SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/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 "kvm.h"
  19 #include "x86.h"
  20 #include "irq.h"
  21
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/gfp.h>
  27 #include <linux/mm.h>
  28 #include <linux/miscdevice.h>
  29 #include <linux/vmalloc.h>
  30 #include <linux/reboot.h>
  31 #include <linux/debugfs.h>
  32 #include <linux/highmem.h>
  33 #include <linux/file.h>
  34 #include <linux/sysdev.h>
  35 #include <linux/cpu.h>
  36 #include <linux/sched.h>
  37 #include <linux/cpumask.h>
  38 #include <linux/smp.h>
  39 #include <linux/anon_inodes.h>
  40 #include <linux/profile.h>
  41 #include <linux/kvm_para.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mman.h>
  44
  45 #include <asm/processor.h>
  46 #include <asm/io.h>
  47 #include <asm/uaccess.h>
  48 #include <asm/desc.h>
  49
  50 MODULE_AUTHOR("Qumranet");
  51 MODULE_LICENSE("GPL");
  52
  53 DEFINE_SPINLOCK(kvm_lock);
  54 LIST_HEAD(vm_list);
  55
  56 static cpumask_t cpus_hardware_enabled;
  57
  58 struct kmem_cache *kvm_vcpu_cache;
  59 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  60
  61 static __read_mostly struct preempt_ops kvm_preempt_ops;
  62
  63 static struct dentry *debugfs_dir;
  64
  65 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  66                            unsigned long arg);
  67
  68 static inline int valid_vcpu(int n)
  69 {
  70         return likely(n >= 0 && n < KVM_MAX_VCPUS);
  71 }
  72
  73 /*
  74  * Switches to specified vcpu, until a matching vcpu_put()
  75  */
  76 void vcpu_load(struct kvm_vcpu *vcpu)
  77 {
  78         int cpu;
  79
  80         mutex_lock(&vcpu->mutex);
  81         cpu = get_cpu();
  82         preempt_notifier_register(&vcpu->preempt_notifier);
  83         kvm_arch_vcpu_load(vcpu, cpu);
  84         put_cpu();
  85 }
  86
  87 void vcpu_put(struct kvm_vcpu *vcpu)
  88 {
  89         preempt_disable();
  90         kvm_arch_vcpu_put(vcpu);
  91         preempt_notifier_unregister(&vcpu->preempt_notifier);
  92         preempt_enable();
  93         mutex_unlock(&vcpu->mutex);
  94 }
  95
  96 static void ack_flush(void *_completed)
  97 {
  98 }
  99
 100 void kvm_flush_remote_tlbs(struct kvm *kvm)
 101 {
 102         int i, cpu;
 103         cpumask_t cpus;
 104         struct kvm_vcpu *vcpu;
 105
 106         cpus_clear(cpus);
 107         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 108                 vcpu = kvm->vcpus[i];
 109                 if (!vcpu)
 110                         continue;
 111                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 112                         continue;
 113                 cpu = vcpu->cpu;
 114                 if (cpu != -1 && cpu != raw_smp_processor_id())
 115                         cpu_set(cpu, cpus);
 116         }
 117         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 118 }
 119
 120 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 121 {
 122         struct page *page;
 123         int r;
 124
 125         mutex_init(&vcpu->mutex);
 126         vcpu->cpu = -1;
 127         vcpu->kvm = kvm;
 128         vcpu->vcpu_id = id;
 129         init_waitqueue_head(&vcpu->wq);
 130
 131         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 132         if (!page) {
 133                 r = -ENOMEM;
 134                 goto fail;
 135         }
 136         vcpu->run = page_address(page);
 137
 138         r = kvm_arch_vcpu_init(vcpu);
 139         if (r < 0)
 140                 goto fail_free_run;
 141         return 0;
 142
 143 fail_free_run:
 144         free_page((unsigned long)vcpu->run);
 145 fail:
 146         return r;
 147 }
 148 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 149
 150 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 151 {
 152         kvm_arch_vcpu_uninit(vcpu);
 153         free_page((unsigned long)vcpu->run);
 154 }
 155 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 156
 157 static struct kvm *kvm_create_vm(void)
 158 {
 159         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 160
 161         if (!kvm)
 162                 return ERR_PTR(-ENOMEM);
 163
 164         kvm_io_bus_init(&kvm->pio_bus);
 165         mutex_init(&kvm->lock);
 166         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 167         kvm_io_bus_init(&kvm->mmio_bus);
 168         spin_lock(&kvm_lock);
 169         list_add(&kvm->vm_list, &vm_list);
 170         spin_unlock(&kvm_lock);
 171         return kvm;
 172 }
 173
 174 /*
 175  * Free any memory in @free but not in @dont.
 176  */
 177 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 178                                   struct kvm_memory_slot *dont)
 179 {
 180         if (!dont || free->rmap != dont->rmap)
 181                 vfree(free->rmap);
 182
 183         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 184                 vfree(free->dirty_bitmap);
 185
 186         free->npages = 0;
 187         free->dirty_bitmap = NULL;
 188         free->rmap = NULL;
 189 }
 190
 191 static void kvm_free_physmem(struct kvm *kvm)
 192 {
 193         int i;
 194
 195         for (i = 0; i < kvm->nmemslots; ++i)
 196                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 197 }
 198
 199 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 200 {
 201         vcpu_load(vcpu);
 202         kvm_mmu_unload(vcpu);
 203         vcpu_put(vcpu);
 204 }
 205
 206 static void kvm_free_vcpus(struct kvm *kvm)
 207 {
 208         unsigned int i;
 209
 210         /*
 211          * Unpin any mmu pages first.
 212          */
 213         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 214                 if (kvm->vcpus[i])
 215                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
 216         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 217                 if (kvm->vcpus[i]) {
 218                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 219                         kvm->vcpus[i] = NULL;
 220                 }
 221         }
 222
 223 }
 224
 225 static void kvm_destroy_vm(struct kvm *kvm)
 226 {
 227         spin_lock(&kvm_lock);
 228         list_del(&kvm->vm_list);
 229         spin_unlock(&kvm_lock);
 230         kvm_io_bus_destroy(&kvm->pio_bus);
 231         kvm_io_bus_destroy(&kvm->mmio_bus);
 232         kfree(kvm->vpic);
 233         kfree(kvm->vioapic);
 234         kvm_free_vcpus(kvm);
 235         kvm_free_physmem(kvm);
 236         kfree(kvm);
 237 }
 238
 239 static int kvm_vm_release(struct inode *inode, struct file *filp)
 240 {
 241         struct kvm *kvm = filp->private_data;
 242
 243         kvm_destroy_vm(kvm);
 244         return 0;
 245 }
 246
 247 /*
 248  * Allocate some memory and give it an address in the guest physical address
 249  * space.
 250  *
 251  * Discontiguous memory is allowed, mostly for framebuffers.
 252  *
 253  * Must be called holding kvm->lock.
 254  */
 255 int __kvm_set_memory_region(struct kvm *kvm,
 256                             struct kvm_userspace_memory_region *mem,
 257                             int user_alloc)
 258 {
 259         int r;
 260         gfn_t base_gfn;
 261         unsigned long npages;
 262         unsigned long i;
 263         struct kvm_memory_slot *memslot;
 264         struct kvm_memory_slot old, new;
 265
 266         r = -EINVAL;
 267         /* General sanity checks */
 268         if (mem->memory_size & (PAGE_SIZE - 1))
 269                 goto out;
 270         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 271                 goto out;
 272         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 273                 goto out;
 274         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 275                 goto out;
 276
 277         memslot = &kvm->memslots[mem->slot];
 278         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 279         npages = mem->memory_size >> PAGE_SHIFT;
 280
 281         if (!npages)
 282                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 283
 284         new = old = *memslot;
 285
 286         new.base_gfn = base_gfn;
 287         new.npages = npages;
 288         new.flags = mem->flags;
 289
 290         /* Disallow changing a memory slot's size. */
 291         r = -EINVAL;
 292         if (npages && old.npages && npages != old.npages)
 293                 goto out_free;
 294
 295         /* Check for overlaps */
 296         r = -EEXIST;
 297         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 298                 struct kvm_memory_slot *s = &kvm->memslots[i];
 299
 300                 if (s == memslot)
 301                         continue;
 302                 if (!((base_gfn + npages <= s->base_gfn) ||
 303                       (base_gfn >= s->base_gfn + s->npages)))
 304                         goto out_free;
 305         }
 306
 307         /* Free page dirty bitmap if unneeded */
 308         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 309                 new.dirty_bitmap = NULL;
 310
 311         r = -ENOMEM;
 312
 313         /* Allocate if a slot is being created */
 314         if (npages && !new.rmap) {
 315                 new.rmap = vmalloc(npages * sizeof(struct page *));
 316
 317                 if (!new.rmap)
 318                         goto out_free;
 319
 320                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 321
 322                 new.user_alloc = user_alloc;
 323                 if (user_alloc)
 324                         new.userspace_addr = mem->userspace_addr;
 325                 else {
 326                         down_write(&current->mm->mmap_sem);
 327                         new.userspace_addr = do_mmap(NULL, 0,
 328                                                      npages * PAGE_SIZE,
 329                                                      PROT_READ | PROT_WRITE,
 330                                                      MAP_SHARED | MAP_ANONYMOUS,
 331                                                      0);
 332                         up_write(&current->mm->mmap_sem);
 333
 334                         if (IS_ERR((void *)new.userspace_addr))
 335                                 goto out_free;
 336                 }
 337         } else {
 338                 if (!old.user_alloc && old.rmap) {
 339                         int ret;
 340
 341                         down_write(&current->mm->mmap_sem);
 342                         ret = do_munmap(current->mm, old.userspace_addr,
 343                                         old.npages * PAGE_SIZE);
 344                         up_write(&current->mm->mmap_sem);
 345                         if (ret < 0)
 346                                 printk(KERN_WARNING
 347                                        "kvm_vm_ioctl_set_memory_region: "
 348                                        "failed to munmap memory\n");
 349                 }
 350         }
 351
 352         /* Allocate page dirty bitmap if needed */
 353         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 354                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 355
 356                 new.dirty_bitmap = vmalloc(dirty_bytes);
 357                 if (!new.dirty_bitmap)
 358                         goto out_free;
 359                 memset(new.dirty_bitmap, 0, dirty_bytes);
 360         }
 361
 362         if (mem->slot >= kvm->nmemslots)
 363                 kvm->nmemslots = mem->slot + 1;
 364
 365         if (!kvm->n_requested_mmu_pages) {
 366                 unsigned int n_pages;
 367
 368                 if (npages) {
 369                         n_pages = npages * KVM_PERMILLE_MMU_PAGES / 1000;
 370                         kvm_mmu_change_mmu_pages(kvm, kvm->n_alloc_mmu_pages +
 371                                                  n_pages);
 372                 } else {
 373                         unsigned int nr_mmu_pages;
 374
 375                         n_pages = old.npages * KVM_PERMILLE_MMU_PAGES / 1000;
 376                         nr_mmu_pages = kvm->n_alloc_mmu_pages - n_pages;
 377                         nr_mmu_pages = max(nr_mmu_pages,
 378                                         (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
 379                         kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
 380                 }
 381         }
 382
 383         *memslot = new;
 384
 385         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 386         kvm_flush_remote_tlbs(kvm);
 387
 388         kvm_free_physmem_slot(&old, &new);
 389         return 0;
 390
 391 out_free:
 392         kvm_free_physmem_slot(&new, &old);
 393 out:
 394         return r;
 395
 396 }
 397 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 398
 399 int kvm_set_memory_region(struct kvm *kvm,
 400                           struct kvm_userspace_memory_region *mem,
 401                           int user_alloc)
 402 {
 403         int r;
 404
 405         mutex_lock(&kvm->lock);
 406         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 407         mutex_unlock(&kvm->lock);
 408         return r;
 409 }
 410 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 411
 412 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 413                                    struct
 414                                    kvm_userspace_memory_region *mem,
 415                                    int user_alloc)
 416 {
 417         if (mem->slot >= KVM_MEMORY_SLOTS)
 418                 return -EINVAL;
 419         return kvm_set_memory_region(kvm, mem, user_alloc);
 420 }
 421
 422 /*
 423  * Get (and clear) the dirty memory log for a memory slot.
 424  */
 425 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 426                                       struct kvm_dirty_log *log)
 427 {
 428         struct kvm_memory_slot *memslot;
 429         int r, i;
 430         int n;
 431         unsigned long any = 0;
 432
 433         mutex_lock(&kvm->lock);
 434
 435         r = -EINVAL;
 436         if (log->slot >= KVM_MEMORY_SLOTS)
 437                 goto out;
 438
 439         memslot = &kvm->memslots[log->slot];
 440         r = -ENOENT;
 441         if (!memslot->dirty_bitmap)
 442                 goto out;
 443
 444         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 445
 446         for (i = 0; !any && i < n/sizeof(long); ++i)
 447                 any = memslot->dirty_bitmap[i];
 448
 449         r = -EFAULT;
 450         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 451                 goto out;
 452
 453         /* If nothing is dirty, don't bother messing with page tables. */
 454         if (any) {
 455                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 456                 kvm_flush_remote_tlbs(kvm);
 457                 memset(memslot->dirty_bitmap, 0, n);
 458         }
 459
 460         r = 0;
 461
 462 out:
 463         mutex_unlock(&kvm->lock);
 464         return r;
 465 }
 466
 467 int is_error_page(struct page *page)
 468 {
 469         return page == bad_page;
 470 }
 471 EXPORT_SYMBOL_GPL(is_error_page);
 472
 473 static inline unsigned long bad_hva(void)
 474 {
 475         return PAGE_OFFSET;
 476 }
 477
 478 int kvm_is_error_hva(unsigned long addr)
 479 {
 480         return addr == bad_hva();
 481 }
 482 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 483
 484 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 485 {
 486         int i;
 487         struct kvm_mem_alias *alias;
 488
 489         for (i = 0; i < kvm->naliases; ++i) {
 490                 alias = &kvm->aliases[i];
 491                 if (gfn >= alias->base_gfn
 492                     && gfn < alias->base_gfn + alias->npages)
 493                         return alias->target_gfn + gfn - alias->base_gfn;
 494         }
 495         return gfn;
 496 }
 497
 498 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 499 {
 500         int i;
 501
 502         for (i = 0; i < kvm->nmemslots; ++i) {
 503                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 504
 505                 if (gfn >= memslot->base_gfn
 506                     && gfn < memslot->base_gfn + memslot->npages)
 507                         return memslot;
 508         }
 509         return NULL;
 510 }
 511
 512 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 513 {
 514         gfn = unalias_gfn(kvm, gfn);
 515         return __gfn_to_memslot(kvm, gfn);
 516 }
 517
 518 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 519 {
 520         int i;
 521
 522         gfn = unalias_gfn(kvm, gfn);
 523         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 524                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 525
 526                 if (gfn >= memslot->base_gfn
 527                     && gfn < memslot->base_gfn + memslot->npages)
 528                         return 1;
 529         }
 530         return 0;
 531 }
 532 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 533
 534 static unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 535 {
 536         struct kvm_memory_slot *slot;
 537
 538         gfn = unalias_gfn(kvm, gfn);
 539         slot = __gfn_to_memslot(kvm, gfn);
 540         if (!slot)
 541                 return bad_hva();
 542         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 543 }
 544
 545 /*
 546  * Requires current->mm->mmap_sem to be held
 547  */
 548 static struct page *__gfn_to_page(struct kvm *kvm, gfn_t gfn)
 549 {
 550         struct page *page[1];
 551         unsigned long addr;
 552         int npages;
 553
 554         might_sleep();
 555
 556         addr = gfn_to_hva(kvm, gfn);
 557         if (kvm_is_error_hva(addr)) {
 558                 get_page(bad_page);
 559                 return bad_page;
 560         }
 561
 562         npages = get_user_pages(current, current->mm, addr, 1, 1, 1, page,
 563                                 NULL);
 564
 565         if (npages != 1) {
 566                 get_page(bad_page);
 567                 return bad_page;
 568         }
 569
 570         return page[0];
 571 }
 572
 573 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 574 {
 575         struct page *page;
 576
 577         down_read(&current->mm->mmap_sem);
 578         page = __gfn_to_page(kvm, gfn);
 579         up_read(&current->mm->mmap_sem);
 580
 581         return page;
 582 }
 583
 584 EXPORT_SYMBOL_GPL(gfn_to_page);
 585
 586 void kvm_release_page(struct page *page)
 587 {
 588         if (!PageReserved(page))
 589                 SetPageDirty(page);
 590         put_page(page);
 591 }
 592 EXPORT_SYMBOL_GPL(kvm_release_page);
 593
 594 static int next_segment(unsigned long len, int offset)
 595 {
 596         if (len > PAGE_SIZE - offset)
 597                 return PAGE_SIZE - offset;
 598         else
 599                 return len;
 600 }
 601
 602 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 603                         int len)
 604 {
 605         int r;
 606         unsigned long addr;
 607
 608         addr = gfn_to_hva(kvm, gfn);
 609         if (kvm_is_error_hva(addr))
 610                 return -EFAULT;
 611         r = copy_from_user(data, (void __user *)addr + offset, len);
 612         if (r)
 613                 return -EFAULT;
 614         return 0;
 615 }
 616 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 617
 618 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 619 {
 620         gfn_t gfn = gpa >> PAGE_SHIFT;
 621         int seg;
 622         int offset = offset_in_page(gpa);
 623         int ret;
 624
 625         while ((seg = next_segment(len, offset)) != 0) {
 626                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 627                 if (ret < 0)
 628                         return ret;
 629                 offset = 0;
 630                 len -= seg;
 631                 data += seg;
 632                 ++gfn;
 633         }
 634         return 0;
 635 }
 636 EXPORT_SYMBOL_GPL(kvm_read_guest);
 637
 638 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
 639                          int offset, int len)
 640 {
 641         int r;
 642         unsigned long addr;
 643
 644         addr = gfn_to_hva(kvm, gfn);
 645         if (kvm_is_error_hva(addr))
 646                 return -EFAULT;
 647         r = copy_to_user((void __user *)addr + offset, data, len);
 648         if (r)
 649                 return -EFAULT;
 650         mark_page_dirty(kvm, gfn);
 651         return 0;
 652 }
 653 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
 654
 655 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
 656                     unsigned long len)
 657 {
 658         gfn_t gfn = gpa >> PAGE_SHIFT;
 659         int seg;
 660         int offset = offset_in_page(gpa);
 661         int ret;
 662
 663         while ((seg = next_segment(len, offset)) != 0) {
 664                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
 665                 if (ret < 0)
 666                         return ret;
 667                 offset = 0;
 668                 len -= seg;
 669                 data += seg;
 670                 ++gfn;
 671         }
 672         return 0;
 673 }
 674
 675 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
 676 {
 677         void *page_virt;
 678         struct page *page;
 679
 680         page = gfn_to_page(kvm, gfn);
 681         if (is_error_page(page)) {
 682                 kvm_release_page(page);
 683                 return -EFAULT;
 684         }
 685         page_virt = kmap_atomic(page, KM_USER0);
 686
 687         memset(page_virt + offset, 0, len);
 688
 689         kunmap_atomic(page_virt, KM_USER0);
 690         kvm_release_page(page);
 691         mark_page_dirty(kvm, gfn);
 692         return 0;
 693 }
 694 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
 695
 696 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
 697 {
 698         gfn_t gfn = gpa >> PAGE_SHIFT;
 699         int seg;
 700         int offset = offset_in_page(gpa);
 701         int ret;
 702
 703         while ((seg = next_segment(len, offset)) != 0) {
 704                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
 705                 if (ret < 0)
 706                         return ret;
 707                 offset = 0;
 708                 len -= seg;
 709                 ++gfn;
 710         }
 711         return 0;
 712 }
 713 EXPORT_SYMBOL_GPL(kvm_clear_guest);
 714
 715 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 716 {
 717         struct kvm_memory_slot *memslot;
 718
 719         gfn = unalias_gfn(kvm, gfn);
 720         memslot = __gfn_to_memslot(kvm, gfn);
 721         if (memslot && memslot->dirty_bitmap) {
 722                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 723
 724                 /* avoid RMW */
 725                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 726                         set_bit(rel_gfn, memslot->dirty_bitmap);
 727         }
 728 }
 729
 730 /*
 731  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
 732  */
 733 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 734 {
 735         DECLARE_WAITQUEUE(wait, current);
 736
 737         add_wait_queue(&vcpu->wq, &wait);
 738
 739         /*
 740          * We will block until either an interrupt or a signal wakes us up
 741          */
 742         while (!kvm_cpu_has_interrupt(vcpu)
 743                && !signal_pending(current)
 744                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
 745                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
 746                 set_current_state(TASK_INTERRUPTIBLE);
 747                 vcpu_put(vcpu);
 748                 schedule();
 749                 vcpu_load(vcpu);
 750         }
 751
 752         __set_current_state(TASK_RUNNING);
 753         remove_wait_queue(&vcpu->wq, &wait);
 754 }
 755
 756 void kvm_resched(struct kvm_vcpu *vcpu)
 757 {
 758         if (!need_resched())
 759                 return;
 760         cond_resched();
 761 }
 762 EXPORT_SYMBOL_GPL(kvm_resched);
 763
 764 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
 765                                     struct kvm_interrupt *irq)
 766 {
 767         if (irq->irq < 0 || irq->irq >= 256)
 768                 return -EINVAL;
 769         if (irqchip_in_kernel(vcpu->kvm))
 770                 return -ENXIO;
 771         vcpu_load(vcpu);
 772
 773         set_bit(irq->irq, vcpu->irq_pending);
 774         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
 775
 776         vcpu_put(vcpu);
 777
 778         return 0;
 779 }
 780
 781 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
 782                                     unsigned long address,
 783                                     int *type)
 784 {
 785         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
 786         unsigned long pgoff;
 787         struct page *page;
 788
 789         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 790         if (pgoff == 0)
 791                 page = virt_to_page(vcpu->run);
 792         else if (pgoff == KVM_PIO_PAGE_OFFSET)
 793                 page = virt_to_page(vcpu->pio_data);
 794         else
 795                 return NOPAGE_SIGBUS;
 796         get_page(page);
 797         if (type != NULL)
 798                 *type = VM_FAULT_MINOR;
 799
 800         return page;
 801 }
 802
 803 static struct vm_operations_struct kvm_vcpu_vm_ops = {
 804         .nopage = kvm_vcpu_nopage,
 805 };
 806
 807 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
 808 {
 809         vma->vm_ops = &kvm_vcpu_vm_ops;
 810         return 0;
 811 }
 812
 813 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
 814 {
 815         struct kvm_vcpu *vcpu = filp->private_data;
 816
 817         fput(vcpu->kvm->filp);
 818         return 0;
 819 }
 820
 821 static struct file_operations kvm_vcpu_fops = {
 822         .release        = kvm_vcpu_release,
 823         .unlocked_ioctl = kvm_vcpu_ioctl,
 824         .compat_ioctl   = kvm_vcpu_ioctl,
 825         .mmap           = kvm_vcpu_mmap,
 826 };
 827
 828 /*
 829  * Allocates an inode for the vcpu.
 830  */
 831 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
 832 {
 833         int fd, r;
 834         struct inode *inode;
 835         struct file *file;
 836
 837         r = anon_inode_getfd(&fd, &inode, &file,
 838                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
 839         if (r)
 840                 return r;
 841         atomic_inc(&vcpu->kvm->filp->f_count);
 842         return fd;
 843 }
 844
 845 /*
 846  * Creates some virtual cpus.  Good luck creating more than one.
 847  */
 848 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
 849 {
 850         int r;
 851         struct kvm_vcpu *vcpu;
 852
 853         if (!valid_vcpu(n))
 854                 return -EINVAL;
 855
 856         vcpu = kvm_arch_vcpu_create(kvm, n);
 857         if (IS_ERR(vcpu))
 858                 return PTR_ERR(vcpu);
 859
 860         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
 861
 862         mutex_lock(&kvm->lock);
 863         if (kvm->vcpus[n]) {
 864                 r = -EEXIST;
 865                 mutex_unlock(&kvm->lock);
 866                 goto vcpu_destroy;
 867         }
 868         kvm->vcpus[n] = vcpu;
 869         mutex_unlock(&kvm->lock);
 870
 871         /* Now it's all set up, let userspace reach it */
 872         r = create_vcpu_fd(vcpu);
 873         if (r < 0)
 874                 goto unlink;
 875         return r;
 876
 877 unlink:
 878         mutex_lock(&kvm->lock);
 879         kvm->vcpus[n] = NULL;
 880         mutex_unlock(&kvm->lock);
 881 vcpu_destroy:
 882         kvm_arch_vcpu_destory(vcpu);
 883         return r;
 884 }
 885
 886 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
 887 {
 888         if (sigset) {
 889                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
 890                 vcpu->sigset_active = 1;
 891                 vcpu->sigset = *sigset;
 892         } else
 893                 vcpu->sigset_active = 0;
 894         return 0;
 895 }
 896
 897 static long kvm_vcpu_ioctl(struct file *filp,
 898                            unsigned int ioctl, unsigned long arg)
 899 {
 900         struct kvm_vcpu *vcpu = filp->private_data;
 901         void __user *argp = (void __user *)arg;
 902         int r;
 903
 904         switch (ioctl) {
 905         case KVM_RUN:
 906                 r = -EINVAL;
 907                 if (arg)
 908                         goto out;
 909                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
 910                 break;
 911         case KVM_GET_REGS: {
 912                 struct kvm_regs kvm_regs;
 913
 914                 memset(&kvm_regs, 0, sizeof kvm_regs);
 915                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
 916                 if (r)
 917                         goto out;
 918                 r = -EFAULT;
 919                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
 920                         goto out;
 921                 r = 0;
 922                 break;
 923         }
 924         case KVM_SET_REGS: {
 925                 struct kvm_regs kvm_regs;
 926
 927                 r = -EFAULT;
 928                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
 929                         goto out;
 930                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
 931                 if (r)
 932                         goto out;
 933                 r = 0;
 934                 break;
 935         }
 936         case KVM_GET_SREGS: {
 937                 struct kvm_sregs kvm_sregs;
 938
 939                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
 940                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
 941                 if (r)
 942                         goto out;
 943                 r = -EFAULT;
 944                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
 945                         goto out;
 946                 r = 0;
 947                 break;
 948         }
 949         case KVM_SET_SREGS: {
 950                 struct kvm_sregs kvm_sregs;
 951
 952                 r = -EFAULT;
 953                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
 954                         goto out;
 955                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
 956                 if (r)
 957                         goto out;
 958                 r = 0;
 959                 break;
 960         }
 961         case KVM_TRANSLATE: {
 962                 struct kvm_translation tr;
 963
 964                 r = -EFAULT;
 965                 if (copy_from_user(&tr, argp, sizeof tr))
 966                         goto out;
 967                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
 968                 if (r)
 969                         goto out;
 970                 r = -EFAULT;
 971                 if (copy_to_user(argp, &tr, sizeof tr))
 972                         goto out;
 973                 r = 0;
 974                 break;
 975         }
 976         case KVM_INTERRUPT: {
 977                 struct kvm_interrupt irq;
 978
 979                 r = -EFAULT;
 980                 if (copy_from_user(&irq, argp, sizeof irq))
 981                         goto out;
 982                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
 983                 if (r)
 984                         goto out;
 985                 r = 0;
 986                 break;
 987         }
 988         case KVM_DEBUG_GUEST: {
 989                 struct kvm_debug_guest dbg;
 990
 991                 r = -EFAULT;
 992                 if (copy_from_user(&dbg, argp, sizeof dbg))
 993                         goto out;
 994                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
 995                 if (r)
 996                         goto out;
 997                 r = 0;
 998                 break;
 999         }
1000         case KVM_SET_SIGNAL_MASK: {
1001                 struct kvm_signal_mask __user *sigmask_arg = argp;
1002                 struct kvm_signal_mask kvm_sigmask;
1003                 sigset_t sigset, *p;
1004
1005                 p = NULL;
1006                 if (argp) {
1007                         r = -EFAULT;
1008                         if (copy_from_user(&kvm_sigmask, argp,
1009                                            sizeof kvm_sigmask))
1010                                 goto out;
1011                         r = -EINVAL;
1012                         if (kvm_sigmask.len != sizeof sigset)
1013                                 goto out;
1014                         r = -EFAULT;
1015                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1016                                            sizeof sigset))
1017                                 goto out;
1018                         p = &sigset;
1019                 }
1020                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1021                 break;
1022         }
1023         case KVM_GET_FPU: {
1024                 struct kvm_fpu fpu;
1025
1026                 memset(&fpu, 0, sizeof fpu);
1027                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
1028                 if (r)
1029                         goto out;
1030                 r = -EFAULT;
1031                 if (copy_to_user(argp, &fpu, sizeof fpu))
1032                         goto out;
1033                 r = 0;
1034                 break;
1035         }
1036         case KVM_SET_FPU: {
1037                 struct kvm_fpu fpu;
1038
1039                 r = -EFAULT;
1040                 if (copy_from_user(&fpu, argp, sizeof fpu))
1041                         goto out;
1042                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
1043                 if (r)
1044                         goto out;
1045                 r = 0;
1046                 break;
1047         }
1048         default:
1049                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1050         }
1051 out:
1052         return r;
1053 }
1054
1055 static long kvm_vm_ioctl(struct file *filp,
1056                            unsigned int ioctl, unsigned long arg)
1057 {
1058         struct kvm *kvm = filp->private_data;
1059         void __user *argp = (void __user *)arg;
1060         int r;
1061
1062         switch (ioctl) {
1063         case KVM_CREATE_VCPU:
1064                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1065                 if (r < 0)
1066                         goto out;
1067                 break;
1068         case KVM_SET_USER_MEMORY_REGION: {
1069                 struct kvm_userspace_memory_region kvm_userspace_mem;
1070
1071                 r = -EFAULT;
1072                 if (copy_from_user(&kvm_userspace_mem, argp,
1073                                                 sizeof kvm_userspace_mem))
1074                         goto out;
1075
1076                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1077                 if (r)
1078                         goto out;
1079                 break;
1080         }
1081         case KVM_GET_DIRTY_LOG: {
1082                 struct kvm_dirty_log log;
1083
1084                 r = -EFAULT;
1085                 if (copy_from_user(&log, argp, sizeof log))
1086                         goto out;
1087                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1088                 if (r)
1089                         goto out;
1090                 break;
1091         }
1092         default:
1093                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1094         }
1095 out:
1096         return r;
1097 }
1098
1099 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
1100                                   unsigned long address,
1101                                   int *type)
1102 {
1103         struct kvm *kvm = vma->vm_file->private_data;
1104         unsigned long pgoff;
1105         struct page *page;
1106
1107         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1108         if (!kvm_is_visible_gfn(kvm, pgoff))
1109                 return NOPAGE_SIGBUS;
1110         /* current->mm->mmap_sem is already held so call lockless version */
1111         page = __gfn_to_page(kvm, pgoff);
1112         if (is_error_page(page)) {
1113                 kvm_release_page(page);
1114                 return NOPAGE_SIGBUS;
1115         }
1116         if (type != NULL)
1117                 *type = VM_FAULT_MINOR;
1118
1119         return page;
1120 }
1121
1122 static struct vm_operations_struct kvm_vm_vm_ops = {
1123         .nopage = kvm_vm_nopage,
1124 };
1125
1126 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1127 {
1128         vma->vm_ops = &kvm_vm_vm_ops;
1129         return 0;
1130 }
1131
1132 static struct file_operations kvm_vm_fops = {
1133         .release        = kvm_vm_release,
1134         .unlocked_ioctl = kvm_vm_ioctl,
1135         .compat_ioctl   = kvm_vm_ioctl,
1136         .mmap           = kvm_vm_mmap,
1137 };
1138
1139 static int kvm_dev_ioctl_create_vm(void)
1140 {
1141         int fd, r;
1142         struct inode *inode;
1143         struct file *file;
1144         struct kvm *kvm;
1145
1146         kvm = kvm_create_vm();
1147         if (IS_ERR(kvm))
1148                 return PTR_ERR(kvm);
1149         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1150         if (r) {
1151                 kvm_destroy_vm(kvm);
1152                 return r;
1153         }
1154
1155         kvm->filp = file;
1156
1157         return fd;
1158 }
1159
1160 static long kvm_dev_ioctl(struct file *filp,
1161                           unsigned int ioctl, unsigned long arg)
1162 {
1163         void __user *argp = (void __user *)arg;
1164         long r = -EINVAL;
1165
1166         switch (ioctl) {
1167         case KVM_GET_API_VERSION:
1168                 r = -EINVAL;
1169                 if (arg)
1170                         goto out;
1171                 r = KVM_API_VERSION;
1172                 break;
1173         case KVM_CREATE_VM:
1174                 r = -EINVAL;
1175                 if (arg)
1176                         goto out;
1177                 r = kvm_dev_ioctl_create_vm();
1178                 break;
1179         case KVM_CHECK_EXTENSION:
1180                 r = kvm_dev_ioctl_check_extension((long)argp);
1181                 break;
1182         case KVM_GET_VCPU_MMAP_SIZE:
1183                 r = -EINVAL;
1184                 if (arg)
1185                         goto out;
1186                 r = 2 * PAGE_SIZE;
1187                 break;
1188         default:
1189                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1190         }
1191 out:
1192         return r;
1193 }
1194
1195 static struct file_operations kvm_chardev_ops = {
1196         .unlocked_ioctl = kvm_dev_ioctl,
1197         .compat_ioctl   = kvm_dev_ioctl,
1198 };
1199
1200 static struct miscdevice kvm_dev = {
1201         KVM_MINOR,
1202         "kvm",
1203         &kvm_chardev_ops,
1204 };
1205
1206 static void hardware_enable(void *junk)
1207 {
1208         int cpu = raw_smp_processor_id();
1209
1210         if (cpu_isset(cpu, cpus_hardware_enabled))
1211                 return;
1212         cpu_set(cpu, cpus_hardware_enabled);
1213         kvm_arch_hardware_enable(NULL);
1214 }
1215
1216 static void hardware_disable(void *junk)
1217 {
1218         int cpu = raw_smp_processor_id();
1219
1220         if (!cpu_isset(cpu, cpus_hardware_enabled))
1221                 return;
1222         cpu_clear(cpu, cpus_hardware_enabled);
1223         decache_vcpus_on_cpu(cpu);
1224         kvm_arch_hardware_disable(NULL);
1225 }
1226
1227 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1228                            void *v)
1229 {
1230         int cpu = (long)v;
1231
1232         val &= ~CPU_TASKS_FROZEN;
1233         switch (val) {
1234         case CPU_DYING:
1235                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1236                        cpu);
1237                 hardware_disable(NULL);
1238                 break;
1239         case CPU_UP_CANCELED:
1240                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1241                        cpu);
1242                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1243                 break;
1244         case CPU_ONLINE:
1245                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1246                        cpu);
1247                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1248                 break;
1249         }
1250         return NOTIFY_OK;
1251 }
1252
1253 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1254                       void *v)
1255 {
1256         if (val == SYS_RESTART) {
1257                 /*
1258                  * Some (well, at least mine) BIOSes hang on reboot if
1259                  * in vmx root mode.
1260                  */
1261                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1262                 on_each_cpu(hardware_disable, NULL, 0, 1);
1263         }
1264         return NOTIFY_OK;
1265 }
1266
1267 static struct notifier_block kvm_reboot_notifier = {
1268         .notifier_call = kvm_reboot,
1269         .priority = 0,
1270 };
1271
1272 void kvm_io_bus_init(struct kvm_io_bus *bus)
1273 {
1274         memset(bus, 0, sizeof(*bus));
1275 }
1276
1277 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1278 {
1279         int i;
1280
1281         for (i = 0; i < bus->dev_count; i++) {
1282                 struct kvm_io_device *pos = bus->devs[i];
1283
1284                 kvm_iodevice_destructor(pos);
1285         }
1286 }
1287
1288 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1289 {
1290         int i;
1291
1292         for (i = 0; i < bus->dev_count; i++) {
1293                 struct kvm_io_device *pos = bus->devs[i];
1294
1295                 if (pos->in_range(pos, addr))
1296                         return pos;
1297         }
1298
1299         return NULL;
1300 }
1301
1302 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1303 {
1304         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1305
1306         bus->devs[bus->dev_count++] = dev;
1307 }
1308
1309 static struct notifier_block kvm_cpu_notifier = {
1310         .notifier_call = kvm_cpu_hotplug,
1311         .priority = 20, /* must be > scheduler priority */
1312 };
1313
1314 static u64 stat_get(void *_offset)
1315 {
1316         unsigned offset = (long)_offset;
1317         u64 total = 0;
1318         struct kvm *kvm;
1319         struct kvm_vcpu *vcpu;
1320         int i;
1321
1322         spin_lock(&kvm_lock);
1323         list_for_each_entry(kvm, &vm_list, vm_list)
1324                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1325                         vcpu = kvm->vcpus[i];
1326                         if (vcpu)
1327                                 total += *(u32 *)((void *)vcpu + offset);
1328                 }
1329         spin_unlock(&kvm_lock);
1330         return total;
1331 }
1332
1333 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
1334
1335 static __init void kvm_init_debug(void)
1336 {
1337         struct kvm_stats_debugfs_item *p;
1338
1339         debugfs_dir = debugfs_create_dir("kvm", NULL);
1340         for (p = debugfs_entries; p->name; ++p)
1341                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1342                                                 (void *)(long)p->offset,
1343                                                 &stat_fops);
1344 }
1345
1346 static void kvm_exit_debug(void)
1347 {
1348         struct kvm_stats_debugfs_item *p;
1349
1350         for (p = debugfs_entries; p->name; ++p)
1351                 debugfs_remove(p->dentry);
1352         debugfs_remove(debugfs_dir);
1353 }
1354
1355 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1356 {
1357         hardware_disable(NULL);
1358         return 0;
1359 }
1360
1361 static int kvm_resume(struct sys_device *dev)
1362 {
1363         hardware_enable(NULL);
1364         return 0;
1365 }
1366
1367 static struct sysdev_class kvm_sysdev_class = {
1368         .name = "kvm",
1369         .suspend = kvm_suspend,
1370         .resume = kvm_resume,
1371 };
1372
1373 static struct sys_device kvm_sysdev = {
1374         .id = 0,
1375         .cls = &kvm_sysdev_class,
1376 };
1377
1378 struct page *bad_page;
1379
1380 static inline
1381 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1382 {
1383         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1384 }
1385
1386 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1387 {
1388         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1389
1390         kvm_arch_vcpu_load(vcpu, cpu);
1391 }
1392
1393 static void kvm_sched_out(struct preempt_notifier *pn,
1394                           struct task_struct *next)
1395 {
1396         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1397
1398         kvm_arch_vcpu_put(vcpu);
1399 }
1400
1401 int kvm_init(void *opaque, unsigned int vcpu_size,
1402                   struct module *module)
1403 {
1404         int r;
1405         int cpu;
1406
1407         r = kvm_mmu_module_init();
1408         if (r)
1409                 goto out4;
1410
1411         kvm_init_debug();
1412
1413         r = kvm_arch_init(opaque);
1414         if (r)
1415                 goto out4;
1416
1417         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1418
1419         if (bad_page == NULL) {
1420                 r = -ENOMEM;
1421                 goto out;
1422         }
1423
1424         r = kvm_arch_hardware_setup();
1425         if (r < 0)
1426                 goto out;
1427
1428         for_each_online_cpu(cpu) {
1429                 smp_call_function_single(cpu,
1430                                 kvm_arch_check_processor_compat,
1431                                 &r, 0, 1);
1432                 if (r < 0)
1433                         goto out_free_0;
1434         }
1435
1436         on_each_cpu(hardware_enable, NULL, 0, 1);
1437         r = register_cpu_notifier(&kvm_cpu_notifier);
1438         if (r)
1439                 goto out_free_1;
1440         register_reboot_notifier(&kvm_reboot_notifier);
1441
1442         r = sysdev_class_register(&kvm_sysdev_class);
1443         if (r)
1444                 goto out_free_2;
1445
1446         r = sysdev_register(&kvm_sysdev);
1447         if (r)
1448                 goto out_free_3;
1449
1450         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1451         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1452                                            __alignof__(struct kvm_vcpu),
1453                                            0, NULL);
1454         if (!kvm_vcpu_cache) {
1455                 r = -ENOMEM;
1456                 goto out_free_4;
1457         }
1458
1459         kvm_chardev_ops.owner = module;
1460
1461         r = misc_register(&kvm_dev);
1462         if (r) {
1463                 printk(KERN_ERR "kvm: misc device register failed\n");
1464                 goto out_free;
1465         }
1466
1467         kvm_preempt_ops.sched_in = kvm_sched_in;
1468         kvm_preempt_ops.sched_out = kvm_sched_out;
1469
1470         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
1471
1472         return 0;
1473
1474 out_free:
1475         kmem_cache_destroy(kvm_vcpu_cache);
1476 out_free_4:
1477         sysdev_unregister(&kvm_sysdev);
1478 out_free_3:
1479         sysdev_class_unregister(&kvm_sysdev_class);
1480 out_free_2:
1481         unregister_reboot_notifier(&kvm_reboot_notifier);
1482         unregister_cpu_notifier(&kvm_cpu_notifier);
1483 out_free_1:
1484         on_each_cpu(hardware_disable, NULL, 0, 1);
1485 out_free_0:
1486         kvm_arch_hardware_unsetup();
1487 out:
1488         kvm_arch_exit();
1489         kvm_exit_debug();
1490         kvm_mmu_module_exit();
1491 out4:
1492         return r;
1493 }
1494 EXPORT_SYMBOL_GPL(kvm_init);
1495
1496 void kvm_exit(void)
1497 {
1498         misc_deregister(&kvm_dev);
1499         kmem_cache_destroy(kvm_vcpu_cache);
1500         sysdev_unregister(&kvm_sysdev);
1501         sysdev_class_unregister(&kvm_sysdev_class);
1502         unregister_reboot_notifier(&kvm_reboot_notifier);
1503         unregister_cpu_notifier(&kvm_cpu_notifier);
1504         on_each_cpu(hardware_disable, NULL, 0, 1);
1505         kvm_arch_hardware_unsetup();
1506         kvm_arch_exit();
1507         kvm_exit_debug();
1508         __free_page(bad_page);
1509         kvm_mmu_module_exit();
1510 }
1511 EXPORT_SYMBOL_GPL(kvm_exit);