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