062                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1063
1064                 if (!new.lpage_info)
1065                         goto out_free;
1066
1067                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1068
1069                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1070                         new.lpage_info[0].write_count = 1;
1071                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1072                         new.lpage_info[largepages-1].write_count = 1;
1073         }
1074
1075         /* Allocate page dirty bitmap if needed */
1076         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1077                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1078
1079                 new.dirty_bitmap = vmalloc(dirty_bytes);
1080                 if (!new.dirty_bitmap)
1081                         goto out_free;
1082                 memset(new.dirty_bitmap, 0, dirty_bytes);
1083         }
1084 #endif /* not defined CONFIG_S390 */
1085
1086         if (!npages)
1087                 kvm_arch_flush_shadow(kvm);
1088
1089         spin_lock(&kvm->mmu_lock);
1090         if (mem->slot >= kvm->nmemslots)
1091                 kvm->nmemslots = mem->slot + 1;
1092
1093         *memslot = new;
1094         spin_unlock(&kvm->mmu_lock);
1095
1096         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1097         if (r) {
1098                 spin_lock(&kvm->mmu_lock);
1099                 *memslot = old;
1100                 spin_unlock(&kvm->mmu_lock);
1101                 goto out_free;
1102         }
1103
1104         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1105         /* Slot deletion case: we have to update the current slot */
1106         if (!npages)
1107                 *memslot = old;
1108 #ifdef CONFIG_DMAR
1109         /* map the pages in iommu page table */
1110         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1111         if (r)
1112                 goto out;
1113 #endif
1114         return 0;
1115
1116 out_free:
1117         kvm_free_physmem_slot(&new, &old);
1118 out:
1119         return r;
1120
1121 }
1122 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1123
1124 int kvm_set_memory_region(struct kvm *kvm,
1125                           struct kvm_userspace_memory_region *mem,
1126                           int user_alloc)
1127 {
1128         int r;
1129
1130         down_write(&kvm->slots_lock);
1131         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1132         up_write(&kvm->slots_lock);
1133         return r;
1134 }
1135 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1136
1137 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1138                                    struct
1139                                    kvm_userspace_memory_region *mem,
1140                                    int user_alloc)
1141 {
1142         if (mem->slot >= KVM_MEMORY_SLOTS)
1143                 return -EINVAL;
1144         return kvm_set_memory_region(kvm, mem, user_alloc);
1145 }
1146
1147 int kvm_get_dirty_log(struct kvm *kvm,
1148                         struct kvm_dirty_log *log, int *is_dirty)
1149 {
1150         struct kvm_memory_slot *memslot;
1151         int r, i;
1152         int n;
1153         unsigned long any = 0;
1154
1155         r = -EINVAL;
1156         if (log->slot >= KVM_MEMORY_SLOTS)
1157                 goto out;
1158
1159         memslot = &kvm->memslots[log->slot];
1160         r = -ENOENT;
1161         if (!memslot->dirty_bitmap)
1162                 goto out;
1163
1164         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1165
1166         for (i = 0; !any && i < n/sizeof(long); ++i)
1167                 any = memslot->dirty_bitmap[i];
1168
1169         r = -EFAULT;
1170         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1171                 goto out;
1172
1173         if (any)
1174                 *is_dirty = 1;
1175
1176         r = 0;
1177 out:
1178         return r;
1179 }
1180
1181 int is_error_page(struct page *page)
1182 {
1183         return page == bad_page;
1184 }
1185 EXPORT_SYMBOL_GPL(is_error_page);
1186
1187 int is_error_pfn(pfn_t pfn)
1188 {
1189         return pfn == bad_pfn;
1190 }
1191 EXPORT_SYMBOL_GPL(is_error_pfn);
1192
1193 static inline unsigned long bad_hva(void)
1194 {
1195         return PAGE_OFFSET;
1196 }
1197
1198 int kvm_is_error_hva(unsigned long addr)
1199 {
1200         return addr == bad_hva();
1201 }
1202 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1203
1204 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1205 {
1206         int i;
1207
1208         for (i = 0; i < kvm->nmemslots; ++i) {
1209                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1210
1211                 if (gfn >= memslot->base_gfn
1212                     && gfn < memslot->base_gfn + memslot->npages)
1213                         return memslot;
1214         }
1215         return NULL;
1216 }
1217 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1218
1219 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1220 {
1221         gfn = unalias_gfn(kvm, gfn);
1222         return gfn_to_memslot_unaliased(kvm, gfn);
1223 }
1224
1225 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1226 {
1227         int i;
1228
1229         gfn = unalias_gfn(kvm, gfn);
1230         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1231                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1232
1233                 if (gfn >= memslot->base_gfn
1234                     && gfn < memslot->base_gfn + memslot->npages)
1235                         return 1;
1236         }
1237         return 0;
1238 }
1239 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1240
1241 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1242 {
1243         struct kvm_memory_slot *slot;
1244
1245         gfn = unalias_gfn(kvm, gfn);
1246         slot = gfn_to_memslot_unaliased(kvm, gfn);
1247         if (!slot)
1248                 return bad_hva();
1249         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1250 }
1251 EXPORT_SYMBOL_GPL(gfn_to_hva);
1252
1253 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1254 {
1255         struct page *page[1];
1256         unsigned long addr;
1257         int npages;
1258         pfn_t pfn;
1259
1260         might_sleep();
1261
1262         addr = gfn_to_hva(kvm, gfn);
1263         if (kvm_is_error_hva(addr)) {
1264                 get_page(bad_page);
1265                 return page_to_pfn(bad_page);
1266         }
1267
1268         npages = get_user_pages_fast(addr, 1, 1, page);
1269
1270         if (unlikely(npages != 1)) {
1271                 struct vm_area_struct *vma;
1272
1273                 down_read(&current->mm->mmap_sem);
1274                 vma = find_vma(current->mm, addr);
1275
1276                 if (vma == NULL || addr < vma->vm_start ||
1277                     !(vma->vm_flags & VM_PFNMAP)) {
1278                         up_read(&current->mm->mmap_sem);
1279                         get_page(bad_page);
1280                         return page_to_pfn(bad_page);
1281                 }
1282
1283                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1284                 up_read(&current->mm->mmap_sem);
1285                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1286         } else
1287                 pfn = page_to_pfn(page[0]);
1288
1289         return pfn;
1290 }
1291
1292 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1293
1294 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1295 {
1296         pfn_t pfn;
1297
1298         pfn = gfn_to_pfn(kvm, gfn);
1299         if (!kvm_is_mmio_pfn(pfn))
1300                 return pfn_to_page(pfn);
1301
1302         WARN_ON(kvm_is_mmio_pfn(pfn));
1303
1304         get_page(bad_page);
1305         return bad_page;
1306 }
1307
1308 EXPORT_SYMBOL_GPL(gfn_to_page);
1309
1310 void kvm_release_page_clean(struct page *page)
1311 {
1312         kvm_release_pfn_clean(page_to_pfn(page));
1313 }
1314 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1315
1316 void kvm_release_pfn_clean(pfn_t pfn)
1317 {
1318         if (!kvm_is_mmio_pfn(pfn))
1319                 put_page(pfn_to_page(pfn));
1320 }
1321 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1322
1323 void kvm_release_page_dirty(struct page *page)
1324 {
1325         kvm_release_pfn_dirty(page_to_pfn(page));
1326 }
1327 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1328
1329 void kvm_release_pfn_dirty(pfn_t pfn)
1330 {
1331         kvm_set_pfn_dirty(pfn);
1332         kvm_release_pfn_clean(pfn);
1333 }
1334 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1335
1336 void kvm_set_page_dirty(struct page *page)
1337 {
1338         kvm_set_pfn_dirty(page_to_pfn(page));
1339 }
1340 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1341
1342 void kvm_set_pfn_dirty(pfn_t pfn)
1343 {
1344         if (!kvm_is_mmio_pfn(pfn)) {
1345                 struct page *page = pfn_to_page(pfn);
1346                 if (!PageReserved(page))
1347                         SetPageDirty(page);
1348         }
1349 }
1350 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1351
1352 void kvm_set_pfn_accessed(pfn_t pfn)
1353 {
1354         if (!kvm_is_mmio_pfn(pfn))
1355                 mark_page_accessed(pfn_to_page(pfn));
1356 }
1357 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1358
1359 void kvm_get_pfn(pfn_t pfn)
1360 {
1361         if (!kvm_is_mmio_pfn(pfn))
1362                 get_page(pfn_to_page(pfn));
1363 }
1364 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1365
1366 static int next_segment(unsigned long len, int offset)
1367 {
1368         if (len > PAGE_SIZE - offset)
1369                 return PAGE_SIZE - offset;
1370         else
1371                 return len;
1372 }
1373
1374 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1375                         int len)
1376 {
1377         int r;
1378         unsigned long addr;
1379
1380         addr = gfn_to_hva(kvm, gfn);
1381         if (kvm_is_error_hva(addr))
1382                 return -EFAULT;
1383         r = copy_from_user(data, (void __user *)addr + offset, len);
1384         if (r)
1385                 return -EFAULT;
1386         return 0;
1387 }
1388 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1389
1390 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1391 {
1392         gfn_t gfn = gpa >> PAGE_SHIFT;
1393         int seg;
1394         int offset = offset_in_page(gpa);
1395         int ret;
1396
1397         while ((seg = next_segment(len, offset)) != 0) {
1398                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1399                 if (ret < 0)
1400                         return ret;
1401                 offset = 0;
1402                 len -= seg;
1403                 data += seg;
1404                 ++gfn;
1405         }
1406         return 0;
1407 }
1408 EXPORT_SYMBOL_GPL(kvm_read_guest);
1409
1410 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1411                           unsigned long len)
1412 {
1413         int r;
1414         unsigned long addr;
1415         gfn_t gfn = gpa >> PAGE_SHIFT;
1416         int offset = offset_in_page(gpa);
1417
1418         addr = gfn_to_hva(kvm, gfn);
1419         if (kvm_is_error_hva(addr))
1420                 return -EFAULT;
1421         pagefault_disable();
1422         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1423         pagefault_enable();
1424         if (r)
1425                 return -EFAULT;
1426         return 0;
1427 }
1428 EXPORT_SYMBOL(kvm_read_guest_atomic);
1429
1430 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1431                          int offset, int len)
1432 {
1433         int r;
1434         unsigned long addr;
1435
1436         addr = gfn_to_hva(kvm, gfn);
1437         if (kvm_is_error_hva(addr))
1438                 return -EFAULT;
1439         r = copy_to_user((void __user *)addr + offset, data, len);
1440         if (r)
1441                 return -EFAULT;
1442         mark_page_dirty(kvm, gfn);
1443         return 0;
1444 }
1445 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1446
1447 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1448                     unsigned long len)
1449 {
1450         gfn_t gfn = gpa >> PAGE_SHIFT;
1451         int seg;
1452         int offset = offset_in_page(gpa);
1453         int ret;
1454
1455         while ((seg = next_segment(len, offset)) != 0) {
1456                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1457                 if (ret < 0)
1458                         return ret;
1459                 offset = 0;
1460                 len -= seg;
1461                 data += seg;
1462                 ++gfn;
1463         }
1464         return 0;
1465 }
1466
1467 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1468 {
1469         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1470 }
1471 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1472
1473 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1474 {
1475         gfn_t gfn = gpa >> PAGE_SHIFT;
1476         int seg;
1477         int offset = offset_in_page(gpa);
1478         int ret;
1479
1480         while ((seg = next_segment(len, offset)) != 0) {
1481                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1482                 if (ret < 0)
1483                         return ret;
1484                 offset = 0;
1485                 len -= seg;
1486                 ++gfn;
1487         }
1488         return 0;
1489 }
1490 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1491
1492 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1493 {
1494         struct kvm_memory_slot *memslot;
1495
1496         gfn = unalias_gfn(kvm, gfn);
1497         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1498         if (memslot && memslot->dirty_bitmap) {
1499                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1500
1501                 /* avoid RMW */
1502                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1503                         set_bit(rel_gfn, memslot->dirty_bitmap);
1504         }
1505 }
1506
1507 /*
1508  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1509  */
1510 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1511 {
1512         DEFINE_WAIT(wait);
1513
1514         for (;;) {
1515                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1516
1517                 if (kvm_cpu_has_interrupt(vcpu) ||
1518                     kvm_cpu_has_pending_timer(vcpu) ||
1519                     kvm_arch_vcpu_runnable(vcpu)) {
1520                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1521                         break;
1522                 }
1523                 if (signal_pending(current))
1524                         break;
1525
1526                 vcpu_put(vcpu);
1527                 schedule();
1528                 vcpu_load(vcpu);
1529         }
1530
1531         finish_wait(&vcpu->wq, &wait);
1532 }
1533
1534 void kvm_resched(struct kvm_vcpu *vcpu)
1535 {
1536         if (!need_resched())
1537                 return;
1538         cond_resched();
1539 }
1540 EXPORT_SYMBOL_GPL(kvm_resched);
1541
1542 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1543 {
1544         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1545         struct page *page;
1546
1547         if (vmf->pgoff == 0)
1548                 page = virt_to_page(vcpu->run);
1549 #ifdef CONFIG_X86
1550         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1551                 page = virt_to_page(vcpu->arch.pio_data);
1552 #endif
1553 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1554         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1555                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1556 #endif
1557         else
1558                 return VM_FAULT_SIGBUS;
1559         get_page(page);
1560         vmf->page = page;
1561         return 0;
1562 }
1563
1564 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1565         .fault = kvm_vcpu_fault,
1566 };
1567
1568 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1569 {
1570         vma->vm_ops = &kvm_vcpu_vm_ops;
1571         return 0;
1572 }
1573
1574 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1575 {
1576         struct kvm_vcpu *vcpu = filp->private_data;
1577
1578         kvm_put_kvm(vcpu->kvm);
1579         return 0;
1580 }
1581
1582 static struct file_operations kvm_vcpu_fops = {
1583         .release        = kvm_vcpu_release,
1584         .unlocked_ioctl = kvm_vcpu_ioctl,
1585         .compat_ioctl   = kvm_vcpu_ioctl,
1586         .mmap           = kvm_vcpu_mmap,
1587 };
1588
1589 /*
1590  * Allocates an inode for the vcpu.
1591  */
1592 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1593 {
1594         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1595         if (fd < 0)
1596                 kvm_put_kvm(vcpu->kvm);
1597         return fd;
1598 }
1599
1600 /*
1601  * Creates some virtual cpus.  Good luck creating more than one.
1602  */
1603 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1604 {
1605         int r;
1606         struct kvm_vcpu *vcpu;
1607
1608         if (!valid_vcpu(n))
1609                 return -EINVAL;
1610
1611         vcpu = kvm_arch_vcpu_create(kvm, n);
1612         if (IS_ERR(vcpu))
1613                 return PTR_ERR(vcpu);
1614
1615         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1616
1617         r = kvm_arch_vcpu_setup(vcpu);
1618         if (r)
1619                 return r;
1620
1621         mutex_lock(&kvm->lock);
1622         if (kvm->vcpus[n]) {
1623                 r = -EEXIST;
1624                 goto vcpu_destroy;
1625         }
1626         kvm->vcpus[n] = vcpu;
1627         mutex_unlock(&kvm->lock);
1628
1629         /* Now it's all set up, let userspace reach it */
1630         kvm_get_kvm(kvm);
1631         r = create_vcpu_fd(vcpu);
1632         if (r < 0)
1633                 goto unlink;
1634         return r;
1635
1636 unlink:
1637         mutex_lock(&kvm->lock);
1638         kvm->vcpus[n] = NULL;
1639 vcpu_destroy:
1640         mutex_unlock(&kvm->lock);
1641         kvm_arch_vcpu_destroy(vcpu);
1642         return r;
1643 }
1644
1645 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1646 {
1647         if (sigset) {
1648                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1649                 vcpu->sigset_active = 1;
1650                 vcpu->sigset = *sigset;
1651         } else
1652                 vcpu->sigset_active = 0;
1653         return 0;
1654 }
1655
1656 static long kvm_vcpu_ioctl(struct file *filp,
1657                            unsigned int ioctl, unsigned long arg)
1658 {
1659         struct kvm_vcpu *vcpu = filp->private_data;
1660         void __user *argp = (void __user *)arg;
1661         int r;
1662         struct kvm_fpu *fpu = NULL;
1663         struct kvm_sregs *kvm_sregs = NULL;
1664
1665         if (vcpu->kvm->mm != current->mm)
1666                 return -EIO;
1667         switch (ioctl) {
1668         case KVM_RUN:
1669                 r = -EINVAL;
1670                 if (arg)
1671                         goto out;
1672                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1673                 break;
1674         case KVM_GET_REGS: {
1675                 struct kvm_regs *kvm_regs;
1676
1677                 r = -ENOMEM;
1678                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1679                 if (!kvm_regs)
1680                         goto out;
1681                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1682                 if (r)
1683                         goto out_free1;
1684                 r = -EFAULT;
1685                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1686                         goto out_free1;
1687                 r = 0;
1688 out_free1:
1689                 kfree(kvm_regs);
1690                 break;
1691         }
1692         case KVM_SET_REGS: {
1693                 struct kvm_regs *kvm_regs;
1694
1695                 r = -ENOMEM;
1696                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1697                 if (!kvm_regs)
1698                         goto out;
1699                 r = -EFAULT;
1700                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1701                         goto out_free2;
1702                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1703                 if (r)
1704                         goto out_free2;
1705                 r = 0;
1706 out_free2:
1707                 kfree(kvm_regs);
1708                 break;
1709         }
1710         case KVM_GET_SREGS: {
1711                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1712                 r = -ENOMEM;
1713                 if (!kvm_sregs)
1714                         goto out;
1715                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1716                 if (r)
1717                         goto out;
1718                 r = -EFAULT;
1719                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1720                         goto out;
1721                 r = 0;
1722                 break;
1723         }
1724         case KVM_SET_SREGS: {
1725                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1726                 r = -ENOMEM;
1727                 if (!kvm_sregs)
1728                         goto out;
1729                 r = -EFAULT;
1730                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1731                         goto out;
1732                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1733                 if (r)
1734                         goto out;
1735                 r = 0;
1736                 break;
1737         }
1738         case KVM_GET_MP_STATE: {
1739                 struct kvm_mp_state mp_state;
1740
1741                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1742                 if (r)
1743                         goto out;
1744                 r = -EFAULT;
1745                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1746                         goto out;
1747                 r = 0;
1748                 break;
1749         }
1750         case KVM_SET_MP_STATE: {
1751                 struct kvm_mp_state mp_state;
1752
1753                 r = -EFAULT;
1754                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1755                         goto out;
1756                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1757                 if (r)
1758                         goto out;
1759                 r = 0;
1760                 break;
1761         }
1762         case KVM_TRANSLATE: {
1763                 struct kvm_translation tr;
1764
1765                 r = -EFAULT;
1766                 if (copy_from_user(&tr, argp, sizeof tr))
1767                         goto out;
1768                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1769                 if (r)
1770                         goto out;
1771                 r = -EFAULT;
1772                 if (copy_to_user(argp, &tr, sizeof tr))
1773                         goto out;
1774                 r = 0;
1775                 break;
1776         }
1777         case KVM_SET_GUEST_DEBUG: {
1778                 struct kvm_guest_debug dbg;
1779
1780                 r = -EFAULT;
1781                 if (copy_from_user(&dbg, argp, sizeof dbg))
1782                         goto out;
1783                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1784                 if (r)
1785                         goto out;
1786                 r = 0;
1787                 break;
1788         }
1789         case KVM_SET_SIGNAL_MASK: {
1790                 struct kvm_signal_mask __user *sigmask_arg = argp;
1791                 struct kvm_signal_mask kvm_sigmask;
1792                 sigset_t sigset, *p;
1793
1794                 p = NULL;
1795                 if (argp) {
1796                         r = -EFAULT;
1797                         if (copy_from_user(&kvm_sigmask, argp,
1798                                            sizeof kvm_sigmask))
1799                                 goto out;
1800                         r = -EINVAL;
1801                         if (kvm_sigmask.len != sizeof sigset)
1802                                 goto out;
1803                         r = -EFAULT;
1804                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1805                                            sizeof sigset))
1806                                 goto out;
1807                         p = &sigset;
1808                 }
1809                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1810                 break;
1811         }
1812         case KVM_GET_FPU: {
1813                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1814                 r = -ENOMEM;
1815                 if (!fpu)
1816                         goto out;
1817                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1818                 if (r)
1819                         goto out;
1820                 r = -EFAULT;
1821                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1822                         goto out;
1823                 r = 0;
1824                 break;
1825         }
1826         case KVM_SET_FPU: {
1827                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1828                 r = -ENOMEM;
1829                 if (!fpu)
1830                         goto out;
1831                 r = -EFAULT;
1832                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1833                         goto out;
1834                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1835                 if (r)
1836                         goto out;
1837                 r = 0;
1838                 break;
1839         }
1840         default:
1841                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1842         }
1843 out:
1844         kfree(fpu);
1845         kfree(kvm_sregs);
1846         return r;
1847 }
1848
1849 static long kvm_vm_ioctl(struct file *filp,
1850                            unsigned int ioctl, unsigned long arg)
1851 {
1852         struct kvm *kvm = filp->private_data;
1853         void __user *argp = (void __user *)arg;
1854         int r;
1855
1856         if (kvm->mm != current->mm)
1857                 return -EIO;
1858         switch (ioctl) {
1859         case KVM_CREATE_VCPU:
1860                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1861                 if (r < 0)
1862                         goto out;
1863                 break;
1864         case KVM_SET_USER_MEMORY_REGION: {
1865                 struct kvm_userspace_memory_region kvm_userspace_mem;
1866
1867                 r = -EFAULT;
1868                 if (copy_from_user(&kvm_userspace_mem, argp,
1869                                                 sizeof kvm_userspace_mem))
1870                         goto out;
1871
1872                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1873                 if (r)
1874                         goto out;
1875                 break;
1876         }
1877         case KVM_GET_DIRTY_LOG: {
1878                 struct kvm_dirty_log log;
1879
1880                 r = -EFAULT;
1881                 if (copy_from_user(&log, argp, sizeof log))
1882                         goto out;
1883                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1884                 if (r)
1885                         goto out;
1886                 break;
1887         }
1888 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1889         case KVM_REGISTER_COALESCED_MMIO: {
1890                 struct kvm_coalesced_mmio_zone zone;
1891                 r = -EFAULT;
1892                 if (copy_from_user(&zone, argp, sizeof zone))
1893                         goto out;
1894                 r = -ENXIO;
1895                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1896                 if (r)
1897                         goto out;
1898                 r = 0;
1899                 break;
1900         }
1901         case KVM_UNREGISTER_COALESCED_MMIO: {
1902                 struct kvm_coalesced_mmio_zone zone;
1903                 r = -EFAULT;
1904                 if (copy_from_user(&zone, argp, sizeof zone))
1905                         goto out;
1906                 r = -ENXIO;
1907                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1908                 if (r)
1909                         goto out;
1910                 r = 0;
1911                 break;
1912         }
1913 #endif
1914 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1915         case KVM_ASSIGN_PCI_DEVICE: {
1916                 struct kvm_assigned_pci_dev assigned_dev;
1917
1918                 r = -EFAULT;
1919                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1920                         goto out;
1921                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1922                 if (r)
1923                         goto out;
1924                 break;
1925         }
1926         case KVM_ASSIGN_IRQ: {
1927                 struct kvm_assigned_irq assigned_irq;
1928
1929                 r = -EFAULT;
1930                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1931                         goto out;
1932                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1933                 if (r)
1934                         goto out;
1935                 break;
1936         }
1937 #endif
1938 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
1939         case KVM_DEASSIGN_PCI_DEVICE: {
1940                 struct kvm_assigned_pci_dev assigned_dev;
1941
1942                 r = -EFAULT;
1943                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1944                         goto out;
1945                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
1946                 if (r)
1947                         goto out;
1948                 break;
1949         }
1950 #endif
1951 #ifdef KVM_CAP_IRQ_ROUTING
1952         case KVM_SET_GSI_ROUTING: {
1953                 struct kvm_irq_routing routing;
1954                 struct kvm_irq_routing __user *urouting;
1955                 struct kvm_irq_routing_entry *entries;
1956
1957                 r = -EFAULT;
1958                 if (copy_from_user(&routing, argp, sizeof(routing)))
1959                         goto out;
1960                 r = -EINVAL;
1961                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
1962                         goto out;
1963                 if (routing.flags)
1964                         goto out;
1965                 r = -ENOMEM;
1966                 entries = vmalloc(routing.nr * sizeof(*entries));
1967                 if (!entries)
1968                         goto out;
1969                 r = -EFAULT;
1970                 urouting = argp;
1971                 if (copy_from_user(entries, urouting->entries,
1972                                    routing.nr * sizeof(*entries)))
1973                         goto out_free_irq_routing;
1974                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
1975                                         routing.flags);
1976         out_free_irq_routing:
1977                 vfree(entries);
1978                 break;
1979         }
1980 #endif
1981         default:
1982                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1983         }
1984 out:
1985         return r;
1986 }
1987
1988 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1989 {
1990         struct page *page[1];
1991         unsigned long addr;
1992         int npages;
1993         gfn_t gfn = vmf->pgoff;
1994         struct kvm *kvm = vma->vm_file->private_data;
1995
1996         addr = gfn_to_hva(kvm, gfn);
1997         if (kvm_is_error_hva(addr))
1998                 return VM_FAULT_SIGBUS;
1999
2000         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2001                                 NULL);
2002         if (unlikely(npages != 1))
2003                 return VM_FAULT_SIGBUS;
2004
2005         vmf->page = page[0];
2006         return 0;
2007 }
2008
2009 static struct vm_operations_struct kvm_vm_vm_ops = {
2010         .fault = kvm_vm_fault,
2011 };
2012
2013 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2014 {
2015         vma->vm_ops = &kvm_vm_vm_ops;
2016         return 0;
2017 }
2018
2019 static struct file_operations kvm_vm_fops = {
2020         .release        = kvm_vm_release,
2021         .unlocked_ioctl = kvm_vm_ioctl,
2022         .compat_ioctl   = kvm_vm_ioctl,
2023         .mmap           = kvm_vm_mmap,
2024 };
2025
2026 static int kvm_dev_ioctl_create_vm(void)
2027 {
2028         int fd;
2029         struct kvm *kvm;
2030
2031         kvm = kvm_create_vm();
2032         if (IS_ERR(kvm))
2033                 return PTR_ERR(kvm);
2034         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
2035         if (fd < 0)
2036                 kvm_put_kvm(kvm);
2037
2038         return fd;
2039 }
2040
2041 static long kvm_dev_ioctl_check_extension_generic(long arg)
2042 {
2043         switch (arg) {
2044         case KVM_CAP_USER_MEMORY:
2045         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2046                 return 1;
2047 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2048         case KVM_CAP_IRQ_ROUTING:
2049                 return 1;
2050 #endif
2051         default:
2052                 break;
2053         }
2054         return kvm_dev_ioctl_check_extension(arg);
2055 }
2056
2057 static long kvm_dev_ioctl(struct file *filp,
2058                           unsigned int ioctl, unsigned long arg)
2059 {
2060         long r = -EINVAL;
2061
2062         switch (ioctl) {
2063         case KVM_GET_API_VERSION:
2064                 r = -EINVAL;
2065                 if (arg)
2066                         goto out;
2067                 r = KVM_API_VERSION;
2068                 break;
2069         case KVM_CREATE_VM:
2070                 r = -EINVAL;
2071                 if (arg)
2072                         goto out;
2073                 r = kvm_dev_ioctl_create_vm();
2074                 break;
2075         case KVM_CHECK_EXTENSION:
2076                 r = kvm_dev_ioctl_check_extension_generic(arg);
2077                 break;
2078         case KVM_GET_VCPU_MMAP_SIZE:
2079                 r = -EINVAL;
2080                 if (arg)
2081                         goto out;
2082                 r = PAGE_SIZE;     /* struct kvm_run */
2083 #ifdef CONFIG_X86
2084                 r += PAGE_SIZE;    /* pio data page */
2085 #endif
2086 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2087                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2088 #endif
2089                 break;
2090         case KVM_TRACE_ENABLE:
2091         case KVM_TRACE_PAUSE:
2092         case KVM_TRACE_DISABLE:
2093                 r = kvm_trace_ioctl(ioctl, arg);
2094                 break;
2095         default:
2096                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2097         }
2098 out:
2099         return r;
2100 }
2101
2102 static struct file_operations kvm_chardev_ops = {
2103         .unlocked_ioctl = kvm_dev_ioctl,
2104         .compat_ioctl   = kvm_dev_ioctl,
2105 };
2106
2107 static struct miscdevice kvm_dev = {
2108         KVM_MINOR,
2109         "kvm",
2110         &kvm_chardev_ops,
2111 };
2112
2113 static void hardware_enable(void *junk)
2114 {
2115         int cpu = raw_smp_processor_id();
2116
2117         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2118                 return;
2119         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2120         kvm_arch_hardware_enable(NULL);
2121 }
2122
2123 static void hardware_disable(void *junk)
2124 {
2125         int cpu = raw_smp_processor_id();
2126
2127         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2128                 return;
2129         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2130         kvm_arch_hardware_disable(NULL);
2131 }
2132
2133 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2134                            void *v)
2135 {
2136         int cpu = (long)v;
2137
2138         val &= ~CPU_TASKS_FROZEN;
2139         switch (val) {
2140         case CPU_DYING:
2141                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2142                        cpu);
2143                 hardware_disable(NULL);
2144                 break;
2145         case CPU_UP_CANCELED:
2146                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2147                        cpu);
2148                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2149                 break;
2150         case CPU_ONLINE:
2151                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2152                        cpu);
2153                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2154                 break;
2155         }
2156         return NOTIFY_OK;
2157 }
2158
2159
2160 asmlinkage void kvm_handle_fault_on_reboot(void)
2161 {
2162         if (kvm_rebooting)
2163                 /* spin while reset goes on */
2164                 while (true)
2165                         ;
2166         /* Fault while not rebooting.  We want the trace. */
2167         BUG();
2168 }
2169 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2170
2171 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2172                       void *v)
2173 {
2174         if (val == SYS_RESTART) {
2175                 /*
2176                  * Some (well, at least mine) BIOSes hang on reboot if
2177                  * in vmx root mode.
2178                  */
2179                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2180                 kvm_rebooting = true;
2181                 on_each_cpu(hardware_disable, NULL, 1);
2182         }
2183         return NOTIFY_OK;
2184 }
2185
2186 static struct notifier_block kvm_reboot_notifier = {
2187         .notifier_call = kvm_reboot,
2188         .priority = 0,
2189 };
2190
2191 void kvm_io_bus_init(struct kvm_io_bus *bus)
2192 {
2193         memset(bus, 0, sizeof(*bus));
2194 }
2195
2196 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2197 {
2198         int i;
2199
2200         for (i = 0; i < bus->dev_count; i++) {
2201                 struct kvm_io_device *pos = bus->devs[i];
2202
2203                 kvm_iodevice_destructor(pos);
2204         }
2205 }
2206
2207 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2208                                           gpa_t addr, int len, int is_write)
2209 {
2210         int i;
2211
2212         for (i = 0; i < bus->dev_count; i++) {
2213                 struct kvm_io_device *pos = bus->devs[i];
2214
2215                 if (pos->in_range(pos, addr, len, is_write))
2216                         return pos;
2217         }
2218
2219         return NULL;
2220 }
2221
2222 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2223 {
2224         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2225
2226         bus->devs[bus->dev_count++] = dev;
2227 }
2228
2229 static struct notifier_block kvm_cpu_notifier = {
2230         .notifier_call = kvm_cpu_hotplug,
2231         .priority = 20, /* must be > scheduler priority */
2232 };
2233
2234 static int vm_stat_get(void *_offset, u64 *val)
2235 {
2236         unsigned offset = (long)_offset;
2237         struct kvm *kvm;
2238
2239         *val = 0;
2240         spin_lock(&kvm_lock);
2241         list_for_each_entry(kvm, &vm_list, vm_list)
2242                 *val += *(u32 *)((void *)kvm + offset);
2243         spin_unlock(&kvm_lock);
2244         return 0;
2245 }
2246
2247 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2248
2249 static int vcpu_stat_get(void *_offset, u64 *val)
2250 {
2251         unsigned offset = (long)_offset;
2252         struct kvm *kvm;
2253         struct kvm_vcpu *vcpu;
2254         int i;
2255
2256         *val = 0;
2257         spin_lock(&kvm_lock);
2258         list_for_each_entry(kvm, &vm_list, vm_list)
2259                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2260                         vcpu = kvm->vcpus[i];
2261                         if (vcpu)
2262                                 *val += *(u32 *)((void *)vcpu + offset);
2263                 }
2264         spin_unlock(&kvm_lock);
2265         return 0;
2266 }
2267
2268 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2269
2270 static struct file_operations *stat_fops[] = {
2271         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2272         [KVM_STAT_VM]   = &vm_stat_fops,
2273 };
2274
2275 static void kvm_init_debug(void)
2276 {
2277         struct kvm_stats_debugfs_item *p;
2278
2279         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2280         for (p = debugfs_entries; p->name; ++p)
2281                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2282                                                 (void *)(long)p->offset,
2283                                                 stat_fops[p->kind]);
2284 }
2285
2286 static void kvm_exit_debug(void)
2287 {
2288         struct kvm_stats_debugfs_item *p;
2289
2290         for (p = debugfs_entries; p->name; ++p)
2291                 debugfs_remove(p->dentry);
2292         debugfs_remove(kvm_debugfs_dir);
2293 }
2294
2295 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2296 {
2297         hardware_disable(NULL);
2298         return 0;
2299 }
2300
2301 static int kvm_resume(struct sys_device *dev)
2302 {
2303         hardware_enable(NULL);
2304         return 0;
2305 }
2306
2307 static struct sysdev_class kvm_sysdev_class = {
2308         .name = "kvm",
2309         .suspend = kvm_suspend,
2310         .resume = kvm_resume,
2311 };
2312
2313 static struct sys_device kvm_sysdev = {
2314         .id = 0,
2315         .cls = &kvm_sysdev_class,
2316 };
2317
2318 struct page *bad_page;
2319 pfn_t bad_pfn;
2320
2321 static inline
2322 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2323 {
2324         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2325 }
2326
2327 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2328 {
2329         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2330
2331         kvm_arch_vcpu_load(vcpu, cpu);
2332 }
2333
2334 static void kvm_sched_out(struct preempt_notifier *pn,
2335                           struct task_struct *next)
2336 {
2337         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2338
2339         kvm_arch_vcpu_put(vcpu);
2340 }
2341
2342 int kvm_init(void *opaque, unsigned int vcpu_size,
2343                   struct module *module)
2344 {
2345         int r;
2346         int cpu;
2347
2348         kvm_init_debug();
2349
2350         r = kvm_arch_init(opaque);
2351         if (r)
2352                 goto out_fail;
2353
2354         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2355
2356         if (bad_page == NULL) {
2357                 r = -ENOMEM;
2358                 goto out;
2359         }
2360
2361         bad_pfn = page_to_pfn(bad_page);
2362
2363         if (!alloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2364                 r = -ENOMEM;
2365                 goto out_free_0;
2366         }
2367
2368         r = kvm_arch_hardware_setup();
2369         if (r < 0)
2370                 goto out_free_0a;
2371
2372         for_each_online_cpu(cpu) {
2373                 smp_call_function_single(cpu,
2374                                 kvm_arch_check_processor_compat,
2375                                 &r, 1);
2376                 if (r < 0)
2377                         goto out_free_1;
2378         }
2379
2380         on_each_cpu(hardware_enable, NULL, 1);
2381         r = register_cpu_notifier(&kvm_cpu_notifier);
2382         if (r)
2383                 goto out_free_2;
2384         register_reboot_notifier(&kvm_reboot_notifier);
2385
2386         r = sysdev_class_register(&kvm_sysdev_class);
2387         if (r)
2388                 goto out_free_3;
2389
2390         r = sysdev_register(&kvm_sysdev);
2391         if (r)
2392                 goto out_free_4;
2393
2394         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2395         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2396                                            __alignof__(struct kvm_vcpu),
2397                                            0, NULL);
2398         if (!kvm_vcpu_cache) {
2399                 r = -ENOMEM;
2400                 goto out_free_5;
2401         }
2402
2403         kvm_chardev_ops.owner = module;
2404         kvm_vm_fops.owner = module;
2405         kvm_vcpu_fops.owner = module;
2406
2407         r = misc_register(&kvm_dev);
2408         if (r) {
2409                 printk(KERN_ERR "kvm: misc device register failed\n");
2410                 goto out_free;
2411         }
2412
2413         kvm_preempt_ops.sched_in = kvm_sched_in;
2414         kvm_preempt_ops.sched_out = kvm_sched_out;
2415 #ifndef CONFIG_X86
2416         msi2intx = 0;
2417 #endif
2418
2419         return 0;
2420
2421 out_free:
2422         kmem_cache_destroy(kvm_vcpu_cache);
2423 out_free_5:
2424         sysdev_unregister(&kvm_sysdev);
2425 out_free_4:
2426         sysdev_class_unregister(&kvm_sysdev_class);
2427 out_free_3:
2428         unregister_reboot_notifier(&kvm_reboot_notifier);
2429         unregister_cpu_notifier(&kvm_cpu_notifier);
2430 out_free_2:
2431         on_each_cpu(hardware_disable, NULL, 1);
2432 out_free_1:
2433         kvm_arch_hardware_unsetup();
2434 out_free_0a:
2435         free_cpumask_var(cpus_hardware_enabled);
2436 out_free_0:
2437         __free_page(bad_page);
2438 out:
2439         kvm_arch_exit();
2440         kvm_exit_debug();
2441 out_fail:
2442         return r;
2443 }
2444 EXPORT_SYMBOL_GPL(kvm_init);
2445
2446 void kvm_exit(void)
2447 {
2448         kvm_trace_cleanup();
2449         misc_deregister(&kvm_dev);
2450         kmem_cache_destroy(kvm_vcpu_cache);
2451         sysdev_unregister(&kvm_sysdev);
2452         sysdev_class_unregister(&kvm_sysdev_class);
2453         unregister_reboot_notifier(&kvm_reboot_notifier);
2454         unregister_cpu_notifier(&kvm_cpu_notifier);
2455         on_each_cpu(hardware_disable, NULL, 1);
2456         kvm_arch_hardware_unsetup();
2457         kvm_arch_exit();
2458         kvm_exit_debug();
2459         free_cpumask_var(cpus_hardware_enabled);
2460         __free_page(bad_page);
2461 }
2462 EXPORT_SYMBOL_GPL(kvm_exit);