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         int 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         kvm->mmu_notifier_seq++;
 293         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 294         spin_unlock(&kvm->mmu_lock);
 295         srcu_read_unlock(&kvm->srcu, idx);
 296
 297         /* we've to flush the tlb before the pages can be freed */
 298         if (need_tlb_flush)
 299                 kvm_flush_remote_tlbs(kvm);
 300
 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         spin_unlock(&kvm->mmu_lock);
 339         srcu_read_unlock(&kvm->srcu, idx);
 340
 341         /* we've to flush the tlb before the pages can be freed */
 342         if (need_tlb_flush)
 343                 kvm_flush_remote_tlbs(kvm);
 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         /*
 361          * The above sequence increase must be visible before the
 362          * below count decrease but both values are read by the kvm
 363          * page fault under mmu_lock spinlock so we don't need to add
 364          * a smb_wmb() here in between the two.
 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         young = kvm_age_hva(kvm, address);
 382         spin_unlock(&kvm->mmu_lock);
 383         srcu_read_unlock(&kvm->srcu, idx);
 384
 385         if (young)
 386                 kvm_flush_remote_tlbs(kvm);
 387
 388         return young;
 389 }
 390
 391 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 392                                        struct mm_struct *mm,
 393                                        unsigned long address)
 394 {
 395         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 396         int young, idx;
 397
 398         idx = srcu_read_lock(&kvm->srcu);
 399         spin_lock(&kvm->mmu_lock);
 400         young = kvm_test_age_hva(kvm, address);
 401         spin_unlock(&kvm->mmu_lock);
 402         srcu_read_unlock(&kvm->srcu, idx);
 403
 404         return young;
 405 }
 406
 407 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 408                                      struct mm_struct *mm)
 409 {
 410         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 411         int idx;
 412
 413         idx = srcu_read_lock(&kvm->srcu);
 414         kvm_arch_flush_shadow(kvm);
 415         srcu_read_unlock(&kvm->srcu, idx);
 416 }
 417
 418 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 419         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 420         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 421         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 422         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 423         .test_young             = kvm_mmu_notifier_test_young,
 424         .change_pte             = kvm_mmu_notifier_change_pte,
 425         .release                = kvm_mmu_notifier_release,
 426 };
 427
 428 static int kvm_init_mmu_notifier(struct kvm *kvm)
 429 {
 430         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 431         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 432 }
 433
 434 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 435
 436 static int kvm_init_mmu_notifier(struct kvm *kvm)
 437 {
 438         return 0;
 439 }
 440
 441 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 442
 443 static struct kvm *kvm_create_vm(void)
 444 {
 445         int r, i;
 446         struct kvm *kvm = kvm_arch_alloc_vm();
 447
 448         if (!kvm)
 449                 return ERR_PTR(-ENOMEM);
 450
 451         r = kvm_arch_init_vm(kvm);
 452         if (r)
 453                 goto out_err_nodisable;
 454
 455         r = hardware_enable_all();
 456         if (r)
 457                 goto out_err_nodisable;
 458
 459 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 460         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 461         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 462 #endif
 463
 464         r = -ENOMEM;
 465         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 466         if (!kvm->memslots)
 467                 goto out_err_nosrcu;
 468         if (init_srcu_struct(&kvm->srcu))
 469                 goto out_err_nosrcu;
 470         for (i = 0; i < KVM_NR_BUSES; i++) {
 471                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 472                                         GFP_KERNEL);
 473                 if (!kvm->buses[i])
 474                         goto out_err;
 475         }
 476
 477         spin_lock_init(&kvm->mmu_lock);
 478         kvm->mm = current->mm;
 479         atomic_inc(&kvm->mm->mm_count);
 480         kvm_eventfd_init(kvm);
 481         mutex_init(&kvm->lock);
 482         mutex_init(&kvm->irq_lock);
 483         mutex_init(&kvm->slots_lock);
 484         atomic_set(&kvm->users_count, 1);
 485
 486         r = kvm_init_mmu_notifier(kvm);
 487         if (r)
 488                 goto out_err;
 489
 490         raw_spin_lock(&kvm_lock);
 491         list_add(&kvm->vm_list, &vm_list);
 492         raw_spin_unlock(&kvm_lock);
 493
 494         return kvm;
 495
 496 out_err:
 497         cleanup_srcu_struct(&kvm->srcu);
 498 out_err_nosrcu:
 499         hardware_disable_all();
 500 out_err_nodisable:
 501         for (i = 0; i < KVM_NR_BUSES; i++)
 502                 kfree(kvm->buses[i]);
 503         kfree(kvm->memslots);
 504         kvm_arch_free_vm(kvm);
 505         return ERR_PTR(r);
 506 }
 507
 508 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 509 {
 510         if (!memslot->dirty_bitmap)
 511                 return;
 512
 513         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
 514                 vfree(memslot->dirty_bitmap_head);
 515         else
 516                 kfree(memslot->dirty_bitmap_head);
 517
 518         memslot->dirty_bitmap = NULL;
 519         memslot->dirty_bitmap_head = NULL;
 520 }
 521
 522 /*
 523  * Free any memory in @free but not in @dont.
 524  */
 525 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 526                                   struct kvm_memory_slot *dont)
 527 {
 528         int i;
 529
 530         if (!dont || free->rmap != dont->rmap)
 531                 vfree(free->rmap);
 532
 533         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 534                 kvm_destroy_dirty_bitmap(free);
 535
 536
 537         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 538                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 539                         vfree(free->lpage_info[i]);
 540                         free->lpage_info[i] = NULL;
 541                 }
 542         }
 543
 544         free->npages = 0;
 545         free->rmap = NULL;
 546 }
 547
 548 void kvm_free_physmem(struct kvm *kvm)
 549 {
 550         struct kvm_memslots *slots = kvm->memslots;
 551         struct kvm_memory_slot *memslot;
 552
 553         kvm_for_each_memslot(memslot, slots)
 554                 kvm_free_physmem_slot(memslot, NULL);
 555
 556         kfree(kvm->memslots);
 557 }
 558
 559 static void kvm_destroy_vm(struct kvm *kvm)
 560 {
 561         int i;
 562         struct mm_struct *mm = kvm->mm;
 563
 564         kvm_arch_sync_events(kvm);
 565         raw_spin_lock(&kvm_lock);
 566         list_del(&kvm->vm_list);
 567         raw_spin_unlock(&kvm_lock);
 568         kvm_free_irq_routing(kvm);
 569         for (i = 0; i < KVM_NR_BUSES; i++)
 570                 kvm_io_bus_destroy(kvm->buses[i]);
 571         kvm_coalesced_mmio_free(kvm);
 572 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 573         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 574 #else
 575         kvm_arch_flush_shadow(kvm);
 576 #endif
 577         kvm_arch_destroy_vm(kvm);
 578         kvm_free_physmem(kvm);
 579         cleanup_srcu_struct(&kvm->srcu);
 580         kvm_arch_free_vm(kvm);
 581         hardware_disable_all();
 582         mmdrop(mm);
 583 }
 584
 585 void kvm_get_kvm(struct kvm *kvm)
 586 {
 587         atomic_inc(&kvm->users_count);
 588 }
 589 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 590
 591 void kvm_put_kvm(struct kvm *kvm)
 592 {
 593         if (atomic_dec_and_test(&kvm->users_count))
 594                 kvm_destroy_vm(kvm);
 595 }
 596 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 597
 598
 599 static int kvm_vm_release(struct inode *inode, struct file *filp)
 600 {
 601         struct kvm *kvm = filp->private_data;
 602
 603         kvm_irqfd_release(kvm);
 604
 605         kvm_put_kvm(kvm);
 606         return 0;
 607 }
 608
 609 #ifndef CONFIG_S390
 610 /*
 611  * Allocation size is twice as large as the actual dirty bitmap size.
 612  * This makes it possible to do double buffering: see x86's
 613  * kvm_vm_ioctl_get_dirty_log().
 614  */
 615 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 616 {
 617         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 618
 619         if (dirty_bytes > PAGE_SIZE)
 620                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
 621         else
 622                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
 623
 624         if (!memslot->dirty_bitmap)
 625                 return -ENOMEM;
 626
 627         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
 628         memslot->nr_dirty_pages = 0;
 629         return 0;
 630 }
 631 #endif /* !CONFIG_S390 */
 632
 633 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
 634 {
 635         if (new) {
 636                 int id = new->id;
 637
 638                 slots->memslots[id] = *new;
 639                 if (id >= slots->nmemslots)
 640                         slots->nmemslots = id + 1;
 641         }
 642
 643         slots->generation++;
 644 }
 645
 646 /*
 647  * Allocate some memory and give it an address in the guest physical address
 648  * space.
 649  *
 650  * Discontiguous memory is allowed, mostly for framebuffers.
 651  *
 652  * Must be called holding mmap_sem for write.
 653  */
 654 int __kvm_set_memory_region(struct kvm *kvm,
 655                             struct kvm_userspace_memory_region *mem,
 656                             int user_alloc)
 657 {
 658         int r;
 659         gfn_t base_gfn;
 660         unsigned long npages;
 661         unsigned long i;
 662         struct kvm_memory_slot *memslot;
 663         struct kvm_memory_slot old, new;
 664         struct kvm_memslots *slots, *old_memslots;
 665
 666         r = -EINVAL;
 667         /* General sanity checks */
 668         if (mem->memory_size & (PAGE_SIZE - 1))
 669                 goto out;
 670         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 671                 goto out;
 672         /* We can read the guest memory with __xxx_user() later on. */
 673         if (user_alloc &&
 674             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 675              !access_ok(VERIFY_WRITE,
 676                         (void __user *)(unsigned long)mem->userspace_addr,
 677                         mem->memory_size)))
 678                 goto out;
 679         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 680                 goto out;
 681         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 682                 goto out;
 683
 684         memslot = &kvm->memslots->memslots[mem->slot];
 685         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 686         npages = mem->memory_size >> PAGE_SHIFT;
 687
 688         r = -EINVAL;
 689         if (npages > KVM_MEM_MAX_NR_PAGES)
 690                 goto out;
 691
 692         if (!npages)
 693                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 694
 695         new = old = *memslot;
 696
 697         new.id = mem->slot;
 698         new.base_gfn = base_gfn;
 699         new.npages = npages;
 700         new.flags = mem->flags;
 701
 702         /* Disallow changing a memory slot's size. */
 703         r = -EINVAL;
 704         if (npages && old.npages && npages != old.npages)
 705                 goto out_free;
 706
 707         /* Check for overlaps */
 708         r = -EEXIST;
 709         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 710                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 711
 712                 if (s == memslot || !s->npages)
 713                         continue;
 714                 if (!((base_gfn + npages <= s->base_gfn) ||
 715                       (base_gfn >= s->base_gfn + s->npages)))
 716                         goto out_free;
 717         }
 718
 719         /* Free page dirty bitmap if unneeded */
 720         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 721                 new.dirty_bitmap = NULL;
 722
 723         r = -ENOMEM;
 724
 725         /* Allocate if a slot is being created */
 726 #ifndef CONFIG_S390
 727         if (npages && !new.rmap) {
 728                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
 729
 730                 if (!new.rmap)
 731                         goto out_free;
 732
 733                 new.user_alloc = user_alloc;
 734                 new.userspace_addr = mem->userspace_addr;
 735         }
 736         if (!npages)
 737                 goto skip_lpage;
 738
 739         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 740                 unsigned long ugfn;
 741                 unsigned long j;
 742                 int lpages;
 743                 int level = i + 2;
 744
 745                 /* Avoid unused variable warning if no large pages */
 746                 (void)level;
 747
 748                 if (new.lpage_info[i])
 749                         continue;
 750
 751                 lpages = 1 + ((base_gfn + npages - 1)
 752                              >> KVM_HPAGE_GFN_SHIFT(level));
 753                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
 754
 755                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
 756
 757                 if (!new.lpage_info[i])
 758                         goto out_free;
 759
 760                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
 761                         new.lpage_info[i][0].write_count = 1;
 762                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
 763                         new.lpage_info[i][lpages - 1].write_count = 1;
 764                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 765                 /*
 766                  * If the gfn and userspace address are not aligned wrt each
 767                  * other, or if explicitly asked to, disable large page
 768                  * support for this slot
 769                  */
 770                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 771                     !largepages_enabled)
 772                         for (j = 0; j < lpages; ++j)
 773                                 new.lpage_info[i][j].write_count = 1;
 774         }
 775
 776 skip_lpage:
 777
 778         /* Allocate page dirty bitmap if needed */
 779         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 780                 if (kvm_create_dirty_bitmap(&new) < 0)
 781                         goto out_free;
 782                 /* destroy any largepage mappings for dirty tracking */
 783         }
 784 #else  /* not defined CONFIG_S390 */
 785         new.user_alloc = user_alloc;
 786         if (user_alloc)
 787                 new.userspace_addr = mem->userspace_addr;
 788 #endif /* not defined CONFIG_S390 */
 789
 790         if (!npages) {
 791                 r = -ENOMEM;
 792                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 793                                 GFP_KERNEL);
 794                 if (!slots)
 795                         goto out_free;
 796                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 797                 update_memslots(slots, NULL);
 798
 799                 old_memslots = kvm->memslots;
 800                 rcu_assign_pointer(kvm->memslots, slots);
 801                 synchronize_srcu_expedited(&kvm->srcu);
 802                 /* From this point no new shadow pages pointing to a deleted
 803                  * memslot will be created.
 804                  *
 805                  * validation of sp->gfn happens in:
 806                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 807                  *      - kvm_is_visible_gfn (mmu_check_roots)
 808                  */
 809                 kvm_arch_flush_shadow(kvm);
 810                 kfree(old_memslots);
 811         }
 812
 813         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 814         if (r)
 815                 goto out_free;
 816
 817         /* map the pages in iommu page table */
 818         if (npages) {
 819                 r = kvm_iommu_map_pages(kvm, &new);
 820                 if (r)
 821                         goto out_free;
 822         }
 823
 824         r = -ENOMEM;
 825         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 826                         GFP_KERNEL);
 827         if (!slots)
 828                 goto out_free;
 829
 830         /* actual memory is freed via old in kvm_free_physmem_slot below */
 831         if (!npages) {
 832                 new.rmap = NULL;
 833                 new.dirty_bitmap = NULL;
 834                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 835                         new.lpage_info[i] = NULL;
 836         }
 837
 838         update_memslots(slots, &new);
 839         old_memslots = kvm->memslots;
 840         rcu_assign_pointer(kvm->memslots, slots);
 841         synchronize_srcu_expedited(&kvm->srcu);
 842
 843         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 844
 845         /*
 846          * If the new memory slot is created, we need to clear all
 847          * mmio sptes.
 848          */
 849         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 850                 kvm_arch_flush_shadow(kvm);
 851
 852         kvm_free_physmem_slot(&old, &new);
 853         kfree(old_memslots);
 854
 855         return 0;
 856
 857 out_free:
 858         kvm_free_physmem_slot(&new, &old);
 859 out:
 860         return r;
 861
 862 }
 863 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 864
 865 int kvm_set_memory_region(struct kvm *kvm,
 866                           struct kvm_userspace_memory_region *mem,
 867                           int user_alloc)
 868 {
 869         int r;
 870
 871         mutex_lock(&kvm->slots_lock);
 872         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 873         mutex_unlock(&kvm->slots_lock);
 874         return r;
 875 }
 876 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 877
 878 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 879                                    struct
 880                                    kvm_userspace_memory_region *mem,
 881                                    int user_alloc)
 882 {
 883         if (mem->slot >= KVM_MEMORY_SLOTS)
 884                 return -EINVAL;
 885         return kvm_set_memory_region(kvm, mem, user_alloc);
 886 }
 887
 888 int kvm_get_dirty_log(struct kvm *kvm,
 889                         struct kvm_dirty_log *log, int *is_dirty)
 890 {
 891         struct kvm_memory_slot *memslot;
 892         int r, i;
 893         unsigned long n;
 894         unsigned long any = 0;
 895
 896         r = -EINVAL;
 897         if (log->slot >= KVM_MEMORY_SLOTS)
 898                 goto out;
 899
 900         memslot = &kvm->memslots->memslots[log->slot];
 901         r = -ENOENT;
 902         if (!memslot->dirty_bitmap)
 903                 goto out;
 904
 905         n = kvm_dirty_bitmap_bytes(memslot);
 906
 907         for (i = 0; !any && i < n/sizeof(long); ++i)
 908                 any = memslot->dirty_bitmap[i];
 909
 910         r = -EFAULT;
 911         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 912                 goto out;
 913
 914         if (any)
 915                 *is_dirty = 1;
 916
 917         r = 0;
 918 out:
 919         return r;
 920 }
 921
 922 void kvm_disable_largepages(void)
 923 {
 924         largepages_enabled = false;
 925 }
 926 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 927
 928 int is_error_page(struct page *page)
 929 {
 930         return page == bad_page || page == hwpoison_page || page == fault_page;
 931 }
 932 EXPORT_SYMBOL_GPL(is_error_page);
 933
 934 int is_error_pfn(pfn_t pfn)
 935 {
 936         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 937 }
 938 EXPORT_SYMBOL_GPL(is_error_pfn);
 939
 940 int is_hwpoison_pfn(pfn_t pfn)
 941 {
 942         return pfn == hwpoison_pfn;
 943 }
 944 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 945
 946 int is_fault_pfn(pfn_t pfn)
 947 {
 948         return pfn == fault_pfn;
 949 }
 950 EXPORT_SYMBOL_GPL(is_fault_pfn);
 951
 952 int is_noslot_pfn(pfn_t pfn)
 953 {
 954         return pfn == bad_pfn;
 955 }
 956 EXPORT_SYMBOL_GPL(is_noslot_pfn);
 957
 958 int is_invalid_pfn(pfn_t pfn)
 959 {
 960         return pfn == hwpoison_pfn || pfn == fault_pfn;
 961 }
 962 EXPORT_SYMBOL_GPL(is_invalid_pfn);
 963
 964 static inline unsigned long bad_hva(void)
 965 {
 966         return PAGE_OFFSET;
 967 }
 968
 969 int kvm_is_error_hva(unsigned long addr)
 970 {
 971         return addr == bad_hva();
 972 }
 973 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 974
 975 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
 976                                                 gfn_t gfn)
 977 {
 978         struct kvm_memory_slot *memslot;
 979
 980         kvm_for_each_memslot(memslot, slots)
 981                 if (gfn >= memslot->base_gfn
 982                     && gfn < memslot->base_gfn + memslot->npages)
 983                         return memslot;
 984
 985         return NULL;
 986 }
 987
 988 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 989 {
 990         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 991 }
 992 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 993
 994 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 995 {
 996         int i;
 997         struct kvm_memslots *slots = kvm_memslots(kvm);
 998
 999         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1000                 struct kvm_memory_slot *memslot = &slots->memslots[i];
1001
1002                 if (memslot->flags & KVM_MEMSLOT_INVALID)
1003                         continue;
1004
1005                 if (gfn >= memslot->base_gfn
1006                     && gfn < memslot->base_gfn + memslot->npages)
1007                         return 1;
1008         }
1009         return 0;
1010 }
1011 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1012
1013 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1014 {
1015         struct vm_area_struct *vma;
1016         unsigned long addr, size;
1017
1018         size = PAGE_SIZE;
1019
1020         addr = gfn_to_hva(kvm, gfn);
1021         if (kvm_is_error_hva(addr))
1022                 return PAGE_SIZE;
1023
1024         down_read(&current->mm->mmap_sem);
1025         vma = find_vma(current->mm, addr);
1026         if (!vma)
1027                 goto out;
1028
1029         size = vma_kernel_pagesize(vma);
1030
1031 out:
1032         up_read(&current->mm->mmap_sem);
1033
1034         return size;
1035 }
1036
1037 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1038                                      gfn_t *nr_pages)
1039 {
1040         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1041                 return bad_hva();
1042
1043         if (nr_pages)
1044                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1045
1046         return gfn_to_hva_memslot(slot, gfn);
1047 }
1048
1049 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1050 {
1051         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1052 }
1053 EXPORT_SYMBOL_GPL(gfn_to_hva);
1054
1055 static pfn_t get_fault_pfn(void)
1056 {
1057         get_page(fault_page);
1058         return fault_pfn;
1059 }
1060
1061 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1062         unsigned long start, int write, struct page **page)
1063 {
1064         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1065
1066         if (write)
1067                 flags |= FOLL_WRITE;
1068
1069         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1070 }
1071
1072 static inline int check_user_page_hwpoison(unsigned long addr)
1073 {
1074         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1075
1076         rc = __get_user_pages(current, current->mm, addr, 1,
1077                               flags, NULL, NULL, NULL);
1078         return rc == -EHWPOISON;
1079 }
1080
1081 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1082                         bool *async, bool write_fault, bool *writable)
1083 {
1084         struct page *page[1];
1085         int npages = 0;
1086         pfn_t pfn;
1087
1088         /* we can do it either atomically or asynchronously, not both */
1089         BUG_ON(atomic && async);
1090
1091         BUG_ON(!write_fault && !writable);
1092
1093         if (writable)
1094                 *writable = true;
1095
1096         if (atomic || async)
1097                 npages = __get_user_pages_fast(addr, 1, 1, page);
1098
1099         if (unlikely(npages != 1) && !atomic) {
1100                 might_sleep();
1101
1102                 if (writable)
1103                         *writable = write_fault;
1104
1105                 if (async) {
1106                         down_read(&current->mm->mmap_sem);
1107                         npages = get_user_page_nowait(current, current->mm,
1108                                                      addr, write_fault, page);
1109                         up_read(&current->mm->mmap_sem);
1110                 } else
1111                         npages = get_user_pages_fast(addr, 1, write_fault,
1112                                                      page);
1113
1114                 /* map read fault as writable if possible */
1115                 if (unlikely(!write_fault) && npages == 1) {
1116                         struct page *wpage[1];
1117
1118                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1119                         if (npages == 1) {
1120                                 *writable = true;
1121                                 put_page(page[0]);
1122                                 page[0] = wpage[0];
1123                         }
1124                         npages = 1;
1125                 }
1126         }
1127
1128         if (unlikely(npages != 1)) {
1129                 struct vm_area_struct *vma;
1130
1131                 if (atomic)
1132                         return get_fault_pfn();
1133
1134                 down_read(&current->mm->mmap_sem);
1135                 if (npages == -EHWPOISON ||
1136                         (!async && check_user_page_hwpoison(addr))) {
1137                         up_read(&current->mm->mmap_sem);
1138                         get_page(hwpoison_page);
1139                         return page_to_pfn(hwpoison_page);
1140                 }
1141
1142                 vma = find_vma_intersection(current->mm, addr, addr+1);
1143
1144                 if (vma == NULL)
1145                         pfn = get_fault_pfn();
1146                 else if ((vma->vm_flags & VM_PFNMAP)) {
1147                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1148                                 vma->vm_pgoff;
1149                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1150                 } else {
1151                         if (async && (vma->vm_flags & VM_WRITE))
1152                                 *async = true;
1153                         pfn = get_fault_pfn();
1154                 }
1155                 up_read(&current->mm->mmap_sem);
1156         } else
1157                 pfn = page_to_pfn(page[0]);
1158
1159         return pfn;
1160 }
1161
1162 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1163 {
1164         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1165 }
1166 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1167
1168 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1169                           bool write_fault, bool *writable)
1170 {
1171         unsigned long addr;
1172
1173         if (async)
1174                 *async = false;
1175
1176         addr = gfn_to_hva(kvm, gfn);
1177         if (kvm_is_error_hva(addr)) {
1178                 get_page(bad_page);
1179                 return page_to_pfn(bad_page);
1180         }
1181
1182         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1183 }
1184
1185 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1186 {
1187         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1188 }
1189 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1190
1191 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1192                        bool write_fault, bool *writable)
1193 {
1194         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1195 }
1196 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1197
1198 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1199 {
1200         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1201 }
1202 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1203
1204 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1205                       bool *writable)
1206 {
1207         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1208 }
1209 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1210
1211 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1212                          struct kvm_memory_slot *slot, gfn_t gfn)
1213 {
1214         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1215         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1216 }
1217
1218 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1219                                                                   int nr_pages)
1220 {
1221         unsigned long addr;
1222         gfn_t entry;
1223
1224         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1225         if (kvm_is_error_hva(addr))
1226                 return -1;
1227
1228         if (entry < nr_pages)
1229                 return 0;
1230
1231         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1232 }
1233 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1234
1235 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1236 {
1237         pfn_t pfn;
1238
1239         pfn = gfn_to_pfn(kvm, gfn);
1240         if (!kvm_is_mmio_pfn(pfn))
1241                 return pfn_to_page(pfn);
1242
1243         WARN_ON(kvm_is_mmio_pfn(pfn));
1244
1245         get_page(bad_page);
1246         return bad_page;
1247 }
1248
1249 EXPORT_SYMBOL_GPL(gfn_to_page);
1250
1251 void kvm_release_page_clean(struct page *page)
1252 {
1253         kvm_release_pfn_clean(page_to_pfn(page));
1254 }
1255 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1256
1257 void kvm_release_pfn_clean(pfn_t pfn)
1258 {
1259         if (!kvm_is_mmio_pfn(pfn))
1260                 put_page(pfn_to_page(pfn));
1261 }
1262 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1263
1264 void kvm_release_page_dirty(struct page *page)
1265 {
1266         kvm_release_pfn_dirty(page_to_pfn(page));
1267 }
1268 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1269
1270 void kvm_release_pfn_dirty(pfn_t pfn)
1271 {
1272         kvm_set_pfn_dirty(pfn);
1273         kvm_release_pfn_clean(pfn);
1274 }
1275 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1276
1277 void kvm_set_page_dirty(struct page *page)
1278 {
1279         kvm_set_pfn_dirty(page_to_pfn(page));
1280 }
1281 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1282
1283 void kvm_set_pfn_dirty(pfn_t pfn)
1284 {
1285         if (!kvm_is_mmio_pfn(pfn)) {
1286                 struct page *page = pfn_to_page(pfn);
1287                 if (!PageReserved(page))
1288                         SetPageDirty(page);
1289         }
1290 }
1291 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1292
1293 void kvm_set_pfn_accessed(pfn_t pfn)
1294 {
1295         if (!kvm_is_mmio_pfn(pfn))
1296                 mark_page_accessed(pfn_to_page(pfn));
1297 }
1298 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1299
1300 void kvm_get_pfn(pfn_t pfn)
1301 {
1302         if (!kvm_is_mmio_pfn(pfn))
1303                 get_page(pfn_to_page(pfn));
1304 }
1305 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1306
1307 static int next_segment(unsigned long len, int offset)
1308 {
1309         if (len > PAGE_SIZE - offset)
1310                 return PAGE_SIZE - offset;
1311         else
1312                 return len;
1313 }
1314
1315 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1316                         int len)
1317 {
1318         int r;
1319         unsigned long addr;
1320
1321         addr = gfn_to_hva(kvm, gfn);
1322         if (kvm_is_error_hva(addr))
1323                 return -EFAULT;
1324         r = __copy_from_user(data, (void __user *)addr + offset, len);
1325         if (r)
1326                 return -EFAULT;
1327         return 0;
1328 }
1329 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1330
1331 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1332 {
1333         gfn_t gfn = gpa >> PAGE_SHIFT;
1334         int seg;
1335         int offset = offset_in_page(gpa);
1336         int ret;
1337
1338         while ((seg = next_segment(len, offset)) != 0) {
1339                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1340                 if (ret < 0)
1341                         return ret;
1342                 offset = 0;
1343                 len -= seg;
1344                 data += seg;
1345                 ++gfn;
1346         }
1347         return 0;
1348 }
1349 EXPORT_SYMBOL_GPL(kvm_read_guest);
1350
1351 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1352                           unsigned long len)
1353 {
1354         int r;
1355         unsigned long addr;
1356         gfn_t gfn = gpa >> PAGE_SHIFT;
1357         int offset = offset_in_page(gpa);
1358
1359         addr = gfn_to_hva(kvm, gfn);
1360         if (kvm_is_error_hva(addr))
1361                 return -EFAULT;
1362         pagefault_disable();
1363         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1364         pagefault_enable();
1365         if (r)
1366                 return -EFAULT;
1367         return 0;
1368 }
1369 EXPORT_SYMBOL(kvm_read_guest_atomic);
1370
1371 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1372                          int offset, int len)
1373 {
1374         int r;
1375         unsigned long addr;
1376
1377         addr = gfn_to_hva(kvm, gfn);
1378         if (kvm_is_error_hva(addr))
1379                 return -EFAULT;
1380         r = __copy_to_user((void __user *)addr + offset, data, len);
1381         if (r)
1382                 return -EFAULT;
1383         mark_page_dirty(kvm, gfn);
1384         return 0;
1385 }
1386 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1387
1388 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1389                     unsigned long len)
1390 {
1391         gfn_t gfn = gpa >> PAGE_SHIFT;
1392         int seg;
1393         int offset = offset_in_page(gpa);
1394         int ret;
1395
1396         while ((seg = next_segment(len, offset)) != 0) {
1397                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1398                 if (ret < 0)
1399                         return ret;
1400                 offset = 0;
1401                 len -= seg;
1402                 data += seg;
1403                 ++gfn;
1404         }
1405         return 0;
1406 }
1407
1408 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1409                               gpa_t gpa)
1410 {
1411         struct kvm_memslots *slots = kvm_memslots(kvm);
1412         int offset = offset_in_page(gpa);
1413         gfn_t gfn = gpa >> PAGE_SHIFT;
1414
1415         ghc->gpa = gpa;
1416         ghc->generation = slots->generation;
1417         ghc->memslot = __gfn_to_memslot(slots, gfn);
1418         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1419         if (!kvm_is_error_hva(ghc->hva))
1420                 ghc->hva += offset;
1421         else
1422                 return -EFAULT;
1423
1424         return 0;
1425 }
1426 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1427
1428 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1429                            void *data, unsigned long len)
1430 {
1431         struct kvm_memslots *slots = kvm_memslots(kvm);
1432         int r;
1433
1434         if (slots->generation != ghc->generation)
1435                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1436
1437         if (kvm_is_error_hva(ghc->hva))
1438                 return -EFAULT;
1439
1440         r = __copy_to_user((void __user *)ghc->hva, data, len);
1441         if (r)
1442                 return -EFAULT;
1443         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1444
1445         return 0;
1446 }
1447 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1448
1449 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1450                            void *data, unsigned long len)
1451 {
1452         struct kvm_memslots *slots = kvm_memslots(kvm);
1453         int r;
1454
1455         if (slots->generation != ghc->generation)
1456                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1457
1458         if (kvm_is_error_hva(ghc->hva))
1459                 return -EFAULT;
1460
1461         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1462         if (r)
1463                 return -EFAULT;
1464
1465         return 0;
1466 }
1467 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1468
1469 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1470 {
1471         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1472                                     offset, len);
1473 }
1474 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1475
1476 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1477 {
1478         gfn_t gfn = gpa >> PAGE_SHIFT;
1479         int seg;
1480         int offset = offset_in_page(gpa);
1481         int ret;
1482
1483         while ((seg = next_segment(len, offset)) != 0) {
1484                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1485                 if (ret < 0)
1486                         return ret;
1487                 offset = 0;
1488                 len -= seg;
1489                 ++gfn;
1490         }
1491         return 0;
1492 }
1493 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1494
1495 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1496                              gfn_t gfn)
1497 {
1498         if (memslot && memslot->dirty_bitmap) {
1499                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1500
1501                 if (!__test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap))
1502                         memslot->nr_dirty_pages++;
1503         }
1504 }
1505
1506 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1507 {
1508         struct kvm_memory_slot *memslot;
1509
1510         memslot = gfn_to_memslot(kvm, gfn);
1511         mark_page_dirty_in_slot(kvm, memslot, gfn);
1512 }
1513
1514 /*
1515  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1516  */
1517 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1518 {
1519         DEFINE_WAIT(wait);
1520
1521         for (;;) {
1522                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1523
1524                 if (kvm_arch_vcpu_runnable(vcpu)) {
1525                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1526                         break;
1527                 }
1528                 if (kvm_cpu_has_pending_timer(vcpu))
1529                         break;
1530                 if (signal_pending(current))
1531                         break;
1532
1533                 schedule();
1534         }
1535
1536         finish_wait(&vcpu->wq, &wait);
1537 }
1538
1539 void kvm_resched(struct kvm_vcpu *vcpu)
1540 {
1541         if (!need_resched())
1542                 return;
1543         cond_resched();
1544 }
1545 EXPORT_SYMBOL_GPL(kvm_resched);
1546
1547 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1548 {
1549         struct kvm *kvm = me->kvm;
1550         struct kvm_vcpu *vcpu;
1551         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1552         int yielded = 0;
1553         int pass;
1554         int i;
1555
1556         /*
1557          * We boost the priority of a VCPU that is runnable but not
1558          * currently running, because it got preempted by something
1559          * else and called schedule in __vcpu_run.  Hopefully that
1560          * VCPU is holding the lock that we need and will release it.
1561          * We approximate round-robin by starting at the last boosted VCPU.
1562          */
1563         for (pass = 0; pass < 2 && !yielded; pass++) {
1564                 kvm_for_each_vcpu(i, vcpu, kvm) {
1565                         struct task_struct *task = NULL;
1566                         struct pid *pid;
1567                         if (!pass && i < last_boosted_vcpu) {
1568                                 i = last_boosted_vcpu;
1569                                 continue;
1570                         } else if (pass && i > last_boosted_vcpu)
1571                                 break;
1572                         if (vcpu == me)
1573                                 continue;
1574                         if (waitqueue_active(&vcpu->wq))
1575                                 continue;
1576                         rcu_read_lock();
1577                         pid = rcu_dereference(vcpu->pid);
1578                         if (pid)
1579                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1580                         rcu_read_unlock();
1581                         if (!task)
1582                                 continue;
1583                         if (task->flags & PF_VCPU) {
1584                                 put_task_struct(task);
1585                                 continue;
1586                         }
1587                         if (yield_to(task, 1)) {
1588                                 put_task_struct(task);
1589                                 kvm->last_boosted_vcpu = i;
1590                                 yielded = 1;
1591                                 break;
1592                         }
1593                         put_task_struct(task);
1594                 }
1595         }
1596 }
1597 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1598
1599 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1600 {
1601         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1602         struct page *page;
1603
1604         if (vmf->pgoff == 0)
1605                 page = virt_to_page(vcpu->run);
1606 #ifdef CONFIG_X86
1607         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1608                 page = virt_to_page(vcpu->arch.pio_data);
1609 #endif
1610 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1611         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1612                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1613 #endif
1614         else
1615                 return VM_FAULT_SIGBUS;
1616         get_page(page);
1617         vmf->page = page;
1618         return 0;
1619 }
1620
1621 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1622         .fault = kvm_vcpu_fault,
1623 };
1624
1625 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1626 {
1627         vma->vm_ops = &kvm_vcpu_vm_ops;
1628         return 0;
1629 }
1630
1631 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1632 {
1633         struct kvm_vcpu *vcpu = filp->private_data;
1634
1635         kvm_put_kvm(vcpu->kvm);
1636         return 0;
1637 }
1638
1639 static struct file_operations kvm_vcpu_fops = {
1640         .release        = kvm_vcpu_release,
1641         .unlocked_ioctl = kvm_vcpu_ioctl,
1642 #ifdef CONFIG_COMPAT
1643         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1644 #endif
1645         .mmap           = kvm_vcpu_mmap,
1646         .llseek         = noop_llseek,
1647 };
1648
1649 /*
1650  * Allocates an inode for the vcpu.
1651  */
1652 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1653 {
1654         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1655 }
1656
1657 /*
1658  * Creates some virtual cpus.  Good luck creating more than one.
1659  */
1660 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1661 {
1662         int r;
1663         struct kvm_vcpu *vcpu, *v;
1664
1665         vcpu = kvm_arch_vcpu_create(kvm, id);
1666         if (IS_ERR(vcpu))
1667                 return PTR_ERR(vcpu);
1668
1669         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1670
1671         r = kvm_arch_vcpu_setup(vcpu);
1672         if (r)
1673                 goto vcpu_destroy;
1674
1675         mutex_lock(&kvm->lock);
1676         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1677                 r = -EINVAL;
1678                 goto unlock_vcpu_destroy;
1679         }
1680
1681         kvm_for_each_vcpu(r, v, kvm)
1682                 if (v->vcpu_id == id) {
1683                         r = -EEXIST;
1684                         goto unlock_vcpu_destroy;
1685                 }
1686
1687         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1688
1689         /* Now it's all set up, let userspace reach it */
1690         kvm_get_kvm(kvm);
1691         r = create_vcpu_fd(vcpu);
1692         if (r < 0) {
1693                 kvm_put_kvm(kvm);
1694                 goto unlock_vcpu_destroy;
1695         }
1696
1697         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1698         smp_wmb();
1699         atomic_inc(&kvm->online_vcpus);
1700
1701 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1702         if (kvm->bsp_vcpu_id == id)
1703                 kvm->bsp_vcpu = vcpu;
1704 #endif
1705         mutex_unlock(&kvm->lock);
1706         return r;
1707
1708 unlock_vcpu_destroy:
1709         mutex_unlock(&kvm->lock);
1710 vcpu_destroy:
1711         kvm_arch_vcpu_destroy(vcpu);
1712         return r;
1713 }
1714
1715 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1716 {
1717         if (sigset) {
1718                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1719                 vcpu->sigset_active = 1;
1720                 vcpu->sigset = *sigset;
1721         } else
1722                 vcpu->sigset_active = 0;
1723         return 0;
1724 }
1725
1726 static long kvm_vcpu_ioctl(struct file *filp,
1727                            unsigned int ioctl, unsigned long arg)
1728 {
1729         struct kvm_vcpu *vcpu = filp->private_data;
1730         void __user *argp = (void __user *)arg;
1731         int r;
1732         struct kvm_fpu *fpu = NULL;
1733         struct kvm_sregs *kvm_sregs = NULL;
1734
1735         if (vcpu->kvm->mm != current->mm)
1736                 return -EIO;
1737
1738 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1739         /*
1740          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1741          * so vcpu_load() would break it.
1742          */
1743         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1744                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1745 #endif
1746
1747
1748         vcpu_load(vcpu);
1749         switch (ioctl) {
1750         case KVM_RUN:
1751                 r = -EINVAL;
1752                 if (arg)
1753                         goto out;
1754                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1755                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1756                 break;
1757         case KVM_GET_REGS: {
1758                 struct kvm_regs *kvm_regs;
1759
1760                 r = -ENOMEM;
1761                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1762                 if (!kvm_regs)
1763                         goto out;
1764                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1765                 if (r)
1766                         goto out_free1;
1767                 r = -EFAULT;
1768                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1769                         goto out_free1;
1770                 r = 0;
1771 out_free1:
1772                 kfree(kvm_regs);
1773                 break;
1774         }
1775         case KVM_SET_REGS: {
1776                 struct kvm_regs *kvm_regs;
1777
1778                 r = -ENOMEM;
1779                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1780                 if (!kvm_regs)
1781                         goto out;
1782                 r = -EFAULT;
1783                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1784                         goto out_free2;
1785                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1786                 if (r)
1787                         goto out_free2;
1788                 r = 0;
1789 out_free2:
1790                 kfree(kvm_regs);
1791                 break;
1792         }
1793         case KVM_GET_SREGS: {
1794                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1795                 r = -ENOMEM;
1796                 if (!kvm_sregs)
1797                         goto out;
1798                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1799                 if (r)
1800                         goto out;
1801                 r = -EFAULT;
1802                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1803                         goto out;
1804                 r = 0;
1805                 break;
1806         }
1807         case KVM_SET_SREGS: {
1808                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1809                 r = -ENOMEM;
1810                 if (!kvm_sregs)
1811                         goto out;
1812                 r = -EFAULT;
1813                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1814                         goto out;
1815                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1816                 if (r)
1817                         goto out;
1818                 r = 0;
1819                 break;
1820         }
1821         case KVM_GET_MP_STATE: {
1822                 struct kvm_mp_state mp_state;
1823
1824                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1825                 if (r)
1826                         goto out;
1827                 r = -EFAULT;
1828                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1829                         goto out;
1830                 r = 0;
1831                 break;
1832         }
1833         case KVM_SET_MP_STATE: {
1834                 struct kvm_mp_state mp_state;
1835
1836                 r = -EFAULT;
1837                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1838                         goto out;
1839                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1840                 if (r)
1841                         goto out;
1842                 r = 0;
1843                 break;
1844         }
1845         case KVM_TRANSLATE: {
1846                 struct kvm_translation tr;
1847
1848                 r = -EFAULT;
1849                 if (copy_from_user(&tr, argp, sizeof tr))
1850                         goto out;
1851                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1852                 if (r)
1853                         goto out;
1854                 r = -EFAULT;
1855                 if (copy_to_user(argp, &tr, sizeof tr))
1856                         goto out;
1857                 r = 0;
1858                 break;
1859         }
1860         case KVM_SET_GUEST_DEBUG: {
1861                 struct kvm_guest_debug dbg;
1862
1863                 r = -EFAULT;
1864                 if (copy_from_user(&dbg, argp, sizeof dbg))
1865                         goto out;
1866                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1867                 if (r)
1868                         goto out;
1869                 r = 0;
1870                 break;
1871         }
1872         case KVM_SET_SIGNAL_MASK: {
1873                 struct kvm_signal_mask __user *sigmask_arg = argp;
1874                 struct kvm_signal_mask kvm_sigmask;
1875                 sigset_t sigset, *p;
1876
1877                 p = NULL;
1878                 if (argp) {
1879                         r = -EFAULT;
1880                         if (copy_from_user(&kvm_sigmask, argp,
1881                                            sizeof kvm_sigmask))
1882                                 goto out;
1883                         r = -EINVAL;
1884                         if (kvm_sigmask.len != sizeof sigset)
1885                                 goto out;
1886                         r = -EFAULT;
1887                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1888                                            sizeof sigset))
1889                                 goto out;
1890                         p = &sigset;
1891                 }
1892                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1893                 break;
1894         }
1895         case KVM_GET_FPU: {
1896                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1897                 r = -ENOMEM;
1898                 if (!fpu)
1899                         goto out;
1900                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1901                 if (r)
1902                         goto out;
1903                 r = -EFAULT;
1904                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1905                         goto out;
1906                 r = 0;
1907                 break;
1908         }
1909         case KVM_SET_FPU: {
1910                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1911                 r = -ENOMEM;
1912                 if (!fpu)
1913                         goto out;
1914                 r = -EFAULT;
1915                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1916                         goto out;
1917                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1918                 if (r)
1919                         goto out;
1920                 r = 0;
1921                 break;
1922         }
1923         default:
1924                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1925         }
1926 out:
1927         vcpu_put(vcpu);
1928         kfree(fpu);
1929         kfree(kvm_sregs);
1930         return r;
1931 }
1932
1933 #ifdef CONFIG_COMPAT
1934 static long kvm_vcpu_compat_ioctl(struct file *filp,
1935                                   unsigned int ioctl, unsigned long arg)
1936 {
1937         struct kvm_vcpu *vcpu = filp->private_data;
1938         void __user *argp = compat_ptr(arg);
1939         int r;
1940
1941         if (vcpu->kvm->mm != current->mm)
1942                 return -EIO;
1943
1944         switch (ioctl) {
1945         case KVM_SET_SIGNAL_MASK: {
1946                 struct kvm_signal_mask __user *sigmask_arg = argp;
1947                 struct kvm_signal_mask kvm_sigmask;
1948                 compat_sigset_t csigset;
1949                 sigset_t sigset;
1950
1951                 if (argp) {
1952                         r = -EFAULT;
1953                         if (copy_from_user(&kvm_sigmask, argp,
1954                                            sizeof kvm_sigmask))
1955                                 goto out;
1956                         r = -EINVAL;
1957                         if (kvm_sigmask.len != sizeof csigset)
1958                                 goto out;
1959                         r = -EFAULT;
1960                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1961                                            sizeof csigset))
1962                                 goto out;
1963                 }
1964                 sigset_from_compat(&sigset, &csigset);
1965                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1966                 break;
1967         }
1968         default:
1969                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1970         }
1971
1972 out:
1973         return r;
1974 }
1975 #endif
1976
1977 static long kvm_vm_ioctl(struct file *filp,
1978                            unsigned int ioctl, unsigned long arg)
1979 {
1980         struct kvm *kvm = filp->private_data;
1981         void __user *argp = (void __user *)arg;
1982         int r;
1983
1984         if (kvm->mm != current->mm)
1985                 return -EIO;
1986         switch (ioctl) {
1987         case KVM_CREATE_VCPU:
1988                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1989                 if (r < 0)
1990                         goto out;
1991                 break;
1992         case KVM_SET_USER_MEMORY_REGION: {
1993                 struct kvm_userspace_memory_region kvm_userspace_mem;
1994
1995                 r = -EFAULT;
1996                 if (copy_from_user(&kvm_userspace_mem, argp,
1997                                                 sizeof kvm_userspace_mem))
1998                         goto out;
1999
2000                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2001                 if (r)
2002                         goto out;
2003                 break;
2004         }
2005         case KVM_GET_DIRTY_LOG: {
2006                 struct kvm_dirty_log log;
2007
2008                 r = -EFAULT;
2009                 if (copy_from_user(&log, argp, sizeof log))
2010                         goto out;
2011                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2012                 if (r)
2013                         goto out;
2014                 break;
2015         }
2016 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2017         case KVM_REGISTER_COALESCED_MMIO: {
2018                 struct kvm_coalesced_mmio_zone zone;
2019                 r = -EFAULT;
2020                 if (copy_from_user(&zone, argp, sizeof zone))
2021                         goto out;
2022                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2023                 if (r)
2024                         goto out;
2025                 r = 0;
2026                 break;
2027         }
2028         case KVM_UNREGISTER_COALESCED_MMIO: {
2029                 struct kvm_coalesced_mmio_zone zone;
2030                 r = -EFAULT;
2031                 if (copy_from_user(&zone, argp, sizeof zone))
2032                         goto out;
2033                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2034                 if (r)
2035                         goto out;
2036                 r = 0;
2037                 break;
2038         }
2039 #endif
2040         case KVM_IRQFD: {
2041                 struct kvm_irqfd data;
2042
2043                 r = -EFAULT;
2044                 if (copy_from_user(&data, argp, sizeof data))
2045                         goto out;
2046                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2047                 break;
2048         }
2049         case KVM_IOEVENTFD: {
2050                 struct kvm_ioeventfd data;
2051
2052                 r = -EFAULT;
2053                 if (copy_from_user(&data, argp, sizeof data))
2054                         goto out;
2055                 r = kvm_ioeventfd(kvm, &data);
2056                 break;
2057         }
2058 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2059         case KVM_SET_BOOT_CPU_ID:
2060                 r = 0;
2061                 mutex_lock(&kvm->lock);
2062                 if (atomic_read(&kvm->online_vcpus) != 0)
2063                         r = -EBUSY;
2064                 else
2065                         kvm->bsp_vcpu_id = arg;
2066                 mutex_unlock(&kvm->lock);
2067                 break;
2068 #endif
2069         default:
2070                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2071                 if (r == -ENOTTY)
2072                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2073         }
2074 out:
2075         return r;
2076 }
2077
2078 #ifdef CONFIG_COMPAT
2079 struct compat_kvm_dirty_log {
2080         __u32 slot;
2081         __u32 padding1;
2082         union {
2083                 compat_uptr_t dirty_bitmap; /* one bit per page */
2084                 __u64 padding2;
2085         };
2086 };
2087
2088 static long kvm_vm_compat_ioctl(struct file *filp,
2089                            unsigned int ioctl, unsigned long arg)
2090 {
2091         struct kvm *kvm = filp->private_data;
2092         int r;
2093
2094         if (kvm->mm != current->mm)
2095                 return -EIO;
2096         switch (ioctl) {
2097         case KVM_GET_DIRTY_LOG: {
2098                 struct compat_kvm_dirty_log compat_log;
2099                 struct kvm_dirty_log log;
2100
2101                 r = -EFAULT;
2102                 if (copy_from_user(&compat_log, (void __user *)arg,
2103                                    sizeof(compat_log)))
2104                         goto out;
2105                 log.slot         = compat_log.slot;
2106                 log.padding1     = compat_log.padding1;
2107                 log.padding2     = compat_log.padding2;
2108                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2109
2110                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2111                 if (r)
2112                         goto out;
2113                 break;
2114         }
2115         default:
2116                 r = kvm_vm_ioctl(filp, ioctl, arg);
2117         }
2118
2119 out:
2120         return r;
2121 }
2122 #endif
2123
2124 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2125 {
2126         struct page *page[1];
2127         unsigned long addr;
2128         int npages;
2129         gfn_t gfn = vmf->pgoff;
2130         struct kvm *kvm = vma->vm_file->private_data;
2131
2132         addr = gfn_to_hva(kvm, gfn);
2133         if (kvm_is_error_hva(addr))
2134                 return VM_FAULT_SIGBUS;
2135
2136         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2137                                 NULL);
2138         if (unlikely(npages != 1))
2139                 return VM_FAULT_SIGBUS;
2140
2141         vmf->page = page[0];
2142         return 0;
2143 }
2144
2145 static const struct vm_operations_struct kvm_vm_vm_ops = {
2146         .fault = kvm_vm_fault,
2147 };
2148
2149 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2150 {
2151         vma->vm_ops = &kvm_vm_vm_ops;
2152         return 0;
2153 }
2154
2155 static struct file_operations kvm_vm_fops = {
2156         .release        = kvm_vm_release,
2157         .unlocked_ioctl = kvm_vm_ioctl,
2158 #ifdef CONFIG_COMPAT
2159         .compat_ioctl   = kvm_vm_compat_ioctl,
2160 #endif
2161         .mmap           = kvm_vm_mmap,
2162         .llseek         = noop_llseek,
2163 };
2164
2165 static int kvm_dev_ioctl_create_vm(void)
2166 {
2167         int r;
2168         struct kvm *kvm;
2169
2170         kvm = kvm_create_vm();
2171         if (IS_ERR(kvm))
2172                 return PTR_ERR(kvm);
2173 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2174         r = kvm_coalesced_mmio_init(kvm);
2175         if (r < 0) {
2176                 kvm_put_kvm(kvm);
2177                 return r;
2178         }
2179 #endif
2180         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2181         if (r < 0)
2182                 kvm_put_kvm(kvm);
2183
2184         return r;
2185 }
2186
2187 static long kvm_dev_ioctl_check_extension_generic(long arg)
2188 {
2189         switch (arg) {
2190         case KVM_CAP_USER_MEMORY:
2191         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2192         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2193 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2194         case KVM_CAP_SET_BOOT_CPU_ID:
2195 #endif
2196         case KVM_CAP_INTERNAL_ERROR_DATA:
2197                 return 1;
2198 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2199         case KVM_CAP_IRQ_ROUTING:
2200                 return KVM_MAX_IRQ_ROUTES;
2201 #endif
2202         default:
2203                 break;
2204         }
2205         return kvm_dev_ioctl_check_extension(arg);
2206 }
2207
2208 static long kvm_dev_ioctl(struct file *filp,
2209                           unsigned int ioctl, unsigned long arg)
2210 {
2211         long r = -EINVAL;
2212
2213         switch (ioctl) {
2214         case KVM_GET_API_VERSION:
2215                 r = -EINVAL;
2216                 if (arg)
2217                         goto out;
2218                 r = KVM_API_VERSION;
2219                 break;
2220         case KVM_CREATE_VM:
2221                 r = -EINVAL;
2222                 if (arg)
2223                         goto out;
2224                 r = kvm_dev_ioctl_create_vm();
2225                 break;
2226         case KVM_CHECK_EXTENSION:
2227                 r = kvm_dev_ioctl_check_extension_generic(arg);
2228                 break;
2229         case KVM_GET_VCPU_MMAP_SIZE:
2230                 r = -EINVAL;
2231                 if (arg)
2232                         goto out;
2233                 r = PAGE_SIZE;     /* struct kvm_run */
2234 #ifdef CONFIG_X86
2235                 r += PAGE_SIZE;    /* pio data page */
2236 #endif
2237 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2238                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2239 #endif
2240                 break;
2241         case KVM_TRACE_ENABLE:
2242         case KVM_TRACE_PAUSE:
2243         case KVM_TRACE_DISABLE:
2244                 r = -EOPNOTSUPP;
2245                 break;
2246         default:
2247                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2248         }
2249 out:
2250         return r;
2251 }
2252
2253 static struct file_operations kvm_chardev_ops = {
2254         .unlocked_ioctl = kvm_dev_ioctl,
2255         .compat_ioctl   = kvm_dev_ioctl,
2256         .llseek         = noop_llseek,
2257 };
2258
2259 static struct miscdevice kvm_dev = {
2260         KVM_MINOR,
2261         "kvm",
2262         &kvm_chardev_ops,
2263 };
2264
2265 static void hardware_enable_nolock(void *junk)
2266 {
2267         int cpu = raw_smp_processor_id();
2268         int r;
2269
2270         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2271                 return;
2272
2273         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2274
2275         r = kvm_arch_hardware_enable(NULL);
2276
2277         if (r) {
2278                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2279                 atomic_inc(&hardware_enable_failed);
2280                 printk(KERN_INFO "kvm: enabling virtualization on "
2281                                  "CPU%d failed\n", cpu);
2282         }
2283 }
2284
2285 static void hardware_enable(void *junk)
2286 {
2287         raw_spin_lock(&kvm_lock);
2288         hardware_enable_nolock(junk);
2289         raw_spin_unlock(&kvm_lock);
2290 }
2291
2292 static void hardware_disable_nolock(void *junk)
2293 {
2294         int cpu = raw_smp_processor_id();
2295
2296         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2297                 return;
2298         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2299         kvm_arch_hardware_disable(NULL);
2300 }
2301
2302 static void hardware_disable(void *junk)
2303 {
2304         raw_spin_lock(&kvm_lock);
2305         hardware_disable_nolock(junk);
2306         raw_spin_unlock(&kvm_lock);
2307 }
2308
2309 static void hardware_disable_all_nolock(void)
2310 {
2311         BUG_ON(!kvm_usage_count);
2312
2313         kvm_usage_count--;
2314         if (!kvm_usage_count)
2315                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2316 }
2317
2318 static void hardware_disable_all(void)
2319 {
2320         raw_spin_lock(&kvm_lock);
2321         hardware_disable_all_nolock();
2322         raw_spin_unlock(&kvm_lock);
2323 }
2324
2325 static int hardware_enable_all(void)
2326 {
2327         int r = 0;
2328
2329         raw_spin_lock(&kvm_lock);
2330
2331         kvm_usage_count++;
2332         if (kvm_usage_count == 1) {
2333                 atomic_set(&hardware_enable_failed, 0);
2334                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2335
2336                 if (atomic_read(&hardware_enable_failed)) {
2337                         hardware_disable_all_nolock();
2338                         r = -EBUSY;
2339                 }
2340         }
2341
2342         raw_spin_unlock(&kvm_lock);
2343
2344         return r;
2345 }
2346
2347 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2348                            void *v)
2349 {
2350         int cpu = (long)v;
2351
2352         if (!kvm_usage_count)
2353                 return NOTIFY_OK;
2354
2355         val &= ~CPU_TASKS_FROZEN;
2356         switch (val) {
2357         case CPU_DYING:
2358                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2359                        cpu);
2360                 hardware_disable(NULL);
2361                 break;
2362         case CPU_STARTING:
2363                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2364                        cpu);
2365                 hardware_enable(NULL);
2366                 break;
2367         }
2368         return NOTIFY_OK;
2369 }
2370
2371
2372 asmlinkage void kvm_spurious_fault(void)
2373 {
2374         /* Fault while not rebooting.  We want the trace. */
2375         BUG();
2376 }
2377 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2378
2379 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2380                       void *v)
2381 {
2382         /*
2383          * Some (well, at least mine) BIOSes hang on reboot if
2384          * in vmx root mode.
2385          *
2386          * And Intel TXT required VMX off for all cpu when system shutdown.
2387          */
2388         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2389         kvm_rebooting = true;
2390         on_each_cpu(hardware_disable_nolock, NULL, 1);
2391         return NOTIFY_OK;
2392 }
2393
2394 static struct notifier_block kvm_reboot_notifier = {
2395         .notifier_call = kvm_reboot,
2396         .priority = 0,
2397 };
2398
2399 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2400 {
2401         int i;
2402
2403         for (i = 0; i < bus->dev_count; i++) {
2404                 struct kvm_io_device *pos = bus->range[i].dev;
2405
2406                 kvm_iodevice_destructor(pos);
2407         }
2408         kfree(bus);
2409 }
2410
2411 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2412 {
2413         const struct kvm_io_range *r1 = p1;
2414         const struct kvm_io_range *r2 = p2;
2415
2416         if (r1->addr < r2->addr)
2417                 return -1;
2418         if (r1->addr + r1->len > r2->addr + r2->len)
2419                 return 1;
2420         return 0;
2421 }
2422
2423 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2424                           gpa_t addr, int len)
2425 {
2426         if (bus->dev_count == NR_IOBUS_DEVS)
2427                 return -ENOSPC;
2428
2429         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2430                 .addr = addr,
2431                 .len = len,
2432                 .dev = dev,
2433         };
2434
2435         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2436                 kvm_io_bus_sort_cmp, NULL);
2437
2438         return 0;
2439 }
2440
2441 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2442                              gpa_t addr, int len)
2443 {
2444         struct kvm_io_range *range, key;
2445         int off;
2446
2447         key = (struct kvm_io_range) {
2448                 .addr = addr,
2449                 .len = len,
2450         };
2451
2452         range = bsearch(&key, bus->range, bus->dev_count,
2453                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2454         if (range == NULL)
2455                 return -ENOENT;
2456
2457         off = range - bus->range;
2458
2459         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2460                 off--;
2461
2462         return off;
2463 }
2464
2465 /* kvm_io_bus_write - called under kvm->slots_lock */
2466 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2467                      int len, const void *val)
2468 {
2469         int idx;
2470         struct kvm_io_bus *bus;
2471         struct kvm_io_range range;
2472
2473         range = (struct kvm_io_range) {
2474                 .addr = addr,
2475                 .len = len,
2476         };
2477
2478         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2479         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2480         if (idx < 0)
2481                 return -EOPNOTSUPP;
2482
2483         while (idx < bus->dev_count &&
2484                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2485                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2486                         return 0;
2487                 idx++;
2488         }
2489
2490         return -EOPNOTSUPP;
2491 }
2492
2493 /* kvm_io_bus_read - called under kvm->slots_lock */
2494 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2495                     int len, void *val)
2496 {
2497         int idx;
2498         struct kvm_io_bus *bus;
2499         struct kvm_io_range range;
2500
2501         range = (struct kvm_io_range) {
2502                 .addr = addr,
2503                 .len = len,
2504         };
2505
2506         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2507         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2508         if (idx < 0)
2509                 return -EOPNOTSUPP;
2510
2511         while (idx < bus->dev_count &&
2512                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2513                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2514                         return 0;
2515                 idx++;
2516         }
2517
2518         return -EOPNOTSUPP;
2519 }
2520
2521 /* Caller must hold slots_lock. */
2522 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2523                             int len, struct kvm_io_device *dev)
2524 {
2525         struct kvm_io_bus *new_bus, *bus;
2526
2527         bus = kvm->buses[bus_idx];
2528         if (bus->dev_count > NR_IOBUS_DEVS-1)
2529                 return -ENOSPC;
2530
2531         new_bus = kmemdup(bus, sizeof(struct kvm_io_bus), GFP_KERNEL);
2532         if (!new_bus)
2533                 return -ENOMEM;
2534         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2535         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2536         synchronize_srcu_expedited(&kvm->srcu);
2537         kfree(bus);
2538
2539         return 0;
2540 }
2541
2542 /* Caller must hold slots_lock. */
2543 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2544                               struct kvm_io_device *dev)
2545 {
2546         int i, r;
2547         struct kvm_io_bus *new_bus, *bus;
2548
2549         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2550         if (!new_bus)
2551                 return -ENOMEM;
2552
2553         bus = kvm->buses[bus_idx];
2554         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2555
2556         r = -ENOENT;
2557         for (i = 0; i < new_bus->dev_count; i++)
2558                 if (new_bus->range[i].dev == dev) {
2559                         r = 0;
2560                         new_bus->dev_count--;
2561                         new_bus->range[i] = new_bus->range[new_bus->dev_count];
2562                         sort(new_bus->range, new_bus->dev_count,
2563                              sizeof(struct kvm_io_range),
2564                              kvm_io_bus_sort_cmp, NULL);
2565                         break;
2566                 }
2567
2568         if (r) {
2569                 kfree(new_bus);
2570                 return r;
2571         }
2572
2573         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2574         synchronize_srcu_expedited(&kvm->srcu);
2575         kfree(bus);
2576         return r;
2577 }
2578
2579 static struct notifier_block kvm_cpu_notifier = {
2580         .notifier_call = kvm_cpu_hotplug,
2581 };
2582
2583 static int vm_stat_get(void *_offset, u64 *val)
2584 {
2585         unsigned offset = (long)_offset;
2586         struct kvm *kvm;
2587
2588         *val = 0;
2589         raw_spin_lock(&kvm_lock);
2590         list_for_each_entry(kvm, &vm_list, vm_list)
2591                 *val += *(u32 *)((void *)kvm + offset);
2592         raw_spin_unlock(&kvm_lock);
2593         return 0;
2594 }
2595
2596 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2597
2598 static int vcpu_stat_get(void *_offset, u64 *val)
2599 {
2600         unsigned offset = (long)_offset;
2601         struct kvm *kvm;
2602         struct kvm_vcpu *vcpu;
2603         int i;
2604
2605         *val = 0;
2606         raw_spin_lock(&kvm_lock);
2607         list_for_each_entry(kvm, &vm_list, vm_list)
2608                 kvm_for_each_vcpu(i, vcpu, kvm)
2609                         *val += *(u32 *)((void *)vcpu + offset);
2610
2611         raw_spin_unlock(&kvm_lock);
2612         return 0;
2613 }
2614
2615 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2616
2617 static const struct file_operations *stat_fops[] = {
2618         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2619         [KVM_STAT_VM]   = &vm_stat_fops,
2620 };
2621
2622 static void kvm_init_debug(void)
2623 {
2624         struct kvm_stats_debugfs_item *p;
2625
2626         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2627         for (p = debugfs_entries; p->name; ++p)
2628                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2629                                                 (void *)(long)p->offset,
2630                                                 stat_fops[p->kind]);
2631 }
2632
2633 static void kvm_exit_debug(void)
2634 {
2635         struct kvm_stats_debugfs_item *p;
2636
2637         for (p = debugfs_entries; p->name; ++p)
2638                 debugfs_remove(p->dentry);
2639         debugfs_remove(kvm_debugfs_dir);
2640 }
2641
2642 static int kvm_suspend(void)
2643 {
2644         if (kvm_usage_count)
2645                 hardware_disable_nolock(NULL);
2646         return 0;
2647 }
2648
2649 static void kvm_resume(void)
2650 {
2651         if (kvm_usage_count) {
2652                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2653                 hardware_enable_nolock(NULL);
2654         }
2655 }
2656
2657 static struct syscore_ops kvm_syscore_ops = {
2658         .suspend = kvm_suspend,
2659         .resume = kvm_resume,
2660 };
2661
2662 struct page *bad_page;
2663 pfn_t bad_pfn;
2664
2665 static inline
2666 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2667 {
2668         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2669 }
2670
2671 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2672 {
2673         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2674
2675         kvm_arch_vcpu_load(vcpu, cpu);
2676 }
2677
2678 static void kvm_sched_out(struct preempt_notifier *pn,
2679                           struct task_struct *next)
2680 {
2681         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2682
2683         kvm_arch_vcpu_put(vcpu);
2684 }
2685
2686 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2687                   struct module *module)
2688 {
2689         int r;
2690         int cpu;
2691
2692         r = kvm_arch_init(opaque);
2693         if (r)
2694                 goto out_fail;
2695
2696         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2697
2698         if (bad_page == NULL) {
2699                 r = -ENOMEM;
2700                 goto out;
2701         }
2702
2703         bad_pfn = page_to_pfn(bad_page);
2704
2705         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2706
2707         if (hwpoison_page == NULL) {
2708                 r = -ENOMEM;
2709                 goto out_free_0;
2710         }
2711
2712         hwpoison_pfn = page_to_pfn(hwpoison_page);
2713
2714         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2715
2716         if (fault_page == NULL) {
2717                 r = -ENOMEM;
2718                 goto out_free_0;
2719         }
2720
2721         fault_pfn = page_to_pfn(fault_page);
2722
2723         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2724                 r = -ENOMEM;
2725                 goto out_free_0;
2726         }
2727
2728         r = kvm_arch_hardware_setup();
2729         if (r < 0)
2730                 goto out_free_0a;
2731
2732         for_each_online_cpu(cpu) {
2733                 smp_call_function_single(cpu,
2734                                 kvm_arch_check_processor_compat,
2735                                 &r, 1);
2736                 if (r < 0)
2737                         goto out_free_1;
2738         }
2739
2740         r = register_cpu_notifier(&kvm_cpu_notifier);
2741         if (r)
2742                 goto out_free_2;
2743         register_reboot_notifier(&kvm_reboot_notifier);
2744
2745         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2746         if (!vcpu_align)
2747                 vcpu_align = __alignof__(struct kvm_vcpu);
2748         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2749                                            0, NULL);
2750         if (!kvm_vcpu_cache) {
2751                 r = -ENOMEM;
2752                 goto out_free_3;
2753         }
2754
2755         r = kvm_async_pf_init();
2756         if (r)
2757                 goto out_free;
2758
2759         kvm_chardev_ops.owner = module;
2760         kvm_vm_fops.owner = module;
2761         kvm_vcpu_fops.owner = module;
2762
2763         r = misc_register(&kvm_dev);
2764         if (r) {
2765                 printk(KERN_ERR "kvm: misc device register failed\n");
2766                 goto out_unreg;
2767         }
2768
2769         register_syscore_ops(&kvm_syscore_ops);
2770
2771         kvm_preempt_ops.sched_in = kvm_sched_in;
2772         kvm_preempt_ops.sched_out = kvm_sched_out;
2773
2774         kvm_init_debug();
2775
2776         return 0;
2777
2778 out_unreg:
2779         kvm_async_pf_deinit();
2780 out_free:
2781         kmem_cache_destroy(kvm_vcpu_cache);
2782 out_free_3:
2783         unregister_reboot_notifier(&kvm_reboot_notifier);
2784         unregister_cpu_notifier(&kvm_cpu_notifier);
2785 out_free_2:
2786 out_free_1:
2787         kvm_arch_hardware_unsetup();
2788 out_free_0a:
2789         free_cpumask_var(cpus_hardware_enabled);
2790 out_free_0:
2791         if (fault_page)
2792                 __free_page(fault_page);
2793         if (hwpoison_page)
2794                 __free_page(hwpoison_page);
2795         __free_page(bad_page);
2796 out:
2797         kvm_arch_exit();
2798 out_fail:
2799         return r;
2800 }
2801 EXPORT_SYMBOL_GPL(kvm_init);
2802
2803 void kvm_exit(void)
2804 {
2805         kvm_exit_debug();
2806         misc_deregister(&kvm_dev);
2807         kmem_cache_destroy(kvm_vcpu_cache);
2808         kvm_async_pf_deinit();
2809         unregister_syscore_ops(&kvm_syscore_ops);
2810         unregister_reboot_notifier(&kvm_reboot_notifier);
2811         unregister_cpu_notifier(&kvm_cpu_notifier);
2812         on_each_cpu(hardware_disable_nolock, NULL, 1);
2813         kvm_arch_hardware_unsetup();
2814         kvm_arch_exit();
2815         free_cpumask_var(cpus_hardware_enabled);
2816         __free_page(hwpoison_page);
2817         __free_page(bad_page);
2818 }
2819 EXPORT_SYMBOL_GPL(kvm_exit);