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  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.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/syscore_ops.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 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/pgtable.h>
  57
  58 #include "coalesced_mmio.h"
  59 #include "async_pf.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/kvm.h>
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 /*
  68  * Ordering of locks:
  69  *
  70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  71  */
  72
  73 DEFINE_RAW_SPINLOCK(kvm_lock);
  74 LIST_HEAD(vm_list);
  75
  76 static cpumask_var_t cpus_hardware_enabled;
  77 static int kvm_usage_count = 0;
  78 static atomic_t hardware_enable_failed;
  79
  80 struct kmem_cache *kvm_vcpu_cache;
  81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  82
  83 static __read_mostly struct preempt_ops kvm_preempt_ops;
  84
  85 struct dentry *kvm_debugfs_dir;
  86
  87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  88                            unsigned long arg);
  89 #ifdef CONFIG_COMPAT
  90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  91                                   unsigned long arg);
  92 #endif
  93 static int hardware_enable_all(void);
  94 static void hardware_disable_all(void);
  95
  96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  97
  98 bool kvm_rebooting;
  99 EXPORT_SYMBOL_GPL(kvm_rebooting);
 100
 101 static bool largepages_enabled = true;
 102
 103 static struct page *hwpoison_page;
 104 static pfn_t hwpoison_pfn;
 105
 106 struct page *fault_page;
 107 pfn_t fault_pfn;
 108
 109 inline int kvm_is_mmio_pfn(pfn_t pfn)
 110 {
 111         if (pfn_valid(pfn)) {
 112                 int reserved;
 113                 struct page *tail = pfn_to_page(pfn);
 114                 struct page *head = compound_trans_head(tail);
 115                 reserved = PageReserved(head);
 116                 if (head != tail) {
 117                         /*
 118                          * "head" is not a dangling pointer
 119                          * (compound_trans_head takes care of that)
 120                          * but the hugepage may have been splitted
 121                          * from under us (and we may not hold a
 122                          * reference count on the head page so it can
 123                          * be reused before we run PageReferenced), so
 124                          * we've to check PageTail before returning
 125                          * what we just read.
 126                          */
 127                         smp_rmb();
 128                         if (PageTail(tail))
 129                                 return reserved;
 130                 }
 131                 return PageReserved(tail);
 132         }
 133
 134         return true;
 135 }
 136
 137 /*
 138  * Switches to specified vcpu, until a matching vcpu_put()
 139  */
 140 void vcpu_load(struct kvm_vcpu *vcpu)
 141 {
 142         int cpu;
 143
 144         mutex_lock(&vcpu->mutex);
 145         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 146                 /* The thread running this VCPU changed. */
 147                 struct pid *oldpid = vcpu->pid;
 148                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 149                 rcu_assign_pointer(vcpu->pid, newpid);
 150                 synchronize_rcu();
 151                 put_pid(oldpid);
 152         }
 153         cpu = get_cpu();
 154         preempt_notifier_register(&vcpu->preempt_notifier);
 155         kvm_arch_vcpu_load(vcpu, cpu);
 156         put_cpu();
 157 }
 158
 159 void vcpu_put(struct kvm_vcpu *vcpu)
 160 {
 161         preempt_disable();
 162         kvm_arch_vcpu_put(vcpu);
 163         preempt_notifier_unregister(&vcpu->preempt_notifier);
 164         preempt_enable();
 165         mutex_unlock(&vcpu->mutex);
 166 }
 167
 168 static void ack_flush(void *_completed)
 169 {
 170 }
 171
 172 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 173 {
 174         int i, cpu, me;
 175         cpumask_var_t cpus;
 176         bool called = true;
 177         struct kvm_vcpu *vcpu;
 178
 179         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 180
 181         me = get_cpu();
 182         kvm_for_each_vcpu(i, vcpu, kvm) {
 183                 kvm_make_request(req, vcpu);
 184                 cpu = vcpu->cpu;
 185
 186                 /* Set ->requests bit before we read ->mode */
 187                 smp_mb();
 188
 189                 if (cpus != NULL && cpu != -1 && cpu != me &&
 190                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 191                         cpumask_set_cpu(cpu, cpus);
 192         }
 193         if (unlikely(cpus == NULL))
 194                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 195         else if (!cpumask_empty(cpus))
 196                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 197         else
 198                 called = false;
 199         put_cpu();
 200         free_cpumask_var(cpus);
 201         return called;
 202 }
 203
 204 void kvm_flush_remote_tlbs(struct kvm *kvm)
 205 {
 206         long dirty_count = kvm->tlbs_dirty;
 207
 208         smp_mb();
 209         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 210                 ++kvm->stat.remote_tlb_flush;
 211         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 212 }
 213
 214 void kvm_reload_remote_mmus(struct kvm *kvm)
 215 {
 216         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 217 }
 218
 219 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 220 {
 221         struct page *page;
 222         int r;
 223
 224         mutex_init(&vcpu->mutex);
 225         vcpu->cpu = -1;
 226         vcpu->kvm = kvm;
 227         vcpu->vcpu_id = id;
 228         vcpu->pid = NULL;
 229         init_waitqueue_head(&vcpu->wq);
 230         kvm_async_pf_vcpu_init(vcpu);
 231
 232         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 233         if (!page) {
 234                 r = -ENOMEM;
 235                 goto fail;
 236         }
 237         vcpu->run = page_address(page);
 238
 239         r = kvm_arch_vcpu_init(vcpu);
 240         if (r < 0)
 241                 goto fail_free_run;
 242         return 0;
 243
 244 fail_free_run:
 245         free_page((unsigned long)vcpu->run);
 246 fail:
 247         return r;
 248 }
 249 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 250
 251 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 252 {
 253         put_pid(vcpu->pid);
 254         kvm_arch_vcpu_uninit(vcpu);
 255         free_page((unsigned long)vcpu->run);
 256 }
 257 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 258
 259 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 260 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 261 {
 262         return container_of(mn, struct kvm, mmu_notifier);
 263 }
 264
 265 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 266                                              struct mm_struct *mm,
 267                                              unsigned long address)
 268 {
 269         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 270         int need_tlb_flush, idx;
 271
 272         /*
 273          * When ->invalidate_page runs, the linux pte has been zapped
 274          * already but the page is still allocated until
 275          * ->invalidate_page returns. So if we increase the sequence
 276          * here the kvm page fault will notice if the spte can't be
 277          * established because the page is going to be freed. If
 278          * instead the kvm page fault establishes the spte before
 279          * ->invalidate_page runs, kvm_unmap_hva will release it
 280          * before returning.
 281          *
 282          * The sequence increase only need to be seen at spin_unlock
 283          * time, and not at spin_lock time.
 284          *
 285          * Increasing the sequence after the spin_unlock would be
 286          * unsafe because the kvm page fault could then establish the
 287          * pte after kvm_unmap_hva returned, without noticing the page
 288          * is going to be freed.
 289          */
 290         idx = srcu_read_lock(&kvm->srcu);
 291         spin_lock(&kvm->mmu_lock);
 292
 293         kvm->mmu_notifier_seq++;
 294         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 295         /* we've to flush the tlb before the pages can be freed */
 296         if (need_tlb_flush)
 297                 kvm_flush_remote_tlbs(kvm);
 298
 299         spin_unlock(&kvm->mmu_lock);
 300         srcu_read_unlock(&kvm->srcu, idx);
 301 }
 302
 303 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 304                                         struct mm_struct *mm,
 305                                         unsigned long address,
 306                                         pte_t pte)
 307 {
 308         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 309         int idx;
 310
 311         idx = srcu_read_lock(&kvm->srcu);
 312         spin_lock(&kvm->mmu_lock);
 313         kvm->mmu_notifier_seq++;
 314         kvm_set_spte_hva(kvm, address, pte);
 315         spin_unlock(&kvm->mmu_lock);
 316         srcu_read_unlock(&kvm->srcu, idx);
 317 }
 318
 319 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 320                                                     struct mm_struct *mm,
 321                                                     unsigned long start,
 322                                                     unsigned long end)
 323 {
 324         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 325         int need_tlb_flush = 0, idx;
 326
 327         idx = srcu_read_lock(&kvm->srcu);
 328         spin_lock(&kvm->mmu_lock);
 329         /*
 330          * The count increase must become visible at unlock time as no
 331          * spte can be established without taking the mmu_lock and
 332          * count is also read inside the mmu_lock critical section.
 333          */
 334         kvm->mmu_notifier_count++;
 335         for (; start < end; start += PAGE_SIZE)
 336                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 337         need_tlb_flush |= kvm->tlbs_dirty;
 338         /* we've to flush the tlb before the pages can be freed */
 339         if (need_tlb_flush)
 340                 kvm_flush_remote_tlbs(kvm);
 341
 342         spin_unlock(&kvm->mmu_lock);
 343         srcu_read_unlock(&kvm->srcu, idx);
 344 }
 345
 346 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 347                                                   struct mm_struct *mm,
 348                                                   unsigned long start,
 349                                                   unsigned long end)
 350 {
 351         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 352
 353         spin_lock(&kvm->mmu_lock);
 354         /*
 355          * This sequence increase will notify the kvm page fault that
 356          * the page that is going to be mapped in the spte could have
 357          * been freed.
 358          */
 359         kvm->mmu_notifier_seq++;
 360         smp_wmb();
 361         /*
 362          * The above sequence increase must be visible before the
 363          * below count decrease, which is ensured by the smp_wmb above
 364          * in conjunction with the smp_rmb in mmu_notifier_retry().
 365          */
 366         kvm->mmu_notifier_count--;
 367         spin_unlock(&kvm->mmu_lock);
 368
 369         BUG_ON(kvm->mmu_notifier_count < 0);
 370 }
 371
 372 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 373                                               struct mm_struct *mm,
 374                                               unsigned long address)
 375 {
 376         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 377         int young, idx;
 378
 379         idx = srcu_read_lock(&kvm->srcu);
 380         spin_lock(&kvm->mmu_lock);
 381
 382         young = kvm_age_hva(kvm, address);
 383         if (young)
 384                 kvm_flush_remote_tlbs(kvm);
 385
 386         spin_unlock(&kvm->mmu_lock);
 387         srcu_read_unlock(&kvm->srcu, idx);
 388
 389         return young;
 390 }
 391
 392 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 393                                        struct mm_struct *mm,
 394                                        unsigned long address)
 395 {
 396         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 397         int young, idx;
 398
 399         idx = srcu_read_lock(&kvm->srcu);
 400         spin_lock(&kvm->mmu_lock);
 401         young = kvm_test_age_hva(kvm, address);
 402         spin_unlock(&kvm->mmu_lock);
 403         srcu_read_unlock(&kvm->srcu, idx);
 404
 405         return young;
 406 }
 407
 408 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 409                                      struct mm_struct *mm)
 410 {
 411         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 412         int idx;
 413
 414         idx = srcu_read_lock(&kvm->srcu);
 415         kvm_arch_flush_shadow(kvm);
 416         srcu_read_unlock(&kvm->srcu, idx);
 417 }
 418
 419 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 420         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 421         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 422         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 423         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 424         .test_young             = kvm_mmu_notifier_test_young,
 425         .change_pte             = kvm_mmu_notifier_change_pte,
 426         .release                = kvm_mmu_notifier_release,
 427 };
 428
 429 static int kvm_init_mmu_notifier(struct kvm *kvm)
 430 {
 431         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 432         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 433 }
 434
 435 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 436
 437 static int kvm_init_mmu_notifier(struct kvm *kvm)
 438 {
 439         return 0;
 440 }
 441
 442 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 443
 444 static void kvm_init_memslots_id(struct kvm *kvm)
 445 {
 446         int i;
 447         struct kvm_memslots *slots = kvm->memslots;
 448
 449         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 450                 slots->id_to_index[i] = slots->memslots[i].id = i;
 451 }
 452
 453 static struct kvm *kvm_create_vm(unsigned long type)
 454 {
 455         int r, i;
 456         struct kvm *kvm = kvm_arch_alloc_vm();
 457
 458         if (!kvm)
 459                 return ERR_PTR(-ENOMEM);
 460
 461         r = kvm_arch_init_vm(kvm, type);
 462         if (r)
 463                 goto out_err_nodisable;
 464
 465         r = hardware_enable_all();
 466         if (r)
 467                 goto out_err_nodisable;
 468
 469 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 470         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 471         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 472 #endif
 473
 474         r = -ENOMEM;
 475         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 476         if (!kvm->memslots)
 477                 goto out_err_nosrcu;
 478         kvm_init_memslots_id(kvm);
 479         if (init_srcu_struct(&kvm->srcu))
 480                 goto out_err_nosrcu;
 481         for (i = 0; i < KVM_NR_BUSES; i++) {
 482                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 483                                         GFP_KERNEL);
 484                 if (!kvm->buses[i])
 485                         goto out_err;
 486         }
 487
 488         spin_lock_init(&kvm->mmu_lock);
 489         kvm->mm = current->mm;
 490         atomic_inc(&kvm->mm->mm_count);
 491         kvm_eventfd_init(kvm);
 492         mutex_init(&kvm->lock);
 493         mutex_init(&kvm->irq_lock);
 494         mutex_init(&kvm->slots_lock);
 495         atomic_set(&kvm->users_count, 1);
 496
 497         r = kvm_init_mmu_notifier(kvm);
 498         if (r)
 499                 goto out_err;
 500
 501         raw_spin_lock(&kvm_lock);
 502         list_add(&kvm->vm_list, &vm_list);
 503         raw_spin_unlock(&kvm_lock);
 504
 505         return kvm;
 506
 507 out_err:
 508         cleanup_srcu_struct(&kvm->srcu);
 509 out_err_nosrcu:
 510         hardware_disable_all();
 511 out_err_nodisable:
 512         for (i = 0; i < KVM_NR_BUSES; i++)
 513                 kfree(kvm->buses[i]);
 514         kfree(kvm->memslots);
 515         kvm_arch_free_vm(kvm);
 516         return ERR_PTR(r);
 517 }
 518
 519 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 520 {
 521         if (!memslot->dirty_bitmap)
 522                 return;
 523
 524         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
 525                 vfree(memslot->dirty_bitmap_head);
 526         else
 527                 kfree(memslot->dirty_bitmap_head);
 528
 529         memslot->dirty_bitmap = NULL;
 530         memslot->dirty_bitmap_head = NULL;
 531 }
 532
 533 /*
 534  * Free any memory in @free but not in @dont.
 535  */
 536 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 537                                   struct kvm_memory_slot *dont)
 538 {
 539         if (!dont || free->rmap != dont->rmap)
 540                 vfree(free->rmap);
 541
 542         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 543                 kvm_destroy_dirty_bitmap(free);
 544
 545         kvm_arch_free_memslot(free, dont);
 546
 547         free->npages = 0;
 548         free->rmap = NULL;
 549 }
 550
 551 void kvm_free_physmem(struct kvm *kvm)
 552 {
 553         struct kvm_memslots *slots = kvm->memslots;
 554         struct kvm_memory_slot *memslot;
 555
 556         kvm_for_each_memslot(memslot, slots)
 557                 kvm_free_physmem_slot(memslot, NULL);
 558
 559         kfree(kvm->memslots);
 560 }
 561
 562 static void kvm_destroy_vm(struct kvm *kvm)
 563 {
 564         int i;
 565         struct mm_struct *mm = kvm->mm;
 566
 567         kvm_arch_sync_events(kvm);
 568         raw_spin_lock(&kvm_lock);
 569         list_del(&kvm->vm_list);
 570         raw_spin_unlock(&kvm_lock);
 571         kvm_free_irq_routing(kvm);
 572         for (i = 0; i < KVM_NR_BUSES; i++)
 573                 kvm_io_bus_destroy(kvm->buses[i]);
 574         kvm_coalesced_mmio_free(kvm);
 575 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 576         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 577 #else
 578         kvm_arch_flush_shadow(kvm);
 579 #endif
 580         kvm_arch_destroy_vm(kvm);
 581         kvm_free_physmem(kvm);
 582         cleanup_srcu_struct(&kvm->srcu);
 583         kvm_arch_free_vm(kvm);
 584         hardware_disable_all();
 585         mmdrop(mm);
 586 }
 587
 588 void kvm_get_kvm(struct kvm *kvm)
 589 {
 590         atomic_inc(&kvm->users_count);
 591 }
 592 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 593
 594 void kvm_put_kvm(struct kvm *kvm)
 595 {
 596         if (atomic_dec_and_test(&kvm->users_count))
 597                 kvm_destroy_vm(kvm);
 598 }
 599 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 600
 601
 602 static int kvm_vm_release(struct inode *inode, struct file *filp)
 603 {
 604         struct kvm *kvm = filp->private_data;
 605
 606         kvm_irqfd_release(kvm);
 607
 608         kvm_put_kvm(kvm);
 609         return 0;
 610 }
 611
 612 /*
 613  * Allocation size is twice as large as the actual dirty bitmap size.
 614  * This makes it possible to do double buffering: see x86's
 615  * kvm_vm_ioctl_get_dirty_log().
 616  */
 617 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 618 {
 619 #ifndef CONFIG_S390
 620         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 621
 622         if (dirty_bytes > PAGE_SIZE)
 623                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
 624         else
 625                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
 626
 627         if (!memslot->dirty_bitmap)
 628                 return -ENOMEM;
 629
 630         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
 631         memslot->nr_dirty_pages = 0;
 632 #endif /* !CONFIG_S390 */
 633         return 0;
 634 }
 635
 636 static int cmp_memslot(const void *slot1, const void *slot2)
 637 {
 638         struct kvm_memory_slot *s1, *s2;
 639
 640         s1 = (struct kvm_memory_slot *)slot1;
 641         s2 = (struct kvm_memory_slot *)slot2;
 642
 643         if (s1->npages < s2->npages)
 644                 return 1;
 645         if (s1->npages > s2->npages)
 646                 return -1;
 647
 648         return 0;
 649 }
 650
 651 /*
 652  * Sort the memslots base on its size, so the larger slots
 653  * will get better fit.
 654  */
 655 static void sort_memslots(struct kvm_memslots *slots)
 656 {
 657         int i;
 658
 659         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 660               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 661
 662         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 663                 slots->id_to_index[slots->memslots[i].id] = i;
 664 }
 665
 666 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
 667 {
 668         if (new) {
 669                 int id = new->id;
 670                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 671                 unsigned long npages = old->npages;
 672
 673                 *old = *new;
 674                 if (new->npages != npages)
 675                         sort_memslots(slots);
 676         }
 677
 678         slots->generation++;
 679 }
 680
 681 /*
 682  * Allocate some memory and give it an address in the guest physical address
 683  * space.
 684  *
 685  * Discontiguous memory is allowed, mostly for framebuffers.
 686  *
 687  * Must be called holding mmap_sem for write.
 688  */
 689 int __kvm_set_memory_region(struct kvm *kvm,
 690                             struct kvm_userspace_memory_region *mem,
 691                             int user_alloc)
 692 {
 693         int r;
 694         gfn_t base_gfn;
 695         unsigned long npages;
 696         unsigned long i;
 697         struct kvm_memory_slot *memslot;
 698         struct kvm_memory_slot old, new;
 699         struct kvm_memslots *slots, *old_memslots;
 700
 701         r = -EINVAL;
 702         /* General sanity checks */
 703         if (mem->memory_size & (PAGE_SIZE - 1))
 704                 goto out;
 705         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 706                 goto out;
 707         /* We can read the guest memory with __xxx_user() later on. */
 708         if (user_alloc &&
 709             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 710              !access_ok(VERIFY_WRITE,
 711                         (void __user *)(unsigned long)mem->userspace_addr,
 712                         mem->memory_size)))
 713                 goto out;
 714         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 715                 goto out;
 716         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 717                 goto out;
 718
 719         memslot = id_to_memslot(kvm->memslots, mem->slot);
 720         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 721         npages = mem->memory_size >> PAGE_SHIFT;
 722
 723         r = -EINVAL;
 724         if (npages > KVM_MEM_MAX_NR_PAGES)
 725                 goto out;
 726
 727         if (!npages)
 728                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 729
 730         new = old = *memslot;
 731
 732         new.id = mem->slot;
 733         new.base_gfn = base_gfn;
 734         new.npages = npages;
 735         new.flags = mem->flags;
 736
 737         /* Disallow changing a memory slot's size. */
 738         r = -EINVAL;
 739         if (npages && old.npages && npages != old.npages)
 740                 goto out_free;
 741
 742         /* Check for overlaps */
 743         r = -EEXIST;
 744         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 745                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 746
 747                 if (s == memslot || !s->npages)
 748                         continue;
 749                 if (!((base_gfn + npages <= s->base_gfn) ||
 750                       (base_gfn >= s->base_gfn + s->npages)))
 751                         goto out_free;
 752         }
 753
 754         /* Free page dirty bitmap if unneeded */
 755         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 756                 new.dirty_bitmap = NULL;
 757
 758         r = -ENOMEM;
 759
 760         /* Allocate if a slot is being created */
 761         if (npages && !old.npages) {
 762                 new.user_alloc = user_alloc;
 763                 new.userspace_addr = mem->userspace_addr;
 764 #ifndef CONFIG_S390
 765                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
 766                 if (!new.rmap)
 767                         goto out_free;
 768 #endif /* not defined CONFIG_S390 */
 769                 if (kvm_arch_create_memslot(&new, npages))
 770                         goto out_free;
 771         }
 772
 773         /* Allocate page dirty bitmap if needed */
 774         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 775                 if (kvm_create_dirty_bitmap(&new) < 0)
 776                         goto out_free;
 777                 /* destroy any largepage mappings for dirty tracking */
 778         }
 779
 780         if (!npages) {
 781                 struct kvm_memory_slot *slot;
 782
 783                 r = -ENOMEM;
 784                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 785                                 GFP_KERNEL);
 786                 if (!slots)
 787                         goto out_free;
 788                 slot = id_to_memslot(slots, mem->slot);
 789                 slot->flags |= KVM_MEMSLOT_INVALID;
 790
 791                 update_memslots(slots, NULL);
 792
 793                 old_memslots = kvm->memslots;
 794                 rcu_assign_pointer(kvm->memslots, slots);
 795                 synchronize_srcu_expedited(&kvm->srcu);
 796                 /* From this point no new shadow pages pointing to a deleted
 797                  * memslot will be created.
 798                  *
 799                  * validation of sp->gfn happens in:
 800                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 801                  *      - kvm_is_visible_gfn (mmu_check_roots)
 802                  */
 803                 kvm_arch_flush_shadow(kvm);
 804                 kfree(old_memslots);
 805         }
 806
 807         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 808         if (r)
 809                 goto out_free;
 810
 811         /* map the pages in iommu page table */
 812         if (npages) {
 813                 r = kvm_iommu_map_pages(kvm, &new);
 814                 if (r)
 815                         goto out_free;
 816         }
 817
 818         r = -ENOMEM;
 819         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 820                         GFP_KERNEL);
 821         if (!slots)
 822                 goto out_free;
 823
 824         /* actual memory is freed via old in kvm_free_physmem_slot below */
 825         if (!npages) {
 826                 new.rmap = NULL;
 827                 new.dirty_bitmap = NULL;
 828                 memset(&new.arch, 0, sizeof(new.arch));
 829         }
 830
 831         update_memslots(slots, &new);
 832         old_memslots = kvm->memslots;
 833         rcu_assign_pointer(kvm->memslots, slots);
 834         synchronize_srcu_expedited(&kvm->srcu);
 835
 836         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 837
 838         /*
 839          * If the new memory slot is created, we need to clear all
 840          * mmio sptes.
 841          */
 842         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 843                 kvm_arch_flush_shadow(kvm);
 844
 845         kvm_free_physmem_slot(&old, &new);
 846         kfree(old_memslots);
 847
 848         return 0;
 849
 850 out_free:
 851         kvm_free_physmem_slot(&new, &old);
 852 out:
 853         return r;
 854
 855 }
 856 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 857
 858 int kvm_set_memory_region(struct kvm *kvm,
 859                           struct kvm_userspace_memory_region *mem,
 860                           int user_alloc)
 861 {
 862         int r;
 863
 864         mutex_lock(&kvm->slots_lock);
 865         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 866         mutex_unlock(&kvm->slots_lock);
 867         return r;
 868 }
 869 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 870
 871 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 872                                    struct
 873                                    kvm_userspace_memory_region *mem,
 874                                    int user_alloc)
 875 {
 876         if (mem->slot >= KVM_MEMORY_SLOTS)
 877                 return -EINVAL;
 878         return kvm_set_memory_region(kvm, mem, user_alloc);
 879 }
 880
 881 int kvm_get_dirty_log(struct kvm *kvm,
 882                         struct kvm_dirty_log *log, int *is_dirty)
 883 {
 884         struct kvm_memory_slot *memslot;
 885         int r, i;
 886         unsigned long n;
 887         unsigned long any = 0;
 888
 889         r = -EINVAL;
 890         if (log->slot >= KVM_MEMORY_SLOTS)
 891                 goto out;
 892
 893         memslot = id_to_memslot(kvm->memslots, log->slot);
 894         r = -ENOENT;
 895         if (!memslot->dirty_bitmap)
 896                 goto out;
 897
 898         n = kvm_dirty_bitmap_bytes(memslot);
 899
 900         for (i = 0; !any && i < n/sizeof(long); ++i)
 901                 any = memslot->dirty_bitmap[i];
 902
 903         r = -EFAULT;
 904         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 905                 goto out;
 906
 907         if (any)
 908                 *is_dirty = 1;
 909
 910         r = 0;
 911 out:
 912         return r;
 913 }
 914
 915 bool kvm_largepages_enabled(void)
 916 {
 917         return largepages_enabled;
 918 }
 919
 920 void kvm_disable_largepages(void)
 921 {
 922         largepages_enabled = false;
 923 }
 924 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 925
 926 int is_error_page(struct page *page)
 927 {
 928         return page == bad_page || page == hwpoison_page || page == fault_page;
 929 }
 930 EXPORT_SYMBOL_GPL(is_error_page);
 931
 932 int is_error_pfn(pfn_t pfn)
 933 {
 934         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 935 }
 936 EXPORT_SYMBOL_GPL(is_error_pfn);
 937
 938 int is_hwpoison_pfn(pfn_t pfn)
 939 {
 940         return pfn == hwpoison_pfn;
 941 }
 942 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 943
 944 int is_fault_pfn(pfn_t pfn)
 945 {
 946         return pfn == fault_pfn;
 947 }
 948 EXPORT_SYMBOL_GPL(is_fault_pfn);
 949
 950 int is_noslot_pfn(pfn_t pfn)
 951 {
 952         return pfn == bad_pfn;
 953 }
 954 EXPORT_SYMBOL_GPL(is_noslot_pfn);
 955
 956 int is_invalid_pfn(pfn_t pfn)
 957 {
 958         return pfn == hwpoison_pfn || pfn == fault_pfn;
 959 }
 960 EXPORT_SYMBOL_GPL(is_invalid_pfn);
 961
 962 static inline unsigned long bad_hva(void)
 963 {
 964         return PAGE_OFFSET;
 965 }
 966
 967 int kvm_is_error_hva(unsigned long addr)
 968 {
 969         return addr == bad_hva();
 970 }
 971 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 972
 973 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 974 {
 975         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 976 }
 977 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 978
 979 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 980 {
 981         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 982
 983         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
 984               memslot->flags & KVM_MEMSLOT_INVALID)
 985                 return 0;
 986
 987         return 1;
 988 }
 989 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 990
 991 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 992 {
 993         struct vm_area_struct *vma;
 994         unsigned long addr, size;
 995
 996         size = PAGE_SIZE;
 997
 998         addr = gfn_to_hva(kvm, gfn);
 999         if (kvm_is_error_hva(addr))
1000                 return PAGE_SIZE;
1001
1002         down_read(&current->mm->mmap_sem);
1003         vma = find_vma(current->mm, addr);
1004         if (!vma)
1005                 goto out;
1006
1007         size = vma_kernel_pagesize(vma);
1008
1009 out:
1010         up_read(&current->mm->mmap_sem);
1011
1012         return size;
1013 }
1014
1015 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1016                                      gfn_t *nr_pages)
1017 {
1018         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1019                 return bad_hva();
1020
1021         if (nr_pages)
1022                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1023
1024         return gfn_to_hva_memslot(slot, gfn);
1025 }
1026
1027 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1028 {
1029         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1030 }
1031 EXPORT_SYMBOL_GPL(gfn_to_hva);
1032
1033 static pfn_t get_fault_pfn(void)
1034 {
1035         get_page(fault_page);
1036         return fault_pfn;
1037 }
1038
1039 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1040         unsigned long start, int write, struct page **page)
1041 {
1042         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1043
1044         if (write)
1045                 flags |= FOLL_WRITE;
1046
1047         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1048 }
1049
1050 static inline int check_user_page_hwpoison(unsigned long addr)
1051 {
1052         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1053
1054         rc = __get_user_pages(current, current->mm, addr, 1,
1055                               flags, NULL, NULL, NULL);
1056         return rc == -EHWPOISON;
1057 }
1058
1059 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1060                         bool *async, bool write_fault, bool *writable)
1061 {
1062         struct page *page[1];
1063         int npages = 0;
1064         pfn_t pfn;
1065
1066         /* we can do it either atomically or asynchronously, not both */
1067         BUG_ON(atomic && async);
1068
1069         BUG_ON(!write_fault && !writable);
1070
1071         if (writable)
1072                 *writable = true;
1073
1074         if (atomic || async)
1075                 npages = __get_user_pages_fast(addr, 1, 1, page);
1076
1077         if (unlikely(npages != 1) && !atomic) {
1078                 might_sleep();
1079
1080                 if (writable)
1081                         *writable = write_fault;
1082
1083                 if (async) {
1084                         down_read(&current->mm->mmap_sem);
1085                         npages = get_user_page_nowait(current, current->mm,
1086                                                      addr, write_fault, page);
1087                         up_read(&current->mm->mmap_sem);
1088                 } else
1089                         npages = get_user_pages_fast(addr, 1, write_fault,
1090                                                      page);
1091
1092                 /* map read fault as writable if possible */
1093                 if (unlikely(!write_fault) && npages == 1) {
1094                         struct page *wpage[1];
1095
1096                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1097                         if (npages == 1) {
1098                                 *writable = true;
1099                                 put_page(page[0]);
1100                                 page[0] = wpage[0];
1101                         }
1102                         npages = 1;
1103                 }
1104         }
1105
1106         if (unlikely(npages != 1)) {
1107                 struct vm_area_struct *vma;
1108
1109                 if (atomic)
1110                         return get_fault_pfn();
1111
1112                 down_read(&current->mm->mmap_sem);
1113                 if (npages == -EHWPOISON ||
1114                         (!async && check_user_page_hwpoison(addr))) {
1115                         up_read(&current->mm->mmap_sem);
1116                         get_page(hwpoison_page);
1117                         return page_to_pfn(hwpoison_page);
1118                 }
1119
1120                 vma = find_vma_intersection(current->mm, addr, addr+1);
1121
1122                 if (vma == NULL)
1123                         pfn = get_fault_pfn();
1124                 else if ((vma->vm_flags & VM_PFNMAP)) {
1125                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1126                                 vma->vm_pgoff;
1127                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1128                 } else {
1129                         if (async && (vma->vm_flags & VM_WRITE))
1130                                 *async = true;
1131                         pfn = get_fault_pfn();
1132                 }
1133                 up_read(&current->mm->mmap_sem);
1134         } else
1135                 pfn = page_to_pfn(page[0]);
1136
1137         return pfn;
1138 }
1139
1140 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1141 {
1142         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1143 }
1144 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1145
1146 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1147                           bool write_fault, bool *writable)
1148 {
1149         unsigned long addr;
1150
1151         if (async)
1152                 *async = false;
1153
1154         addr = gfn_to_hva(kvm, gfn);
1155         if (kvm_is_error_hva(addr)) {
1156                 get_page(bad_page);
1157                 return page_to_pfn(bad_page);
1158         }
1159
1160         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1161 }
1162
1163 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1164 {
1165         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1166 }
1167 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1168
1169 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1170                        bool write_fault, bool *writable)
1171 {
1172         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1173 }
1174 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1175
1176 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1177 {
1178         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1179 }
1180 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1181
1182 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1183                       bool *writable)
1184 {
1185         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1186 }
1187 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1188
1189 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1190                          struct kvm_memory_slot *slot, gfn_t gfn)
1191 {
1192         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1193         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1194 }
1195
1196 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1197                                                                   int nr_pages)
1198 {
1199         unsigned long addr;
1200         gfn_t entry;
1201
1202         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1203         if (kvm_is_error_hva(addr))
1204                 return -1;
1205
1206         if (entry < nr_pages)
1207                 return 0;
1208
1209         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1210 }
1211 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1212
1213 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1214 {
1215         pfn_t pfn;
1216
1217         pfn = gfn_to_pfn(kvm, gfn);
1218         if (!kvm_is_mmio_pfn(pfn))
1219                 return pfn_to_page(pfn);
1220
1221         WARN_ON(kvm_is_mmio_pfn(pfn));
1222
1223         get_page(bad_page);
1224         return bad_page;
1225 }
1226
1227 EXPORT_SYMBOL_GPL(gfn_to_page);
1228
1229 void kvm_release_page_clean(struct page *page)
1230 {
1231         kvm_release_pfn_clean(page_to_pfn(page));
1232 }
1233 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1234
1235 void kvm_release_pfn_clean(pfn_t pfn)
1236 {
1237         if (!kvm_is_mmio_pfn(pfn))
1238                 put_page(pfn_to_page(pfn));
1239 }
1240 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1241
1242 void kvm_release_page_dirty(struct page *page)
1243 {
1244         kvm_release_pfn_dirty(page_to_pfn(page));
1245 }
1246 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1247
1248 void kvm_release_pfn_dirty(pfn_t pfn)
1249 {
1250         kvm_set_pfn_dirty(pfn);
1251         kvm_release_pfn_clean(pfn);
1252 }
1253 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1254
1255 void kvm_set_page_dirty(struct page *page)
1256 {
1257         kvm_set_pfn_dirty(page_to_pfn(page));
1258 }
1259 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1260
1261 void kvm_set_pfn_dirty(pfn_t pfn)
1262 {
1263         if (!kvm_is_mmio_pfn(pfn)) {
1264                 struct page *page = pfn_to_page(pfn);
1265                 if (!PageReserved(page))
1266                         SetPageDirty(page);
1267         }
1268 }
1269 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1270
1271 void kvm_set_pfn_accessed(pfn_t pfn)
1272 {
1273         if (!kvm_is_mmio_pfn(pfn))
1274                 mark_page_accessed(pfn_to_page(pfn));
1275 }
1276 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1277
1278 void kvm_get_pfn(pfn_t pfn)
1279 {
1280         if (!kvm_is_mmio_pfn(pfn))
1281                 get_page(pfn_to_page(pfn));
1282 }
1283 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1284
1285 static int next_segment(unsigned long len, int offset)
1286 {
1287         if (len > PAGE_SIZE - offset)
1288                 return PAGE_SIZE - offset;
1289         else
1290                 return len;
1291 }
1292
1293 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1294                         int len)
1295 {
1296         int r;
1297         unsigned long addr;
1298
1299         addr = gfn_to_hva(kvm, gfn);
1300         if (kvm_is_error_hva(addr))
1301                 return -EFAULT;
1302         r = __copy_from_user(data, (void __user *)addr + offset, len);
1303         if (r)
1304                 return -EFAULT;
1305         return 0;
1306 }
1307 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1308
1309 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1310 {
1311         gfn_t gfn = gpa >> PAGE_SHIFT;
1312         int seg;
1313         int offset = offset_in_page(gpa);
1314         int ret;
1315
1316         while ((seg = next_segment(len, offset)) != 0) {
1317                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1318                 if (ret < 0)
1319                         return ret;
1320                 offset = 0;
1321                 len -= seg;
1322                 data += seg;
1323                 ++gfn;
1324         }
1325         return 0;
1326 }
1327 EXPORT_SYMBOL_GPL(kvm_read_guest);
1328
1329 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1330                           unsigned long len)
1331 {
1332         int r;
1333         unsigned long addr;
1334         gfn_t gfn = gpa >> PAGE_SHIFT;
1335         int offset = offset_in_page(gpa);
1336
1337         addr = gfn_to_hva(kvm, gfn);
1338         if (kvm_is_error_hva(addr))
1339                 return -EFAULT;
1340         pagefault_disable();
1341         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1342         pagefault_enable();
1343         if (r)
1344                 return -EFAULT;
1345         return 0;
1346 }
1347 EXPORT_SYMBOL(kvm_read_guest_atomic);
1348
1349 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1350                          int offset, int len)
1351 {
1352         int r;
1353         unsigned long addr;
1354
1355         addr = gfn_to_hva(kvm, gfn);
1356         if (kvm_is_error_hva(addr))
1357                 return -EFAULT;
1358         r = __copy_to_user((void __user *)addr + offset, data, len);
1359         if (r)
1360                 return -EFAULT;
1361         mark_page_dirty(kvm, gfn);
1362         return 0;
1363 }
1364 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1365
1366 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1367                     unsigned long len)
1368 {
1369         gfn_t gfn = gpa >> PAGE_SHIFT;
1370         int seg;
1371         int offset = offset_in_page(gpa);
1372         int ret;
1373
1374         while ((seg = next_segment(len, offset)) != 0) {
1375                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1376                 if (ret < 0)
1377                         return ret;
1378                 offset = 0;
1379                 len -= seg;
1380                 data += seg;
1381                 ++gfn;
1382         }
1383         return 0;
1384 }
1385
1386 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1387                               gpa_t gpa)
1388 {
1389         struct kvm_memslots *slots = kvm_memslots(kvm);
1390         int offset = offset_in_page(gpa);
1391         gfn_t gfn = gpa >> PAGE_SHIFT;
1392
1393         ghc->gpa = gpa;
1394         ghc->generation = slots->generation;
1395         ghc->memslot = gfn_to_memslot(kvm, gfn);
1396         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1397         if (!kvm_is_error_hva(ghc->hva))
1398                 ghc->hva += offset;
1399         else
1400                 return -EFAULT;
1401
1402         return 0;
1403 }
1404 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1405
1406 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1407                            void *data, unsigned long len)
1408 {
1409         struct kvm_memslots *slots = kvm_memslots(kvm);
1410         int r;
1411
1412         if (slots->generation != ghc->generation)
1413                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1414
1415         if (kvm_is_error_hva(ghc->hva))
1416                 return -EFAULT;
1417
1418         r = __copy_to_user((void __user *)ghc->hva, data, len);
1419         if (r)
1420                 return -EFAULT;
1421         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1422
1423         return 0;
1424 }
1425 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1426
1427 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1428                            void *data, unsigned long len)
1429 {
1430         struct kvm_memslots *slots = kvm_memslots(kvm);
1431         int r;
1432
1433         if (slots->generation != ghc->generation)
1434                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1435
1436         if (kvm_is_error_hva(ghc->hva))
1437                 return -EFAULT;
1438
1439         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1440         if (r)
1441                 return -EFAULT;
1442
1443         return 0;
1444 }
1445 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1446
1447 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1448 {
1449         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1450                                     offset, len);
1451 }
1452 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1453
1454 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1455 {
1456         gfn_t gfn = gpa >> PAGE_SHIFT;
1457         int seg;
1458         int offset = offset_in_page(gpa);
1459         int ret;
1460
1461         while ((seg = next_segment(len, offset)) != 0) {
1462                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1463                 if (ret < 0)
1464                         return ret;
1465                 offset = 0;
1466                 len -= seg;
1467                 ++gfn;
1468         }
1469         return 0;
1470 }
1471 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1472
1473 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1474                              gfn_t gfn)
1475 {
1476         if (memslot && memslot->dirty_bitmap) {
1477                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1478
1479                 if (!test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap))
1480                         memslot->nr_dirty_pages++;
1481         }
1482 }
1483
1484 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1485 {
1486         struct kvm_memory_slot *memslot;
1487
1488         memslot = gfn_to_memslot(kvm, gfn);
1489         mark_page_dirty_in_slot(kvm, memslot, gfn);
1490 }
1491
1492 /*
1493  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1494  */
1495 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1496 {
1497         DEFINE_WAIT(wait);
1498
1499         for (;;) {
1500                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1501
1502                 if (kvm_arch_vcpu_runnable(vcpu)) {
1503                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1504                         break;
1505                 }
1506                 if (kvm_cpu_has_pending_timer(vcpu))
1507                         break;
1508                 if (signal_pending(current))
1509                         break;
1510
1511                 schedule();
1512         }
1513
1514         finish_wait(&vcpu->wq, &wait);
1515 }
1516
1517 void kvm_resched(struct kvm_vcpu *vcpu)
1518 {
1519         if (!need_resched())
1520                 return;
1521         cond_resched();
1522 }
1523 EXPORT_SYMBOL_GPL(kvm_resched);
1524
1525 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1526 {
1527         struct kvm *kvm = me->kvm;
1528         struct kvm_vcpu *vcpu;
1529         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1530         int yielded = 0;
1531         int pass;
1532         int i;
1533
1534         /*
1535          * We boost the priority of a VCPU that is runnable but not
1536          * currently running, because it got preempted by something
1537          * else and called schedule in __vcpu_run.  Hopefully that
1538          * VCPU is holding the lock that we need and will release it.
1539          * We approximate round-robin by starting at the last boosted VCPU.
1540          */
1541         for (pass = 0; pass < 2 && !yielded; pass++) {
1542                 kvm_for_each_vcpu(i, vcpu, kvm) {
1543                         struct task_struct *task = NULL;
1544                         struct pid *pid;
1545                         if (!pass && i < last_boosted_vcpu) {
1546                                 i = last_boosted_vcpu;
1547                                 continue;
1548                         } else if (pass && i > last_boosted_vcpu)
1549                                 break;
1550                         if (vcpu == me)
1551                                 continue;
1552                         if (waitqueue_active(&vcpu->wq))
1553                                 continue;
1554                         rcu_read_lock();
1555                         pid = rcu_dereference(vcpu->pid);
1556                         if (pid)
1557                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1558                         rcu_read_unlock();
1559                         if (!task)
1560                                 continue;
1561                         if (task->flags & PF_VCPU) {
1562                                 put_task_struct(task);
1563                                 continue;
1564                         }
1565                         if (yield_to(task, 1)) {
1566                                 put_task_struct(task);
1567                                 kvm->last_boosted_vcpu = i;
1568                                 yielded = 1;
1569                                 break;
1570                         }
1571                         put_task_struct(task);
1572                 }
1573         }
1574 }
1575 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1576
1577 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1578 {
1579         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1580         struct page *page;
1581
1582         if (vmf->pgoff == 0)
1583                 page = virt_to_page(vcpu->run);
1584 #ifdef CONFIG_X86
1585         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1586                 page = virt_to_page(vcpu->arch.pio_data);
1587 #endif
1588 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1589         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1590                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1591 #endif
1592         else
1593                 return kvm_arch_vcpu_fault(vcpu, vmf);
1594         get_page(page);
1595         vmf->page = page;
1596         return 0;
1597 }
1598
1599 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1600         .fault = kvm_vcpu_fault,
1601 };
1602
1603 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1604 {
1605         vma->vm_ops = &kvm_vcpu_vm_ops;
1606         return 0;
1607 }
1608
1609 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1610 {
1611         struct kvm_vcpu *vcpu = filp->private_data;
1612
1613         kvm_put_kvm(vcpu->kvm);
1614         return 0;
1615 }
1616
1617 static struct file_operations kvm_vcpu_fops = {
1618         .release        = kvm_vcpu_release,
1619         .unlocked_ioctl = kvm_vcpu_ioctl,
1620 #ifdef CONFIG_COMPAT
1621         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1622 #endif
1623         .mmap           = kvm_vcpu_mmap,
1624         .llseek         = noop_llseek,
1625 };
1626
1627 /*
1628  * Allocates an inode for the vcpu.
1629  */
1630 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1631 {
1632         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1633 }
1634
1635 /*
1636  * Creates some virtual cpus.  Good luck creating more than one.
1637  */
1638 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1639 {
1640         int r;
1641         struct kvm_vcpu *vcpu, *v;
1642
1643         vcpu = kvm_arch_vcpu_create(kvm, id);
1644         if (IS_ERR(vcpu))
1645                 return PTR_ERR(vcpu);
1646
1647         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1648
1649         r = kvm_arch_vcpu_setup(vcpu);
1650         if (r)
1651                 goto vcpu_destroy;
1652
1653         mutex_lock(&kvm->lock);
1654         if (!kvm_vcpu_compatible(vcpu)) {
1655                 r = -EINVAL;
1656                 goto unlock_vcpu_destroy;
1657         }
1658         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1659                 r = -EINVAL;
1660                 goto unlock_vcpu_destroy;
1661         }
1662
1663         kvm_for_each_vcpu(r, v, kvm)
1664                 if (v->vcpu_id == id) {
1665                         r = -EEXIST;
1666                         goto unlock_vcpu_destroy;
1667                 }
1668
1669         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1670
1671         /* Now it's all set up, let userspace reach it */
1672         kvm_get_kvm(kvm);
1673         r = create_vcpu_fd(vcpu);
1674         if (r < 0) {
1675                 kvm_put_kvm(kvm);
1676                 goto unlock_vcpu_destroy;
1677         }
1678
1679         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1680         smp_wmb();
1681         atomic_inc(&kvm->online_vcpus);
1682
1683         mutex_unlock(&kvm->lock);
1684         return r;
1685
1686 unlock_vcpu_destroy:
1687         mutex_unlock(&kvm->lock);
1688 vcpu_destroy:
1689         kvm_arch_vcpu_destroy(vcpu);
1690         return r;
1691 }
1692
1693 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1694 {
1695         if (sigset) {
1696                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1697                 vcpu->sigset_active = 1;
1698                 vcpu->sigset = *sigset;
1699         } else
1700                 vcpu->sigset_active = 0;
1701         return 0;
1702 }
1703
1704 static long kvm_vcpu_ioctl(struct file *filp,
1705                            unsigned int ioctl, unsigned long arg)
1706 {
1707         struct kvm_vcpu *vcpu = filp->private_data;
1708         void __user *argp = (void __user *)arg;
1709         int r;
1710         struct kvm_fpu *fpu = NULL;
1711         struct kvm_sregs *kvm_sregs = NULL;
1712
1713         if (vcpu->kvm->mm != current->mm)
1714                 return -EIO;
1715
1716 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1717         /*
1718          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1719          * so vcpu_load() would break it.
1720          */
1721         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1722                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1723 #endif
1724
1725
1726         vcpu_load(vcpu);
1727         switch (ioctl) {
1728         case KVM_RUN:
1729                 r = -EINVAL;
1730                 if (arg)
1731                         goto out;
1732                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1733                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1734                 break;
1735         case KVM_GET_REGS: {
1736                 struct kvm_regs *kvm_regs;
1737
1738                 r = -ENOMEM;
1739                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1740                 if (!kvm_regs)
1741                         goto out;
1742                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1743                 if (r)
1744                         goto out_free1;
1745                 r = -EFAULT;
1746                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1747                         goto out_free1;
1748                 r = 0;
1749 out_free1:
1750                 kfree(kvm_regs);
1751                 break;
1752         }
1753         case KVM_SET_REGS: {
1754                 struct kvm_regs *kvm_regs;
1755
1756                 r = -ENOMEM;
1757                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1758                 if (IS_ERR(kvm_regs)) {
1759                         r = PTR_ERR(kvm_regs);
1760                         goto out;
1761                 }
1762                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1763                 if (r)
1764                         goto out_free2;
1765                 r = 0;
1766 out_free2:
1767                 kfree(kvm_regs);
1768                 break;
1769         }
1770         case KVM_GET_SREGS: {
1771                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1772                 r = -ENOMEM;
1773                 if (!kvm_sregs)
1774                         goto out;
1775                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1776                 if (r)
1777                         goto out;
1778                 r = -EFAULT;
1779                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1780                         goto out;
1781                 r = 0;
1782                 break;
1783         }
1784         case KVM_SET_SREGS: {
1785                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1786                 if (IS_ERR(kvm_sregs)) {
1787                         r = PTR_ERR(kvm_sregs);
1788                         goto out;
1789                 }
1790                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1791                 if (r)
1792                         goto out;
1793                 r = 0;
1794                 break;
1795         }
1796         case KVM_GET_MP_STATE: {
1797                 struct kvm_mp_state mp_state;
1798
1799                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1800                 if (r)
1801                         goto out;
1802                 r = -EFAULT;
1803                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1804                         goto out;
1805                 r = 0;
1806                 break;
1807         }
1808         case KVM_SET_MP_STATE: {
1809                 struct kvm_mp_state mp_state;
1810
1811                 r = -EFAULT;
1812                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1813                         goto out;
1814                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1815                 if (r)
1816                         goto out;
1817                 r = 0;
1818                 break;
1819         }
1820         case KVM_TRANSLATE: {
1821                 struct kvm_translation tr;
1822
1823                 r = -EFAULT;
1824                 if (copy_from_user(&tr, argp, sizeof tr))
1825                         goto out;
1826                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1827                 if (r)
1828                         goto out;
1829                 r = -EFAULT;
1830                 if (copy_to_user(argp, &tr, sizeof tr))
1831                         goto out;
1832                 r = 0;
1833                 break;
1834         }
1835         case KVM_SET_GUEST_DEBUG: {
1836                 struct kvm_guest_debug dbg;
1837
1838                 r = -EFAULT;
1839                 if (copy_from_user(&dbg, argp, sizeof dbg))
1840                         goto out;
1841                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1842                 if (r)
1843                         goto out;
1844                 r = 0;
1845                 break;
1846         }
1847         case KVM_SET_SIGNAL_MASK: {
1848                 struct kvm_signal_mask __user *sigmask_arg = argp;
1849                 struct kvm_signal_mask kvm_sigmask;
1850                 sigset_t sigset, *p;
1851
1852                 p = NULL;
1853                 if (argp) {
1854                         r = -EFAULT;
1855                         if (copy_from_user(&kvm_sigmask, argp,
1856                                            sizeof kvm_sigmask))
1857                                 goto out;
1858                         r = -EINVAL;
1859                         if (kvm_sigmask.len != sizeof sigset)
1860                                 goto out;
1861                         r = -EFAULT;
1862                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1863                                            sizeof sigset))
1864                                 goto out;
1865                         p = &sigset;
1866                 }
1867                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1868                 break;
1869         }
1870         case KVM_GET_FPU: {
1871                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1872                 r = -ENOMEM;
1873                 if (!fpu)
1874                         goto out;
1875                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1876                 if (r)
1877                         goto out;
1878                 r = -EFAULT;
1879                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1880                         goto out;
1881                 r = 0;
1882                 break;
1883         }
1884         case KVM_SET_FPU: {
1885                 fpu = memdup_user(argp, sizeof(*fpu));
1886                 if (IS_ERR(fpu)) {
1887                         r = PTR_ERR(fpu);
1888                         goto out;
1889                 }
1890                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1891                 if (r)
1892                         goto out;
1893                 r = 0;
1894                 break;
1895         }
1896         default:
1897                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1898         }
1899 out:
1900         vcpu_put(vcpu);
1901         kfree(fpu);
1902         kfree(kvm_sregs);
1903         return r;
1904 }
1905
1906 #ifdef CONFIG_COMPAT
1907 static long kvm_vcpu_compat_ioctl(struct file *filp,
1908                                   unsigned int ioctl, unsigned long arg)
1909 {
1910         struct kvm_vcpu *vcpu = filp->private_data;
1911         void __user *argp = compat_ptr(arg);
1912         int r;
1913
1914         if (vcpu->kvm->mm != current->mm)
1915                 return -EIO;
1916
1917         switch (ioctl) {
1918         case KVM_SET_SIGNAL_MASK: {
1919                 struct kvm_signal_mask __user *sigmask_arg = argp;
1920                 struct kvm_signal_mask kvm_sigmask;
1921                 compat_sigset_t csigset;
1922                 sigset_t sigset;
1923
1924                 if (argp) {
1925                         r = -EFAULT;
1926                         if (copy_from_user(&kvm_sigmask, argp,
1927                                            sizeof kvm_sigmask))
1928                                 goto out;
1929                         r = -EINVAL;
1930                         if (kvm_sigmask.len != sizeof csigset)
1931                                 goto out;
1932                         r = -EFAULT;
1933                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1934                                            sizeof csigset))
1935                                 goto out;
1936                 }
1937                 sigset_from_compat(&sigset, &csigset);
1938                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1939                 break;
1940         }
1941         default:
1942                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1943         }
1944
1945 out:
1946         return r;
1947 }
1948 #endif
1949
1950 static long kvm_vm_ioctl(struct file *filp,
1951                            unsigned int ioctl, unsigned long arg)
1952 {
1953         struct kvm *kvm = filp->private_data;
1954         void __user *argp = (void __user *)arg;
1955         int r;
1956
1957         if (kvm->mm != current->mm)
1958                 return -EIO;
1959         switch (ioctl) {
1960         case KVM_CREATE_VCPU:
1961                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1962                 if (r < 0)
1963                         goto out;
1964                 break;
1965         case KVM_SET_USER_MEMORY_REGION: {
1966                 struct kvm_userspace_memory_region kvm_userspace_mem;
1967
1968                 r = -EFAULT;
1969                 if (copy_from_user(&kvm_userspace_mem, argp,
1970                                                 sizeof kvm_userspace_mem))
1971                         goto out;
1972
1973                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1974                 if (r)
1975                         goto out;
1976                 break;
1977         }
1978         case KVM_GET_DIRTY_LOG: {
1979                 struct kvm_dirty_log log;
1980
1981                 r = -EFAULT;
1982                 if (copy_from_user(&log, argp, sizeof log))
1983                         goto out;
1984                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1985                 if (r)
1986                         goto out;
1987                 break;
1988         }
1989 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1990         case KVM_REGISTER_COALESCED_MMIO: {
1991                 struct kvm_coalesced_mmio_zone zone;
1992                 r = -EFAULT;
1993                 if (copy_from_user(&zone, argp, sizeof zone))
1994                         goto out;
1995                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1996                 if (r)
1997                         goto out;
1998                 r = 0;
1999                 break;
2000         }
2001         case KVM_UNREGISTER_COALESCED_MMIO: {
2002                 struct kvm_coalesced_mmio_zone zone;
2003                 r = -EFAULT;
2004                 if (copy_from_user(&zone, argp, sizeof zone))
2005                         goto out;
2006                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2007                 if (r)
2008                         goto out;
2009                 r = 0;
2010                 break;
2011         }
2012 #endif
2013         case KVM_IRQFD: {
2014                 struct kvm_irqfd data;
2015
2016                 r = -EFAULT;
2017                 if (copy_from_user(&data, argp, sizeof data))
2018                         goto out;
2019                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2020                 break;
2021         }
2022         case KVM_IOEVENTFD: {
2023                 struct kvm_ioeventfd data;
2024
2025                 r = -EFAULT;
2026                 if (copy_from_user(&data, argp, sizeof data))
2027                         goto out;
2028                 r = kvm_ioeventfd(kvm, &data);
2029                 break;
2030         }
2031 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2032         case KVM_SET_BOOT_CPU_ID:
2033                 r = 0;
2034                 mutex_lock(&kvm->lock);
2035                 if (atomic_read(&kvm->online_vcpus) != 0)
2036                         r = -EBUSY;
2037                 else
2038                         kvm->bsp_vcpu_id = arg;
2039                 mutex_unlock(&kvm->lock);
2040                 break;
2041 #endif
2042         default:
2043                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2044                 if (r == -ENOTTY)
2045                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2046         }
2047 out:
2048         return r;
2049 }
2050
2051 #ifdef CONFIG_COMPAT
2052 struct compat_kvm_dirty_log {
2053         __u32 slot;
2054         __u32 padding1;
2055         union {
2056                 compat_uptr_t dirty_bitmap; /* one bit per page */
2057                 __u64 padding2;
2058         };
2059 };
2060
2061 static long kvm_vm_compat_ioctl(struct file *filp,
2062                            unsigned int ioctl, unsigned long arg)
2063 {
2064         struct kvm *kvm = filp->private_data;
2065         int r;
2066
2067         if (kvm->mm != current->mm)
2068                 return -EIO;
2069         switch (ioctl) {
2070         case KVM_GET_DIRTY_LOG: {
2071                 struct compat_kvm_dirty_log compat_log;
2072                 struct kvm_dirty_log log;
2073
2074                 r = -EFAULT;
2075                 if (copy_from_user(&compat_log, (void __user *)arg,
2076                                    sizeof(compat_log)))
2077                         goto out;
2078                 log.slot         = compat_log.slot;
2079                 log.padding1     = compat_log.padding1;
2080                 log.padding2     = compat_log.padding2;
2081                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2082
2083                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2084                 if (r)
2085                         goto out;
2086                 break;
2087         }
2088         default:
2089                 r = kvm_vm_ioctl(filp, ioctl, arg);
2090         }
2091
2092 out:
2093         return r;
2094 }
2095 #endif
2096
2097 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2098 {
2099         struct page *page[1];
2100         unsigned long addr;
2101         int npages;
2102         gfn_t gfn = vmf->pgoff;
2103         struct kvm *kvm = vma->vm_file->private_data;
2104
2105         addr = gfn_to_hva(kvm, gfn);
2106         if (kvm_is_error_hva(addr))
2107                 return VM_FAULT_SIGBUS;
2108
2109         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2110                                 NULL);
2111         if (unlikely(npages != 1))
2112                 return VM_FAULT_SIGBUS;
2113
2114         vmf->page = page[0];
2115         return 0;
2116 }
2117
2118 static const struct vm_operations_struct kvm_vm_vm_ops = {
2119         .fault = kvm_vm_fault,
2120 };
2121
2122 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2123 {
2124         vma->vm_ops = &kvm_vm_vm_ops;
2125         return 0;
2126 }
2127
2128 static struct file_operations kvm_vm_fops = {
2129         .release        = kvm_vm_release,
2130         .unlocked_ioctl = kvm_vm_ioctl,
2131 #ifdef CONFIG_COMPAT
2132         .compat_ioctl   = kvm_vm_compat_ioctl,
2133 #endif
2134         .mmap           = kvm_vm_mmap,
2135         .llseek         = noop_llseek,
2136 };
2137
2138 static int kvm_dev_ioctl_create_vm(unsigned long type)
2139 {
2140         int r;
2141         struct kvm *kvm;
2142
2143         kvm = kvm_create_vm(type);
2144         if (IS_ERR(kvm))
2145                 return PTR_ERR(kvm);
2146 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2147         r = kvm_coalesced_mmio_init(kvm);
2148         if (r < 0) {
2149                 kvm_put_kvm(kvm);
2150                 return r;
2151         }
2152 #endif
2153         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2154         if (r < 0)
2155                 kvm_put_kvm(kvm);
2156
2157         return r;
2158 }
2159
2160 static long kvm_dev_ioctl_check_extension_generic(long arg)
2161 {
2162         switch (arg) {
2163         case KVM_CAP_USER_MEMORY:
2164         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2165         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2166 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2167         case KVM_CAP_SET_BOOT_CPU_ID:
2168 #endif
2169         case KVM_CAP_INTERNAL_ERROR_DATA:
2170                 return 1;
2171 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2172         case KVM_CAP_IRQ_ROUTING:
2173                 return KVM_MAX_IRQ_ROUTES;
2174 #endif
2175         default:
2176                 break;
2177         }
2178         return kvm_dev_ioctl_check_extension(arg);
2179 }
2180
2181 static long kvm_dev_ioctl(struct file *filp,
2182                           unsigned int ioctl, unsigned long arg)
2183 {
2184         long r = -EINVAL;
2185
2186         switch (ioctl) {
2187         case KVM_GET_API_VERSION:
2188                 r = -EINVAL;
2189                 if (arg)
2190                         goto out;
2191                 r = KVM_API_VERSION;
2192                 break;
2193         case KVM_CREATE_VM:
2194                 r = kvm_dev_ioctl_create_vm(arg);
2195                 break;
2196         case KVM_CHECK_EXTENSION:
2197                 r = kvm_dev_ioctl_check_extension_generic(arg);
2198                 break;
2199         case KVM_GET_VCPU_MMAP_SIZE:
2200                 r = -EINVAL;
2201                 if (arg)
2202                         goto out;
2203                 r = PAGE_SIZE;     /* struct kvm_run */
2204 #ifdef CONFIG_X86
2205                 r += PAGE_SIZE;    /* pio data page */
2206 #endif
2207 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2208                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2209 #endif
2210                 break;
2211         case KVM_TRACE_ENABLE:
2212         case KVM_TRACE_PAUSE:
2213         case KVM_TRACE_DISABLE:
2214                 r = -EOPNOTSUPP;
2215                 break;
2216         default:
2217                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2218         }
2219 out:
2220         return r;
2221 }
2222
2223 static struct file_operations kvm_chardev_ops = {
2224         .unlocked_ioctl = kvm_dev_ioctl,
2225         .compat_ioctl   = kvm_dev_ioctl,
2226         .llseek         = noop_llseek,
2227 };
2228
2229 static struct miscdevice kvm_dev = {
2230         KVM_MINOR,
2231         "kvm",
2232         &kvm_chardev_ops,
2233 };
2234
2235 static void hardware_enable_nolock(void *junk)
2236 {
2237         int cpu = raw_smp_processor_id();
2238         int r;
2239
2240         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2241                 return;
2242
2243         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2244
2245         r = kvm_arch_hardware_enable(NULL);
2246
2247         if (r) {
2248                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2249                 atomic_inc(&hardware_enable_failed);
2250                 printk(KERN_INFO "kvm: enabling virtualization on "
2251                                  "CPU%d failed\n", cpu);
2252         }
2253 }
2254
2255 static void hardware_enable(void *junk)
2256 {
2257         raw_spin_lock(&kvm_lock);
2258         hardware_enable_nolock(junk);
2259         raw_spin_unlock(&kvm_lock);
2260 }
2261
2262 static void hardware_disable_nolock(void *junk)
2263 {
2264         int cpu = raw_smp_processor_id();
2265
2266         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2267                 return;
2268         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2269         kvm_arch_hardware_disable(NULL);
2270 }
2271
2272 static void hardware_disable(void *junk)
2273 {
2274         raw_spin_lock(&kvm_lock);
2275         hardware_disable_nolock(junk);
2276         raw_spin_unlock(&kvm_lock);
2277 }
2278
2279 static void hardware_disable_all_nolock(void)
2280 {
2281         BUG_ON(!kvm_usage_count);
2282
2283         kvm_usage_count--;
2284         if (!kvm_usage_count)
2285                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2286 }
2287
2288 static void hardware_disable_all(void)
2289 {
2290         raw_spin_lock(&kvm_lock);
2291         hardware_disable_all_nolock();
2292         raw_spin_unlock(&kvm_lock);
2293 }
2294
2295 static int hardware_enable_all(void)
2296 {
2297         int r = 0;
2298
2299         raw_spin_lock(&kvm_lock);
2300
2301         kvm_usage_count++;
2302         if (kvm_usage_count == 1) {
2303                 atomic_set(&hardware_enable_failed, 0);
2304                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2305
2306                 if (atomic_read(&hardware_enable_failed)) {
2307                         hardware_disable_all_nolock();
2308                         r = -EBUSY;
2309                 }
2310         }
2311
2312         raw_spin_unlock(&kvm_lock);
2313
2314         return r;
2315 }
2316
2317 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2318                            void *v)
2319 {
2320         int cpu = (long)v;
2321
2322         if (!kvm_usage_count)
2323                 return NOTIFY_OK;
2324
2325         val &= ~CPU_TASKS_FROZEN;
2326         switch (val) {
2327         case CPU_DYING:
2328                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2329                        cpu);
2330                 hardware_disable(NULL);
2331                 break;
2332         case CPU_STARTING:
2333                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2334                        cpu);
2335                 hardware_enable(NULL);
2336                 break;
2337         }
2338         return NOTIFY_OK;
2339 }
2340
2341
2342 asmlinkage void kvm_spurious_fault(void)
2343 {
2344         /* Fault while not rebooting.  We want the trace. */
2345         BUG();
2346 }
2347 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2348
2349 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2350                       void *v)
2351 {
2352         /*
2353          * Some (well, at least mine) BIOSes hang on reboot if
2354          * in vmx root mode.
2355          *
2356          * And Intel TXT required VMX off for all cpu when system shutdown.
2357          */
2358         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2359         kvm_rebooting = true;
2360         on_each_cpu(hardware_disable_nolock, NULL, 1);
2361         return NOTIFY_OK;
2362 }
2363
2364 static struct notifier_block kvm_reboot_notifier = {
2365         .notifier_call = kvm_reboot,
2366         .priority = 0,
2367 };
2368
2369 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2370 {
2371         int i;
2372
2373         for (i = 0; i < bus->dev_count; i++) {
2374                 struct kvm_io_device *pos = bus->range[i].dev;
2375
2376                 kvm_iodevice_destructor(pos);
2377         }
2378         kfree(bus);
2379 }
2380
2381 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2382 {
2383         const struct kvm_io_range *r1 = p1;
2384         const struct kvm_io_range *r2 = p2;
2385
2386         if (r1->addr < r2->addr)
2387                 return -1;
2388         if (r1->addr + r1->len > r2->addr + r2->len)
2389                 return 1;
2390         return 0;
2391 }
2392
2393 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2394                           gpa_t addr, int len)
2395 {
2396         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2397                 .addr = addr,
2398                 .len = len,
2399                 .dev = dev,
2400         };
2401
2402         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2403                 kvm_io_bus_sort_cmp, NULL);
2404
2405         return 0;
2406 }
2407
2408 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2409                              gpa_t addr, int len)
2410 {
2411         struct kvm_io_range *range, key;
2412         int off;
2413
2414         key = (struct kvm_io_range) {
2415                 .addr = addr,
2416                 .len = len,
2417         };
2418
2419         range = bsearch(&key, bus->range, bus->dev_count,
2420                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2421         if (range == NULL)
2422                 return -ENOENT;
2423
2424         off = range - bus->range;
2425
2426         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2427                 off--;
2428
2429         return off;
2430 }
2431
2432 /* kvm_io_bus_write - called under kvm->slots_lock */
2433 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2434                      int len, const void *val)
2435 {
2436         int idx;
2437         struct kvm_io_bus *bus;
2438         struct kvm_io_range range;
2439
2440         range = (struct kvm_io_range) {
2441                 .addr = addr,
2442                 .len = len,
2443         };
2444
2445         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2446         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2447         if (idx < 0)
2448                 return -EOPNOTSUPP;
2449
2450         while (idx < bus->dev_count &&
2451                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2452                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2453                         return 0;
2454                 idx++;
2455         }
2456
2457         return -EOPNOTSUPP;
2458 }
2459
2460 /* kvm_io_bus_read - called under kvm->slots_lock */
2461 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2462                     int len, void *val)
2463 {
2464         int idx;
2465         struct kvm_io_bus *bus;
2466         struct kvm_io_range range;
2467
2468         range = (struct kvm_io_range) {
2469                 .addr = addr,
2470                 .len = len,
2471         };
2472
2473         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2474         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2475         if (idx < 0)
2476                 return -EOPNOTSUPP;
2477
2478         while (idx < bus->dev_count &&
2479                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2480                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2481                         return 0;
2482                 idx++;
2483         }
2484
2485         return -EOPNOTSUPP;
2486 }
2487
2488 /* Caller must hold slots_lock. */
2489 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2490                             int len, struct kvm_io_device *dev)
2491 {
2492         struct kvm_io_bus *new_bus, *bus;
2493
2494         bus = kvm->buses[bus_idx];
2495         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2496                 return -ENOSPC;
2497
2498         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2499                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2500         if (!new_bus)
2501                 return -ENOMEM;
2502         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2503                sizeof(struct kvm_io_range)));
2504         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2505         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2506         synchronize_srcu_expedited(&kvm->srcu);
2507         kfree(bus);
2508
2509         return 0;
2510 }
2511
2512 /* Caller must hold slots_lock. */
2513 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2514                               struct kvm_io_device *dev)
2515 {
2516         int i, r;
2517         struct kvm_io_bus *new_bus, *bus;
2518
2519         bus = kvm->buses[bus_idx];
2520         r = -ENOENT;
2521         for (i = 0; i < bus->dev_count; i++)
2522                 if (bus->range[i].dev == dev) {
2523                         r = 0;
2524                         break;
2525                 }
2526
2527         if (r)
2528                 return r;
2529
2530         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2531                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2532         if (!new_bus)
2533                 return -ENOMEM;
2534
2535         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2536         new_bus->dev_count--;
2537         memcpy(new_bus->range + i, bus->range + i + 1,
2538                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2539
2540         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2541         synchronize_srcu_expedited(&kvm->srcu);
2542         kfree(bus);
2543         return r;
2544 }
2545
2546 static struct notifier_block kvm_cpu_notifier = {
2547         .notifier_call = kvm_cpu_hotplug,
2548 };
2549
2550 static int vm_stat_get(void *_offset, u64 *val)
2551 {
2552         unsigned offset = (long)_offset;
2553         struct kvm *kvm;
2554
2555         *val = 0;
2556         raw_spin_lock(&kvm_lock);
2557         list_for_each_entry(kvm, &vm_list, vm_list)
2558                 *val += *(u32 *)((void *)kvm + offset);
2559         raw_spin_unlock(&kvm_lock);
2560         return 0;
2561 }
2562
2563 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2564
2565 static int vcpu_stat_get(void *_offset, u64 *val)
2566 {
2567         unsigned offset = (long)_offset;
2568         struct kvm *kvm;
2569         struct kvm_vcpu *vcpu;
2570         int i;
2571
2572         *val = 0;
2573         raw_spin_lock(&kvm_lock);
2574         list_for_each_entry(kvm, &vm_list, vm_list)
2575                 kvm_for_each_vcpu(i, vcpu, kvm)
2576                         *val += *(u32 *)((void *)vcpu + offset);
2577
2578         raw_spin_unlock(&kvm_lock);
2579         return 0;
2580 }
2581
2582 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2583
2584 static const struct file_operations *stat_fops[] = {
2585         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2586         [KVM_STAT_VM]   = &vm_stat_fops,
2587 };
2588
2589 static int kvm_init_debug(void)
2590 {
2591         int r = -EFAULT;
2592         struct kvm_stats_debugfs_item *p;
2593
2594         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2595         if (kvm_debugfs_dir == NULL)
2596                 goto out;
2597
2598         for (p = debugfs_entries; p->name; ++p) {
2599                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2600                                                 (void *)(long)p->offset,
2601                                                 stat_fops[p->kind]);
2602                 if (p->dentry == NULL)
2603                         goto out_dir;
2604         }
2605
2606         return 0;
2607
2608 out_dir:
2609         debugfs_remove_recursive(kvm_debugfs_dir);
2610 out:
2611         return r;
2612 }
2613
2614 static void kvm_exit_debug(void)
2615 {
2616         struct kvm_stats_debugfs_item *p;
2617
2618         for (p = debugfs_entries; p->name; ++p)
2619                 debugfs_remove(p->dentry);
2620         debugfs_remove(kvm_debugfs_dir);
2621 }
2622
2623 static int kvm_suspend(void)
2624 {
2625         if (kvm_usage_count)
2626                 hardware_disable_nolock(NULL);
2627         return 0;
2628 }
2629
2630 static void kvm_resume(void)
2631 {
2632         if (kvm_usage_count) {
2633                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2634                 hardware_enable_nolock(NULL);
2635         }
2636 }
2637
2638 static struct syscore_ops kvm_syscore_ops = {
2639         .suspend = kvm_suspend,
2640         .resume = kvm_resume,
2641 };
2642
2643 struct page *bad_page;
2644 pfn_t bad_pfn;
2645
2646 static inline
2647 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2648 {
2649         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2650 }
2651
2652 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2653 {
2654         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2655
2656         kvm_arch_vcpu_load(vcpu, cpu);
2657 }
2658
2659 static void kvm_sched_out(struct preempt_notifier *pn,
2660                           struct task_struct *next)
2661 {
2662         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2663
2664         kvm_arch_vcpu_put(vcpu);
2665 }
2666
2667 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2668                   struct module *module)
2669 {
2670         int r;
2671         int cpu;
2672
2673         r = kvm_arch_init(opaque);
2674         if (r)
2675                 goto out_fail;
2676
2677         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2678
2679         if (bad_page == NULL) {
2680                 r = -ENOMEM;
2681                 goto out;
2682         }
2683
2684         bad_pfn = page_to_pfn(bad_page);
2685
2686         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2687
2688         if (hwpoison_page == NULL) {
2689                 r = -ENOMEM;
2690                 goto out_free_0;
2691         }
2692
2693         hwpoison_pfn = page_to_pfn(hwpoison_page);
2694
2695         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2696
2697         if (fault_page == NULL) {
2698                 r = -ENOMEM;
2699                 goto out_free_0;
2700         }
2701
2702         fault_pfn = page_to_pfn(fault_page);
2703
2704         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2705                 r = -ENOMEM;
2706                 goto out_free_0;
2707         }
2708
2709         r = kvm_arch_hardware_setup();
2710         if (r < 0)
2711                 goto out_free_0a;
2712
2713         for_each_online_cpu(cpu) {
2714                 smp_call_function_single(cpu,
2715                                 kvm_arch_check_processor_compat,
2716                                 &r, 1);
2717                 if (r < 0)
2718                         goto out_free_1;
2719         }
2720
2721         r = register_cpu_notifier(&kvm_cpu_notifier);
2722         if (r)
2723                 goto out_free_2;
2724         register_reboot_notifier(&kvm_reboot_notifier);
2725
2726         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2727         if (!vcpu_align)
2728                 vcpu_align = __alignof__(struct kvm_vcpu);
2729         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2730                                            0, NULL);
2731         if (!kvm_vcpu_cache) {
2732                 r = -ENOMEM;
2733                 goto out_free_3;
2734         }
2735
2736         r = kvm_async_pf_init();
2737         if (r)
2738                 goto out_free;
2739
2740         kvm_chardev_ops.owner = module;
2741         kvm_vm_fops.owner = module;
2742         kvm_vcpu_fops.owner = module;
2743
2744         r = misc_register(&kvm_dev);
2745         if (r) {
2746                 printk(KERN_ERR "kvm: misc device register failed\n");
2747                 goto out_unreg;
2748         }
2749
2750         register_syscore_ops(&kvm_syscore_ops);
2751
2752         kvm_preempt_ops.sched_in = kvm_sched_in;
2753         kvm_preempt_ops.sched_out = kvm_sched_out;
2754
2755         r = kvm_init_debug();
2756         if (r) {
2757                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2758                 goto out_undebugfs;
2759         }
2760
2761         return 0;
2762
2763 out_undebugfs:
2764         unregister_syscore_ops(&kvm_syscore_ops);
2765 out_unreg:
2766         kvm_async_pf_deinit();
2767 out_free:
2768         kmem_cache_destroy(kvm_vcpu_cache);
2769 out_free_3:
2770         unregister_reboot_notifier(&kvm_reboot_notifier);
2771         unregister_cpu_notifier(&kvm_cpu_notifier);
2772 out_free_2:
2773 out_free_1:
2774         kvm_arch_hardware_unsetup();
2775 out_free_0a:
2776         free_cpumask_var(cpus_hardware_enabled);
2777 out_free_0:
2778         if (fault_page)
2779                 __free_page(fault_page);
2780         if (hwpoison_page)
2781                 __free_page(hwpoison_page);
2782         __free_page(bad_page);
2783 out:
2784         kvm_arch_exit();
2785 out_fail:
2786         return r;
2787 }
2788 EXPORT_SYMBOL_GPL(kvm_init);
2789
2790 void kvm_exit(void)
2791 {
2792         kvm_exit_debug();
2793         misc_deregister(&kvm_dev);
2794         kmem_cache_destroy(kvm_vcpu_cache);
2795         kvm_async_pf_deinit();
2796         unregister_syscore_ops(&kvm_syscore_ops);
2797         unregister_reboot_notifier(&kvm_reboot_notifier);
2798         unregister_cpu_notifier(&kvm_cpu_notifier);
2799         on_each_cpu(hardware_disable_nolock, NULL, 1);
2800         kvm_arch_hardware_unsetup();
2801         kvm_arch_exit();
2802         free_cpumask_var(cpus_hardware_enabled);
2803         __free_page(hwpoison_page);
2804         __free_page(bad_page);
2805 }
2806 EXPORT_SYMBOL_GPL(kvm_exit);