SAFE public projects git trees. - safe/jmp/linux-2.6/blob - virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "iodev.h"
  19
  20 #include <linux/kvm_host.h>
  21 #include <linux/kvm.h>
  22 #include <linux/module.h>
  23 #include <linux/errno.h>
  24 #include <linux/percpu.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/sysdev.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44
  45 #include <asm/processor.h>
  46 #include <asm/io.h>
  47 #include <asm/uaccess.h>
  48 #include <asm/pgtable.h>
  49
  50 #ifdef CONFIG_X86
  51 #include <asm/msidef.h>
  52 #endif
  53
  54 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
  55 #include "coalesced_mmio.h"
  56 #endif
  57
  58 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
  59 #include <linux/pci.h>
  60 #include <linux/interrupt.h>
  61 #include "irq.h"
  62 #endif
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 static int msi2intx = 1;
  68 module_param(msi2intx, bool, 0);
  69
  70 DEFINE_SPINLOCK(kvm_lock);
  71 LIST_HEAD(vm_list);
  72
  73 static cpumask_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_device(struct kvm *kvm,
 204                                      struct kvm_assigned_dev_kernel
 205                                      *assigned_dev)
 206 {
 207         if (irqchip_in_kernel(kvm) && assigned_dev->irq_requested_type)
 208                 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
 209         if (assigned_dev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
 210                 pci_disable_msi(assigned_dev->dev);
 211
 212         kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
 213
 214         if (assigned_dev->irq_source_id != -1)
 215                 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
 216         assigned_dev->irq_source_id = -1;
 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         pci_reset_function(assigned_dev->dev);
 225
 226         pci_release_regions(assigned_dev->dev);
 227         pci_disable_device(assigned_dev->dev);
 228         pci_dev_put(assigned_dev->dev);
 229
 230         list_del(&assigned_dev->list);
 231         kfree(assigned_dev);
 232 }
 233
 234 void kvm_free_all_assigned_devices(struct kvm *kvm)
 235 {
 236         struct list_head *ptr, *ptr2;
 237         struct kvm_assigned_dev_kernel *assigned_dev;
 238
 239         list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
 240                 assigned_dev = list_entry(ptr,
 241                                           struct kvm_assigned_dev_kernel,
 242                                           list);
 243
 244                 kvm_free_assigned_device(kvm, assigned_dev);
 245         }
 246 }
 247
 248 static int assigned_device_update_intx(struct kvm *kvm,
 249                         struct kvm_assigned_dev_kernel *adev,
 250                         struct kvm_assigned_irq *airq)
 251 {
 252         adev->guest_irq = airq->guest_irq;
 253         adev->ack_notifier.gsi = airq->guest_irq;
 254
 255         if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_INTX)
 256                 return 0;
 257
 258         if (irqchip_in_kernel(kvm)) {
 259                 if (!msi2intx &&
 260                     adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI) {
 261                         free_irq(adev->host_irq, (void *)kvm);
 262                         pci_disable_msi(adev->dev);
 263                 }
 264
 265                 if (!capable(CAP_SYS_RAWIO))
 266                         return -EPERM;
 267
 268                 if (airq->host_irq)
 269                         adev->host_irq = airq->host_irq;
 270                 else
 271                         adev->host_irq = adev->dev->irq;
 272
 273                 /* Even though this is PCI, we don't want to use shared
 274                  * interrupts. Sharing host devices with guest-assigned devices
 275                  * on the same interrupt line is not a happy situation: there
 276                  * are going to be long delays in accepting, acking, etc.
 277                  */
 278                 if (request_irq(adev->host_irq, kvm_assigned_dev_intr,
 279                                 0, "kvm_assigned_intx_device", (void *)adev))
 280                         return -EIO;
 281         }
 282
 283         adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_INTX |
 284                                    KVM_ASSIGNED_DEV_HOST_INTX;
 285         return 0;
 286 }
 287
 288 #ifdef CONFIG_X86
 289 static int assigned_device_update_msi(struct kvm *kvm,
 290                         struct kvm_assigned_dev_kernel *adev,
 291                         struct kvm_assigned_irq *airq)
 292 {
 293         int r;
 294
 295         if (airq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI) {
 296                 /* x86 don't care upper address of guest msi message addr */
 297                 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_MSI;
 298                 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_INTX;
 299                 adev->guest_msi.address_lo = airq->guest_msi.addr_lo;
 300                 adev->guest_msi.data = airq->guest_msi.data;
 301                 adev->ack_notifier.gsi = -1;
 302         } else if (msi2intx) {
 303                 adev->irq_requested_type |= KVM_ASSIGNED_DEV_GUEST_INTX;
 304                 adev->irq_requested_type &= ~KVM_ASSIGNED_DEV_GUEST_MSI;
 305                 adev->guest_irq = airq->guest_irq;
 306                 adev->ack_notifier.gsi = airq->guest_irq;
 307         }
 308
 309         if (adev->irq_requested_type & KVM_ASSIGNED_DEV_HOST_MSI)
 310                 return 0;
 311
 312         if (irqchip_in_kernel(kvm)) {
 313                 if (!msi2intx) {
 314                         if (adev->irq_requested_type &
 315                                         KVM_ASSIGNED_DEV_HOST_INTX)
 316                                 free_irq(adev->host_irq, (void *)adev);
 317
 318                         r = pci_enable_msi(adev->dev);
 319                         if (r)
 320                                 return r;
 321                 }
 322
 323                 adev->host_irq = adev->dev->irq;
 324                 if (request_irq(adev->host_irq, kvm_assigned_dev_intr, 0,
 325                                 "kvm_assigned_msi_device", (void *)adev))
 326                         return -EIO;
 327         }
 328
 329         if (!msi2intx)
 330                 adev->irq_requested_type = KVM_ASSIGNED_DEV_GUEST_MSI;
 331
 332         adev->irq_requested_type |= KVM_ASSIGNED_DEV_HOST_MSI;
 333         return 0;
 334 }
 335 #endif
 336
 337 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
 338                                    struct kvm_assigned_irq
 339                                    *assigned_irq)
 340 {
 341         int r = 0;
 342         struct kvm_assigned_dev_kernel *match;
 343
 344         mutex_lock(&kvm->lock);
 345
 346         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 347                                       assigned_irq->assigned_dev_id);
 348         if (!match) {
 349                 mutex_unlock(&kvm->lock);
 350                 return -EINVAL;
 351         }
 352
 353         if (!match->irq_requested_type) {
 354                 INIT_WORK(&match->interrupt_work,
 355                                 kvm_assigned_dev_interrupt_work_handler);
 356                 if (irqchip_in_kernel(kvm)) {
 357                         /* Register ack nofitier */
 358                         match->ack_notifier.gsi = -1;
 359                         match->ack_notifier.irq_acked =
 360                                         kvm_assigned_dev_ack_irq;
 361                         kvm_register_irq_ack_notifier(kvm,
 362                                         &match->ack_notifier);
 363
 364                         /* Request IRQ source ID */
 365                         r = kvm_request_irq_source_id(kvm);
 366                         if (r < 0)
 367                                 goto out_release;
 368                         else
 369                                 match->irq_source_id = r;
 370
 371 #ifdef CONFIG_X86
 372                         /* Determine host device irq type, we can know the
 373                          * result from dev->msi_enabled */
 374                         if (msi2intx)
 375                                 pci_enable_msi(match->dev);
 376 #endif
 377                 }
 378         }
 379
 380         if ((!msi2intx &&
 381              (assigned_irq->flags & KVM_DEV_IRQ_ASSIGN_ENABLE_MSI)) ||
 382             (msi2intx && match->dev->msi_enabled)) {
 383 #ifdef CONFIG_X86
 384                 r = assigned_device_update_msi(kvm, match, assigned_irq);
 385                 if (r) {
 386                         printk(KERN_WARNING "kvm: failed to enable "
 387                                         "MSI device!\n");
 388                         goto out_release;
 389                 }
 390 #else
 391                 r = -ENOTTY;
 392 #endif
 393         } else if (assigned_irq->host_irq == 0 && match->dev->irq == 0) {
 394                 /* Host device IRQ 0 means don't support INTx */
 395                 if (!msi2intx) {
 396                         printk(KERN_WARNING
 397                                "kvm: wait device to enable MSI!\n");
 398                         r = 0;
 399                 } else {
 400                         printk(KERN_WARNING
 401                                "kvm: failed to enable MSI device!\n");
 402                         r = -ENOTTY;
 403                         goto out_release;
 404                 }
 405         } else {
 406                 /* Non-sharing INTx mode */
 407                 r = assigned_device_update_intx(kvm, match, assigned_irq);
 408                 if (r) {
 409                         printk(KERN_WARNING "kvm: failed to enable "
 410                                         "INTx device!\n");
 411                         goto out_release;
 412                 }
 413         }
 414
 415         mutex_unlock(&kvm->lock);
 416         return r;
 417 out_release:
 418         mutex_unlock(&kvm->lock);
 419         kvm_free_assigned_device(kvm, match);
 420         return r;
 421 }
 422
 423 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
 424                                       struct kvm_assigned_pci_dev *assigned_dev)
 425 {
 426         int r = 0;
 427         struct kvm_assigned_dev_kernel *match;
 428         struct pci_dev *dev;
 429
 430         mutex_lock(&kvm->lock);
 431
 432         match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
 433                                       assigned_dev->assigned_dev_id);
 434         if (match) {
 435                 /* device already assigned */
 436                 r = -EINVAL;
 437                 goto out;
 438         }
 439
 440         match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
 441         if (match == NULL) {
 442                 printk(KERN_INFO "%s: Couldn't allocate memory\n",
 443                        __func__);
 444                 r = -ENOMEM;
 445                 goto out;
 446         }
 447         dev = pci_get_bus_and_slot(assigned_dev->busnr,
 448                                    assigned_dev->devfn);
 449         if (!dev) {
 450                 printk(KERN_INFO "%s: host device not found\n", __func__);
 451                 r = -EINVAL;
 452                 goto out_free;
 453         }
 454         if (pci_enable_device(dev)) {
 455                 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
 456                 r = -EBUSY;
 457                 goto out_put;
 458         }
 459         r = pci_request_regions(dev, "kvm_assigned_device");
 460         if (r) {
 461                 printk(KERN_INFO "%s: Could not get access to device regions\n",
 462                        __func__);
 463                 goto out_disable;
 464         }
 465
 466         pci_reset_function(dev);
 467
 468         match->assigned_dev_id = assigned_dev->assigned_dev_id;
 469         match->host_busnr = assigned_dev->busnr;
 470         match->host_devfn = assigned_dev->devfn;
 471         match->dev = dev;
 472         match->irq_source_id = -1;
 473         match->kvm = kvm;
 474
 475         list_add(&match->list, &kvm->arch.assigned_dev_head);
 476
 477         if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
 478                 r = kvm_iommu_map_guest(kvm, match);
 479                 if (r)
 480                         goto out_list_del;
 481         }
 482
 483 out:
 484         mutex_unlock(&kvm->lock);
 485         return r;
 486 out_list_del:
 487         list_del(&match->list);
 488         pci_release_regions(dev);
 489 out_disable:
 490         pci_disable_device(dev);
 491 out_put:
 492         pci_dev_put(dev);
 493 out_free:
 494         kfree(match);
 495         mutex_unlock(&kvm->lock);
 496         return r;
 497 }
 498 #endif
 499
 500 static inline int valid_vcpu(int n)
 501 {
 502         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 503 }
 504
 505 inline int kvm_is_mmio_pfn(pfn_t pfn)
 506 {
 507         if (pfn_valid(pfn))
 508                 return PageReserved(pfn_to_page(pfn));
 509
 510         return true;
 511 }
 512
 513 /*
 514  * Switches to specified vcpu, until a matching vcpu_put()
 515  */
 516 void vcpu_load(struct kvm_vcpu *vcpu)
 517 {
 518         int cpu;
 519
 520         mutex_lock(&vcpu->mutex);
 521         cpu = get_cpu();
 522         preempt_notifier_register(&vcpu->preempt_notifier);
 523         kvm_arch_vcpu_load(vcpu, cpu);
 524         put_cpu();
 525 }
 526
 527 void vcpu_put(struct kvm_vcpu *vcpu)
 528 {
 529         preempt_disable();
 530         kvm_arch_vcpu_put(vcpu);
 531         preempt_notifier_unregister(&vcpu->preempt_notifier);
 532         preempt_enable();
 533         mutex_unlock(&vcpu->mutex);
 534 }
 535
 536 static void ack_flush(void *_completed)
 537 {
 538 }
 539
 540 void kvm_flush_remote_tlbs(struct kvm *kvm)
 541 {
 542         int i, cpu, me;
 543         cpumask_t cpus;
 544         struct kvm_vcpu *vcpu;
 545
 546         me = get_cpu();
 547         cpus_clear(cpus);
 548         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 549                 vcpu = kvm->vcpus[i];
 550                 if (!vcpu)
 551                         continue;
 552                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 553                         continue;
 554                 cpu = vcpu->cpu;
 555                 if (cpu != -1 && cpu != me)
 556                         cpu_set(cpu, cpus);
 557         }
 558         if (cpus_empty(cpus))
 559                 goto out;
 560         ++kvm->stat.remote_tlb_flush;
 561         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 562 out:
 563         put_cpu();
 564 }
 565
 566 void kvm_reload_remote_mmus(struct kvm *kvm)
 567 {
 568         int i, cpu, me;
 569         cpumask_t cpus;
 570         struct kvm_vcpu *vcpu;
 571
 572         me = get_cpu();
 573         cpus_clear(cpus);
 574         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 575                 vcpu = kvm->vcpus[i];
 576                 if (!vcpu)
 577                         continue;
 578                 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
 579                         continue;
 580                 cpu = vcpu->cpu;
 581                 if (cpu != -1 && cpu != me)
 582                         cpu_set(cpu, cpus);
 583         }
 584         if (cpus_empty(cpus))
 585                 goto out;
 586         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 587 out:
 588         put_cpu();
 589 }
 590
 591
 592 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 593 {
 594         struct page *page;
 595         int r;
 596
 597         mutex_init(&vcpu->mutex);
 598         vcpu->cpu = -1;
 599         vcpu->kvm = kvm;
 600         vcpu->vcpu_id = id;
 601         init_waitqueue_head(&vcpu->wq);
 602
 603         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 604         if (!page) {
 605                 r = -ENOMEM;
 606                 goto fail;
 607         }
 608         vcpu->run = page_address(page);
 609
 610         r = kvm_arch_vcpu_init(vcpu);
 611         if (r < 0)
 612                 goto fail_free_run;
 613         return 0;
 614
 615 fail_free_run:
 616         free_page((unsigned long)vcpu->run);
 617 fail:
 618         return r;
 619 }
 620 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 621
 622 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 623 {
 624         kvm_arch_vcpu_uninit(vcpu);
 625         free_page((unsigned long)vcpu->run);
 626 }
 627 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 628
 629 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 630 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 631 {
 632         return container_of(mn, struct kvm, mmu_notifier);
 633 }
 634
 635 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 636                                              struct mm_struct *mm,
 637                                              unsigned long address)
 638 {
 639         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 640         int need_tlb_flush;
 641
 642         /*
 643          * When ->invalidate_page runs, the linux pte has been zapped
 644          * already but the page is still allocated until
 645          * ->invalidate_page returns. So if we increase the sequence
 646          * here the kvm page fault will notice if the spte can't be
 647          * established because the page is going to be freed. If
 648          * instead the kvm page fault establishes the spte before
 649          * ->invalidate_page runs, kvm_unmap_hva will release it
 650          * before returning.
 651          *
 652          * The sequence increase only need to be seen at spin_unlock
 653          * time, and not at spin_lock time.
 654          *
 655          * Increasing the sequence after the spin_unlock would be
 656          * unsafe because the kvm page fault could then establish the
 657          * pte after kvm_unmap_hva returned, without noticing the page
 658          * is going to be freed.
 659          */
 660         spin_lock(&kvm->mmu_lock);
 661         kvm->mmu_notifier_seq++;
 662         need_tlb_flush = kvm_unmap_hva(kvm, address);
 663         spin_unlock(&kvm->mmu_lock);
 664
 665         /* we've to flush the tlb before the pages can be freed */
 666         if (need_tlb_flush)
 667                 kvm_flush_remote_tlbs(kvm);
 668
 669 }
 670
 671 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 672                                                     struct mm_struct *mm,
 673                                                     unsigned long start,
 674                                                     unsigned long end)
 675 {
 676         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 677         int need_tlb_flush = 0;
 678
 679         spin_lock(&kvm->mmu_lock);
 680         /*
 681          * The count increase must become visible at unlock time as no
 682          * spte can be established without taking the mmu_lock and
 683          * count is also read inside the mmu_lock critical section.
 684          */
 685         kvm->mmu_notifier_count++;
 686         for (; start < end; start += PAGE_SIZE)
 687                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 688         spin_unlock(&kvm->mmu_lock);
 689
 690         /* we've to flush the tlb before the pages can be freed */
 691         if (need_tlb_flush)
 692                 kvm_flush_remote_tlbs(kvm);
 693 }
 694
 695 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 696                                                   struct mm_struct *mm,
 697                                                   unsigned long start,
 698                                                   unsigned long end)
 699 {
 700         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 701
 702         spin_lock(&kvm->mmu_lock);
 703         /*
 704          * This sequence increase will notify the kvm page fault that
 705          * the page that is going to be mapped in the spte could have
 706          * been freed.
 707          */
 708         kvm->mmu_notifier_seq++;
 709         /*
 710          * The above sequence increase must be visible before the
 711          * below count decrease but both values are read by the kvm
 712          * page fault under mmu_lock spinlock so we don't need to add
 713          * a smb_wmb() here in between the two.
 714          */
 715         kvm->mmu_notifier_count--;
 716         spin_unlock(&kvm->mmu_lock);
 717
 718         BUG_ON(kvm->mmu_notifier_count < 0);
 719 }
 720
 721 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 722                                               struct mm_struct *mm,
 723                                               unsigned long address)
 724 {
 725         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 726         int young;
 727
 728         spin_lock(&kvm->mmu_lock);
 729         young = kvm_age_hva(kvm, address);
 730         spin_unlock(&kvm->mmu_lock);
 731
 732         if (young)
 733                 kvm_flush_remote_tlbs(kvm);
 734
 735         return young;
 736 }
 737
 738 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 739         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 740         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 741         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 742         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 743 };
 744 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 745
 746 static struct kvm *kvm_create_vm(void)
 747 {
 748         struct kvm *kvm = kvm_arch_create_vm();
 749 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 750         struct page *page;
 751 #endif
 752
 753         if (IS_ERR(kvm))
 754                 goto out;
 755
 756 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 757         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 758         if (!page) {
 759                 kfree(kvm);
 760                 return ERR_PTR(-ENOMEM);
 761         }
 762         kvm->coalesced_mmio_ring =
 763                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 764 #endif
 765
 766 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 767         {
 768                 int err;
 769                 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 770                 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 771                 if (err) {
 772 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 773                         put_page(page);
 774 #endif
 775                         kfree(kvm);
 776                         return ERR_PTR(err);
 777                 }
 778         }
 779 #endif
 780
 781         kvm->mm = current->mm;
 782         atomic_inc(&kvm->mm->mm_count);
 783         spin_lock_init(&kvm->mmu_lock);
 784         kvm_io_bus_init(&kvm->pio_bus);
 785         mutex_init(&kvm->lock);
 786         kvm_io_bus_init(&kvm->mmio_bus);
 787         init_rwsem(&kvm->slots_lock);
 788         atomic_set(&kvm->users_count, 1);
 789         spin_lock(&kvm_lock);
 790         list_add(&kvm->vm_list, &vm_list);
 791         spin_unlock(&kvm_lock);
 792 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 793         kvm_coalesced_mmio_init(kvm);
 794 #endif
 795 out:
 796         return kvm;
 797 }
 798
 799 /*
 800  * Free any memory in @free but not in @dont.
 801  */
 802 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 803                                   struct kvm_memory_slot *dont)
 804 {
 805         if (!dont || free->rmap != dont->rmap)
 806                 vfree(free->rmap);
 807
 808         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 809                 vfree(free->dirty_bitmap);
 810
 811         if (!dont || free->lpage_info != dont->lpage_info)
 812                 vfree(free->lpage_info);
 813
 814         free->npages = 0;
 815         free->dirty_bitmap = NULL;
 816         free->rmap = NULL;
 817         free->lpage_info = NULL;
 818 }
 819
 820 void kvm_free_physmem(struct kvm *kvm)
 821 {
 822         int i;
 823
 824         for (i = 0; i < kvm->nmemslots; ++i)
 825                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 826 }
 827
 828 static void kvm_destroy_vm(struct kvm *kvm)
 829 {
 830         struct mm_struct *mm = kvm->mm;
 831
 832         spin_lock(&kvm_lock);
 833         list_del(&kvm->vm_list);
 834         spin_unlock(&kvm_lock);
 835         kvm_io_bus_destroy(&kvm->pio_bus);
 836         kvm_io_bus_destroy(&kvm->mmio_bus);
 837 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 838         if (kvm->coalesced_mmio_ring != NULL)
 839                 free_page((unsigned long)kvm->coalesced_mmio_ring);
 840 #endif
 841 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 842         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 843 #endif
 844         kvm_arch_destroy_vm(kvm);
 845         mmdrop(mm);
 846 }
 847
 848 void kvm_get_kvm(struct kvm *kvm)
 849 {
 850         atomic_inc(&kvm->users_count);
 851 }
 852 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 853
 854 void kvm_put_kvm(struct kvm *kvm)
 855 {
 856         if (atomic_dec_and_test(&kvm->users_count))
 857                 kvm_destroy_vm(kvm);
 858 }
 859 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 860
 861
 862 static int kvm_vm_release(struct inode *inode, struct file *filp)
 863 {
 864         struct kvm *kvm = filp->private_data;
 865
 866         kvm_put_kvm(kvm);
 867         return 0;
 868 }
 869
 870 /*
 871  * Allocate some memory and give it an address in the guest physical address
 872  * space.
 873  *
 874  * Discontiguous memory is allowed, mostly for framebuffers.
 875  *
 876  * Must be called holding mmap_sem for write.
 877  */
 878 int __kvm_set_memory_region(struct kvm *kvm,
 879                             struct kvm_userspace_memory_region *mem,
 880                             int user_alloc)
 881 {
 882         int r;
 883         gfn_t base_gfn;
 884         unsigned long npages;
 885         unsigned long i;
 886         struct kvm_memory_slot *memslot;
 887         struct kvm_memory_slot old, new;
 888
 889         r = -EINVAL;
 890         /* General sanity checks */
 891         if (mem->memory_size & (PAGE_SIZE - 1))
 892                 goto out;
 893         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 894                 goto out;
 895         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 896                 goto out;
 897         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 898                 goto out;
 899         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 900                 goto out;
 901
 902         memslot = &kvm->memslots[mem->slot];
 903         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 904         npages = mem->memory_size >> PAGE_SHIFT;
 905
 906         if (!npages)
 907                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 908
 909         new = old = *memslot;
 910
 911         new.base_gfn = base_gfn;
 912         new.npages = npages;
 913         new.flags = mem->flags;
 914
 915         /* Disallow changing a memory slot's size. */
 916         r = -EINVAL;
 917         if (npages && old.npages && npages != old.npages)
 918                 goto out_free;
 919
 920         /* Check for overlaps */
 921         r = -EEXIST;
 922         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 923                 struct kvm_memory_slot *s = &kvm->memslots[i];
 924
 925                 if (s == memslot)
 926                         continue;
 927                 if (!((base_gfn + npages <= s->base_gfn) ||
 928                       (base_gfn >= s->base_gfn + s->npages)))
 929                         goto out_free;
 930         }
 931
 932         /* Free page dirty bitmap if unneeded */
 933         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 934                 new.dirty_bitmap = NULL;
 935
 936         r = -ENOMEM;
 937
 938         /* Allocate if a slot is being created */
 939 #ifndef CONFIG_S390
 940         if (npages && !new.rmap) {
 941                 new.rmap = vmalloc(npages * sizeof(struct page *));
 942
 943                 if (!new.rmap)
 944                         goto out_free;
 945
 946                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 947
 948                 new.user_alloc = user_alloc;
 949                 /*
 950                  * hva_to_rmmap() serialzies with the mmu_lock and to be
 951                  * safe it has to ignore memslots with !user_alloc &&
 952                  * !userspace_addr.
 953                  */
 954                 if (user_alloc)
 955                         new.userspace_addr = mem->userspace_addr;
 956                 else
 957                         new.userspace_addr = 0;
 958         }
 959         if (npages && !new.lpage_info) {
 960                 int largepages = npages / KVM_PAGES_PER_HPAGE;
 961                 if (npages % KVM_PAGES_PER_HPAGE)
 962                         largepages++;
 963                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 964                         largepages++;
 965
 966                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
 967
 968                 if (!new.lpage_info)
 969                         goto out_free;
 970
 971                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
 972
 973                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 974                         new.lpage_info[0].write_count = 1;
 975                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
 976                         new.lpage_info[largepages-1].write_count = 1;
 977         }
 978
 979         /* Allocate page dirty bitmap if needed */
 980         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 981                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 982
 983                 new.dirty_bitmap = vmalloc(dirty_bytes);
 984                 if (!new.dirty_bitmap)
 985                         goto out_free;
 986                 memset(new.dirty_bitmap, 0, dirty_bytes);
 987         }
 988 #endif /* not defined CONFIG_S390 */
 989
 990         if (!npages)
 991                 kvm_arch_flush_shadow(kvm);
 992
 993         spin_lock(&kvm->mmu_lock);
 994         if (mem->slot >= kvm->nmemslots)
 995                 kvm->nmemslots = mem->slot + 1;
 996
 997         *memslot = new;
 998         spin_unlock(&kvm->mmu_lock);
 999
1000         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
1001         if (r) {
1002                 spin_lock(&kvm->mmu_lock);
1003                 *memslot = old;
1004                 spin_unlock(&kvm->mmu_lock);
1005                 goto out_free;
1006         }
1007
1008         kvm_free_physmem_slot(&old, &new);
1009 #ifdef CONFIG_DMAR
1010         /* map the pages in iommu page table */
1011         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
1012         if (r)
1013                 goto out;
1014 #endif
1015         return 0;
1016
1017 out_free:
1018         kvm_free_physmem_slot(&new, &old);
1019 out:
1020         return r;
1021
1022 }
1023 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1024
1025 int kvm_set_memory_region(struct kvm *kvm,
1026                           struct kvm_userspace_memory_region *mem,
1027                           int user_alloc)
1028 {
1029         int r;
1030
1031         down_write(&kvm->slots_lock);
1032         r = __kvm_set_memory_region(kvm, mem, user_alloc);
1033         up_write(&kvm->slots_lock);
1034         return r;
1035 }
1036 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1037
1038 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1039                                    struct
1040                                    kvm_userspace_memory_region *mem,
1041                                    int user_alloc)
1042 {
1043         if (mem->slot >= KVM_MEMORY_SLOTS)
1044                 return -EINVAL;
1045         return kvm_set_memory_region(kvm, mem, user_alloc);
1046 }
1047
1048 int kvm_get_dirty_log(struct kvm *kvm,
1049                         struct kvm_dirty_log *log, int *is_dirty)
1050 {
1051         struct kvm_memory_slot *memslot;
1052         int r, i;
1053         int n;
1054         unsigned long any = 0;
1055
1056         r = -EINVAL;
1057         if (log->slot >= KVM_MEMORY_SLOTS)
1058                 goto out;
1059
1060         memslot = &kvm->memslots[log->slot];
1061         r = -ENOENT;
1062         if (!memslot->dirty_bitmap)
1063                 goto out;
1064
1065         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
1066
1067         for (i = 0; !any && i < n/sizeof(long); ++i)
1068                 any = memslot->dirty_bitmap[i];
1069
1070         r = -EFAULT;
1071         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1072                 goto out;
1073
1074         if (any)
1075                 *is_dirty = 1;
1076
1077         r = 0;
1078 out:
1079         return r;
1080 }
1081
1082 int is_error_page(struct page *page)
1083 {
1084         return page == bad_page;
1085 }
1086 EXPORT_SYMBOL_GPL(is_error_page);
1087
1088 int is_error_pfn(pfn_t pfn)
1089 {
1090         return pfn == bad_pfn;
1091 }
1092 EXPORT_SYMBOL_GPL(is_error_pfn);
1093
1094 static inline unsigned long bad_hva(void)
1095 {
1096         return PAGE_OFFSET;
1097 }
1098
1099 int kvm_is_error_hva(unsigned long addr)
1100 {
1101         return addr == bad_hva();
1102 }
1103 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
1104
1105 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
1106 {
1107         int i;
1108
1109         for (i = 0; i < kvm->nmemslots; ++i) {
1110                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1111
1112                 if (gfn >= memslot->base_gfn
1113                     && gfn < memslot->base_gfn + memslot->npages)
1114                         return memslot;
1115         }
1116         return NULL;
1117 }
1118 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
1119
1120 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1121 {
1122         gfn = unalias_gfn(kvm, gfn);
1123         return gfn_to_memslot_unaliased(kvm, gfn);
1124 }
1125
1126 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1127 {
1128         int i;
1129
1130         gfn = unalias_gfn(kvm, gfn);
1131         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1132                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
1133
1134                 if (gfn >= memslot->base_gfn
1135                     && gfn < memslot->base_gfn + memslot->npages)
1136                         return 1;
1137         }
1138         return 0;
1139 }
1140 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1141
1142 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1143 {
1144         struct kvm_memory_slot *slot;
1145
1146         gfn = unalias_gfn(kvm, gfn);
1147         slot = gfn_to_memslot_unaliased(kvm, gfn);
1148         if (!slot)
1149                 return bad_hva();
1150         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
1151 }
1152 EXPORT_SYMBOL_GPL(gfn_to_hva);
1153
1154 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1155 {
1156         struct page *page[1];
1157         unsigned long addr;
1158         int npages;
1159         pfn_t pfn;
1160
1161         might_sleep();
1162
1163         addr = gfn_to_hva(kvm, gfn);
1164         if (kvm_is_error_hva(addr)) {
1165                 get_page(bad_page);
1166                 return page_to_pfn(bad_page);
1167         }
1168
1169         npages = get_user_pages_fast(addr, 1, 1, page);
1170
1171         if (unlikely(npages != 1)) {
1172                 struct vm_area_struct *vma;
1173
1174                 down_read(&current->mm->mmap_sem);
1175                 vma = find_vma(current->mm, addr);
1176
1177                 if (vma == NULL || addr < vma->vm_start ||
1178                     !(vma->vm_flags & VM_PFNMAP)) {
1179                         up_read(&current->mm->mmap_sem);
1180                         get_page(bad_page);
1181                         return page_to_pfn(bad_page);
1182                 }
1183
1184                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1185                 up_read(&current->mm->mmap_sem);
1186                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1187         } else
1188                 pfn = page_to_pfn(page[0]);
1189
1190         return pfn;
1191 }
1192
1193 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1194
1195 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1196 {
1197         pfn_t pfn;
1198
1199         pfn = gfn_to_pfn(kvm, gfn);
1200         if (!kvm_is_mmio_pfn(pfn))
1201                 return pfn_to_page(pfn);
1202
1203         WARN_ON(kvm_is_mmio_pfn(pfn));
1204
1205         get_page(bad_page);
1206         return bad_page;
1207 }
1208
1209 EXPORT_SYMBOL_GPL(gfn_to_page);
1210
1211 void kvm_release_page_clean(struct page *page)
1212 {
1213         kvm_release_pfn_clean(page_to_pfn(page));
1214 }
1215 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1216
1217 void kvm_release_pfn_clean(pfn_t pfn)
1218 {
1219         if (!kvm_is_mmio_pfn(pfn))
1220                 put_page(pfn_to_page(pfn));
1221 }
1222 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1223
1224 void kvm_release_page_dirty(struct page *page)
1225 {
1226         kvm_release_pfn_dirty(page_to_pfn(page));
1227 }
1228 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1229
1230 void kvm_release_pfn_dirty(pfn_t pfn)
1231 {
1232         kvm_set_pfn_dirty(pfn);
1233         kvm_release_pfn_clean(pfn);
1234 }
1235 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1236
1237 void kvm_set_page_dirty(struct page *page)
1238 {
1239         kvm_set_pfn_dirty(page_to_pfn(page));
1240 }
1241 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1242
1243 void kvm_set_pfn_dirty(pfn_t pfn)
1244 {
1245         if (!kvm_is_mmio_pfn(pfn)) {
1246                 struct page *page = pfn_to_page(pfn);
1247                 if (!PageReserved(page))
1248                         SetPageDirty(page);
1249         }
1250 }
1251 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1252
1253 void kvm_set_pfn_accessed(pfn_t pfn)
1254 {
1255         if (!kvm_is_mmio_pfn(pfn))
1256                 mark_page_accessed(pfn_to_page(pfn));
1257 }
1258 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1259
1260 void kvm_get_pfn(pfn_t pfn)
1261 {
1262         if (!kvm_is_mmio_pfn(pfn))
1263                 get_page(pfn_to_page(pfn));
1264 }
1265 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1266
1267 static int next_segment(unsigned long len, int offset)
1268 {
1269         if (len > PAGE_SIZE - offset)
1270                 return PAGE_SIZE - offset;
1271         else
1272                 return len;
1273 }
1274
1275 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1276                         int len)
1277 {
1278         int r;
1279         unsigned long addr;
1280
1281         addr = gfn_to_hva(kvm, gfn);
1282         if (kvm_is_error_hva(addr))
1283                 return -EFAULT;
1284         r = copy_from_user(data, (void __user *)addr + offset, len);
1285         if (r)
1286                 return -EFAULT;
1287         return 0;
1288 }
1289 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1290
1291 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1292 {
1293         gfn_t gfn = gpa >> PAGE_SHIFT;
1294         int seg;
1295         int offset = offset_in_page(gpa);
1296         int ret;
1297
1298         while ((seg = next_segment(len, offset)) != 0) {
1299                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1300                 if (ret < 0)
1301                         return ret;
1302                 offset = 0;
1303                 len -= seg;
1304                 data += seg;
1305                 ++gfn;
1306         }
1307         return 0;
1308 }
1309 EXPORT_SYMBOL_GPL(kvm_read_guest);
1310
1311 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1312                           unsigned long len)
1313 {
1314         int r;
1315         unsigned long addr;
1316         gfn_t gfn = gpa >> PAGE_SHIFT;
1317         int offset = offset_in_page(gpa);
1318
1319         addr = gfn_to_hva(kvm, gfn);
1320         if (kvm_is_error_hva(addr))
1321                 return -EFAULT;
1322         pagefault_disable();
1323         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1324         pagefault_enable();
1325         if (r)
1326                 return -EFAULT;
1327         return 0;
1328 }
1329 EXPORT_SYMBOL(kvm_read_guest_atomic);
1330
1331 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1332                          int offset, int len)
1333 {
1334         int r;
1335         unsigned long addr;
1336
1337         addr = gfn_to_hva(kvm, gfn);
1338         if (kvm_is_error_hva(addr))
1339                 return -EFAULT;
1340         r = copy_to_user((void __user *)addr + offset, data, len);
1341         if (r)
1342                 return -EFAULT;
1343         mark_page_dirty(kvm, gfn);
1344         return 0;
1345 }
1346 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1347
1348 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1349                     unsigned long len)
1350 {
1351         gfn_t gfn = gpa >> PAGE_SHIFT;
1352         int seg;
1353         int offset = offset_in_page(gpa);
1354         int ret;
1355
1356         while ((seg = next_segment(len, offset)) != 0) {
1357                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1358                 if (ret < 0)
1359                         return ret;
1360                 offset = 0;
1361                 len -= seg;
1362                 data += seg;
1363                 ++gfn;
1364         }
1365         return 0;
1366 }
1367
1368 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1369 {
1370         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1371 }
1372 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1373
1374 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1375 {
1376         gfn_t gfn = gpa >> PAGE_SHIFT;
1377         int seg;
1378         int offset = offset_in_page(gpa);
1379         int ret;
1380
1381         while ((seg = next_segment(len, offset)) != 0) {
1382                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1383                 if (ret < 0)
1384                         return ret;
1385                 offset = 0;
1386                 len -= seg;
1387                 ++gfn;
1388         }
1389         return 0;
1390 }
1391 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1392
1393 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1394 {
1395         struct kvm_memory_slot *memslot;
1396
1397         gfn = unalias_gfn(kvm, gfn);
1398         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1399         if (memslot && memslot->dirty_bitmap) {
1400                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1401
1402                 /* avoid RMW */
1403                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1404                         set_bit(rel_gfn, memslot->dirty_bitmap);
1405         }
1406 }
1407
1408 /*
1409  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1410  */
1411 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1412 {
1413         DEFINE_WAIT(wait);
1414
1415         for (;;) {
1416                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1417
1418                 if (kvm_cpu_has_interrupt(vcpu) ||
1419                     kvm_cpu_has_pending_timer(vcpu) ||
1420                     kvm_arch_vcpu_runnable(vcpu)) {
1421                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1422                         break;
1423                 }
1424                 if (signal_pending(current))
1425                         break;
1426
1427                 vcpu_put(vcpu);
1428                 schedule();
1429                 vcpu_load(vcpu);
1430         }
1431
1432         finish_wait(&vcpu->wq, &wait);
1433 }
1434
1435 void kvm_resched(struct kvm_vcpu *vcpu)
1436 {
1437         if (!need_resched())
1438                 return;
1439         cond_resched();
1440 }
1441 EXPORT_SYMBOL_GPL(kvm_resched);
1442
1443 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1444 {
1445         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1446         struct page *page;
1447
1448         if (vmf->pgoff == 0)
1449                 page = virt_to_page(vcpu->run);
1450 #ifdef CONFIG_X86
1451         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1452                 page = virt_to_page(vcpu->arch.pio_data);
1453 #endif
1454 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1455         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1456                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1457 #endif
1458         else
1459                 return VM_FAULT_SIGBUS;
1460         get_page(page);
1461         vmf->page = page;
1462         return 0;
1463 }
1464
1465 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1466         .fault = kvm_vcpu_fault,
1467 };
1468
1469 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1470 {
1471         vma->vm_ops = &kvm_vcpu_vm_ops;
1472         return 0;
1473 }
1474
1475 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1476 {
1477         struct kvm_vcpu *vcpu = filp->private_data;
1478
1479         kvm_put_kvm(vcpu->kvm);
1480         return 0;
1481 }
1482
1483 static const struct file_operations kvm_vcpu_fops = {
1484         .release        = kvm_vcpu_release,
1485         .unlocked_ioctl = kvm_vcpu_ioctl,
1486         .compat_ioctl   = kvm_vcpu_ioctl,
1487         .mmap           = kvm_vcpu_mmap,
1488 };
1489
1490 /*
1491  * Allocates an inode for the vcpu.
1492  */
1493 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1494 {
1495         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1496         if (fd < 0)
1497                 kvm_put_kvm(vcpu->kvm);
1498         return fd;
1499 }
1500
1501 /*
1502  * Creates some virtual cpus.  Good luck creating more than one.
1503  */
1504 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1505 {
1506         int r;
1507         struct kvm_vcpu *vcpu;
1508
1509         if (!valid_vcpu(n))
1510                 return -EINVAL;
1511
1512         vcpu = kvm_arch_vcpu_create(kvm, n);
1513         if (IS_ERR(vcpu))
1514                 return PTR_ERR(vcpu);
1515
1516         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1517
1518         r = kvm_arch_vcpu_setup(vcpu);
1519         if (r)
1520                 return r;
1521
1522         mutex_lock(&kvm->lock);
1523         if (kvm->vcpus[n]) {
1524                 r = -EEXIST;
1525                 goto vcpu_destroy;
1526         }
1527         kvm->vcpus[n] = vcpu;
1528         mutex_unlock(&kvm->lock);
1529
1530         /* Now it's all set up, let userspace reach it */
1531         kvm_get_kvm(kvm);
1532         r = create_vcpu_fd(vcpu);
1533         if (r < 0)
1534                 goto unlink;
1535         return r;
1536
1537 unlink:
1538         mutex_lock(&kvm->lock);
1539         kvm->vcpus[n] = NULL;
1540 vcpu_destroy:
1541         mutex_unlock(&kvm->lock);
1542         kvm_arch_vcpu_destroy(vcpu);
1543         return r;
1544 }
1545
1546 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1547 {
1548         if (sigset) {
1549                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1550                 vcpu->sigset_active = 1;
1551                 vcpu->sigset = *sigset;
1552         } else
1553                 vcpu->sigset_active = 0;
1554         return 0;
1555 }
1556
1557 static long kvm_vcpu_ioctl(struct file *filp,
1558                            unsigned int ioctl, unsigned long arg)
1559 {
1560         struct kvm_vcpu *vcpu = filp->private_data;
1561         void __user *argp = (void __user *)arg;
1562         int r;
1563         struct kvm_fpu *fpu = NULL;
1564         struct kvm_sregs *kvm_sregs = NULL;
1565
1566         if (vcpu->kvm->mm != current->mm)
1567                 return -EIO;
1568         switch (ioctl) {
1569         case KVM_RUN:
1570                 r = -EINVAL;
1571                 if (arg)
1572                         goto out;
1573                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1574                 break;
1575         case KVM_GET_REGS: {
1576                 struct kvm_regs *kvm_regs;
1577
1578                 r = -ENOMEM;
1579                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1580                 if (!kvm_regs)
1581                         goto out;
1582                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1583                 if (r)
1584                         goto out_free1;
1585                 r = -EFAULT;
1586                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1587                         goto out_free1;
1588                 r = 0;
1589 out_free1:
1590                 kfree(kvm_regs);
1591                 break;
1592         }
1593         case KVM_SET_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 = -EFAULT;
1601                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1602                         goto out_free2;
1603                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1604                 if (r)
1605                         goto out_free2;
1606                 r = 0;
1607 out_free2:
1608                 kfree(kvm_regs);
1609                 break;
1610         }
1611         case KVM_GET_SREGS: {
1612                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1613                 r = -ENOMEM;
1614                 if (!kvm_sregs)
1615                         goto out;
1616                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1617                 if (r)
1618                         goto out;
1619                 r = -EFAULT;
1620                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1621                         goto out;
1622                 r = 0;
1623                 break;
1624         }
1625         case KVM_SET_SREGS: {
1626                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1627                 r = -ENOMEM;
1628                 if (!kvm_sregs)
1629                         goto out;
1630                 r = -EFAULT;
1631                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1632                         goto out;
1633                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1634                 if (r)
1635                         goto out;
1636                 r = 0;
1637                 break;
1638         }
1639         case KVM_GET_MP_STATE: {
1640                 struct kvm_mp_state mp_state;
1641
1642                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1643                 if (r)
1644                         goto out;
1645                 r = -EFAULT;
1646                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1647                         goto out;
1648                 r = 0;
1649                 break;
1650         }
1651         case KVM_SET_MP_STATE: {
1652                 struct kvm_mp_state mp_state;
1653
1654                 r = -EFAULT;
1655                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1656                         goto out;
1657                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1658                 if (r)
1659                         goto out;
1660                 r = 0;
1661                 break;
1662         }
1663         case KVM_TRANSLATE: {
1664                 struct kvm_translation tr;
1665
1666                 r = -EFAULT;
1667                 if (copy_from_user(&tr, argp, sizeof tr))
1668                         goto out;
1669                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1670                 if (r)
1671                         goto out;
1672                 r = -EFAULT;
1673                 if (copy_to_user(argp, &tr, sizeof tr))
1674                         goto out;
1675                 r = 0;
1676                 break;
1677         }
1678         case KVM_DEBUG_GUEST: {
1679                 struct kvm_debug_guest dbg;
1680
1681                 r = -EFAULT;
1682                 if (copy_from_user(&dbg, argp, sizeof dbg))
1683                         goto out;
1684                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1685                 if (r)
1686                         goto out;
1687                 r = 0;
1688                 break;
1689         }
1690         case KVM_SET_SIGNAL_MASK: {
1691                 struct kvm_signal_mask __user *sigmask_arg = argp;
1692                 struct kvm_signal_mask kvm_sigmask;
1693                 sigset_t sigset, *p;
1694
1695                 p = NULL;
1696                 if (argp) {
1697                         r = -EFAULT;
1698                         if (copy_from_user(&kvm_sigmask, argp,
1699                                            sizeof kvm_sigmask))
1700                                 goto out;
1701                         r = -EINVAL;
1702                         if (kvm_sigmask.len != sizeof sigset)
1703                                 goto out;
1704                         r = -EFAULT;
1705                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1706                                            sizeof sigset))
1707                                 goto out;
1708                         p = &sigset;
1709                 }
1710                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1711                 break;
1712         }
1713         case KVM_GET_FPU: {
1714                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1715                 r = -ENOMEM;
1716                 if (!fpu)
1717                         goto out;
1718                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1719                 if (r)
1720                         goto out;
1721                 r = -EFAULT;
1722                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1723                         goto out;
1724                 r = 0;
1725                 break;
1726         }
1727         case KVM_SET_FPU: {
1728                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1729                 r = -ENOMEM;
1730                 if (!fpu)
1731                         goto out;
1732                 r = -EFAULT;
1733                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1734                         goto out;
1735                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1736                 if (r)
1737                         goto out;
1738                 r = 0;
1739                 break;
1740         }
1741         default:
1742                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1743         }
1744 out:
1745         kfree(fpu);
1746         kfree(kvm_sregs);
1747         return r;
1748 }
1749
1750 static long kvm_vm_ioctl(struct file *filp,
1751                            unsigned int ioctl, unsigned long arg)
1752 {
1753         struct kvm *kvm = filp->private_data;
1754         void __user *argp = (void __user *)arg;
1755         int r;
1756
1757         if (kvm->mm != current->mm)
1758                 return -EIO;
1759         switch (ioctl) {
1760         case KVM_CREATE_VCPU:
1761                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1762                 if (r < 0)
1763                         goto out;
1764                 break;
1765         case KVM_SET_USER_MEMORY_REGION: {
1766                 struct kvm_userspace_memory_region kvm_userspace_mem;
1767
1768                 r = -EFAULT;
1769                 if (copy_from_user(&kvm_userspace_mem, argp,
1770                                                 sizeof kvm_userspace_mem))
1771                         goto out;
1772
1773                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1774                 if (r)
1775                         goto out;
1776                 break;
1777         }
1778         case KVM_GET_DIRTY_LOG: {
1779                 struct kvm_dirty_log log;
1780
1781                 r = -EFAULT;
1782                 if (copy_from_user(&log, argp, sizeof log))
1783                         goto out;
1784                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1785                 if (r)
1786                         goto out;
1787                 break;
1788         }
1789 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1790         case KVM_REGISTER_COALESCED_MMIO: {
1791                 struct kvm_coalesced_mmio_zone zone;
1792                 r = -EFAULT;
1793                 if (copy_from_user(&zone, argp, sizeof zone))
1794                         goto out;
1795                 r = -ENXIO;
1796                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1797                 if (r)
1798                         goto out;
1799                 r = 0;
1800                 break;
1801         }
1802         case KVM_UNREGISTER_COALESCED_MMIO: {
1803                 struct kvm_coalesced_mmio_zone zone;
1804                 r = -EFAULT;
1805                 if (copy_from_user(&zone, argp, sizeof zone))
1806                         goto out;
1807                 r = -ENXIO;
1808                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1809                 if (r)
1810                         goto out;
1811                 r = 0;
1812                 break;
1813         }
1814 #endif
1815 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1816         case KVM_ASSIGN_PCI_DEVICE: {
1817                 struct kvm_assigned_pci_dev assigned_dev;
1818
1819                 r = -EFAULT;
1820                 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1821                         goto out;
1822                 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1823                 if (r)
1824                         goto out;
1825                 break;
1826         }
1827         case KVM_ASSIGN_IRQ: {
1828                 struct kvm_assigned_irq assigned_irq;
1829
1830                 r = -EFAULT;
1831                 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1832                         goto out;
1833                 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1834                 if (r)
1835                         goto out;
1836                 break;
1837         }
1838 #endif
1839         default:
1840                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1841         }
1842 out:
1843         return r;
1844 }
1845
1846 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1847 {
1848         struct page *page[1];
1849         unsigned long addr;
1850         int npages;
1851         gfn_t gfn = vmf->pgoff;
1852         struct kvm *kvm = vma->vm_file->private_data;
1853
1854         addr = gfn_to_hva(kvm, gfn);
1855         if (kvm_is_error_hva(addr))
1856                 return VM_FAULT_SIGBUS;
1857
1858         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1859                                 NULL);
1860         if (unlikely(npages != 1))
1861                 return VM_FAULT_SIGBUS;
1862
1863         vmf->page = page[0];
1864         return 0;
1865 }
1866
1867 static struct vm_operations_struct kvm_vm_vm_ops = {
1868         .fault = kvm_vm_fault,
1869 };
1870
1871 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1872 {
1873         vma->vm_ops = &kvm_vm_vm_ops;
1874         return 0;
1875 }
1876
1877 static const struct file_operations kvm_vm_fops = {
1878         .release        = kvm_vm_release,
1879         .unlocked_ioctl = kvm_vm_ioctl,
1880         .compat_ioctl   = kvm_vm_ioctl,
1881         .mmap           = kvm_vm_mmap,
1882 };
1883
1884 static int kvm_dev_ioctl_create_vm(void)
1885 {
1886         int fd;
1887         struct kvm *kvm;
1888
1889         kvm = kvm_create_vm();
1890         if (IS_ERR(kvm))
1891                 return PTR_ERR(kvm);
1892         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1893         if (fd < 0)
1894                 kvm_put_kvm(kvm);
1895
1896         return fd;
1897 }
1898
1899 static long kvm_dev_ioctl(struct file *filp,
1900                           unsigned int ioctl, unsigned long arg)
1901 {
1902         long r = -EINVAL;
1903
1904         switch (ioctl) {
1905         case KVM_GET_API_VERSION:
1906                 r = -EINVAL;
1907                 if (arg)
1908                         goto out;
1909                 r = KVM_API_VERSION;
1910                 break;
1911         case KVM_CREATE_VM:
1912                 r = -EINVAL;
1913                 if (arg)
1914                         goto out;
1915                 r = kvm_dev_ioctl_create_vm();
1916                 break;
1917         case KVM_CHECK_EXTENSION:
1918                 r = kvm_dev_ioctl_check_extension(arg);
1919                 break;
1920         case KVM_GET_VCPU_MMAP_SIZE:
1921                 r = -EINVAL;
1922                 if (arg)
1923                         goto out;
1924                 r = PAGE_SIZE;     /* struct kvm_run */
1925 #ifdef CONFIG_X86
1926                 r += PAGE_SIZE;    /* pio data page */
1927 #endif
1928 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1929                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1930 #endif
1931                 break;
1932         case KVM_TRACE_ENABLE:
1933         case KVM_TRACE_PAUSE:
1934         case KVM_TRACE_DISABLE:
1935                 r = kvm_trace_ioctl(ioctl, arg);
1936                 break;
1937         default:
1938                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1939         }
1940 out:
1941         return r;
1942 }
1943
1944 static struct file_operations kvm_chardev_ops = {
1945         .unlocked_ioctl = kvm_dev_ioctl,
1946         .compat_ioctl   = kvm_dev_ioctl,
1947 };
1948
1949 static struct miscdevice kvm_dev = {
1950         KVM_MINOR,
1951         "kvm",
1952         &kvm_chardev_ops,
1953 };
1954
1955 static void hardware_enable(void *junk)
1956 {
1957         int cpu = raw_smp_processor_id();
1958
1959         if (cpu_isset(cpu, cpus_hardware_enabled))
1960                 return;
1961         cpu_set(cpu, cpus_hardware_enabled);
1962         kvm_arch_hardware_enable(NULL);
1963 }
1964
1965 static void hardware_disable(void *junk)
1966 {
1967         int cpu = raw_smp_processor_id();
1968
1969         if (!cpu_isset(cpu, cpus_hardware_enabled))
1970                 return;
1971         cpu_clear(cpu, cpus_hardware_enabled);
1972         kvm_arch_hardware_disable(NULL);
1973 }
1974
1975 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1976                            void *v)
1977 {
1978         int cpu = (long)v;
1979
1980         val &= ~CPU_TASKS_FROZEN;
1981         switch (val) {
1982         case CPU_DYING:
1983                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1984                        cpu);
1985                 hardware_disable(NULL);
1986                 break;
1987         case CPU_UP_CANCELED:
1988                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1989                        cpu);
1990                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1991                 break;
1992         case CPU_ONLINE:
1993                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1994                        cpu);
1995                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1996                 break;
1997         }
1998         return NOTIFY_OK;
1999 }
2000
2001
2002 asmlinkage void kvm_handle_fault_on_reboot(void)
2003 {
2004         if (kvm_rebooting)
2005                 /* spin while reset goes on */
2006                 while (true)
2007                         ;
2008         /* Fault while not rebooting.  We want the trace. */
2009         BUG();
2010 }
2011 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2012
2013 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2014                       void *v)
2015 {
2016         if (val == SYS_RESTART) {
2017                 /*
2018                  * Some (well, at least mine) BIOSes hang on reboot if
2019                  * in vmx root mode.
2020                  */
2021                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2022                 kvm_rebooting = true;
2023                 on_each_cpu(hardware_disable, NULL, 1);
2024         }
2025         return NOTIFY_OK;
2026 }
2027
2028 static struct notifier_block kvm_reboot_notifier = {
2029         .notifier_call = kvm_reboot,
2030         .priority = 0,
2031 };
2032
2033 void kvm_io_bus_init(struct kvm_io_bus *bus)
2034 {
2035         memset(bus, 0, sizeof(*bus));
2036 }
2037
2038 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2039 {
2040         int i;
2041
2042         for (i = 0; i < bus->dev_count; i++) {
2043                 struct kvm_io_device *pos = bus->devs[i];
2044
2045                 kvm_iodevice_destructor(pos);
2046         }
2047 }
2048
2049 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
2050                                           gpa_t addr, int len, int is_write)
2051 {
2052         int i;
2053
2054         for (i = 0; i < bus->dev_count; i++) {
2055                 struct kvm_io_device *pos = bus->devs[i];
2056
2057                 if (pos->in_range(pos, addr, len, is_write))
2058                         return pos;
2059         }
2060
2061         return NULL;
2062 }
2063
2064 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2065 {
2066         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2067
2068         bus->devs[bus->dev_count++] = dev;
2069 }
2070
2071 static struct notifier_block kvm_cpu_notifier = {
2072         .notifier_call = kvm_cpu_hotplug,
2073         .priority = 20, /* must be > scheduler priority */
2074 };
2075
2076 static int vm_stat_get(void *_offset, u64 *val)
2077 {
2078         unsigned offset = (long)_offset;
2079         struct kvm *kvm;
2080
2081         *val = 0;
2082         spin_lock(&kvm_lock);
2083         list_for_each_entry(kvm, &vm_list, vm_list)
2084                 *val += *(u32 *)((void *)kvm + offset);
2085         spin_unlock(&kvm_lock);
2086         return 0;
2087 }
2088
2089 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2090
2091 static int vcpu_stat_get(void *_offset, u64 *val)
2092 {
2093         unsigned offset = (long)_offset;
2094         struct kvm *kvm;
2095         struct kvm_vcpu *vcpu;
2096         int i;
2097
2098         *val = 0;
2099         spin_lock(&kvm_lock);
2100         list_for_each_entry(kvm, &vm_list, vm_list)
2101                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2102                         vcpu = kvm->vcpus[i];
2103                         if (vcpu)
2104                                 *val += *(u32 *)((void *)vcpu + offset);
2105                 }
2106         spin_unlock(&kvm_lock);
2107         return 0;
2108 }
2109
2110 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2111
2112 static struct file_operations *stat_fops[] = {
2113         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2114         [KVM_STAT_VM]   = &vm_stat_fops,
2115 };
2116
2117 static void kvm_init_debug(void)
2118 {
2119         struct kvm_stats_debugfs_item *p;
2120
2121         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2122         for (p = debugfs_entries; p->name; ++p)
2123                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2124                                                 (void *)(long)p->offset,
2125                                                 stat_fops[p->kind]);
2126 }
2127
2128 static void kvm_exit_debug(void)
2129 {
2130         struct kvm_stats_debugfs_item *p;
2131
2132         for (p = debugfs_entries; p->name; ++p)
2133                 debugfs_remove(p->dentry);
2134         debugfs_remove(kvm_debugfs_dir);
2135 }
2136
2137 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2138 {
2139         hardware_disable(NULL);
2140         return 0;
2141 }
2142
2143 static int kvm_resume(struct sys_device *dev)
2144 {
2145         hardware_enable(NULL);
2146         return 0;
2147 }
2148
2149 static struct sysdev_class kvm_sysdev_class = {
2150         .name = "kvm",
2151         .suspend = kvm_suspend,
2152         .resume = kvm_resume,
2153 };
2154
2155 static struct sys_device kvm_sysdev = {
2156         .id = 0,
2157         .cls = &kvm_sysdev_class,
2158 };
2159
2160 struct page *bad_page;
2161 pfn_t bad_pfn;
2162
2163 static inline
2164 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2165 {
2166         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2167 }
2168
2169 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2170 {
2171         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2172
2173         kvm_arch_vcpu_load(vcpu, cpu);
2174 }
2175
2176 static void kvm_sched_out(struct preempt_notifier *pn,
2177                           struct task_struct *next)
2178 {
2179         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2180
2181         kvm_arch_vcpu_put(vcpu);
2182 }
2183
2184 int kvm_init(void *opaque, unsigned int vcpu_size,
2185                   struct module *module)
2186 {
2187         int r;
2188         int cpu;
2189
2190         kvm_init_debug();
2191
2192         r = kvm_arch_init(opaque);
2193         if (r)
2194                 goto out_fail;
2195
2196         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2197
2198         if (bad_page == NULL) {
2199                 r = -ENOMEM;
2200                 goto out;
2201         }
2202
2203         bad_pfn = page_to_pfn(bad_page);
2204
2205         r = kvm_arch_hardware_setup();
2206         if (r < 0)
2207                 goto out_free_0;
2208
2209         for_each_online_cpu(cpu) {
2210                 smp_call_function_single(cpu,
2211                                 kvm_arch_check_processor_compat,
2212                                 &r, 1);
2213                 if (r < 0)
2214                         goto out_free_1;
2215         }
2216
2217         on_each_cpu(hardware_enable, NULL, 1);
2218         r = register_cpu_notifier(&kvm_cpu_notifier);
2219         if (r)
2220                 goto out_free_2;
2221         register_reboot_notifier(&kvm_reboot_notifier);
2222
2223         r = sysdev_class_register(&kvm_sysdev_class);
2224         if (r)
2225                 goto out_free_3;
2226
2227         r = sysdev_register(&kvm_sysdev);
2228         if (r)
2229                 goto out_free_4;
2230
2231         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2232         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2233                                            __alignof__(struct kvm_vcpu),
2234                                            0, NULL);
2235         if (!kvm_vcpu_cache) {
2236                 r = -ENOMEM;
2237                 goto out_free_5;
2238         }
2239
2240         kvm_chardev_ops.owner = module;
2241
2242         r = misc_register(&kvm_dev);
2243         if (r) {
2244                 printk(KERN_ERR "kvm: misc device register failed\n");
2245                 goto out_free;
2246         }
2247
2248         kvm_preempt_ops.sched_in = kvm_sched_in;
2249         kvm_preempt_ops.sched_out = kvm_sched_out;
2250 #ifndef CONFIG_X86
2251         msi2intx = 0;
2252 #endif
2253
2254         return 0;
2255
2256 out_free:
2257         kmem_cache_destroy(kvm_vcpu_cache);
2258 out_free_5:
2259         sysdev_unregister(&kvm_sysdev);
2260 out_free_4:
2261         sysdev_class_unregister(&kvm_sysdev_class);
2262 out_free_3:
2263         unregister_reboot_notifier(&kvm_reboot_notifier);
2264         unregister_cpu_notifier(&kvm_cpu_notifier);
2265 out_free_2:
2266         on_each_cpu(hardware_disable, NULL, 1);
2267 out_free_1:
2268         kvm_arch_hardware_unsetup();
2269 out_free_0:
2270         __free_page(bad_page);
2271 out:
2272         kvm_arch_exit();
2273         kvm_exit_debug();
2274 out_fail:
2275         return r;
2276 }
2277 EXPORT_SYMBOL_GPL(kvm_init);
2278
2279 void kvm_exit(void)
2280 {
2281         kvm_trace_cleanup();
2282         misc_deregister(&kvm_dev);
2283         kmem_cache_destroy(kvm_vcpu_cache);
2284         sysdev_unregister(&kvm_sysdev);
2285         sysdev_class_unregister(&kvm_sysdev_class);
2286         unregister_reboot_notifier(&kvm_reboot_notifier);
2287         unregister_cpu_notifier(&kvm_cpu_notifier);
2288         on_each_cpu(hardware_disable, NULL, 1);
2289         kvm_arch_hardware_unsetup();
2290         kvm_arch_exit();
2291         kvm_exit_debug();
2292         __free_page(bad_page);
2293 }
2294 EXPORT_SYMBOL_GPL(kvm_exit);