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