virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/syscore_ops.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/pgtable.h>
  57
  58 #include "coalesced_mmio.h"
  59 #include "async_pf.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/kvm.h>
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 /*
  68  * Ordering of locks:
  69  *
  70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  71  */
  72
  73 DEFINE_RAW_SPINLOCK(kvm_lock);
  74 LIST_HEAD(vm_list);
  75
  76 static cpumask_var_t cpus_hardware_enabled;
  77 static int kvm_usage_count = 0;
  78 static atomic_t hardware_enable_failed;
  79
  80 struct kmem_cache *kvm_vcpu_cache;
  81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  82
  83 static __read_mostly struct preempt_ops kvm_preempt_ops;
  84
  85 struct dentry *kvm_debugfs_dir;
  86
  87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  88                            unsigned long arg);
  89 #ifdef CONFIG_COMPAT
  90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  91                                   unsigned long arg);
  92 #endif
  93 static int hardware_enable_all(void);
  94 static void hardware_disable_all(void);
  95
  96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  97
  98 bool kvm_rebooting;
  99 EXPORT_SYMBOL_GPL(kvm_rebooting);
 100
 101 static bool largepages_enabled = true;
 102
 103 static struct page *hwpoison_page;
 104 static pfn_t hwpoison_pfn;
 105
 106 struct page *fault_page;
 107 pfn_t fault_pfn;
 108
 109 inline int kvm_is_mmio_pfn(pfn_t pfn)
 110 {
 111         if (pfn_valid(pfn)) {
 112                 int reserved;
 113                 struct page *tail = pfn_to_page(pfn);
 114                 struct page *head = compound_trans_head(tail);
 115                 reserved = PageReserved(head);
 116                 if (head != tail) {
 117                         /*
 118                          * "head" is not a dangling pointer
 119                          * (compound_trans_head takes care of that)
 120                          * but the hugepage may have been splitted
 121                          * from under us (and we may not hold a
 122                          * reference count on the head page so it can
 123                          * be reused before we run PageReferenced), so
 124                          * we've to check PageTail before returning
 125                          * what we just read.
 126                          */
 127                         smp_rmb();
 128                         if (PageTail(tail))
 129                                 return reserved;
 130                 }
 131                 return PageReserved(tail);
 132         }
 133
 134         return true;
 135 }
 136
 137 /*
 138  * Switches to specified vcpu, until a matching vcpu_put()
 139  */
 140 void vcpu_load(struct kvm_vcpu *vcpu)
 141 {
 142         int cpu;
 143
 144         mutex_lock(&vcpu->mutex);
 145         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 146                 /* The thread running this VCPU changed. */
 147                 struct pid *oldpid = vcpu->pid;
 148                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 149                 rcu_assign_pointer(vcpu->pid, newpid);
 150                 synchronize_rcu();
 151                 put_pid(oldpid);
 152         }
 153         cpu = get_cpu();
 154         preempt_notifier_register(&vcpu->preempt_notifier);
 155         kvm_arch_vcpu_load(vcpu, cpu);
 156         put_cpu();
 157 }
 158
 159 void vcpu_put(struct kvm_vcpu *vcpu)
 160 {
 161         preempt_disable();
 162         kvm_arch_vcpu_put(vcpu);
 163         preempt_notifier_unregister(&vcpu->preempt_notifier);
 164         preempt_enable();
 165         mutex_unlock(&vcpu->mutex);
 166 }
 167
 168 static void ack_flush(void *_completed)
 169 {
 170 }
 171
 172 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 173 {
 174         int i, cpu, me;
 175         cpumask_var_t cpus;
 176         bool called = true;
 177         struct kvm_vcpu *vcpu;
 178
 179         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 180
 181         me = get_cpu();
 182         kvm_for_each_vcpu(i, vcpu, kvm) {
 183                 kvm_make_request(req, vcpu);
 184                 cpu = vcpu->cpu;
 185
 186                 /* Set ->requests bit before we read ->mode */
 187                 smp_mb();
 188
 189                 if (cpus != NULL && cpu != -1 && cpu != me &&
 190                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 191                         cpumask_set_cpu(cpu, cpus);
 192         }
 193         if (unlikely(cpus == NULL))
 194                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 195         else if (!cpumask_empty(cpus))
 196                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 197         else
 198                 called = false;
 199         put_cpu();
 200         free_cpumask_var(cpus);
 201         return called;
 202 }
 203
 204 void kvm_flush_remote_tlbs(struct kvm *kvm)
 205 {
 206         long dirty_count = kvm->tlbs_dirty;
 207
 208         smp_mb();
 209         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 210                 ++kvm->stat.remote_tlb_flush;
 211         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 212 }
 213
 214 void kvm_reload_remote_mmus(struct kvm *kvm)
 215 {
 216         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 217 }
 218
 219 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 220 {
 221         struct page *page;
 222         int r;
 223
 224         mutex_init(&vcpu->mutex);
 225         vcpu->cpu = -1;
 226         vcpu->kvm = kvm;
 227         vcpu->vcpu_id = id;
 228         vcpu->pid = NULL;
 229         init_waitqueue_head(&vcpu->wq);
 230         kvm_async_pf_vcpu_init(vcpu);
 231
 232         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 233         if (!page) {
 234                 r = -ENOMEM;
 235                 goto fail;
 236         }
 237         vcpu->run = page_address(page);
 238
 239         r = kvm_arch_vcpu_init(vcpu);
 240         if (r < 0)
 241                 goto fail_free_run;
 242         return 0;
 243
 244 fail_free_run:
 245         free_page((unsigned long)vcpu->run);
 246 fail:
 247         return r;
 248 }
 249 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 250
 251 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 252 {
 253         put_pid(vcpu->pid);
 254         kvm_arch_vcpu_uninit(vcpu);
 255         free_page((unsigned long)vcpu->run);
 256 }
 257 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 258
 259 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 260 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 261 {
 262         return container_of(mn, struct kvm, mmu_notifier);
 263 }
 264
 265 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 266                                              struct mm_struct *mm,
 267                                              unsigned long address)
 268 {
 269         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 270         int need_tlb_flush, idx;
 271
 272         /*
 273          * When ->invalidate_page runs, the linux pte has been zapped
 274          * already but the page is still allocated until
 275          * ->invalidate_page returns. So if we increase the sequence
 276          * here the kvm page fault will notice if the spte can't be
 277          * established because the page is going to be freed. If
 278          * instead the kvm page fault establishes the spte before
 279          * ->invalidate_page runs, kvm_unmap_hva will release it
 280          * before returning.
 281          *
 282          * The sequence increase only need to be seen at spin_unlock
 283          * time, and not at spin_lock time.
 284          *
 285          * Increasing the sequence after the spin_unlock would be
 286          * unsafe because the kvm page fault could then establish the
 287          * pte after kvm_unmap_hva returned, without noticing the page
 288          * is going to be freed.
 289          */
 290         idx = srcu_read_lock(&kvm->srcu);
 291         spin_lock(&kvm->mmu_lock);
 292
 293         kvm->mmu_notifier_seq++;
 294         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 295         /* we've to flush the tlb before the pages can be freed */
 296         if (need_tlb_flush)
 297                 kvm_flush_remote_tlbs(kvm);
 298
 299         spin_unlock(&kvm->mmu_lock);
 300         srcu_read_unlock(&kvm->srcu, idx);
 301 }
 302
 303 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 304                                         struct mm_struct *mm,
 305                                         unsigned long address,
 306                                         pte_t pte)
 307 {
 308         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 309         int idx;
 310
 311         idx = srcu_read_lock(&kvm->srcu);
 312         spin_lock(&kvm->mmu_lock);
 313         kvm->mmu_notifier_seq++;
 314         kvm_set_spte_hva(kvm, address, pte);
 315         spin_unlock(&kvm->mmu_lock);
 316         srcu_read_unlock(&kvm->srcu, idx);
 317 }
 318
 319 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 320                                                     struct mm_struct *mm,
 321                                                     unsigned long start,
 322                                                     unsigned long end)
 323 {
 324         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 325         int need_tlb_flush = 0, idx;
 326
 327         idx = srcu_read_lock(&kvm->srcu);
 328         spin_lock(&kvm->mmu_lock);
 329         /*
 330          * The count increase must become visible at unlock time as no
 331          * spte can be established without taking the mmu_lock and
 332          * count is also read inside the mmu_lock critical section.
 333          */
 334         kvm->mmu_notifier_count++;
 335         for (; start < end; start += PAGE_SIZE)
 336                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 337         need_tlb_flush |= kvm->tlbs_dirty;
 338         /* we've to flush the tlb before the pages can be freed */
 339         if (need_tlb_flush)
 340                 kvm_flush_remote_tlbs(kvm);
 341
 342         spin_unlock(&kvm->mmu_lock);
 343         srcu_read_unlock(&kvm->srcu, idx);
 344 }
 345
 346 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 347                                                   struct mm_struct *mm,
 348                                                   unsigned long start,
 349                                                   unsigned long end)
 350 {
 351         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 352
 353         spin_lock(&kvm->mmu_lock);
 354         /*
 355          * This sequence increase will notify the kvm page fault that
 356          * the page that is going to be mapped in the spte could have
 357          * been freed.
 358          */
 359         kvm->mmu_notifier_seq++;
 360         smp_wmb();
 361         /*
 362          * The above sequence increase must be visible before the
 363          * below count decrease, which is ensured by the smp_wmb above
 364          * in conjunction with the smp_rmb in mmu_notifier_retry().
 365          */
 366         kvm->mmu_notifier_count--;
 367         spin_unlock(&kvm->mmu_lock);
 368
 369         BUG_ON(kvm->mmu_notifier_count < 0);
 370 }
 371
 372 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 373                                               struct mm_struct *mm,
 374                                               unsigned long address)
 375 {
 376         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 377         int young, idx;
 378
 379         idx = srcu_read_lock(&kvm->srcu);
 380         spin_lock(&kvm->mmu_lock);
 381
 382         young = kvm_age_hva(kvm, address);
 383         if (young)
 384                 kvm_flush_remote_tlbs(kvm);
 385
 386         spin_unlock(&kvm->mmu_lock);
 387         srcu_read_unlock(&kvm->srcu, idx);
 388
 389         return young;
 390 }
 391
 392 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 393                                        struct mm_struct *mm,
 394                                        unsigned long address)
 395 {
 396         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 397         int young, idx;
 398
 399         idx = srcu_read_lock(&kvm->srcu);
 400         spin_lock(&kvm->mmu_lock);
 401         young = kvm_test_age_hva(kvm, address);
 402         spin_unlock(&kvm->mmu_lock);
 403         srcu_read_unlock(&kvm->srcu, idx);
 404
 405         return young;
 406 }
 407
 408 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 409                                      struct mm_struct *mm)
 410 {
 411         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 412         int idx;
 413
 414         idx = srcu_read_lock(&kvm->srcu);
 415         kvm_arch_flush_shadow(kvm);
 416         srcu_read_unlock(&kvm->srcu, idx);
 417 }
 418
 419 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 420         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 421         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 422         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 423         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 424         .test_young             = kvm_mmu_notifier_test_young,
 425         .change_pte             = kvm_mmu_notifier_change_pte,
 426         .release                = kvm_mmu_notifier_release,
 427 };
 428
 429 static int kvm_init_mmu_notifier(struct kvm *kvm)
 430 {
 431         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 432         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 433 }
 434
 435 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 436
 437 static int kvm_init_mmu_notifier(struct kvm *kvm)
 438 {
 439         return 0;
 440 }
 441
 442 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 443
 444 static void kvm_init_memslots_id(struct kvm *kvm)
 445 {
 446         int i;
 447         struct kvm_memslots *slots = kvm->memslots;
 448
 449         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 450                 slots->id_to_index[i] = slots->memslots[i].id = i;
 451 }
 452
 453 static struct kvm *kvm_create_vm(unsigned long type)
 454 {
 455         int r, i;
 456         struct kvm *kvm = kvm_arch_alloc_vm();
 457
 458         if (!kvm)
 459                 return ERR_PTR(-ENOMEM);
 460
 461         r = kvm_arch_init_vm(kvm, type);
 462         if (r)
 463                 goto out_err_nodisable;
 464
 465         r = hardware_enable_all();
 466         if (r)
 467                 goto out_err_nodisable;
 468
 469 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 470         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 471         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 472 #endif
 473
 474         r = -ENOMEM;
 475         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 476         if (!kvm->memslots)
 477                 goto out_err_nosrcu;
 478         kvm_init_memslots_id(kvm);
 479         if (init_srcu_struct(&kvm->srcu))
 480                 goto out_err_nosrcu;
 481         for (i = 0; i < KVM_NR_BUSES; i++) {
 482                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 483                                         GFP_KERNEL);
 484                 if (!kvm->buses[i])
 485                         goto out_err;
 486         }
 487
 488         spin_lock_init(&kvm->mmu_lock);
 489         kvm->mm = current->mm;
 490         atomic_inc(&kvm->mm->mm_count);
 491         kvm_eventfd_init(kvm);
 492         mutex_init(&kvm->lock);
 493         mutex_init(&kvm->irq_lock);
 494         mutex_init(&kvm->slots_lock);
 495         atomic_set(&kvm->users_count, 1);
 496
 497         r = kvm_init_mmu_notifier(kvm);
 498         if (r)
 499                 goto out_err;
 500
 501         raw_spin_lock(&kvm_lock);
 502         list_add(&kvm->vm_list, &vm_list);
 503         raw_spin_unlock(&kvm_lock);
 504
 505         return kvm;
 506
 507 out_err:
 508         cleanup_srcu_struct(&kvm->srcu);
 509 out_err_nosrcu:
 510         hardware_disable_all();
 511 out_err_nodisable:
 512         for (i = 0; i < KVM_NR_BUSES; i++)
 513                 kfree(kvm->buses[i]);
 514         kfree(kvm->memslots);
 515         kvm_arch_free_vm(kvm);
 516         return ERR_PTR(r);
 517 }
 518
 519 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 520 {
 521         if (!memslot->dirty_bitmap)
 522                 return;
 523
 524         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
 525                 vfree(memslot->dirty_bitmap);
 526         else
 527                 kfree(memslot->dirty_bitmap);
 528
 529         memslot->dirty_bitmap = NULL;
 530 }
 531
 532 /*
 533  * Free any memory in @free but not in @dont.
 534  */
 535 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 536                                   struct kvm_memory_slot *dont)
 537 {
 538         if (!dont || free->rmap != dont->rmap)
 539                 vfree(free->rmap);
 540
 541         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 542                 kvm_destroy_dirty_bitmap(free);
 543
 544         kvm_arch_free_memslot(free, dont);
 545
 546         free->npages = 0;
 547         free->rmap = NULL;
 548 }
 549
 550 void kvm_free_physmem(struct kvm *kvm)
 551 {
 552         struct kvm_memslots *slots = kvm->memslots;
 553         struct kvm_memory_slot *memslot;
 554
 555         kvm_for_each_memslot(memslot, slots)
 556                 kvm_free_physmem_slot(memslot, NULL);
 557
 558         kfree(kvm->memslots);
 559 }
 560
 561 static void kvm_destroy_vm(struct kvm *kvm)
 562 {
 563         int i;
 564         struct mm_struct *mm = kvm->mm;
 565
 566         kvm_arch_sync_events(kvm);
 567         raw_spin_lock(&kvm_lock);
 568         list_del(&kvm->vm_list);
 569         raw_spin_unlock(&kvm_lock);
 570         kvm_free_irq_routing(kvm);
 571         for (i = 0; i < KVM_NR_BUSES; i++)
 572                 kvm_io_bus_destroy(kvm->buses[i]);
 573         kvm_coalesced_mmio_free(kvm);
 574 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 575         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 576 #else
 577         kvm_arch_flush_shadow(kvm);
 578 #endif
 579         kvm_arch_destroy_vm(kvm);
 580         kvm_free_physmem(kvm);
 581         cleanup_srcu_struct(&kvm->srcu);
 582         kvm_arch_free_vm(kvm);
 583         hardware_disable_all();
 584         mmdrop(mm);
 585 }
 586
 587 void kvm_get_kvm(struct kvm *kvm)
 588 {
 589         atomic_inc(&kvm->users_count);
 590 }
 591 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 592
 593 void kvm_put_kvm(struct kvm *kvm)
 594 {
 595         if (atomic_dec_and_test(&kvm->users_count))
 596                 kvm_destroy_vm(kvm);
 597 }
 598 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 599
 600
 601 static int kvm_vm_release(struct inode *inode, struct file *filp)
 602 {
 603         struct kvm *kvm = filp->private_data;
 604
 605         kvm_irqfd_release(kvm);
 606
 607         kvm_put_kvm(kvm);
 608         return 0;
 609 }
 610
 611 /*
 612  * Allocation size is twice as large as the actual dirty bitmap size.
 613  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 614  */
 615 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 616 {
 617 #ifndef CONFIG_S390
 618         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 619
 620         if (dirty_bytes > PAGE_SIZE)
 621                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
 622         else
 623                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
 624
 625         if (!memslot->dirty_bitmap)
 626                 return -ENOMEM;
 627
 628 #endif /* !CONFIG_S390 */
 629         return 0;
 630 }
 631
 632 static int cmp_memslot(const void *slot1, const void *slot2)
 633 {
 634         struct kvm_memory_slot *s1, *s2;
 635
 636         s1 = (struct kvm_memory_slot *)slot1;
 637         s2 = (struct kvm_memory_slot *)slot2;
 638
 639         if (s1->npages < s2->npages)
 640                 return 1;
 641         if (s1->npages > s2->npages)
 642                 return -1;
 643
 644         return 0;
 645 }
 646
 647 /*
 648  * Sort the memslots base on its size, so the larger slots
 649  * will get better fit.
 650  */
 651 static void sort_memslots(struct kvm_memslots *slots)
 652 {
 653         int i;
 654
 655         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 656               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 657
 658         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 659                 slots->id_to_index[slots->memslots[i].id] = i;
 660 }
 661
 662 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
 663 {
 664         if (new) {
 665                 int id = new->id;
 666                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 667                 unsigned long npages = old->npages;
 668
 669                 *old = *new;
 670                 if (new->npages != npages)
 671                         sort_memslots(slots);
 672         }
 673
 674         slots->generation++;
 675 }
 676
 677 /*
 678  * Allocate some memory and give it an address in the guest physical address
 679  * space.
 680  *
 681  * Discontiguous memory is allowed, mostly for framebuffers.
 682  *
 683  * Must be called holding mmap_sem for write.
 684  */
 685 int __kvm_set_memory_region(struct kvm *kvm,
 686                             struct kvm_userspace_memory_region *mem,
 687                             int user_alloc)
 688 {
 689         int r;
 690         gfn_t base_gfn;
 691         unsigned long npages;
 692         unsigned long i;
 693         struct kvm_memory_slot *memslot;
 694         struct kvm_memory_slot old, new;
 695         struct kvm_memslots *slots, *old_memslots;
 696
 697         r = -EINVAL;
 698         /* General sanity checks */
 699         if (mem->memory_size & (PAGE_SIZE - 1))
 700                 goto out;
 701         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 702                 goto out;
 703         /* We can read the guest memory with __xxx_user() later on. */
 704         if (user_alloc &&
 705             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 706              !access_ok(VERIFY_WRITE,
 707                         (void __user *)(unsigned long)mem->userspace_addr,
 708                         mem->memory_size)))
 709                 goto out;
 710         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 711                 goto out;
 712         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 713                 goto out;
 714
 715         memslot = id_to_memslot(kvm->memslots, mem->slot);
 716         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 717         npages = mem->memory_size >> PAGE_SHIFT;
 718
 719         r = -EINVAL;
 720         if (npages > KVM_MEM_MAX_NR_PAGES)
 721                 goto out;
 722
 723         if (!npages)
 724                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 725
 726         new = old = *memslot;
 727
 728         new.id = mem->slot;
 729         new.base_gfn = base_gfn;
 730         new.npages = npages;
 731         new.flags = mem->flags;
 732
 733         /* Disallow changing a memory slot's size. */
 734         r = -EINVAL;
 735         if (npages && old.npages && npages != old.npages)
 736                 goto out_free;
 737
 738         /* Check for overlaps */
 739         r = -EEXIST;
 740         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 741                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 742
 743                 if (s == memslot || !s->npages)
 744                         continue;
 745                 if (!((base_gfn + npages <= s->base_gfn) ||
 746                       (base_gfn >= s->base_gfn + s->npages)))
 747                         goto out_free;
 748         }
 749
 750         /* Free page dirty bitmap if unneeded */
 751         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 752                 new.dirty_bitmap = NULL;
 753
 754         r = -ENOMEM;
 755
 756         /* Allocate if a slot is being created */
 757         if (npages && !old.npages) {
 758                 new.user_alloc = user_alloc;
 759                 new.userspace_addr = mem->userspace_addr;
 760 #ifndef CONFIG_S390
 761                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
 762                 if (!new.rmap)
 763                         goto out_free;
 764 #endif /* not defined CONFIG_S390 */
 765                 if (kvm_arch_create_memslot(&new, npages))
 766                         goto out_free;
 767         }
 768
 769         /* Allocate page dirty bitmap if needed */
 770         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 771                 if (kvm_create_dirty_bitmap(&new) < 0)
 772                         goto out_free;
 773                 /* destroy any largepage mappings for dirty tracking */
 774         }
 775
 776         if (!npages) {
 777                 struct kvm_memory_slot *slot;
 778
 779                 r = -ENOMEM;
 780                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 781                                 GFP_KERNEL);
 782                 if (!slots)
 783                         goto out_free;
 784                 slot = id_to_memslot(slots, mem->slot);
 785                 slot->flags |= KVM_MEMSLOT_INVALID;
 786
 787                 update_memslots(slots, NULL);
 788
 789                 old_memslots = kvm->memslots;
 790                 rcu_assign_pointer(kvm->memslots, slots);
 791                 synchronize_srcu_expedited(&kvm->srcu);
 792                 /* From this point no new shadow pages pointing to a deleted
 793                  * memslot will be created.
 794                  *
 795                  * validation of sp->gfn happens in:
 796                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 797                  *      - kvm_is_visible_gfn (mmu_check_roots)
 798                  */
 799                 kvm_arch_flush_shadow(kvm);
 800                 kfree(old_memslots);
 801         }
 802
 803         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 804         if (r)
 805                 goto out_free;
 806
 807         /* map/unmap the pages in iommu page table */
 808         if (npages) {
 809                 r = kvm_iommu_map_pages(kvm, &new);
 810                 if (r)
 811                         goto out_free;
 812         } else
 813                 kvm_iommu_unmap_pages(kvm, &old);
 814
 815         r = -ENOMEM;
 816         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 817                         GFP_KERNEL);
 818         if (!slots)
 819                 goto out_free;
 820
 821         /* actual memory is freed via old in kvm_free_physmem_slot below */
 822         if (!npages) {
 823                 new.rmap = NULL;
 824                 new.dirty_bitmap = NULL;
 825                 memset(&new.arch, 0, sizeof(new.arch));
 826         }
 827
 828         update_memslots(slots, &new);
 829         old_memslots = kvm->memslots;
 830         rcu_assign_pointer(kvm->memslots, slots);
 831         synchronize_srcu_expedited(&kvm->srcu);
 832
 833         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 834
 835         /*
 836          * If the new memory slot is created, we need to clear all
 837          * mmio sptes.
 838          */
 839         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 840                 kvm_arch_flush_shadow(kvm);
 841
 842         kvm_free_physmem_slot(&old, &new);
 843         kfree(old_memslots);
 844
 845         return 0;
 846
 847 out_free:
 848         kvm_free_physmem_slot(&new, &old);
 849 out:
 850         return r;
 851
 852 }
 853 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 854
 855 int kvm_set_memory_region(struct kvm *kvm,
 856                           struct kvm_userspace_memory_region *mem,
 857                           int user_alloc)
 858 {
 859         int r;
 860
 861         mutex_lock(&kvm->slots_lock);
 862         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 863         mutex_unlock(&kvm->slots_lock);
 864         return r;
 865 }
 866 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 867
 868 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 869                                    struct
 870                                    kvm_userspace_memory_region *mem,
 871                                    int user_alloc)
 872 {
 873         if (mem->slot >= KVM_MEMORY_SLOTS)
 874                 return -EINVAL;
 875         return kvm_set_memory_region(kvm, mem, user_alloc);
 876 }
 877
 878 int kvm_get_dirty_log(struct kvm *kvm,
 879                         struct kvm_dirty_log *log, int *is_dirty)
 880 {
 881         struct kvm_memory_slot *memslot;
 882         int r, i;
 883         unsigned long n;
 884         unsigned long any = 0;
 885
 886         r = -EINVAL;
 887         if (log->slot >= KVM_MEMORY_SLOTS)
 888                 goto out;
 889
 890         memslot = id_to_memslot(kvm->memslots, log->slot);
 891         r = -ENOENT;
 892         if (!memslot->dirty_bitmap)
 893                 goto out;
 894
 895         n = kvm_dirty_bitmap_bytes(memslot);
 896
 897         for (i = 0; !any && i < n/sizeof(long); ++i)
 898                 any = memslot->dirty_bitmap[i];
 899
 900         r = -EFAULT;
 901         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 902                 goto out;
 903
 904         if (any)
 905                 *is_dirty = 1;
 906
 907         r = 0;
 908 out:
 909         return r;
 910 }
 911
 912 bool kvm_largepages_enabled(void)
 913 {
 914         return largepages_enabled;
 915 }
 916
 917 void kvm_disable_largepages(void)
 918 {
 919         largepages_enabled = false;
 920 }
 921 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 922
 923 int is_error_page(struct page *page)
 924 {
 925         return page == bad_page || page == hwpoison_page || page == fault_page;
 926 }
 927 EXPORT_SYMBOL_GPL(is_error_page);
 928
 929 int is_error_pfn(pfn_t pfn)
 930 {
 931         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 932 }
 933 EXPORT_SYMBOL_GPL(is_error_pfn);
 934
 935 int is_hwpoison_pfn(pfn_t pfn)
 936 {
 937         return pfn == hwpoison_pfn;
 938 }
 939 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 940
 941 int is_fault_pfn(pfn_t pfn)
 942 {
 943         return pfn == fault_pfn;
 944 }
 945 EXPORT_SYMBOL_GPL(is_fault_pfn);
 946
 947 int is_noslot_pfn(pfn_t pfn)
 948 {
 949         return pfn == bad_pfn;
 950 }
 951 EXPORT_SYMBOL_GPL(is_noslot_pfn);
 952
 953 int is_invalid_pfn(pfn_t pfn)
 954 {
 955         return pfn == hwpoison_pfn || pfn == fault_pfn;
 956 }
 957 EXPORT_SYMBOL_GPL(is_invalid_pfn);
 958
 959 static inline unsigned long bad_hva(void)
 960 {
 961         return PAGE_OFFSET;
 962 }
 963
 964 int kvm_is_error_hva(unsigned long addr)
 965 {
 966         return addr == bad_hva();
 967 }
 968 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 969
 970 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 971 {
 972         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 973 }
 974 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 975
 976 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 977 {
 978         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 979
 980         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
 981               memslot->flags & KVM_MEMSLOT_INVALID)
 982                 return 0;
 983
 984         return 1;
 985 }
 986 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 987
 988 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 989 {
 990         struct vm_area_struct *vma;
 991         unsigned long addr, size;
 992
 993         size = PAGE_SIZE;
 994
 995         addr = gfn_to_hva(kvm, gfn);
 996         if (kvm_is_error_hva(addr))
 997                 return PAGE_SIZE;
 998
 999         down_read(&current->mm->mmap_sem);
1000         vma = find_vma(current->mm, addr);
1001         if (!vma)
1002                 goto out;
1003
1004         size = vma_kernel_pagesize(vma);
1005
1006 out:
1007         up_read(&current->mm->mmap_sem);
1008
1009         return size;
1010 }
1011
1012 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1013                                      gfn_t *nr_pages)
1014 {
1015         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1016                 return bad_hva();
1017
1018         if (nr_pages)
1019                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1020
1021         return gfn_to_hva_memslot(slot, gfn);
1022 }
1023
1024 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1025 {
1026         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1027 }
1028 EXPORT_SYMBOL_GPL(gfn_to_hva);
1029
1030 static pfn_t get_fault_pfn(void)
1031 {
1032         get_page(fault_page);
1033         return fault_pfn;
1034 }
1035
1036 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1037         unsigned long start, int write, struct page **page)
1038 {
1039         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1040
1041         if (write)
1042                 flags |= FOLL_WRITE;
1043
1044         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1045 }
1046
1047 static inline int check_user_page_hwpoison(unsigned long addr)
1048 {
1049         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1050
1051         rc = __get_user_pages(current, current->mm, addr, 1,
1052                               flags, NULL, NULL, NULL);
1053         return rc == -EHWPOISON;
1054 }
1055
1056 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1057                         bool *async, bool write_fault, bool *writable)
1058 {
1059         struct page *page[1];
1060         int npages = 0;
1061         pfn_t pfn;
1062
1063         /* we can do it either atomically or asynchronously, not both */
1064         BUG_ON(atomic && async);
1065
1066         BUG_ON(!write_fault && !writable);
1067
1068         if (writable)
1069                 *writable = true;
1070
1071         if (atomic || async)
1072                 npages = __get_user_pages_fast(addr, 1, 1, page);
1073
1074         if (unlikely(npages != 1) && !atomic) {
1075                 might_sleep();
1076
1077                 if (writable)
1078                         *writable = write_fault;
1079
1080                 if (async) {
1081                         down_read(&current->mm->mmap_sem);
1082                         npages = get_user_page_nowait(current, current->mm,
1083                                                      addr, write_fault, page);
1084                         up_read(&current->mm->mmap_sem);
1085                 } else
1086                         npages = get_user_pages_fast(addr, 1, write_fault,
1087                                                      page);
1088
1089                 /* map read fault as writable if possible */
1090                 if (unlikely(!write_fault) && npages == 1) {
1091                         struct page *wpage[1];
1092
1093                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1094                         if (npages == 1) {
1095                                 *writable = true;
1096                                 put_page(page[0]);
1097                                 page[0] = wpage[0];
1098                         }
1099                         npages = 1;
1100                 }
1101         }
1102
1103         if (unlikely(npages != 1)) {
1104                 struct vm_area_struct *vma;
1105
1106                 if (atomic)
1107                         return get_fault_pfn();
1108
1109                 down_read(&current->mm->mmap_sem);
1110                 if (npages == -EHWPOISON ||
1111                         (!async && check_user_page_hwpoison(addr))) {
1112                         up_read(&current->mm->mmap_sem);
1113                         get_page(hwpoison_page);
1114                         return page_to_pfn(hwpoison_page);
1115                 }
1116
1117                 vma = find_vma_intersection(current->mm, addr, addr+1);
1118
1119                 if (vma == NULL)
1120                         pfn = get_fault_pfn();
1121                 else if ((vma->vm_flags & VM_PFNMAP)) {
1122                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1123                                 vma->vm_pgoff;
1124                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1125                 } else {
1126                         if (async && (vma->vm_flags & VM_WRITE))
1127                                 *async = true;
1128                         pfn = get_fault_pfn();
1129                 }
1130                 up_read(&current->mm->mmap_sem);
1131         } else
1132                 pfn = page_to_pfn(page[0]);
1133
1134         return pfn;
1135 }
1136
1137 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1138 {
1139         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1140 }
1141 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1142
1143 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1144                           bool write_fault, bool *writable)
1145 {
1146         unsigned long addr;
1147
1148         if (async)
1149                 *async = false;
1150
1151         addr = gfn_to_hva(kvm, gfn);
1152         if (kvm_is_error_hva(addr)) {
1153                 get_page(bad_page);
1154                 return page_to_pfn(bad_page);
1155         }
1156
1157         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1158 }
1159
1160 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1161 {
1162         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1163 }
1164 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1165
1166 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1167                        bool write_fault, bool *writable)
1168 {
1169         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1170 }
1171 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1172
1173 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1174 {
1175         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1176 }
1177 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1178
1179 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1180                       bool *writable)
1181 {
1182         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1183 }
1184 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1185
1186 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1187                          struct kvm_memory_slot *slot, gfn_t gfn)
1188 {
1189         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1190         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1191 }
1192
1193 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1194                                                                   int nr_pages)
1195 {
1196         unsigned long addr;
1197         gfn_t entry;
1198
1199         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1200         if (kvm_is_error_hva(addr))
1201                 return -1;
1202
1203         if (entry < nr_pages)
1204                 return 0;
1205
1206         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1207 }
1208 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1209
1210 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1211 {
1212         pfn_t pfn;
1213
1214         pfn = gfn_to_pfn(kvm, gfn);
1215         if (!kvm_is_mmio_pfn(pfn))
1216                 return pfn_to_page(pfn);
1217
1218         WARN_ON(kvm_is_mmio_pfn(pfn));
1219
1220         get_page(bad_page);
1221         return bad_page;
1222 }
1223
1224 EXPORT_SYMBOL_GPL(gfn_to_page);
1225
1226 void kvm_release_page_clean(struct page *page)
1227 {
1228         kvm_release_pfn_clean(page_to_pfn(page));
1229 }
1230 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1231
1232 void kvm_release_pfn_clean(pfn_t pfn)
1233 {
1234         if (!kvm_is_mmio_pfn(pfn))
1235                 put_page(pfn_to_page(pfn));
1236 }
1237 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1238
1239 void kvm_release_page_dirty(struct page *page)
1240 {
1241         kvm_release_pfn_dirty(page_to_pfn(page));
1242 }
1243 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1244
1245 void kvm_release_pfn_dirty(pfn_t pfn)
1246 {
1247         kvm_set_pfn_dirty(pfn);
1248         kvm_release_pfn_clean(pfn);
1249 }
1250 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1251
1252 void kvm_set_page_dirty(struct page *page)
1253 {
1254         kvm_set_pfn_dirty(page_to_pfn(page));
1255 }
1256 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1257
1258 void kvm_set_pfn_dirty(pfn_t pfn)
1259 {
1260         if (!kvm_is_mmio_pfn(pfn)) {
1261                 struct page *page = pfn_to_page(pfn);
1262                 if (!PageReserved(page))
1263                         SetPageDirty(page);
1264         }
1265 }
1266 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1267
1268 void kvm_set_pfn_accessed(pfn_t pfn)
1269 {
1270         if (!kvm_is_mmio_pfn(pfn))
1271                 mark_page_accessed(pfn_to_page(pfn));
1272 }
1273 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1274
1275 void kvm_get_pfn(pfn_t pfn)
1276 {
1277         if (!kvm_is_mmio_pfn(pfn))
1278                 get_page(pfn_to_page(pfn));
1279 }
1280 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1281
1282 static int next_segment(unsigned long len, int offset)
1283 {
1284         if (len > PAGE_SIZE - offset)
1285                 return PAGE_SIZE - offset;
1286         else
1287                 return len;
1288 }
1289
1290 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1291                         int len)
1292 {
1293         int r;
1294         unsigned long addr;
1295
1296         addr = gfn_to_hva(kvm, gfn);
1297         if (kvm_is_error_hva(addr))
1298                 return -EFAULT;
1299         r = __copy_from_user(data, (void __user *)addr + offset, len);
1300         if (r)
1301                 return -EFAULT;
1302         return 0;
1303 }
1304 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1305
1306 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1307 {
1308         gfn_t gfn = gpa >> PAGE_SHIFT;
1309         int seg;
1310         int offset = offset_in_page(gpa);
1311         int ret;
1312
1313         while ((seg = next_segment(len, offset)) != 0) {
1314                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1315                 if (ret < 0)
1316                         return ret;
1317                 offset = 0;
1318                 len -= seg;
1319                 data += seg;
1320                 ++gfn;
1321         }
1322         return 0;
1323 }
1324 EXPORT_SYMBOL_GPL(kvm_read_guest);
1325
1326 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1327                           unsigned long len)
1328 {
1329         int r;
1330         unsigned long addr;
1331         gfn_t gfn = gpa >> PAGE_SHIFT;
1332         int offset = offset_in_page(gpa);
1333
1334         addr = gfn_to_hva(kvm, gfn);
1335         if (kvm_is_error_hva(addr))
1336                 return -EFAULT;
1337         pagefault_disable();
1338         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1339         pagefault_enable();
1340         if (r)
1341                 return -EFAULT;
1342         return 0;
1343 }
1344 EXPORT_SYMBOL(kvm_read_guest_atomic);
1345
1346 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1347                          int offset, int len)
1348 {
1349         int r;
1350         unsigned long addr;
1351
1352         addr = gfn_to_hva(kvm, gfn);
1353         if (kvm_is_error_hva(addr))
1354                 return -EFAULT;
1355         r = __copy_to_user((void __user *)addr + offset, data, len);
1356         if (r)
1357                 return -EFAULT;
1358         mark_page_dirty(kvm, gfn);
1359         return 0;
1360 }
1361 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1362
1363 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1364                     unsigned long len)
1365 {
1366         gfn_t gfn = gpa >> PAGE_SHIFT;
1367         int seg;
1368         int offset = offset_in_page(gpa);
1369         int ret;
1370
1371         while ((seg = next_segment(len, offset)) != 0) {
1372                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1373                 if (ret < 0)
1374                         return ret;
1375                 offset = 0;
1376                 len -= seg;
1377                 data += seg;
1378                 ++gfn;
1379         }
1380         return 0;
1381 }
1382
1383 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1384                               gpa_t gpa)
1385 {
1386         struct kvm_memslots *slots = kvm_memslots(kvm);
1387         int offset = offset_in_page(gpa);
1388         gfn_t gfn = gpa >> PAGE_SHIFT;
1389
1390         ghc->gpa = gpa;
1391         ghc->generation = slots->generation;
1392         ghc->memslot = gfn_to_memslot(kvm, gfn);
1393         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1394         if (!kvm_is_error_hva(ghc->hva))
1395                 ghc->hva += offset;
1396         else
1397                 return -EFAULT;
1398
1399         return 0;
1400 }
1401 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1402
1403 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1404                            void *data, unsigned long len)
1405 {
1406         struct kvm_memslots *slots = kvm_memslots(kvm);
1407         int r;
1408
1409         if (slots->generation != ghc->generation)
1410                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1411
1412         if (kvm_is_error_hva(ghc->hva))
1413                 return -EFAULT;
1414
1415         r = __copy_to_user((void __user *)ghc->hva, data, len);
1416         if (r)
1417                 return -EFAULT;
1418         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1419
1420         return 0;
1421 }
1422 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1423
1424 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1425                            void *data, unsigned long len)
1426 {
1427         struct kvm_memslots *slots = kvm_memslots(kvm);
1428         int r;
1429
1430         if (slots->generation != ghc->generation)
1431                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1432
1433         if (kvm_is_error_hva(ghc->hva))
1434                 return -EFAULT;
1435
1436         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1437         if (r)
1438                 return -EFAULT;
1439
1440         return 0;
1441 }
1442 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1443
1444 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1445 {
1446         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1447                                     offset, len);
1448 }
1449 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1450
1451 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1452 {
1453         gfn_t gfn = gpa >> PAGE_SHIFT;
1454         int seg;
1455         int offset = offset_in_page(gpa);
1456         int ret;
1457
1458         while ((seg = next_segment(len, offset)) != 0) {
1459                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1460                 if (ret < 0)
1461                         return ret;
1462                 offset = 0;
1463                 len -= seg;
1464                 ++gfn;
1465         }
1466         return 0;
1467 }
1468 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1469
1470 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1471                              gfn_t gfn)
1472 {
1473         if (memslot && memslot->dirty_bitmap) {
1474                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1475
1476                 /* TODO: introduce set_bit_le() and use it */
1477                 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1478         }
1479 }
1480
1481 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1482 {
1483         struct kvm_memory_slot *memslot;
1484
1485         memslot = gfn_to_memslot(kvm, gfn);
1486         mark_page_dirty_in_slot(kvm, memslot, gfn);
1487 }
1488
1489 /*
1490  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1491  */
1492 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1493 {
1494         DEFINE_WAIT(wait);
1495
1496         for (;;) {
1497                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1498
1499                 if (kvm_arch_vcpu_runnable(vcpu)) {
1500                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1501                         break;
1502                 }
1503                 if (kvm_cpu_has_pending_timer(vcpu))
1504                         break;
1505                 if (signal_pending(current))
1506                         break;
1507
1508                 schedule();
1509         }
1510
1511         finish_wait(&vcpu->wq, &wait);
1512 }
1513
1514 #ifndef CONFIG_S390
1515 /*
1516  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1517  */
1518 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1519 {
1520         int me;
1521         int cpu = vcpu->cpu;
1522         wait_queue_head_t *wqp;
1523
1524         wqp = kvm_arch_vcpu_wq(vcpu);
1525         if (waitqueue_active(wqp)) {
1526                 wake_up_interruptible(wqp);
1527                 ++vcpu->stat.halt_wakeup;
1528         }
1529
1530         me = get_cpu();
1531         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1532                 if (kvm_arch_vcpu_should_kick(vcpu))
1533                         smp_send_reschedule(cpu);
1534         put_cpu();
1535 }
1536 #endif /* !CONFIG_S390 */
1537
1538 void kvm_resched(struct kvm_vcpu *vcpu)
1539 {
1540         if (!need_resched())
1541                 return;
1542         cond_resched();
1543 }
1544 EXPORT_SYMBOL_GPL(kvm_resched);
1545
1546 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1547 {
1548         struct pid *pid;
1549         struct task_struct *task = NULL;
1550
1551         rcu_read_lock();
1552         pid = rcu_dereference(target->pid);
1553         if (pid)
1554                 task = get_pid_task(target->pid, PIDTYPE_PID);
1555         rcu_read_unlock();
1556         if (!task)
1557                 return false;
1558         if (task->flags & PF_VCPU) {
1559                 put_task_struct(task);
1560                 return false;
1561         }
1562         if (yield_to(task, 1)) {
1563                 put_task_struct(task);
1564                 return true;
1565         }
1566         put_task_struct(task);
1567         return false;
1568 }
1569 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1570
1571 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1572 {
1573         struct kvm *kvm = me->kvm;
1574         struct kvm_vcpu *vcpu;
1575         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1576         int yielded = 0;
1577         int pass;
1578         int i;
1579
1580         /*
1581          * We boost the priority of a VCPU that is runnable but not
1582          * currently running, because it got preempted by something
1583          * else and called schedule in __vcpu_run.  Hopefully that
1584          * VCPU is holding the lock that we need and will release it.
1585          * We approximate round-robin by starting at the last boosted VCPU.
1586          */
1587         for (pass = 0; pass < 2 && !yielded; pass++) {
1588                 kvm_for_each_vcpu(i, vcpu, kvm) {
1589                         if (!pass && i < last_boosted_vcpu) {
1590                                 i = last_boosted_vcpu;
1591                                 continue;
1592                         } else if (pass && i > last_boosted_vcpu)
1593                                 break;
1594                         if (vcpu == me)
1595                                 continue;
1596                         if (waitqueue_active(&vcpu->wq))
1597                                 continue;
1598                         if (kvm_vcpu_yield_to(vcpu)) {
1599                                 kvm->last_boosted_vcpu = i;
1600                                 yielded = 1;
1601                                 break;
1602                         }
1603                 }
1604         }
1605 }
1606 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1607
1608 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1609 {
1610         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1611         struct page *page;
1612
1613         if (vmf->pgoff == 0)
1614                 page = virt_to_page(vcpu->run);
1615 #ifdef CONFIG_X86
1616         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1617                 page = virt_to_page(vcpu->arch.pio_data);
1618 #endif
1619 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1620         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1621                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1622 #endif
1623         else
1624                 return kvm_arch_vcpu_fault(vcpu, vmf);
1625         get_page(page);
1626         vmf->page = page;
1627         return 0;
1628 }
1629
1630 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1631         .fault = kvm_vcpu_fault,
1632 };
1633
1634 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1635 {
1636         vma->vm_ops = &kvm_vcpu_vm_ops;
1637         return 0;
1638 }
1639
1640 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1641 {
1642         struct kvm_vcpu *vcpu = filp->private_data;
1643
1644         kvm_put_kvm(vcpu->kvm);
1645         return 0;
1646 }
1647
1648 static struct file_operations kvm_vcpu_fops = {
1649         .release        = kvm_vcpu_release,
1650         .unlocked_ioctl = kvm_vcpu_ioctl,
1651 #ifdef CONFIG_COMPAT
1652         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1653 #endif
1654         .mmap           = kvm_vcpu_mmap,
1655         .llseek         = noop_llseek,
1656 };
1657
1658 /*
1659  * Allocates an inode for the vcpu.
1660  */
1661 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1662 {
1663         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1664 }
1665
1666 /*
1667  * Creates some virtual cpus.  Good luck creating more than one.
1668  */
1669 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1670 {
1671         int r;
1672         struct kvm_vcpu *vcpu, *v;
1673
1674         vcpu = kvm_arch_vcpu_create(kvm, id);
1675         if (IS_ERR(vcpu))
1676                 return PTR_ERR(vcpu);
1677
1678         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1679
1680         r = kvm_arch_vcpu_setup(vcpu);
1681         if (r)
1682                 goto vcpu_destroy;
1683
1684         mutex_lock(&kvm->lock);
1685         if (!kvm_vcpu_compatible(vcpu)) {
1686                 r = -EINVAL;
1687                 goto unlock_vcpu_destroy;
1688         }
1689         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1690                 r = -EINVAL;
1691                 goto unlock_vcpu_destroy;
1692         }
1693
1694         kvm_for_each_vcpu(r, v, kvm)
1695                 if (v->vcpu_id == id) {
1696                         r = -EEXIST;
1697                         goto unlock_vcpu_destroy;
1698                 }
1699
1700         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1701
1702         /* Now it's all set up, let userspace reach it */
1703         kvm_get_kvm(kvm);
1704         r = create_vcpu_fd(vcpu);
1705         if (r < 0) {
1706                 kvm_put_kvm(kvm);
1707                 goto unlock_vcpu_destroy;
1708         }
1709
1710         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1711         smp_wmb();
1712         atomic_inc(&kvm->online_vcpus);
1713
1714         mutex_unlock(&kvm->lock);
1715         return r;
1716
1717 unlock_vcpu_destroy:
1718         mutex_unlock(&kvm->lock);
1719 vcpu_destroy:
1720         kvm_arch_vcpu_destroy(vcpu);
1721         return r;
1722 }
1723
1724 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1725 {
1726         if (sigset) {
1727                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1728                 vcpu->sigset_active = 1;
1729                 vcpu->sigset = *sigset;
1730         } else
1731                 vcpu->sigset_active = 0;
1732         return 0;
1733 }
1734
1735 static long kvm_vcpu_ioctl(struct file *filp,
1736                            unsigned int ioctl, unsigned long arg)
1737 {
1738         struct kvm_vcpu *vcpu = filp->private_data;
1739         void __user *argp = (void __user *)arg;
1740         int r;
1741         struct kvm_fpu *fpu = NULL;
1742         struct kvm_sregs *kvm_sregs = NULL;
1743
1744         if (vcpu->kvm->mm != current->mm)
1745                 return -EIO;
1746
1747 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1748         /*
1749          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1750          * so vcpu_load() would break it.
1751          */
1752         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1753                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1754 #endif
1755
1756
1757         vcpu_load(vcpu);
1758         switch (ioctl) {
1759         case KVM_RUN:
1760                 r = -EINVAL;
1761                 if (arg)
1762                         goto out;
1763                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1764                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1765                 break;
1766         case KVM_GET_REGS: {
1767                 struct kvm_regs *kvm_regs;
1768
1769                 r = -ENOMEM;
1770                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1771                 if (!kvm_regs)
1772                         goto out;
1773                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1774                 if (r)
1775                         goto out_free1;
1776                 r = -EFAULT;
1777                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1778                         goto out_free1;
1779                 r = 0;
1780 out_free1:
1781                 kfree(kvm_regs);
1782                 break;
1783         }
1784         case KVM_SET_REGS: {
1785                 struct kvm_regs *kvm_regs;
1786
1787                 r = -ENOMEM;
1788                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1789                 if (IS_ERR(kvm_regs)) {
1790                         r = PTR_ERR(kvm_regs);
1791                         goto out;
1792                 }
1793                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1794                 if (r)
1795                         goto out_free2;
1796                 r = 0;
1797 out_free2:
1798                 kfree(kvm_regs);
1799                 break;
1800         }
1801         case KVM_GET_SREGS: {
1802                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1803                 r = -ENOMEM;
1804                 if (!kvm_sregs)
1805                         goto out;
1806                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1807                 if (r)
1808                         goto out;
1809                 r = -EFAULT;
1810                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1811                         goto out;
1812                 r = 0;
1813                 break;
1814         }
1815         case KVM_SET_SREGS: {
1816                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1817                 if (IS_ERR(kvm_sregs)) {
1818                         r = PTR_ERR(kvm_sregs);
1819                         goto out;
1820                 }
1821                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1822                 if (r)
1823                         goto out;
1824                 r = 0;
1825                 break;
1826         }
1827         case KVM_GET_MP_STATE: {
1828                 struct kvm_mp_state mp_state;
1829
1830                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1831                 if (r)
1832                         goto out;
1833                 r = -EFAULT;
1834                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1835                         goto out;
1836                 r = 0;
1837                 break;
1838         }
1839         case KVM_SET_MP_STATE: {
1840                 struct kvm_mp_state mp_state;
1841
1842                 r = -EFAULT;
1843                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1844                         goto out;
1845                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1846                 if (r)
1847                         goto out;
1848                 r = 0;
1849                 break;
1850         }
1851         case KVM_TRANSLATE: {
1852                 struct kvm_translation tr;
1853
1854                 r = -EFAULT;
1855                 if (copy_from_user(&tr, argp, sizeof tr))
1856                         goto out;
1857                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1858                 if (r)
1859                         goto out;
1860                 r = -EFAULT;
1861                 if (copy_to_user(argp, &tr, sizeof tr))
1862                         goto out;
1863                 r = 0;
1864                 break;
1865         }
1866         case KVM_SET_GUEST_DEBUG: {
1867                 struct kvm_guest_debug dbg;
1868
1869                 r = -EFAULT;
1870                 if (copy_from_user(&dbg, argp, sizeof dbg))
1871                         goto out;
1872                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1873                 if (r)
1874                         goto out;
1875                 r = 0;
1876                 break;
1877         }
1878         case KVM_SET_SIGNAL_MASK: {
1879                 struct kvm_signal_mask __user *sigmask_arg = argp;
1880                 struct kvm_signal_mask kvm_sigmask;
1881                 sigset_t sigset, *p;
1882
1883                 p = NULL;
1884                 if (argp) {
1885                         r = -EFAULT;
1886                         if (copy_from_user(&kvm_sigmask, argp,
1887                                            sizeof kvm_sigmask))
1888                                 goto out;
1889                         r = -EINVAL;
1890                         if (kvm_sigmask.len != sizeof sigset)
1891                                 goto out;
1892                         r = -EFAULT;
1893                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1894                                            sizeof sigset))
1895                                 goto out;
1896                         p = &sigset;
1897                 }
1898                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1899                 break;
1900         }
1901         case KVM_GET_FPU: {
1902                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1903                 r = -ENOMEM;
1904                 if (!fpu)
1905                         goto out;
1906                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1907                 if (r)
1908                         goto out;
1909                 r = -EFAULT;
1910                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1911                         goto out;
1912                 r = 0;
1913                 break;
1914         }
1915         case KVM_SET_FPU: {
1916                 fpu = memdup_user(argp, sizeof(*fpu));
1917                 if (IS_ERR(fpu)) {
1918                         r = PTR_ERR(fpu);
1919                         goto out;
1920                 }
1921                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1922                 if (r)
1923                         goto out;
1924                 r = 0;
1925                 break;
1926         }
1927         default:
1928                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1929         }
1930 out:
1931         vcpu_put(vcpu);
1932         kfree(fpu);
1933         kfree(kvm_sregs);
1934         return r;
1935 }
1936
1937 #ifdef CONFIG_COMPAT
1938 static long kvm_vcpu_compat_ioctl(struct file *filp,
1939                                   unsigned int ioctl, unsigned long arg)
1940 {
1941         struct kvm_vcpu *vcpu = filp->private_data;
1942         void __user *argp = compat_ptr(arg);
1943         int r;
1944
1945         if (vcpu->kvm->mm != current->mm)
1946                 return -EIO;
1947
1948         switch (ioctl) {
1949         case KVM_SET_SIGNAL_MASK: {
1950                 struct kvm_signal_mask __user *sigmask_arg = argp;
1951                 struct kvm_signal_mask kvm_sigmask;
1952                 compat_sigset_t csigset;
1953                 sigset_t sigset;
1954
1955                 if (argp) {
1956                         r = -EFAULT;
1957                         if (copy_from_user(&kvm_sigmask, argp,
1958                                            sizeof kvm_sigmask))
1959                                 goto out;
1960                         r = -EINVAL;
1961                         if (kvm_sigmask.len != sizeof csigset)
1962                                 goto out;
1963                         r = -EFAULT;
1964                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1965                                            sizeof csigset))
1966                                 goto out;
1967                 }
1968                 sigset_from_compat(&sigset, &csigset);
1969                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1970                 break;
1971         }
1972         default:
1973                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1974         }
1975
1976 out:
1977         return r;
1978 }
1979 #endif
1980
1981 static long kvm_vm_ioctl(struct file *filp,
1982                            unsigned int ioctl, unsigned long arg)
1983 {
1984         struct kvm *kvm = filp->private_data;
1985         void __user *argp = (void __user *)arg;
1986         int r;
1987
1988         if (kvm->mm != current->mm)
1989                 return -EIO;
1990         switch (ioctl) {
1991         case KVM_CREATE_VCPU:
1992                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1993                 if (r < 0)
1994                         goto out;
1995                 break;
1996         case KVM_SET_USER_MEMORY_REGION: {
1997                 struct kvm_userspace_memory_region kvm_userspace_mem;
1998
1999                 r = -EFAULT;
2000                 if (copy_from_user(&kvm_userspace_mem, argp,
2001                                                 sizeof kvm_userspace_mem))
2002                         goto out;
2003
2004                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2005                 if (r)
2006                         goto out;
2007                 break;
2008         }
2009         case KVM_GET_DIRTY_LOG: {
2010                 struct kvm_dirty_log log;
2011
2012                 r = -EFAULT;
2013                 if (copy_from_user(&log, argp, sizeof log))
2014                         goto out;
2015                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2016                 if (r)
2017                         goto out;
2018                 break;
2019         }
2020 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2021         case KVM_REGISTER_COALESCED_MMIO: {
2022                 struct kvm_coalesced_mmio_zone zone;
2023                 r = -EFAULT;
2024                 if (copy_from_user(&zone, argp, sizeof zone))
2025                         goto out;
2026                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2027                 if (r)
2028                         goto out;
2029                 r = 0;
2030                 break;
2031         }
2032         case KVM_UNREGISTER_COALESCED_MMIO: {
2033                 struct kvm_coalesced_mmio_zone zone;
2034                 r = -EFAULT;
2035                 if (copy_from_user(&zone, argp, sizeof zone))
2036                         goto out;
2037                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2038                 if (r)
2039                         goto out;
2040                 r = 0;
2041                 break;
2042         }
2043 #endif
2044         case KVM_IRQFD: {
2045                 struct kvm_irqfd data;
2046
2047                 r = -EFAULT;
2048                 if (copy_from_user(&data, argp, sizeof data))
2049                         goto out;
2050                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2051                 break;
2052         }
2053         case KVM_IOEVENTFD: {
2054                 struct kvm_ioeventfd data;
2055
2056                 r = -EFAULT;
2057                 if (copy_from_user(&data, argp, sizeof data))
2058                         goto out;
2059                 r = kvm_ioeventfd(kvm, &data);
2060                 break;
2061         }
2062 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2063         case KVM_SET_BOOT_CPU_ID:
2064                 r = 0;
2065                 mutex_lock(&kvm->lock);
2066                 if (atomic_read(&kvm->online_vcpus) != 0)
2067                         r = -EBUSY;
2068                 else
2069                         kvm->bsp_vcpu_id = arg;
2070                 mutex_unlock(&kvm->lock);
2071                 break;
2072 #endif
2073 #ifdef CONFIG_HAVE_KVM_MSI
2074         case KVM_SIGNAL_MSI: {
2075                 struct kvm_msi msi;
2076
2077                 r = -EFAULT;
2078                 if (copy_from_user(&msi, argp, sizeof msi))
2079                         goto out;
2080                 r = kvm_send_userspace_msi(kvm, &msi);
2081                 break;
2082         }
2083 #endif
2084         default:
2085                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2086                 if (r == -ENOTTY)
2087                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2088         }
2089 out:
2090         return r;
2091 }
2092
2093 #ifdef CONFIG_COMPAT
2094 struct compat_kvm_dirty_log {
2095         __u32 slot;
2096         __u32 padding1;
2097         union {
2098                 compat_uptr_t dirty_bitmap; /* one bit per page */
2099                 __u64 padding2;
2100         };
2101 };
2102
2103 static long kvm_vm_compat_ioctl(struct file *filp,
2104                            unsigned int ioctl, unsigned long arg)
2105 {
2106         struct kvm *kvm = filp->private_data;
2107         int r;
2108
2109         if (kvm->mm != current->mm)
2110                 return -EIO;
2111         switch (ioctl) {
2112         case KVM_GET_DIRTY_LOG: {
2113                 struct compat_kvm_dirty_log compat_log;
2114                 struct kvm_dirty_log log;
2115
2116                 r = -EFAULT;
2117                 if (copy_from_user(&compat_log, (void __user *)arg,
2118                                    sizeof(compat_log)))
2119                         goto out;
2120                 log.slot         = compat_log.slot;
2121                 log.padding1     = compat_log.padding1;
2122                 log.padding2     = compat_log.padding2;
2123                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2124
2125                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2126                 if (r)
2127                         goto out;
2128                 break;
2129         }
2130         default:
2131                 r = kvm_vm_ioctl(filp, ioctl, arg);
2132         }
2133
2134 out:
2135         return r;
2136 }
2137 #endif
2138
2139 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2140 {
2141         struct page *page[1];
2142         unsigned long addr;
2143         int npages;
2144         gfn_t gfn = vmf->pgoff;
2145         struct kvm *kvm = vma->vm_file->private_data;
2146
2147         addr = gfn_to_hva(kvm, gfn);
2148         if (kvm_is_error_hva(addr))
2149                 return VM_FAULT_SIGBUS;
2150
2151         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2152                                 NULL);
2153         if (unlikely(npages != 1))
2154                 return VM_FAULT_SIGBUS;
2155
2156         vmf->page = page[0];
2157         return 0;
2158 }
2159
2160 static const struct vm_operations_struct kvm_vm_vm_ops = {
2161         .fault = kvm_vm_fault,
2162 };
2163
2164 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2165 {
2166         vma->vm_ops = &kvm_vm_vm_ops;
2167         return 0;
2168 }
2169
2170 static struct file_operations kvm_vm_fops = {
2171         .release        = kvm_vm_release,
2172         .unlocked_ioctl = kvm_vm_ioctl,
2173 #ifdef CONFIG_COMPAT
2174         .compat_ioctl   = kvm_vm_compat_ioctl,
2175 #endif
2176         .mmap           = kvm_vm_mmap,
2177         .llseek         = noop_llseek,
2178 };
2179
2180 static int kvm_dev_ioctl_create_vm(unsigned long type)
2181 {
2182         int r;
2183         struct kvm *kvm;
2184
2185         kvm = kvm_create_vm(type);
2186         if (IS_ERR(kvm))
2187                 return PTR_ERR(kvm);
2188 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2189         r = kvm_coalesced_mmio_init(kvm);
2190         if (r < 0) {
2191                 kvm_put_kvm(kvm);
2192                 return r;
2193         }
2194 #endif
2195         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2196         if (r < 0)
2197                 kvm_put_kvm(kvm);
2198
2199         return r;
2200 }
2201
2202 static long kvm_dev_ioctl_check_extension_generic(long arg)
2203 {
2204         switch (arg) {
2205         case KVM_CAP_USER_MEMORY:
2206         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2207         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2208 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2209         case KVM_CAP_SET_BOOT_CPU_ID:
2210 #endif
2211         case KVM_CAP_INTERNAL_ERROR_DATA:
2212 #ifdef CONFIG_HAVE_KVM_MSI
2213         case KVM_CAP_SIGNAL_MSI:
2214 #endif
2215                 return 1;
2216 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2217         case KVM_CAP_IRQ_ROUTING:
2218                 return KVM_MAX_IRQ_ROUTES;
2219 #endif
2220         default:
2221                 break;
2222         }
2223         return kvm_dev_ioctl_check_extension(arg);
2224 }
2225
2226 static long kvm_dev_ioctl(struct file *filp,
2227                           unsigned int ioctl, unsigned long arg)
2228 {
2229         long r = -EINVAL;
2230
2231         switch (ioctl) {
2232         case KVM_GET_API_VERSION:
2233                 r = -EINVAL;
2234                 if (arg)
2235                         goto out;
2236                 r = KVM_API_VERSION;
2237                 break;
2238         case KVM_CREATE_VM:
2239                 r = kvm_dev_ioctl_create_vm(arg);
2240                 break;
2241         case KVM_CHECK_EXTENSION:
2242                 r = kvm_dev_ioctl_check_extension_generic(arg);
2243                 break;
2244         case KVM_GET_VCPU_MMAP_SIZE:
2245                 r = -EINVAL;
2246                 if (arg)
2247                         goto out;
2248                 r = PAGE_SIZE;     /* struct kvm_run */
2249 #ifdef CONFIG_X86
2250                 r += PAGE_SIZE;    /* pio data page */
2251 #endif
2252 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2253                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2254 #endif
2255                 break;
2256         case KVM_TRACE_ENABLE:
2257         case KVM_TRACE_PAUSE:
2258         case KVM_TRACE_DISABLE:
2259                 r = -EOPNOTSUPP;
2260                 break;
2261         default:
2262                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2263         }
2264 out:
2265         return r;
2266 }
2267
2268 static struct file_operations kvm_chardev_ops = {
2269         .unlocked_ioctl = kvm_dev_ioctl,
2270         .compat_ioctl   = kvm_dev_ioctl,
2271         .llseek         = noop_llseek,
2272 };
2273
2274 static struct miscdevice kvm_dev = {
2275         KVM_MINOR,
2276         "kvm",
2277         &kvm_chardev_ops,
2278 };
2279
2280 static void hardware_enable_nolock(void *junk)
2281 {
2282         int cpu = raw_smp_processor_id();
2283         int r;
2284
2285         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2286                 return;
2287
2288         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2289
2290         r = kvm_arch_hardware_enable(NULL);
2291
2292         if (r) {
2293                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2294                 atomic_inc(&hardware_enable_failed);
2295                 printk(KERN_INFO "kvm: enabling virtualization on "
2296                                  "CPU%d failed\n", cpu);
2297         }
2298 }
2299
2300 static void hardware_enable(void *junk)
2301 {
2302         raw_spin_lock(&kvm_lock);
2303         hardware_enable_nolock(junk);
2304         raw_spin_unlock(&kvm_lock);
2305 }
2306
2307 static void hardware_disable_nolock(void *junk)
2308 {
2309         int cpu = raw_smp_processor_id();
2310
2311         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2312                 return;
2313         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2314         kvm_arch_hardware_disable(NULL);
2315 }
2316
2317 static void hardware_disable(void *junk)
2318 {
2319         raw_spin_lock(&kvm_lock);
2320         hardware_disable_nolock(junk);
2321         raw_spin_unlock(&kvm_lock);
2322 }
2323
2324 static void hardware_disable_all_nolock(void)
2325 {
2326         BUG_ON(!kvm_usage_count);
2327
2328         kvm_usage_count--;
2329         if (!kvm_usage_count)
2330                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2331 }
2332
2333 static void hardware_disable_all(void)
2334 {
2335         raw_spin_lock(&kvm_lock);
2336         hardware_disable_all_nolock();
2337         raw_spin_unlock(&kvm_lock);
2338 }
2339
2340 static int hardware_enable_all(void)
2341 {
2342         int r = 0;
2343
2344         raw_spin_lock(&kvm_lock);
2345
2346         kvm_usage_count++;
2347         if (kvm_usage_count == 1) {
2348                 atomic_set(&hardware_enable_failed, 0);
2349                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2350
2351                 if (atomic_read(&hardware_enable_failed)) {
2352                         hardware_disable_all_nolock();
2353                         r = -EBUSY;
2354                 }
2355         }
2356
2357         raw_spin_unlock(&kvm_lock);
2358
2359         return r;
2360 }
2361
2362 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2363                            void *v)
2364 {
2365         int cpu = (long)v;
2366
2367         if (!kvm_usage_count)
2368                 return NOTIFY_OK;
2369
2370         val &= ~CPU_TASKS_FROZEN;
2371         switch (val) {
2372         case CPU_DYING:
2373                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2374                        cpu);
2375                 hardware_disable(NULL);
2376                 break;
2377         case CPU_STARTING:
2378                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2379                        cpu);
2380                 hardware_enable(NULL);
2381                 break;
2382         }
2383         return NOTIFY_OK;
2384 }
2385
2386
2387 asmlinkage void kvm_spurious_fault(void)
2388 {
2389         /* Fault while not rebooting.  We want the trace. */
2390         BUG();
2391 }
2392 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2393
2394 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2395                       void *v)
2396 {
2397         /*
2398          * Some (well, at least mine) BIOSes hang on reboot if
2399          * in vmx root mode.
2400          *
2401          * And Intel TXT required VMX off for all cpu when system shutdown.
2402          */
2403         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2404         kvm_rebooting = true;
2405         on_each_cpu(hardware_disable_nolock, NULL, 1);
2406         return NOTIFY_OK;
2407 }
2408
2409 static struct notifier_block kvm_reboot_notifier = {
2410         .notifier_call = kvm_reboot,
2411         .priority = 0,
2412 };
2413
2414 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2415 {
2416         int i;
2417
2418         for (i = 0; i < bus->dev_count; i++) {
2419                 struct kvm_io_device *pos = bus->range[i].dev;
2420
2421                 kvm_iodevice_destructor(pos);
2422         }
2423         kfree(bus);
2424 }
2425
2426 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2427 {
2428         const struct kvm_io_range *r1 = p1;
2429         const struct kvm_io_range *r2 = p2;
2430
2431         if (r1->addr < r2->addr)
2432                 return -1;
2433         if (r1->addr + r1->len > r2->addr + r2->len)
2434                 return 1;
2435         return 0;
2436 }
2437
2438 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2439                           gpa_t addr, int len)
2440 {
2441         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2442                 .addr = addr,
2443                 .len = len,
2444                 .dev = dev,
2445         };
2446
2447         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2448                 kvm_io_bus_sort_cmp, NULL);
2449
2450         return 0;
2451 }
2452
2453 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2454                              gpa_t addr, int len)
2455 {
2456         struct kvm_io_range *range, key;
2457         int off;
2458
2459         key = (struct kvm_io_range) {
2460                 .addr = addr,
2461                 .len = len,
2462         };
2463
2464         range = bsearch(&key, bus->range, bus->dev_count,
2465                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2466         if (range == NULL)
2467                 return -ENOENT;
2468
2469         off = range - bus->range;
2470
2471         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2472                 off--;
2473
2474         return off;
2475 }
2476
2477 /* kvm_io_bus_write - called under kvm->slots_lock */
2478 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2479                      int len, const void *val)
2480 {
2481         int idx;
2482         struct kvm_io_bus *bus;
2483         struct kvm_io_range range;
2484
2485         range = (struct kvm_io_range) {
2486                 .addr = addr,
2487                 .len = len,
2488         };
2489
2490         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2491         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2492         if (idx < 0)
2493                 return -EOPNOTSUPP;
2494
2495         while (idx < bus->dev_count &&
2496                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2497                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2498                         return 0;
2499                 idx++;
2500         }
2501
2502         return -EOPNOTSUPP;
2503 }
2504
2505 /* kvm_io_bus_read - called under kvm->slots_lock */
2506 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2507                     int len, void *val)
2508 {
2509         int idx;
2510         struct kvm_io_bus *bus;
2511         struct kvm_io_range range;
2512
2513         range = (struct kvm_io_range) {
2514                 .addr = addr,
2515                 .len = len,
2516         };
2517
2518         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2519         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2520         if (idx < 0)
2521                 return -EOPNOTSUPP;
2522
2523         while (idx < bus->dev_count &&
2524                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2525                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2526                         return 0;
2527                 idx++;
2528         }
2529
2530         return -EOPNOTSUPP;
2531 }
2532
2533 /* Caller must hold slots_lock. */
2534 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2535                             int len, struct kvm_io_device *dev)
2536 {
2537         struct kvm_io_bus *new_bus, *bus;
2538
2539         bus = kvm->buses[bus_idx];
2540         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2541                 return -ENOSPC;
2542
2543         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2544                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2545         if (!new_bus)
2546                 return -ENOMEM;
2547         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2548                sizeof(struct kvm_io_range)));
2549         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2550         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2551         synchronize_srcu_expedited(&kvm->srcu);
2552         kfree(bus);
2553
2554         return 0;
2555 }
2556
2557 /* Caller must hold slots_lock. */
2558 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2559                               struct kvm_io_device *dev)
2560 {
2561         int i, r;
2562         struct kvm_io_bus *new_bus, *bus;
2563
2564         bus = kvm->buses[bus_idx];
2565         r = -ENOENT;
2566         for (i = 0; i < bus->dev_count; i++)
2567                 if (bus->range[i].dev == dev) {
2568                         r = 0;
2569                         break;
2570                 }
2571
2572         if (r)
2573                 return r;
2574
2575         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2576                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2577         if (!new_bus)
2578                 return -ENOMEM;
2579
2580         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2581         new_bus->dev_count--;
2582         memcpy(new_bus->range + i, bus->range + i + 1,
2583                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2584
2585         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2586         synchronize_srcu_expedited(&kvm->srcu);
2587         kfree(bus);
2588         return r;
2589 }
2590
2591 static struct notifier_block kvm_cpu_notifier = {
2592         .notifier_call = kvm_cpu_hotplug,
2593 };
2594
2595 static int vm_stat_get(void *_offset, u64 *val)
2596 {
2597         unsigned offset = (long)_offset;
2598         struct kvm *kvm;
2599
2600         *val = 0;
2601         raw_spin_lock(&kvm_lock);
2602         list_for_each_entry(kvm, &vm_list, vm_list)
2603                 *val += *(u32 *)((void *)kvm + offset);
2604         raw_spin_unlock(&kvm_lock);
2605         return 0;
2606 }
2607
2608 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2609
2610 static int vcpu_stat_get(void *_offset, u64 *val)
2611 {
2612         unsigned offset = (long)_offset;
2613         struct kvm *kvm;
2614         struct kvm_vcpu *vcpu;
2615         int i;
2616
2617         *val = 0;
2618         raw_spin_lock(&kvm_lock);
2619         list_for_each_entry(kvm, &vm_list, vm_list)
2620                 kvm_for_each_vcpu(i, vcpu, kvm)
2621                         *val += *(u32 *)((void *)vcpu + offset);
2622
2623         raw_spin_unlock(&kvm_lock);
2624         return 0;
2625 }
2626
2627 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2628
2629 static const struct file_operations *stat_fops[] = {
2630         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2631         [KVM_STAT_VM]   = &vm_stat_fops,
2632 };
2633
2634 static int kvm_init_debug(void)
2635 {
2636         int r = -EFAULT;
2637         struct kvm_stats_debugfs_item *p;
2638
2639         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2640         if (kvm_debugfs_dir == NULL)
2641                 goto out;
2642
2643         for (p = debugfs_entries; p->name; ++p) {
2644                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2645                                                 (void *)(long)p->offset,
2646                                                 stat_fops[p->kind]);
2647                 if (p->dentry == NULL)
2648                         goto out_dir;
2649         }
2650
2651         return 0;
2652
2653 out_dir:
2654         debugfs_remove_recursive(kvm_debugfs_dir);
2655 out:
2656         return r;
2657 }
2658
2659 static void kvm_exit_debug(void)
2660 {
2661         struct kvm_stats_debugfs_item *p;
2662
2663         for (p = debugfs_entries; p->name; ++p)
2664                 debugfs_remove(p->dentry);
2665         debugfs_remove(kvm_debugfs_dir);
2666 }
2667
2668 static int kvm_suspend(void)
2669 {
2670         if (kvm_usage_count)
2671                 hardware_disable_nolock(NULL);
2672         return 0;
2673 }
2674
2675 static void kvm_resume(void)
2676 {
2677         if (kvm_usage_count) {
2678                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2679                 hardware_enable_nolock(NULL);
2680         }
2681 }
2682
2683 static struct syscore_ops kvm_syscore_ops = {
2684         .suspend = kvm_suspend,
2685         .resume = kvm_resume,
2686 };
2687
2688 struct page *bad_page;
2689 pfn_t bad_pfn;
2690
2691 static inline
2692 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2693 {
2694         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2695 }
2696
2697 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2698 {
2699         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2700
2701         kvm_arch_vcpu_load(vcpu, cpu);
2702 }
2703
2704 static void kvm_sched_out(struct preempt_notifier *pn,
2705                           struct task_struct *next)
2706 {
2707         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2708
2709         kvm_arch_vcpu_put(vcpu);
2710 }
2711
2712 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2713                   struct module *module)
2714 {
2715         int r;
2716         int cpu;
2717
2718         r = kvm_arch_init(opaque);
2719         if (r)
2720                 goto out_fail;
2721
2722         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2723
2724         if (bad_page == NULL) {
2725                 r = -ENOMEM;
2726                 goto out;
2727         }
2728
2729         bad_pfn = page_to_pfn(bad_page);
2730
2731         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2732
2733         if (hwpoison_page == NULL) {
2734                 r = -ENOMEM;
2735                 goto out_free_0;
2736         }
2737
2738         hwpoison_pfn = page_to_pfn(hwpoison_page);
2739
2740         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2741
2742         if (fault_page == NULL) {
2743                 r = -ENOMEM;
2744                 goto out_free_0;
2745         }
2746
2747         fault_pfn = page_to_pfn(fault_page);
2748
2749         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2750                 r = -ENOMEM;
2751                 goto out_free_0;
2752         }
2753
2754         r = kvm_arch_hardware_setup();
2755         if (r < 0)
2756                 goto out_free_0a;
2757
2758         for_each_online_cpu(cpu) {
2759                 smp_call_function_single(cpu,
2760                                 kvm_arch_check_processor_compat,
2761                                 &r, 1);
2762                 if (r < 0)
2763                         goto out_free_1;
2764         }
2765
2766         r = register_cpu_notifier(&kvm_cpu_notifier);
2767         if (r)
2768                 goto out_free_2;
2769         register_reboot_notifier(&kvm_reboot_notifier);
2770
2771         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2772         if (!vcpu_align)
2773                 vcpu_align = __alignof__(struct kvm_vcpu);
2774         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2775                                            0, NULL);
2776         if (!kvm_vcpu_cache) {
2777                 r = -ENOMEM;
2778                 goto out_free_3;
2779         }
2780
2781         r = kvm_async_pf_init();
2782         if (r)
2783                 goto out_free;
2784
2785         kvm_chardev_ops.owner = module;
2786         kvm_vm_fops.owner = module;
2787         kvm_vcpu_fops.owner = module;
2788
2789         r = misc_register(&kvm_dev);
2790         if (r) {
2791                 printk(KERN_ERR "kvm: misc device register failed\n");
2792                 goto out_unreg;
2793         }
2794
2795         register_syscore_ops(&kvm_syscore_ops);
2796
2797         kvm_preempt_ops.sched_in = kvm_sched_in;
2798         kvm_preempt_ops.sched_out = kvm_sched_out;
2799
2800         r = kvm_init_debug();
2801         if (r) {
2802                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2803                 goto out_undebugfs;
2804         }
2805
2806         return 0;
2807
2808 out_undebugfs:
2809         unregister_syscore_ops(&kvm_syscore_ops);
2810 out_unreg:
2811         kvm_async_pf_deinit();
2812 out_free:
2813         kmem_cache_destroy(kvm_vcpu_cache);
2814 out_free_3:
2815         unregister_reboot_notifier(&kvm_reboot_notifier);
2816         unregister_cpu_notifier(&kvm_cpu_notifier);
2817 out_free_2:
2818 out_free_1:
2819         kvm_arch_hardware_unsetup();
2820 out_free_0a:
2821         free_cpumask_var(cpus_hardware_enabled);
2822 out_free_0:
2823         if (fault_page)
2824                 __free_page(fault_page);
2825         if (hwpoison_page)
2826                 __free_page(hwpoison_page);
2827         __free_page(bad_page);
2828 out:
2829         kvm_arch_exit();
2830 out_fail:
2831         return r;
2832 }
2833 EXPORT_SYMBOL_GPL(kvm_init);
2834
2835 void kvm_exit(void)
2836 {
2837         kvm_exit_debug();
2838         misc_deregister(&kvm_dev);
2839         kmem_cache_destroy(kvm_vcpu_cache);
2840         kvm_async_pf_deinit();
2841         unregister_syscore_ops(&kvm_syscore_ops);
2842         unregister_reboot_notifier(&kvm_reboot_notifier);
2843         unregister_cpu_notifier(&kvm_cpu_notifier);
2844         on_each_cpu(hardware_disable_nolock, NULL, 1);
2845         kvm_arch_hardware_unsetup();
2846         kvm_arch_exit();
2847         free_cpumask_var(cpus_hardware_enabled);
2848         __free_page(hwpoison_page);
2849         __free_page(bad_page);
2850 }
2851 EXPORT_SYMBOL_GPL(kvm_exit);