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