Pileus Git - ~andy/linux/blob - 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 /*
 520  * Avoid using vmalloc for a small buffer.
 521  * Should not be used when the size is statically known.
 522  */
 523 void *kvm_kvzalloc(unsigned long size)
 524 {
 525         if (size > PAGE_SIZE)
 526                 return vzalloc(size);
 527         else
 528                 return kzalloc(size, GFP_KERNEL);
 529 }
 530
 531 void kvm_kvfree(const void *addr)
 532 {
 533         if (is_vmalloc_addr(addr))
 534                 vfree(addr);
 535         else
 536                 kfree(addr);
 537 }
 538
 539 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 540 {
 541         if (!memslot->dirty_bitmap)
 542                 return;
 543
 544         kvm_kvfree(memslot->dirty_bitmap);
 545         memslot->dirty_bitmap = NULL;
 546 }
 547
 548 /*
 549  * Free any memory in @free but not in @dont.
 550  */
 551 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 552                                   struct kvm_memory_slot *dont)
 553 {
 554         if (!dont || free->rmap != dont->rmap)
 555                 vfree(free->rmap);
 556
 557         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 558                 kvm_destroy_dirty_bitmap(free);
 559
 560         kvm_arch_free_memslot(free, dont);
 561
 562         free->npages = 0;
 563         free->rmap = NULL;
 564 }
 565
 566 void kvm_free_physmem(struct kvm *kvm)
 567 {
 568         struct kvm_memslots *slots = kvm->memslots;
 569         struct kvm_memory_slot *memslot;
 570
 571         kvm_for_each_memslot(memslot, slots)
 572                 kvm_free_physmem_slot(memslot, NULL);
 573
 574         kfree(kvm->memslots);
 575 }
 576
 577 static void kvm_destroy_vm(struct kvm *kvm)
 578 {
 579         int i;
 580         struct mm_struct *mm = kvm->mm;
 581
 582         kvm_arch_sync_events(kvm);
 583         raw_spin_lock(&kvm_lock);
 584         list_del(&kvm->vm_list);
 585         raw_spin_unlock(&kvm_lock);
 586         kvm_free_irq_routing(kvm);
 587         for (i = 0; i < KVM_NR_BUSES; i++)
 588                 kvm_io_bus_destroy(kvm->buses[i]);
 589         kvm_coalesced_mmio_free(kvm);
 590 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 591         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 592 #else
 593         kvm_arch_flush_shadow(kvm);
 594 #endif
 595         kvm_arch_destroy_vm(kvm);
 596         kvm_free_physmem(kvm);
 597         cleanup_srcu_struct(&kvm->srcu);
 598         kvm_arch_free_vm(kvm);
 599         hardware_disable_all();
 600         mmdrop(mm);
 601 }
 602
 603 void kvm_get_kvm(struct kvm *kvm)
 604 {
 605         atomic_inc(&kvm->users_count);
 606 }
 607 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 608
 609 void kvm_put_kvm(struct kvm *kvm)
 610 {
 611         if (atomic_dec_and_test(&kvm->users_count))
 612                 kvm_destroy_vm(kvm);
 613 }
 614 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 615
 616
 617 static int kvm_vm_release(struct inode *inode, struct file *filp)
 618 {
 619         struct kvm *kvm = filp->private_data;
 620
 621         kvm_irqfd_release(kvm);
 622
 623         kvm_put_kvm(kvm);
 624         return 0;
 625 }
 626
 627 /*
 628  * Allocation size is twice as large as the actual dirty bitmap size.
 629  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 630  */
 631 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 632 {
 633 #ifndef CONFIG_S390
 634         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 635
 636         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 637         if (!memslot->dirty_bitmap)
 638                 return -ENOMEM;
 639
 640 #endif /* !CONFIG_S390 */
 641         return 0;
 642 }
 643
 644 static int cmp_memslot(const void *slot1, const void *slot2)
 645 {
 646         struct kvm_memory_slot *s1, *s2;
 647
 648         s1 = (struct kvm_memory_slot *)slot1;
 649         s2 = (struct kvm_memory_slot *)slot2;
 650
 651         if (s1->npages < s2->npages)
 652                 return 1;
 653         if (s1->npages > s2->npages)
 654                 return -1;
 655
 656         return 0;
 657 }
 658
 659 /*
 660  * Sort the memslots base on its size, so the larger slots
 661  * will get better fit.
 662  */
 663 static void sort_memslots(struct kvm_memslots *slots)
 664 {
 665         int i;
 666
 667         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 668               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 669
 670         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 671                 slots->id_to_index[slots->memslots[i].id] = i;
 672 }
 673
 674 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
 675 {
 676         if (new) {
 677                 int id = new->id;
 678                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 679                 unsigned long npages = old->npages;
 680
 681                 *old = *new;
 682                 if (new->npages != npages)
 683                         sort_memslots(slots);
 684         }
 685
 686         slots->generation++;
 687 }
 688
 689 /*
 690  * Allocate some memory and give it an address in the guest physical address
 691  * space.
 692  *
 693  * Discontiguous memory is allowed, mostly for framebuffers.
 694  *
 695  * Must be called holding mmap_sem for write.
 696  */
 697 int __kvm_set_memory_region(struct kvm *kvm,
 698                             struct kvm_userspace_memory_region *mem,
 699                             int user_alloc)
 700 {
 701         int r;
 702         gfn_t base_gfn;
 703         unsigned long npages;
 704         unsigned long i;
 705         struct kvm_memory_slot *memslot;
 706         struct kvm_memory_slot old, new;
 707         struct kvm_memslots *slots, *old_memslots;
 708
 709         r = -EINVAL;
 710         /* General sanity checks */
 711         if (mem->memory_size & (PAGE_SIZE - 1))
 712                 goto out;
 713         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 714                 goto out;
 715         /* We can read the guest memory with __xxx_user() later on. */
 716         if (user_alloc &&
 717             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 718              !access_ok(VERIFY_WRITE,
 719                         (void __user *)(unsigned long)mem->userspace_addr,
 720                         mem->memory_size)))
 721                 goto out;
 722         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 723                 goto out;
 724         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 725                 goto out;
 726
 727         memslot = id_to_memslot(kvm->memslots, mem->slot);
 728         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 729         npages = mem->memory_size >> PAGE_SHIFT;
 730
 731         r = -EINVAL;
 732         if (npages > KVM_MEM_MAX_NR_PAGES)
 733                 goto out;
 734
 735         if (!npages)
 736                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 737
 738         new = old = *memslot;
 739
 740         new.id = mem->slot;
 741         new.base_gfn = base_gfn;
 742         new.npages = npages;
 743         new.flags = mem->flags;
 744
 745         /* Disallow changing a memory slot's size. */
 746         r = -EINVAL;
 747         if (npages && old.npages && npages != old.npages)
 748                 goto out_free;
 749
 750         /* Check for overlaps */
 751         r = -EEXIST;
 752         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 753                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 754
 755                 if (s == memslot || !s->npages)
 756                         continue;
 757                 if (!((base_gfn + npages <= s->base_gfn) ||
 758                       (base_gfn >= s->base_gfn + s->npages)))
 759                         goto out_free;
 760         }
 761
 762         /* Free page dirty bitmap if unneeded */
 763         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 764                 new.dirty_bitmap = NULL;
 765
 766         r = -ENOMEM;
 767
 768         /* Allocate if a slot is being created */
 769         if (npages && !old.npages) {
 770                 new.user_alloc = user_alloc;
 771                 new.userspace_addr = mem->userspace_addr;
 772 #ifndef CONFIG_S390
 773                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
 774                 if (!new.rmap)
 775                         goto out_free;
 776 #endif /* not defined CONFIG_S390 */
 777                 if (kvm_arch_create_memslot(&new, npages))
 778                         goto out_free;
 779         }
 780
 781         /* Allocate page dirty bitmap if needed */
 782         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 783                 if (kvm_create_dirty_bitmap(&new) < 0)
 784                         goto out_free;
 785                 /* destroy any largepage mappings for dirty tracking */
 786         }
 787
 788         if (!npages) {
 789                 struct kvm_memory_slot *slot;
 790
 791                 r = -ENOMEM;
 792                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 793                                 GFP_KERNEL);
 794                 if (!slots)
 795                         goto out_free;
 796                 slot = id_to_memslot(slots, mem->slot);
 797                 slot->flags |= KVM_MEMSLOT_INVALID;
 798
 799                 update_memslots(slots, NULL);
 800
 801                 old_memslots = kvm->memslots;
 802                 rcu_assign_pointer(kvm->memslots, slots);
 803                 synchronize_srcu_expedited(&kvm->srcu);
 804                 /* From this point no new shadow pages pointing to a deleted
 805                  * memslot will be created.
 806                  *
 807                  * validation of sp->gfn happens in:
 808                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 809                  *      - kvm_is_visible_gfn (mmu_check_roots)
 810                  */
 811                 kvm_arch_flush_shadow(kvm);
 812                 kfree(old_memslots);
 813         }
 814
 815         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 816         if (r)
 817                 goto out_free;
 818
 819         /* map/unmap the pages in iommu page table */
 820         if (npages) {
 821                 r = kvm_iommu_map_pages(kvm, &new);
 822                 if (r)
 823                         goto out_free;
 824         } else
 825                 kvm_iommu_unmap_pages(kvm, &old);
 826
 827         r = -ENOMEM;
 828         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 829                         GFP_KERNEL);
 830         if (!slots)
 831                 goto out_free;
 832
 833         /* actual memory is freed via old in kvm_free_physmem_slot below */
 834         if (!npages) {
 835                 new.rmap = NULL;
 836                 new.dirty_bitmap = NULL;
 837                 memset(&new.arch, 0, sizeof(new.arch));
 838         }
 839
 840         update_memslots(slots, &new);
 841         old_memslots = kvm->memslots;
 842         rcu_assign_pointer(kvm->memslots, slots);
 843         synchronize_srcu_expedited(&kvm->srcu);
 844
 845         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 846
 847         /*
 848          * If the new memory slot is created, we need to clear all
 849          * mmio sptes.
 850          */
 851         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 852                 kvm_arch_flush_shadow(kvm);
 853
 854         kvm_free_physmem_slot(&old, &new);
 855         kfree(old_memslots);
 856
 857         return 0;
 858
 859 out_free:
 860         kvm_free_physmem_slot(&new, &old);
 861 out:
 862         return r;
 863
 864 }
 865 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 866
 867 int kvm_set_memory_region(struct kvm *kvm,
 868                           struct kvm_userspace_memory_region *mem,
 869                           int user_alloc)
 870 {
 871         int r;
 872
 873         mutex_lock(&kvm->slots_lock);
 874         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 875         mutex_unlock(&kvm->slots_lock);
 876         return r;
 877 }
 878 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 879
 880 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 881                                    struct
 882                                    kvm_userspace_memory_region *mem,
 883                                    int user_alloc)
 884 {
 885         if (mem->slot >= KVM_MEMORY_SLOTS)
 886                 return -EINVAL;
 887         return kvm_set_memory_region(kvm, mem, user_alloc);
 888 }
 889
 890 int kvm_get_dirty_log(struct kvm *kvm,
 891                         struct kvm_dirty_log *log, int *is_dirty)
 892 {
 893         struct kvm_memory_slot *memslot;
 894         int r, i;
 895         unsigned long n;
 896         unsigned long any = 0;
 897
 898         r = -EINVAL;
 899         if (log->slot >= KVM_MEMORY_SLOTS)
 900                 goto out;
 901
 902         memslot = id_to_memslot(kvm->memslots, log->slot);
 903         r = -ENOENT;
 904         if (!memslot->dirty_bitmap)
 905                 goto out;
 906
 907         n = kvm_dirty_bitmap_bytes(memslot);
 908
 909         for (i = 0; !any && i < n/sizeof(long); ++i)
 910                 any = memslot->dirty_bitmap[i];
 911
 912         r = -EFAULT;
 913         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 914                 goto out;
 915
 916         if (any)
 917                 *is_dirty = 1;
 918
 919         r = 0;
 920 out:
 921         return r;
 922 }
 923
 924 bool kvm_largepages_enabled(void)
 925 {
 926         return largepages_enabled;
 927 }
 928
 929 void kvm_disable_largepages(void)
 930 {
 931         largepages_enabled = false;
 932 }
 933 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 934
 935 int is_error_page(struct page *page)
 936 {
 937         return page == bad_page || page == hwpoison_page || page == fault_page;
 938 }
 939 EXPORT_SYMBOL_GPL(is_error_page);
 940
 941 int is_error_pfn(pfn_t pfn)
 942 {
 943         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 944 }
 945 EXPORT_SYMBOL_GPL(is_error_pfn);
 946
 947 int is_hwpoison_pfn(pfn_t pfn)
 948 {
 949         return pfn == hwpoison_pfn;
 950 }
 951 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 952
 953 int is_fault_pfn(pfn_t pfn)
 954 {
 955         return pfn == fault_pfn;
 956 }
 957 EXPORT_SYMBOL_GPL(is_fault_pfn);
 958
 959 int is_noslot_pfn(pfn_t pfn)
 960 {
 961         return pfn == bad_pfn;
 962 }
 963 EXPORT_SYMBOL_GPL(is_noslot_pfn);
 964
 965 int is_invalid_pfn(pfn_t pfn)
 966 {
 967         return pfn == hwpoison_pfn || pfn == fault_pfn;
 968 }
 969 EXPORT_SYMBOL_GPL(is_invalid_pfn);
 970
 971 static inline unsigned long bad_hva(void)
 972 {
 973         return PAGE_OFFSET;
 974 }
 975
 976 int kvm_is_error_hva(unsigned long addr)
 977 {
 978         return addr == bad_hva();
 979 }
 980 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 981
 982 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 983 {
 984         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 985 }
 986 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 987
 988 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 989 {
 990         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 991
 992         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
 993               memslot->flags & KVM_MEMSLOT_INVALID)
 994                 return 0;
 995
 996         return 1;
 997 }
 998 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 999
1000 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1001 {
1002         struct vm_area_struct *vma;
1003         unsigned long addr, size;
1004
1005         size = PAGE_SIZE;
1006
1007         addr = gfn_to_hva(kvm, gfn);
1008         if (kvm_is_error_hva(addr))
1009                 return PAGE_SIZE;
1010
1011         down_read(&current->mm->mmap_sem);
1012         vma = find_vma(current->mm, addr);
1013         if (!vma)
1014                 goto out;
1015
1016         size = vma_kernel_pagesize(vma);
1017
1018 out:
1019         up_read(&current->mm->mmap_sem);
1020
1021         return size;
1022 }
1023
1024 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1025                                      gfn_t *nr_pages)
1026 {
1027         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1028                 return bad_hva();
1029
1030         if (nr_pages)
1031                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1032
1033         return gfn_to_hva_memslot(slot, gfn);
1034 }
1035
1036 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1037 {
1038         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1039 }
1040 EXPORT_SYMBOL_GPL(gfn_to_hva);
1041
1042 static pfn_t get_fault_pfn(void)
1043 {
1044         get_page(fault_page);
1045         return fault_pfn;
1046 }
1047
1048 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1049         unsigned long start, int write, struct page **page)
1050 {
1051         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1052
1053         if (write)
1054                 flags |= FOLL_WRITE;
1055
1056         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1057 }
1058
1059 static inline int check_user_page_hwpoison(unsigned long addr)
1060 {
1061         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1062
1063         rc = __get_user_pages(current, current->mm, addr, 1,
1064                               flags, NULL, NULL, NULL);
1065         return rc == -EHWPOISON;
1066 }
1067
1068 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1069                         bool *async, bool write_fault, bool *writable)
1070 {
1071         struct page *page[1];
1072         int npages = 0;
1073         pfn_t pfn;
1074
1075         /* we can do it either atomically or asynchronously, not both */
1076         BUG_ON(atomic && async);
1077
1078         BUG_ON(!write_fault && !writable);
1079
1080         if (writable)
1081                 *writable = true;
1082
1083         if (atomic || async)
1084                 npages = __get_user_pages_fast(addr, 1, 1, page);
1085
1086         if (unlikely(npages != 1) && !atomic) {
1087                 might_sleep();
1088
1089                 if (writable)
1090                         *writable = write_fault;
1091
1092                 if (async) {
1093                         down_read(&current->mm->mmap_sem);
1094                         npages = get_user_page_nowait(current, current->mm,
1095                                                      addr, write_fault, page);
1096                         up_read(&current->mm->mmap_sem);
1097                 } else
1098                         npages = get_user_pages_fast(addr, 1, write_fault,
1099                                                      page);
1100
1101                 /* map read fault as writable if possible */
1102                 if (unlikely(!write_fault) && npages == 1) {
1103                         struct page *wpage[1];
1104
1105                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1106                         if (npages == 1) {
1107                                 *writable = true;
1108                                 put_page(page[0]);
1109                                 page[0] = wpage[0];
1110                         }
1111                         npages = 1;
1112                 }
1113         }
1114
1115         if (unlikely(npages != 1)) {
1116                 struct vm_area_struct *vma;
1117
1118                 if (atomic)
1119                         return get_fault_pfn();
1120
1121                 down_read(&current->mm->mmap_sem);
1122                 if (npages == -EHWPOISON ||
1123                         (!async && check_user_page_hwpoison(addr))) {
1124                         up_read(&current->mm->mmap_sem);
1125                         get_page(hwpoison_page);
1126                         return page_to_pfn(hwpoison_page);
1127                 }
1128
1129                 vma = find_vma_intersection(current->mm, addr, addr+1);
1130
1131                 if (vma == NULL)
1132                         pfn = get_fault_pfn();
1133                 else if ((vma->vm_flags & VM_PFNMAP)) {
1134                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1135                                 vma->vm_pgoff;
1136                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1137                 } else {
1138                         if (async && (vma->vm_flags & VM_WRITE))
1139                                 *async = true;
1140                         pfn = get_fault_pfn();
1141                 }
1142                 up_read(&current->mm->mmap_sem);
1143         } else
1144                 pfn = page_to_pfn(page[0]);
1145
1146         return pfn;
1147 }
1148
1149 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1150 {
1151         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1152 }
1153 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1154
1155 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1156                           bool write_fault, bool *writable)
1157 {
1158         unsigned long addr;
1159
1160         if (async)
1161                 *async = false;
1162
1163         addr = gfn_to_hva(kvm, gfn);
1164         if (kvm_is_error_hva(addr)) {
1165                 get_page(bad_page);
1166                 return page_to_pfn(bad_page);
1167         }
1168
1169         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1170 }
1171
1172 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1173 {
1174         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1175 }
1176 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1177
1178 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1179                        bool write_fault, bool *writable)
1180 {
1181         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1182 }
1183 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1184
1185 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1186 {
1187         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1188 }
1189 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1190
1191 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1192                       bool *writable)
1193 {
1194         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1195 }
1196 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1197
1198 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1199                          struct kvm_memory_slot *slot, gfn_t gfn)
1200 {
1201         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1202         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1203 }
1204
1205 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1206                                                                   int nr_pages)
1207 {
1208         unsigned long addr;
1209         gfn_t entry;
1210
1211         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1212         if (kvm_is_error_hva(addr))
1213                 return -1;
1214
1215         if (entry < nr_pages)
1216                 return 0;
1217
1218         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1219 }
1220 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1221
1222 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1223 {
1224         pfn_t pfn;
1225
1226         pfn = gfn_to_pfn(kvm, gfn);
1227         if (!kvm_is_mmio_pfn(pfn))
1228                 return pfn_to_page(pfn);
1229
1230         WARN_ON(kvm_is_mmio_pfn(pfn));
1231
1232         get_page(bad_page);
1233         return bad_page;
1234 }
1235
1236 EXPORT_SYMBOL_GPL(gfn_to_page);
1237
1238 void kvm_release_page_clean(struct page *page)
1239 {
1240         kvm_release_pfn_clean(page_to_pfn(page));
1241 }
1242 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1243
1244 void kvm_release_pfn_clean(pfn_t pfn)
1245 {
1246         if (!kvm_is_mmio_pfn(pfn))
1247                 put_page(pfn_to_page(pfn));
1248 }
1249 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1250
1251 void kvm_release_page_dirty(struct page *page)
1252 {
1253         kvm_release_pfn_dirty(page_to_pfn(page));
1254 }
1255 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1256
1257 void kvm_release_pfn_dirty(pfn_t pfn)
1258 {
1259         kvm_set_pfn_dirty(pfn);
1260         kvm_release_pfn_clean(pfn);
1261 }
1262 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1263
1264 void kvm_set_page_dirty(struct page *page)
1265 {
1266         kvm_set_pfn_dirty(page_to_pfn(page));
1267 }
1268 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1269
1270 void kvm_set_pfn_dirty(pfn_t pfn)
1271 {
1272         if (!kvm_is_mmio_pfn(pfn)) {
1273                 struct page *page = pfn_to_page(pfn);
1274                 if (!PageReserved(page))
1275                         SetPageDirty(page);
1276         }
1277 }
1278 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1279
1280 void kvm_set_pfn_accessed(pfn_t pfn)
1281 {
1282         if (!kvm_is_mmio_pfn(pfn))
1283                 mark_page_accessed(pfn_to_page(pfn));
1284 }
1285 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1286
1287 void kvm_get_pfn(pfn_t pfn)
1288 {
1289         if (!kvm_is_mmio_pfn(pfn))
1290                 get_page(pfn_to_page(pfn));
1291 }
1292 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1293
1294 static int next_segment(unsigned long len, int offset)
1295 {
1296         if (len > PAGE_SIZE - offset)
1297                 return PAGE_SIZE - offset;
1298         else
1299                 return len;
1300 }
1301
1302 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1303                         int len)
1304 {
1305         int r;
1306         unsigned long addr;
1307
1308         addr = gfn_to_hva(kvm, gfn);
1309         if (kvm_is_error_hva(addr))
1310                 return -EFAULT;
1311         r = __copy_from_user(data, (void __user *)addr + offset, len);
1312         if (r)
1313                 return -EFAULT;
1314         return 0;
1315 }
1316 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1317
1318 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1319 {
1320         gfn_t gfn = gpa >> PAGE_SHIFT;
1321         int seg;
1322         int offset = offset_in_page(gpa);
1323         int ret;
1324
1325         while ((seg = next_segment(len, offset)) != 0) {
1326                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1327                 if (ret < 0)
1328                         return ret;
1329                 offset = 0;
1330                 len -= seg;
1331                 data += seg;
1332                 ++gfn;
1333         }
1334         return 0;
1335 }
1336 EXPORT_SYMBOL_GPL(kvm_read_guest);
1337
1338 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1339                           unsigned long len)
1340 {
1341         int r;
1342         unsigned long addr;
1343         gfn_t gfn = gpa >> PAGE_SHIFT;
1344         int offset = offset_in_page(gpa);
1345
1346         addr = gfn_to_hva(kvm, gfn);
1347         if (kvm_is_error_hva(addr))
1348                 return -EFAULT;
1349         pagefault_disable();
1350         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1351         pagefault_enable();
1352         if (r)
1353                 return -EFAULT;
1354         return 0;
1355 }
1356 EXPORT_SYMBOL(kvm_read_guest_atomic);
1357
1358 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1359                          int offset, int len)
1360 {
1361         int r;
1362         unsigned long addr;
1363
1364         addr = gfn_to_hva(kvm, gfn);
1365         if (kvm_is_error_hva(addr))
1366                 return -EFAULT;
1367         r = __copy_to_user((void __user *)addr + offset, data, len);
1368         if (r)
1369                 return -EFAULT;
1370         mark_page_dirty(kvm, gfn);
1371         return 0;
1372 }
1373 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1374
1375 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1376                     unsigned long len)
1377 {
1378         gfn_t gfn = gpa >> PAGE_SHIFT;
1379         int seg;
1380         int offset = offset_in_page(gpa);
1381         int ret;
1382
1383         while ((seg = next_segment(len, offset)) != 0) {
1384                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1385                 if (ret < 0)
1386                         return ret;
1387                 offset = 0;
1388                 len -= seg;
1389                 data += seg;
1390                 ++gfn;
1391         }
1392         return 0;
1393 }
1394
1395 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1396                               gpa_t gpa)
1397 {
1398         struct kvm_memslots *slots = kvm_memslots(kvm);
1399         int offset = offset_in_page(gpa);
1400         gfn_t gfn = gpa >> PAGE_SHIFT;
1401
1402         ghc->gpa = gpa;
1403         ghc->generation = slots->generation;
1404         ghc->memslot = gfn_to_memslot(kvm, gfn);
1405         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1406         if (!kvm_is_error_hva(ghc->hva))
1407                 ghc->hva += offset;
1408         else
1409                 return -EFAULT;
1410
1411         return 0;
1412 }
1413 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1414
1415 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1416                            void *data, unsigned long len)
1417 {
1418         struct kvm_memslots *slots = kvm_memslots(kvm);
1419         int r;
1420
1421         if (slots->generation != ghc->generation)
1422                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1423
1424         if (kvm_is_error_hva(ghc->hva))
1425                 return -EFAULT;
1426
1427         r = __copy_to_user((void __user *)ghc->hva, data, len);
1428         if (r)
1429                 return -EFAULT;
1430         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1431
1432         return 0;
1433 }
1434 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1435
1436 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1437                            void *data, unsigned long len)
1438 {
1439         struct kvm_memslots *slots = kvm_memslots(kvm);
1440         int r;
1441
1442         if (slots->generation != ghc->generation)
1443                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1444
1445         if (kvm_is_error_hva(ghc->hva))
1446                 return -EFAULT;
1447
1448         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1449         if (r)
1450                 return -EFAULT;
1451
1452         return 0;
1453 }
1454 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1455
1456 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1457 {
1458         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1459                                     offset, len);
1460 }
1461 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1462
1463 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1464 {
1465         gfn_t gfn = gpa >> PAGE_SHIFT;
1466         int seg;
1467         int offset = offset_in_page(gpa);
1468         int ret;
1469
1470         while ((seg = next_segment(len, offset)) != 0) {
1471                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1472                 if (ret < 0)
1473                         return ret;
1474                 offset = 0;
1475                 len -= seg;
1476                 ++gfn;
1477         }
1478         return 0;
1479 }
1480 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1481
1482 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1483                              gfn_t gfn)
1484 {
1485         if (memslot && memslot->dirty_bitmap) {
1486                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1487
1488                 /* TODO: introduce set_bit_le() and use it */
1489                 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1490         }
1491 }
1492
1493 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1494 {
1495         struct kvm_memory_slot *memslot;
1496
1497         memslot = gfn_to_memslot(kvm, gfn);
1498         mark_page_dirty_in_slot(kvm, memslot, gfn);
1499 }
1500
1501 /*
1502  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1503  */
1504 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1505 {
1506         DEFINE_WAIT(wait);
1507
1508         for (;;) {
1509                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1510
1511                 if (kvm_arch_vcpu_runnable(vcpu)) {
1512                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1513                         break;
1514                 }
1515                 if (kvm_cpu_has_pending_timer(vcpu))
1516                         break;
1517                 if (signal_pending(current))
1518                         break;
1519
1520                 schedule();
1521         }
1522
1523         finish_wait(&vcpu->wq, &wait);
1524 }
1525
1526 #ifndef CONFIG_S390
1527 /*
1528  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1529  */
1530 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1531 {
1532         int me;
1533         int cpu = vcpu->cpu;
1534         wait_queue_head_t *wqp;
1535
1536         wqp = kvm_arch_vcpu_wq(vcpu);
1537         if (waitqueue_active(wqp)) {
1538                 wake_up_interruptible(wqp);
1539                 ++vcpu->stat.halt_wakeup;
1540         }
1541
1542         me = get_cpu();
1543         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1544                 if (kvm_arch_vcpu_should_kick(vcpu))
1545                         smp_send_reschedule(cpu);
1546         put_cpu();
1547 }
1548 #endif /* !CONFIG_S390 */
1549
1550 void kvm_resched(struct kvm_vcpu *vcpu)
1551 {
1552         if (!need_resched())
1553                 return;
1554         cond_resched();
1555 }
1556 EXPORT_SYMBOL_GPL(kvm_resched);
1557
1558 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1559 {
1560         struct pid *pid;
1561         struct task_struct *task = NULL;
1562
1563         rcu_read_lock();
1564         pid = rcu_dereference(target->pid);
1565         if (pid)
1566                 task = get_pid_task(target->pid, PIDTYPE_PID);
1567         rcu_read_unlock();
1568         if (!task)
1569                 return false;
1570         if (task->flags & PF_VCPU) {
1571                 put_task_struct(task);
1572                 return false;
1573         }
1574         if (yield_to(task, 1)) {
1575                 put_task_struct(task);
1576                 return true;
1577         }
1578         put_task_struct(task);
1579         return false;
1580 }
1581 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1582
1583 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1584 {
1585         struct kvm *kvm = me->kvm;
1586         struct kvm_vcpu *vcpu;
1587         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1588         int yielded = 0;
1589         int pass;
1590         int i;
1591
1592         /*
1593          * We boost the priority of a VCPU that is runnable but not
1594          * currently running, because it got preempted by something
1595          * else and called schedule in __vcpu_run.  Hopefully that
1596          * VCPU is holding the lock that we need and will release it.
1597          * We approximate round-robin by starting at the last boosted VCPU.
1598          */
1599         for (pass = 0; pass < 2 && !yielded; pass++) {
1600                 kvm_for_each_vcpu(i, vcpu, kvm) {
1601                         if (!pass && i <= last_boosted_vcpu) {
1602                                 i = last_boosted_vcpu;
1603                                 continue;
1604                         } else if (pass && i > last_boosted_vcpu)
1605                                 break;
1606                         if (vcpu == me)
1607                                 continue;
1608                         if (waitqueue_active(&vcpu->wq))
1609                                 continue;
1610                         if (kvm_vcpu_yield_to(vcpu)) {
1611                                 kvm->last_boosted_vcpu = i;
1612                                 yielded = 1;
1613                                 break;
1614                         }
1615                 }
1616         }
1617 }
1618 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1619
1620 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1621 {
1622         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1623         struct page *page;
1624
1625         if (vmf->pgoff == 0)
1626                 page = virt_to_page(vcpu->run);
1627 #ifdef CONFIG_X86
1628         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1629                 page = virt_to_page(vcpu->arch.pio_data);
1630 #endif
1631 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1632         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1633                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1634 #endif
1635         else
1636                 return kvm_arch_vcpu_fault(vcpu, vmf);
1637         get_page(page);
1638         vmf->page = page;
1639         return 0;
1640 }
1641
1642 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1643         .fault = kvm_vcpu_fault,
1644 };
1645
1646 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1647 {
1648         vma->vm_ops = &kvm_vcpu_vm_ops;
1649         return 0;
1650 }
1651
1652 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1653 {
1654         struct kvm_vcpu *vcpu = filp->private_data;
1655
1656         kvm_put_kvm(vcpu->kvm);
1657         return 0;
1658 }
1659
1660 static struct file_operations kvm_vcpu_fops = {
1661         .release        = kvm_vcpu_release,
1662         .unlocked_ioctl = kvm_vcpu_ioctl,
1663 #ifdef CONFIG_COMPAT
1664         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1665 #endif
1666         .mmap           = kvm_vcpu_mmap,
1667         .llseek         = noop_llseek,
1668 };
1669
1670 /*
1671  * Allocates an inode for the vcpu.
1672  */
1673 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1674 {
1675         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1676 }
1677
1678 /*
1679  * Creates some virtual cpus.  Good luck creating more than one.
1680  */
1681 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1682 {
1683         int r;
1684         struct kvm_vcpu *vcpu, *v;
1685
1686         vcpu = kvm_arch_vcpu_create(kvm, id);
1687         if (IS_ERR(vcpu))
1688                 return PTR_ERR(vcpu);
1689
1690         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1691
1692         r = kvm_arch_vcpu_setup(vcpu);
1693         if (r)
1694                 goto vcpu_destroy;
1695
1696         mutex_lock(&kvm->lock);
1697         if (!kvm_vcpu_compatible(vcpu)) {
1698                 r = -EINVAL;
1699                 goto unlock_vcpu_destroy;
1700         }
1701         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1702                 r = -EINVAL;
1703                 goto unlock_vcpu_destroy;
1704         }
1705
1706         kvm_for_each_vcpu(r, v, kvm)
1707                 if (v->vcpu_id == id) {
1708                         r = -EEXIST;
1709                         goto unlock_vcpu_destroy;
1710                 }
1711
1712         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1713
1714         /* Now it's all set up, let userspace reach it */
1715         kvm_get_kvm(kvm);
1716         r = create_vcpu_fd(vcpu);
1717         if (r < 0) {
1718                 kvm_put_kvm(kvm);
1719                 goto unlock_vcpu_destroy;
1720         }
1721
1722         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1723         smp_wmb();
1724         atomic_inc(&kvm->online_vcpus);
1725
1726         mutex_unlock(&kvm->lock);
1727         return r;
1728
1729 unlock_vcpu_destroy:
1730         mutex_unlock(&kvm->lock);
1731 vcpu_destroy:
1732         kvm_arch_vcpu_destroy(vcpu);
1733         return r;
1734 }
1735
1736 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1737 {
1738         if (sigset) {
1739                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1740                 vcpu->sigset_active = 1;
1741                 vcpu->sigset = *sigset;
1742         } else
1743                 vcpu->sigset_active = 0;
1744         return 0;
1745 }
1746
1747 static long kvm_vcpu_ioctl(struct file *filp,
1748                            unsigned int ioctl, unsigned long arg)
1749 {
1750         struct kvm_vcpu *vcpu = filp->private_data;
1751         void __user *argp = (void __user *)arg;
1752         int r;
1753         struct kvm_fpu *fpu = NULL;
1754         struct kvm_sregs *kvm_sregs = NULL;
1755
1756         if (vcpu->kvm->mm != current->mm)
1757                 return -EIO;
1758
1759 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1760         /*
1761          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1762          * so vcpu_load() would break it.
1763          */
1764         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1765                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1766 #endif
1767
1768
1769         vcpu_load(vcpu);
1770         switch (ioctl) {
1771         case KVM_RUN:
1772                 r = -EINVAL;
1773                 if (arg)
1774                         goto out;
1775                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1776                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1777                 break;
1778         case KVM_GET_REGS: {
1779                 struct kvm_regs *kvm_regs;
1780
1781                 r = -ENOMEM;
1782                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1783                 if (!kvm_regs)
1784                         goto out;
1785                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1786                 if (r)
1787                         goto out_free1;
1788                 r = -EFAULT;
1789                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1790                         goto out_free1;
1791                 r = 0;
1792 out_free1:
1793                 kfree(kvm_regs);
1794                 break;
1795         }
1796         case KVM_SET_REGS: {
1797                 struct kvm_regs *kvm_regs;
1798
1799                 r = -ENOMEM;
1800                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1801                 if (IS_ERR(kvm_regs)) {
1802                         r = PTR_ERR(kvm_regs);
1803                         goto out;
1804                 }
1805                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1806                 if (r)
1807                         goto out_free2;
1808                 r = 0;
1809 out_free2:
1810                 kfree(kvm_regs);
1811                 break;
1812         }
1813         case KVM_GET_SREGS: {
1814                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1815                 r = -ENOMEM;
1816                 if (!kvm_sregs)
1817                         goto out;
1818                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1819                 if (r)
1820                         goto out;
1821                 r = -EFAULT;
1822                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1823                         goto out;
1824                 r = 0;
1825                 break;
1826         }
1827         case KVM_SET_SREGS: {
1828                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1829                 if (IS_ERR(kvm_sregs)) {
1830                         r = PTR_ERR(kvm_sregs);
1831                         goto out;
1832                 }
1833                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1834                 if (r)
1835                         goto out;
1836                 r = 0;
1837                 break;
1838         }
1839         case KVM_GET_MP_STATE: {
1840                 struct kvm_mp_state mp_state;
1841
1842                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1843                 if (r)
1844                         goto out;
1845                 r = -EFAULT;
1846                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1847                         goto out;
1848                 r = 0;
1849                 break;
1850         }
1851         case KVM_SET_MP_STATE: {
1852                 struct kvm_mp_state mp_state;
1853
1854                 r = -EFAULT;
1855                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1856                         goto out;
1857                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1858                 if (r)
1859                         goto out;
1860                 r = 0;
1861                 break;
1862         }
1863         case KVM_TRANSLATE: {
1864                 struct kvm_translation tr;
1865
1866                 r = -EFAULT;
1867                 if (copy_from_user(&tr, argp, sizeof tr))
1868                         goto out;
1869                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1870                 if (r)
1871                         goto out;
1872                 r = -EFAULT;
1873                 if (copy_to_user(argp, &tr, sizeof tr))
1874                         goto out;
1875                 r = 0;
1876                 break;
1877         }
1878         case KVM_SET_GUEST_DEBUG: {
1879                 struct kvm_guest_debug dbg;
1880
1881                 r = -EFAULT;
1882                 if (copy_from_user(&dbg, argp, sizeof dbg))
1883                         goto out;
1884                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1885                 if (r)
1886                         goto out;
1887                 r = 0;
1888                 break;
1889         }
1890         case KVM_SET_SIGNAL_MASK: {
1891                 struct kvm_signal_mask __user *sigmask_arg = argp;
1892                 struct kvm_signal_mask kvm_sigmask;
1893                 sigset_t sigset, *p;
1894
1895                 p = NULL;
1896                 if (argp) {
1897                         r = -EFAULT;
1898                         if (copy_from_user(&kvm_sigmask, argp,
1899                                            sizeof kvm_sigmask))
1900                                 goto out;
1901                         r = -EINVAL;
1902                         if (kvm_sigmask.len != sizeof sigset)
1903                                 goto out;
1904                         r = -EFAULT;
1905                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1906                                            sizeof sigset))
1907                                 goto out;
1908                         p = &sigset;
1909                 }
1910                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1911                 break;
1912         }
1913         case KVM_GET_FPU: {
1914                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1915                 r = -ENOMEM;
1916                 if (!fpu)
1917                         goto out;
1918                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1919                 if (r)
1920                         goto out;
1921                 r = -EFAULT;
1922                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1923                         goto out;
1924                 r = 0;
1925                 break;
1926         }
1927         case KVM_SET_FPU: {
1928                 fpu = memdup_user(argp, sizeof(*fpu));
1929                 if (IS_ERR(fpu)) {
1930                         r = PTR_ERR(fpu);
1931                         goto out;
1932                 }
1933                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1934                 if (r)
1935                         goto out;
1936                 r = 0;
1937                 break;
1938         }
1939         default:
1940                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1941         }
1942 out:
1943         vcpu_put(vcpu);
1944         kfree(fpu);
1945         kfree(kvm_sregs);
1946         return r;
1947 }
1948
1949 #ifdef CONFIG_COMPAT
1950 static long kvm_vcpu_compat_ioctl(struct file *filp,
1951                                   unsigned int ioctl, unsigned long arg)
1952 {
1953         struct kvm_vcpu *vcpu = filp->private_data;
1954         void __user *argp = compat_ptr(arg);
1955         int r;
1956
1957         if (vcpu->kvm->mm != current->mm)
1958                 return -EIO;
1959
1960         switch (ioctl) {
1961         case KVM_SET_SIGNAL_MASK: {
1962                 struct kvm_signal_mask __user *sigmask_arg = argp;
1963                 struct kvm_signal_mask kvm_sigmask;
1964                 compat_sigset_t csigset;
1965                 sigset_t sigset;
1966
1967                 if (argp) {
1968                         r = -EFAULT;
1969                         if (copy_from_user(&kvm_sigmask, argp,
1970                                            sizeof kvm_sigmask))
1971                                 goto out;
1972                         r = -EINVAL;
1973                         if (kvm_sigmask.len != sizeof csigset)
1974                                 goto out;
1975                         r = -EFAULT;
1976                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1977                                            sizeof csigset))
1978                                 goto out;
1979                 }
1980                 sigset_from_compat(&sigset, &csigset);
1981                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1982                 break;
1983         }
1984         default:
1985                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1986         }
1987
1988 out:
1989         return r;
1990 }
1991 #endif
1992
1993 static long kvm_vm_ioctl(struct file *filp,
1994                            unsigned int ioctl, unsigned long arg)
1995 {
1996         struct kvm *kvm = filp->private_data;
1997         void __user *argp = (void __user *)arg;
1998         int r;
1999
2000         if (kvm->mm != current->mm)
2001                 return -EIO;
2002         switch (ioctl) {
2003         case KVM_CREATE_VCPU:
2004                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2005                 if (r < 0)
2006                         goto out;
2007                 break;
2008         case KVM_SET_USER_MEMORY_REGION: {
2009                 struct kvm_userspace_memory_region kvm_userspace_mem;
2010
2011                 r = -EFAULT;
2012                 if (copy_from_user(&kvm_userspace_mem, argp,
2013                                                 sizeof kvm_userspace_mem))
2014                         goto out;
2015
2016                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2017                 if (r)
2018                         goto out;
2019                 break;
2020         }
2021         case KVM_GET_DIRTY_LOG: {
2022                 struct kvm_dirty_log log;
2023
2024                 r = -EFAULT;
2025                 if (copy_from_user(&log, argp, sizeof log))
2026                         goto out;
2027                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2028                 if (r)
2029                         goto out;
2030                 break;
2031         }
2032 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2033         case KVM_REGISTER_COALESCED_MMIO: {
2034                 struct kvm_coalesced_mmio_zone zone;
2035                 r = -EFAULT;
2036                 if (copy_from_user(&zone, argp, sizeof zone))
2037                         goto out;
2038                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2039                 if (r)
2040                         goto out;
2041                 r = 0;
2042                 break;
2043         }
2044         case KVM_UNREGISTER_COALESCED_MMIO: {
2045                 struct kvm_coalesced_mmio_zone zone;
2046                 r = -EFAULT;
2047                 if (copy_from_user(&zone, argp, sizeof zone))
2048                         goto out;
2049                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2050                 if (r)
2051                         goto out;
2052                 r = 0;
2053                 break;
2054         }
2055 #endif
2056         case KVM_IRQFD: {
2057                 struct kvm_irqfd data;
2058
2059                 r = -EFAULT;
2060                 if (copy_from_user(&data, argp, sizeof data))
2061                         goto out;
2062                 r = kvm_irqfd(kvm, &data);
2063                 break;
2064         }
2065         case KVM_IOEVENTFD: {
2066                 struct kvm_ioeventfd data;
2067
2068                 r = -EFAULT;
2069                 if (copy_from_user(&data, argp, sizeof data))
2070                         goto out;
2071                 r = kvm_ioeventfd(kvm, &data);
2072                 break;
2073         }
2074 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2075         case KVM_SET_BOOT_CPU_ID:
2076                 r = 0;
2077                 mutex_lock(&kvm->lock);
2078                 if (atomic_read(&kvm->online_vcpus) != 0)
2079                         r = -EBUSY;
2080                 else
2081                         kvm->bsp_vcpu_id = arg;
2082                 mutex_unlock(&kvm->lock);
2083                 break;
2084 #endif
2085 #ifdef CONFIG_HAVE_KVM_MSI
2086         case KVM_SIGNAL_MSI: {
2087                 struct kvm_msi msi;
2088
2089                 r = -EFAULT;
2090                 if (copy_from_user(&msi, argp, sizeof msi))
2091                         goto out;
2092                 r = kvm_send_userspace_msi(kvm, &msi);
2093                 break;
2094         }
2095 #endif
2096         default:
2097                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2098                 if (r == -ENOTTY)
2099                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2100         }
2101 out:
2102         return r;
2103 }
2104
2105 #ifdef CONFIG_COMPAT
2106 struct compat_kvm_dirty_log {
2107         __u32 slot;
2108         __u32 padding1;
2109         union {
2110                 compat_uptr_t dirty_bitmap; /* one bit per page */
2111                 __u64 padding2;
2112         };
2113 };
2114
2115 static long kvm_vm_compat_ioctl(struct file *filp,
2116                            unsigned int ioctl, unsigned long arg)
2117 {
2118         struct kvm *kvm = filp->private_data;
2119         int r;
2120
2121         if (kvm->mm != current->mm)
2122                 return -EIO;
2123         switch (ioctl) {
2124         case KVM_GET_DIRTY_LOG: {
2125                 struct compat_kvm_dirty_log compat_log;
2126                 struct kvm_dirty_log log;
2127
2128                 r = -EFAULT;
2129                 if (copy_from_user(&compat_log, (void __user *)arg,
2130                                    sizeof(compat_log)))
2131                         goto out;
2132                 log.slot         = compat_log.slot;
2133                 log.padding1     = compat_log.padding1;
2134                 log.padding2     = compat_log.padding2;
2135                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2136
2137                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2138                 if (r)
2139                         goto out;
2140                 break;
2141         }
2142         default:
2143                 r = kvm_vm_ioctl(filp, ioctl, arg);
2144         }
2145
2146 out:
2147         return r;
2148 }
2149 #endif
2150
2151 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2152 {
2153         struct page *page[1];
2154         unsigned long addr;
2155         int npages;
2156         gfn_t gfn = vmf->pgoff;
2157         struct kvm *kvm = vma->vm_file->private_data;
2158
2159         addr = gfn_to_hva(kvm, gfn);
2160         if (kvm_is_error_hva(addr))
2161                 return VM_FAULT_SIGBUS;
2162
2163         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2164                                 NULL);
2165         if (unlikely(npages != 1))
2166                 return VM_FAULT_SIGBUS;
2167
2168         vmf->page = page[0];
2169         return 0;
2170 }
2171
2172 static const struct vm_operations_struct kvm_vm_vm_ops = {
2173         .fault = kvm_vm_fault,
2174 };
2175
2176 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2177 {
2178         vma->vm_ops = &kvm_vm_vm_ops;
2179         return 0;
2180 }
2181
2182 static struct file_operations kvm_vm_fops = {
2183         .release        = kvm_vm_release,
2184         .unlocked_ioctl = kvm_vm_ioctl,
2185 #ifdef CONFIG_COMPAT
2186         .compat_ioctl   = kvm_vm_compat_ioctl,
2187 #endif
2188         .mmap           = kvm_vm_mmap,
2189         .llseek         = noop_llseek,
2190 };
2191
2192 static int kvm_dev_ioctl_create_vm(unsigned long type)
2193 {
2194         int r;
2195         struct kvm *kvm;
2196
2197         kvm = kvm_create_vm(type);
2198         if (IS_ERR(kvm))
2199                 return PTR_ERR(kvm);
2200 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2201         r = kvm_coalesced_mmio_init(kvm);
2202         if (r < 0) {
2203                 kvm_put_kvm(kvm);
2204                 return r;
2205         }
2206 #endif
2207         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2208         if (r < 0)
2209                 kvm_put_kvm(kvm);
2210
2211         return r;
2212 }
2213
2214 static long kvm_dev_ioctl_check_extension_generic(long arg)
2215 {
2216         switch (arg) {
2217         case KVM_CAP_USER_MEMORY:
2218         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2219         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2220 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2221         case KVM_CAP_SET_BOOT_CPU_ID:
2222 #endif
2223         case KVM_CAP_INTERNAL_ERROR_DATA:
2224 #ifdef CONFIG_HAVE_KVM_MSI
2225         case KVM_CAP_SIGNAL_MSI:
2226 #endif
2227                 return 1;
2228 #ifdef KVM_CAP_IRQ_ROUTING
2229         case KVM_CAP_IRQ_ROUTING:
2230                 return KVM_MAX_IRQ_ROUTES;
2231 #endif
2232         default:
2233                 break;
2234         }
2235         return kvm_dev_ioctl_check_extension(arg);
2236 }
2237
2238 static long kvm_dev_ioctl(struct file *filp,
2239                           unsigned int ioctl, unsigned long arg)
2240 {
2241         long r = -EINVAL;
2242
2243         switch (ioctl) {
2244         case KVM_GET_API_VERSION:
2245                 r = -EINVAL;
2246                 if (arg)
2247                         goto out;
2248                 r = KVM_API_VERSION;
2249                 break;
2250         case KVM_CREATE_VM:
2251                 r = kvm_dev_ioctl_create_vm(arg);
2252                 break;
2253         case KVM_CHECK_EXTENSION:
2254                 r = kvm_dev_ioctl_check_extension_generic(arg);
2255                 break;
2256         case KVM_GET_VCPU_MMAP_SIZE:
2257                 r = -EINVAL;
2258                 if (arg)
2259                         goto out;
2260                 r = PAGE_SIZE;     /* struct kvm_run */
2261 #ifdef CONFIG_X86
2262                 r += PAGE_SIZE;    /* pio data page */
2263 #endif
2264 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2265                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2266 #endif
2267                 break;
2268         case KVM_TRACE_ENABLE:
2269         case KVM_TRACE_PAUSE:
2270         case KVM_TRACE_DISABLE:
2271                 r = -EOPNOTSUPP;
2272                 break;
2273         default:
2274                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2275         }
2276 out:
2277         return r;
2278 }
2279
2280 static struct file_operations kvm_chardev_ops = {
2281         .unlocked_ioctl = kvm_dev_ioctl,
2282         .compat_ioctl   = kvm_dev_ioctl,
2283         .llseek         = noop_llseek,
2284 };
2285
2286 static struct miscdevice kvm_dev = {
2287         KVM_MINOR,
2288         "kvm",
2289         &kvm_chardev_ops,
2290 };
2291
2292 static void hardware_enable_nolock(void *junk)
2293 {
2294         int cpu = raw_smp_processor_id();
2295         int r;
2296
2297         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2298                 return;
2299
2300         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2301
2302         r = kvm_arch_hardware_enable(NULL);
2303
2304         if (r) {
2305                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2306                 atomic_inc(&hardware_enable_failed);
2307                 printk(KERN_INFO "kvm: enabling virtualization on "
2308                                  "CPU%d failed\n", cpu);
2309         }
2310 }
2311
2312 static void hardware_enable(void *junk)
2313 {
2314         raw_spin_lock(&kvm_lock);
2315         hardware_enable_nolock(junk);
2316         raw_spin_unlock(&kvm_lock);
2317 }
2318
2319 static void hardware_disable_nolock(void *junk)
2320 {
2321         int cpu = raw_smp_processor_id();
2322
2323         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2324                 return;
2325         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2326         kvm_arch_hardware_disable(NULL);
2327 }
2328
2329 static void hardware_disable(void *junk)
2330 {
2331         raw_spin_lock(&kvm_lock);
2332         hardware_disable_nolock(junk);
2333         raw_spin_unlock(&kvm_lock);
2334 }
2335
2336 static void hardware_disable_all_nolock(void)
2337 {
2338         BUG_ON(!kvm_usage_count);
2339
2340         kvm_usage_count--;
2341         if (!kvm_usage_count)
2342                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2343 }
2344
2345 static void hardware_disable_all(void)
2346 {
2347         raw_spin_lock(&kvm_lock);
2348         hardware_disable_all_nolock();
2349         raw_spin_unlock(&kvm_lock);
2350 }
2351
2352 static int hardware_enable_all(void)
2353 {
2354         int r = 0;
2355
2356         raw_spin_lock(&kvm_lock);
2357
2358         kvm_usage_count++;
2359         if (kvm_usage_count == 1) {
2360                 atomic_set(&hardware_enable_failed, 0);
2361                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2362
2363                 if (atomic_read(&hardware_enable_failed)) {
2364                         hardware_disable_all_nolock();
2365                         r = -EBUSY;
2366                 }
2367         }
2368
2369         raw_spin_unlock(&kvm_lock);
2370
2371         return r;
2372 }
2373
2374 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2375                            void *v)
2376 {
2377         int cpu = (long)v;
2378
2379         if (!kvm_usage_count)
2380                 return NOTIFY_OK;
2381
2382         val &= ~CPU_TASKS_FROZEN;
2383         switch (val) {
2384         case CPU_DYING:
2385                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2386                        cpu);
2387                 hardware_disable(NULL);
2388                 break;
2389         case CPU_STARTING:
2390                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2391                        cpu);
2392                 hardware_enable(NULL);
2393                 break;
2394         }
2395         return NOTIFY_OK;
2396 }
2397
2398
2399 asmlinkage void kvm_spurious_fault(void)
2400 {
2401         /* Fault while not rebooting.  We want the trace. */
2402         BUG();
2403 }
2404 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2405
2406 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2407                       void *v)
2408 {
2409         /*
2410          * Some (well, at least mine) BIOSes hang on reboot if
2411          * in vmx root mode.
2412          *
2413          * And Intel TXT required VMX off for all cpu when system shutdown.
2414          */
2415         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2416         kvm_rebooting = true;
2417         on_each_cpu(hardware_disable_nolock, NULL, 1);
2418         return NOTIFY_OK;
2419 }
2420
2421 static struct notifier_block kvm_reboot_notifier = {
2422         .notifier_call = kvm_reboot,
2423         .priority = 0,
2424 };
2425
2426 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2427 {
2428         int i;
2429
2430         for (i = 0; i < bus->dev_count; i++) {
2431                 struct kvm_io_device *pos = bus->range[i].dev;
2432
2433                 kvm_iodevice_destructor(pos);
2434         }
2435         kfree(bus);
2436 }
2437
2438 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2439 {
2440         const struct kvm_io_range *r1 = p1;
2441         const struct kvm_io_range *r2 = p2;
2442
2443         if (r1->addr < r2->addr)
2444                 return -1;
2445         if (r1->addr + r1->len > r2->addr + r2->len)
2446                 return 1;
2447         return 0;
2448 }
2449
2450 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2451                           gpa_t addr, int len)
2452 {
2453         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2454                 .addr = addr,
2455                 .len = len,
2456                 .dev = dev,
2457         };
2458
2459         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2460                 kvm_io_bus_sort_cmp, NULL);
2461
2462         return 0;
2463 }
2464
2465 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2466                              gpa_t addr, int len)
2467 {
2468         struct kvm_io_range *range, key;
2469         int off;
2470
2471         key = (struct kvm_io_range) {
2472                 .addr = addr,
2473                 .len = len,
2474         };
2475
2476         range = bsearch(&key, bus->range, bus->dev_count,
2477                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2478         if (range == NULL)
2479                 return -ENOENT;
2480
2481         off = range - bus->range;
2482
2483         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2484                 off--;
2485
2486         return off;
2487 }
2488
2489 /* kvm_io_bus_write - called under kvm->slots_lock */
2490 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2491                      int len, const void *val)
2492 {
2493         int idx;
2494         struct kvm_io_bus *bus;
2495         struct kvm_io_range range;
2496
2497         range = (struct kvm_io_range) {
2498                 .addr = addr,
2499                 .len = len,
2500         };
2501
2502         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2503         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2504         if (idx < 0)
2505                 return -EOPNOTSUPP;
2506
2507         while (idx < bus->dev_count &&
2508                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2509                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2510                         return 0;
2511                 idx++;
2512         }
2513
2514         return -EOPNOTSUPP;
2515 }
2516
2517 /* kvm_io_bus_read - called under kvm->slots_lock */
2518 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2519                     int len, void *val)
2520 {
2521         int idx;
2522         struct kvm_io_bus *bus;
2523         struct kvm_io_range range;
2524
2525         range = (struct kvm_io_range) {
2526                 .addr = addr,
2527                 .len = len,
2528         };
2529
2530         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2531         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2532         if (idx < 0)
2533                 return -EOPNOTSUPP;
2534
2535         while (idx < bus->dev_count &&
2536                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2537                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2538                         return 0;
2539                 idx++;
2540         }
2541
2542         return -EOPNOTSUPP;
2543 }
2544
2545 /* Caller must hold slots_lock. */
2546 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2547                             int len, struct kvm_io_device *dev)
2548 {
2549         struct kvm_io_bus *new_bus, *bus;
2550
2551         bus = kvm->buses[bus_idx];
2552         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2553                 return -ENOSPC;
2554
2555         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2556                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2557         if (!new_bus)
2558                 return -ENOMEM;
2559         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2560                sizeof(struct kvm_io_range)));
2561         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2562         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2563         synchronize_srcu_expedited(&kvm->srcu);
2564         kfree(bus);
2565
2566         return 0;
2567 }
2568
2569 /* Caller must hold slots_lock. */
2570 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2571                               struct kvm_io_device *dev)
2572 {
2573         int i, r;
2574         struct kvm_io_bus *new_bus, *bus;
2575
2576         bus = kvm->buses[bus_idx];
2577         r = -ENOENT;
2578         for (i = 0; i < bus->dev_count; i++)
2579                 if (bus->range[i].dev == dev) {
2580                         r = 0;
2581                         break;
2582                 }
2583
2584         if (r)
2585                 return r;
2586
2587         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2588                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2589         if (!new_bus)
2590                 return -ENOMEM;
2591
2592         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2593         new_bus->dev_count--;
2594         memcpy(new_bus->range + i, bus->range + i + 1,
2595                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2596
2597         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2598         synchronize_srcu_expedited(&kvm->srcu);
2599         kfree(bus);
2600         return r;
2601 }
2602
2603 static struct notifier_block kvm_cpu_notifier = {
2604         .notifier_call = kvm_cpu_hotplug,
2605 };
2606
2607 static int vm_stat_get(void *_offset, u64 *val)
2608 {
2609         unsigned offset = (long)_offset;
2610         struct kvm *kvm;
2611
2612         *val = 0;
2613         raw_spin_lock(&kvm_lock);
2614         list_for_each_entry(kvm, &vm_list, vm_list)
2615                 *val += *(u32 *)((void *)kvm + offset);
2616         raw_spin_unlock(&kvm_lock);
2617         return 0;
2618 }
2619
2620 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2621
2622 static int vcpu_stat_get(void *_offset, u64 *val)
2623 {
2624         unsigned offset = (long)_offset;
2625         struct kvm *kvm;
2626         struct kvm_vcpu *vcpu;
2627         int i;
2628
2629         *val = 0;
2630         raw_spin_lock(&kvm_lock);
2631         list_for_each_entry(kvm, &vm_list, vm_list)
2632                 kvm_for_each_vcpu(i, vcpu, kvm)
2633                         *val += *(u32 *)((void *)vcpu + offset);
2634
2635         raw_spin_unlock(&kvm_lock);
2636         return 0;
2637 }
2638
2639 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2640
2641 static const struct file_operations *stat_fops[] = {
2642         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2643         [KVM_STAT_VM]   = &vm_stat_fops,
2644 };
2645
2646 static int kvm_init_debug(void)
2647 {
2648         int r = -EFAULT;
2649         struct kvm_stats_debugfs_item *p;
2650
2651         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2652         if (kvm_debugfs_dir == NULL)
2653                 goto out;
2654
2655         for (p = debugfs_entries; p->name; ++p) {
2656                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2657                                                 (void *)(long)p->offset,
2658                                                 stat_fops[p->kind]);
2659                 if (p->dentry == NULL)
2660                         goto out_dir;
2661         }
2662
2663         return 0;
2664
2665 out_dir:
2666         debugfs_remove_recursive(kvm_debugfs_dir);
2667 out:
2668         return r;
2669 }
2670
2671 static void kvm_exit_debug(void)
2672 {
2673         struct kvm_stats_debugfs_item *p;
2674
2675         for (p = debugfs_entries; p->name; ++p)
2676                 debugfs_remove(p->dentry);
2677         debugfs_remove(kvm_debugfs_dir);
2678 }
2679
2680 static int kvm_suspend(void)
2681 {
2682         if (kvm_usage_count)
2683                 hardware_disable_nolock(NULL);
2684         return 0;
2685 }
2686
2687 static void kvm_resume(void)
2688 {
2689         if (kvm_usage_count) {
2690                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2691                 hardware_enable_nolock(NULL);
2692         }
2693 }
2694
2695 static struct syscore_ops kvm_syscore_ops = {
2696         .suspend = kvm_suspend,
2697         .resume = kvm_resume,
2698 };
2699
2700 struct page *bad_page;
2701 pfn_t bad_pfn;
2702
2703 static inline
2704 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2705 {
2706         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2707 }
2708
2709 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2710 {
2711         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2712
2713         kvm_arch_vcpu_load(vcpu, cpu);
2714 }
2715
2716 static void kvm_sched_out(struct preempt_notifier *pn,
2717                           struct task_struct *next)
2718 {
2719         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2720
2721         kvm_arch_vcpu_put(vcpu);
2722 }
2723
2724 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2725                   struct module *module)
2726 {
2727         int r;
2728         int cpu;
2729
2730         r = kvm_arch_init(opaque);
2731         if (r)
2732                 goto out_fail;
2733
2734         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2735
2736         if (bad_page == NULL) {
2737                 r = -ENOMEM;
2738                 goto out;
2739         }
2740
2741         bad_pfn = page_to_pfn(bad_page);
2742
2743         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2744
2745         if (hwpoison_page == NULL) {
2746                 r = -ENOMEM;
2747                 goto out_free_0;
2748         }
2749
2750         hwpoison_pfn = page_to_pfn(hwpoison_page);
2751
2752         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2753
2754         if (fault_page == NULL) {
2755                 r = -ENOMEM;
2756                 goto out_free_0;
2757         }
2758
2759         fault_pfn = page_to_pfn(fault_page);
2760
2761         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2762                 r = -ENOMEM;
2763                 goto out_free_0;
2764         }
2765
2766         r = kvm_arch_hardware_setup();
2767         if (r < 0)
2768                 goto out_free_0a;
2769
2770         for_each_online_cpu(cpu) {
2771                 smp_call_function_single(cpu,
2772                                 kvm_arch_check_processor_compat,
2773                                 &r, 1);
2774                 if (r < 0)
2775                         goto out_free_1;
2776         }
2777
2778         r = register_cpu_notifier(&kvm_cpu_notifier);
2779         if (r)
2780                 goto out_free_2;
2781         register_reboot_notifier(&kvm_reboot_notifier);
2782
2783         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2784         if (!vcpu_align)
2785                 vcpu_align = __alignof__(struct kvm_vcpu);
2786         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2787                                            0, NULL);
2788         if (!kvm_vcpu_cache) {
2789                 r = -ENOMEM;
2790                 goto out_free_3;
2791         }
2792
2793         r = kvm_async_pf_init();
2794         if (r)
2795                 goto out_free;
2796
2797         kvm_chardev_ops.owner = module;
2798         kvm_vm_fops.owner = module;
2799         kvm_vcpu_fops.owner = module;
2800
2801         r = misc_register(&kvm_dev);
2802         if (r) {
2803                 printk(KERN_ERR "kvm: misc device register failed\n");
2804                 goto out_unreg;
2805         }
2806
2807         register_syscore_ops(&kvm_syscore_ops);
2808
2809         kvm_preempt_ops.sched_in = kvm_sched_in;
2810         kvm_preempt_ops.sched_out = kvm_sched_out;
2811
2812         r = kvm_init_debug();
2813         if (r) {
2814                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2815                 goto out_undebugfs;
2816         }
2817
2818         return 0;
2819
2820 out_undebugfs:
2821         unregister_syscore_ops(&kvm_syscore_ops);
2822 out_unreg:
2823         kvm_async_pf_deinit();
2824 out_free:
2825         kmem_cache_destroy(kvm_vcpu_cache);
2826 out_free_3:
2827         unregister_reboot_notifier(&kvm_reboot_notifier);
2828         unregister_cpu_notifier(&kvm_cpu_notifier);
2829 out_free_2:
2830 out_free_1:
2831         kvm_arch_hardware_unsetup();
2832 out_free_0a:
2833         free_cpumask_var(cpus_hardware_enabled);
2834 out_free_0:
2835         if (fault_page)
2836                 __free_page(fault_page);
2837         if (hwpoison_page)
2838                 __free_page(hwpoison_page);
2839         __free_page(bad_page);
2840 out:
2841         kvm_arch_exit();
2842 out_fail:
2843         return r;
2844 }
2845 EXPORT_SYMBOL_GPL(kvm_init);
2846
2847 void kvm_exit(void)
2848 {
2849         kvm_exit_debug();
2850         misc_deregister(&kvm_dev);
2851         kmem_cache_destroy(kvm_vcpu_cache);
2852         kvm_async_pf_deinit();
2853         unregister_syscore_ops(&kvm_syscore_ops);
2854         unregister_reboot_notifier(&kvm_reboot_notifier);
2855         unregister_cpu_notifier(&kvm_cpu_notifier);
2856         on_each_cpu(hardware_disable_nolock, NULL, 1);
2857         kvm_arch_hardware_unsetup();
2858         kvm_arch_exit();
2859         free_cpumask_var(cpus_hardware_enabled);
2860         __free_page(fault_page);
2861         __free_page(hwpoison_page);
2862         __free_page(bad_page);
2863 }
2864 EXPORT_SYMBOL_GPL(kvm_exit);