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 bool kvm_is_mmio_pfn(pfn_t pfn)
 104 {
 105         if (pfn_valid(pfn)) {
 106                 int reserved;
 107                 struct page *tail = pfn_to_page(pfn);
 108                 struct page *head = compound_trans_head(tail);
 109                 reserved = PageReserved(head);
 110                 if (head != tail) {
 111                         /*
 112                          * "head" is not a dangling pointer
 113                          * (compound_trans_head takes care of that)
 114                          * but the hugepage may have been splitted
 115                          * from under us (and we may not hold a
 116                          * reference count on the head page so it can
 117                          * be reused before we run PageReferenced), so
 118                          * we've to check PageTail before returning
 119                          * what we just read.
 120                          */
 121                         smp_rmb();
 122                         if (PageTail(tail))
 123                                 return reserved;
 124                 }
 125                 return PageReserved(tail);
 126         }
 127
 128         return true;
 129 }
 130
 131 /*
 132  * Switches to specified vcpu, until a matching vcpu_put()
 133  */
 134 int vcpu_load(struct kvm_vcpu *vcpu)
 135 {
 136         int cpu;
 137
 138         if (mutex_lock_killable(&vcpu->mutex))
 139                 return -EINTR;
 140         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 141                 /* The thread running this VCPU changed. */
 142                 struct pid *oldpid = vcpu->pid;
 143                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 144                 rcu_assign_pointer(vcpu->pid, newpid);
 145                 synchronize_rcu();
 146                 put_pid(oldpid);
 147         }
 148         cpu = get_cpu();
 149         preempt_notifier_register(&vcpu->preempt_notifier);
 150         kvm_arch_vcpu_load(vcpu, cpu);
 151         put_cpu();
 152         return 0;
 153 }
 154
 155 void vcpu_put(struct kvm_vcpu *vcpu)
 156 {
 157         preempt_disable();
 158         kvm_arch_vcpu_put(vcpu);
 159         preempt_notifier_unregister(&vcpu->preempt_notifier);
 160         preempt_enable();
 161         mutex_unlock(&vcpu->mutex);
 162 }
 163
 164 static void ack_flush(void *_completed)
 165 {
 166 }
 167
 168 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 169 {
 170         int i, cpu, me;
 171         cpumask_var_t cpus;
 172         bool called = true;
 173         struct kvm_vcpu *vcpu;
 174
 175         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 176
 177         me = get_cpu();
 178         kvm_for_each_vcpu(i, vcpu, kvm) {
 179                 kvm_make_request(req, vcpu);
 180                 cpu = vcpu->cpu;
 181
 182                 /* Set ->requests bit before we read ->mode */
 183                 smp_mb();
 184
 185                 if (cpus != NULL && cpu != -1 && cpu != me &&
 186                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 187                         cpumask_set_cpu(cpu, cpus);
 188         }
 189         if (unlikely(cpus == NULL))
 190                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 191         else if (!cpumask_empty(cpus))
 192                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 193         else
 194                 called = false;
 195         put_cpu();
 196         free_cpumask_var(cpus);
 197         return called;
 198 }
 199
 200 void kvm_flush_remote_tlbs(struct kvm *kvm)
 201 {
 202         long dirty_count = kvm->tlbs_dirty;
 203
 204         smp_mb();
 205         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 206                 ++kvm->stat.remote_tlb_flush;
 207         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 208 }
 209
 210 void kvm_reload_remote_mmus(struct kvm *kvm)
 211 {
 212         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 213 }
 214
 215 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 216 {
 217         struct page *page;
 218         int r;
 219
 220         mutex_init(&vcpu->mutex);
 221         vcpu->cpu = -1;
 222         vcpu->kvm = kvm;
 223         vcpu->vcpu_id = id;
 224         vcpu->pid = NULL;
 225         init_waitqueue_head(&vcpu->wq);
 226         kvm_async_pf_vcpu_init(vcpu);
 227
 228         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 229         if (!page) {
 230                 r = -ENOMEM;
 231                 goto fail;
 232         }
 233         vcpu->run = page_address(page);
 234
 235         kvm_vcpu_set_in_spin_loop(vcpu, false);
 236         kvm_vcpu_set_dy_eligible(vcpu, false);
 237
 238         r = kvm_arch_vcpu_init(vcpu);
 239         if (r < 0)
 240                 goto fail_free_run;
 241         return 0;
 242
 243 fail_free_run:
 244         free_page((unsigned long)vcpu->run);
 245 fail:
 246         return r;
 247 }
 248 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 249
 250 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 251 {
 252         put_pid(vcpu->pid);
 253         kvm_arch_vcpu_uninit(vcpu);
 254         free_page((unsigned long)vcpu->run);
 255 }
 256 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 257
 258 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 259 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 260 {
 261         return container_of(mn, struct kvm, mmu_notifier);
 262 }
 263
 264 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 265                                              struct mm_struct *mm,
 266                                              unsigned long address)
 267 {
 268         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 269         int need_tlb_flush, idx;
 270
 271         /*
 272          * When ->invalidate_page runs, the linux pte has been zapped
 273          * already but the page is still allocated until
 274          * ->invalidate_page returns. So if we increase the sequence
 275          * here the kvm page fault will notice if the spte can't be
 276          * established because the page is going to be freed. If
 277          * instead the kvm page fault establishes the spte before
 278          * ->invalidate_page runs, kvm_unmap_hva will release it
 279          * before returning.
 280          *
 281          * The sequence increase only need to be seen at spin_unlock
 282          * time, and not at spin_lock time.
 283          *
 284          * Increasing the sequence after the spin_unlock would be
 285          * unsafe because the kvm page fault could then establish the
 286          * pte after kvm_unmap_hva returned, without noticing the page
 287          * is going to be freed.
 288          */
 289         idx = srcu_read_lock(&kvm->srcu);
 290         spin_lock(&kvm->mmu_lock);
 291
 292         kvm->mmu_notifier_seq++;
 293         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 294         /* we've to flush the tlb before the pages can be freed */
 295         if (need_tlb_flush)
 296                 kvm_flush_remote_tlbs(kvm);
 297
 298         spin_unlock(&kvm->mmu_lock);
 299         srcu_read_unlock(&kvm->srcu, idx);
 300 }
 301
 302 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 303                                         struct mm_struct *mm,
 304                                         unsigned long address,
 305                                         pte_t pte)
 306 {
 307         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 308         int idx;
 309
 310         idx = srcu_read_lock(&kvm->srcu);
 311         spin_lock(&kvm->mmu_lock);
 312         kvm->mmu_notifier_seq++;
 313         kvm_set_spte_hva(kvm, address, pte);
 314         spin_unlock(&kvm->mmu_lock);
 315         srcu_read_unlock(&kvm->srcu, idx);
 316 }
 317
 318 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 319                                                     struct mm_struct *mm,
 320                                                     unsigned long start,
 321                                                     unsigned long end)
 322 {
 323         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 324         int need_tlb_flush = 0, idx;
 325
 326         idx = srcu_read_lock(&kvm->srcu);
 327         spin_lock(&kvm->mmu_lock);
 328         /*
 329          * The count increase must become visible at unlock time as no
 330          * spte can be established without taking the mmu_lock and
 331          * count is also read inside the mmu_lock critical section.
 332          */
 333         kvm->mmu_notifier_count++;
 334         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 335         need_tlb_flush |= kvm->tlbs_dirty;
 336         /* we've to flush the tlb before the pages can be freed */
 337         if (need_tlb_flush)
 338                 kvm_flush_remote_tlbs(kvm);
 339
 340         spin_unlock(&kvm->mmu_lock);
 341         srcu_read_unlock(&kvm->srcu, idx);
 342 }
 343
 344 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 345                                                   struct mm_struct *mm,
 346                                                   unsigned long start,
 347                                                   unsigned long end)
 348 {
 349         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 350
 351         spin_lock(&kvm->mmu_lock);
 352         /*
 353          * This sequence increase will notify the kvm page fault that
 354          * the page that is going to be mapped in the spte could have
 355          * been freed.
 356          */
 357         kvm->mmu_notifier_seq++;
 358         smp_wmb();
 359         /*
 360          * The above sequence increase must be visible before the
 361          * below count decrease, which is ensured by the smp_wmb above
 362          * in conjunction with the smp_rmb in mmu_notifier_retry().
 363          */
 364         kvm->mmu_notifier_count--;
 365         spin_unlock(&kvm->mmu_lock);
 366
 367         BUG_ON(kvm->mmu_notifier_count < 0);
 368 }
 369
 370 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 371                                               struct mm_struct *mm,
 372                                               unsigned long address)
 373 {
 374         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 375         int young, idx;
 376
 377         idx = srcu_read_lock(&kvm->srcu);
 378         spin_lock(&kvm->mmu_lock);
 379
 380         young = kvm_age_hva(kvm, address);
 381         if (young)
 382                 kvm_flush_remote_tlbs(kvm);
 383
 384         spin_unlock(&kvm->mmu_lock);
 385         srcu_read_unlock(&kvm->srcu, idx);
 386
 387         return young;
 388 }
 389
 390 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 391                                        struct mm_struct *mm,
 392                                        unsigned long address)
 393 {
 394         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 395         int young, idx;
 396
 397         idx = srcu_read_lock(&kvm->srcu);
 398         spin_lock(&kvm->mmu_lock);
 399         young = kvm_test_age_hva(kvm, address);
 400         spin_unlock(&kvm->mmu_lock);
 401         srcu_read_unlock(&kvm->srcu, idx);
 402
 403         return young;
 404 }
 405
 406 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 407                                      struct mm_struct *mm)
 408 {
 409         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 410         int idx;
 411
 412         idx = srcu_read_lock(&kvm->srcu);
 413         kvm_arch_flush_shadow_all(kvm);
 414         srcu_read_unlock(&kvm->srcu, idx);
 415 }
 416
 417 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 418         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 419         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 420         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 421         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 422         .test_young             = kvm_mmu_notifier_test_young,
 423         .change_pte             = kvm_mmu_notifier_change_pte,
 424         .release                = kvm_mmu_notifier_release,
 425 };
 426
 427 static int kvm_init_mmu_notifier(struct kvm *kvm)
 428 {
 429         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 430         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 431 }
 432
 433 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 434
 435 static int kvm_init_mmu_notifier(struct kvm *kvm)
 436 {
 437         return 0;
 438 }
 439
 440 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 441
 442 static void kvm_init_memslots_id(struct kvm *kvm)
 443 {
 444         int i;
 445         struct kvm_memslots *slots = kvm->memslots;
 446
 447         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 448                 slots->id_to_index[i] = slots->memslots[i].id = i;
 449 }
 450
 451 static struct kvm *kvm_create_vm(unsigned long type)
 452 {
 453         int r, i;
 454         struct kvm *kvm = kvm_arch_alloc_vm();
 455
 456         if (!kvm)
 457                 return ERR_PTR(-ENOMEM);
 458
 459         r = kvm_arch_init_vm(kvm, type);
 460         if (r)
 461                 goto out_err_nodisable;
 462
 463         r = hardware_enable_all();
 464         if (r)
 465                 goto out_err_nodisable;
 466
 467 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 468         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 469         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 470 #endif
 471
 472         r = -ENOMEM;
 473         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 474         if (!kvm->memslots)
 475                 goto out_err_nosrcu;
 476         kvm_init_memslots_id(kvm);
 477         if (init_srcu_struct(&kvm->srcu))
 478                 goto out_err_nosrcu;
 479         for (i = 0; i < KVM_NR_BUSES; i++) {
 480                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 481                                         GFP_KERNEL);
 482                 if (!kvm->buses[i])
 483                         goto out_err;
 484         }
 485
 486         spin_lock_init(&kvm->mmu_lock);
 487         kvm->mm = current->mm;
 488         atomic_inc(&kvm->mm->mm_count);
 489         kvm_eventfd_init(kvm);
 490         mutex_init(&kvm->lock);
 491         mutex_init(&kvm->irq_lock);
 492         mutex_init(&kvm->slots_lock);
 493         atomic_set(&kvm->users_count, 1);
 494
 495         r = kvm_init_mmu_notifier(kvm);
 496         if (r)
 497                 goto out_err;
 498
 499         raw_spin_lock(&kvm_lock);
 500         list_add(&kvm->vm_list, &vm_list);
 501         raw_spin_unlock(&kvm_lock);
 502
 503         return kvm;
 504
 505 out_err:
 506         cleanup_srcu_struct(&kvm->srcu);
 507 out_err_nosrcu:
 508         hardware_disable_all();
 509 out_err_nodisable:
 510         for (i = 0; i < KVM_NR_BUSES; i++)
 511                 kfree(kvm->buses[i]);
 512         kfree(kvm->memslots);
 513         kvm_arch_free_vm(kvm);
 514         return ERR_PTR(r);
 515 }
 516
 517 /*
 518  * Avoid using vmalloc for a small buffer.
 519  * Should not be used when the size is statically known.
 520  */
 521 void *kvm_kvzalloc(unsigned long size)
 522 {
 523         if (size > PAGE_SIZE)
 524                 return vzalloc(size);
 525         else
 526                 return kzalloc(size, GFP_KERNEL);
 527 }
 528
 529 void kvm_kvfree(const void *addr)
 530 {
 531         if (is_vmalloc_addr(addr))
 532                 vfree(addr);
 533         else
 534                 kfree(addr);
 535 }
 536
 537 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 538 {
 539         if (!memslot->dirty_bitmap)
 540                 return;
 541
 542         kvm_kvfree(memslot->dirty_bitmap);
 543         memslot->dirty_bitmap = NULL;
 544 }
 545
 546 /*
 547  * Free any memory in @free but not in @dont.
 548  */
 549 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 550                                   struct kvm_memory_slot *dont)
 551 {
 552         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 553                 kvm_destroy_dirty_bitmap(free);
 554
 555         kvm_arch_free_memslot(free, dont);
 556
 557         free->npages = 0;
 558 }
 559
 560 void kvm_free_physmem(struct kvm *kvm)
 561 {
 562         struct kvm_memslots *slots = kvm->memslots;
 563         struct kvm_memory_slot *memslot;
 564
 565         kvm_for_each_memslot(memslot, slots)
 566                 kvm_free_physmem_slot(memslot, NULL);
 567
 568         kfree(kvm->memslots);
 569 }
 570
 571 static void kvm_destroy_vm(struct kvm *kvm)
 572 {
 573         int i;
 574         struct mm_struct *mm = kvm->mm;
 575
 576         kvm_arch_sync_events(kvm);
 577         raw_spin_lock(&kvm_lock);
 578         list_del(&kvm->vm_list);
 579         raw_spin_unlock(&kvm_lock);
 580         kvm_free_irq_routing(kvm);
 581         for (i = 0; i < KVM_NR_BUSES; i++)
 582                 kvm_io_bus_destroy(kvm->buses[i]);
 583         kvm_coalesced_mmio_free(kvm);
 584 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 585         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 586 #else
 587         kvm_arch_flush_shadow_all(kvm);
 588 #endif
 589         kvm_arch_destroy_vm(kvm);
 590         kvm_free_physmem(kvm);
 591         cleanup_srcu_struct(&kvm->srcu);
 592         kvm_arch_free_vm(kvm);
 593         hardware_disable_all();
 594         mmdrop(mm);
 595 }
 596
 597 void kvm_get_kvm(struct kvm *kvm)
 598 {
 599         atomic_inc(&kvm->users_count);
 600 }
 601 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 602
 603 void kvm_put_kvm(struct kvm *kvm)
 604 {
 605         if (atomic_dec_and_test(&kvm->users_count))
 606                 kvm_destroy_vm(kvm);
 607 }
 608 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 609
 610
 611 static int kvm_vm_release(struct inode *inode, struct file *filp)
 612 {
 613         struct kvm *kvm = filp->private_data;
 614
 615         kvm_irqfd_release(kvm);
 616
 617         kvm_put_kvm(kvm);
 618         return 0;
 619 }
 620
 621 /*
 622  * Allocation size is twice as large as the actual dirty bitmap size.
 623  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 624  */
 625 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 626 {
 627 #ifndef CONFIG_S390
 628         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 629
 630         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 631         if (!memslot->dirty_bitmap)
 632                 return -ENOMEM;
 633
 634 #endif /* !CONFIG_S390 */
 635         return 0;
 636 }
 637
 638 static int cmp_memslot(const void *slot1, const void *slot2)
 639 {
 640         struct kvm_memory_slot *s1, *s2;
 641
 642         s1 = (struct kvm_memory_slot *)slot1;
 643         s2 = (struct kvm_memory_slot *)slot2;
 644
 645         if (s1->npages < s2->npages)
 646                 return 1;
 647         if (s1->npages > s2->npages)
 648                 return -1;
 649
 650         return 0;
 651 }
 652
 653 /*
 654  * Sort the memslots base on its size, so the larger slots
 655  * will get better fit.
 656  */
 657 static void sort_memslots(struct kvm_memslots *slots)
 658 {
 659         int i;
 660
 661         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 662               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 663
 664         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 665                 slots->id_to_index[slots->memslots[i].id] = i;
 666 }
 667
 668 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
 669 {
 670         if (new) {
 671                 int id = new->id;
 672                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 673                 unsigned long npages = old->npages;
 674
 675                 *old = *new;
 676                 if (new->npages != npages)
 677                         sort_memslots(slots);
 678         }
 679
 680         slots->generation++;
 681 }
 682
 683 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 684 {
 685         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 686
 687 #ifdef KVM_CAP_READONLY_MEM
 688         valid_flags |= KVM_MEM_READONLY;
 689 #endif
 690
 691         if (mem->flags & ~valid_flags)
 692                 return -EINVAL;
 693
 694         return 0;
 695 }
 696
 697 /*
 698  * Allocate some memory and give it an address in the guest physical address
 699  * space.
 700  *
 701  * Discontiguous memory is allowed, mostly for framebuffers.
 702  *
 703  * Must be called holding mmap_sem for write.
 704  */
 705 int __kvm_set_memory_region(struct kvm *kvm,
 706                             struct kvm_userspace_memory_region *mem,
 707                             int user_alloc)
 708 {
 709         int r;
 710         gfn_t base_gfn;
 711         unsigned long npages;
 712         unsigned long i;
 713         struct kvm_memory_slot *memslot;
 714         struct kvm_memory_slot old, new;
 715         struct kvm_memslots *slots, *old_memslots;
 716
 717         r = check_memory_region_flags(mem);
 718         if (r)
 719                 goto out;
 720
 721         r = -EINVAL;
 722         /* General sanity checks */
 723         if (mem->memory_size & (PAGE_SIZE - 1))
 724                 goto out;
 725         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 726                 goto out;
 727         /* We can read the guest memory with __xxx_user() later on. */
 728         if (user_alloc &&
 729             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 730              !access_ok(VERIFY_WRITE,
 731                         (void __user *)(unsigned long)mem->userspace_addr,
 732                         mem->memory_size)))
 733                 goto out;
 734         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 735                 goto out;
 736         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 737                 goto out;
 738
 739         memslot = id_to_memslot(kvm->memslots, mem->slot);
 740         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 741         npages = mem->memory_size >> PAGE_SHIFT;
 742
 743         r = -EINVAL;
 744         if (npages > KVM_MEM_MAX_NR_PAGES)
 745                 goto out;
 746
 747         if (!npages)
 748                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 749
 750         new = old = *memslot;
 751
 752         new.id = mem->slot;
 753         new.base_gfn = base_gfn;
 754         new.npages = npages;
 755         new.flags = mem->flags;
 756
 757         /* Disallow changing a memory slot's size. */
 758         r = -EINVAL;
 759         if (npages && old.npages && npages != old.npages)
 760                 goto out_free;
 761
 762         /* Check for overlaps */
 763         r = -EEXIST;
 764         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 765                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 766
 767                 if (s == memslot || !s->npages)
 768                         continue;
 769                 if (!((base_gfn + npages <= s->base_gfn) ||
 770                       (base_gfn >= s->base_gfn + s->npages)))
 771                         goto out_free;
 772         }
 773
 774         /* Free page dirty bitmap if unneeded */
 775         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 776                 new.dirty_bitmap = NULL;
 777
 778         r = -ENOMEM;
 779
 780         /* Allocate if a slot is being created */
 781         if (npages && !old.npages) {
 782                 new.user_alloc = user_alloc;
 783                 new.userspace_addr = mem->userspace_addr;
 784
 785                 if (kvm_arch_create_memslot(&new, npages))
 786                         goto out_free;
 787         }
 788
 789         /* Allocate page dirty bitmap if needed */
 790         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 791                 if (kvm_create_dirty_bitmap(&new) < 0)
 792                         goto out_free;
 793                 /* destroy any largepage mappings for dirty tracking */
 794         }
 795
 796         if (!npages || base_gfn != old.base_gfn) {
 797                 struct kvm_memory_slot *slot;
 798
 799                 r = -ENOMEM;
 800                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 801                                 GFP_KERNEL);
 802                 if (!slots)
 803                         goto out_free;
 804                 slot = id_to_memslot(slots, mem->slot);
 805                 slot->flags |= KVM_MEMSLOT_INVALID;
 806
 807                 update_memslots(slots, NULL);
 808
 809                 old_memslots = kvm->memslots;
 810                 rcu_assign_pointer(kvm->memslots, slots);
 811                 synchronize_srcu_expedited(&kvm->srcu);
 812                 /* From this point no new shadow pages pointing to a deleted,
 813                  * or moved, memslot will be created.
 814                  *
 815                  * validation of sp->gfn happens in:
 816                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 817                  *      - kvm_is_visible_gfn (mmu_check_roots)
 818                  */
 819                 kvm_arch_flush_shadow_memslot(kvm, slot);
 820                 kfree(old_memslots);
 821         }
 822
 823         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 824         if (r)
 825                 goto out_free;
 826
 827         /* map/unmap the pages in iommu page table */
 828         if (npages) {
 829                 r = kvm_iommu_map_pages(kvm, &new);
 830                 if (r)
 831                         goto out_free;
 832         } else
 833                 kvm_iommu_unmap_pages(kvm, &old);
 834
 835         r = -ENOMEM;
 836         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 837                         GFP_KERNEL);
 838         if (!slots)
 839                 goto out_free;
 840
 841         /* actual memory is freed via old in kvm_free_physmem_slot below */
 842         if (!npages) {
 843                 new.dirty_bitmap = NULL;
 844                 memset(&new.arch, 0, sizeof(new.arch));
 845         }
 846
 847         update_memslots(slots, &new);
 848         old_memslots = kvm->memslots;
 849         rcu_assign_pointer(kvm->memslots, slots);
 850         synchronize_srcu_expedited(&kvm->srcu);
 851
 852         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 853
 854         kvm_free_physmem_slot(&old, &new);
 855         kfree(old_memslots);
 856
 857         return 0;
 858
 859 out_free:
 860         kvm_free_physmem_slot(&new, &old);
 861 out:
 862         return r;
 863
 864 }
 865 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 866
 867 int kvm_set_memory_region(struct kvm *kvm,
 868                           struct kvm_userspace_memory_region *mem,
 869                           int user_alloc)
 870 {
 871         int r;
 872
 873         mutex_lock(&kvm->slots_lock);
 874         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 875         mutex_unlock(&kvm->slots_lock);
 876         return r;
 877 }
 878 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 879
 880 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 881                                    struct
 882                                    kvm_userspace_memory_region *mem,
 883                                    int user_alloc)
 884 {
 885         if (mem->slot >= KVM_MEMORY_SLOTS)
 886                 return -EINVAL;
 887         return kvm_set_memory_region(kvm, mem, user_alloc);
 888 }
 889
 890 int kvm_get_dirty_log(struct kvm *kvm,
 891                         struct kvm_dirty_log *log, int *is_dirty)
 892 {
 893         struct kvm_memory_slot *memslot;
 894         int r, i;
 895         unsigned long n;
 896         unsigned long any = 0;
 897
 898         r = -EINVAL;
 899         if (log->slot >= KVM_MEMORY_SLOTS)
 900                 goto out;
 901
 902         memslot = id_to_memslot(kvm->memslots, log->slot);
 903         r = -ENOENT;
 904         if (!memslot->dirty_bitmap)
 905                 goto out;
 906
 907         n = kvm_dirty_bitmap_bytes(memslot);
 908
 909         for (i = 0; !any && i < n/sizeof(long); ++i)
 910                 any = memslot->dirty_bitmap[i];
 911
 912         r = -EFAULT;
 913         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 914                 goto out;
 915
 916         if (any)
 917                 *is_dirty = 1;
 918
 919         r = 0;
 920 out:
 921         return r;
 922 }
 923
 924 bool kvm_largepages_enabled(void)
 925 {
 926         return largepages_enabled;
 927 }
 928
 929 void kvm_disable_largepages(void)
 930 {
 931         largepages_enabled = false;
 932 }
 933 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 934
 935 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 936 {
 937         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 938 }
 939 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 940
 941 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 942 {
 943         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 944
 945         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
 946               memslot->flags & KVM_MEMSLOT_INVALID)
 947                 return 0;
 948
 949         return 1;
 950 }
 951 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 952
 953 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 954 {
 955         struct vm_area_struct *vma;
 956         unsigned long addr, size;
 957
 958         size = PAGE_SIZE;
 959
 960         addr = gfn_to_hva(kvm, gfn);
 961         if (kvm_is_error_hva(addr))
 962                 return PAGE_SIZE;
 963
 964         down_read(&current->mm->mmap_sem);
 965         vma = find_vma(current->mm, addr);
 966         if (!vma)
 967                 goto out;
 968
 969         size = vma_kernel_pagesize(vma);
 970
 971 out:
 972         up_read(&current->mm->mmap_sem);
 973
 974         return size;
 975 }
 976
 977 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
 978 {
 979         return slot->flags & KVM_MEM_READONLY;
 980 }
 981
 982 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
 983                                        gfn_t *nr_pages, bool write)
 984 {
 985         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 986                 return KVM_HVA_ERR_BAD;
 987
 988         if (memslot_is_readonly(slot) && write)
 989                 return KVM_HVA_ERR_RO_BAD;
 990
 991         if (nr_pages)
 992                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
 993
 994         return __gfn_to_hva_memslot(slot, gfn);
 995 }
 996
 997 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
 998                                      gfn_t *nr_pages)
 999 {
1000         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1001 }
1002
1003 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1004                                  gfn_t gfn)
1005 {
1006         return gfn_to_hva_many(slot, gfn, NULL);
1007 }
1008 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1009
1010 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1011 {
1012         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1013 }
1014 EXPORT_SYMBOL_GPL(gfn_to_hva);
1015
1016 /*
1017  * The hva returned by this function is only allowed to be read.
1018  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1019  */
1020 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1021 {
1022         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1023 }
1024
1025 static int kvm_read_hva(void *data, void __user *hva, int len)
1026 {
1027         return __copy_from_user(data, hva, len);
1028 }
1029
1030 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1031 {
1032         return __copy_from_user_inatomic(data, hva, len);
1033 }
1034
1035 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1036         unsigned long start, int write, struct page **page)
1037 {
1038         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1039
1040         if (write)
1041                 flags |= FOLL_WRITE;
1042
1043         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1044 }
1045
1046 static inline int check_user_page_hwpoison(unsigned long addr)
1047 {
1048         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1049
1050         rc = __get_user_pages(current, current->mm, addr, 1,
1051                               flags, NULL, NULL, NULL);
1052         return rc == -EHWPOISON;
1053 }
1054
1055 /*
1056  * The atomic path to get the writable pfn which will be stored in @pfn,
1057  * true indicates success, otherwise false is returned.
1058  */
1059 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1060                             bool write_fault, bool *writable, pfn_t *pfn)
1061 {
1062         struct page *page[1];
1063         int npages;
1064
1065         if (!(async || atomic))
1066                 return false;
1067
1068         /*
1069          * Fast pin a writable pfn only if it is a write fault request
1070          * or the caller allows to map a writable pfn for a read fault
1071          * request.
1072          */
1073         if (!(write_fault || writable))
1074                 return false;
1075
1076         npages = __get_user_pages_fast(addr, 1, 1, page);
1077         if (npages == 1) {
1078                 *pfn = page_to_pfn(page[0]);
1079
1080                 if (writable)
1081                         *writable = true;
1082                 return true;
1083         }
1084
1085         return false;
1086 }
1087
1088 /*
1089  * The slow path to get the pfn of the specified host virtual address,
1090  * 1 indicates success, -errno is returned if error is detected.
1091  */
1092 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1093                            bool *writable, pfn_t *pfn)
1094 {
1095         struct page *page[1];
1096         int npages = 0;
1097
1098         might_sleep();
1099
1100         if (writable)
1101                 *writable = write_fault;
1102
1103         if (async) {
1104                 down_read(&current->mm->mmap_sem);
1105                 npages = get_user_page_nowait(current, current->mm,
1106                                               addr, write_fault, page);
1107                 up_read(&current->mm->mmap_sem);
1108         } else
1109                 npages = get_user_pages_fast(addr, 1, write_fault,
1110                                              page);
1111         if (npages != 1)
1112                 return npages;
1113
1114         /* map read fault as writable if possible */
1115         if (unlikely(!write_fault) && writable) {
1116                 struct page *wpage[1];
1117
1118                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1119                 if (npages == 1) {
1120                         *writable = true;
1121                         put_page(page[0]);
1122                         page[0] = wpage[0];
1123                 }
1124
1125                 npages = 1;
1126         }
1127         *pfn = page_to_pfn(page[0]);
1128         return npages;
1129 }
1130
1131 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1132 {
1133         if (unlikely(!(vma->vm_flags & VM_READ)))
1134                 return false;
1135
1136         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1137                 return false;
1138
1139         return true;
1140 }
1141
1142 /*
1143  * Pin guest page in memory and return its pfn.
1144  * @addr: host virtual address which maps memory to the guest
1145  * @atomic: whether this function can sleep
1146  * @async: whether this function need to wait IO complete if the
1147  *         host page is not in the memory
1148  * @write_fault: whether we should get a writable host page
1149  * @writable: whether it allows to map a writable host page for !@write_fault
1150  *
1151  * The function will map a writable host page for these two cases:
1152  * 1): @write_fault = true
1153  * 2): @write_fault = false && @writable, @writable will tell the caller
1154  *     whether the mapping is writable.
1155  */
1156 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1157                         bool write_fault, bool *writable)
1158 {
1159         struct vm_area_struct *vma;
1160         pfn_t pfn = 0;
1161         int npages;
1162
1163         /* we can do it either atomically or asynchronously, not both */
1164         BUG_ON(atomic && async);
1165
1166         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1167                 return pfn;
1168
1169         if (atomic)
1170                 return KVM_PFN_ERR_FAULT;
1171
1172         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1173         if (npages == 1)
1174                 return pfn;
1175
1176         down_read(&current->mm->mmap_sem);
1177         if (npages == -EHWPOISON ||
1178               (!async && check_user_page_hwpoison(addr))) {
1179                 pfn = KVM_PFN_ERR_HWPOISON;
1180                 goto exit;
1181         }
1182
1183         vma = find_vma_intersection(current->mm, addr, addr + 1);
1184
1185         if (vma == NULL)
1186                 pfn = KVM_PFN_ERR_FAULT;
1187         else if ((vma->vm_flags & VM_PFNMAP)) {
1188                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1189                         vma->vm_pgoff;
1190                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1191         } else {
1192                 if (async && vma_is_valid(vma, write_fault))
1193                         *async = true;
1194                 pfn = KVM_PFN_ERR_FAULT;
1195         }
1196 exit:
1197         up_read(&current->mm->mmap_sem);
1198         return pfn;
1199 }
1200
1201 static pfn_t
1202 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1203                      bool *async, bool write_fault, bool *writable)
1204 {
1205         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1206
1207         if (addr == KVM_HVA_ERR_RO_BAD)
1208                 return KVM_PFN_ERR_RO_FAULT;
1209
1210         if (kvm_is_error_hva(addr))
1211                 return KVM_PFN_ERR_BAD;
1212
1213         /* Do not map writable pfn in the readonly memslot. */
1214         if (writable && memslot_is_readonly(slot)) {
1215                 *writable = false;
1216                 writable = NULL;
1217         }
1218
1219         return hva_to_pfn(addr, atomic, async, write_fault,
1220                           writable);
1221 }
1222
1223 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1224                           bool write_fault, bool *writable)
1225 {
1226         struct kvm_memory_slot *slot;
1227
1228         if (async)
1229                 *async = false;
1230
1231         slot = gfn_to_memslot(kvm, gfn);
1232
1233         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1234                                     writable);
1235 }
1236
1237 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1238 {
1239         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1240 }
1241 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1242
1243 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1244                        bool write_fault, bool *writable)
1245 {
1246         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1247 }
1248 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1249
1250 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1251 {
1252         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1253 }
1254 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1255
1256 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1257                       bool *writable)
1258 {
1259         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1260 }
1261 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1262
1263 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1264 {
1265         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1266 }
1267
1268 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1269 {
1270         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1271 }
1272 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1273
1274 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1275                                                                   int nr_pages)
1276 {
1277         unsigned long addr;
1278         gfn_t entry;
1279
1280         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1281         if (kvm_is_error_hva(addr))
1282                 return -1;
1283
1284         if (entry < nr_pages)
1285                 return 0;
1286
1287         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1288 }
1289 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1290
1291 static struct page *kvm_pfn_to_page(pfn_t pfn)
1292 {
1293         if (is_error_pfn(pfn))
1294                 return KVM_ERR_PTR_BAD_PAGE;
1295
1296         if (kvm_is_mmio_pfn(pfn)) {
1297                 WARN_ON(1);
1298                 return KVM_ERR_PTR_BAD_PAGE;
1299         }
1300
1301         return pfn_to_page(pfn);
1302 }
1303
1304 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1305 {
1306         pfn_t pfn;
1307
1308         pfn = gfn_to_pfn(kvm, gfn);
1309
1310         return kvm_pfn_to_page(pfn);
1311 }
1312
1313 EXPORT_SYMBOL_GPL(gfn_to_page);
1314
1315 void kvm_release_page_clean(struct page *page)
1316 {
1317         WARN_ON(is_error_page(page));
1318
1319         kvm_release_pfn_clean(page_to_pfn(page));
1320 }
1321 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1322
1323 void kvm_release_pfn_clean(pfn_t pfn)
1324 {
1325         WARN_ON(is_error_pfn(pfn));
1326
1327         if (!kvm_is_mmio_pfn(pfn))
1328                 put_page(pfn_to_page(pfn));
1329 }
1330 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1331
1332 void kvm_release_page_dirty(struct page *page)
1333 {
1334         WARN_ON(is_error_page(page));
1335
1336         kvm_release_pfn_dirty(page_to_pfn(page));
1337 }
1338 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1339
1340 void kvm_release_pfn_dirty(pfn_t pfn)
1341 {
1342         kvm_set_pfn_dirty(pfn);
1343         kvm_release_pfn_clean(pfn);
1344 }
1345 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1346
1347 void kvm_set_page_dirty(struct page *page)
1348 {
1349         kvm_set_pfn_dirty(page_to_pfn(page));
1350 }
1351 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1352
1353 void kvm_set_pfn_dirty(pfn_t pfn)
1354 {
1355         if (!kvm_is_mmio_pfn(pfn)) {
1356                 struct page *page = pfn_to_page(pfn);
1357                 if (!PageReserved(page))
1358                         SetPageDirty(page);
1359         }
1360 }
1361 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1362
1363 void kvm_set_pfn_accessed(pfn_t pfn)
1364 {
1365         if (!kvm_is_mmio_pfn(pfn))
1366                 mark_page_accessed(pfn_to_page(pfn));
1367 }
1368 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1369
1370 void kvm_get_pfn(pfn_t pfn)
1371 {
1372         if (!kvm_is_mmio_pfn(pfn))
1373                 get_page(pfn_to_page(pfn));
1374 }
1375 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1376
1377 static int next_segment(unsigned long len, int offset)
1378 {
1379         if (len > PAGE_SIZE - offset)
1380                 return PAGE_SIZE - offset;
1381         else
1382                 return len;
1383 }
1384
1385 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1386                         int len)
1387 {
1388         int r;
1389         unsigned long addr;
1390
1391         addr = gfn_to_hva_read(kvm, gfn);
1392         if (kvm_is_error_hva(addr))
1393                 return -EFAULT;
1394         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1395         if (r)
1396                 return -EFAULT;
1397         return 0;
1398 }
1399 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1400
1401 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1402 {
1403         gfn_t gfn = gpa >> PAGE_SHIFT;
1404         int seg;
1405         int offset = offset_in_page(gpa);
1406         int ret;
1407
1408         while ((seg = next_segment(len, offset)) != 0) {
1409                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1410                 if (ret < 0)
1411                         return ret;
1412                 offset = 0;
1413                 len -= seg;
1414                 data += seg;
1415                 ++gfn;
1416         }
1417         return 0;
1418 }
1419 EXPORT_SYMBOL_GPL(kvm_read_guest);
1420
1421 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1422                           unsigned long len)
1423 {
1424         int r;
1425         unsigned long addr;
1426         gfn_t gfn = gpa >> PAGE_SHIFT;
1427         int offset = offset_in_page(gpa);
1428
1429         addr = gfn_to_hva_read(kvm, gfn);
1430         if (kvm_is_error_hva(addr))
1431                 return -EFAULT;
1432         pagefault_disable();
1433         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1434         pagefault_enable();
1435         if (r)
1436                 return -EFAULT;
1437         return 0;
1438 }
1439 EXPORT_SYMBOL(kvm_read_guest_atomic);
1440
1441 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1442                          int offset, int len)
1443 {
1444         int r;
1445         unsigned long addr;
1446
1447         addr = gfn_to_hva(kvm, gfn);
1448         if (kvm_is_error_hva(addr))
1449                 return -EFAULT;
1450         r = __copy_to_user((void __user *)addr + offset, data, len);
1451         if (r)
1452                 return -EFAULT;
1453         mark_page_dirty(kvm, gfn);
1454         return 0;
1455 }
1456 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1457
1458 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1459                     unsigned long len)
1460 {
1461         gfn_t gfn = gpa >> PAGE_SHIFT;
1462         int seg;
1463         int offset = offset_in_page(gpa);
1464         int ret;
1465
1466         while ((seg = next_segment(len, offset)) != 0) {
1467                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1468                 if (ret < 0)
1469                         return ret;
1470                 offset = 0;
1471                 len -= seg;
1472                 data += seg;
1473                 ++gfn;
1474         }
1475         return 0;
1476 }
1477
1478 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1479                               gpa_t gpa)
1480 {
1481         struct kvm_memslots *slots = kvm_memslots(kvm);
1482         int offset = offset_in_page(gpa);
1483         gfn_t gfn = gpa >> PAGE_SHIFT;
1484
1485         ghc->gpa = gpa;
1486         ghc->generation = slots->generation;
1487         ghc->memslot = gfn_to_memslot(kvm, gfn);
1488         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1489         if (!kvm_is_error_hva(ghc->hva))
1490                 ghc->hva += offset;
1491         else
1492                 return -EFAULT;
1493
1494         return 0;
1495 }
1496 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1497
1498 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1499                            void *data, unsigned long len)
1500 {
1501         struct kvm_memslots *slots = kvm_memslots(kvm);
1502         int r;
1503
1504         if (slots->generation != ghc->generation)
1505                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1506
1507         if (kvm_is_error_hva(ghc->hva))
1508                 return -EFAULT;
1509
1510         r = __copy_to_user((void __user *)ghc->hva, data, len);
1511         if (r)
1512                 return -EFAULT;
1513         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1514
1515         return 0;
1516 }
1517 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1518
1519 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1520                            void *data, unsigned long len)
1521 {
1522         struct kvm_memslots *slots = kvm_memslots(kvm);
1523         int r;
1524
1525         if (slots->generation != ghc->generation)
1526                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1527
1528         if (kvm_is_error_hva(ghc->hva))
1529                 return -EFAULT;
1530
1531         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1532         if (r)
1533                 return -EFAULT;
1534
1535         return 0;
1536 }
1537 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1538
1539 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1540 {
1541         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1542                                     offset, len);
1543 }
1544 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1545
1546 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1547 {
1548         gfn_t gfn = gpa >> PAGE_SHIFT;
1549         int seg;
1550         int offset = offset_in_page(gpa);
1551         int ret;
1552
1553         while ((seg = next_segment(len, offset)) != 0) {
1554                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1555                 if (ret < 0)
1556                         return ret;
1557                 offset = 0;
1558                 len -= seg;
1559                 ++gfn;
1560         }
1561         return 0;
1562 }
1563 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1564
1565 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1566                              gfn_t gfn)
1567 {
1568         if (memslot && memslot->dirty_bitmap) {
1569                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1570
1571                 /* TODO: introduce set_bit_le() and use it */
1572                 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1573         }
1574 }
1575
1576 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1577 {
1578         struct kvm_memory_slot *memslot;
1579
1580         memslot = gfn_to_memslot(kvm, gfn);
1581         mark_page_dirty_in_slot(kvm, memslot, gfn);
1582 }
1583
1584 /*
1585  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1586  */
1587 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1588 {
1589         DEFINE_WAIT(wait);
1590
1591         for (;;) {
1592                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1593
1594                 if (kvm_arch_vcpu_runnable(vcpu)) {
1595                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1596                         break;
1597                 }
1598                 if (kvm_cpu_has_pending_timer(vcpu))
1599                         break;
1600                 if (signal_pending(current))
1601                         break;
1602
1603                 schedule();
1604         }
1605
1606         finish_wait(&vcpu->wq, &wait);
1607 }
1608
1609 #ifndef CONFIG_S390
1610 /*
1611  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1612  */
1613 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1614 {
1615         int me;
1616         int cpu = vcpu->cpu;
1617         wait_queue_head_t *wqp;
1618
1619         wqp = kvm_arch_vcpu_wq(vcpu);
1620         if (waitqueue_active(wqp)) {
1621                 wake_up_interruptible(wqp);
1622                 ++vcpu->stat.halt_wakeup;
1623         }
1624
1625         me = get_cpu();
1626         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1627                 if (kvm_arch_vcpu_should_kick(vcpu))
1628                         smp_send_reschedule(cpu);
1629         put_cpu();
1630 }
1631 #endif /* !CONFIG_S390 */
1632
1633 void kvm_resched(struct kvm_vcpu *vcpu)
1634 {
1635         if (!need_resched())
1636                 return;
1637         cond_resched();
1638 }
1639 EXPORT_SYMBOL_GPL(kvm_resched);
1640
1641 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1642 {
1643         struct pid *pid;
1644         struct task_struct *task = NULL;
1645
1646         rcu_read_lock();
1647         pid = rcu_dereference(target->pid);
1648         if (pid)
1649                 task = get_pid_task(target->pid, PIDTYPE_PID);
1650         rcu_read_unlock();
1651         if (!task)
1652                 return false;
1653         if (task->flags & PF_VCPU) {
1654                 put_task_struct(task);
1655                 return false;
1656         }
1657         if (yield_to(task, 1)) {
1658                 put_task_struct(task);
1659                 return true;
1660         }
1661         put_task_struct(task);
1662         return false;
1663 }
1664 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1665
1666 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1667 /*
1668  * Helper that checks whether a VCPU is eligible for directed yield.
1669  * Most eligible candidate to yield is decided by following heuristics:
1670  *
1671  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1672  *  (preempted lock holder), indicated by @in_spin_loop.
1673  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1674  *
1675  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1676  *  chance last time (mostly it has become eligible now since we have probably
1677  *  yielded to lockholder in last iteration. This is done by toggling
1678  *  @dy_eligible each time a VCPU checked for eligibility.)
1679  *
1680  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1681  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1682  *  burning. Giving priority for a potential lock-holder increases lock
1683  *  progress.
1684  *
1685  *  Since algorithm is based on heuristics, accessing another VCPU data without
1686  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1687  *  and continue with next VCPU and so on.
1688  */
1689 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1690 {
1691         bool eligible;
1692
1693         eligible = !vcpu->spin_loop.in_spin_loop ||
1694                         (vcpu->spin_loop.in_spin_loop &&
1695                          vcpu->spin_loop.dy_eligible);
1696
1697         if (vcpu->spin_loop.in_spin_loop)
1698                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1699
1700         return eligible;
1701 }
1702 #endif
1703 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1704 {
1705         struct kvm *kvm = me->kvm;
1706         struct kvm_vcpu *vcpu;
1707         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1708         int yielded = 0;
1709         int pass;
1710         int i;
1711
1712         kvm_vcpu_set_in_spin_loop(me, true);
1713         /*
1714          * We boost the priority of a VCPU that is runnable but not
1715          * currently running, because it got preempted by something
1716          * else and called schedule in __vcpu_run.  Hopefully that
1717          * VCPU is holding the lock that we need and will release it.
1718          * We approximate round-robin by starting at the last boosted VCPU.
1719          */
1720         for (pass = 0; pass < 2 && !yielded; pass++) {
1721                 kvm_for_each_vcpu(i, vcpu, kvm) {
1722                         if (!pass && i <= last_boosted_vcpu) {
1723                                 i = last_boosted_vcpu;
1724                                 continue;
1725                         } else if (pass && i > last_boosted_vcpu)
1726                                 break;
1727                         if (vcpu == me)
1728                                 continue;
1729                         if (waitqueue_active(&vcpu->wq))
1730                                 continue;
1731                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1732                                 continue;
1733                         if (kvm_vcpu_yield_to(vcpu)) {
1734                                 kvm->last_boosted_vcpu = i;
1735                                 yielded = 1;
1736                                 break;
1737                         }
1738                 }
1739         }
1740         kvm_vcpu_set_in_spin_loop(me, false);
1741
1742         /* Ensure vcpu is not eligible during next spinloop */
1743         kvm_vcpu_set_dy_eligible(me, false);
1744 }
1745 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1746
1747 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1748 {
1749         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1750         struct page *page;
1751
1752         if (vmf->pgoff == 0)
1753                 page = virt_to_page(vcpu->run);
1754 #ifdef CONFIG_X86
1755         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1756                 page = virt_to_page(vcpu->arch.pio_data);
1757 #endif
1758 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1759         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1760                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1761 #endif
1762         else
1763                 return kvm_arch_vcpu_fault(vcpu, vmf);
1764         get_page(page);
1765         vmf->page = page;
1766         return 0;
1767 }
1768
1769 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1770         .fault = kvm_vcpu_fault,
1771 };
1772
1773 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1774 {
1775         vma->vm_ops = &kvm_vcpu_vm_ops;
1776         return 0;
1777 }
1778
1779 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1780 {
1781         struct kvm_vcpu *vcpu = filp->private_data;
1782
1783         kvm_put_kvm(vcpu->kvm);
1784         return 0;
1785 }
1786
1787 static struct file_operations kvm_vcpu_fops = {
1788         .release        = kvm_vcpu_release,
1789         .unlocked_ioctl = kvm_vcpu_ioctl,
1790 #ifdef CONFIG_COMPAT
1791         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1792 #endif
1793         .mmap           = kvm_vcpu_mmap,
1794         .llseek         = noop_llseek,
1795 };
1796
1797 /*
1798  * Allocates an inode for the vcpu.
1799  */
1800 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1801 {
1802         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1803 }
1804
1805 /*
1806  * Creates some virtual cpus.  Good luck creating more than one.
1807  */
1808 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1809 {
1810         int r;
1811         struct kvm_vcpu *vcpu, *v;
1812
1813         vcpu = kvm_arch_vcpu_create(kvm, id);
1814         if (IS_ERR(vcpu))
1815                 return PTR_ERR(vcpu);
1816
1817         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1818
1819         r = kvm_arch_vcpu_setup(vcpu);
1820         if (r)
1821                 goto vcpu_destroy;
1822
1823         mutex_lock(&kvm->lock);
1824         if (!kvm_vcpu_compatible(vcpu)) {
1825                 r = -EINVAL;
1826                 goto unlock_vcpu_destroy;
1827         }
1828         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1829                 r = -EINVAL;
1830                 goto unlock_vcpu_destroy;
1831         }
1832
1833         kvm_for_each_vcpu(r, v, kvm)
1834                 if (v->vcpu_id == id) {
1835                         r = -EEXIST;
1836                         goto unlock_vcpu_destroy;
1837                 }
1838
1839         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1840
1841         /* Now it's all set up, let userspace reach it */
1842         kvm_get_kvm(kvm);
1843         r = create_vcpu_fd(vcpu);
1844         if (r < 0) {
1845                 kvm_put_kvm(kvm);
1846                 goto unlock_vcpu_destroy;
1847         }
1848
1849         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1850         smp_wmb();
1851         atomic_inc(&kvm->online_vcpus);
1852
1853         mutex_unlock(&kvm->lock);
1854         return r;
1855
1856 unlock_vcpu_destroy:
1857         mutex_unlock(&kvm->lock);
1858 vcpu_destroy:
1859         kvm_arch_vcpu_destroy(vcpu);
1860         return r;
1861 }
1862
1863 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1864 {
1865         if (sigset) {
1866                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1867                 vcpu->sigset_active = 1;
1868                 vcpu->sigset = *sigset;
1869         } else
1870                 vcpu->sigset_active = 0;
1871         return 0;
1872 }
1873
1874 static long kvm_vcpu_ioctl(struct file *filp,
1875                            unsigned int ioctl, unsigned long arg)
1876 {
1877         struct kvm_vcpu *vcpu = filp->private_data;
1878         void __user *argp = (void __user *)arg;
1879         int r;
1880         struct kvm_fpu *fpu = NULL;
1881         struct kvm_sregs *kvm_sregs = NULL;
1882
1883         if (vcpu->kvm->mm != current->mm)
1884                 return -EIO;
1885
1886 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1887         /*
1888          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1889          * so vcpu_load() would break it.
1890          */
1891         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1892                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1893 #endif
1894
1895
1896         r = vcpu_load(vcpu);
1897         if (r)
1898                 return r;
1899         switch (ioctl) {
1900         case KVM_RUN:
1901                 r = -EINVAL;
1902                 if (arg)
1903                         goto out;
1904                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1905                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1906                 break;
1907         case KVM_GET_REGS: {
1908                 struct kvm_regs *kvm_regs;
1909
1910                 r = -ENOMEM;
1911                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1912                 if (!kvm_regs)
1913                         goto out;
1914                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1915                 if (r)
1916                         goto out_free1;
1917                 r = -EFAULT;
1918                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1919                         goto out_free1;
1920                 r = 0;
1921 out_free1:
1922                 kfree(kvm_regs);
1923                 break;
1924         }
1925         case KVM_SET_REGS: {
1926                 struct kvm_regs *kvm_regs;
1927
1928                 r = -ENOMEM;
1929                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1930                 if (IS_ERR(kvm_regs)) {
1931                         r = PTR_ERR(kvm_regs);
1932                         goto out;
1933                 }
1934                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1935                 if (r)
1936                         goto out_free2;
1937                 r = 0;
1938 out_free2:
1939                 kfree(kvm_regs);
1940                 break;
1941         }
1942         case KVM_GET_SREGS: {
1943                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1944                 r = -ENOMEM;
1945                 if (!kvm_sregs)
1946                         goto out;
1947                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1948                 if (r)
1949                         goto out;
1950                 r = -EFAULT;
1951                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1952                         goto out;
1953                 r = 0;
1954                 break;
1955         }
1956         case KVM_SET_SREGS: {
1957                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1958                 if (IS_ERR(kvm_sregs)) {
1959                         r = PTR_ERR(kvm_sregs);
1960                         goto out;
1961                 }
1962                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1963                 if (r)
1964                         goto out;
1965                 r = 0;
1966                 break;
1967         }
1968         case KVM_GET_MP_STATE: {
1969                 struct kvm_mp_state mp_state;
1970
1971                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1972                 if (r)
1973                         goto out;
1974                 r = -EFAULT;
1975                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1976                         goto out;
1977                 r = 0;
1978                 break;
1979         }
1980         case KVM_SET_MP_STATE: {
1981                 struct kvm_mp_state mp_state;
1982
1983                 r = -EFAULT;
1984                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1985                         goto out;
1986                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1987                 if (r)
1988                         goto out;
1989                 r = 0;
1990                 break;
1991         }
1992         case KVM_TRANSLATE: {
1993                 struct kvm_translation tr;
1994
1995                 r = -EFAULT;
1996                 if (copy_from_user(&tr, argp, sizeof tr))
1997                         goto out;
1998                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1999                 if (r)
2000                         goto out;
2001                 r = -EFAULT;
2002                 if (copy_to_user(argp, &tr, sizeof tr))
2003                         goto out;
2004                 r = 0;
2005                 break;
2006         }
2007         case KVM_SET_GUEST_DEBUG: {
2008                 struct kvm_guest_debug dbg;
2009
2010                 r = -EFAULT;
2011                 if (copy_from_user(&dbg, argp, sizeof dbg))
2012                         goto out;
2013                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2014                 if (r)
2015                         goto out;
2016                 r = 0;
2017                 break;
2018         }
2019         case KVM_SET_SIGNAL_MASK: {
2020                 struct kvm_signal_mask __user *sigmask_arg = argp;
2021                 struct kvm_signal_mask kvm_sigmask;
2022                 sigset_t sigset, *p;
2023
2024                 p = NULL;
2025                 if (argp) {
2026                         r = -EFAULT;
2027                         if (copy_from_user(&kvm_sigmask, argp,
2028                                            sizeof kvm_sigmask))
2029                                 goto out;
2030                         r = -EINVAL;
2031                         if (kvm_sigmask.len != sizeof sigset)
2032                                 goto out;
2033                         r = -EFAULT;
2034                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2035                                            sizeof sigset))
2036                                 goto out;
2037                         p = &sigset;
2038                 }
2039                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2040                 break;
2041         }
2042         case KVM_GET_FPU: {
2043                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2044                 r = -ENOMEM;
2045                 if (!fpu)
2046                         goto out;
2047                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2048                 if (r)
2049                         goto out;
2050                 r = -EFAULT;
2051                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2052                         goto out;
2053                 r = 0;
2054                 break;
2055         }
2056         case KVM_SET_FPU: {
2057                 fpu = memdup_user(argp, sizeof(*fpu));
2058                 if (IS_ERR(fpu)) {
2059                         r = PTR_ERR(fpu);
2060                         goto out;
2061                 }
2062                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2063                 if (r)
2064                         goto out;
2065                 r = 0;
2066                 break;
2067         }
2068         default:
2069                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2070         }
2071 out:
2072         vcpu_put(vcpu);
2073         kfree(fpu);
2074         kfree(kvm_sregs);
2075         return r;
2076 }
2077
2078 #ifdef CONFIG_COMPAT
2079 static long kvm_vcpu_compat_ioctl(struct file *filp,
2080                                   unsigned int ioctl, unsigned long arg)
2081 {
2082         struct kvm_vcpu *vcpu = filp->private_data;
2083         void __user *argp = compat_ptr(arg);
2084         int r;
2085
2086         if (vcpu->kvm->mm != current->mm)
2087                 return -EIO;
2088
2089         switch (ioctl) {
2090         case KVM_SET_SIGNAL_MASK: {
2091                 struct kvm_signal_mask __user *sigmask_arg = argp;
2092                 struct kvm_signal_mask kvm_sigmask;
2093                 compat_sigset_t csigset;
2094                 sigset_t sigset;
2095
2096                 if (argp) {
2097                         r = -EFAULT;
2098                         if (copy_from_user(&kvm_sigmask, argp,
2099                                            sizeof kvm_sigmask))
2100                                 goto out;
2101                         r = -EINVAL;
2102                         if (kvm_sigmask.len != sizeof csigset)
2103                                 goto out;
2104                         r = -EFAULT;
2105                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2106                                            sizeof csigset))
2107                                 goto out;
2108                         sigset_from_compat(&sigset, &csigset);
2109                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2110                 } else
2111                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2112                 break;
2113         }
2114         default:
2115                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2116         }
2117
2118 out:
2119         return r;
2120 }
2121 #endif
2122
2123 static long kvm_vm_ioctl(struct file *filp,
2124                            unsigned int ioctl, unsigned long arg)
2125 {
2126         struct kvm *kvm = filp->private_data;
2127         void __user *argp = (void __user *)arg;
2128         int r;
2129
2130         if (kvm->mm != current->mm)
2131                 return -EIO;
2132         switch (ioctl) {
2133         case KVM_CREATE_VCPU:
2134                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2135                 if (r < 0)
2136                         goto out;
2137                 break;
2138         case KVM_SET_USER_MEMORY_REGION: {
2139                 struct kvm_userspace_memory_region kvm_userspace_mem;
2140
2141                 r = -EFAULT;
2142                 if (copy_from_user(&kvm_userspace_mem, argp,
2143                                                 sizeof kvm_userspace_mem))
2144                         goto out;
2145
2146                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2147                 if (r)
2148                         goto out;
2149                 break;
2150         }
2151         case KVM_GET_DIRTY_LOG: {
2152                 struct kvm_dirty_log log;
2153
2154                 r = -EFAULT;
2155                 if (copy_from_user(&log, argp, sizeof log))
2156                         goto out;
2157                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2158                 if (r)
2159                         goto out;
2160                 break;
2161         }
2162 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2163         case KVM_REGISTER_COALESCED_MMIO: {
2164                 struct kvm_coalesced_mmio_zone zone;
2165                 r = -EFAULT;
2166                 if (copy_from_user(&zone, argp, sizeof zone))
2167                         goto out;
2168                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2169                 if (r)
2170                         goto out;
2171                 r = 0;
2172                 break;
2173         }
2174         case KVM_UNREGISTER_COALESCED_MMIO: {
2175                 struct kvm_coalesced_mmio_zone zone;
2176                 r = -EFAULT;
2177                 if (copy_from_user(&zone, argp, sizeof zone))
2178                         goto out;
2179                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2180                 if (r)
2181                         goto out;
2182                 r = 0;
2183                 break;
2184         }
2185 #endif
2186         case KVM_IRQFD: {
2187                 struct kvm_irqfd data;
2188
2189                 r = -EFAULT;
2190                 if (copy_from_user(&data, argp, sizeof data))
2191                         goto out;
2192                 r = kvm_irqfd(kvm, &data);
2193                 break;
2194         }
2195         case KVM_IOEVENTFD: {
2196                 struct kvm_ioeventfd data;
2197
2198                 r = -EFAULT;
2199                 if (copy_from_user(&data, argp, sizeof data))
2200                         goto out;
2201                 r = kvm_ioeventfd(kvm, &data);
2202                 break;
2203         }
2204 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2205         case KVM_SET_BOOT_CPU_ID:
2206                 r = 0;
2207                 mutex_lock(&kvm->lock);
2208                 if (atomic_read(&kvm->online_vcpus) != 0)
2209                         r = -EBUSY;
2210                 else
2211                         kvm->bsp_vcpu_id = arg;
2212                 mutex_unlock(&kvm->lock);
2213                 break;
2214 #endif
2215 #ifdef CONFIG_HAVE_KVM_MSI
2216         case KVM_SIGNAL_MSI: {
2217                 struct kvm_msi msi;
2218
2219                 r = -EFAULT;
2220                 if (copy_from_user(&msi, argp, sizeof msi))
2221                         goto out;
2222                 r = kvm_send_userspace_msi(kvm, &msi);
2223                 break;
2224         }
2225 #endif
2226 #ifdef __KVM_HAVE_IRQ_LINE
2227         case KVM_IRQ_LINE_STATUS:
2228         case KVM_IRQ_LINE: {
2229                 struct kvm_irq_level irq_event;
2230
2231                 r = -EFAULT;
2232                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2233                         goto out;
2234
2235                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2236                 if (r)
2237                         goto out;
2238
2239                 r = -EFAULT;
2240                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2241                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2242                                 goto out;
2243                 }
2244
2245                 r = 0;
2246                 break;
2247         }
2248 #endif
2249         default:
2250                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2251                 if (r == -ENOTTY)
2252                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2253         }
2254 out:
2255         return r;
2256 }
2257
2258 #ifdef CONFIG_COMPAT
2259 struct compat_kvm_dirty_log {
2260         __u32 slot;
2261         __u32 padding1;
2262         union {
2263                 compat_uptr_t dirty_bitmap; /* one bit per page */
2264                 __u64 padding2;
2265         };
2266 };
2267
2268 static long kvm_vm_compat_ioctl(struct file *filp,
2269                            unsigned int ioctl, unsigned long arg)
2270 {
2271         struct kvm *kvm = filp->private_data;
2272         int r;
2273
2274         if (kvm->mm != current->mm)
2275                 return -EIO;
2276         switch (ioctl) {
2277         case KVM_GET_DIRTY_LOG: {
2278                 struct compat_kvm_dirty_log compat_log;
2279                 struct kvm_dirty_log log;
2280
2281                 r = -EFAULT;
2282                 if (copy_from_user(&compat_log, (void __user *)arg,
2283                                    sizeof(compat_log)))
2284                         goto out;
2285                 log.slot         = compat_log.slot;
2286                 log.padding1     = compat_log.padding1;
2287                 log.padding2     = compat_log.padding2;
2288                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2289
2290                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2291                 if (r)
2292                         goto out;
2293                 break;
2294         }
2295         default:
2296                 r = kvm_vm_ioctl(filp, ioctl, arg);
2297         }
2298
2299 out:
2300         return r;
2301 }
2302 #endif
2303
2304 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2305 {
2306         struct page *page[1];
2307         unsigned long addr;
2308         int npages;
2309         gfn_t gfn = vmf->pgoff;
2310         struct kvm *kvm = vma->vm_file->private_data;
2311
2312         addr = gfn_to_hva(kvm, gfn);
2313         if (kvm_is_error_hva(addr))
2314                 return VM_FAULT_SIGBUS;
2315
2316         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2317                                 NULL);
2318         if (unlikely(npages != 1))
2319                 return VM_FAULT_SIGBUS;
2320
2321         vmf->page = page[0];
2322         return 0;
2323 }
2324
2325 static const struct vm_operations_struct kvm_vm_vm_ops = {
2326         .fault = kvm_vm_fault,
2327 };
2328
2329 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2330 {
2331         vma->vm_ops = &kvm_vm_vm_ops;
2332         return 0;
2333 }
2334
2335 static struct file_operations kvm_vm_fops = {
2336         .release        = kvm_vm_release,
2337         .unlocked_ioctl = kvm_vm_ioctl,
2338 #ifdef CONFIG_COMPAT
2339         .compat_ioctl   = kvm_vm_compat_ioctl,
2340 #endif
2341         .mmap           = kvm_vm_mmap,
2342         .llseek         = noop_llseek,
2343 };
2344
2345 static int kvm_dev_ioctl_create_vm(unsigned long type)
2346 {
2347         int r;
2348         struct kvm *kvm;
2349
2350         kvm = kvm_create_vm(type);
2351         if (IS_ERR(kvm))
2352                 return PTR_ERR(kvm);
2353 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2354         r = kvm_coalesced_mmio_init(kvm);
2355         if (r < 0) {
2356                 kvm_put_kvm(kvm);
2357                 return r;
2358         }
2359 #endif
2360         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2361         if (r < 0)
2362                 kvm_put_kvm(kvm);
2363
2364         return r;
2365 }
2366
2367 static long kvm_dev_ioctl_check_extension_generic(long arg)
2368 {
2369         switch (arg) {
2370         case KVM_CAP_USER_MEMORY:
2371         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2372         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2373 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2374         case KVM_CAP_SET_BOOT_CPU_ID:
2375 #endif
2376         case KVM_CAP_INTERNAL_ERROR_DATA:
2377 #ifdef CONFIG_HAVE_KVM_MSI
2378         case KVM_CAP_SIGNAL_MSI:
2379 #endif
2380                 return 1;
2381 #ifdef KVM_CAP_IRQ_ROUTING
2382         case KVM_CAP_IRQ_ROUTING:
2383                 return KVM_MAX_IRQ_ROUTES;
2384 #endif
2385         default:
2386                 break;
2387         }
2388         return kvm_dev_ioctl_check_extension(arg);
2389 }
2390
2391 static long kvm_dev_ioctl(struct file *filp,
2392                           unsigned int ioctl, unsigned long arg)
2393 {
2394         long r = -EINVAL;
2395
2396         switch (ioctl) {
2397         case KVM_GET_API_VERSION:
2398                 r = -EINVAL;
2399                 if (arg)
2400                         goto out;
2401                 r = KVM_API_VERSION;
2402                 break;
2403         case KVM_CREATE_VM:
2404                 r = kvm_dev_ioctl_create_vm(arg);
2405                 break;
2406         case KVM_CHECK_EXTENSION:
2407                 r = kvm_dev_ioctl_check_extension_generic(arg);
2408                 break;
2409         case KVM_GET_VCPU_MMAP_SIZE:
2410                 r = -EINVAL;
2411                 if (arg)
2412                         goto out;
2413                 r = PAGE_SIZE;     /* struct kvm_run */
2414 #ifdef CONFIG_X86
2415                 r += PAGE_SIZE;    /* pio data page */
2416 #endif
2417 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2418                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2419 #endif
2420                 break;
2421         case KVM_TRACE_ENABLE:
2422         case KVM_TRACE_PAUSE:
2423         case KVM_TRACE_DISABLE:
2424                 r = -EOPNOTSUPP;
2425                 break;
2426         default:
2427                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2428         }
2429 out:
2430         return r;
2431 }
2432
2433 static struct file_operations kvm_chardev_ops = {
2434         .unlocked_ioctl = kvm_dev_ioctl,
2435         .compat_ioctl   = kvm_dev_ioctl,
2436         .llseek         = noop_llseek,
2437 };
2438
2439 static struct miscdevice kvm_dev = {
2440         KVM_MINOR,
2441         "kvm",
2442         &kvm_chardev_ops,
2443 };
2444
2445 static void hardware_enable_nolock(void *junk)
2446 {
2447         int cpu = raw_smp_processor_id();
2448         int r;
2449
2450         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2451                 return;
2452
2453         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2454
2455         r = kvm_arch_hardware_enable(NULL);
2456
2457         if (r) {
2458                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2459                 atomic_inc(&hardware_enable_failed);
2460                 printk(KERN_INFO "kvm: enabling virtualization on "
2461                                  "CPU%d failed\n", cpu);
2462         }
2463 }
2464
2465 static void hardware_enable(void *junk)
2466 {
2467         raw_spin_lock(&kvm_lock);
2468         hardware_enable_nolock(junk);
2469         raw_spin_unlock(&kvm_lock);
2470 }
2471
2472 static void hardware_disable_nolock(void *junk)
2473 {
2474         int cpu = raw_smp_processor_id();
2475
2476         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2477                 return;
2478         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2479         kvm_arch_hardware_disable(NULL);
2480 }
2481
2482 static void hardware_disable(void *junk)
2483 {
2484         raw_spin_lock(&kvm_lock);
2485         hardware_disable_nolock(junk);
2486         raw_spin_unlock(&kvm_lock);
2487 }
2488
2489 static void hardware_disable_all_nolock(void)
2490 {
2491         BUG_ON(!kvm_usage_count);
2492
2493         kvm_usage_count--;
2494         if (!kvm_usage_count)
2495                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2496 }
2497
2498 static void hardware_disable_all(void)
2499 {
2500         raw_spin_lock(&kvm_lock);
2501         hardware_disable_all_nolock();
2502         raw_spin_unlock(&kvm_lock);
2503 }
2504
2505 static int hardware_enable_all(void)
2506 {
2507         int r = 0;
2508
2509         raw_spin_lock(&kvm_lock);
2510
2511         kvm_usage_count++;
2512         if (kvm_usage_count == 1) {
2513                 atomic_set(&hardware_enable_failed, 0);
2514                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2515
2516                 if (atomic_read(&hardware_enable_failed)) {
2517                         hardware_disable_all_nolock();
2518                         r = -EBUSY;
2519                 }
2520         }
2521
2522         raw_spin_unlock(&kvm_lock);
2523
2524         return r;
2525 }
2526
2527 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2528                            void *v)
2529 {
2530         int cpu = (long)v;
2531
2532         if (!kvm_usage_count)
2533                 return NOTIFY_OK;
2534
2535         val &= ~CPU_TASKS_FROZEN;
2536         switch (val) {
2537         case CPU_DYING:
2538                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2539                        cpu);
2540                 hardware_disable(NULL);
2541                 break;
2542         case CPU_STARTING:
2543                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2544                        cpu);
2545                 hardware_enable(NULL);
2546                 break;
2547         }
2548         return NOTIFY_OK;
2549 }
2550
2551
2552 asmlinkage void kvm_spurious_fault(void)
2553 {
2554         /* Fault while not rebooting.  We want the trace. */
2555         BUG();
2556 }
2557 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2558
2559 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2560                       void *v)
2561 {
2562         /*
2563          * Some (well, at least mine) BIOSes hang on reboot if
2564          * in vmx root mode.
2565          *
2566          * And Intel TXT required VMX off for all cpu when system shutdown.
2567          */
2568         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2569         kvm_rebooting = true;
2570         on_each_cpu(hardware_disable_nolock, NULL, 1);
2571         return NOTIFY_OK;
2572 }
2573
2574 static struct notifier_block kvm_reboot_notifier = {
2575         .notifier_call = kvm_reboot,
2576         .priority = 0,
2577 };
2578
2579 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2580 {
2581         int i;
2582
2583         for (i = 0; i < bus->dev_count; i++) {
2584                 struct kvm_io_device *pos = bus->range[i].dev;
2585
2586                 kvm_iodevice_destructor(pos);
2587         }
2588         kfree(bus);
2589 }
2590
2591 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2592 {
2593         const struct kvm_io_range *r1 = p1;
2594         const struct kvm_io_range *r2 = p2;
2595
2596         if (r1->addr < r2->addr)
2597                 return -1;
2598         if (r1->addr + r1->len > r2->addr + r2->len)
2599                 return 1;
2600         return 0;
2601 }
2602
2603 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2604                           gpa_t addr, int len)
2605 {
2606         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2607                 .addr = addr,
2608                 .len = len,
2609                 .dev = dev,
2610         };
2611
2612         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2613                 kvm_io_bus_sort_cmp, NULL);
2614
2615         return 0;
2616 }
2617
2618 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2619                              gpa_t addr, int len)
2620 {
2621         struct kvm_io_range *range, key;
2622         int off;
2623
2624         key = (struct kvm_io_range) {
2625                 .addr = addr,
2626                 .len = len,
2627         };
2628
2629         range = bsearch(&key, bus->range, bus->dev_count,
2630                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2631         if (range == NULL)
2632                 return -ENOENT;
2633
2634         off = range - bus->range;
2635
2636         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2637                 off--;
2638
2639         return off;
2640 }
2641
2642 /* kvm_io_bus_write - called under kvm->slots_lock */
2643 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2644                      int len, const void *val)
2645 {
2646         int idx;
2647         struct kvm_io_bus *bus;
2648         struct kvm_io_range range;
2649
2650         range = (struct kvm_io_range) {
2651                 .addr = addr,
2652                 .len = len,
2653         };
2654
2655         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2656         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2657         if (idx < 0)
2658                 return -EOPNOTSUPP;
2659
2660         while (idx < bus->dev_count &&
2661                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2662                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2663                         return 0;
2664                 idx++;
2665         }
2666
2667         return -EOPNOTSUPP;
2668 }
2669
2670 /* kvm_io_bus_read - called under kvm->slots_lock */
2671 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2672                     int len, void *val)
2673 {
2674         int idx;
2675         struct kvm_io_bus *bus;
2676         struct kvm_io_range range;
2677
2678         range = (struct kvm_io_range) {
2679                 .addr = addr,
2680                 .len = len,
2681         };
2682
2683         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2684         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2685         if (idx < 0)
2686                 return -EOPNOTSUPP;
2687
2688         while (idx < bus->dev_count &&
2689                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2690                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2691                         return 0;
2692                 idx++;
2693         }
2694
2695         return -EOPNOTSUPP;
2696 }
2697
2698 /* Caller must hold slots_lock. */
2699 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2700                             int len, struct kvm_io_device *dev)
2701 {
2702         struct kvm_io_bus *new_bus, *bus;
2703
2704         bus = kvm->buses[bus_idx];
2705         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2706                 return -ENOSPC;
2707
2708         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2709                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2710         if (!new_bus)
2711                 return -ENOMEM;
2712         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2713                sizeof(struct kvm_io_range)));
2714         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2715         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2716         synchronize_srcu_expedited(&kvm->srcu);
2717         kfree(bus);
2718
2719         return 0;
2720 }
2721
2722 /* Caller must hold slots_lock. */
2723 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2724                               struct kvm_io_device *dev)
2725 {
2726         int i, r;
2727         struct kvm_io_bus *new_bus, *bus;
2728
2729         bus = kvm->buses[bus_idx];
2730         r = -ENOENT;
2731         for (i = 0; i < bus->dev_count; i++)
2732                 if (bus->range[i].dev == dev) {
2733                         r = 0;
2734                         break;
2735                 }
2736
2737         if (r)
2738                 return r;
2739
2740         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2741                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2742         if (!new_bus)
2743                 return -ENOMEM;
2744
2745         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2746         new_bus->dev_count--;
2747         memcpy(new_bus->range + i, bus->range + i + 1,
2748                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2749
2750         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2751         synchronize_srcu_expedited(&kvm->srcu);
2752         kfree(bus);
2753         return r;
2754 }
2755
2756 static struct notifier_block kvm_cpu_notifier = {
2757         .notifier_call = kvm_cpu_hotplug,
2758 };
2759
2760 static int vm_stat_get(void *_offset, u64 *val)
2761 {
2762         unsigned offset = (long)_offset;
2763         struct kvm *kvm;
2764
2765         *val = 0;
2766         raw_spin_lock(&kvm_lock);
2767         list_for_each_entry(kvm, &vm_list, vm_list)
2768                 *val += *(u32 *)((void *)kvm + offset);
2769         raw_spin_unlock(&kvm_lock);
2770         return 0;
2771 }
2772
2773 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2774
2775 static int vcpu_stat_get(void *_offset, u64 *val)
2776 {
2777         unsigned offset = (long)_offset;
2778         struct kvm *kvm;
2779         struct kvm_vcpu *vcpu;
2780         int i;
2781
2782         *val = 0;
2783         raw_spin_lock(&kvm_lock);
2784         list_for_each_entry(kvm, &vm_list, vm_list)
2785                 kvm_for_each_vcpu(i, vcpu, kvm)
2786                         *val += *(u32 *)((void *)vcpu + offset);
2787
2788         raw_spin_unlock(&kvm_lock);
2789         return 0;
2790 }
2791
2792 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2793
2794 static const struct file_operations *stat_fops[] = {
2795         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2796         [KVM_STAT_VM]   = &vm_stat_fops,
2797 };
2798
2799 static int kvm_init_debug(void)
2800 {
2801         int r = -EFAULT;
2802         struct kvm_stats_debugfs_item *p;
2803
2804         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2805         if (kvm_debugfs_dir == NULL)
2806                 goto out;
2807
2808         for (p = debugfs_entries; p->name; ++p) {
2809                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2810                                                 (void *)(long)p->offset,
2811                                                 stat_fops[p->kind]);
2812                 if (p->dentry == NULL)
2813                         goto out_dir;
2814         }
2815
2816         return 0;
2817
2818 out_dir:
2819         debugfs_remove_recursive(kvm_debugfs_dir);
2820 out:
2821         return r;
2822 }
2823
2824 static void kvm_exit_debug(void)
2825 {
2826         struct kvm_stats_debugfs_item *p;
2827
2828         for (p = debugfs_entries; p->name; ++p)
2829                 debugfs_remove(p->dentry);
2830         debugfs_remove(kvm_debugfs_dir);
2831 }
2832
2833 static int kvm_suspend(void)
2834 {
2835         if (kvm_usage_count)
2836                 hardware_disable_nolock(NULL);
2837         return 0;
2838 }
2839
2840 static void kvm_resume(void)
2841 {
2842         if (kvm_usage_count) {
2843                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2844                 hardware_enable_nolock(NULL);
2845         }
2846 }
2847
2848 static struct syscore_ops kvm_syscore_ops = {
2849         .suspend = kvm_suspend,
2850         .resume = kvm_resume,
2851 };
2852
2853 static inline
2854 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2855 {
2856         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2857 }
2858
2859 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2860 {
2861         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2862
2863         kvm_arch_vcpu_load(vcpu, cpu);
2864 }
2865
2866 static void kvm_sched_out(struct preempt_notifier *pn,
2867                           struct task_struct *next)
2868 {
2869         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2870
2871         kvm_arch_vcpu_put(vcpu);
2872 }
2873
2874 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2875                   struct module *module)
2876 {
2877         int r;
2878         int cpu;
2879
2880         r = kvm_arch_init(opaque);
2881         if (r)
2882                 goto out_fail;
2883
2884         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2885                 r = -ENOMEM;
2886                 goto out_free_0;
2887         }
2888
2889         r = kvm_arch_hardware_setup();
2890         if (r < 0)
2891                 goto out_free_0a;
2892
2893         for_each_online_cpu(cpu) {
2894                 smp_call_function_single(cpu,
2895                                 kvm_arch_check_processor_compat,
2896                                 &r, 1);
2897                 if (r < 0)
2898                         goto out_free_1;
2899         }
2900
2901         r = register_cpu_notifier(&kvm_cpu_notifier);
2902         if (r)
2903                 goto out_free_2;
2904         register_reboot_notifier(&kvm_reboot_notifier);
2905
2906         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2907         if (!vcpu_align)
2908                 vcpu_align = __alignof__(struct kvm_vcpu);
2909         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2910                                            0, NULL);
2911         if (!kvm_vcpu_cache) {
2912                 r = -ENOMEM;
2913                 goto out_free_3;
2914         }
2915
2916         r = kvm_async_pf_init();
2917         if (r)
2918                 goto out_free;
2919
2920         kvm_chardev_ops.owner = module;
2921         kvm_vm_fops.owner = module;
2922         kvm_vcpu_fops.owner = module;
2923
2924         r = misc_register(&kvm_dev);
2925         if (r) {
2926                 printk(KERN_ERR "kvm: misc device register failed\n");
2927                 goto out_unreg;
2928         }
2929
2930         register_syscore_ops(&kvm_syscore_ops);
2931
2932         kvm_preempt_ops.sched_in = kvm_sched_in;
2933         kvm_preempt_ops.sched_out = kvm_sched_out;
2934
2935         r = kvm_init_debug();
2936         if (r) {
2937                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2938                 goto out_undebugfs;
2939         }
2940
2941         return 0;
2942
2943 out_undebugfs:
2944         unregister_syscore_ops(&kvm_syscore_ops);
2945 out_unreg:
2946         kvm_async_pf_deinit();
2947 out_free:
2948         kmem_cache_destroy(kvm_vcpu_cache);
2949 out_free_3:
2950         unregister_reboot_notifier(&kvm_reboot_notifier);
2951         unregister_cpu_notifier(&kvm_cpu_notifier);
2952 out_free_2:
2953 out_free_1:
2954         kvm_arch_hardware_unsetup();
2955 out_free_0a:
2956         free_cpumask_var(cpus_hardware_enabled);
2957 out_free_0:
2958         kvm_arch_exit();
2959 out_fail:
2960         return r;
2961 }
2962 EXPORT_SYMBOL_GPL(kvm_init);
2963
2964 void kvm_exit(void)
2965 {
2966         kvm_exit_debug();
2967         misc_deregister(&kvm_dev);
2968         kmem_cache_destroy(kvm_vcpu_cache);
2969         kvm_async_pf_deinit();
2970         unregister_syscore_ops(&kvm_syscore_ops);
2971         unregister_reboot_notifier(&kvm_reboot_notifier);
2972         unregister_cpu_notifier(&kvm_cpu_notifier);
2973         on_each_cpu(hardware_disable_nolock, NULL, 1);
2974         kvm_arch_hardware_unsetup();
2975         kvm_arch_exit();
2976         free_cpumask_var(cpus_hardware_enabled);
2977 }
2978 EXPORT_SYMBOL_GPL(kvm_exit);