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