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