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
 293         kvm->mmu_notifier_seq++;
 294         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 295         /* we've to flush the tlb before the pages can be freed */
 296         if (need_tlb_flush)
 297                 kvm_flush_remote_tlbs(kvm);
 298
 299         spin_unlock(&kvm->mmu_lock);
 300         srcu_read_unlock(&kvm->srcu, idx);
 301 }
 302
 303 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 304                                         struct mm_struct *mm,
 305                                         unsigned long address,
 306                                         pte_t pte)
 307 {
 308         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 309         int idx;
 310
 311         idx = srcu_read_lock(&kvm->srcu);
 312         spin_lock(&kvm->mmu_lock);
 313         kvm->mmu_notifier_seq++;
 314         kvm_set_spte_hva(kvm, address, pte);
 315         spin_unlock(&kvm->mmu_lock);
 316         srcu_read_unlock(&kvm->srcu, idx);
 317 }
 318
 319 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 320                                                     struct mm_struct *mm,
 321                                                     unsigned long start,
 322                                                     unsigned long end)
 323 {
 324         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 325         int need_tlb_flush = 0, idx;
 326
 327         idx = srcu_read_lock(&kvm->srcu);
 328         spin_lock(&kvm->mmu_lock);
 329         /*
 330          * The count increase must become visible at unlock time as no
 331          * spte can be established without taking the mmu_lock and
 332          * count is also read inside the mmu_lock critical section.
 333          */
 334         kvm->mmu_notifier_count++;
 335         for (; start < end; start += PAGE_SIZE)
 336                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 337         need_tlb_flush |= kvm->tlbs_dirty;
 338         /* we've to flush the tlb before the pages can be freed */
 339         if (need_tlb_flush)
 340                 kvm_flush_remote_tlbs(kvm);
 341
 342         spin_unlock(&kvm->mmu_lock);
 343         srcu_read_unlock(&kvm->srcu, idx);
 344 }
 345
 346 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 347                                                   struct mm_struct *mm,
 348                                                   unsigned long start,
 349                                                   unsigned long end)
 350 {
 351         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 352
 353         spin_lock(&kvm->mmu_lock);
 354         /*
 355          * This sequence increase will notify the kvm page fault that
 356          * the page that is going to be mapped in the spte could have
 357          * been freed.
 358          */
 359         kvm->mmu_notifier_seq++;
 360         /*
 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
 382         young = kvm_age_hva(kvm, address);
 383         if (young)
 384                 kvm_flush_remote_tlbs(kvm);
 385
 386         spin_unlock(&kvm->mmu_lock);
 387         srcu_read_unlock(&kvm->srcu, idx);
 388
 389         return young;
 390 }
 391
 392 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 393                                        struct mm_struct *mm,
 394                                        unsigned long address)
 395 {
 396         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 397         int young, idx;
 398
 399         idx = srcu_read_lock(&kvm->srcu);
 400         spin_lock(&kvm->mmu_lock);
 401         young = kvm_test_age_hva(kvm, address);
 402         spin_unlock(&kvm->mmu_lock);
 403         srcu_read_unlock(&kvm->srcu, idx);
 404
 405         return young;
 406 }
 407
 408 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 409                                      struct mm_struct *mm)
 410 {
 411         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 412         int idx;
 413
 414         idx = srcu_read_lock(&kvm->srcu);
 415         kvm_arch_flush_shadow(kvm);
 416         srcu_read_unlock(&kvm->srcu, idx);
 417 }
 418
 419 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 420         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 421         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 422         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 423         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 424         .test_young             = kvm_mmu_notifier_test_young,
 425         .change_pte             = kvm_mmu_notifier_change_pte,
 426         .release                = kvm_mmu_notifier_release,
 427 };
 428
 429 static int kvm_init_mmu_notifier(struct kvm *kvm)
 430 {
 431         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 432         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 433 }
 434
 435 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 436
 437 static int kvm_init_mmu_notifier(struct kvm *kvm)
 438 {
 439         return 0;
 440 }
 441
 442 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 443
 444 static struct kvm *kvm_create_vm(void)
 445 {
 446         int r, i;
 447         struct kvm *kvm = kvm_arch_alloc_vm();
 448
 449         if (!kvm)
 450                 return ERR_PTR(-ENOMEM);
 451
 452         r = kvm_arch_init_vm(kvm);
 453         if (r)
 454                 goto out_err_nodisable;
 455
 456         r = hardware_enable_all();
 457         if (r)
 458                 goto out_err_nodisable;
 459
 460 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 461         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 462         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 463 #endif
 464
 465         r = -ENOMEM;
 466         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 467         if (!kvm->memslots)
 468                 goto out_err_nosrcu;
 469         if (init_srcu_struct(&kvm->srcu))
 470                 goto out_err_nosrcu;
 471         for (i = 0; i < KVM_NR_BUSES; i++) {
 472                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 473                                         GFP_KERNEL);
 474                 if (!kvm->buses[i])
 475                         goto out_err;
 476         }
 477
 478         spin_lock_init(&kvm->mmu_lock);
 479         kvm->mm = current->mm;
 480         atomic_inc(&kvm->mm->mm_count);
 481         kvm_eventfd_init(kvm);
 482         mutex_init(&kvm->lock);
 483         mutex_init(&kvm->irq_lock);
 484         mutex_init(&kvm->slots_lock);
 485         atomic_set(&kvm->users_count, 1);
 486
 487         r = kvm_init_mmu_notifier(kvm);
 488         if (r)
 489                 goto out_err;
 490
 491         raw_spin_lock(&kvm_lock);
 492         list_add(&kvm->vm_list, &vm_list);
 493         raw_spin_unlock(&kvm_lock);
 494
 495         return kvm;
 496
 497 out_err:
 498         cleanup_srcu_struct(&kvm->srcu);
 499 out_err_nosrcu:
 500         hardware_disable_all();
 501 out_err_nodisable:
 502         for (i = 0; i < KVM_NR_BUSES; i++)
 503                 kfree(kvm->buses[i]);
 504         kfree(kvm->memslots);
 505         kvm_arch_free_vm(kvm);
 506         return ERR_PTR(r);
 507 }
 508
 509 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 510 {
 511         if (!memslot->dirty_bitmap)
 512                 return;
 513
 514         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
 515                 vfree(memslot->dirty_bitmap_head);
 516         else
 517                 kfree(memslot->dirty_bitmap_head);
 518
 519         memslot->dirty_bitmap = NULL;
 520         memslot->dirty_bitmap_head = NULL;
 521 }
 522
 523 /*
 524  * Free any memory in @free but not in @dont.
 525  */
 526 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 527                                   struct kvm_memory_slot *dont)
 528 {
 529         int i;
 530
 531         if (!dont || free->rmap != dont->rmap)
 532                 vfree(free->rmap);
 533
 534         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 535                 kvm_destroy_dirty_bitmap(free);
 536
 537
 538         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 539                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 540                         vfree(free->lpage_info[i]);
 541                         free->lpage_info[i] = NULL;
 542                 }
 543         }
 544
 545         free->npages = 0;
 546         free->rmap = NULL;
 547 }
 548
 549 void kvm_free_physmem(struct kvm *kvm)
 550 {
 551         int i;
 552         struct kvm_memslots *slots = kvm->memslots;
 553
 554         for (i = 0; i < slots->nmemslots; ++i)
 555                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 556
 557         kfree(kvm->memslots);
 558 }
 559
 560 static void kvm_destroy_vm(struct kvm *kvm)
 561 {
 562         int i;
 563         struct mm_struct *mm = kvm->mm;
 564
 565         kvm_arch_sync_events(kvm);
 566         raw_spin_lock(&kvm_lock);
 567         list_del(&kvm->vm_list);
 568         raw_spin_unlock(&kvm_lock);
 569         kvm_free_irq_routing(kvm);
 570         for (i = 0; i < KVM_NR_BUSES; i++)
 571                 kvm_io_bus_destroy(kvm->buses[i]);
 572         kvm_coalesced_mmio_free(kvm);
 573 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 574         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 575 #else
 576         kvm_arch_flush_shadow(kvm);
 577 #endif
 578         kvm_arch_destroy_vm(kvm);
 579         kvm_free_physmem(kvm);
 580         cleanup_srcu_struct(&kvm->srcu);
 581         kvm_arch_free_vm(kvm);
 582         hardware_disable_all();
 583         mmdrop(mm);
 584 }
 585
 586 void kvm_get_kvm(struct kvm *kvm)
 587 {
 588         atomic_inc(&kvm->users_count);
 589 }
 590 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 591
 592 void kvm_put_kvm(struct kvm *kvm)
 593 {
 594         if (atomic_dec_and_test(&kvm->users_count))
 595                 kvm_destroy_vm(kvm);
 596 }
 597 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 598
 599
 600 static int kvm_vm_release(struct inode *inode, struct file *filp)
 601 {
 602         struct kvm *kvm = filp->private_data;
 603
 604         kvm_irqfd_release(kvm);
 605
 606         kvm_put_kvm(kvm);
 607         return 0;
 608 }
 609
 610 #ifndef CONFIG_S390
 611 /*
 612  * Allocation size is twice as large as the actual dirty bitmap size.
 613  * This makes it possible to do double buffering: see x86's
 614  * kvm_vm_ioctl_get_dirty_log().
 615  */
 616 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 617 {
 618         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 619
 620         if (dirty_bytes > PAGE_SIZE)
 621                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
 622         else
 623                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
 624
 625         if (!memslot->dirty_bitmap)
 626                 return -ENOMEM;
 627
 628         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
 629         return 0;
 630 }
 631 #endif /* !CONFIG_S390 */
 632
 633 /*
 634  * Allocate some memory and give it an address in the guest physical address
 635  * space.
 636  *
 637  * Discontiguous memory is allowed, mostly for framebuffers.
 638  *
 639  * Must be called holding mmap_sem for write.
 640  */
 641 int __kvm_set_memory_region(struct kvm *kvm,
 642                             struct kvm_userspace_memory_region *mem,
 643                             int user_alloc)
 644 {
 645         int r;
 646         gfn_t base_gfn;
 647         unsigned long npages;
 648         unsigned long i;
 649         struct kvm_memory_slot *memslot;
 650         struct kvm_memory_slot old, new;
 651         struct kvm_memslots *slots, *old_memslots;
 652
 653         r = -EINVAL;
 654         /* General sanity checks */
 655         if (mem->memory_size & (PAGE_SIZE - 1))
 656                 goto out;
 657         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 658                 goto out;
 659         /* We can read the guest memory with __xxx_user() later on. */
 660         if (user_alloc &&
 661             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 662              !access_ok(VERIFY_WRITE,
 663                         (void __user *)(unsigned long)mem->userspace_addr,
 664                         mem->memory_size)))
 665                 goto out;
 666         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 667                 goto out;
 668         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 669                 goto out;
 670
 671         memslot = &kvm->memslots->memslots[mem->slot];
 672         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 673         npages = mem->memory_size >> PAGE_SHIFT;
 674
 675         r = -EINVAL;
 676         if (npages > KVM_MEM_MAX_NR_PAGES)
 677                 goto out;
 678
 679         if (!npages)
 680                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 681
 682         new = old = *memslot;
 683
 684         new.id = mem->slot;
 685         new.base_gfn = base_gfn;
 686         new.npages = npages;
 687         new.flags = mem->flags;
 688
 689         /* Disallow changing a memory slot's size. */
 690         r = -EINVAL;
 691         if (npages && old.npages && npages != old.npages)
 692                 goto out_free;
 693
 694         /* Check for overlaps */
 695         r = -EEXIST;
 696         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 697                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 698
 699                 if (s == memslot || !s->npages)
 700                         continue;
 701                 if (!((base_gfn + npages <= s->base_gfn) ||
 702                       (base_gfn >= s->base_gfn + s->npages)))
 703                         goto out_free;
 704         }
 705
 706         /* Free page dirty bitmap if unneeded */
 707         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 708                 new.dirty_bitmap = NULL;
 709
 710         r = -ENOMEM;
 711
 712         /* Allocate if a slot is being created */
 713 #ifndef CONFIG_S390
 714         if (npages && !new.rmap) {
 715                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
 716
 717                 if (!new.rmap)
 718                         goto out_free;
 719
 720                 new.user_alloc = user_alloc;
 721                 new.userspace_addr = mem->userspace_addr;
 722         }
 723         if (!npages)
 724                 goto skip_lpage;
 725
 726         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 727                 unsigned long ugfn;
 728                 unsigned long j;
 729                 int lpages;
 730                 int level = i + 2;
 731
 732                 /* Avoid unused variable warning if no large pages */
 733                 (void)level;
 734
 735                 if (new.lpage_info[i])
 736                         continue;
 737
 738                 lpages = 1 + ((base_gfn + npages - 1)
 739                              >> KVM_HPAGE_GFN_SHIFT(level));
 740                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
 741
 742                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
 743
 744                 if (!new.lpage_info[i])
 745                         goto out_free;
 746
 747                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
 748                         new.lpage_info[i][0].write_count = 1;
 749                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
 750                         new.lpage_info[i][lpages - 1].write_count = 1;
 751                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 752                 /*
 753                  * If the gfn and userspace address are not aligned wrt each
 754                  * other, or if explicitly asked to, disable large page
 755                  * support for this slot
 756                  */
 757                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 758                     !largepages_enabled)
 759                         for (j = 0; j < lpages; ++j)
 760                                 new.lpage_info[i][j].write_count = 1;
 761         }
 762
 763 skip_lpage:
 764
 765         /* Allocate page dirty bitmap if needed */
 766         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 767                 if (kvm_create_dirty_bitmap(&new) < 0)
 768                         goto out_free;
 769                 /* destroy any largepage mappings for dirty tracking */
 770         }
 771 #else  /* not defined CONFIG_S390 */
 772         new.user_alloc = user_alloc;
 773         if (user_alloc)
 774                 new.userspace_addr = mem->userspace_addr;
 775 #endif /* not defined CONFIG_S390 */
 776
 777         if (!npages) {
 778                 r = -ENOMEM;
 779                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 780                 if (!slots)
 781                         goto out_free;
 782                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 783                 if (mem->slot >= slots->nmemslots)
 784                         slots->nmemslots = mem->slot + 1;
 785                 slots->generation++;
 786                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 787
 788                 old_memslots = kvm->memslots;
 789                 rcu_assign_pointer(kvm->memslots, slots);
 790                 synchronize_srcu_expedited(&kvm->srcu);
 791                 /* From this point no new shadow pages pointing to a deleted
 792                  * memslot will be created.
 793                  *
 794                  * validation of sp->gfn happens in:
 795                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 796                  *      - kvm_is_visible_gfn (mmu_check_roots)
 797                  */
 798                 kvm_arch_flush_shadow(kvm);
 799                 kfree(old_memslots);
 800         }
 801
 802         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 803         if (r)
 804                 goto out_free;
 805
 806         /* map/unmap the pages in iommu page table */
 807         if (npages) {
 808                 r = kvm_iommu_map_pages(kvm, &new);
 809                 if (r)
 810                         goto out_free;
 811         } else
 812                 kvm_iommu_unmap_pages(kvm, &old);
 813
 814         r = -ENOMEM;
 815         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 816         if (!slots)
 817                 goto out_free;
 818         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 819         if (mem->slot >= slots->nmemslots)
 820                 slots->nmemslots = mem->slot + 1;
 821         slots->generation++;
 822
 823         /* actual memory is freed via old in kvm_free_physmem_slot below */
 824         if (!npages) {
 825                 new.rmap = NULL;
 826                 new.dirty_bitmap = NULL;
 827                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 828                         new.lpage_info[i] = NULL;
 829         }
 830
 831         slots->memslots[mem->slot] = new;
 832         old_memslots = kvm->memslots;
 833         rcu_assign_pointer(kvm->memslots, slots);
 834         synchronize_srcu_expedited(&kvm->srcu);
 835
 836         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 837
 838         /*
 839          * If the new memory slot is created, we need to clear all
 840          * mmio sptes.
 841          */
 842         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 843                 kvm_arch_flush_shadow(kvm);
 844
 845         kvm_free_physmem_slot(&old, &new);
 846         kfree(old_memslots);
 847
 848         return 0;
 849
 850 out_free:
 851         kvm_free_physmem_slot(&new, &old);
 852 out:
 853         return r;
 854
 855 }
 856 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 857
 858 int kvm_set_memory_region(struct kvm *kvm,
 859                           struct kvm_userspace_memory_region *mem,
 860                           int user_alloc)
 861 {
 862         int r;
 863
 864         mutex_lock(&kvm->slots_lock);
 865         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 866         mutex_unlock(&kvm->slots_lock);
 867         return r;
 868 }
 869 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 870
 871 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 872                                    struct
 873                                    kvm_userspace_memory_region *mem,
 874                                    int user_alloc)
 875 {
 876         if (mem->slot >= KVM_MEMORY_SLOTS)
 877                 return -EINVAL;
 878         return kvm_set_memory_region(kvm, mem, user_alloc);
 879 }
 880
 881 int kvm_get_dirty_log(struct kvm *kvm,
 882                         struct kvm_dirty_log *log, int *is_dirty)
 883 {
 884         struct kvm_memory_slot *memslot;
 885         int r, i;
 886         unsigned long n;
 887         unsigned long any = 0;
 888
 889         r = -EINVAL;
 890         if (log->slot >= KVM_MEMORY_SLOTS)
 891                 goto out;
 892
 893         memslot = &kvm->memslots->memslots[log->slot];
 894         r = -ENOENT;
 895         if (!memslot->dirty_bitmap)
 896                 goto out;
 897
 898         n = kvm_dirty_bitmap_bytes(memslot);
 899
 900         for (i = 0; !any && i < n/sizeof(long); ++i)
 901                 any = memslot->dirty_bitmap[i];
 902
 903         r = -EFAULT;
 904         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 905                 goto out;
 906
 907         if (any)
 908                 *is_dirty = 1;
 909
 910         r = 0;
 911 out:
 912         return r;
 913 }
 914
 915 void kvm_disable_largepages(void)
 916 {
 917         largepages_enabled = false;
 918 }
 919 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 920
 921 int is_error_page(struct page *page)
 922 {
 923         return page == bad_page || page == hwpoison_page || page == fault_page;
 924 }
 925 EXPORT_SYMBOL_GPL(is_error_page);
 926
 927 int is_error_pfn(pfn_t pfn)
 928 {
 929         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 930 }
 931 EXPORT_SYMBOL_GPL(is_error_pfn);
 932
 933 int is_hwpoison_pfn(pfn_t pfn)
 934 {
 935         return pfn == hwpoison_pfn;
 936 }
 937 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 938
 939 int is_fault_pfn(pfn_t pfn)
 940 {
 941         return pfn == fault_pfn;
 942 }
 943 EXPORT_SYMBOL_GPL(is_fault_pfn);
 944
 945 int is_noslot_pfn(pfn_t pfn)
 946 {
 947         return pfn == bad_pfn;
 948 }
 949 EXPORT_SYMBOL_GPL(is_noslot_pfn);
 950
 951 int is_invalid_pfn(pfn_t pfn)
 952 {
 953         return pfn == hwpoison_pfn || pfn == fault_pfn;
 954 }
 955 EXPORT_SYMBOL_GPL(is_invalid_pfn);
 956
 957 static inline unsigned long bad_hva(void)
 958 {
 959         return PAGE_OFFSET;
 960 }
 961
 962 int kvm_is_error_hva(unsigned long addr)
 963 {
 964         return addr == bad_hva();
 965 }
 966 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 967
 968 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
 969                                                 gfn_t gfn)
 970 {
 971         int i;
 972
 973         for (i = 0; i < slots->nmemslots; ++i) {
 974                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 975
 976                 if (gfn >= memslot->base_gfn
 977                     && gfn < memslot->base_gfn + memslot->npages)
 978                         return memslot;
 979         }
 980         return NULL;
 981 }
 982
 983 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 984 {
 985         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 986 }
 987 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 988
 989 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 990 {
 991         int i;
 992         struct kvm_memslots *slots = kvm_memslots(kvm);
 993
 994         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 995                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 996
 997                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 998                         continue;
 999
1000                 if (gfn >= memslot->base_gfn
1001                     && gfn < memslot->base_gfn + memslot->npages)
1002                         return 1;
1003         }
1004         return 0;
1005 }
1006 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1007
1008 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1009 {
1010         struct vm_area_struct *vma;
1011         unsigned long addr, size;
1012
1013         size = PAGE_SIZE;
1014
1015         addr = gfn_to_hva(kvm, gfn);
1016         if (kvm_is_error_hva(addr))
1017                 return PAGE_SIZE;
1018
1019         down_read(&current->mm->mmap_sem);
1020         vma = find_vma(current->mm, addr);
1021         if (!vma)
1022                 goto out;
1023
1024         size = vma_kernel_pagesize(vma);
1025
1026 out:
1027         up_read(&current->mm->mmap_sem);
1028
1029         return size;
1030 }
1031
1032 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1033                                      gfn_t *nr_pages)
1034 {
1035         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1036                 return bad_hva();
1037
1038         if (nr_pages)
1039                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1040
1041         return gfn_to_hva_memslot(slot, gfn);
1042 }
1043
1044 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1045 {
1046         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1047 }
1048 EXPORT_SYMBOL_GPL(gfn_to_hva);
1049
1050 static pfn_t get_fault_pfn(void)
1051 {
1052         get_page(fault_page);
1053         return fault_pfn;
1054 }
1055
1056 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1057         unsigned long start, int write, struct page **page)
1058 {
1059         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1060
1061         if (write)
1062                 flags |= FOLL_WRITE;
1063
1064         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1065 }
1066
1067 static inline int check_user_page_hwpoison(unsigned long addr)
1068 {
1069         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1070
1071         rc = __get_user_pages(current, current->mm, addr, 1,
1072                               flags, NULL, NULL, NULL);
1073         return rc == -EHWPOISON;
1074 }
1075
1076 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1077                         bool *async, bool write_fault, bool *writable)
1078 {
1079         struct page *page[1];
1080         int npages = 0;
1081         pfn_t pfn;
1082
1083         /* we can do it either atomically or asynchronously, not both */
1084         BUG_ON(atomic && async);
1085
1086         BUG_ON(!write_fault && !writable);
1087
1088         if (writable)
1089                 *writable = true;
1090
1091         if (atomic || async)
1092                 npages = __get_user_pages_fast(addr, 1, 1, page);
1093
1094         if (unlikely(npages != 1) && !atomic) {
1095                 might_sleep();
1096
1097                 if (writable)
1098                         *writable = write_fault;
1099
1100                 if (async) {
1101                         down_read(&current->mm->mmap_sem);
1102                         npages = get_user_page_nowait(current, current->mm,
1103                                                      addr, write_fault, page);
1104                         up_read(&current->mm->mmap_sem);
1105                 } else
1106                         npages = get_user_pages_fast(addr, 1, write_fault,
1107                                                      page);
1108
1109                 /* map read fault as writable if possible */
1110                 if (unlikely(!write_fault) && npages == 1) {
1111                         struct page *wpage[1];
1112
1113                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1114                         if (npages == 1) {
1115                                 *writable = true;
1116                                 put_page(page[0]);
1117                                 page[0] = wpage[0];
1118                         }
1119                         npages = 1;
1120                 }
1121         }
1122
1123         if (unlikely(npages != 1)) {
1124                 struct vm_area_struct *vma;
1125
1126                 if (atomic)
1127                         return get_fault_pfn();
1128
1129                 down_read(&current->mm->mmap_sem);
1130                 if (npages == -EHWPOISON ||
1131                         (!async && check_user_page_hwpoison(addr))) {
1132                         up_read(&current->mm->mmap_sem);
1133                         get_page(hwpoison_page);
1134                         return page_to_pfn(hwpoison_page);
1135                 }
1136
1137                 vma = find_vma_intersection(current->mm, addr, addr+1);
1138
1139                 if (vma == NULL)
1140                         pfn = get_fault_pfn();
1141                 else if ((vma->vm_flags & VM_PFNMAP)) {
1142                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1143                                 vma->vm_pgoff;
1144                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1145                 } else {
1146                         if (async && (vma->vm_flags & VM_WRITE))
1147                                 *async = true;
1148                         pfn = get_fault_pfn();
1149                 }
1150                 up_read(&current->mm->mmap_sem);
1151         } else
1152                 pfn = page_to_pfn(page[0]);
1153
1154         return pfn;
1155 }
1156
1157 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1158 {
1159         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1160 }
1161 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1162
1163 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1164                           bool write_fault, bool *writable)
1165 {
1166         unsigned long addr;
1167
1168         if (async)
1169                 *async = false;
1170
1171         addr = gfn_to_hva(kvm, gfn);
1172         if (kvm_is_error_hva(addr)) {
1173                 get_page(bad_page);
1174                 return page_to_pfn(bad_page);
1175         }
1176
1177         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1178 }
1179
1180 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1181 {
1182         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1183 }
1184 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1185
1186 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1187                        bool write_fault, bool *writable)
1188 {
1189         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1190 }
1191 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1192
1193 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1194 {
1195         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1196 }
1197 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1198
1199 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1200                       bool *writable)
1201 {
1202         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1203 }
1204 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1205
1206 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1207                          struct kvm_memory_slot *slot, gfn_t gfn)
1208 {
1209         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1210         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1211 }
1212
1213 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1214                                                                   int nr_pages)
1215 {
1216         unsigned long addr;
1217         gfn_t entry;
1218
1219         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1220         if (kvm_is_error_hva(addr))
1221                 return -1;
1222
1223         if (entry < nr_pages)
1224                 return 0;
1225
1226         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1227 }
1228 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1229
1230 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1231 {
1232         pfn_t pfn;
1233
1234         pfn = gfn_to_pfn(kvm, gfn);
1235         if (!kvm_is_mmio_pfn(pfn))
1236                 return pfn_to_page(pfn);
1237
1238         WARN_ON(kvm_is_mmio_pfn(pfn));
1239
1240         get_page(bad_page);
1241         return bad_page;
1242 }
1243
1244 EXPORT_SYMBOL_GPL(gfn_to_page);
1245
1246 void kvm_release_page_clean(struct page *page)
1247 {
1248         kvm_release_pfn_clean(page_to_pfn(page));
1249 }
1250 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1251
1252 void kvm_release_pfn_clean(pfn_t pfn)
1253 {
1254         if (!kvm_is_mmio_pfn(pfn))
1255                 put_page(pfn_to_page(pfn));
1256 }
1257 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1258
1259 void kvm_release_page_dirty(struct page *page)
1260 {
1261         kvm_release_pfn_dirty(page_to_pfn(page));
1262 }
1263 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1264
1265 void kvm_release_pfn_dirty(pfn_t pfn)
1266 {
1267         kvm_set_pfn_dirty(pfn);
1268         kvm_release_pfn_clean(pfn);
1269 }
1270 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1271
1272 void kvm_set_page_dirty(struct page *page)
1273 {
1274         kvm_set_pfn_dirty(page_to_pfn(page));
1275 }
1276 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1277
1278 void kvm_set_pfn_dirty(pfn_t pfn)
1279 {
1280         if (!kvm_is_mmio_pfn(pfn)) {
1281                 struct page *page = pfn_to_page(pfn);
1282                 if (!PageReserved(page))
1283                         SetPageDirty(page);
1284         }
1285 }
1286 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1287
1288 void kvm_set_pfn_accessed(pfn_t pfn)
1289 {
1290         if (!kvm_is_mmio_pfn(pfn))
1291                 mark_page_accessed(pfn_to_page(pfn));
1292 }
1293 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1294
1295 void kvm_get_pfn(pfn_t pfn)
1296 {
1297         if (!kvm_is_mmio_pfn(pfn))
1298                 get_page(pfn_to_page(pfn));
1299 }
1300 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1301
1302 static int next_segment(unsigned long len, int offset)
1303 {
1304         if (len > PAGE_SIZE - offset)
1305                 return PAGE_SIZE - offset;
1306         else
1307                 return len;
1308 }
1309
1310 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1311                         int len)
1312 {
1313         int r;
1314         unsigned long addr;
1315
1316         addr = gfn_to_hva(kvm, gfn);
1317         if (kvm_is_error_hva(addr))
1318                 return -EFAULT;
1319         r = __copy_from_user(data, (void __user *)addr + offset, len);
1320         if (r)
1321                 return -EFAULT;
1322         return 0;
1323 }
1324 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1325
1326 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1327 {
1328         gfn_t gfn = gpa >> PAGE_SHIFT;
1329         int seg;
1330         int offset = offset_in_page(gpa);
1331         int ret;
1332
1333         while ((seg = next_segment(len, offset)) != 0) {
1334                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1335                 if (ret < 0)
1336                         return ret;
1337                 offset = 0;
1338                 len -= seg;
1339                 data += seg;
1340                 ++gfn;
1341         }
1342         return 0;
1343 }
1344 EXPORT_SYMBOL_GPL(kvm_read_guest);
1345
1346 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1347                           unsigned long len)
1348 {
1349         int r;
1350         unsigned long addr;
1351         gfn_t gfn = gpa >> PAGE_SHIFT;
1352         int offset = offset_in_page(gpa);
1353
1354         addr = gfn_to_hva(kvm, gfn);
1355         if (kvm_is_error_hva(addr))
1356                 return -EFAULT;
1357         pagefault_disable();
1358         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1359         pagefault_enable();
1360         if (r)
1361                 return -EFAULT;
1362         return 0;
1363 }
1364 EXPORT_SYMBOL(kvm_read_guest_atomic);
1365
1366 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1367                          int offset, int len)
1368 {
1369         int r;
1370         unsigned long addr;
1371
1372         addr = gfn_to_hva(kvm, gfn);
1373         if (kvm_is_error_hva(addr))
1374                 return -EFAULT;
1375         r = __copy_to_user((void __user *)addr + offset, data, len);
1376         if (r)
1377                 return -EFAULT;
1378         mark_page_dirty(kvm, gfn);
1379         return 0;
1380 }
1381 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1382
1383 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1384                     unsigned long len)
1385 {
1386         gfn_t gfn = gpa >> PAGE_SHIFT;
1387         int seg;
1388         int offset = offset_in_page(gpa);
1389         int ret;
1390
1391         while ((seg = next_segment(len, offset)) != 0) {
1392                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1393                 if (ret < 0)
1394                         return ret;
1395                 offset = 0;
1396                 len -= seg;
1397                 data += seg;
1398                 ++gfn;
1399         }
1400         return 0;
1401 }
1402
1403 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1404                               gpa_t gpa)
1405 {
1406         struct kvm_memslots *slots = kvm_memslots(kvm);
1407         int offset = offset_in_page(gpa);
1408         gfn_t gfn = gpa >> PAGE_SHIFT;
1409
1410         ghc->gpa = gpa;
1411         ghc->generation = slots->generation;
1412         ghc->memslot = __gfn_to_memslot(slots, gfn);
1413         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1414         if (!kvm_is_error_hva(ghc->hva))
1415                 ghc->hva += offset;
1416         else
1417                 return -EFAULT;
1418
1419         return 0;
1420 }
1421 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1422
1423 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1424                            void *data, unsigned long len)
1425 {
1426         struct kvm_memslots *slots = kvm_memslots(kvm);
1427         int r;
1428
1429         if (slots->generation != ghc->generation)
1430                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1431
1432         if (kvm_is_error_hva(ghc->hva))
1433                 return -EFAULT;
1434
1435         r = __copy_to_user((void __user *)ghc->hva, data, len);
1436         if (r)
1437                 return -EFAULT;
1438         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1439
1440         return 0;
1441 }
1442 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1443
1444 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1445                            void *data, unsigned long len)
1446 {
1447         struct kvm_memslots *slots = kvm_memslots(kvm);
1448         int r;
1449
1450         if (slots->generation != ghc->generation)
1451                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1452
1453         if (kvm_is_error_hva(ghc->hva))
1454                 return -EFAULT;
1455
1456         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1457         if (r)
1458                 return -EFAULT;
1459
1460         return 0;
1461 }
1462 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1463
1464 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1465 {
1466         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1467                                     offset, len);
1468 }
1469 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1470
1471 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1472 {
1473         gfn_t gfn = gpa >> PAGE_SHIFT;
1474         int seg;
1475         int offset = offset_in_page(gpa);
1476         int ret;
1477
1478         while ((seg = next_segment(len, offset)) != 0) {
1479                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1480                 if (ret < 0)
1481                         return ret;
1482                 offset = 0;
1483                 len -= seg;
1484                 ++gfn;
1485         }
1486         return 0;
1487 }
1488 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1489
1490 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1491                              gfn_t gfn)
1492 {
1493         if (memslot && memslot->dirty_bitmap) {
1494                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1495
1496                 __set_bit_le(rel_gfn, memslot->dirty_bitmap);
1497         }
1498 }
1499
1500 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1501 {
1502         struct kvm_memory_slot *memslot;
1503
1504         memslot = gfn_to_memslot(kvm, gfn);
1505         mark_page_dirty_in_slot(kvm, memslot, gfn);
1506 }
1507
1508 /*
1509  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1510  */
1511 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1512 {
1513         DEFINE_WAIT(wait);
1514
1515         for (;;) {
1516                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1517
1518                 if (kvm_arch_vcpu_runnable(vcpu)) {
1519                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1520                         break;
1521                 }
1522                 if (kvm_cpu_has_pending_timer(vcpu))
1523                         break;
1524                 if (signal_pending(current))
1525                         break;
1526
1527                 schedule();
1528         }
1529
1530         finish_wait(&vcpu->wq, &wait);
1531 }
1532
1533 void kvm_resched(struct kvm_vcpu *vcpu)
1534 {
1535         if (!need_resched())
1536                 return;
1537         cond_resched();
1538 }
1539 EXPORT_SYMBOL_GPL(kvm_resched);
1540
1541 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1542 {
1543         struct kvm *kvm = me->kvm;
1544         struct kvm_vcpu *vcpu;
1545         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1546         int yielded = 0;
1547         int pass;
1548         int i;
1549
1550         /*
1551          * We boost the priority of a VCPU that is runnable but not
1552          * currently running, because it got preempted by something
1553          * else and called schedule in __vcpu_run.  Hopefully that
1554          * VCPU is holding the lock that we need and will release it.
1555          * We approximate round-robin by starting at the last boosted VCPU.
1556          */
1557         for (pass = 0; pass < 2 && !yielded; pass++) {
1558                 kvm_for_each_vcpu(i, vcpu, kvm) {
1559                         struct task_struct *task = NULL;
1560                         struct pid *pid;
1561                         if (!pass && i < last_boosted_vcpu) {
1562                                 i = last_boosted_vcpu;
1563                                 continue;
1564                         } else if (pass && i > last_boosted_vcpu)
1565                                 break;
1566                         if (vcpu == me)
1567                                 continue;
1568                         if (waitqueue_active(&vcpu->wq))
1569                                 continue;
1570                         rcu_read_lock();
1571                         pid = rcu_dereference(vcpu->pid);
1572                         if (pid)
1573                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1574                         rcu_read_unlock();
1575                         if (!task)
1576                                 continue;
1577                         if (task->flags & PF_VCPU) {
1578                                 put_task_struct(task);
1579                                 continue;
1580                         }
1581                         if (yield_to(task, 1)) {
1582                                 put_task_struct(task);
1583                                 kvm->last_boosted_vcpu = i;
1584                                 yielded = 1;
1585                                 break;
1586                         }
1587                         put_task_struct(task);
1588                 }
1589         }
1590 }
1591 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1592
1593 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1594 {
1595         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1596         struct page *page;
1597
1598         if (vmf->pgoff == 0)
1599                 page = virt_to_page(vcpu->run);
1600 #ifdef CONFIG_X86
1601         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1602                 page = virt_to_page(vcpu->arch.pio_data);
1603 #endif
1604 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1605         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1606                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1607 #endif
1608         else
1609                 return VM_FAULT_SIGBUS;
1610         get_page(page);
1611         vmf->page = page;
1612         return 0;
1613 }
1614
1615 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1616         .fault = kvm_vcpu_fault,
1617 };
1618
1619 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1620 {
1621         vma->vm_ops = &kvm_vcpu_vm_ops;
1622         return 0;
1623 }
1624
1625 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1626 {
1627         struct kvm_vcpu *vcpu = filp->private_data;
1628
1629         kvm_put_kvm(vcpu->kvm);
1630         return 0;
1631 }
1632
1633 static struct file_operations kvm_vcpu_fops = {
1634         .release        = kvm_vcpu_release,
1635         .unlocked_ioctl = kvm_vcpu_ioctl,
1636 #ifdef CONFIG_COMPAT
1637         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1638 #endif
1639         .mmap           = kvm_vcpu_mmap,
1640         .llseek         = noop_llseek,
1641 };
1642
1643 /*
1644  * Allocates an inode for the vcpu.
1645  */
1646 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1647 {
1648         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1649 }
1650
1651 /*
1652  * Creates some virtual cpus.  Good luck creating more than one.
1653  */
1654 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1655 {
1656         int r;
1657         struct kvm_vcpu *vcpu, *v;
1658
1659         vcpu = kvm_arch_vcpu_create(kvm, id);
1660         if (IS_ERR(vcpu))
1661                 return PTR_ERR(vcpu);
1662
1663         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1664
1665         r = kvm_arch_vcpu_setup(vcpu);
1666         if (r)
1667                 goto vcpu_destroy;
1668
1669         mutex_lock(&kvm->lock);
1670         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1671                 r = -EINVAL;
1672                 goto unlock_vcpu_destroy;
1673         }
1674
1675         kvm_for_each_vcpu(r, v, kvm)
1676                 if (v->vcpu_id == id) {
1677                         r = -EEXIST;
1678                         goto unlock_vcpu_destroy;
1679                 }
1680
1681         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1682
1683         /* Now it's all set up, let userspace reach it */
1684         kvm_get_kvm(kvm);
1685         r = create_vcpu_fd(vcpu);
1686         if (r < 0) {
1687                 kvm_put_kvm(kvm);
1688                 goto unlock_vcpu_destroy;
1689         }
1690
1691         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1692         smp_wmb();
1693         atomic_inc(&kvm->online_vcpus);
1694
1695 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1696         if (kvm->bsp_vcpu_id == id)
1697                 kvm->bsp_vcpu = vcpu;
1698 #endif
1699         mutex_unlock(&kvm->lock);
1700         return r;
1701
1702 unlock_vcpu_destroy:
1703         mutex_unlock(&kvm->lock);
1704 vcpu_destroy:
1705         kvm_arch_vcpu_destroy(vcpu);
1706         return r;
1707 }
1708
1709 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1710 {
1711         if (sigset) {
1712                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1713                 vcpu->sigset_active = 1;
1714                 vcpu->sigset = *sigset;
1715         } else
1716                 vcpu->sigset_active = 0;
1717         return 0;
1718 }
1719
1720 static long kvm_vcpu_ioctl(struct file *filp,
1721                            unsigned int ioctl, unsigned long arg)
1722 {
1723         struct kvm_vcpu *vcpu = filp->private_data;
1724         void __user *argp = (void __user *)arg;
1725         int r;
1726         struct kvm_fpu *fpu = NULL;
1727         struct kvm_sregs *kvm_sregs = NULL;
1728
1729         if (vcpu->kvm->mm != current->mm)
1730                 return -EIO;
1731
1732 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1733         /*
1734          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1735          * so vcpu_load() would break it.
1736          */
1737         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1738                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1739 #endif
1740
1741
1742         vcpu_load(vcpu);
1743         switch (ioctl) {
1744         case KVM_RUN:
1745                 r = -EINVAL;
1746                 if (arg)
1747                         goto out;
1748                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1749                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1750                 break;
1751         case KVM_GET_REGS: {
1752                 struct kvm_regs *kvm_regs;
1753
1754                 r = -ENOMEM;
1755                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1756                 if (!kvm_regs)
1757                         goto out;
1758                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1759                 if (r)
1760                         goto out_free1;
1761                 r = -EFAULT;
1762                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1763                         goto out_free1;
1764                 r = 0;
1765 out_free1:
1766                 kfree(kvm_regs);
1767                 break;
1768         }
1769         case KVM_SET_REGS: {
1770                 struct kvm_regs *kvm_regs;
1771
1772                 r = -ENOMEM;
1773                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1774                 if (!kvm_regs)
1775                         goto out;
1776                 r = -EFAULT;
1777                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1778                         goto out_free2;
1779                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1780                 if (r)
1781                         goto out_free2;
1782                 r = 0;
1783 out_free2:
1784                 kfree(kvm_regs);
1785                 break;
1786         }
1787         case KVM_GET_SREGS: {
1788                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1789                 r = -ENOMEM;
1790                 if (!kvm_sregs)
1791                         goto out;
1792                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1793                 if (r)
1794                         goto out;
1795                 r = -EFAULT;
1796                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1797                         goto out;
1798                 r = 0;
1799                 break;
1800         }
1801         case KVM_SET_SREGS: {
1802                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1803                 r = -ENOMEM;
1804                 if (!kvm_sregs)
1805                         goto out;
1806                 r = -EFAULT;
1807                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1808                         goto out;
1809                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1810                 if (r)
1811                         goto out;
1812                 r = 0;
1813                 break;
1814         }
1815         case KVM_GET_MP_STATE: {
1816                 struct kvm_mp_state mp_state;
1817
1818                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1819                 if (r)
1820                         goto out;
1821                 r = -EFAULT;
1822                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1823                         goto out;
1824                 r = 0;
1825                 break;
1826         }
1827         case KVM_SET_MP_STATE: {
1828                 struct kvm_mp_state mp_state;
1829
1830                 r = -EFAULT;
1831                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1832                         goto out;
1833                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1834                 if (r)
1835                         goto out;
1836                 r = 0;
1837                 break;
1838         }
1839         case KVM_TRANSLATE: {
1840                 struct kvm_translation tr;
1841
1842                 r = -EFAULT;
1843                 if (copy_from_user(&tr, argp, sizeof tr))
1844                         goto out;
1845                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1846                 if (r)
1847                         goto out;
1848                 r = -EFAULT;
1849                 if (copy_to_user(argp, &tr, sizeof tr))
1850                         goto out;
1851                 r = 0;
1852                 break;
1853         }
1854         case KVM_SET_GUEST_DEBUG: {
1855                 struct kvm_guest_debug dbg;
1856
1857                 r = -EFAULT;
1858                 if (copy_from_user(&dbg, argp, sizeof dbg))
1859                         goto out;
1860                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1861                 if (r)
1862                         goto out;
1863                 r = 0;
1864                 break;
1865         }
1866         case KVM_SET_SIGNAL_MASK: {
1867                 struct kvm_signal_mask __user *sigmask_arg = argp;
1868                 struct kvm_signal_mask kvm_sigmask;
1869                 sigset_t sigset, *p;
1870
1871                 p = NULL;
1872                 if (argp) {
1873                         r = -EFAULT;
1874                         if (copy_from_user(&kvm_sigmask, argp,
1875                                            sizeof kvm_sigmask))
1876                                 goto out;
1877                         r = -EINVAL;
1878                         if (kvm_sigmask.len != sizeof sigset)
1879                                 goto out;
1880                         r = -EFAULT;
1881                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1882                                            sizeof sigset))
1883                                 goto out;
1884                         p = &sigset;
1885                 }
1886                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1887                 break;
1888         }
1889         case KVM_GET_FPU: {
1890                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1891                 r = -ENOMEM;
1892                 if (!fpu)
1893                         goto out;
1894                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1895                 if (r)
1896                         goto out;
1897                 r = -EFAULT;
1898                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1899                         goto out;
1900                 r = 0;
1901                 break;
1902         }
1903         case KVM_SET_FPU: {
1904                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1905                 r = -ENOMEM;
1906                 if (!fpu)
1907                         goto out;
1908                 r = -EFAULT;
1909                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1910                         goto out;
1911                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1912                 if (r)
1913                         goto out;
1914                 r = 0;
1915                 break;
1916         }
1917         default:
1918                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1919         }
1920 out:
1921         vcpu_put(vcpu);
1922         kfree(fpu);
1923         kfree(kvm_sregs);
1924         return r;
1925 }
1926
1927 #ifdef CONFIG_COMPAT
1928 static long kvm_vcpu_compat_ioctl(struct file *filp,
1929                                   unsigned int ioctl, unsigned long arg)
1930 {
1931         struct kvm_vcpu *vcpu = filp->private_data;
1932         void __user *argp = compat_ptr(arg);
1933         int r;
1934
1935         if (vcpu->kvm->mm != current->mm)
1936                 return -EIO;
1937
1938         switch (ioctl) {
1939         case KVM_SET_SIGNAL_MASK: {
1940                 struct kvm_signal_mask __user *sigmask_arg = argp;
1941                 struct kvm_signal_mask kvm_sigmask;
1942                 compat_sigset_t csigset;
1943                 sigset_t sigset;
1944
1945                 if (argp) {
1946                         r = -EFAULT;
1947                         if (copy_from_user(&kvm_sigmask, argp,
1948                                            sizeof kvm_sigmask))
1949                                 goto out;
1950                         r = -EINVAL;
1951                         if (kvm_sigmask.len != sizeof csigset)
1952                                 goto out;
1953                         r = -EFAULT;
1954                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1955                                            sizeof csigset))
1956                                 goto out;
1957                 }
1958                 sigset_from_compat(&sigset, &csigset);
1959                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1960                 break;
1961         }
1962         default:
1963                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1964         }
1965
1966 out:
1967         return r;
1968 }
1969 #endif
1970
1971 static long kvm_vm_ioctl(struct file *filp,
1972                            unsigned int ioctl, unsigned long arg)
1973 {
1974         struct kvm *kvm = filp->private_data;
1975         void __user *argp = (void __user *)arg;
1976         int r;
1977
1978         if (kvm->mm != current->mm)
1979                 return -EIO;
1980         switch (ioctl) {
1981         case KVM_CREATE_VCPU:
1982                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1983                 if (r < 0)
1984                         goto out;
1985                 break;
1986         case KVM_SET_USER_MEMORY_REGION: {
1987                 struct kvm_userspace_memory_region kvm_userspace_mem;
1988
1989                 r = -EFAULT;
1990                 if (copy_from_user(&kvm_userspace_mem, argp,
1991                                                 sizeof kvm_userspace_mem))
1992                         goto out;
1993
1994                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1995                 if (r)
1996                         goto out;
1997                 break;
1998         }
1999         case KVM_GET_DIRTY_LOG: {
2000                 struct kvm_dirty_log log;
2001
2002                 r = -EFAULT;
2003                 if (copy_from_user(&log, argp, sizeof log))
2004                         goto out;
2005                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2006                 if (r)
2007                         goto out;
2008                 break;
2009         }
2010 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2011         case KVM_REGISTER_COALESCED_MMIO: {
2012                 struct kvm_coalesced_mmio_zone zone;
2013                 r = -EFAULT;
2014                 if (copy_from_user(&zone, argp, sizeof zone))
2015                         goto out;
2016                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2017                 if (r)
2018                         goto out;
2019                 r = 0;
2020                 break;
2021         }
2022         case KVM_UNREGISTER_COALESCED_MMIO: {
2023                 struct kvm_coalesced_mmio_zone zone;
2024                 r = -EFAULT;
2025                 if (copy_from_user(&zone, argp, sizeof zone))
2026                         goto out;
2027                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2028                 if (r)
2029                         goto out;
2030                 r = 0;
2031                 break;
2032         }
2033 #endif
2034         case KVM_IRQFD: {
2035                 struct kvm_irqfd data;
2036
2037                 r = -EFAULT;
2038                 if (copy_from_user(&data, argp, sizeof data))
2039                         goto out;
2040                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2041                 break;
2042         }
2043         case KVM_IOEVENTFD: {
2044                 struct kvm_ioeventfd data;
2045
2046                 r = -EFAULT;
2047                 if (copy_from_user(&data, argp, sizeof data))
2048                         goto out;
2049                 r = kvm_ioeventfd(kvm, &data);
2050                 break;
2051         }
2052 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2053         case KVM_SET_BOOT_CPU_ID:
2054                 r = 0;
2055                 mutex_lock(&kvm->lock);
2056                 if (atomic_read(&kvm->online_vcpus) != 0)
2057                         r = -EBUSY;
2058                 else
2059                         kvm->bsp_vcpu_id = arg;
2060                 mutex_unlock(&kvm->lock);
2061                 break;
2062 #endif
2063         default:
2064                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2065                 if (r == -ENOTTY)
2066                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2067         }
2068 out:
2069         return r;
2070 }
2071
2072 #ifdef CONFIG_COMPAT
2073 struct compat_kvm_dirty_log {
2074         __u32 slot;
2075         __u32 padding1;
2076         union {
2077                 compat_uptr_t dirty_bitmap; /* one bit per page */
2078                 __u64 padding2;
2079         };
2080 };
2081
2082 static long kvm_vm_compat_ioctl(struct file *filp,
2083                            unsigned int ioctl, unsigned long arg)
2084 {
2085         struct kvm *kvm = filp->private_data;
2086         int r;
2087
2088         if (kvm->mm != current->mm)
2089                 return -EIO;
2090         switch (ioctl) {
2091         case KVM_GET_DIRTY_LOG: {
2092                 struct compat_kvm_dirty_log compat_log;
2093                 struct kvm_dirty_log log;
2094
2095                 r = -EFAULT;
2096                 if (copy_from_user(&compat_log, (void __user *)arg,
2097                                    sizeof(compat_log)))
2098                         goto out;
2099                 log.slot         = compat_log.slot;
2100                 log.padding1     = compat_log.padding1;
2101                 log.padding2     = compat_log.padding2;
2102                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2103
2104                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2105                 if (r)
2106                         goto out;
2107                 break;
2108         }
2109         default:
2110                 r = kvm_vm_ioctl(filp, ioctl, arg);
2111         }
2112
2113 out:
2114         return r;
2115 }
2116 #endif
2117
2118 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2119 {
2120         struct page *page[1];
2121         unsigned long addr;
2122         int npages;
2123         gfn_t gfn = vmf->pgoff;
2124         struct kvm *kvm = vma->vm_file->private_data;
2125
2126         addr = gfn_to_hva(kvm, gfn);
2127         if (kvm_is_error_hva(addr))
2128                 return VM_FAULT_SIGBUS;
2129
2130         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2131                                 NULL);
2132         if (unlikely(npages != 1))
2133                 return VM_FAULT_SIGBUS;
2134
2135         vmf->page = page[0];
2136         return 0;
2137 }
2138
2139 static const struct vm_operations_struct kvm_vm_vm_ops = {
2140         .fault = kvm_vm_fault,
2141 };
2142
2143 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2144 {
2145         vma->vm_ops = &kvm_vm_vm_ops;
2146         return 0;
2147 }
2148
2149 static struct file_operations kvm_vm_fops = {
2150         .release        = kvm_vm_release,
2151         .unlocked_ioctl = kvm_vm_ioctl,
2152 #ifdef CONFIG_COMPAT
2153         .compat_ioctl   = kvm_vm_compat_ioctl,
2154 #endif
2155         .mmap           = kvm_vm_mmap,
2156         .llseek         = noop_llseek,
2157 };
2158
2159 static int kvm_dev_ioctl_create_vm(void)
2160 {
2161         int r;
2162         struct kvm *kvm;
2163
2164         kvm = kvm_create_vm();
2165         if (IS_ERR(kvm))
2166                 return PTR_ERR(kvm);
2167 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2168         r = kvm_coalesced_mmio_init(kvm);
2169         if (r < 0) {
2170                 kvm_put_kvm(kvm);
2171                 return r;
2172         }
2173 #endif
2174         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2175         if (r < 0)
2176                 kvm_put_kvm(kvm);
2177
2178         return r;
2179 }
2180
2181 static long kvm_dev_ioctl_check_extension_generic(long arg)
2182 {
2183         switch (arg) {
2184         case KVM_CAP_USER_MEMORY:
2185         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2186         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2187 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2188         case KVM_CAP_SET_BOOT_CPU_ID:
2189 #endif
2190         case KVM_CAP_INTERNAL_ERROR_DATA:
2191                 return 1;
2192 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2193         case KVM_CAP_IRQ_ROUTING:
2194                 return KVM_MAX_IRQ_ROUTES;
2195 #endif
2196         default:
2197                 break;
2198         }
2199         return kvm_dev_ioctl_check_extension(arg);
2200 }
2201
2202 static long kvm_dev_ioctl(struct file *filp,
2203                           unsigned int ioctl, unsigned long arg)
2204 {
2205         long r = -EINVAL;
2206
2207         switch (ioctl) {
2208         case KVM_GET_API_VERSION:
2209                 r = -EINVAL;
2210                 if (arg)
2211                         goto out;
2212                 r = KVM_API_VERSION;
2213                 break;
2214         case KVM_CREATE_VM:
2215                 r = -EINVAL;
2216                 if (arg)
2217                         goto out;
2218                 r = kvm_dev_ioctl_create_vm();
2219                 break;
2220         case KVM_CHECK_EXTENSION:
2221                 r = kvm_dev_ioctl_check_extension_generic(arg);
2222                 break;
2223         case KVM_GET_VCPU_MMAP_SIZE:
2224                 r = -EINVAL;
2225                 if (arg)
2226                         goto out;
2227                 r = PAGE_SIZE;     /* struct kvm_run */
2228 #ifdef CONFIG_X86
2229                 r += PAGE_SIZE;    /* pio data page */
2230 #endif
2231 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2232                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2233 #endif
2234                 break;
2235         case KVM_TRACE_ENABLE:
2236         case KVM_TRACE_PAUSE:
2237         case KVM_TRACE_DISABLE:
2238                 r = -EOPNOTSUPP;
2239                 break;
2240         default:
2241                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2242         }
2243 out:
2244         return r;
2245 }
2246
2247 static struct file_operations kvm_chardev_ops = {
2248         .unlocked_ioctl = kvm_dev_ioctl,
2249         .compat_ioctl   = kvm_dev_ioctl,
2250         .llseek         = noop_llseek,
2251 };
2252
2253 static struct miscdevice kvm_dev = {
2254         KVM_MINOR,
2255         "kvm",
2256         &kvm_chardev_ops,
2257 };
2258
2259 static void hardware_enable_nolock(void *junk)
2260 {
2261         int cpu = raw_smp_processor_id();
2262         int r;
2263
2264         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2265                 return;
2266
2267         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2268
2269         r = kvm_arch_hardware_enable(NULL);
2270
2271         if (r) {
2272                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2273                 atomic_inc(&hardware_enable_failed);
2274                 printk(KERN_INFO "kvm: enabling virtualization on "
2275                                  "CPU%d failed\n", cpu);
2276         }
2277 }
2278
2279 static void hardware_enable(void *junk)
2280 {
2281         raw_spin_lock(&kvm_lock);
2282         hardware_enable_nolock(junk);
2283         raw_spin_unlock(&kvm_lock);
2284 }
2285
2286 static void hardware_disable_nolock(void *junk)
2287 {
2288         int cpu = raw_smp_processor_id();
2289
2290         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2291                 return;
2292         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2293         kvm_arch_hardware_disable(NULL);
2294 }
2295
2296 static void hardware_disable(void *junk)
2297 {
2298         raw_spin_lock(&kvm_lock);
2299         hardware_disable_nolock(junk);
2300         raw_spin_unlock(&kvm_lock);
2301 }
2302
2303 static void hardware_disable_all_nolock(void)
2304 {
2305         BUG_ON(!kvm_usage_count);
2306
2307         kvm_usage_count--;
2308         if (!kvm_usage_count)
2309                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2310 }
2311
2312 static void hardware_disable_all(void)
2313 {
2314         raw_spin_lock(&kvm_lock);
2315         hardware_disable_all_nolock();
2316         raw_spin_unlock(&kvm_lock);
2317 }
2318
2319 static int hardware_enable_all(void)
2320 {
2321         int r = 0;
2322
2323         raw_spin_lock(&kvm_lock);
2324
2325         kvm_usage_count++;
2326         if (kvm_usage_count == 1) {
2327                 atomic_set(&hardware_enable_failed, 0);
2328                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2329
2330                 if (atomic_read(&hardware_enable_failed)) {
2331                         hardware_disable_all_nolock();
2332                         r = -EBUSY;
2333                 }
2334         }
2335
2336         raw_spin_unlock(&kvm_lock);
2337
2338         return r;
2339 }
2340
2341 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2342                            void *v)
2343 {
2344         int cpu = (long)v;
2345
2346         if (!kvm_usage_count)
2347                 return NOTIFY_OK;
2348
2349         val &= ~CPU_TASKS_FROZEN;
2350         switch (val) {
2351         case CPU_DYING:
2352                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2353                        cpu);
2354                 hardware_disable(NULL);
2355                 break;
2356         case CPU_STARTING:
2357                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2358                        cpu);
2359                 hardware_enable(NULL);
2360                 break;
2361         }
2362         return NOTIFY_OK;
2363 }
2364
2365
2366 asmlinkage void kvm_spurious_fault(void)
2367 {
2368         /* Fault while not rebooting.  We want the trace. */
2369         BUG();
2370 }
2371 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2372
2373 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2374                       void *v)
2375 {
2376         /*
2377          * Some (well, at least mine) BIOSes hang on reboot if
2378          * in vmx root mode.
2379          *
2380          * And Intel TXT required VMX off for all cpu when system shutdown.
2381          */
2382         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2383         kvm_rebooting = true;
2384         on_each_cpu(hardware_disable_nolock, NULL, 1);
2385         return NOTIFY_OK;
2386 }
2387
2388 static struct notifier_block kvm_reboot_notifier = {
2389         .notifier_call = kvm_reboot,
2390         .priority = 0,
2391 };
2392
2393 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2394 {
2395         int i;
2396
2397         for (i = 0; i < bus->dev_count; i++) {
2398                 struct kvm_io_device *pos = bus->range[i].dev;
2399
2400                 kvm_iodevice_destructor(pos);
2401         }
2402         kfree(bus);
2403 }
2404
2405 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2406 {
2407         const struct kvm_io_range *r1 = p1;
2408         const struct kvm_io_range *r2 = p2;
2409
2410         if (r1->addr < r2->addr)
2411                 return -1;
2412         if (r1->addr + r1->len > r2->addr + r2->len)
2413                 return 1;
2414         return 0;
2415 }
2416
2417 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2418                           gpa_t addr, int len)
2419 {
2420         if (bus->dev_count == NR_IOBUS_DEVS)
2421                 return -ENOSPC;
2422
2423         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2424                 .addr = addr,
2425                 .len = len,
2426                 .dev = dev,
2427         };
2428
2429         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2430                 kvm_io_bus_sort_cmp, NULL);
2431
2432         return 0;
2433 }
2434
2435 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2436                              gpa_t addr, int len)
2437 {
2438         struct kvm_io_range *range, key;
2439         int off;
2440
2441         key = (struct kvm_io_range) {
2442                 .addr = addr,
2443                 .len = len,
2444         };
2445
2446         range = bsearch(&key, bus->range, bus->dev_count,
2447                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2448         if (range == NULL)
2449                 return -ENOENT;
2450
2451         off = range - bus->range;
2452
2453         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2454                 off--;
2455
2456         return off;
2457 }
2458
2459 /* kvm_io_bus_write - called under kvm->slots_lock */
2460 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2461                      int len, const void *val)
2462 {
2463         int idx;
2464         struct kvm_io_bus *bus;
2465         struct kvm_io_range range;
2466
2467         range = (struct kvm_io_range) {
2468                 .addr = addr,
2469                 .len = len,
2470         };
2471
2472         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2473         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2474         if (idx < 0)
2475                 return -EOPNOTSUPP;
2476
2477         while (idx < bus->dev_count &&
2478                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2479                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2480                         return 0;
2481                 idx++;
2482         }
2483
2484         return -EOPNOTSUPP;
2485 }
2486
2487 /* kvm_io_bus_read - called under kvm->slots_lock */
2488 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2489                     int len, void *val)
2490 {
2491         int idx;
2492         struct kvm_io_bus *bus;
2493         struct kvm_io_range range;
2494
2495         range = (struct kvm_io_range) {
2496                 .addr = addr,
2497                 .len = len,
2498         };
2499
2500         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2501         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2502         if (idx < 0)
2503                 return -EOPNOTSUPP;
2504
2505         while (idx < bus->dev_count &&
2506                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2507                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2508                         return 0;
2509                 idx++;
2510         }
2511
2512         return -EOPNOTSUPP;
2513 }
2514
2515 /* Caller must hold slots_lock. */
2516 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2517                             int len, struct kvm_io_device *dev)
2518 {
2519         struct kvm_io_bus *new_bus, *bus;
2520
2521         bus = kvm->buses[bus_idx];
2522         if (bus->dev_count > NR_IOBUS_DEVS-1)
2523                 return -ENOSPC;
2524
2525         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2526         if (!new_bus)
2527                 return -ENOMEM;
2528         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2529         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2530         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2531         synchronize_srcu_expedited(&kvm->srcu);
2532         kfree(bus);
2533
2534         return 0;
2535 }
2536
2537 /* Caller must hold slots_lock. */
2538 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2539                               struct kvm_io_device *dev)
2540 {
2541         int i, r;
2542         struct kvm_io_bus *new_bus, *bus;
2543
2544         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2545         if (!new_bus)
2546                 return -ENOMEM;
2547
2548         bus = kvm->buses[bus_idx];
2549         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2550
2551         r = -ENOENT;
2552         for (i = 0; i < new_bus->dev_count; i++)
2553                 if (new_bus->range[i].dev == dev) {
2554                         r = 0;
2555                         new_bus->dev_count--;
2556                         new_bus->range[i] = new_bus->range[new_bus->dev_count];
2557                         sort(new_bus->range, new_bus->dev_count,
2558                              sizeof(struct kvm_io_range),
2559                              kvm_io_bus_sort_cmp, NULL);
2560                         break;
2561                 }
2562
2563         if (r) {
2564                 kfree(new_bus);
2565                 return r;
2566         }
2567
2568         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2569         synchronize_srcu_expedited(&kvm->srcu);
2570         kfree(bus);
2571         return r;
2572 }
2573
2574 static struct notifier_block kvm_cpu_notifier = {
2575         .notifier_call = kvm_cpu_hotplug,
2576 };
2577
2578 static int vm_stat_get(void *_offset, u64 *val)
2579 {
2580         unsigned offset = (long)_offset;
2581         struct kvm *kvm;
2582
2583         *val = 0;
2584         raw_spin_lock(&kvm_lock);
2585         list_for_each_entry(kvm, &vm_list, vm_list)
2586                 *val += *(u32 *)((void *)kvm + offset);
2587         raw_spin_unlock(&kvm_lock);
2588         return 0;
2589 }
2590
2591 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2592
2593 static int vcpu_stat_get(void *_offset, u64 *val)
2594 {
2595         unsigned offset = (long)_offset;
2596         struct kvm *kvm;
2597         struct kvm_vcpu *vcpu;
2598         int i;
2599
2600         *val = 0;
2601         raw_spin_lock(&kvm_lock);
2602         list_for_each_entry(kvm, &vm_list, vm_list)
2603                 kvm_for_each_vcpu(i, vcpu, kvm)
2604                         *val += *(u32 *)((void *)vcpu + offset);
2605
2606         raw_spin_unlock(&kvm_lock);
2607         return 0;
2608 }
2609
2610 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2611
2612 static const struct file_operations *stat_fops[] = {
2613         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2614         [KVM_STAT_VM]   = &vm_stat_fops,
2615 };
2616
2617 static void kvm_init_debug(void)
2618 {
2619         struct kvm_stats_debugfs_item *p;
2620
2621         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2622         for (p = debugfs_entries; p->name; ++p)
2623                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2624                                                 (void *)(long)p->offset,
2625                                                 stat_fops[p->kind]);
2626 }
2627
2628 static void kvm_exit_debug(void)
2629 {
2630         struct kvm_stats_debugfs_item *p;
2631
2632         for (p = debugfs_entries; p->name; ++p)
2633                 debugfs_remove(p->dentry);
2634         debugfs_remove(kvm_debugfs_dir);
2635 }
2636
2637 static int kvm_suspend(void)
2638 {
2639         if (kvm_usage_count)
2640                 hardware_disable_nolock(NULL);
2641         return 0;
2642 }
2643
2644 static void kvm_resume(void)
2645 {
2646         if (kvm_usage_count) {
2647                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2648                 hardware_enable_nolock(NULL);
2649         }
2650 }
2651
2652 static struct syscore_ops kvm_syscore_ops = {
2653         .suspend = kvm_suspend,
2654         .resume = kvm_resume,
2655 };
2656
2657 struct page *bad_page;
2658 pfn_t bad_pfn;
2659
2660 static inline
2661 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2662 {
2663         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2664 }
2665
2666 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2667 {
2668         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2669
2670         kvm_arch_vcpu_load(vcpu, cpu);
2671 }
2672
2673 static void kvm_sched_out(struct preempt_notifier *pn,
2674                           struct task_struct *next)
2675 {
2676         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2677
2678         kvm_arch_vcpu_put(vcpu);
2679 }
2680
2681 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2682                   struct module *module)
2683 {
2684         int r;
2685         int cpu;
2686
2687         r = kvm_arch_init(opaque);
2688         if (r)
2689                 goto out_fail;
2690
2691         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2692
2693         if (bad_page == NULL) {
2694                 r = -ENOMEM;
2695                 goto out;
2696         }
2697
2698         bad_pfn = page_to_pfn(bad_page);
2699
2700         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2701
2702         if (hwpoison_page == NULL) {
2703                 r = -ENOMEM;
2704                 goto out_free_0;
2705         }
2706
2707         hwpoison_pfn = page_to_pfn(hwpoison_page);
2708
2709         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2710
2711         if (fault_page == NULL) {
2712                 r = -ENOMEM;
2713                 goto out_free_0;
2714         }
2715
2716         fault_pfn = page_to_pfn(fault_page);
2717
2718         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2719                 r = -ENOMEM;
2720                 goto out_free_0;
2721         }
2722
2723         r = kvm_arch_hardware_setup();
2724         if (r < 0)
2725                 goto out_free_0a;
2726
2727         for_each_online_cpu(cpu) {
2728                 smp_call_function_single(cpu,
2729                                 kvm_arch_check_processor_compat,
2730                                 &r, 1);
2731                 if (r < 0)
2732                         goto out_free_1;
2733         }
2734
2735         r = register_cpu_notifier(&kvm_cpu_notifier);
2736         if (r)
2737                 goto out_free_2;
2738         register_reboot_notifier(&kvm_reboot_notifier);
2739
2740         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2741         if (!vcpu_align)
2742                 vcpu_align = __alignof__(struct kvm_vcpu);
2743         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2744                                            0, NULL);
2745         if (!kvm_vcpu_cache) {
2746                 r = -ENOMEM;
2747                 goto out_free_3;
2748         }
2749
2750         r = kvm_async_pf_init();
2751         if (r)
2752                 goto out_free;
2753
2754         kvm_chardev_ops.owner = module;
2755         kvm_vm_fops.owner = module;
2756         kvm_vcpu_fops.owner = module;
2757
2758         r = misc_register(&kvm_dev);
2759         if (r) {
2760                 printk(KERN_ERR "kvm: misc device register failed\n");
2761                 goto out_unreg;
2762         }
2763
2764         register_syscore_ops(&kvm_syscore_ops);
2765
2766         kvm_preempt_ops.sched_in = kvm_sched_in;
2767         kvm_preempt_ops.sched_out = kvm_sched_out;
2768
2769         kvm_init_debug();
2770
2771         return 0;
2772
2773 out_unreg:
2774         kvm_async_pf_deinit();
2775 out_free:
2776         kmem_cache_destroy(kvm_vcpu_cache);
2777 out_free_3:
2778         unregister_reboot_notifier(&kvm_reboot_notifier);
2779         unregister_cpu_notifier(&kvm_cpu_notifier);
2780 out_free_2:
2781 out_free_1:
2782         kvm_arch_hardware_unsetup();
2783 out_free_0a:
2784         free_cpumask_var(cpus_hardware_enabled);
2785 out_free_0:
2786         if (fault_page)
2787                 __free_page(fault_page);
2788         if (hwpoison_page)
2789                 __free_page(hwpoison_page);
2790         __free_page(bad_page);
2791 out:
2792         kvm_arch_exit();
2793 out_fail:
2794         return r;
2795 }
2796 EXPORT_SYMBOL_GPL(kvm_init);
2797
2798 void kvm_exit(void)
2799 {
2800         kvm_exit_debug();
2801         misc_deregister(&kvm_dev);
2802         kmem_cache_destroy(kvm_vcpu_cache);
2803         kvm_async_pf_deinit();
2804         unregister_syscore_ops(&kvm_syscore_ops);
2805         unregister_reboot_notifier(&kvm_reboot_notifier);
2806         unregister_cpu_notifier(&kvm_cpu_notifier);
2807         on_each_cpu(hardware_disable_nolock, NULL, 1);
2808         kvm_arch_hardware_unsetup();
2809         kvm_arch_exit();
2810         free_cpumask_var(cpus_hardware_enabled);
2811         __free_page(hwpoison_page);
2812         __free_page(bad_page);
2813 }
2814 EXPORT_SYMBOL_GPL(kvm_exit);