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 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 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         if (cpus_empty(cpus))
 118                 return;
 119         ++kvm->stat.remote_tlb_flush;
 120         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 121 }
 122
 123 void kvm_reload_remote_mmus(struct kvm *kvm)
 124 {
 125         int i, cpu;
 126         cpumask_t cpus;
 127         struct kvm_vcpu *vcpu;
 128
 129         cpus_clear(cpus);
 130         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 131                 vcpu = kvm->vcpus[i];
 132                 if (!vcpu)
 133                         continue;
 134                 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
 135                         continue;
 136                 cpu = vcpu->cpu;
 137                 if (cpu != -1 && cpu != raw_smp_processor_id())
 138                         cpu_set(cpu, cpus);
 139         }
 140         if (cpus_empty(cpus))
 141                 return;
 142         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 143 }
 144
 145
 146 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 147 {
 148         struct page *page;
 149         int r;
 150
 151         mutex_init(&vcpu->mutex);
 152         vcpu->cpu = -1;
 153         vcpu->kvm = kvm;
 154         vcpu->vcpu_id = id;
 155         init_waitqueue_head(&vcpu->wq);
 156
 157         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 158         if (!page) {
 159                 r = -ENOMEM;
 160                 goto fail;
 161         }
 162         vcpu->run = page_address(page);
 163
 164         r = kvm_arch_vcpu_init(vcpu);
 165         if (r < 0)
 166                 goto fail_free_run;
 167         return 0;
 168
 169 fail_free_run:
 170         free_page((unsigned long)vcpu->run);
 171 fail:
 172         return r;
 173 }
 174 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 175
 176 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 177 {
 178         kvm_arch_vcpu_uninit(vcpu);
 179         free_page((unsigned long)vcpu->run);
 180 }
 181 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 182
 183 static struct kvm *kvm_create_vm(void)
 184 {
 185         struct kvm *kvm = kvm_arch_create_vm();
 186
 187         if (IS_ERR(kvm))
 188                 goto out;
 189
 190         kvm->mm = current->mm;
 191         atomic_inc(&kvm->mm->mm_count);
 192         spin_lock_init(&kvm->mmu_lock);
 193         kvm_io_bus_init(&kvm->pio_bus);
 194         mutex_init(&kvm->lock);
 195         kvm_io_bus_init(&kvm->mmio_bus);
 196         init_rwsem(&kvm->slots_lock);
 197         atomic_set(&kvm->users_count, 1);
 198         spin_lock(&kvm_lock);
 199         list_add(&kvm->vm_list, &vm_list);
 200         spin_unlock(&kvm_lock);
 201 out:
 202         return kvm;
 203 }
 204
 205 /*
 206  * Free any memory in @free but not in @dont.
 207  */
 208 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 209                                   struct kvm_memory_slot *dont)
 210 {
 211         if (!dont || free->rmap != dont->rmap)
 212                 vfree(free->rmap);
 213
 214         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 215                 vfree(free->dirty_bitmap);
 216
 217         if (!dont || free->lpage_info != dont->lpage_info)
 218                 vfree(free->lpage_info);
 219
 220         free->npages = 0;
 221         free->dirty_bitmap = NULL;
 222         free->rmap = NULL;
 223         free->lpage_info = NULL;
 224 }
 225
 226 void kvm_free_physmem(struct kvm *kvm)
 227 {
 228         int i;
 229
 230         for (i = 0; i < kvm->nmemslots; ++i)
 231                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 232 }
 233
 234 static void kvm_destroy_vm(struct kvm *kvm)
 235 {
 236         struct mm_struct *mm = kvm->mm;
 237
 238         spin_lock(&kvm_lock);
 239         list_del(&kvm->vm_list);
 240         spin_unlock(&kvm_lock);
 241         kvm_io_bus_destroy(&kvm->pio_bus);
 242         kvm_io_bus_destroy(&kvm->mmio_bus);
 243         kvm_arch_destroy_vm(kvm);
 244         mmdrop(mm);
 245 }
 246
 247 void kvm_get_kvm(struct kvm *kvm)
 248 {
 249         atomic_inc(&kvm->users_count);
 250 }
 251 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 252
 253 void kvm_put_kvm(struct kvm *kvm)
 254 {
 255         if (atomic_dec_and_test(&kvm->users_count))
 256                 kvm_destroy_vm(kvm);
 257 }
 258 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 259
 260
 261 static int kvm_vm_release(struct inode *inode, struct file *filp)
 262 {
 263         struct kvm *kvm = filp->private_data;
 264
 265         kvm_put_kvm(kvm);
 266         return 0;
 267 }
 268
 269 /*
 270  * Allocate some memory and give it an address in the guest physical address
 271  * space.
 272  *
 273  * Discontiguous memory is allowed, mostly for framebuffers.
 274  *
 275  * Must be called holding mmap_sem for write.
 276  */
 277 int __kvm_set_memory_region(struct kvm *kvm,
 278                             struct kvm_userspace_memory_region *mem,
 279                             int user_alloc)
 280 {
 281         int r;
 282         gfn_t base_gfn;
 283         unsigned long npages;
 284         unsigned long i;
 285         struct kvm_memory_slot *memslot;
 286         struct kvm_memory_slot old, new;
 287
 288         r = -EINVAL;
 289         /* General sanity checks */
 290         if (mem->memory_size & (PAGE_SIZE - 1))
 291                 goto out;
 292         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 293                 goto out;
 294         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 295                 goto out;
 296         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 297                 goto out;
 298
 299         memslot = &kvm->memslots[mem->slot];
 300         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 301         npages = mem->memory_size >> PAGE_SHIFT;
 302
 303         if (!npages)
 304                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 305
 306         new = old = *memslot;
 307
 308         new.base_gfn = base_gfn;
 309         new.npages = npages;
 310         new.flags = mem->flags;
 311
 312         /* Disallow changing a memory slot's size. */
 313         r = -EINVAL;
 314         if (npages && old.npages && npages != old.npages)
 315                 goto out_free;
 316
 317         /* Check for overlaps */
 318         r = -EEXIST;
 319         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 320                 struct kvm_memory_slot *s = &kvm->memslots[i];
 321
 322                 if (s == memslot)
 323                         continue;
 324                 if (!((base_gfn + npages <= s->base_gfn) ||
 325                       (base_gfn >= s->base_gfn + s->npages)))
 326                         goto out_free;
 327         }
 328
 329         /* Free page dirty bitmap if unneeded */
 330         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 331                 new.dirty_bitmap = NULL;
 332
 333         r = -ENOMEM;
 334
 335         /* Allocate if a slot is being created */
 336         if (npages && !new.rmap) {
 337                 new.rmap = vmalloc(npages * sizeof(struct page *));
 338
 339                 if (!new.rmap)
 340                         goto out_free;
 341
 342                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 343
 344                 new.user_alloc = user_alloc;
 345                 new.userspace_addr = mem->userspace_addr;
 346         }
 347         if (npages && !new.lpage_info) {
 348                 int largepages = npages / KVM_PAGES_PER_HPAGE;
 349                 if (npages % KVM_PAGES_PER_HPAGE)
 350                         largepages++;
 351                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 352                         largepages++;
 353
 354                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
 355
 356                 if (!new.lpage_info)
 357                         goto out_free;
 358
 359                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
 360
 361                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 362                         new.lpage_info[0].write_count = 1;
 363                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
 364                         new.lpage_info[largepages-1].write_count = 1;
 365         }
 366
 367         /* Allocate page dirty bitmap if needed */
 368         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 369                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 370
 371                 new.dirty_bitmap = vmalloc(dirty_bytes);
 372                 if (!new.dirty_bitmap)
 373                         goto out_free;
 374                 memset(new.dirty_bitmap, 0, dirty_bytes);
 375         }
 376
 377         if (mem->slot >= kvm->nmemslots)
 378                 kvm->nmemslots = mem->slot + 1;
 379
 380         *memslot = new;
 381
 382         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
 383         if (r) {
 384                 *memslot = old;
 385                 goto out_free;
 386         }
 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         down_write(&kvm->slots_lock);
 406         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 407         up_write(&kvm->slots_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 int kvm_get_dirty_log(struct kvm *kvm,
 423                         struct kvm_dirty_log *log, int *is_dirty)
 424 {
 425         struct kvm_memory_slot *memslot;
 426         int r, i;
 427         int n;
 428         unsigned long any = 0;
 429
 430         r = -EINVAL;
 431         if (log->slot >= KVM_MEMORY_SLOTS)
 432                 goto out;
 433
 434         memslot = &kvm->memslots[log->slot];
 435         r = -ENOENT;
 436         if (!memslot->dirty_bitmap)
 437                 goto out;
 438
 439         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 440
 441         for (i = 0; !any && i < n/sizeof(long); ++i)
 442                 any = memslot->dirty_bitmap[i];
 443
 444         r = -EFAULT;
 445         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 446                 goto out;
 447
 448         if (any)
 449                 *is_dirty = 1;
 450
 451         r = 0;
 452 out:
 453         return r;
 454 }
 455
 456 int is_error_page(struct page *page)
 457 {
 458         return page == bad_page;
 459 }
 460 EXPORT_SYMBOL_GPL(is_error_page);
 461
 462 int is_error_pfn(pfn_t pfn)
 463 {
 464         return pfn == bad_pfn;
 465 }
 466 EXPORT_SYMBOL_GPL(is_error_pfn);
 467
 468 static inline unsigned long bad_hva(void)
 469 {
 470         return PAGE_OFFSET;
 471 }
 472
 473 int kvm_is_error_hva(unsigned long addr)
 474 {
 475         return addr == bad_hva();
 476 }
 477 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 478
 479 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 480 {
 481         int i;
 482
 483         for (i = 0; i < kvm->nmemslots; ++i) {
 484                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 485
 486                 if (gfn >= memslot->base_gfn
 487                     && gfn < memslot->base_gfn + memslot->npages)
 488                         return memslot;
 489         }
 490         return NULL;
 491 }
 492
 493 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 494 {
 495         gfn = unalias_gfn(kvm, gfn);
 496         return __gfn_to_memslot(kvm, gfn);
 497 }
 498
 499 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 500 {
 501         int i;
 502
 503         gfn = unalias_gfn(kvm, gfn);
 504         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 505                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 506
 507                 if (gfn >= memslot->base_gfn
 508                     && gfn < memslot->base_gfn + memslot->npages)
 509                         return 1;
 510         }
 511         return 0;
 512 }
 513 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 514
 515 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 516 {
 517         struct kvm_memory_slot *slot;
 518
 519         gfn = unalias_gfn(kvm, gfn);
 520         slot = __gfn_to_memslot(kvm, gfn);
 521         if (!slot)
 522                 return bad_hva();
 523         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 524 }
 525
 526 /*
 527  * Requires current->mm->mmap_sem to be held
 528  */
 529 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 530 {
 531         struct page *page[1];
 532         unsigned long addr;
 533         int npages;
 534
 535         might_sleep();
 536
 537         addr = gfn_to_hva(kvm, gfn);
 538         if (kvm_is_error_hva(addr)) {
 539                 get_page(bad_page);
 540                 return page_to_pfn(bad_page);
 541         }
 542
 543         npages = get_user_pages(current, current->mm, addr, 1, 1, 1, page,
 544                                 NULL);
 545
 546         if (npages != 1) {
 547                 get_page(bad_page);
 548                 return page_to_pfn(bad_page);
 549         }
 550
 551         return page_to_pfn(page[0]);
 552 }
 553
 554 EXPORT_SYMBOL_GPL(gfn_to_pfn);
 555
 556 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 557 {
 558         return pfn_to_page(gfn_to_pfn(kvm, gfn));
 559 }
 560
 561 EXPORT_SYMBOL_GPL(gfn_to_page);
 562
 563 void kvm_release_page_clean(struct page *page)
 564 {
 565         kvm_release_pfn_clean(page_to_pfn(page));
 566 }
 567 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
 568
 569 void kvm_release_pfn_clean(pfn_t pfn)
 570 {
 571         put_page(pfn_to_page(pfn));
 572 }
 573 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
 574
 575 void kvm_release_page_dirty(struct page *page)
 576 {
 577         kvm_release_pfn_dirty(page_to_pfn(page));
 578 }
 579 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
 580
 581 void kvm_release_pfn_dirty(pfn_t pfn)
 582 {
 583         kvm_set_pfn_dirty(pfn);
 584         kvm_release_pfn_clean(pfn);
 585 }
 586 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
 587
 588 void kvm_set_page_dirty(struct page *page)
 589 {
 590         kvm_set_pfn_dirty(page_to_pfn(page));
 591 }
 592 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
 593
 594 void kvm_set_pfn_dirty(pfn_t pfn)
 595 {
 596         struct page *page = pfn_to_page(pfn);
 597         if (!PageReserved(page))
 598                 SetPageDirty(page);
 599 }
 600 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
 601
 602 void kvm_set_pfn_accessed(pfn_t pfn)
 603 {
 604         mark_page_accessed(pfn_to_page(pfn));
 605 }
 606 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
 607
 608 void kvm_get_pfn(pfn_t pfn)
 609 {
 610         get_page(pfn_to_page(pfn));
 611 }
 612 EXPORT_SYMBOL_GPL(kvm_get_pfn);
 613
 614 static int next_segment(unsigned long len, int offset)
 615 {
 616         if (len > PAGE_SIZE - offset)
 617                 return PAGE_SIZE - offset;
 618         else
 619                 return len;
 620 }
 621
 622 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 623                         int len)
 624 {
 625         int r;
 626         unsigned long addr;
 627
 628         addr = gfn_to_hva(kvm, gfn);
 629         if (kvm_is_error_hva(addr))
 630                 return -EFAULT;
 631         r = copy_from_user(data, (void __user *)addr + offset, len);
 632         if (r)
 633                 return -EFAULT;
 634         return 0;
 635 }
 636 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 637
 638 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 639 {
 640         gfn_t gfn = gpa >> PAGE_SHIFT;
 641         int seg;
 642         int offset = offset_in_page(gpa);
 643         int ret;
 644
 645         while ((seg = next_segment(len, offset)) != 0) {
 646                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 647                 if (ret < 0)
 648                         return ret;
 649                 offset = 0;
 650                 len -= seg;
 651                 data += seg;
 652                 ++gfn;
 653         }
 654         return 0;
 655 }
 656 EXPORT_SYMBOL_GPL(kvm_read_guest);
 657
 658 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
 659                           unsigned long len)
 660 {
 661         int r;
 662         unsigned long addr;
 663         gfn_t gfn = gpa >> PAGE_SHIFT;
 664         int offset = offset_in_page(gpa);
 665
 666         addr = gfn_to_hva(kvm, gfn);
 667         if (kvm_is_error_hva(addr))
 668                 return -EFAULT;
 669         pagefault_disable();
 670         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
 671         pagefault_enable();
 672         if (r)
 673                 return -EFAULT;
 674         return 0;
 675 }
 676 EXPORT_SYMBOL(kvm_read_guest_atomic);
 677
 678 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
 679                          int offset, int len)
 680 {
 681         int r;
 682         unsigned long addr;
 683
 684         addr = gfn_to_hva(kvm, gfn);
 685         if (kvm_is_error_hva(addr))
 686                 return -EFAULT;
 687         r = copy_to_user((void __user *)addr + offset, data, len);
 688         if (r)
 689                 return -EFAULT;
 690         mark_page_dirty(kvm, gfn);
 691         return 0;
 692 }
 693 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
 694
 695 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
 696                     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_write_guest_page(kvm, gfn, data, offset, seg);
 705                 if (ret < 0)
 706                         return ret;
 707                 offset = 0;
 708                 len -= seg;
 709                 data += seg;
 710                 ++gfn;
 711         }
 712         return 0;
 713 }
 714
 715 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
 716 {
 717         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
 718 }
 719 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
 720
 721 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
 722 {
 723         gfn_t gfn = gpa >> PAGE_SHIFT;
 724         int seg;
 725         int offset = offset_in_page(gpa);
 726         int ret;
 727
 728         while ((seg = next_segment(len, offset)) != 0) {
 729                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
 730                 if (ret < 0)
 731                         return ret;
 732                 offset = 0;
 733                 len -= seg;
 734                 ++gfn;
 735         }
 736         return 0;
 737 }
 738 EXPORT_SYMBOL_GPL(kvm_clear_guest);
 739
 740 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 741 {
 742         struct kvm_memory_slot *memslot;
 743
 744         gfn = unalias_gfn(kvm, gfn);
 745         memslot = __gfn_to_memslot(kvm, gfn);
 746         if (memslot && memslot->dirty_bitmap) {
 747                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 748
 749                 /* avoid RMW */
 750                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 751                         set_bit(rel_gfn, memslot->dirty_bitmap);
 752         }
 753 }
 754
 755 /*
 756  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
 757  */
 758 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 759 {
 760         DECLARE_WAITQUEUE(wait, current);
 761
 762         add_wait_queue(&vcpu->wq, &wait);
 763
 764         /*
 765          * We will block until either an interrupt or a signal wakes us up
 766          */
 767         while (!kvm_cpu_has_interrupt(vcpu)
 768                && !signal_pending(current)
 769                && !kvm_arch_vcpu_runnable(vcpu)) {
 770                 set_current_state(TASK_INTERRUPTIBLE);
 771                 vcpu_put(vcpu);
 772                 schedule();
 773                 vcpu_load(vcpu);
 774         }
 775
 776         __set_current_state(TASK_RUNNING);
 777         remove_wait_queue(&vcpu->wq, &wait);
 778 }
 779
 780 void kvm_resched(struct kvm_vcpu *vcpu)
 781 {
 782         if (!need_resched())
 783                 return;
 784         cond_resched();
 785 }
 786 EXPORT_SYMBOL_GPL(kvm_resched);
 787
 788 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 789 {
 790         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
 791         struct page *page;
 792
 793         if (vmf->pgoff == 0)
 794                 page = virt_to_page(vcpu->run);
 795 #ifdef CONFIG_X86
 796         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
 797                 page = virt_to_page(vcpu->arch.pio_data);
 798 #endif
 799         else
 800                 return VM_FAULT_SIGBUS;
 801         get_page(page);
 802         vmf->page = page;
 803         return 0;
 804 }
 805
 806 static struct vm_operations_struct kvm_vcpu_vm_ops = {
 807         .fault = kvm_vcpu_fault,
 808 };
 809
 810 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
 811 {
 812         vma->vm_ops = &kvm_vcpu_vm_ops;
 813         return 0;
 814 }
 815
 816 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
 817 {
 818         struct kvm_vcpu *vcpu = filp->private_data;
 819
 820         fput(vcpu->kvm->filp);
 821         return 0;
 822 }
 823
 824 static const struct file_operations kvm_vcpu_fops = {
 825         .release        = kvm_vcpu_release,
 826         .unlocked_ioctl = kvm_vcpu_ioctl,
 827         .compat_ioctl   = kvm_vcpu_ioctl,
 828         .mmap           = kvm_vcpu_mmap,
 829 };
 830
 831 /*
 832  * Allocates an inode for the vcpu.
 833  */
 834 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
 835 {
 836         int fd, r;
 837         struct inode *inode;
 838         struct file *file;
 839
 840         r = anon_inode_getfd(&fd, &inode, &file,
 841                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
 842         if (r)
 843                 return r;
 844         atomic_inc(&vcpu->kvm->filp->f_count);
 845         return fd;
 846 }
 847
 848 /*
 849  * Creates some virtual cpus.  Good luck creating more than one.
 850  */
 851 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
 852 {
 853         int r;
 854         struct kvm_vcpu *vcpu;
 855
 856         if (!valid_vcpu(n))
 857                 return -EINVAL;
 858
 859         vcpu = kvm_arch_vcpu_create(kvm, n);
 860         if (IS_ERR(vcpu))
 861                 return PTR_ERR(vcpu);
 862
 863         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
 864
 865         r = kvm_arch_vcpu_setup(vcpu);
 866         if (r)
 867                 goto vcpu_destroy;
 868
 869         mutex_lock(&kvm->lock);
 870         if (kvm->vcpus[n]) {
 871                 r = -EEXIST;
 872                 mutex_unlock(&kvm->lock);
 873                 goto vcpu_destroy;
 874         }
 875         kvm->vcpus[n] = vcpu;
 876         mutex_unlock(&kvm->lock);
 877
 878         /* Now it's all set up, let userspace reach it */
 879         r = create_vcpu_fd(vcpu);
 880         if (r < 0)
 881                 goto unlink;
 882         return r;
 883
 884 unlink:
 885         mutex_lock(&kvm->lock);
 886         kvm->vcpus[n] = NULL;
 887         mutex_unlock(&kvm->lock);
 888 vcpu_destroy:
 889         kvm_arch_vcpu_destroy(vcpu);
 890         return r;
 891 }
 892
 893 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
 894 {
 895         if (sigset) {
 896                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
 897                 vcpu->sigset_active = 1;
 898                 vcpu->sigset = *sigset;
 899         } else
 900                 vcpu->sigset_active = 0;
 901         return 0;
 902 }
 903
 904 static long kvm_vcpu_ioctl(struct file *filp,
 905                            unsigned int ioctl, unsigned long arg)
 906 {
 907         struct kvm_vcpu *vcpu = filp->private_data;
 908         void __user *argp = (void __user *)arg;
 909         int r;
 910
 911         if (vcpu->kvm->mm != current->mm)
 912                 return -EIO;
 913         switch (ioctl) {
 914         case KVM_RUN:
 915                 r = -EINVAL;
 916                 if (arg)
 917                         goto out;
 918                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
 919                 break;
 920         case KVM_GET_REGS: {
 921                 struct kvm_regs *kvm_regs;
 922
 923                 r = -ENOMEM;
 924                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
 925                 if (!kvm_regs)
 926                         goto out;
 927                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
 928                 if (r)
 929                         goto out_free1;
 930                 r = -EFAULT;
 931                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
 932                         goto out_free1;
 933                 r = 0;
 934 out_free1:
 935                 kfree(kvm_regs);
 936                 break;
 937         }
 938         case KVM_SET_REGS: {
 939                 struct kvm_regs *kvm_regs;
 940
 941                 r = -ENOMEM;
 942                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
 943                 if (!kvm_regs)
 944                         goto out;
 945                 r = -EFAULT;
 946                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
 947                         goto out_free2;
 948                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
 949                 if (r)
 950                         goto out_free2;
 951                 r = 0;
 952 out_free2:
 953                 kfree(kvm_regs);
 954                 break;
 955         }
 956         case KVM_GET_SREGS: {
 957                 struct kvm_sregs kvm_sregs;
 958
 959                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
 960                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
 961                 if (r)
 962                         goto out;
 963                 r = -EFAULT;
 964                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
 965                         goto out;
 966                 r = 0;
 967                 break;
 968         }
 969         case KVM_SET_SREGS: {
 970                 struct kvm_sregs kvm_sregs;
 971
 972                 r = -EFAULT;
 973                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
 974                         goto out;
 975                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
 976                 if (r)
 977                         goto out;
 978                 r = 0;
 979                 break;
 980         }
 981         case KVM_TRANSLATE: {
 982                 struct kvm_translation tr;
 983
 984                 r = -EFAULT;
 985                 if (copy_from_user(&tr, argp, sizeof tr))
 986                         goto out;
 987                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
 988                 if (r)
 989                         goto out;
 990                 r = -EFAULT;
 991                 if (copy_to_user(argp, &tr, sizeof tr))
 992                         goto out;
 993                 r = 0;
 994                 break;
 995         }
 996         case KVM_DEBUG_GUEST: {
 997                 struct kvm_debug_guest dbg;
 998
 999                 r = -EFAULT;
1000                 if (copy_from_user(&dbg, argp, sizeof dbg))
1001                         goto out;
1002                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1003                 if (r)
1004                         goto out;
1005                 r = 0;
1006                 break;
1007         }
1008         case KVM_SET_SIGNAL_MASK: {
1009                 struct kvm_signal_mask __user *sigmask_arg = argp;
1010                 struct kvm_signal_mask kvm_sigmask;
1011                 sigset_t sigset, *p;
1012
1013                 p = NULL;
1014                 if (argp) {
1015                         r = -EFAULT;
1016                         if (copy_from_user(&kvm_sigmask, argp,
1017                                            sizeof kvm_sigmask))
1018                                 goto out;
1019                         r = -EINVAL;
1020                         if (kvm_sigmask.len != sizeof sigset)
1021                                 goto out;
1022                         r = -EFAULT;
1023                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1024                                            sizeof sigset))
1025                                 goto out;
1026                         p = &sigset;
1027                 }
1028                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1029                 break;
1030         }
1031         case KVM_GET_FPU: {
1032                 struct kvm_fpu fpu;
1033
1034                 memset(&fpu, 0, sizeof fpu);
1035                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
1036                 if (r)
1037                         goto out;
1038                 r = -EFAULT;
1039                 if (copy_to_user(argp, &fpu, sizeof fpu))
1040                         goto out;
1041                 r = 0;
1042                 break;
1043         }
1044         case KVM_SET_FPU: {
1045                 struct kvm_fpu fpu;
1046
1047                 r = -EFAULT;
1048                 if (copy_from_user(&fpu, argp, sizeof fpu))
1049                         goto out;
1050                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
1051                 if (r)
1052                         goto out;
1053                 r = 0;
1054                 break;
1055         }
1056         default:
1057                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1058         }
1059 out:
1060         return r;
1061 }
1062
1063 static long kvm_vm_ioctl(struct file *filp,
1064                            unsigned int ioctl, unsigned long arg)
1065 {
1066         struct kvm *kvm = filp->private_data;
1067         void __user *argp = (void __user *)arg;
1068         int r;
1069
1070         if (kvm->mm != current->mm)
1071                 return -EIO;
1072         switch (ioctl) {
1073         case KVM_CREATE_VCPU:
1074                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1075                 if (r < 0)
1076                         goto out;
1077                 break;
1078         case KVM_SET_USER_MEMORY_REGION: {
1079                 struct kvm_userspace_memory_region kvm_userspace_mem;
1080
1081                 r = -EFAULT;
1082                 if (copy_from_user(&kvm_userspace_mem, argp,
1083                                                 sizeof kvm_userspace_mem))
1084                         goto out;
1085
1086                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1087                 if (r)
1088                         goto out;
1089                 break;
1090         }
1091         case KVM_GET_DIRTY_LOG: {
1092                 struct kvm_dirty_log log;
1093
1094                 r = -EFAULT;
1095                 if (copy_from_user(&log, argp, sizeof log))
1096                         goto out;
1097                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1098                 if (r)
1099                         goto out;
1100                 break;
1101         }
1102         default:
1103                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1104         }
1105 out:
1106         return r;
1107 }
1108
1109 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1110 {
1111         struct kvm *kvm = vma->vm_file->private_data;
1112         struct page *page;
1113
1114         if (!kvm_is_visible_gfn(kvm, vmf->pgoff))
1115                 return VM_FAULT_SIGBUS;
1116         page = gfn_to_page(kvm, vmf->pgoff);
1117         if (is_error_page(page)) {
1118                 kvm_release_page_clean(page);
1119                 return VM_FAULT_SIGBUS;
1120         }
1121         vmf->page = page;
1122         return 0;
1123 }
1124
1125 static struct vm_operations_struct kvm_vm_vm_ops = {
1126         .fault = kvm_vm_fault,
1127 };
1128
1129 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1130 {
1131         vma->vm_ops = &kvm_vm_vm_ops;
1132         return 0;
1133 }
1134
1135 static const struct file_operations kvm_vm_fops = {
1136         .release        = kvm_vm_release,
1137         .unlocked_ioctl = kvm_vm_ioctl,
1138         .compat_ioctl   = kvm_vm_ioctl,
1139         .mmap           = kvm_vm_mmap,
1140 };
1141
1142 static int kvm_dev_ioctl_create_vm(void)
1143 {
1144         int fd, r;
1145         struct inode *inode;
1146         struct file *file;
1147         struct kvm *kvm;
1148
1149         kvm = kvm_create_vm();
1150         if (IS_ERR(kvm))
1151                 return PTR_ERR(kvm);
1152         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1153         if (r) {
1154                 kvm_destroy_vm(kvm);
1155                 return r;
1156         }
1157
1158         kvm->filp = file;
1159
1160         return fd;
1161 }
1162
1163 static long kvm_dev_ioctl(struct file *filp,
1164                           unsigned int ioctl, unsigned long arg)
1165 {
1166         void __user *argp = (void __user *)arg;
1167         long r = -EINVAL;
1168
1169         switch (ioctl) {
1170         case KVM_GET_API_VERSION:
1171                 r = -EINVAL;
1172                 if (arg)
1173                         goto out;
1174                 r = KVM_API_VERSION;
1175                 break;
1176         case KVM_CREATE_VM:
1177                 r = -EINVAL;
1178                 if (arg)
1179                         goto out;
1180                 r = kvm_dev_ioctl_create_vm();
1181                 break;
1182         case KVM_CHECK_EXTENSION:
1183                 r = kvm_dev_ioctl_check_extension((long)argp);
1184                 break;
1185         case KVM_GET_VCPU_MMAP_SIZE:
1186                 r = -EINVAL;
1187                 if (arg)
1188                         goto out;
1189                 r = PAGE_SIZE;     /* struct kvm_run */
1190 #ifdef CONFIG_X86
1191                 r += PAGE_SIZE;    /* pio data page */
1192 #endif
1193                 break;
1194         case KVM_TRACE_ENABLE:
1195         case KVM_TRACE_PAUSE:
1196         case KVM_TRACE_DISABLE:
1197                 r = kvm_trace_ioctl(ioctl, arg);
1198                 break;
1199         default:
1200                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1201         }
1202 out:
1203         return r;
1204 }
1205
1206 static struct file_operations kvm_chardev_ops = {
1207         .unlocked_ioctl = kvm_dev_ioctl,
1208         .compat_ioctl   = kvm_dev_ioctl,
1209 };
1210
1211 static struct miscdevice kvm_dev = {
1212         KVM_MINOR,
1213         "kvm",
1214         &kvm_chardev_ops,
1215 };
1216
1217 static void hardware_enable(void *junk)
1218 {
1219         int cpu = raw_smp_processor_id();
1220
1221         if (cpu_isset(cpu, cpus_hardware_enabled))
1222                 return;
1223         cpu_set(cpu, cpus_hardware_enabled);
1224         kvm_arch_hardware_enable(NULL);
1225 }
1226
1227 static void hardware_disable(void *junk)
1228 {
1229         int cpu = raw_smp_processor_id();
1230
1231         if (!cpu_isset(cpu, cpus_hardware_enabled))
1232                 return;
1233         cpu_clear(cpu, cpus_hardware_enabled);
1234         decache_vcpus_on_cpu(cpu);
1235         kvm_arch_hardware_disable(NULL);
1236 }
1237
1238 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1239                            void *v)
1240 {
1241         int cpu = (long)v;
1242
1243         val &= ~CPU_TASKS_FROZEN;
1244         switch (val) {
1245         case CPU_DYING:
1246                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1247                        cpu);
1248                 hardware_disable(NULL);
1249                 break;
1250         case CPU_UP_CANCELED:
1251                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1252                        cpu);
1253                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1254                 break;
1255         case CPU_ONLINE:
1256                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1257                        cpu);
1258                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1259                 break;
1260         }
1261         return NOTIFY_OK;
1262 }
1263
1264 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1265                       void *v)
1266 {
1267         if (val == SYS_RESTART) {
1268                 /*
1269                  * Some (well, at least mine) BIOSes hang on reboot if
1270                  * in vmx root mode.
1271                  */
1272                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1273                 on_each_cpu(hardware_disable, NULL, 0, 1);
1274         }
1275         return NOTIFY_OK;
1276 }
1277
1278 static struct notifier_block kvm_reboot_notifier = {
1279         .notifier_call = kvm_reboot,
1280         .priority = 0,
1281 };
1282
1283 void kvm_io_bus_init(struct kvm_io_bus *bus)
1284 {
1285         memset(bus, 0, sizeof(*bus));
1286 }
1287
1288 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
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                 kvm_iodevice_destructor(pos);
1296         }
1297 }
1298
1299 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1300 {
1301         int i;
1302
1303         for (i = 0; i < bus->dev_count; i++) {
1304                 struct kvm_io_device *pos = bus->devs[i];
1305
1306                 if (pos->in_range(pos, addr))
1307                         return pos;
1308         }
1309
1310         return NULL;
1311 }
1312
1313 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1314 {
1315         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1316
1317         bus->devs[bus->dev_count++] = dev;
1318 }
1319
1320 static struct notifier_block kvm_cpu_notifier = {
1321         .notifier_call = kvm_cpu_hotplug,
1322         .priority = 20, /* must be > scheduler priority */
1323 };
1324
1325 static int vm_stat_get(void *_offset, u64 *val)
1326 {
1327         unsigned offset = (long)_offset;
1328         struct kvm *kvm;
1329
1330         *val = 0;
1331         spin_lock(&kvm_lock);
1332         list_for_each_entry(kvm, &vm_list, vm_list)
1333                 *val += *(u32 *)((void *)kvm + offset);
1334         spin_unlock(&kvm_lock);
1335         return 0;
1336 }
1337
1338 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1339
1340 static int vcpu_stat_get(void *_offset, u64 *val)
1341 {
1342         unsigned offset = (long)_offset;
1343         struct kvm *kvm;
1344         struct kvm_vcpu *vcpu;
1345         int i;
1346
1347         *val = 0;
1348         spin_lock(&kvm_lock);
1349         list_for_each_entry(kvm, &vm_list, vm_list)
1350                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1351                         vcpu = kvm->vcpus[i];
1352                         if (vcpu)
1353                                 *val += *(u32 *)((void *)vcpu + offset);
1354                 }
1355         spin_unlock(&kvm_lock);
1356         return 0;
1357 }
1358
1359 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1360
1361 static struct file_operations *stat_fops[] = {
1362         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1363         [KVM_STAT_VM]   = &vm_stat_fops,
1364 };
1365
1366 static void kvm_init_debug(void)
1367 {
1368         struct kvm_stats_debugfs_item *p;
1369
1370         debugfs_dir = debugfs_create_dir("kvm", NULL);
1371         for (p = debugfs_entries; p->name; ++p)
1372                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1373                                                 (void *)(long)p->offset,
1374                                                 stat_fops[p->kind]);
1375 }
1376
1377 static void kvm_exit_debug(void)
1378 {
1379         struct kvm_stats_debugfs_item *p;
1380
1381         for (p = debugfs_entries; p->name; ++p)
1382                 debugfs_remove(p->dentry);
1383         debugfs_remove(debugfs_dir);
1384 }
1385
1386 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1387 {
1388         hardware_disable(NULL);
1389         return 0;
1390 }
1391
1392 static int kvm_resume(struct sys_device *dev)
1393 {
1394         hardware_enable(NULL);
1395         return 0;
1396 }
1397
1398 static struct sysdev_class kvm_sysdev_class = {
1399         .name = "kvm",
1400         .suspend = kvm_suspend,
1401         .resume = kvm_resume,
1402 };
1403
1404 static struct sys_device kvm_sysdev = {
1405         .id = 0,
1406         .cls = &kvm_sysdev_class,
1407 };
1408
1409 struct page *bad_page;
1410 pfn_t bad_pfn;
1411
1412 static inline
1413 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1414 {
1415         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1416 }
1417
1418 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1419 {
1420         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1421
1422         kvm_arch_vcpu_load(vcpu, cpu);
1423 }
1424
1425 static void kvm_sched_out(struct preempt_notifier *pn,
1426                           struct task_struct *next)
1427 {
1428         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1429
1430         kvm_arch_vcpu_put(vcpu);
1431 }
1432
1433 int kvm_init(void *opaque, unsigned int vcpu_size,
1434                   struct module *module)
1435 {
1436         int r;
1437         int cpu;
1438
1439         kvm_init_debug();
1440
1441         r = kvm_arch_init(opaque);
1442         if (r)
1443                 goto out_fail;
1444
1445         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1446
1447         if (bad_page == NULL) {
1448                 r = -ENOMEM;
1449                 goto out;
1450         }
1451
1452         bad_pfn = page_to_pfn(bad_page);
1453
1454         r = kvm_arch_hardware_setup();
1455         if (r < 0)
1456                 goto out_free_0;
1457
1458         for_each_online_cpu(cpu) {
1459                 smp_call_function_single(cpu,
1460                                 kvm_arch_check_processor_compat,
1461                                 &r, 0, 1);
1462                 if (r < 0)
1463                         goto out_free_1;
1464         }
1465
1466         on_each_cpu(hardware_enable, NULL, 0, 1);
1467         r = register_cpu_notifier(&kvm_cpu_notifier);
1468         if (r)
1469                 goto out_free_2;
1470         register_reboot_notifier(&kvm_reboot_notifier);
1471
1472         r = sysdev_class_register(&kvm_sysdev_class);
1473         if (r)
1474                 goto out_free_3;
1475
1476         r = sysdev_register(&kvm_sysdev);
1477         if (r)
1478                 goto out_free_4;
1479
1480         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1481         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1482                                            __alignof__(struct kvm_vcpu),
1483                                            0, NULL);
1484         if (!kvm_vcpu_cache) {
1485                 r = -ENOMEM;
1486                 goto out_free_5;
1487         }
1488
1489         kvm_chardev_ops.owner = module;
1490
1491         r = misc_register(&kvm_dev);
1492         if (r) {
1493                 printk(KERN_ERR "kvm: misc device register failed\n");
1494                 goto out_free;
1495         }
1496
1497         kvm_preempt_ops.sched_in = kvm_sched_in;
1498         kvm_preempt_ops.sched_out = kvm_sched_out;
1499
1500         return 0;
1501
1502 out_free:
1503         kmem_cache_destroy(kvm_vcpu_cache);
1504 out_free_5:
1505         sysdev_unregister(&kvm_sysdev);
1506 out_free_4:
1507         sysdev_class_unregister(&kvm_sysdev_class);
1508 out_free_3:
1509         unregister_reboot_notifier(&kvm_reboot_notifier);
1510         unregister_cpu_notifier(&kvm_cpu_notifier);
1511 out_free_2:
1512         on_each_cpu(hardware_disable, NULL, 0, 1);
1513 out_free_1:
1514         kvm_arch_hardware_unsetup();
1515 out_free_0:
1516         __free_page(bad_page);
1517 out:
1518         kvm_arch_exit();
1519         kvm_exit_debug();
1520 out_fail:
1521         return r;
1522 }
1523 EXPORT_SYMBOL_GPL(kvm_init);
1524
1525 void kvm_exit(void)
1526 {
1527         kvm_trace_cleanup();
1528         misc_deregister(&kvm_dev);
1529         kmem_cache_destroy(kvm_vcpu_cache);
1530         sysdev_unregister(&kvm_sysdev);
1531         sysdev_class_unregister(&kvm_sysdev_class);
1532         unregister_reboot_notifier(&kvm_reboot_notifier);
1533         unregister_cpu_notifier(&kvm_cpu_notifier);
1534         on_each_cpu(hardware_disable, NULL, 0, 1);
1535         kvm_arch_hardware_unsetup();
1536         kvm_arch_exit();
1537         kvm_exit_debug();
1538         __free_page(bad_page);
1539 }
1540 EXPORT_SYMBOL_GPL(kvm_exit);