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 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
 534 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
 535                 struct kvm_assigned_pci_dev *assigned_dev)
 536 {
 537         int r = 0;
 538         struct kvm_assigned_dev_kernel *match;
 539
 540         mutex_lock(&kvm->lock);
 541
 542         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 543                                       assigned_dev->assigned_dev_id);
 544         if (!match) {
 545                 printk(KERN_INFO "%s: device hasn't been assigned before, "
 546                   "so cannot be deassigned\n", __func__);
 547                 r = -EINVAL;
 548                 goto out;
 549         }
 550
 551         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
 552                 kvm_deassign_device(kvm, match);
 553
 554         kvm_free_assigned_device(kvm, match);
 555
 556 out:
 557         mutex_unlock(&kvm->lock);
 558         return r;
 559 }
 560 #endif
 561
 562 static inline int valid_vcpu(int n)
 563 {
 564         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 565 }
 566
 567 inline int kvm_is_mmio_pfn(pfn_t pfn)
 568 {
 569         if (pfn_valid(pfn))
 570                 return PageReserved(pfn_to_page(pfn));
 571
 572         return true;
 573 }
 574
 575 /*
 576  * Switches to specified vcpu, until a matching vcpu_put()
 577  */
 578 void vcpu_load(struct kvm_vcpu *vcpu)
 579 {
 580         int cpu;
 581
 582         mutex_lock(&vcpu->mutex);
 583         cpu = get_cpu();
 584         preempt_notifier_register(&vcpu->preempt_notifier);
 585         kvm_arch_vcpu_load(vcpu, cpu);
 586         put_cpu();
 587 }
 588
 589 void vcpu_put(struct kvm_vcpu *vcpu)
 590 {
 591         preempt_disable();
 592         kvm_arch_vcpu_put(vcpu);
 593         preempt_notifier_unregister(&vcpu->preempt_notifier);
 594         preempt_enable();
 595         mutex_unlock(&vcpu->mutex);
 596 }
 597
 598 static void ack_flush(void *_completed)
 599 {
 600 }
 601
 602 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 603 {
 604         int i, cpu, me;
 605         cpumask_var_t cpus;
 606         bool called = true;
 607         struct kvm_vcpu *vcpu;
 608
 609         if (alloc_cpumask_var(&cpus, GFP_ATOMIC))
 610                 cpumask_clear(cpus);
 611
 612         me = get_cpu();
 613         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 614                 vcpu = kvm->vcpus[i];
 615                 if (!vcpu)
 616                         continue;
 617                 if (test_and_set_bit(req, &vcpu->requests))
 618                         continue;
 619                 cpu = vcpu->cpu;
 620                 if (cpus != NULL && cpu != -1 && cpu != me)
 621                         cpumask_set_cpu(cpu, cpus);
 622         }
 623         if (unlikely(cpus == NULL))
 624                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 625         else if (!cpumask_empty(cpus))
 626                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 627         else
 628                 called = false;
 629         put_cpu();
 630         free_cpumask_var(cpus);
 631         return called;
 632 }
 633
 634 void kvm_flush_remote_tlbs(struct kvm *kvm)
 635 {
 636         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 637                 ++kvm->stat.remote_tlb_flush;
 638 }
 639
 640 void kvm_reload_remote_mmus(struct kvm *kvm)
 641 {
 642         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 643 }
 644
 645 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 646 {
 647         struct page *page;
 648         int r;
 649
 650         mutex_init(&vcpu->mutex);
 651         vcpu->cpu = -1;
 652         vcpu->kvm = kvm;
 653         vcpu->vcpu_id = id;
 654         init_waitqueue_head(&vcpu->wq);
 655
 656         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 657         if (!page) {
 658                 r = -ENOMEM;
 659                 goto fail;
 660         }
 661         vcpu->run = page_address(page);
 662
 663         r = kvm_arch_vcpu_init(vcpu);
 664         if (r < 0)
 665                 goto fail_free_run;
 666         return 0;
 667
 668 fail_free_run:
 669         free_page((unsigned long)vcpu->run);
 670 fail:
 671         return r;
 672 }
 673 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 674
 675 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 676 {
 677         kvm_arch_vcpu_uninit(vcpu);
 678         free_page((unsigned long)vcpu->run);
 679 }
 680 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 681
 682 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 683 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 684 {
 685         return container_of(mn, struct kvm, mmu_notifier);
 686 }
 687
 688 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 689                                              struct mm_struct *mm,
 690                                              unsigned long address)
 691 {
 692         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 693         int need_tlb_flush;
 694
 695         /*
 696          * When ->invalidate_page runs, the linux pte has been zapped
 697          * already but the page is still allocated until
 698          * ->invalidate_page returns. So if we increase the sequence
 699          * here the kvm page fault will notice if the spte can't be
 700          * established because the page is going to be freed. If
 701          * instead the kvm page fault establishes the spte before
 702          * ->invalidate_page runs, kvm_unmap_hva will release it
 703          * before returning.
 704          *
 705          * The sequence increase only need to be seen at spin_unlock
 706          * time, and not at spin_lock time.
 707          *
 708          * Increasing the sequence after the spin_unlock would be
 709          * unsafe because the kvm page fault could then establish the
 710          * pte after kvm_unmap_hva returned, without noticing the page
 711          * is going to be freed.
 712          */
 713         spin_lock(&kvm->mmu_lock);
 714         kvm->mmu_notifier_seq++;
 715         need_tlb_flush = kvm_unmap_hva(kvm, address);
 716         spin_unlock(&kvm->mmu_lock);
 717
 718         /* we've to flush the tlb before the pages can be freed */
 719         if (need_tlb_flush)
 720                 kvm_flush_remote_tlbs(kvm);
 721
 722 }
 723
 724 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 725                                                     struct mm_struct *mm,
 726                                                     unsigned long start,
 727                                                     unsigned long end)
 728 {
 729         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 730         int need_tlb_flush = 0;
 731
 732         spin_lock(&kvm->mmu_lock);
 733         /*
 734          * The count increase must become visible at unlock time as no
 735          * spte can be established without taking the mmu_lock and
 736          * count is also read inside the mmu_lock critical section.
 737          */
 738         kvm->mmu_notifier_count++;
 739         for (; start < end; start += PAGE_SIZE)
 740                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 741         spin_unlock(&kvm->mmu_lock);
 742
 743         /* we've to flush the tlb before the pages can be freed */
 744         if (need_tlb_flush)
 745                 kvm_flush_remote_tlbs(kvm);
 746 }
 747
 748 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 749                                                   struct mm_struct *mm,
 750                                                   unsigned long start,
 751                                                   unsigned long end)
 752 {
 753         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 754
 755         spin_lock(&kvm->mmu_lock);
 756         /*
 757          * This sequence increase will notify the kvm page fault that
 758          * the page that is going to be mapped in the spte could have
 759          * been freed.
 760          */
 761         kvm->mmu_notifier_seq++;
 762         /*
 763          * The above sequence increase must be visible before the
 764          * below count decrease but both values are read by the kvm
 765          * page fault under mmu_lock spinlock so we don't need to add
 766          * a smb_wmb() here in between the two.
 767          */
 768         kvm->mmu_notifier_count--;
 769         spin_unlock(&kvm->mmu_lock);
 770
 771         BUG_ON(kvm->mmu_notifier_count < 0);
 772 }
 773
 774 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 775                                               struct mm_struct *mm,
 776                                               unsigned long address)
 777 {
 778         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 779         int young;
 780
 781         spin_lock(&kvm->mmu_lock);
 782         young = kvm_age_hva(kvm, address);
 783         spin_unlock(&kvm->mmu_lock);
 784
 785         if (young)
 786                 kvm_flush_remote_tlbs(kvm);
 787
 788         return young;
 789 }
 790
 791 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 792         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 793         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 794         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 795         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 796 };
 797 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 798
 799 static struct kvm *kvm_create_vm(void)
 800 {
 801         struct kvm *kvm = kvm_arch_create_vm();
 802 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 803         struct page *page;
 804 #endif
 805
 806         if (IS_ERR(kvm))
 807                 goto out;
 808
 809 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 810         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 811         if (!page) {
 812                 kfree(kvm);
 813                 return ERR_PTR(-ENOMEM);
 814         }
 815         kvm->coalesced_mmio_ring =
 816                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 817 #endif
 818
 819 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 820         {
 821                 int err;
 822                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 823                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 824                 if (err) {
 825 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 826                         put_page(page);
 827 #endif
 828                         kfree(kvm);
 829                         return ERR_PTR(err);
 830                 }
 831         }
 832 #endif
 833
 834         kvm->mm = current->mm;
 835         atomic_inc(&kvm->mm->mm_count);
 836         spin_lock_init(&kvm->mmu_lock);
 837         kvm_io_bus_init(&kvm->pio_bus);
 838         mutex_init(&kvm->lock);
 839         kvm_io_bus_init(&kvm->mmio_bus);
 840         init_rwsem(&kvm->slots_lock);
 841         atomic_set(&kvm->users_count, 1);
 842         spin_lock(&kvm_lock);
 843         list_add(&kvm->vm_list, &vm_list);
 844         spin_unlock(&kvm_lock);
 845 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 846         kvm_coalesced_mmio_init(kvm);
 847 #endif
 848 out:
 849         return kvm;
 850 }
 851
 852 /*
 853  * Free any memory in @free but not in @dont.
 854  */
 855 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 856                                   struct kvm_memory_slot *dont)
 857 {
 858         if (!dont || free->rmap != dont->rmap)
 859                 vfree(free->rmap);
 860
 861         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 862                 vfree(free->dirty_bitmap);
 863
 864         if (!dont || free->lpage_info != dont->lpage_info)
 865                 vfree(free->lpage_info);
 866
 867         free->npages = 0;
 868         free->dirty_bitmap = NULL;
 869         free->rmap = NULL;
 870         free->lpage_info = NULL;
 871 }
 872
 873 void kvm_free_physmem(struct kvm *kvm)
 874 {
 875         int i;
 876
 877         for (i = 0; i < kvm->nmemslots; ++i)
 878                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 879 }
 880
 881 static void kvm_destroy_vm(struct kvm *kvm)
 882 {
 883         struct mm_struct *mm = kvm->mm;
 884
 885         spin_lock(&kvm_lock);
 886         list_del(&kvm->vm_list);
 887         spin_unlock(&kvm_lock);
 888         kvm_io_bus_destroy(&kvm->pio_bus);
 889         kvm_io_bus_destroy(&kvm->mmio_bus);
 890 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 891         if (kvm->coalesced_mmio_ring != NULL)
 892                 free_page((unsigned long)kvm->coalesced_mmio_ring);
 893 #endif
 894 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 895         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 896 #endif
 897         kvm_arch_destroy_vm(kvm);
 898         mmdrop(mm);
 899 }
 900
 901 void kvm_get_kvm(struct kvm *kvm)
 902 {
 903         atomic_inc(&kvm->users_count);
 904 }
 905 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 906
 907 void kvm_put_kvm(struct kvm *kvm)
 908 {
 909         if (atomic_dec_and_test(&kvm->users_count))
 910                 kvm_destroy_vm(kvm);
 911 }
 912 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 913
 914
 915 static int kvm_vm_release(struct inode *inode, struct file *filp)
 916 {
 917         struct kvm *kvm = filp->private_data;
 918
 919         kvm_put_kvm(kvm);
 920         return 0;
 921 }
 922
 923 /*
 924  * Allocate some memory and give it an address in the guest physical address
 925  * space.
 926  *
 927  * Discontiguous memory is allowed, mostly for framebuffers.
 928  *
 929  * Must be called holding mmap_sem for write.
 930  */
 931 int __kvm_set_memory_region(struct kvm *kvm,
 932                             struct kvm_userspace_memory_region *mem,
 933                             int user_alloc)
 934 {
 935         int r;
 936         gfn_t base_gfn;
 937         unsigned long npages;
 938         unsigned long i;
 939         struct kvm_memory_slot *memslot;
 940         struct kvm_memory_slot old, new;
 941
 942         r = -EINVAL;
 943         /* General sanity checks */
 944         if (mem->memory_size & (PAGE_SIZE - 1))
 945                 goto out;
 946         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 947                 goto out;
 948         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 949                 goto out;
 950         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 951                 goto out;
 952         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 953                 goto out;
 954
 955         memslot = &kvm->memslots[mem->slot];
 956         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 957         npages = mem->memory_size >> PAGE_SHIFT;
 958
 959         if (!npages)
 960                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 961
 962         new = old = *memslot;
 963
 964         new.base_gfn = base_gfn;
 965         new.npages = npages;
 966         new.flags = mem->flags;
 967
 968         /* Disallow changing a memory slot's size. */
 969         r = -EINVAL;
 970         if (npages && old.npages && npages != old.npages)
 971                 goto out_free;
 972
 973         /* Check for overlaps */
 974         r = -EEXIST;
 975         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 976                 struct kvm_memory_slot *s = &kvm->memslots[i];
 977
 978                 if (s == memslot)
 979                         continue;
 980                 if (!((base_gfn + npages <= s->base_gfn) ||
 981                       (base_gfn >= s->base_gfn + s->npages)))
 982                         goto out_free;
 983         }
 984
 985         /* Free page dirty bitmap if unneeded */
 986         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 987                 new.dirty_bitmap = NULL;
 988
 989         r = -ENOMEM;
 990
 991         /* Allocate if a slot is being created */
 992 #ifndef CONFIG_S390
 993         if (npages && !new.rmap) {
 994                 new.rmap = vmalloc(npages * sizeof(struct page *));
 995
 996                 if (!new.rmap)
 997                         goto out_free;
 998
 999                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
1000
1001                 new.user_alloc = user_alloc;
1002                 /*
1003                  * hva_to_rmmap() serialzies with the mmu_lock and to be
1004                  * safe it has to ignore memslots with !user_alloc &&
1005                  * !userspace_addr.
1006                  */
1007                 if (user_alloc)
1008                         new.userspace_addr = mem->userspace_addr;
1009                 else
1010                         new.userspace_addr = 0;
1011         }
1012         if (npages && !new.lpage_info) {
1013                 int largepages = npages / KVM_PAGES_PER_HPAGE;
1014                 if (npages % KVM_PAGES_PER_HPAGE)
1015                         largepages++;
1016                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1017                         largepages++;
1018
1019                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
1020
1021                 if (!new.lpage_info)
1022                         goto out_free;
1023
1024                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
1025
1026                 if (base_gfn % KVM_PAGES_PER_HPAGE)
1027                         new.lpage_info[0].write_count = 1;
1028                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
1029                         new.lpage_info[largepages-1].write_count = 1;
1030         }
1031
1032         /* Allocate page dirty bitmap if needed */
1033         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1034                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
1035
1036                 new.dirty_bitmap = vmalloc(dirty_bytes);
1037                 if (!new.dirty_bitmap)
1038                         goto out_free;
1039                 memset(new.dirty_bitmap, 0, dirty_bytes);
1040         }
1041 #endif /* not defined CONFIG_S390 */
1042
1043         if (!npages)
1044                 kvm_arch_flush_shadow(kvm);
1045
1046         spin_lock(&kvm->mmu_lock);
1047         if (mem->slot >= kvm->nmemslots)
1048                 kvm->nmemslots = mem->slot + 1;
1049
1050         *memslot = new;
1051         spin_unlock(&kvm->mmu_lock);
1052
1053         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1054         if (r) {
1055                 spin_lock(&kvm->mmu_lock);
1056                 *memslot = old;
1057                 spin_unlock(&kvm->mmu_lock);
1058                 goto out_free;
1059         }
1060
1061         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1062         /* Slot deletion case: we have to update the current slot */
1063         if (!npages)
1064                 *memslot = old;
1065 #ifdef CONFIG_DMAR
1066         /* map the pages in iommu page table */
1067         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1068         if (r)
1069                 goto out;
1070 #endif
1071         return 0;
1072
1073 out_free:
1074         kvm_free_physmem_slot(&new, &old);
1075 out:
1076         return r;
1077
1078 }
1079 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1080
1081 int kvm_set_memory_region(struct kvm *kvm,
1082                           struct kvm_userspace_memory_region *mem,
1083                           int user_alloc)
1084 {
1085         int r;
1086
1087         down_write(&kvm->slots_lock);
1088         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1089         up_write(&kvm->slots_lock);
1090         return r;
1091 }
1092 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1093
1094 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1095                                    struct
1096                                    kvm_userspace_memory_region *mem,
1097                                    int user_alloc)
1098 {
1099         if (mem->slot >= KVM_MEMORY_SLOTS)
1100                 return -EINVAL;
1101         return kvm_set_memory_region(kvm, mem, user_alloc);
1102 }
1103
1104 int kvm_get_dirty_log(struct kvm *kvm,
1105                         struct kvm_dirty_log *log, int *is_dirty)
1106 {
1107         struct kvm_memory_slot *memslot;
1108         int r, i;
1109         int n;
1110         unsigned long any = 0;
1111
1112         r = -EINVAL;
1113         if (log->slot >= KVM_MEMORY_SLOTS)
1114                 goto out;
1115
1116         memslot = &kvm->memslots[log->slot];
1117         r = -ENOENT;
1118         if (!memslot->dirty_bitmap)
1119                 goto out;
1120
1121         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1122
1123         for (i = 0; !any && i < n/sizeof(long); ++i)
1124                 any = memslot->dirty_bitmap[i];
1125
1126         r = -EFAULT;
1127         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1128                 goto out;
1129
1130         if (any)
1131                 *is_dirty = 1;
1132
1133         r = 0;
1134 out:
1135         return r;
1136 }
1137
1138 int is_error_page(struct page *page)
1139 {
1140         return page == bad_page;
1141 }
1142 EXPORT_SYMBOL_GPL(is_error_page);
1143
1144 int is_error_pfn(pfn_t pfn)
1145 {
1146         return pfn == bad_pfn;
1147 }
1148 EXPORT_SYMBOL_GPL(is_error_pfn);
1149
1150 static inline unsigned long bad_hva(void)
1151 {
1152         return PAGE_OFFSET;
1153 }
1154
1155 int kvm_is_error_hva(unsigned long addr)
1156 {
1157         return addr == bad_hva();
1158 }
1159 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1160
1161 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1162 {
1163         int i;
1164
1165         for (i = 0; i < kvm->nmemslots; ++i) {
1166                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1167
1168                 if (gfn >= memslot->base_gfn
1169                     && gfn < memslot->base_gfn + memslot->npages)
1170                         return memslot;
1171         }
1172         return NULL;
1173 }
1174 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1175
1176 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1177 {
1178         gfn = unalias_gfn(kvm, gfn);
1179         return gfn_to_memslot_unaliased(kvm, gfn);
1180 }
1181
1182 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1183 {
1184         int i;
1185
1186         gfn = unalias_gfn(kvm, gfn);
1187         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1188                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1189
1190                 if (gfn >= memslot->base_gfn
1191                     && gfn < memslot->base_gfn + memslot->npages)
1192                         return 1;
1193         }
1194         return 0;
1195 }
1196 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1197
1198 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1199 {
1200         struct kvm_memory_slot *slot;
1201
1202         gfn = unalias_gfn(kvm, gfn);
1203         slot = gfn_to_memslot_unaliased(kvm, gfn);
1204         if (!slot)
1205                 return bad_hva();
1206         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1207 }
1208 EXPORT_SYMBOL_GPL(gfn_to_hva);
1209
1210 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1211 {
1212         struct page *page[1];
1213         unsigned long addr;
1214         int npages;
1215         pfn_t pfn;
1216
1217         might_sleep();
1218
1219         addr = gfn_to_hva(kvm, gfn);
1220         if (kvm_is_error_hva(addr)) {
1221                 get_page(bad_page);
1222                 return page_to_pfn(bad_page);
1223         }
1224
1225         npages = get_user_pages_fast(addr, 1, 1, page);
1226
1227         if (unlikely(npages != 1)) {
1228                 struct vm_area_struct *vma;
1229
1230                 down_read(&current->mm->mmap_sem);
1231                 vma = find_vma(current->mm, addr);
1232
1233                 if (vma == NULL || addr < vma->vm_start ||
1234                     !(vma->vm_flags & VM_PFNMAP)) {
1235                         up_read(&current->mm->mmap_sem);
1236                         get_page(bad_page);
1237                         return page_to_pfn(bad_page);
1238                 }
1239
1240                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1241                 up_read(&current->mm->mmap_sem);
1242                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1243         } else
1244                 pfn = page_to_pfn(page[0]);
1245
1246         return pfn;
1247 }
1248
1249 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1250
1251 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1252 {
1253         pfn_t pfn;
1254
1255         pfn = gfn_to_pfn(kvm, gfn);
1256         if (!kvm_is_mmio_pfn(pfn))
1257                 return pfn_to_page(pfn);
1258
1259         WARN_ON(kvm_is_mmio_pfn(pfn));
1260
1261         get_page(bad_page);
1262         return bad_page;
1263 }
1264
1265 EXPORT_SYMBOL_GPL(gfn_to_page);
1266
1267 void kvm_release_page_clean(struct page *page)
1268 {
1269         kvm_release_pfn_clean(page_to_pfn(page));
1270 }
1271 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1272
1273 void kvm_release_pfn_clean(pfn_t pfn)
1274 {
1275         if (!kvm_is_mmio_pfn(pfn))
1276                 put_page(pfn_to_page(pfn));
1277 }
1278 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1279
1280 void kvm_release_page_dirty(struct page *page)
1281 {
1282         kvm_release_pfn_dirty(page_to_pfn(page));
1283 }
1284 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1285
1286 void kvm_release_pfn_dirty(pfn_t pfn)
1287 {
1288         kvm_set_pfn_dirty(pfn);
1289         kvm_release_pfn_clean(pfn);
1290 }
1291 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1292
1293 void kvm_set_page_dirty(struct page *page)
1294 {
1295         kvm_set_pfn_dirty(page_to_pfn(page));
1296 }
1297 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1298
1299 void kvm_set_pfn_dirty(pfn_t pfn)
1300 {
1301         if (!kvm_is_mmio_pfn(pfn)) {
1302                 struct page *page = pfn_to_page(pfn);
1303                 if (!PageReserved(page))
1304                         SetPageDirty(page);
1305         }
1306 }
1307 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1308
1309 void kvm_set_pfn_accessed(pfn_t pfn)
1310 {
1311         if (!kvm_is_mmio_pfn(pfn))
1312                 mark_page_accessed(pfn_to_page(pfn));
1313 }
1314 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1315
1316 void kvm_get_pfn(pfn_t pfn)
1317 {
1318         if (!kvm_is_mmio_pfn(pfn))
1319                 get_page(pfn_to_page(pfn));
1320 }
1321 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1322
1323 static int next_segment(unsigned long len, int offset)
1324 {
1325         if (len > PAGE_SIZE - offset)
1326                 return PAGE_SIZE - offset;
1327         else
1328                 return len;
1329 }
1330
1331 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1332                         int len)
1333 {
1334         int r;
1335         unsigned long addr;
1336
1337         addr = gfn_to_hva(kvm, gfn);
1338         if (kvm_is_error_hva(addr))
1339                 return -EFAULT;
1340         r = copy_from_user(data, (void __user *)addr + offset, len);
1341         if (r)
1342                 return -EFAULT;
1343         return 0;
1344 }
1345 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1346
1347 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1348 {
1349         gfn_t gfn = gpa >> PAGE_SHIFT;
1350         int seg;
1351         int offset = offset_in_page(gpa);
1352         int ret;
1353
1354         while ((seg = next_segment(len, offset)) != 0) {
1355                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1356                 if (ret < 0)
1357                         return ret;
1358                 offset = 0;
1359                 len -= seg;
1360                 data += seg;
1361                 ++gfn;
1362         }
1363         return 0;
1364 }
1365 EXPORT_SYMBOL_GPL(kvm_read_guest);
1366
1367 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1368                           unsigned long len)
1369 {
1370         int r;
1371         unsigned long addr;
1372         gfn_t gfn = gpa >> PAGE_SHIFT;
1373         int offset = offset_in_page(gpa);
1374
1375         addr = gfn_to_hva(kvm, gfn);
1376         if (kvm_is_error_hva(addr))
1377                 return -EFAULT;
1378         pagefault_disable();
1379         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1380         pagefault_enable();
1381         if (r)
1382                 return -EFAULT;
1383         return 0;
1384 }
1385 EXPORT_SYMBOL(kvm_read_guest_atomic);
1386
1387 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1388                          int offset, int len)
1389 {
1390         int r;
1391         unsigned long addr;
1392
1393         addr = gfn_to_hva(kvm, gfn);
1394         if (kvm_is_error_hva(addr))
1395                 return -EFAULT;
1396         r = copy_to_user((void __user *)addr + offset, data, len);
1397         if (r)
1398                 return -EFAULT;
1399         mark_page_dirty(kvm, gfn);
1400         return 0;
1401 }
1402 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1403
1404 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1405                     unsigned long len)
1406 {
1407         gfn_t gfn = gpa >> PAGE_SHIFT;
1408         int seg;
1409         int offset = offset_in_page(gpa);
1410         int ret;
1411
1412         while ((seg = next_segment(len, offset)) != 0) {
1413                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1414                 if (ret < 0)
1415                         return ret;
1416                 offset = 0;
1417                 len -= seg;
1418                 data += seg;
1419                 ++gfn;
1420         }
1421         return 0;
1422 }
1423
1424 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1425 {
1426         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1427 }
1428 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1429
1430 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1431 {
1432         gfn_t gfn = gpa >> PAGE_SHIFT;
1433         int seg;
1434         int offset = offset_in_page(gpa);
1435         int ret;
1436
1437         while ((seg = next_segment(len, offset)) != 0) {
1438                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1439                 if (ret < 0)
1440                         return ret;
1441                 offset = 0;
1442                 len -= seg;
1443                 ++gfn;
1444         }
1445         return 0;
1446 }
1447 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1448
1449 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1450 {
1451         struct kvm_memory_slot *memslot;
1452
1453         gfn = unalias_gfn(kvm, gfn);
1454         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1455         if (memslot && memslot->dirty_bitmap) {
1456                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1457
1458                 /* avoid RMW */
1459                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1460                         set_bit(rel_gfn, memslot->dirty_bitmap);
1461         }
1462 }
1463
1464 /*
1465  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1466  */
1467 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1468 {
1469         DEFINE_WAIT(wait);
1470
1471         for (;;) {
1472                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1473
1474                 if (kvm_cpu_has_interrupt(vcpu) ||
1475                     kvm_cpu_has_pending_timer(vcpu) ||
1476                     kvm_arch_vcpu_runnable(vcpu)) {
1477                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1478                         break;
1479                 }
1480                 if (signal_pending(current))
1481                         break;
1482
1483                 vcpu_put(vcpu);
1484                 schedule();
1485                 vcpu_load(vcpu);
1486         }
1487
1488         finish_wait(&vcpu->wq, &wait);
1489 }
1490
1491 void kvm_resched(struct kvm_vcpu *vcpu)
1492 {
1493         if (!need_resched())
1494                 return;
1495         cond_resched();
1496 }
1497 EXPORT_SYMBOL_GPL(kvm_resched);
1498
1499 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1500 {
1501         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1502         struct page *page;
1503
1504         if (vmf->pgoff == 0)
1505                 page = virt_to_page(vcpu->run);
1506 #ifdef CONFIG_X86
1507         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1508                 page = virt_to_page(vcpu->arch.pio_data);
1509 #endif
1510 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1511         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1512                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1513 #endif
1514         else
1515                 return VM_FAULT_SIGBUS;
1516         get_page(page);
1517         vmf->page = page;
1518         return 0;
1519 }
1520
1521 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1522         .fault = kvm_vcpu_fault,
1523 };
1524
1525 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1526 {
1527         vma->vm_ops = &kvm_vcpu_vm_ops;
1528         return 0;
1529 }
1530
1531 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1532 {
1533         struct kvm_vcpu *vcpu = filp->private_data;
1534
1535         kvm_put_kvm(vcpu->kvm);
1536         return 0;
1537 }
1538
1539 static struct file_operations kvm_vcpu_fops = {
1540         .release        = kvm_vcpu_release,
1541         .unlocked_ioctl = kvm_vcpu_ioctl,
1542         .compat_ioctl   = kvm_vcpu_ioctl,
1543         .mmap           = kvm_vcpu_mmap,
1544 };
1545
1546 /*
1547  * Allocates an inode for the vcpu.
1548  */
1549 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1550 {
1551         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1552         if (fd < 0)
1553                 kvm_put_kvm(vcpu->kvm);
1554         return fd;
1555 }
1556
1557 /*
1558  * Creates some virtual cpus.  Good luck creating more than one.
1559  */
1560 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1561 {
1562         int r;
1563         struct kvm_vcpu *vcpu;
1564
1565         if (!valid_vcpu(n))
1566                 return -EINVAL;
1567
1568         vcpu = kvm_arch_vcpu_create(kvm, n);
1569         if (IS_ERR(vcpu))
1570                 return PTR_ERR(vcpu);
1571
1572         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1573
1574         r = kvm_arch_vcpu_setup(vcpu);
1575         if (r)
1576                 return r;
1577
1578         mutex_lock(&kvm->lock);
1579         if (kvm->vcpus[n]) {
1580                 r = -EEXIST;
1581                 goto vcpu_destroy;
1582         }
1583         kvm->vcpus[n] = vcpu;
1584         mutex_unlock(&kvm->lock);
1585
1586         /* Now it's all set up, let userspace reach it */
1587         kvm_get_kvm(kvm);
1588         r = create_vcpu_fd(vcpu);
1589         if (r < 0)
1590                 goto unlink;
1591         return r;
1592
1593 unlink:
1594         mutex_lock(&kvm->lock);
1595         kvm->vcpus[n] = NULL;
1596 vcpu_destroy:
1597         mutex_unlock(&kvm->lock);
1598         kvm_arch_vcpu_destroy(vcpu);
1599         return r;
1600 }
1601
1602 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1603 {
1604         if (sigset) {
1605                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1606                 vcpu->sigset_active = 1;
1607                 vcpu->sigset = *sigset;
1608         } else
1609                 vcpu->sigset_active = 0;
1610         return 0;
1611 }
1612
1613 static long kvm_vcpu_ioctl(struct file *filp,
1614                            unsigned int ioctl, unsigned long arg)
1615 {
1616         struct kvm_vcpu *vcpu = filp->private_data;
1617         void __user *argp = (void __user *)arg;
1618         int r;
1619         struct kvm_fpu *fpu = NULL;
1620         struct kvm_sregs *kvm_sregs = NULL;
1621
1622         if (vcpu->kvm->mm != current->mm)
1623                 return -EIO;
1624         switch (ioctl) {
1625         case KVM_RUN:
1626                 r = -EINVAL;
1627                 if (arg)
1628                         goto out;
1629                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1630                 break;
1631         case KVM_GET_REGS: {
1632                 struct kvm_regs *kvm_regs;
1633
1634                 r = -ENOMEM;
1635                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1636                 if (!kvm_regs)
1637                         goto out;
1638                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1639                 if (r)
1640                         goto out_free1;
1641                 r = -EFAULT;
1642                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1643                         goto out_free1;
1644                 r = 0;
1645 out_free1:
1646                 kfree(kvm_regs);
1647                 break;
1648         }
1649         case KVM_SET_REGS: {
1650                 struct kvm_regs *kvm_regs;
1651
1652                 r = -ENOMEM;
1653                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1654                 if (!kvm_regs)
1655                         goto out;
1656                 r = -EFAULT;
1657                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1658                         goto out_free2;
1659                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1660                 if (r)
1661                         goto out_free2;
1662                 r = 0;
1663 out_free2:
1664                 kfree(kvm_regs);
1665                 break;
1666         }
1667         case KVM_GET_SREGS: {
1668                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1669                 r = -ENOMEM;
1670                 if (!kvm_sregs)
1671                         goto out;
1672                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1673                 if (r)
1674                         goto out;
1675                 r = -EFAULT;
1676                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1677                         goto out;
1678                 r = 0;
1679                 break;
1680         }
1681         case KVM_SET_SREGS: {
1682                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1683                 r = -ENOMEM;
1684                 if (!kvm_sregs)
1685                         goto out;
1686                 r = -EFAULT;
1687                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1688                         goto out;
1689                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1690                 if (r)
1691                         goto out;
1692                 r = 0;
1693                 break;
1694         }
1695         case KVM_GET_MP_STATE: {
1696                 struct kvm_mp_state mp_state;
1697
1698                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1699                 if (r)
1700                         goto out;
1701                 r = -EFAULT;
1702                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1703                         goto out;
1704                 r = 0;
1705                 break;
1706         }
1707         case KVM_SET_MP_STATE: {
1708                 struct kvm_mp_state mp_state;
1709
1710                 r = -EFAULT;
1711                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1712                         goto out;
1713                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1714                 if (r)
1715                         goto out;
1716                 r = 0;
1717                 break;
1718         }
1719         case KVM_TRANSLATE: {
1720                 struct kvm_translation tr;
1721
1722                 r = -EFAULT;
1723                 if (copy_from_user(&tr, argp, sizeof tr))
1724                         goto out;
1725                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1726                 if (r)
1727                         goto out;
1728                 r = -EFAULT;
1729                 if (copy_to_user(argp, &tr, sizeof tr))
1730                         goto out;
1731                 r = 0;
1732                 break;
1733         }
1734         case KVM_DEBUG_GUEST: {
1735                 struct kvm_debug_guest dbg;
1736
1737                 r = -EFAULT;
1738                 if (copy_from_user(&dbg, argp, sizeof dbg))
1739                         goto out;
1740                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1741                 if (r)
1742                         goto out;
1743                 r = 0;
1744                 break;
1745         }
1746         case KVM_SET_SIGNAL_MASK: {
1747                 struct kvm_signal_mask __user *sigmask_arg = argp;
1748                 struct kvm_signal_mask kvm_sigmask;
1749                 sigset_t sigset, *p;
1750
1751                 p = NULL;
1752                 if (argp) {
1753                         r = -EFAULT;
1754                         if (copy_from_user(&kvm_sigmask, argp,
1755                                            sizeof kvm_sigmask))
1756                                 goto out;
1757                         r = -EINVAL;
1758                         if (kvm_sigmask.len != sizeof sigset)
1759                                 goto out;
1760                         r = -EFAULT;
1761                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1762                                            sizeof sigset))
1763                                 goto out;
1764                         p = &sigset;
1765                 }
1766                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1767                 break;
1768         }
1769         case KVM_GET_FPU: {
1770                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1771                 r = -ENOMEM;
1772                 if (!fpu)
1773                         goto out;
1774                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1775                 if (r)
1776                         goto out;
1777                 r = -EFAULT;
1778                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1779                         goto out;
1780                 r = 0;
1781                 break;
1782         }
1783         case KVM_SET_FPU: {
1784                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1785                 r = -ENOMEM;
1786                 if (!fpu)
1787                         goto out;
1788                 r = -EFAULT;
1789                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1790                         goto out;
1791                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1792                 if (r)
1793                         goto out;
1794                 r = 0;
1795                 break;
1796         }
1797         default:
1798                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1799         }
1800 out:
1801         kfree(fpu);
1802         kfree(kvm_sregs);
1803         return r;
1804 }
1805
1806 static long kvm_vm_ioctl(struct file *filp,
1807                            unsigned int ioctl, unsigned long arg)
1808 {
1809         struct kvm *kvm = filp->private_data;
1810         void __user *argp = (void __user *)arg;
1811         int r;
1812
1813         if (kvm->mm != current->mm)
1814                 return -EIO;
1815         switch (ioctl) {
1816         case KVM_CREATE_VCPU:
1817                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1818                 if (r < 0)
1819                         goto out;
1820                 break;
1821         case KVM_SET_USER_MEMORY_REGION: {
1822                 struct kvm_userspace_memory_region kvm_userspace_mem;
1823
1824                 r = -EFAULT;
1825                 if (copy_from_user(&kvm_userspace_mem, argp,
1826                                                 sizeof kvm_userspace_mem))
1827                         goto out;
1828
1829                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1830                 if (r)
1831                         goto out;
1832                 break;
1833         }
1834         case KVM_GET_DIRTY_LOG: {
1835                 struct kvm_dirty_log log;
1836
1837                 r = -EFAULT;
1838                 if (copy_from_user(&log, argp, sizeof log))
1839                         goto out;
1840                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1841                 if (r)
1842                         goto out;
1843                 break;
1844         }
1845 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1846         case KVM_REGISTER_COALESCED_MMIO: {
1847                 struct kvm_coalesced_mmio_zone zone;
1848                 r = -EFAULT;
1849                 if (copy_from_user(&zone, argp, sizeof zone))
1850                         goto out;
1851                 r = -ENXIO;
1852                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1853                 if (r)
1854                         goto out;
1855                 r = 0;
1856                 break;
1857         }
1858         case KVM_UNREGISTER_COALESCED_MMIO: {
1859                 struct kvm_coalesced_mmio_zone zone;
1860                 r = -EFAULT;
1861                 if (copy_from_user(&zone, argp, sizeof zone))
1862                         goto out;
1863                 r = -ENXIO;
1864                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1865                 if (r)
1866                         goto out;
1867                 r = 0;
1868                 break;
1869         }
1870 #endif
1871 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1872         case KVM_ASSIGN_PCI_DEVICE: {
1873                 struct kvm_assigned_pci_dev assigned_dev;
1874
1875                 r = -EFAULT;
1876                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1877                         goto out;
1878                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1879                 if (r)
1880                         goto out;
1881                 break;
1882         }
1883         case KVM_ASSIGN_IRQ: {
1884                 struct kvm_assigned_irq assigned_irq;
1885
1886                 r = -EFAULT;
1887                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1888                         goto out;
1889                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1890                 if (r)
1891                         goto out;
1892                 break;
1893         }
1894 #endif
1895 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
1896         case KVM_DEASSIGN_PCI_DEVICE: {
1897                 struct kvm_assigned_pci_dev assigned_dev;
1898
1899                 r = -EFAULT;
1900                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1901                         goto out;
1902                 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
1903                 if (r)
1904                         goto out;
1905                 break;
1906         }
1907 #endif
1908         default:
1909                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1910         }
1911 out:
1912         return r;
1913 }
1914
1915 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1916 {
1917         struct page *page[1];
1918         unsigned long addr;
1919         int npages;
1920         gfn_t gfn = vmf->pgoff;
1921         struct kvm *kvm = vma->vm_file->private_data;
1922
1923         addr = gfn_to_hva(kvm, gfn);
1924         if (kvm_is_error_hva(addr))
1925                 return VM_FAULT_SIGBUS;
1926
1927         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1928                                 NULL);
1929         if (unlikely(npages != 1))
1930                 return VM_FAULT_SIGBUS;
1931
1932         vmf->page = page[0];
1933         return 0;
1934 }
1935
1936 static struct vm_operations_struct kvm_vm_vm_ops = {
1937         .fault = kvm_vm_fault,
1938 };
1939
1940 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1941 {
1942         vma->vm_ops = &kvm_vm_vm_ops;
1943         return 0;
1944 }
1945
1946 static struct file_operations kvm_vm_fops = {
1947         .release        = kvm_vm_release,
1948         .unlocked_ioctl = kvm_vm_ioctl,
1949         .compat_ioctl   = kvm_vm_ioctl,
1950         .mmap           = kvm_vm_mmap,
1951 };
1952
1953 static int kvm_dev_ioctl_create_vm(void)
1954 {
1955         int fd;
1956         struct kvm *kvm;
1957
1958         kvm = kvm_create_vm();
1959         if (IS_ERR(kvm))
1960                 return PTR_ERR(kvm);
1961         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1962         if (fd < 0)
1963                 kvm_put_kvm(kvm);
1964
1965         return fd;
1966 }
1967
1968 static long kvm_dev_ioctl_check_extension_generic(long arg)
1969 {
1970         switch (arg) {
1971         case KVM_CAP_USER_MEMORY:
1972         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1973                 return 1;
1974         default:
1975                 break;
1976         }
1977         return kvm_dev_ioctl_check_extension(arg);
1978 }
1979
1980 static long kvm_dev_ioctl(struct file *filp,
1981                           unsigned int ioctl, unsigned long arg)
1982 {
1983         long r = -EINVAL;
1984
1985         switch (ioctl) {
1986         case KVM_GET_API_VERSION:
1987                 r = -EINVAL;
1988                 if (arg)
1989                         goto out;
1990                 r = KVM_API_VERSION;
1991                 break;
1992         case KVM_CREATE_VM:
1993                 r = -EINVAL;
1994                 if (arg)
1995                         goto out;
1996                 r = kvm_dev_ioctl_create_vm();
1997                 break;
1998         case KVM_CHECK_EXTENSION:
1999                 r = kvm_dev_ioctl_check_extension_generic(arg);
2000                 break;
2001         case KVM_GET_VCPU_MMAP_SIZE:
2002                 r = -EINVAL;
2003                 if (arg)
2004                         goto out;
2005                 r = PAGE_SIZE;     /* struct kvm_run */
2006 #ifdef CONFIG_X86
2007                 r += PAGE_SIZE;    /* pio data page */
2008 #endif
2009 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2010                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2011 #endif
2012                 break;
2013         case KVM_TRACE_ENABLE:
2014         case KVM_TRACE_PAUSE:
2015         case KVM_TRACE_DISABLE:
2016                 r = kvm_trace_ioctl(ioctl, arg);
2017                 break;
2018         default:
2019                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2020         }
2021 out:
2022         return r;
2023 }
2024
2025 static struct file_operations kvm_chardev_ops = {
2026         .unlocked_ioctl = kvm_dev_ioctl,
2027         .compat_ioctl   = kvm_dev_ioctl,
2028 };
2029
2030 static struct miscdevice kvm_dev = {
2031         KVM_MINOR,
2032         "kvm",
2033         &kvm_chardev_ops,
2034 };
2035
2036 static void hardware_enable(void *junk)
2037 {
2038         int cpu = raw_smp_processor_id();
2039
2040         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2041                 return;
2042         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2043         kvm_arch_hardware_enable(NULL);
2044 }
2045
2046 static void hardware_disable(void *junk)
2047 {
2048         int cpu = raw_smp_processor_id();
2049
2050         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2051                 return;
2052         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2053         kvm_arch_hardware_disable(NULL);
2054 }
2055
2056 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2057                            void *v)
2058 {
2059         int cpu = (long)v;
2060
2061         val &= ~CPU_TASKS_FROZEN;
2062         switch (val) {
2063         case CPU_DYING:
2064                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2065                        cpu);
2066                 hardware_disable(NULL);
2067                 break;
2068         case CPU_UP_CANCELED:
2069                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2070                        cpu);
2071                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2072                 break;
2073         case CPU_ONLINE:
2074                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2075                        cpu);
2076                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2077                 break;
2078         }
2079         return NOTIFY_OK;
2080 }
2081
2082
2083 asmlinkage void kvm_handle_fault_on_reboot(void)
2084 {
2085         if (kvm_rebooting)
2086                 /* spin while reset goes on */
2087                 while (true)
2088                         ;
2089         /* Fault while not rebooting.  We want the trace. */
2090         BUG();
2091 }
2092 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2093
2094 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2095                       void *v)
2096 {
2097         if (val == SYS_RESTART) {
2098                 /*
2099                  * Some (well, at least mine) BIOSes hang on reboot if
2100                  * in vmx root mode.
2101                  */
2102                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2103                 kvm_rebooting = true;
2104                 on_each_cpu(hardware_disable, NULL, 1);
2105         }
2106         return NOTIFY_OK;
2107 }
2108
2109 static struct notifier_block kvm_reboot_notifier = {
2110         .notifier_call = kvm_reboot,
2111         .priority = 0,
2112 };
2113
2114 void kvm_io_bus_init(struct kvm_io_bus *bus)
2115 {
2116         memset(bus, 0, sizeof(*bus));
2117 }
2118
2119 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2120 {
2121         int i;
2122
2123         for (i = 0; i < bus->dev_count; i++) {
2124                 struct kvm_io_device *pos = bus->devs[i];
2125
2126                 kvm_iodevice_destructor(pos);
2127         }
2128 }
2129
2130 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2131                                           gpa_t addr, int len, int is_write)
2132 {
2133         int i;
2134
2135         for (i = 0; i < bus->dev_count; i++) {
2136                 struct kvm_io_device *pos = bus->devs[i];
2137
2138                 if (pos->in_range(pos, addr, len, is_write))
2139                         return pos;
2140         }
2141
2142         return NULL;
2143 }
2144
2145 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2146 {
2147         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2148
2149         bus->devs[bus->dev_count++] = dev;
2150 }
2151
2152 static struct notifier_block kvm_cpu_notifier = {
2153         .notifier_call = kvm_cpu_hotplug,
2154         .priority = 20, /* must be > scheduler priority */
2155 };
2156
2157 static int vm_stat_get(void *_offset, u64 *val)
2158 {
2159         unsigned offset = (long)_offset;
2160         struct kvm *kvm;
2161
2162         *val = 0;
2163         spin_lock(&kvm_lock);
2164         list_for_each_entry(kvm, &vm_list, vm_list)
2165                 *val += *(u32 *)((void *)kvm + offset);
2166         spin_unlock(&kvm_lock);
2167         return 0;
2168 }
2169
2170 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2171
2172 static int vcpu_stat_get(void *_offset, u64 *val)
2173 {
2174         unsigned offset = (long)_offset;
2175         struct kvm *kvm;
2176         struct kvm_vcpu *vcpu;
2177         int i;
2178
2179         *val = 0;
2180         spin_lock(&kvm_lock);
2181         list_for_each_entry(kvm, &vm_list, vm_list)
2182                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2183                         vcpu = kvm->vcpus[i];
2184                         if (vcpu)
2185                                 *val += *(u32 *)((void *)vcpu + offset);
2186                 }
2187         spin_unlock(&kvm_lock);
2188         return 0;
2189 }
2190
2191 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2192
2193 static struct file_operations *stat_fops[] = {
2194         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2195         [KVM_STAT_VM]   = &vm_stat_fops,
2196 };
2197
2198 static void kvm_init_debug(void)
2199 {
2200         struct kvm_stats_debugfs_item *p;
2201
2202         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2203         for (p = debugfs_entries; p->name; ++p)
2204                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2205                                                 (void *)(long)p->offset,
2206                                                 stat_fops[p->kind]);
2207 }
2208
2209 static void kvm_exit_debug(void)
2210 {
2211         struct kvm_stats_debugfs_item *p;
2212
2213         for (p = debugfs_entries; p->name; ++p)
2214                 debugfs_remove(p->dentry);
2215         debugfs_remove(kvm_debugfs_dir);
2216 }
2217
2218 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2219 {
2220         hardware_disable(NULL);
2221         return 0;
2222 }
2223
2224 static int kvm_resume(struct sys_device *dev)
2225 {
2226         hardware_enable(NULL);
2227         return 0;
2228 }
2229
2230 static struct sysdev_class kvm_sysdev_class = {
2231         .name = "kvm",
2232         .suspend = kvm_suspend,
2233         .resume = kvm_resume,
2234 };
2235
2236 static struct sys_device kvm_sysdev = {
2237         .id = 0,
2238         .cls = &kvm_sysdev_class,
2239 };
2240
2241 struct page *bad_page;
2242 pfn_t bad_pfn;
2243
2244 static inline
2245 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2246 {
2247         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2248 }
2249
2250 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2251 {
2252         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2253
2254         kvm_arch_vcpu_load(vcpu, cpu);
2255 }
2256
2257 static void kvm_sched_out(struct preempt_notifier *pn,
2258                           struct task_struct *next)
2259 {
2260         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2261
2262         kvm_arch_vcpu_put(vcpu);
2263 }
2264
2265 int kvm_init(void *opaque, unsigned int vcpu_size,
2266                   struct module *module)
2267 {
2268         int r;
2269         int cpu;
2270
2271         kvm_init_debug();
2272
2273         r = kvm_arch_init(opaque);
2274         if (r)
2275                 goto out_fail;
2276
2277         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2278
2279         if (bad_page == NULL) {
2280                 r = -ENOMEM;
2281                 goto out;
2282         }
2283
2284         bad_pfn = page_to_pfn(bad_page);
2285
2286         if (!alloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2287                 r = -ENOMEM;
2288                 goto out_free_0;
2289         }
2290
2291         r = kvm_arch_hardware_setup();
2292         if (r < 0)
2293                 goto out_free_0a;
2294
2295         for_each_online_cpu(cpu) {
2296                 smp_call_function_single(cpu,
2297                                 kvm_arch_check_processor_compat,
2298                                 &r, 1);
2299                 if (r < 0)
2300                         goto out_free_1;
2301         }
2302
2303         on_each_cpu(hardware_enable, NULL, 1);
2304         r = register_cpu_notifier(&kvm_cpu_notifier);
2305         if (r)
2306                 goto out_free_2;
2307         register_reboot_notifier(&kvm_reboot_notifier);
2308
2309         r = sysdev_class_register(&kvm_sysdev_class);
2310         if (r)
2311                 goto out_free_3;
2312
2313         r = sysdev_register(&kvm_sysdev);
2314         if (r)
2315                 goto out_free_4;
2316
2317         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2318         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2319                                            __alignof__(struct kvm_vcpu),
2320                                            0, NULL);
2321         if (!kvm_vcpu_cache) {
2322                 r = -ENOMEM;
2323                 goto out_free_5;
2324         }
2325
2326         kvm_chardev_ops.owner = module;
2327         kvm_vm_fops.owner = module;
2328         kvm_vcpu_fops.owner = module;
2329
2330         r = misc_register(&kvm_dev);
2331         if (r) {
2332                 printk(KERN_ERR "kvm: misc device register failed\n");
2333                 goto out_free;
2334         }
2335
2336         kvm_preempt_ops.sched_in = kvm_sched_in;
2337         kvm_preempt_ops.sched_out = kvm_sched_out;
2338 #ifndef CONFIG_X86
2339         msi2intx = 0;
2340 #endif
2341
2342         return 0;
2343
2344 out_free:
2345         kmem_cache_destroy(kvm_vcpu_cache);
2346 out_free_5:
2347         sysdev_unregister(&kvm_sysdev);
2348 out_free_4:
2349         sysdev_class_unregister(&kvm_sysdev_class);
2350 out_free_3:
2351         unregister_reboot_notifier(&kvm_reboot_notifier);
2352         unregister_cpu_notifier(&kvm_cpu_notifier);
2353 out_free_2:
2354         on_each_cpu(hardware_disable, NULL, 1);
2355 out_free_1:
2356         kvm_arch_hardware_unsetup();
2357 out_free_0a:
2358         free_cpumask_var(cpus_hardware_enabled);
2359 out_free_0:
2360         __free_page(bad_page);
2361 out:
2362         kvm_arch_exit();
2363         kvm_exit_debug();
2364 out_fail:
2365         return r;
2366 }
2367 EXPORT_SYMBOL_GPL(kvm_init);
2368
2369 void kvm_exit(void)
2370 {
2371         kvm_trace_cleanup();
2372         misc_deregister(&kvm_dev);
2373         kmem_cache_destroy(kvm_vcpu_cache);
2374         sysdev_unregister(&kvm_sysdev);
2375         sysdev_class_unregister(&kvm_sysdev_class);
2376         unregister_reboot_notifier(&kvm_reboot_notifier);
2377         unregister_cpu_notifier(&kvm_cpu_notifier);
2378         on_each_cpu(hardware_disable, NULL, 1);
2379         kvm_arch_hardware_unsetup();
2380         kvm_arch_exit();
2381         kvm_exit_debug();
2382         free_cpumask_var(cpus_hardware_enabled);
2383         __free_page(bad_page);
2384 }
2385 EXPORT_SYMBOL_GPL(kvm_exit);