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