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 *kvm_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                && !kvm_cpu_has_pending_timer(vcpu)
 769                && !signal_pending(current)
 770                && !kvm_arch_vcpu_runnable(vcpu)) {
 771                 set_current_state(TASK_INTERRUPTIBLE);
 772                 vcpu_put(vcpu);
 773                 schedule();
 774                 vcpu_load(vcpu);
 775         }
 776
 777         __set_current_state(TASK_RUNNING);
 778         remove_wait_queue(&vcpu->wq, &wait);
 779 }
 780
 781 void kvm_resched(struct kvm_vcpu *vcpu)
 782 {
 783         if (!need_resched())
 784                 return;
 785         cond_resched();
 786 }
 787 EXPORT_SYMBOL_GPL(kvm_resched);
 788
 789 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 790 {
 791         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
 792         struct page *page;
 793
 794         if (vmf->pgoff == 0)
 795                 page = virt_to_page(vcpu->run);
 796 #ifdef CONFIG_X86
 797         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
 798                 page = virt_to_page(vcpu->arch.pio_data);
 799 #endif
 800         else
 801                 return VM_FAULT_SIGBUS;
 802         get_page(page);
 803         vmf->page = page;
 804         return 0;
 805 }
 806
 807 static struct vm_operations_struct kvm_vcpu_vm_ops = {
 808         .fault = kvm_vcpu_fault,
 809 };
 810
 811 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
 812 {
 813         vma->vm_ops = &kvm_vcpu_vm_ops;
 814         return 0;
 815 }
 816
 817 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
 818 {
 819         struct kvm_vcpu *vcpu = filp->private_data;
 820
 821         kvm_put_kvm(vcpu->kvm);
 822         return 0;
 823 }
 824
 825 static const struct file_operations kvm_vcpu_fops = {
 826         .release        = kvm_vcpu_release,
 827         .unlocked_ioctl = kvm_vcpu_ioctl,
 828         .compat_ioctl   = kvm_vcpu_ioctl,
 829         .mmap           = kvm_vcpu_mmap,
 830 };
 831
 832 /*
 833  * Allocates an inode for the vcpu.
 834  */
 835 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
 836 {
 837         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu);
 838         if (fd < 0)
 839                 kvm_put_kvm(vcpu->kvm);
 840         return fd;
 841 }
 842
 843 /*
 844  * Creates some virtual cpus.  Good luck creating more than one.
 845  */
 846 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
 847 {
 848         int r;
 849         struct kvm_vcpu *vcpu;
 850
 851         if (!valid_vcpu(n))
 852                 return -EINVAL;
 853
 854         vcpu = kvm_arch_vcpu_create(kvm, n);
 855         if (IS_ERR(vcpu))
 856                 return PTR_ERR(vcpu);
 857
 858         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
 859
 860         r = kvm_arch_vcpu_setup(vcpu);
 861         if (r)
 862                 goto vcpu_destroy;
 863
 864         mutex_lock(&kvm->lock);
 865         if (kvm->vcpus[n]) {
 866                 r = -EEXIST;
 867                 mutex_unlock(&kvm->lock);
 868                 goto vcpu_destroy;
 869         }
 870         kvm->vcpus[n] = vcpu;
 871         mutex_unlock(&kvm->lock);
 872
 873         /* Now it's all set up, let userspace reach it */
 874         kvm_get_kvm(kvm);
 875         r = create_vcpu_fd(vcpu);
 876         if (r < 0)
 877                 goto unlink;
 878         return r;
 879
 880 unlink:
 881         mutex_lock(&kvm->lock);
 882         kvm->vcpus[n] = NULL;
 883         mutex_unlock(&kvm->lock);
 884 vcpu_destroy:
 885         kvm_arch_vcpu_destroy(vcpu);
 886         return r;
 887 }
 888
 889 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
 890 {
 891         if (sigset) {
 892                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
 893                 vcpu->sigset_active = 1;
 894                 vcpu->sigset = *sigset;
 895         } else
 896                 vcpu->sigset_active = 0;
 897         return 0;
 898 }
 899
 900 static long kvm_vcpu_ioctl(struct file *filp,
 901                            unsigned int ioctl, unsigned long arg)
 902 {
 903         struct kvm_vcpu *vcpu = filp->private_data;
 904         void __user *argp = (void __user *)arg;
 905         int r;
 906
 907         if (vcpu->kvm->mm != current->mm)
 908                 return -EIO;
 909         switch (ioctl) {
 910         case KVM_RUN:
 911                 r = -EINVAL;
 912                 if (arg)
 913                         goto out;
 914                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
 915                 break;
 916         case KVM_GET_REGS: {
 917                 struct kvm_regs *kvm_regs;
 918
 919                 r = -ENOMEM;
 920                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
 921                 if (!kvm_regs)
 922                         goto out;
 923                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
 924                 if (r)
 925                         goto out_free1;
 926                 r = -EFAULT;
 927                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
 928                         goto out_free1;
 929                 r = 0;
 930 out_free1:
 931                 kfree(kvm_regs);
 932                 break;
 933         }
 934         case KVM_SET_REGS: {
 935                 struct kvm_regs *kvm_regs;
 936
 937                 r = -ENOMEM;
 938                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
 939                 if (!kvm_regs)
 940                         goto out;
 941                 r = -EFAULT;
 942                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
 943                         goto out_free2;
 944                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
 945                 if (r)
 946                         goto out_free2;
 947                 r = 0;
 948 out_free2:
 949                 kfree(kvm_regs);
 950                 break;
 951         }
 952         case KVM_GET_SREGS: {
 953                 struct kvm_sregs kvm_sregs;
 954
 955                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
 956                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
 957                 if (r)
 958                         goto out;
 959                 r = -EFAULT;
 960                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
 961                         goto out;
 962                 r = 0;
 963                 break;
 964         }
 965         case KVM_SET_SREGS: {
 966                 struct kvm_sregs kvm_sregs;
 967
 968                 r = -EFAULT;
 969                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
 970                         goto out;
 971                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
 972                 if (r)
 973                         goto out;
 974                 r = 0;
 975                 break;
 976         }
 977         case KVM_GET_MP_STATE: {
 978                 struct kvm_mp_state mp_state;
 979
 980                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
 981                 if (r)
 982                         goto out;
 983                 r = -EFAULT;
 984                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
 985                         goto out;
 986                 r = 0;
 987                 break;
 988         }
 989         case KVM_SET_MP_STATE: {
 990                 struct kvm_mp_state mp_state;
 991
 992                 r = -EFAULT;
 993                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
 994                         goto out;
 995                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
 996                 if (r)
 997                         goto out;
 998                 r = 0;
 999                 break;
1000         }
1001         case KVM_TRANSLATE: {
1002                 struct kvm_translation tr;
1003
1004                 r = -EFAULT;
1005                 if (copy_from_user(&tr, argp, sizeof tr))
1006                         goto out;
1007                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1008                 if (r)
1009                         goto out;
1010                 r = -EFAULT;
1011                 if (copy_to_user(argp, &tr, sizeof tr))
1012                         goto out;
1013                 r = 0;
1014                 break;
1015         }
1016         case KVM_DEBUG_GUEST: {
1017                 struct kvm_debug_guest dbg;
1018
1019                 r = -EFAULT;
1020                 if (copy_from_user(&dbg, argp, sizeof dbg))
1021                         goto out;
1022                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1023                 if (r)
1024                         goto out;
1025                 r = 0;
1026                 break;
1027         }
1028         case KVM_SET_SIGNAL_MASK: {
1029                 struct kvm_signal_mask __user *sigmask_arg = argp;
1030                 struct kvm_signal_mask kvm_sigmask;
1031                 sigset_t sigset, *p;
1032
1033                 p = NULL;
1034                 if (argp) {
1035                         r = -EFAULT;
1036                         if (copy_from_user(&kvm_sigmask, argp,
1037                                            sizeof kvm_sigmask))
1038                                 goto out;
1039                         r = -EINVAL;
1040                         if (kvm_sigmask.len != sizeof sigset)
1041                                 goto out;
1042                         r = -EFAULT;
1043                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1044                                            sizeof sigset))
1045                                 goto out;
1046                         p = &sigset;
1047                 }
1048                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1049                 break;
1050         }
1051         case KVM_GET_FPU: {
1052                 struct kvm_fpu fpu;
1053
1054                 memset(&fpu, 0, sizeof fpu);
1055                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
1056                 if (r)
1057                         goto out;
1058                 r = -EFAULT;
1059                 if (copy_to_user(argp, &fpu, sizeof fpu))
1060                         goto out;
1061                 r = 0;
1062                 break;
1063         }
1064         case KVM_SET_FPU: {
1065                 struct kvm_fpu fpu;
1066
1067                 r = -EFAULT;
1068                 if (copy_from_user(&fpu, argp, sizeof fpu))
1069                         goto out;
1070                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
1071                 if (r)
1072                         goto out;
1073                 r = 0;
1074                 break;
1075         }
1076         default:
1077                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1078         }
1079 out:
1080         return r;
1081 }
1082
1083 static long kvm_vm_ioctl(struct file *filp,
1084                            unsigned int ioctl, unsigned long arg)
1085 {
1086         struct kvm *kvm = filp->private_data;
1087         void __user *argp = (void __user *)arg;
1088         int r;
1089
1090         if (kvm->mm != current->mm)
1091                 return -EIO;
1092         switch (ioctl) {
1093         case KVM_CREATE_VCPU:
1094                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1095                 if (r < 0)
1096                         goto out;
1097                 break;
1098         case KVM_SET_USER_MEMORY_REGION: {
1099                 struct kvm_userspace_memory_region kvm_userspace_mem;
1100
1101                 r = -EFAULT;
1102                 if (copy_from_user(&kvm_userspace_mem, argp,
1103                                                 sizeof kvm_userspace_mem))
1104                         goto out;
1105
1106                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1107                 if (r)
1108                         goto out;
1109                 break;
1110         }
1111         case KVM_GET_DIRTY_LOG: {
1112                 struct kvm_dirty_log log;
1113
1114                 r = -EFAULT;
1115                 if (copy_from_user(&log, argp, sizeof log))
1116                         goto out;
1117                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1118                 if (r)
1119                         goto out;
1120                 break;
1121         }
1122         default:
1123                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1124         }
1125 out:
1126         return r;
1127 }
1128
1129 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1130 {
1131         struct kvm *kvm = vma->vm_file->private_data;
1132         struct page *page;
1133
1134         if (!kvm_is_visible_gfn(kvm, vmf->pgoff))
1135                 return VM_FAULT_SIGBUS;
1136         page = gfn_to_page(kvm, vmf->pgoff);
1137         if (is_error_page(page)) {
1138                 kvm_release_page_clean(page);
1139                 return VM_FAULT_SIGBUS;
1140         }
1141         vmf->page = page;
1142         return 0;
1143 }
1144
1145 static struct vm_operations_struct kvm_vm_vm_ops = {
1146         .fault = kvm_vm_fault,
1147 };
1148
1149 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1150 {
1151         vma->vm_ops = &kvm_vm_vm_ops;
1152         return 0;
1153 }
1154
1155 static const struct file_operations kvm_vm_fops = {
1156         .release        = kvm_vm_release,
1157         .unlocked_ioctl = kvm_vm_ioctl,
1158         .compat_ioctl   = kvm_vm_ioctl,
1159         .mmap           = kvm_vm_mmap,
1160 };
1161
1162 static int kvm_dev_ioctl_create_vm(void)
1163 {
1164         int fd;
1165         struct kvm *kvm;
1166
1167         kvm = kvm_create_vm();
1168         if (IS_ERR(kvm))
1169                 return PTR_ERR(kvm);
1170         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm);
1171         if (fd < 0)
1172                 kvm_put_kvm(kvm);
1173
1174         return fd;
1175 }
1176
1177 static long kvm_dev_ioctl(struct file *filp,
1178                           unsigned int ioctl, unsigned long arg)
1179 {
1180         void __user *argp = (void __user *)arg;
1181         long r = -EINVAL;
1182
1183         switch (ioctl) {
1184         case KVM_GET_API_VERSION:
1185                 r = -EINVAL;
1186                 if (arg)
1187                         goto out;
1188                 r = KVM_API_VERSION;
1189                 break;
1190         case KVM_CREATE_VM:
1191                 r = -EINVAL;
1192                 if (arg)
1193                         goto out;
1194                 r = kvm_dev_ioctl_create_vm();
1195                 break;
1196         case KVM_CHECK_EXTENSION:
1197                 r = kvm_dev_ioctl_check_extension((long)argp);
1198                 break;
1199         case KVM_GET_VCPU_MMAP_SIZE:
1200                 r = -EINVAL;
1201                 if (arg)
1202                         goto out;
1203                 r = PAGE_SIZE;     /* struct kvm_run */
1204 #ifdef CONFIG_X86
1205                 r += PAGE_SIZE;    /* pio data page */
1206 #endif
1207                 break;
1208         case KVM_TRACE_ENABLE:
1209         case KVM_TRACE_PAUSE:
1210         case KVM_TRACE_DISABLE:
1211                 r = kvm_trace_ioctl(ioctl, arg);
1212                 break;
1213         default:
1214                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1215         }
1216 out:
1217         return r;
1218 }
1219
1220 static struct file_operations kvm_chardev_ops = {
1221         .unlocked_ioctl = kvm_dev_ioctl,
1222         .compat_ioctl   = kvm_dev_ioctl,
1223 };
1224
1225 static struct miscdevice kvm_dev = {
1226         KVM_MINOR,
1227         "kvm",
1228         &kvm_chardev_ops,
1229 };
1230
1231 static void hardware_enable(void *junk)
1232 {
1233         int cpu = raw_smp_processor_id();
1234
1235         if (cpu_isset(cpu, cpus_hardware_enabled))
1236                 return;
1237         cpu_set(cpu, cpus_hardware_enabled);
1238         kvm_arch_hardware_enable(NULL);
1239 }
1240
1241 static void hardware_disable(void *junk)
1242 {
1243         int cpu = raw_smp_processor_id();
1244
1245         if (!cpu_isset(cpu, cpus_hardware_enabled))
1246                 return;
1247         cpu_clear(cpu, cpus_hardware_enabled);
1248         decache_vcpus_on_cpu(cpu);
1249         kvm_arch_hardware_disable(NULL);
1250 }
1251
1252 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1253                            void *v)
1254 {
1255         int cpu = (long)v;
1256
1257         val &= ~CPU_TASKS_FROZEN;
1258         switch (val) {
1259         case CPU_DYING:
1260                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1261                        cpu);
1262                 hardware_disable(NULL);
1263                 break;
1264         case CPU_UP_CANCELED:
1265                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1266                        cpu);
1267                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1268                 break;
1269         case CPU_ONLINE:
1270                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1271                        cpu);
1272                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1273                 break;
1274         }
1275         return NOTIFY_OK;
1276 }
1277
1278 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1279                       void *v)
1280 {
1281         if (val == SYS_RESTART) {
1282                 /*
1283                  * Some (well, at least mine) BIOSes hang on reboot if
1284                  * in vmx root mode.
1285                  */
1286                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1287                 on_each_cpu(hardware_disable, NULL, 0, 1);
1288         }
1289         return NOTIFY_OK;
1290 }
1291
1292 static struct notifier_block kvm_reboot_notifier = {
1293         .notifier_call = kvm_reboot,
1294         .priority = 0,
1295 };
1296
1297 void kvm_io_bus_init(struct kvm_io_bus *bus)
1298 {
1299         memset(bus, 0, sizeof(*bus));
1300 }
1301
1302 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1303 {
1304         int i;
1305
1306         for (i = 0; i < bus->dev_count; i++) {
1307                 struct kvm_io_device *pos = bus->devs[i];
1308
1309                 kvm_iodevice_destructor(pos);
1310         }
1311 }
1312
1313 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1314 {
1315         int i;
1316
1317         for (i = 0; i < bus->dev_count; i++) {
1318                 struct kvm_io_device *pos = bus->devs[i];
1319
1320                 if (pos->in_range(pos, addr))
1321                         return pos;
1322         }
1323
1324         return NULL;
1325 }
1326
1327 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1328 {
1329         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1330
1331         bus->devs[bus->dev_count++] = dev;
1332 }
1333
1334 static struct notifier_block kvm_cpu_notifier = {
1335         .notifier_call = kvm_cpu_hotplug,
1336         .priority = 20, /* must be > scheduler priority */
1337 };
1338
1339 static int vm_stat_get(void *_offset, u64 *val)
1340 {
1341         unsigned offset = (long)_offset;
1342         struct kvm *kvm;
1343
1344         *val = 0;
1345         spin_lock(&kvm_lock);
1346         list_for_each_entry(kvm, &vm_list, vm_list)
1347                 *val += *(u32 *)((void *)kvm + offset);
1348         spin_unlock(&kvm_lock);
1349         return 0;
1350 }
1351
1352 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1353
1354 static int vcpu_stat_get(void *_offset, u64 *val)
1355 {
1356         unsigned offset = (long)_offset;
1357         struct kvm *kvm;
1358         struct kvm_vcpu *vcpu;
1359         int i;
1360
1361         *val = 0;
1362         spin_lock(&kvm_lock);
1363         list_for_each_entry(kvm, &vm_list, vm_list)
1364                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1365                         vcpu = kvm->vcpus[i];
1366                         if (vcpu)
1367                                 *val += *(u32 *)((void *)vcpu + offset);
1368                 }
1369         spin_unlock(&kvm_lock);
1370         return 0;
1371 }
1372
1373 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1374
1375 static struct file_operations *stat_fops[] = {
1376         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1377         [KVM_STAT_VM]   = &vm_stat_fops,
1378 };
1379
1380 static void kvm_init_debug(void)
1381 {
1382         struct kvm_stats_debugfs_item *p;
1383
1384         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1385         for (p = debugfs_entries; p->name; ++p)
1386                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1387                                                 (void *)(long)p->offset,
1388                                                 stat_fops[p->kind]);
1389 }
1390
1391 static void kvm_exit_debug(void)
1392 {
1393         struct kvm_stats_debugfs_item *p;
1394
1395         for (p = debugfs_entries; p->name; ++p)
1396                 debugfs_remove(p->dentry);
1397         debugfs_remove(kvm_debugfs_dir);
1398 }
1399
1400 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1401 {
1402         hardware_disable(NULL);
1403         return 0;
1404 }
1405
1406 static int kvm_resume(struct sys_device *dev)
1407 {
1408         hardware_enable(NULL);
1409         return 0;
1410 }
1411
1412 static struct sysdev_class kvm_sysdev_class = {
1413         .name = "kvm",
1414         .suspend = kvm_suspend,
1415         .resume = kvm_resume,
1416 };
1417
1418 static struct sys_device kvm_sysdev = {
1419         .id = 0,
1420         .cls = &kvm_sysdev_class,
1421 };
1422
1423 struct page *bad_page;
1424 pfn_t bad_pfn;
1425
1426 static inline
1427 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1428 {
1429         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1430 }
1431
1432 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1433 {
1434         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1435
1436         kvm_arch_vcpu_load(vcpu, cpu);
1437 }
1438
1439 static void kvm_sched_out(struct preempt_notifier *pn,
1440                           struct task_struct *next)
1441 {
1442         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1443
1444         kvm_arch_vcpu_put(vcpu);
1445 }
1446
1447 int kvm_init(void *opaque, unsigned int vcpu_size,
1448                   struct module *module)
1449 {
1450         int r;
1451         int cpu;
1452
1453         kvm_init_debug();
1454
1455         r = kvm_arch_init(opaque);
1456         if (r)
1457                 goto out_fail;
1458
1459         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1460
1461         if (bad_page == NULL) {
1462                 r = -ENOMEM;
1463                 goto out;
1464         }
1465
1466         bad_pfn = page_to_pfn(bad_page);
1467
1468         r = kvm_arch_hardware_setup();
1469         if (r < 0)
1470                 goto out_free_0;
1471
1472         for_each_online_cpu(cpu) {
1473                 smp_call_function_single(cpu,
1474                                 kvm_arch_check_processor_compat,
1475                                 &r, 0, 1);
1476                 if (r < 0)
1477                         goto out_free_1;
1478         }
1479
1480         on_each_cpu(hardware_enable, NULL, 0, 1);
1481         r = register_cpu_notifier(&kvm_cpu_notifier);
1482         if (r)
1483                 goto out_free_2;
1484         register_reboot_notifier(&kvm_reboot_notifier);
1485
1486         r = sysdev_class_register(&kvm_sysdev_class);
1487         if (r)
1488                 goto out_free_3;
1489
1490         r = sysdev_register(&kvm_sysdev);
1491         if (r)
1492                 goto out_free_4;
1493
1494         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1495         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1496                                            __alignof__(struct kvm_vcpu),
1497                                            0, NULL);
1498         if (!kvm_vcpu_cache) {
1499                 r = -ENOMEM;
1500                 goto out_free_5;
1501         }
1502
1503         kvm_chardev_ops.owner = module;
1504
1505         r = misc_register(&kvm_dev);
1506         if (r) {
1507                 printk(KERN_ERR "kvm: misc device register failed\n");
1508                 goto out_free;
1509         }
1510
1511         kvm_preempt_ops.sched_in = kvm_sched_in;
1512         kvm_preempt_ops.sched_out = kvm_sched_out;
1513
1514         return 0;
1515
1516 out_free:
1517         kmem_cache_destroy(kvm_vcpu_cache);
1518 out_free_5:
1519         sysdev_unregister(&kvm_sysdev);
1520 out_free_4:
1521         sysdev_class_unregister(&kvm_sysdev_class);
1522 out_free_3:
1523         unregister_reboot_notifier(&kvm_reboot_notifier);
1524         unregister_cpu_notifier(&kvm_cpu_notifier);
1525 out_free_2:
1526         on_each_cpu(hardware_disable, NULL, 0, 1);
1527 out_free_1:
1528         kvm_arch_hardware_unsetup();
1529 out_free_0:
1530         __free_page(bad_page);
1531 out:
1532         kvm_arch_exit();
1533         kvm_exit_debug();
1534 out_fail:
1535         return r;
1536 }
1537 EXPORT_SYMBOL_GPL(kvm_init);
1538
1539 void kvm_exit(void)
1540 {
1541         kvm_trace_cleanup();
1542         misc_deregister(&kvm_dev);
1543         kmem_cache_destroy(kvm_vcpu_cache);
1544         sysdev_unregister(&kvm_sysdev);
1545         sysdev_class_unregister(&kvm_sysdev_class);
1546         unregister_reboot_notifier(&kvm_reboot_notifier);
1547         unregister_cpu_notifier(&kvm_cpu_notifier);
1548         on_each_cpu(hardware_disable, NULL, 0, 1);
1549         kvm_arch_hardware_unsetup();
1550         kvm_arch_exit();
1551         kvm_exit_debug();
1552         __free_page(bad_page);
1553 }
1554 EXPORT_SYMBOL_GPL(kvm_exit);