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_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         }
 174         mutex_unlock(&assigned_dev->kvm->lock);
 175         kvm_put_kvm(assigned_dev->kvm);
 176 }
 177
 178 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
 179 {
 180         struct kvm_assigned_dev_kernel *assigned_dev =
 181                 (struct kvm_assigned_dev_kernel *) dev_id;
 182
 183         kvm_get_kvm(assigned_dev->kvm);
 184         schedule_work(&assigned_dev->interrupt_work);
 185         disable_irq_nosync(irq);
 186         return IRQ_HANDLED;
 187 }
 188
 189 /* Ack the irq line for an assigned device */
 190 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
 191 {
 192         struct kvm_assigned_dev_kernel *dev;
 193
 194         if (kian->gsi == -1)
 195                 return;
 196
 197         dev = container_of(kian, struct kvm_assigned_dev_kernel,
 198                            ack_notifier);
 199         kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
 200         enable_irq(dev->host_irq);
 201 }
 202
 203 static void kvm_free_assigned_irq(struct kvm *kvm,
 204                                   struct kvm_assigned_dev_kernel *assigned_dev)
 205 {
 206         if (!irqchip_in_kernel(kvm))
 207                 return;
 208
 209         kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
 210
 211         if (assigned_dev->irq_source_id != -1)
 212                 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
 213         assigned_dev->irq_source_id = -1;
 214
 215         if (!assigned_dev->irq_requested_type)
 216                 return;
 217
 218         if (cancel_work_sync(&assigned_dev->interrupt_work))
 219                 /* We had pending work. That means we will have to take
 220                  * care of kvm_put_kvm.
 221                  */
 222                 kvm_put_kvm(kvm);
 223
 224         free_irq(assigned_dev->host_irq, (void *)assigned_dev);
 225
 226         if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
 227                 pci_disable_msi(assigned_dev->dev);
 228
 229         assigned_dev->irq_requested_type = 0;
 230 }
 231
 232
 233 static void kvm_free_assigned_device(struct kvm *kvm,
 234                                      struct kvm_assigned_dev_kernel
 235                                      *assigned_dev)
 236 {
 237         kvm_free_assigned_irq(kvm, assigned_dev);
 238
 239         pci_reset_function(assigned_dev->dev);
 240
 241         pci_release_regions(assigned_dev->dev);
 242         pci_disable_device(assigned_dev->dev);
 243         pci_dev_put(assigned_dev->dev);
 244
 245         list_del(&assigned_dev->list);
 246         kfree(assigned_dev);
 247 }
 248
 249 void kvm_free_all_assigned_devices(struct kvm *kvm)
 250 {
 251         struct list_head *ptr, *ptr2;
 252         struct kvm_assigned_dev_kernel *assigned_dev;
 253
 254         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
 255                 assigned_dev = list_entry(ptr,
 256                                           struct kvm_assigned_dev_kernel,
 257                                           list);
 258
 259                 kvm_free_assigned_device(kvm, assigned_dev);
 260         }
 261 }
 262
 263 static int assigned_device_update_intx(struct kvm *kvm,
 264                         struct kvm_assigned_dev_kernel *adev,
 265                         struct kvm_assigned_irq *airq)
 266 {
 267         adev->guest_irq = airq->guest_irq;
 268         adev->ack_notifier.gsi = airq->guest_irq;
 269
 270         if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_INTX)
 271                 return 0;
 272
 273         if (irqchip_in_kernel(kvm)) {
 274                 if (!msi2intx &&
 275                     adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI) {
 276                         free_irq(adev->host_irq, (void *)kvm);
 277                         pci_disable_msi(adev->dev);
 278                 }
 279
 280                 if (!capable(CAP_SYS_RAWIO))
 281                         return -EPERM;
 282
 283                 if (airq->host_irq)
 284                         adev->host_irq = airq->host_irq;
 285                 else
 286                         adev->host_irq = adev->dev->irq;
 287
 288                 /* Even though this is PCI, we don't want to use shared
 289                  * interrupts. Sharing host devices with guest-assigned devices
 290                  * on the same interrupt line is not a happy situation: there
 291                  * are going to be long delays in accepting, acking, etc.
 292                  */
 293                 if (request_irq(adev->host_irq, kvm_assigned_dev_intr,
 294                                 0, "kvm_assigned_intx_device", (void *)adev))
 295                         return -EIO;
 296         }
 297
 298         adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_INTX |
 299                                    KVM_ASSIGNED_DEV_HOST_INTX;
 300         return 0;
 301 }
 302
 303 #ifdef CONFIG_X86
 304 static int assigned_device_update_msi(struct kvm *kvm,
 305                         struct kvm_assigned_dev_kernel *adev,
 306                         struct kvm_assigned_irq *airq)
 307 {
 308         int r;
 309
 310         if (airq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI) {
 311                 /* x86 don't care upper address of guest msi message addr */
 312                 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_MSI;
 313                 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_INTX;
 314                 adev->guest_msi.address_lo = airq->guest_msi.addr_lo;
 315                 adev->guest_msi.data = airq->guest_msi.data;
 316                 adev->ack_notifier.gsi = -1;
 317         } else if (msi2intx) {
 318                 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_INTX;
 319                 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_MSI;
 320                 adev->guest_irq = airq->guest_irq;
 321                 adev->ack_notifier.gsi = airq->guest_irq;
 322         }
 323
 324         if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
 325                 return 0;
 326
 327         if (irqchip_in_kernel(kvm)) {
 328                 if (!msi2intx) {
 329                         if (adev->irq_requested_type &
 330                                         KVM_ASSIGNED_DEV_HOST_INTX)
 331                                 free_irq(adev->host_irq, (void *)adev);
 332
 333                         r = pci_enable_msi(adev->dev);
 334                         if (r)
 335                                 return r;
 336                 }
 337
 338                 adev->host_irq = adev->dev->irq;
 339                 if (request_irq(adev->host_irq, kvm_assigned_dev_intr, 0,
 340                                 "kvm_assigned_msi_device", (void *)adev))
 341                         return -EIO;
 342         }
 343
 344         if (!msi2intx)
 345                 adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_MSI;
 346
 347         adev->irq_requested_type |= KVM_ASSIGNED_DEV_HOST_MSI;
 348         return 0;
 349 }
 350 #endif
 351
 352 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 353                                    struct kvm_assigned_irq
 354                                    *assigned_irq)
 355 {
 356         int r = 0;
 357         struct kvm_assigned_dev_kernel *match;
 358
 359         mutex_lock(&kvm->lock);
 360
 361         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 362                                       assigned_irq->assigned_dev_id);
 363         if (!match) {
 364                 mutex_unlock(&kvm->lock);
 365                 return -EINVAL;
 366         }
 367
 368         if (!match->irq_requested_type) {
 369                 INIT_WORK(&match->interrupt_work,
 370                                 kvm_assigned_dev_interrupt_work_handler);
 371                 if (irqchip_in_kernel(kvm)) {
 372                         /* Register ack nofitier */
 373                         match->ack_notifier.gsi = -1;
 374                         match->ack_notifier.irq_acked =
 375                                         kvm_assigned_dev_ack_irq;
 376                         kvm_register_irq_ack_notifier(kvm,
 377                                         &match->ack_notifier);
 378
 379                         /* Request IRQ source ID */
 380                         r = kvm_request_irq_source_id(kvm);
 381                         if (r < 0)
 382                                 goto out_release;
 383                         else
 384                                 match->irq_source_id = r;
 385
 386 #ifdef CONFIG_X86
 387                         /* Determine host device irq type, we can know the
 388                          * result from dev->msi_enabled */
 389                         if (msi2intx)
 390                                 pci_enable_msi(match->dev);
 391 #endif
 392                 }
 393         }
 394
 395         if ((!msi2intx &&
 396              (assigned_irq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI)) ||
 397             (msi2intx && match->dev->msi_enabled)) {
 398 #ifdef CONFIG_X86
 399                 r = assigned_device_update_msi(kvm, match, assigned_irq);
 400                 if (r) {
 401                         printk(KERN_WARNING "kvm: failed to enable "
 402                                         "MSI device!\n");
 403                         goto out_release;
 404                 }
 405 #else
 406                 r = -ENOTTY;
 407 #endif
 408         } else if (assigned_irq->host_irq == 0 && match->dev->irq == 0) {
 409                 /* Host device IRQ 0 means don't support INTx */
 410                 if (!msi2intx) {
 411                         printk(KERN_WARNING
 412                                "kvm: wait device to enable MSI!\n");
 413                         r = 0;
 414                 } else {
 415                         printk(KERN_WARNING
 416                                "kvm: failed to enable MSI device!\n");
 417                         r = -ENOTTY;
 418                         goto out_release;
 419                 }
 420         } else {
 421                 /* Non-sharing INTx mode */
 422                 r = assigned_device_update_intx(kvm, match, assigned_irq);
 423                 if (r) {
 424                         printk(KERN_WARNING "kvm: failed to enable "
 425                                         "INTx device!\n");
 426                         goto out_release;
 427                 }
 428         }
 429
 430         mutex_unlock(&kvm->lock);
 431         return r;
 432 out_release:
 433         mutex_unlock(&kvm->lock);
 434         kvm_free_assigned_device(kvm, match);
 435         return r;
 436 }
 437
 438 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 439                                       struct kvm_assigned_pci_dev *assigned_dev)
 440 {
 441         int r = 0;
 442         struct kvm_assigned_dev_kernel *match;
 443         struct pci_dev *dev;
 444
 445         mutex_lock(&kvm->lock);
 446
 447         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 448                                       assigned_dev->assigned_dev_id);
 449         if (match) {
 450                 /* device already assigned */
 451                 r = -EINVAL;
 452                 goto out;
 453         }
 454
 455         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 456         if (match == NULL) {
 457                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 458                        __func__);
 459                 r = -ENOMEM;
 460                 goto out;
 461         }
 462         dev = pci_get_bus_and_slot(assigned_dev->busnr,
 463                                    assigned_dev->devfn);
 464         if (!dev) {
 465                 printk(KERN_INFO "%s: host device not found\n", __func__);
 466                 r = -EINVAL;
 467                 goto out_free;
 468         }
 469         if (pci_enable_device(dev)) {
 470                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 471                 r = -EBUSY;
 472                 goto out_put;
 473         }
 474         r = pci_request_regions(dev, "kvm_assigned_device");
 475         if (r) {
 476                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 477                        __func__);
 478                 goto out_disable;
 479         }
 480
 481         pci_reset_function(dev);
 482
 483         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 484         match->host_busnr = assigned_dev->busnr;
 485         match->host_devfn = assigned_dev->devfn;
 486         match->dev = dev;
 487         match->irq_source_id = -1;
 488         match->kvm = kvm;
 489
 490         list_add(&match->list, &kvm->arch.assigned_dev_head);
 491
 492         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
 493                 r = kvm_iommu_map_guest(kvm, match);
 494                 if (r)
 495                         goto out_list_del;
 496         }
 497
 498 out:
 499         mutex_unlock(&kvm->lock);
 500         return r;
 501 out_list_del:
 502         list_del(&match->list);
 503         pci_release_regions(dev);
 504 out_disable:
 505         pci_disable_device(dev);
 506 out_put:
 507         pci_dev_put(dev);
 508 out_free:
 509         kfree(match);
 510         mutex_unlock(&kvm->lock);
 511         return r;
 512 }
 513 #endif
 514
 515 static inline int valid_vcpu(int n)
 516 {
 517         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 518 }
 519
 520 inline int kvm_is_mmio_pfn(pfn_t pfn)
 521 {
 522         if (pfn_valid(pfn))
 523                 return PageReserved(pfn_to_page(pfn));
 524
 525         return true;
 526 }
 527
 528 /*
 529  * Switches to specified vcpu, until a matching vcpu_put()
 530  */
 531 void vcpu_load(struct kvm_vcpu *vcpu)
 532 {
 533         int cpu;
 534
 535         mutex_lock(&vcpu->mutex);
 536         cpu = get_cpu();
 537         preempt_notifier_register(&vcpu->preempt_notifier);
 538         kvm_arch_vcpu_load(vcpu, cpu);
 539         put_cpu();
 540 }
 541
 542 void vcpu_put(struct kvm_vcpu *vcpu)
 543 {
 544         preempt_disable();
 545         kvm_arch_vcpu_put(vcpu);
 546         preempt_notifier_unregister(&vcpu->preempt_notifier);
 547         preempt_enable();
 548         mutex_unlock(&vcpu->mutex);
 549 }
 550
 551 static void ack_flush(void *_completed)
 552 {
 553 }
 554
 555 void kvm_flush_remote_tlbs(struct kvm *kvm)
 556 {
 557         int i, cpu, me;
 558         cpumask_t cpus;
 559         struct kvm_vcpu *vcpu;
 560
 561         me = get_cpu();
 562         cpus_clear(cpus);
 563         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 564                 vcpu = kvm->vcpus[i];
 565                 if (!vcpu)
 566                         continue;
 567                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 568                         continue;
 569                 cpu = vcpu->cpu;
 570                 if (cpu != -1 && cpu != me)
 571                         cpu_set(cpu, cpus);
 572         }
 573         if (cpus_empty(cpus))
 574                 goto out;
 575         ++kvm->stat.remote_tlb_flush;
 576         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 577 out:
 578         put_cpu();
 579 }
 580
 581 void kvm_reload_remote_mmus(struct kvm *kvm)
 582 {
 583         int i, cpu, me;
 584         cpumask_t cpus;
 585         struct kvm_vcpu *vcpu;
 586
 587         me = get_cpu();
 588         cpus_clear(cpus);
 589         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 590                 vcpu = kvm->vcpus[i];
 591                 if (!vcpu)
 592                         continue;
 593                 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
 594                         continue;
 595                 cpu = vcpu->cpu;
 596                 if (cpu != -1 && cpu != me)
 597                         cpu_set(cpu, cpus);
 598         }
 599         if (cpus_empty(cpus))
 600                 goto out;
 601         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 602 out:
 603         put_cpu();
 604 }
 605
 606
 607 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 608 {
 609         struct page *page;
 610         int r;
 611
 612         mutex_init(&vcpu->mutex);
 613         vcpu->cpu = -1;
 614         vcpu->kvm = kvm;
 615         vcpu->vcpu_id = id;
 616         init_waitqueue_head(&vcpu->wq);
 617
 618         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 619         if (!page) {
 620                 r = -ENOMEM;
 621                 goto fail;
 622         }
 623         vcpu->run = page_address(page);
 624
 625         r = kvm_arch_vcpu_init(vcpu);
 626         if (r < 0)
 627                 goto fail_free_run;
 628         return 0;
 629
 630 fail_free_run:
 631         free_page((unsigned long)vcpu->run);
 632 fail:
 633         return r;
 634 }
 635 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 636
 637 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 638 {
 639         kvm_arch_vcpu_uninit(vcpu);
 640         free_page((unsigned long)vcpu->run);
 641 }
 642 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 643
 644 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 645 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 646 {
 647         return container_of(mn, struct kvm, mmu_notifier);
 648 }
 649
 650 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 651                                              struct mm_struct *mm,
 652                                              unsigned long address)
 653 {
 654         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 655         int need_tlb_flush;
 656
 657         /*
 658          * When ->invalidate_page runs, the linux pte has been zapped
 659          * already but the page is still allocated until
 660          * ->invalidate_page returns. So if we increase the sequence
 661          * here the kvm page fault will notice if the spte can't be
 662          * established because the page is going to be freed. If
 663          * instead the kvm page fault establishes the spte before
 664          * ->invalidate_page runs, kvm_unmap_hva will release it
 665          * before returning.
 666          *
 667          * The sequence increase only need to be seen at spin_unlock
 668          * time, and not at spin_lock time.
 669          *
 670          * Increasing the sequence after the spin_unlock would be
 671          * unsafe because the kvm page fault could then establish the
 672          * pte after kvm_unmap_hva returned, without noticing the page
 673          * is going to be freed.
 674          */
 675         spin_lock(&kvm->mmu_lock);
 676         kvm->mmu_notifier_seq++;
 677         need_tlb_flush = kvm_unmap_hva(kvm, address);
 678         spin_unlock(&kvm->mmu_lock);
 679
 680         /* we've to flush the tlb before the pages can be freed */
 681         if (need_tlb_flush)
 682                 kvm_flush_remote_tlbs(kvm);
 683
 684 }
 685
 686 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 687                                                     struct mm_struct *mm,
 688                                                     unsigned long start,
 689                                                     unsigned long end)
 690 {
 691         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 692         int need_tlb_flush = 0;
 693
 694         spin_lock(&kvm->mmu_lock);
 695         /*
 696          * The count increase must become visible at unlock time as no
 697          * spte can be established without taking the mmu_lock and
 698          * count is also read inside the mmu_lock critical section.
 699          */
 700         kvm->mmu_notifier_count++;
 701         for (; start < end; start += PAGE_SIZE)
 702                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 703         spin_unlock(&kvm->mmu_lock);
 704
 705         /* we've to flush the tlb before the pages can be freed */
 706         if (need_tlb_flush)
 707                 kvm_flush_remote_tlbs(kvm);
 708 }
 709
 710 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 711                                                   struct mm_struct *mm,
 712                                                   unsigned long start,
 713                                                   unsigned long end)
 714 {
 715         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 716
 717         spin_lock(&kvm->mmu_lock);
 718         /*
 719          * This sequence increase will notify the kvm page fault that
 720          * the page that is going to be mapped in the spte could have
 721          * been freed.
 722          */
 723         kvm->mmu_notifier_seq++;
 724         /*
 725          * The above sequence increase must be visible before the
 726          * below count decrease but both values are read by the kvm
 727          * page fault under mmu_lock spinlock so we don't need to add
 728          * a smb_wmb() here in between the two.
 729          */
 730         kvm->mmu_notifier_count--;
 731         spin_unlock(&kvm->mmu_lock);
 732
 733         BUG_ON(kvm->mmu_notifier_count < 0);
 734 }
 735
 736 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 737                                               struct mm_struct *mm,
 738                                               unsigned long address)
 739 {
 740         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 741         int young;
 742
 743         spin_lock(&kvm->mmu_lock);
 744         young = kvm_age_hva(kvm, address);
 745         spin_unlock(&kvm->mmu_lock);
 746
 747         if (young)
 748                 kvm_flush_remote_tlbs(kvm);
 749
 750         return young;
 751 }
 752
 753 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 754         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 755         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 756         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 757         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 758 };
 759 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 760
 761 static struct kvm *kvm_create_vm(void)
 762 {
 763         struct kvm *kvm = kvm_arch_create_vm();
 764 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 765         struct page *page;
 766 #endif
 767
 768         if (IS_ERR(kvm))
 769                 goto out;
 770
 771 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 772         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 773         if (!page) {
 774                 kfree(kvm);
 775                 return ERR_PTR(-ENOMEM);
 776         }
 777         kvm->coalesced_mmio_ring =
 778                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 779 #endif
 780
 781 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 782         {
 783                 int err;
 784                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 785                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 786                 if (err) {
 787 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 788                         put_page(page);
 789 #endif
 790                         kfree(kvm);
 791                         return ERR_PTR(err);
 792                 }
 793         }
 794 #endif
 795
 796         kvm->mm = current->mm;
 797         atomic_inc(&kvm->mm->mm_count);
 798         spin_lock_init(&kvm->mmu_lock);
 799         kvm_io_bus_init(&kvm->pio_bus);
 800         mutex_init(&kvm->lock);
 801         kvm_io_bus_init(&kvm->mmio_bus);
 802         init_rwsem(&kvm->slots_lock);
 803         atomic_set(&kvm->users_count, 1);
 804         spin_lock(&kvm_lock);
 805         list_add(&kvm->vm_list, &vm_list);
 806         spin_unlock(&kvm_lock);
 807 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 808         kvm_coalesced_mmio_init(kvm);
 809 #endif
 810 out:
 811         return kvm;
 812 }
 813
 814 /*
 815  * Free any memory in @free but not in @dont.
 816  */
 817 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 818                                   struct kvm_memory_slot *dont)
 819 {
 820         if (!dont || free->rmap != dont->rmap)
 821                 vfree(free->rmap);
 822
 823         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 824                 vfree(free->dirty_bitmap);
 825
 826         if (!dont || free->lpage_info != dont->lpage_info)
 827                 vfree(free->lpage_info);
 828
 829         free->npages = 0;
 830         free->dirty_bitmap = NULL;
 831         free->rmap = NULL;
 832         free->lpage_info = NULL;
 833 }
 834
 835 void kvm_free_physmem(struct kvm *kvm)
 836 {
 837         int i;
 838
 839         for (i = 0; i < kvm->nmemslots; ++i)
 840                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 841 }
 842
 843 static void kvm_destroy_vm(struct kvm *kvm)
 844 {
 845         struct mm_struct *mm = kvm->mm;
 846
 847         spin_lock(&kvm_lock);
 848         list_del(&kvm->vm_list);
 849         spin_unlock(&kvm_lock);
 850         kvm_io_bus_destroy(&kvm->pio_bus);
 851         kvm_io_bus_destroy(&kvm->mmio_bus);
 852 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 853         if (kvm->coalesced_mmio_ring != NULL)
 854                 free_page((unsigned long)kvm->coalesced_mmio_ring);
 855 #endif
 856 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 857         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 858 #endif
 859         kvm_arch_destroy_vm(kvm);
 860         mmdrop(mm);
 861 }
 862
 863 void kvm_get_kvm(struct kvm *kvm)
 864 {
 865         atomic_inc(&kvm->users_count);
 866 }
 867 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 868
 869 void kvm_put_kvm(struct kvm *kvm)
 870 {
 871         if (atomic_dec_and_test(&kvm->users_count))
 872                 kvm_destroy_vm(kvm);
 873 }
 874 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 875
 876
 877 static int kvm_vm_release(struct inode *inode, struct file *filp)
 878 {
 879         struct kvm *kvm = filp->private_data;
 880
 881         kvm_put_kvm(kvm);
 882         return 0;
 883 }
 884
 885 /*
 886  * Allocate some memory and give it an address in the guest physical address
 887  * space.
 888  *
 889  * Discontiguous memory is allowed, mostly for framebuffers.
 890  *
 891  * Must be called holding mmap_sem for write.
 892  */
 893 int __kvm_set_memory_region(struct kvm *kvm,
 894                             struct kvm_userspace_memory_region *mem,
 895                             int user_alloc)
 896 {
 897         int r;
 898         gfn_t base_gfn;
 899         unsigned long npages;
 900         unsigned long i;
 901         struct kvm_memory_slot *memslot;
 902         struct kvm_memory_slot old, new;
 903
 904         r = -EINVAL;
 905         /* General sanity checks */
 906         if (mem->memory_size & (PAGE_SIZE - 1))
 907                 goto out;
 908         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 909                 goto out;
 910         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 911                 goto out;
 912         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 913                 goto out;
 914         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 915                 goto out;
 916
 917         memslot = &kvm->memslots[mem->slot];
 918         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 919         npages = mem->memory_size >> PAGE_SHIFT;
 920
 921         if (!npages)
 922                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 923
 924         new = old = *memslot;
 925
 926         new.base_gfn = base_gfn;
 927         new.npages = npages;
 928         new.flags = mem->flags;
 929
 930         /* Disallow changing a memory slot's size. */
 931         r = -EINVAL;
 932         if (npages && old.npages && npages != old.npages)
 933                 goto out_free;
 934
 935         /* Check for overlaps */
 936         r = -EEXIST;
 937         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 938                 struct kvm_memory_slot *s = &kvm->memslots[i];
 939
 940                 if (s == memslot)
 941                         continue;
 942                 if (!((base_gfn + npages <= s->base_gfn) ||
 943                       (base_gfn >= s->base_gfn + s->npages)))
 944                         goto out_free;
 945         }
 946
 947         /* Free page dirty bitmap if unneeded */
 948         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 949                 new.dirty_bitmap = NULL;
 950
 951         r = -ENOMEM;
 952
 953         /* Allocate if a slot is being created */
 954 #ifndef CONFIG_S390
 955         if (npages && !new.rmap) {
 956                 new.rmap = vmalloc(npages * sizeof(struct page *));
 957
 958                 if (!new.rmap)
 959                         goto out_free;
 960
 961                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 962
 963                 new.user_alloc = user_alloc;
 964                 /*
 965                  * hva_to_rmmap() serialzies with the mmu_lock and to be
 966                  * safe it has to ignore memslots with !user_alloc &&
 967                  * !userspace_addr.
 968                  */
 969                 if (user_alloc)
 970                         new.userspace_addr = mem->userspace_addr;
 971                 else
 972                         new.userspace_addr = 0;
 973         }
 974         if (npages && !new.lpage_info) {
 975                 int largepages = npages / KVM_PAGES_PER_HPAGE;
 976                 if (npages % KVM_PAGES_PER_HPAGE)
 977                         largepages++;
 978                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 979                         largepages++;
 980
 981                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
 982
 983                 if (!new.lpage_info)
 984                         goto out_free;
 985
 986                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
 987
 988                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 989                         new.lpage_info[0].write_count = 1;
 990                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
 991                         new.lpage_info[largepages-1].write_count = 1;
 992         }
 993
 994         /* Allocate page dirty bitmap if needed */
 995         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 996                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 997
 998                 new.dirty_bitmap = vmalloc(dirty_bytes);
 999                 if (!new.dirty_bitmap)
1000                         goto out_free;
1001                 memset(new.dirty_bitmap, 0, dirty_bytes);
1002         }
1003 #endif /* not defined CONFIG_S390 */
1004
1005         if (!npages)
1006                 kvm_arch_flush_shadow(kvm);
1007
1008         spin_lock(&kvm->mmu_lock);
1009         if (mem->slot >= kvm->nmemslots)
1010                 kvm->nmemslots = mem->slot + 1;
1011
1012         *memslot = new;
1013         spin_unlock(&kvm->mmu_lock);
1014
1015         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1016         if (r) {
1017                 spin_lock(&kvm->mmu_lock);
1018                 *memslot = old;
1019                 spin_unlock(&kvm->mmu_lock);
1020                 goto out_free;
1021         }
1022
1023         kvm_free_physmem_slot(&old, npages ? &new : NULL);
1024         /* Slot deletion case: we have to update the current slot */
1025         if (!npages)
1026                 *memslot = old;
1027 #ifdef CONFIG_DMAR
1028         /* map the pages in iommu page table */
1029         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1030         if (r)
1031                 goto out;
1032 #endif
1033         return 0;
1034
1035 out_free:
1036         kvm_free_physmem_slot(&new, &old);
1037 out:
1038         return r;
1039
1040 }
1041 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1042
1043 int kvm_set_memory_region(struct kvm *kvm,
1044                           struct kvm_userspace_memory_region *mem,
1045                           int user_alloc)
1046 {
1047         int r;
1048
1049         down_write(&kvm->slots_lock);
1050         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1051         up_write(&kvm->slots_lock);
1052         return r;
1053 }
1054 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1055
1056 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1057                                    struct
1058                                    kvm_userspace_memory_region *mem,
1059                                    int user_alloc)
1060 {
1061         if (mem->slot >= KVM_MEMORY_SLOTS)
1062                 return -EINVAL;
1063         return kvm_set_memory_region(kvm, mem, user_alloc);
1064 }
1065
1066 int kvm_get_dirty_log(struct kvm *kvm,
1067                         struct kvm_dirty_log *log, int *is_dirty)
1068 {
1069         struct kvm_memory_slot *memslot;
1070         int r, i;
1071         int n;
1072         unsigned long any = 0;
1073
1074         r = -EINVAL;
1075         if (log->slot >= KVM_MEMORY_SLOTS)
1076                 goto out;
1077
1078         memslot = &kvm->memslots[log->slot];
1079         r = -ENOENT;
1080         if (!memslot->dirty_bitmap)
1081                 goto out;
1082
1083         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1084
1085         for (i = 0; !any && i < n/sizeof(long); ++i)
1086                 any = memslot->dirty_bitmap[i];
1087
1088         r = -EFAULT;
1089         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1090                 goto out;
1091
1092         if (any)
1093                 *is_dirty = 1;
1094
1095         r = 0;
1096 out:
1097         return r;
1098 }
1099
1100 int is_error_page(struct page *page)
1101 {
1102         return page == bad_page;
1103 }
1104 EXPORT_SYMBOL_GPL(is_error_page);
1105
1106 int is_error_pfn(pfn_t pfn)
1107 {
1108         return pfn == bad_pfn;
1109 }
1110 EXPORT_SYMBOL_GPL(is_error_pfn);
1111
1112 static inline unsigned long bad_hva(void)
1113 {
1114         return PAGE_OFFSET;
1115 }
1116
1117 int kvm_is_error_hva(unsigned long addr)
1118 {
1119         return addr == bad_hva();
1120 }
1121 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1122
1123 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1124 {
1125         int i;
1126
1127         for (i = 0; i < kvm->nmemslots; ++i) {
1128                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1129
1130                 if (gfn >= memslot->base_gfn
1131                     && gfn < memslot->base_gfn + memslot->npages)
1132                         return memslot;
1133         }
1134         return NULL;
1135 }
1136 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1137
1138 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1139 {
1140         gfn = unalias_gfn(kvm, gfn);
1141         return gfn_to_memslot_unaliased(kvm, gfn);
1142 }
1143
1144 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1145 {
1146         int i;
1147
1148         gfn = unalias_gfn(kvm, gfn);
1149         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1150                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1151
1152                 if (gfn >= memslot->base_gfn
1153                     && gfn < memslot->base_gfn + memslot->npages)
1154                         return 1;
1155         }
1156         return 0;
1157 }
1158 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1159
1160 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1161 {
1162         struct kvm_memory_slot *slot;
1163
1164         gfn = unalias_gfn(kvm, gfn);
1165         slot = gfn_to_memslot_unaliased(kvm, gfn);
1166         if (!slot)
1167                 return bad_hva();
1168         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1169 }
1170 EXPORT_SYMBOL_GPL(gfn_to_hva);
1171
1172 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1173 {
1174         struct page *page[1];
1175         unsigned long addr;
1176         int npages;
1177         pfn_t pfn;
1178
1179         might_sleep();
1180
1181         addr = gfn_to_hva(kvm, gfn);
1182         if (kvm_is_error_hva(addr)) {
1183                 get_page(bad_page);
1184                 return page_to_pfn(bad_page);
1185         }
1186
1187         npages = get_user_pages_fast(addr, 1, 1, page);
1188
1189         if (unlikely(npages != 1)) {
1190                 struct vm_area_struct *vma;
1191
1192                 down_read(&current->mm->mmap_sem);
1193                 vma = find_vma(current->mm, addr);
1194
1195                 if (vma == NULL || addr < vma->vm_start ||
1196                     !(vma->vm_flags & VM_PFNMAP)) {
1197                         up_read(&current->mm->mmap_sem);
1198                         get_page(bad_page);
1199                         return page_to_pfn(bad_page);
1200                 }
1201
1202                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1203                 up_read(&current->mm->mmap_sem);
1204                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1205         } else
1206                 pfn = page_to_pfn(page[0]);
1207
1208         return pfn;
1209 }
1210
1211 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1212
1213 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1214 {
1215         pfn_t pfn;
1216
1217         pfn = gfn_to_pfn(kvm, gfn);
1218         if (!kvm_is_mmio_pfn(pfn))
1219                 return pfn_to_page(pfn);
1220
1221         WARN_ON(kvm_is_mmio_pfn(pfn));
1222
1223         get_page(bad_page);
1224         return bad_page;
1225 }
1226
1227 EXPORT_SYMBOL_GPL(gfn_to_page);
1228
1229 void kvm_release_page_clean(struct page *page)
1230 {
1231         kvm_release_pfn_clean(page_to_pfn(page));
1232 }
1233 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1234
1235 void kvm_release_pfn_clean(pfn_t pfn)
1236 {
1237         if (!kvm_is_mmio_pfn(pfn))
1238                 put_page(pfn_to_page(pfn));
1239 }
1240 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1241
1242 void kvm_release_page_dirty(struct page *page)
1243 {
1244         kvm_release_pfn_dirty(page_to_pfn(page));
1245 }
1246 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1247
1248 void kvm_release_pfn_dirty(pfn_t pfn)
1249 {
1250         kvm_set_pfn_dirty(pfn);
1251         kvm_release_pfn_clean(pfn);
1252 }
1253 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1254
1255 void kvm_set_page_dirty(struct page *page)
1256 {
1257         kvm_set_pfn_dirty(page_to_pfn(page));
1258 }
1259 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1260
1261 void kvm_set_pfn_dirty(pfn_t pfn)
1262 {
1263         if (!kvm_is_mmio_pfn(pfn)) {
1264                 struct page *page = pfn_to_page(pfn);
1265                 if (!PageReserved(page))
1266                         SetPageDirty(page);
1267         }
1268 }
1269 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1270
1271 void kvm_set_pfn_accessed(pfn_t pfn)
1272 {
1273         if (!kvm_is_mmio_pfn(pfn))
1274                 mark_page_accessed(pfn_to_page(pfn));
1275 }
1276 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1277
1278 void kvm_get_pfn(pfn_t pfn)
1279 {
1280         if (!kvm_is_mmio_pfn(pfn))
1281                 get_page(pfn_to_page(pfn));
1282 }
1283 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1284
1285 static int next_segment(unsigned long len, int offset)
1286 {
1287         if (len > PAGE_SIZE - offset)
1288                 return PAGE_SIZE - offset;
1289         else
1290                 return len;
1291 }
1292
1293 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1294                         int len)
1295 {
1296         int r;
1297         unsigned long addr;
1298
1299         addr = gfn_to_hva(kvm, gfn);
1300         if (kvm_is_error_hva(addr))
1301                 return -EFAULT;
1302         r = copy_from_user(data, (void __user *)addr + offset, len);
1303         if (r)
1304                 return -EFAULT;
1305         return 0;
1306 }
1307 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1308
1309 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1310 {
1311         gfn_t gfn = gpa >> PAGE_SHIFT;
1312         int seg;
1313         int offset = offset_in_page(gpa);
1314         int ret;
1315
1316         while ((seg = next_segment(len, offset)) != 0) {
1317                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1318                 if (ret < 0)
1319                         return ret;
1320                 offset = 0;
1321                 len -= seg;
1322                 data += seg;
1323                 ++gfn;
1324         }
1325         return 0;
1326 }
1327 EXPORT_SYMBOL_GPL(kvm_read_guest);
1328
1329 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1330                           unsigned long len)
1331 {
1332         int r;
1333         unsigned long addr;
1334         gfn_t gfn = gpa >> PAGE_SHIFT;
1335         int offset = offset_in_page(gpa);
1336
1337         addr = gfn_to_hva(kvm, gfn);
1338         if (kvm_is_error_hva(addr))
1339                 return -EFAULT;
1340         pagefault_disable();
1341         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1342         pagefault_enable();
1343         if (r)
1344                 return -EFAULT;
1345         return 0;
1346 }
1347 EXPORT_SYMBOL(kvm_read_guest_atomic);
1348
1349 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1350                          int offset, int len)
1351 {
1352         int r;
1353         unsigned long addr;
1354
1355         addr = gfn_to_hva(kvm, gfn);
1356         if (kvm_is_error_hva(addr))
1357                 return -EFAULT;
1358         r = copy_to_user((void __user *)addr + offset, data, len);
1359         if (r)
1360                 return -EFAULT;
1361         mark_page_dirty(kvm, gfn);
1362         return 0;
1363 }
1364 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1365
1366 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1367                     unsigned long len)
1368 {
1369         gfn_t gfn = gpa >> PAGE_SHIFT;
1370         int seg;
1371         int offset = offset_in_page(gpa);
1372         int ret;
1373
1374         while ((seg = next_segment(len, offset)) != 0) {
1375                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1376                 if (ret < 0)
1377                         return ret;
1378                 offset = 0;
1379                 len -= seg;
1380                 data += seg;
1381                 ++gfn;
1382         }
1383         return 0;
1384 }
1385
1386 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1387 {
1388         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1389 }
1390 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1391
1392 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1393 {
1394         gfn_t gfn = gpa >> PAGE_SHIFT;
1395         int seg;
1396         int offset = offset_in_page(gpa);
1397         int ret;
1398
1399         while ((seg = next_segment(len, offset)) != 0) {
1400                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1401                 if (ret < 0)
1402                         return ret;
1403                 offset = 0;
1404                 len -= seg;
1405                 ++gfn;
1406         }
1407         return 0;
1408 }
1409 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1410
1411 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1412 {
1413         struct kvm_memory_slot *memslot;
1414
1415         gfn = unalias_gfn(kvm, gfn);
1416         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1417         if (memslot && memslot->dirty_bitmap) {
1418                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1419
1420                 /* avoid RMW */
1421                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1422                         set_bit(rel_gfn, memslot->dirty_bitmap);
1423         }
1424 }
1425
1426 /*
1427  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1428  */
1429 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1430 {
1431         DEFINE_WAIT(wait);
1432
1433         for (;;) {
1434                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1435
1436                 if (kvm_cpu_has_interrupt(vcpu) ||
1437                     kvm_cpu_has_pending_timer(vcpu) ||
1438                     kvm_arch_vcpu_runnable(vcpu)) {
1439                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1440                         break;
1441                 }
1442                 if (signal_pending(current))
1443                         break;
1444
1445                 vcpu_put(vcpu);
1446                 schedule();
1447                 vcpu_load(vcpu);
1448         }
1449
1450         finish_wait(&vcpu->wq, &wait);
1451 }
1452
1453 void kvm_resched(struct kvm_vcpu *vcpu)
1454 {
1455         if (!need_resched())
1456                 return;
1457         cond_resched();
1458 }
1459 EXPORT_SYMBOL_GPL(kvm_resched);
1460
1461 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1462 {
1463         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1464         struct page *page;
1465
1466         if (vmf->pgoff == 0)
1467                 page = virt_to_page(vcpu->run);
1468 #ifdef CONFIG_X86
1469         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1470                 page = virt_to_page(vcpu->arch.pio_data);
1471 #endif
1472 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1473         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1474                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1475 #endif
1476         else
1477                 return VM_FAULT_SIGBUS;
1478         get_page(page);
1479         vmf->page = page;
1480         return 0;
1481 }
1482
1483 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1484         .fault = kvm_vcpu_fault,
1485 };
1486
1487 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1488 {
1489         vma->vm_ops = &kvm_vcpu_vm_ops;
1490         return 0;
1491 }
1492
1493 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1494 {
1495         struct kvm_vcpu *vcpu = filp->private_data;
1496
1497         kvm_put_kvm(vcpu->kvm);
1498         return 0;
1499 }
1500
1501 static const struct file_operations kvm_vcpu_fops = {
1502         .release        = kvm_vcpu_release,
1503         .unlocked_ioctl = kvm_vcpu_ioctl,
1504         .compat_ioctl   = kvm_vcpu_ioctl,
1505         .mmap           = kvm_vcpu_mmap,
1506 };
1507
1508 /*
1509  * Allocates an inode for the vcpu.
1510  */
1511 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1512 {
1513         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1514         if (fd < 0)
1515                 kvm_put_kvm(vcpu->kvm);
1516         return fd;
1517 }
1518
1519 /*
1520  * Creates some virtual cpus.  Good luck creating more than one.
1521  */
1522 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1523 {
1524         int r;
1525         struct kvm_vcpu *vcpu;
1526
1527         if (!valid_vcpu(n))
1528                 return -EINVAL;
1529
1530         vcpu = kvm_arch_vcpu_create(kvm, n);
1531         if (IS_ERR(vcpu))
1532                 return PTR_ERR(vcpu);
1533
1534         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1535
1536         r = kvm_arch_vcpu_setup(vcpu);
1537         if (r)
1538                 return r;
1539
1540         mutex_lock(&kvm->lock);
1541         if (kvm->vcpus[n]) {
1542                 r = -EEXIST;
1543                 goto vcpu_destroy;
1544         }
1545         kvm->vcpus[n] = vcpu;
1546         mutex_unlock(&kvm->lock);
1547
1548         /* Now it's all set up, let userspace reach it */
1549         kvm_get_kvm(kvm);
1550         r = create_vcpu_fd(vcpu);
1551         if (r < 0)
1552                 goto unlink;
1553         return r;
1554
1555 unlink:
1556         mutex_lock(&kvm->lock);
1557         kvm->vcpus[n] = NULL;
1558 vcpu_destroy:
1559         mutex_unlock(&kvm->lock);
1560         kvm_arch_vcpu_destroy(vcpu);
1561         return r;
1562 }
1563
1564 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1565 {
1566         if (sigset) {
1567                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1568                 vcpu->sigset_active = 1;
1569                 vcpu->sigset = *sigset;
1570         } else
1571                 vcpu->sigset_active = 0;
1572         return 0;
1573 }
1574
1575 static long kvm_vcpu_ioctl(struct file *filp,
1576                            unsigned int ioctl, unsigned long arg)
1577 {
1578         struct kvm_vcpu *vcpu = filp->private_data;
1579         void __user *argp = (void __user *)arg;
1580         int r;
1581         struct kvm_fpu *fpu = NULL;
1582         struct kvm_sregs *kvm_sregs = NULL;
1583
1584         if (vcpu->kvm->mm != current->mm)
1585                 return -EIO;
1586         switch (ioctl) {
1587         case KVM_RUN:
1588                 r = -EINVAL;
1589                 if (arg)
1590                         goto out;
1591                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1592                 break;
1593         case KVM_GET_REGS: {
1594                 struct kvm_regs *kvm_regs;
1595
1596                 r = -ENOMEM;
1597                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1598                 if (!kvm_regs)
1599                         goto out;
1600                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1601                 if (r)
1602                         goto out_free1;
1603                 r = -EFAULT;
1604                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1605                         goto out_free1;
1606                 r = 0;
1607 out_free1:
1608                 kfree(kvm_regs);
1609                 break;
1610         }
1611         case KVM_SET_REGS: {
1612                 struct kvm_regs *kvm_regs;
1613
1614                 r = -ENOMEM;
1615                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1616                 if (!kvm_regs)
1617                         goto out;
1618                 r = -EFAULT;
1619                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1620                         goto out_free2;
1621                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1622                 if (r)
1623                         goto out_free2;
1624                 r = 0;
1625 out_free2:
1626                 kfree(kvm_regs);
1627                 break;
1628         }
1629         case KVM_GET_SREGS: {
1630                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1631                 r = -ENOMEM;
1632                 if (!kvm_sregs)
1633                         goto out;
1634                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1635                 if (r)
1636                         goto out;
1637                 r = -EFAULT;
1638                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1639                         goto out;
1640                 r = 0;
1641                 break;
1642         }
1643         case KVM_SET_SREGS: {
1644                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1645                 r = -ENOMEM;
1646                 if (!kvm_sregs)
1647                         goto out;
1648                 r = -EFAULT;
1649                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1650                         goto out;
1651                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1652                 if (r)
1653                         goto out;
1654                 r = 0;
1655                 break;
1656         }
1657         case KVM_GET_MP_STATE: {
1658                 struct kvm_mp_state mp_state;
1659
1660                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1661                 if (r)
1662                         goto out;
1663                 r = -EFAULT;
1664                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1665                         goto out;
1666                 r = 0;
1667                 break;
1668         }
1669         case KVM_SET_MP_STATE: {
1670                 struct kvm_mp_state mp_state;
1671
1672                 r = -EFAULT;
1673                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1674                         goto out;
1675                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1676                 if (r)
1677                         goto out;
1678                 r = 0;
1679                 break;
1680         }
1681         case KVM_TRANSLATE: {
1682                 struct kvm_translation tr;
1683
1684                 r = -EFAULT;
1685                 if (copy_from_user(&tr, argp, sizeof tr))
1686                         goto out;
1687                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1688                 if (r)
1689                         goto out;
1690                 r = -EFAULT;
1691                 if (copy_to_user(argp, &tr, sizeof tr))
1692                         goto out;
1693                 r = 0;
1694                 break;
1695         }
1696         case KVM_DEBUG_GUEST: {
1697                 struct kvm_debug_guest dbg;
1698
1699                 r = -EFAULT;
1700                 if (copy_from_user(&dbg, argp, sizeof dbg))
1701                         goto out;
1702                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1703                 if (r)
1704                         goto out;
1705                 r = 0;
1706                 break;
1707         }
1708         case KVM_SET_SIGNAL_MASK: {
1709                 struct kvm_signal_mask __user *sigmask_arg = argp;
1710                 struct kvm_signal_mask kvm_sigmask;
1711                 sigset_t sigset, *p;
1712
1713                 p = NULL;
1714                 if (argp) {
1715                         r = -EFAULT;
1716                         if (copy_from_user(&kvm_sigmask, argp,
1717                                            sizeof kvm_sigmask))
1718                                 goto out;
1719                         r = -EINVAL;
1720                         if (kvm_sigmask.len != sizeof sigset)
1721                                 goto out;
1722                         r = -EFAULT;
1723                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1724                                            sizeof sigset))
1725                                 goto out;
1726                         p = &sigset;
1727                 }
1728                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1729                 break;
1730         }
1731         case KVM_GET_FPU: {
1732                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1733                 r = -ENOMEM;
1734                 if (!fpu)
1735                         goto out;
1736                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1737                 if (r)
1738                         goto out;
1739                 r = -EFAULT;
1740                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1741                         goto out;
1742                 r = 0;
1743                 break;
1744         }
1745         case KVM_SET_FPU: {
1746                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1747                 r = -ENOMEM;
1748                 if (!fpu)
1749                         goto out;
1750                 r = -EFAULT;
1751                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1752                         goto out;
1753                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1754                 if (r)
1755                         goto out;
1756                 r = 0;
1757                 break;
1758         }
1759         default:
1760                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1761         }
1762 out:
1763         kfree(fpu);
1764         kfree(kvm_sregs);
1765         return r;
1766 }
1767
1768 static long kvm_vm_ioctl(struct file *filp,
1769                            unsigned int ioctl, unsigned long arg)
1770 {
1771         struct kvm *kvm = filp->private_data;
1772         void __user *argp = (void __user *)arg;
1773         int r;
1774
1775         if (kvm->mm != current->mm)
1776                 return -EIO;
1777         switch (ioctl) {
1778         case KVM_CREATE_VCPU:
1779                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1780                 if (r < 0)
1781                         goto out;
1782                 break;
1783         case KVM_SET_USER_MEMORY_REGION: {
1784                 struct kvm_userspace_memory_region kvm_userspace_mem;
1785
1786                 r = -EFAULT;
1787                 if (copy_from_user(&kvm_userspace_mem, argp,
1788                                                 sizeof kvm_userspace_mem))
1789                         goto out;
1790
1791                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1792                 if (r)
1793                         goto out;
1794                 break;
1795         }
1796         case KVM_GET_DIRTY_LOG: {
1797                 struct kvm_dirty_log log;
1798
1799                 r = -EFAULT;
1800                 if (copy_from_user(&log, argp, sizeof log))
1801                         goto out;
1802                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1803                 if (r)
1804                         goto out;
1805                 break;
1806         }
1807 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1808         case KVM_REGISTER_COALESCED_MMIO: {
1809                 struct kvm_coalesced_mmio_zone zone;
1810                 r = -EFAULT;
1811                 if (copy_from_user(&zone, argp, sizeof zone))
1812                         goto out;
1813                 r = -ENXIO;
1814                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1815                 if (r)
1816                         goto out;
1817                 r = 0;
1818                 break;
1819         }
1820         case KVM_UNREGISTER_COALESCED_MMIO: {
1821                 struct kvm_coalesced_mmio_zone zone;
1822                 r = -EFAULT;
1823                 if (copy_from_user(&zone, argp, sizeof zone))
1824                         goto out;
1825                 r = -ENXIO;
1826                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1827                 if (r)
1828                         goto out;
1829                 r = 0;
1830                 break;
1831         }
1832 #endif
1833 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1834         case KVM_ASSIGN_PCI_DEVICE: {
1835                 struct kvm_assigned_pci_dev assigned_dev;
1836
1837                 r = -EFAULT;
1838                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1839                         goto out;
1840                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1841                 if (r)
1842                         goto out;
1843                 break;
1844         }
1845         case KVM_ASSIGN_IRQ: {
1846                 struct kvm_assigned_irq assigned_irq;
1847
1848                 r = -EFAULT;
1849                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1850                         goto out;
1851                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1852                 if (r)
1853                         goto out;
1854                 break;
1855         }
1856 #endif
1857         default:
1858                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1859         }
1860 out:
1861         return r;
1862 }
1863
1864 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1865 {
1866         struct page *page[1];
1867         unsigned long addr;
1868         int npages;
1869         gfn_t gfn = vmf->pgoff;
1870         struct kvm *kvm = vma->vm_file->private_data;
1871
1872         addr = gfn_to_hva(kvm, gfn);
1873         if (kvm_is_error_hva(addr))
1874                 return VM_FAULT_SIGBUS;
1875
1876         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1877                                 NULL);
1878         if (unlikely(npages != 1))
1879                 return VM_FAULT_SIGBUS;
1880
1881         vmf->page = page[0];
1882         return 0;
1883 }
1884
1885 static struct vm_operations_struct kvm_vm_vm_ops = {
1886         .fault = kvm_vm_fault,
1887 };
1888
1889 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1890 {
1891         vma->vm_ops = &kvm_vm_vm_ops;
1892         return 0;
1893 }
1894
1895 static const struct file_operations kvm_vm_fops = {
1896         .release        = kvm_vm_release,
1897         .unlocked_ioctl = kvm_vm_ioctl,
1898         .compat_ioctl   = kvm_vm_ioctl,
1899         .mmap           = kvm_vm_mmap,
1900 };
1901
1902 static int kvm_dev_ioctl_create_vm(void)
1903 {
1904         int fd;
1905         struct kvm *kvm;
1906
1907         kvm = kvm_create_vm();
1908         if (IS_ERR(kvm))
1909                 return PTR_ERR(kvm);
1910         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1911         if (fd < 0)
1912                 kvm_put_kvm(kvm);
1913
1914         return fd;
1915 }
1916
1917 static long kvm_dev_ioctl(struct file *filp,
1918                           unsigned int ioctl, unsigned long arg)
1919 {
1920         long r = -EINVAL;
1921
1922         switch (ioctl) {
1923         case KVM_GET_API_VERSION:
1924                 r = -EINVAL;
1925                 if (arg)
1926                         goto out;
1927                 r = KVM_API_VERSION;
1928                 break;
1929         case KVM_CREATE_VM:
1930                 r = -EINVAL;
1931                 if (arg)
1932                         goto out;
1933                 r = kvm_dev_ioctl_create_vm();
1934                 break;
1935         case KVM_CHECK_EXTENSION:
1936                 r = kvm_dev_ioctl_check_extension(arg);
1937                 break;
1938         case KVM_GET_VCPU_MMAP_SIZE:
1939                 r = -EINVAL;
1940                 if (arg)
1941                         goto out;
1942                 r = PAGE_SIZE;     /* struct kvm_run */
1943 #ifdef CONFIG_X86
1944                 r += PAGE_SIZE;    /* pio data page */
1945 #endif
1946 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1947                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1948 #endif
1949                 break;
1950         case KVM_TRACE_ENABLE:
1951         case KVM_TRACE_PAUSE:
1952         case KVM_TRACE_DISABLE:
1953                 r = kvm_trace_ioctl(ioctl, arg);
1954                 break;
1955         default:
1956                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1957         }
1958 out:
1959         return r;
1960 }
1961
1962 static struct file_operations kvm_chardev_ops = {
1963         .unlocked_ioctl = kvm_dev_ioctl,
1964         .compat_ioctl   = kvm_dev_ioctl,
1965 };
1966
1967 static struct miscdevice kvm_dev = {
1968         KVM_MINOR,
1969         "kvm",
1970         &kvm_chardev_ops,
1971 };
1972
1973 static void hardware_enable(void *junk)
1974 {
1975         int cpu = raw_smp_processor_id();
1976
1977         if (cpu_isset(cpu, cpus_hardware_enabled))
1978                 return;
1979         cpu_set(cpu, cpus_hardware_enabled);
1980         kvm_arch_hardware_enable(NULL);
1981 }
1982
1983 static void hardware_disable(void *junk)
1984 {
1985         int cpu = raw_smp_processor_id();
1986
1987         if (!cpu_isset(cpu, cpus_hardware_enabled))
1988                 return;
1989         cpu_clear(cpu, cpus_hardware_enabled);
1990         kvm_arch_hardware_disable(NULL);
1991 }
1992
1993 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1994                            void *v)
1995 {
1996         int cpu = (long)v;
1997
1998         val &= ~CPU_TASKS_FROZEN;
1999         switch (val) {
2000         case CPU_DYING:
2001                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2002                        cpu);
2003                 hardware_disable(NULL);
2004                 break;
2005         case CPU_UP_CANCELED:
2006                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2007                        cpu);
2008                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
2009                 break;
2010         case CPU_ONLINE:
2011                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2012                        cpu);
2013                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
2014                 break;
2015         }
2016         return NOTIFY_OK;
2017 }
2018
2019
2020 asmlinkage void kvm_handle_fault_on_reboot(void)
2021 {
2022         if (kvm_rebooting)
2023                 /* spin while reset goes on */
2024                 while (true)
2025                         ;
2026         /* Fault while not rebooting.  We want the trace. */
2027         BUG();
2028 }
2029 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2030
2031 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2032                       void *v)
2033 {
2034         if (val == SYS_RESTART) {
2035                 /*
2036                  * Some (well, at least mine) BIOSes hang on reboot if
2037                  * in vmx root mode.
2038                  */
2039                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2040                 kvm_rebooting = true;
2041                 on_each_cpu(hardware_disable, NULL, 1);
2042         }
2043         return NOTIFY_OK;
2044 }
2045
2046 static struct notifier_block kvm_reboot_notifier = {
2047         .notifier_call = kvm_reboot,
2048         .priority = 0,
2049 };
2050
2051 void kvm_io_bus_init(struct kvm_io_bus *bus)
2052 {
2053         memset(bus, 0, sizeof(*bus));
2054 }
2055
2056 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2057 {
2058         int i;
2059
2060         for (i = 0; i < bus->dev_count; i++) {
2061                 struct kvm_io_device *pos = bus->devs[i];
2062
2063                 kvm_iodevice_destructor(pos);
2064         }
2065 }
2066
2067 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2068                                           gpa_t addr, int len, int is_write)
2069 {
2070         int i;
2071
2072         for (i = 0; i < bus->dev_count; i++) {
2073                 struct kvm_io_device *pos = bus->devs[i];
2074
2075                 if (pos->in_range(pos, addr, len, is_write))
2076                         return pos;
2077         }
2078
2079         return NULL;
2080 }
2081
2082 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2083 {
2084         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2085
2086         bus->devs[bus->dev_count++] = dev;
2087 }
2088
2089 static struct notifier_block kvm_cpu_notifier = {
2090         .notifier_call = kvm_cpu_hotplug,
2091         .priority = 20, /* must be > scheduler priority */
2092 };
2093
2094 static int vm_stat_get(void *_offset, u64 *val)
2095 {
2096         unsigned offset = (long)_offset;
2097         struct kvm *kvm;
2098
2099         *val = 0;
2100         spin_lock(&kvm_lock);
2101         list_for_each_entry(kvm, &vm_list, vm_list)
2102                 *val += *(u32 *)((void *)kvm + offset);
2103         spin_unlock(&kvm_lock);
2104         return 0;
2105 }
2106
2107 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2108
2109 static int vcpu_stat_get(void *_offset, u64 *val)
2110 {
2111         unsigned offset = (long)_offset;
2112         struct kvm *kvm;
2113         struct kvm_vcpu *vcpu;
2114         int i;
2115
2116         *val = 0;
2117         spin_lock(&kvm_lock);
2118         list_for_each_entry(kvm, &vm_list, vm_list)
2119                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2120                         vcpu = kvm->vcpus[i];
2121                         if (vcpu)
2122                                 *val += *(u32 *)((void *)vcpu + offset);
2123                 }
2124         spin_unlock(&kvm_lock);
2125         return 0;
2126 }
2127
2128 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2129
2130 static struct file_operations *stat_fops[] = {
2131         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2132         [KVM_STAT_VM]   = &vm_stat_fops,
2133 };
2134
2135 static void kvm_init_debug(void)
2136 {
2137         struct kvm_stats_debugfs_item *p;
2138
2139         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2140         for (p = debugfs_entries; p->name; ++p)
2141                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2142                                                 (void *)(long)p->offset,
2143                                                 stat_fops[p->kind]);
2144 }
2145
2146 static void kvm_exit_debug(void)
2147 {
2148         struct kvm_stats_debugfs_item *p;
2149
2150         for (p = debugfs_entries; p->name; ++p)
2151                 debugfs_remove(p->dentry);
2152         debugfs_remove(kvm_debugfs_dir);
2153 }
2154
2155 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2156 {
2157         hardware_disable(NULL);
2158         return 0;
2159 }
2160
2161 static int kvm_resume(struct sys_device *dev)
2162 {
2163         hardware_enable(NULL);
2164         return 0;
2165 }
2166
2167 static struct sysdev_class kvm_sysdev_class = {
2168         .name = "kvm",
2169         .suspend = kvm_suspend,
2170         .resume = kvm_resume,
2171 };
2172
2173 static struct sys_device kvm_sysdev = {
2174         .id = 0,
2175         .cls = &kvm_sysdev_class,
2176 };
2177
2178 struct page *bad_page;
2179 pfn_t bad_pfn;
2180
2181 static inline
2182 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2183 {
2184         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2185 }
2186
2187 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2188 {
2189         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2190
2191         kvm_arch_vcpu_load(vcpu, cpu);
2192 }
2193
2194 static void kvm_sched_out(struct preempt_notifier *pn,
2195                           struct task_struct *next)
2196 {
2197         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2198
2199         kvm_arch_vcpu_put(vcpu);
2200 }
2201
2202 int kvm_init(void *opaque, unsigned int vcpu_size,
2203                   struct module *module)
2204 {
2205         int r;
2206         int cpu;
2207
2208         kvm_init_debug();
2209
2210         r = kvm_arch_init(opaque);
2211         if (r)
2212                 goto out_fail;
2213
2214         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2215
2216         if (bad_page == NULL) {
2217                 r = -ENOMEM;
2218                 goto out;
2219         }
2220
2221         bad_pfn = page_to_pfn(bad_page);
2222
2223         r = kvm_arch_hardware_setup();
2224         if (r < 0)
2225                 goto out_free_0;
2226
2227         for_each_online_cpu(cpu) {
2228                 smp_call_function_single(cpu,
2229                                 kvm_arch_check_processor_compat,
2230                                 &r, 1);
2231                 if (r < 0)
2232                         goto out_free_1;
2233         }
2234
2235         on_each_cpu(hardware_enable, NULL, 1);
2236         r = register_cpu_notifier(&kvm_cpu_notifier);
2237         if (r)
2238                 goto out_free_2;
2239         register_reboot_notifier(&kvm_reboot_notifier);
2240
2241         r = sysdev_class_register(&kvm_sysdev_class);
2242         if (r)
2243                 goto out_free_3;
2244
2245         r = sysdev_register(&kvm_sysdev);
2246         if (r)
2247                 goto out_free_4;
2248
2249         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2250         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2251                                            __alignof__(struct kvm_vcpu),
2252                                            0, NULL);
2253         if (!kvm_vcpu_cache) {
2254                 r = -ENOMEM;
2255                 goto out_free_5;
2256         }
2257
2258         kvm_chardev_ops.owner = module;
2259
2260         r = misc_register(&kvm_dev);
2261         if (r) {
2262                 printk(KERN_ERR "kvm: misc device register failed\n");
2263                 goto out_free;
2264         }
2265
2266         kvm_preempt_ops.sched_in = kvm_sched_in;
2267         kvm_preempt_ops.sched_out = kvm_sched_out;
2268 #ifndef CONFIG_X86
2269         msi2intx = 0;
2270 #endif
2271
2272         return 0;
2273
2274 out_free:
2275         kmem_cache_destroy(kvm_vcpu_cache);
2276 out_free_5:
2277         sysdev_unregister(&kvm_sysdev);
2278 out_free_4:
2279         sysdev_class_unregister(&kvm_sysdev_class);
2280 out_free_3:
2281         unregister_reboot_notifier(&kvm_reboot_notifier);
2282         unregister_cpu_notifier(&kvm_cpu_notifier);
2283 out_free_2:
2284         on_each_cpu(hardware_disable, NULL, 1);
2285 out_free_1:
2286         kvm_arch_hardware_unsetup();
2287 out_free_0:
2288         __free_page(bad_page);
2289 out:
2290         kvm_arch_exit();
2291         kvm_exit_debug();
2292 out_fail:
2293         return r;
2294 }
2295 EXPORT_SYMBOL_GPL(kvm_init);
2296
2297 void kvm_exit(void)
2298 {
2299         kvm_trace_cleanup();
2300         misc_deregister(&kvm_dev);
2301         kmem_cache_destroy(kvm_vcpu_cache);
2302         sysdev_unregister(&kvm_sysdev);
2303         sysdev_class_unregister(&kvm_sysdev_class);
2304         unregister_reboot_notifier(&kvm_reboot_notifier);
2305         unregister_cpu_notifier(&kvm_cpu_notifier);
2306         on_each_cpu(hardware_disable, NULL, 1);
2307         kvm_arch_hardware_unsetup();
2308         kvm_arch_exit();
2309         kvm_exit_debug();
2310         __free_page(bad_page);
2311 }
2312 EXPORT_SYMBOL_GPL(kvm_exit);