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 static pfn_t get_hwpoison_pfn(void)
 943 {
 944         return -EHWPOISON;
 945 }
 946
 947 int is_hwpoison_pfn(pfn_t pfn)
 948 {
 949         return pfn == -EHWPOISON;
 950 }
 951 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 952
 953 int is_noslot_pfn(pfn_t pfn)
 954 {
 955         return pfn == -ENOENT;
 956 }
 957 EXPORT_SYMBOL_GPL(is_noslot_pfn);
 958
 959 int is_invalid_pfn(pfn_t pfn)
 960 {
 961         return !is_noslot_pfn(pfn) && is_error_pfn(pfn);
 962 }
 963 EXPORT_SYMBOL_GPL(is_invalid_pfn);
 964
 965 struct page *get_bad_page(void)
 966 {
 967         return ERR_PTR(-ENOENT);
 968 }
 969
 970 static inline unsigned long bad_hva(void)
 971 {
 972         return PAGE_OFFSET;
 973 }
 974
 975 int kvm_is_error_hva(unsigned long addr)
 976 {
 977         return addr == bad_hva();
 978 }
 979 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 980
 981 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 982 {
 983         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 984 }
 985 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 986
 987 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 988 {
 989         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 990
 991         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
 992               memslot->flags & KVM_MEMSLOT_INVALID)
 993                 return 0;
 994
 995         return 1;
 996 }
 997 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 998
 999 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1000 {
1001         struct vm_area_struct *vma;
1002         unsigned long addr, size;
1003
1004         size = PAGE_SIZE;
1005
1006         addr = gfn_to_hva(kvm, gfn);
1007         if (kvm_is_error_hva(addr))
1008                 return PAGE_SIZE;
1009
1010         down_read(&current->mm->mmap_sem);
1011         vma = find_vma(current->mm, addr);
1012         if (!vma)
1013                 goto out;
1014
1015         size = vma_kernel_pagesize(vma);
1016
1017 out:
1018         up_read(&current->mm->mmap_sem);
1019
1020         return size;
1021 }
1022
1023 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1024                                      gfn_t *nr_pages)
1025 {
1026         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1027                 return bad_hva();
1028
1029         if (nr_pages)
1030                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1031
1032         return gfn_to_hva_memslot(slot, gfn);
1033 }
1034
1035 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1036 {
1037         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1038 }
1039 EXPORT_SYMBOL_GPL(gfn_to_hva);
1040
1041 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1042         unsigned long start, int write, struct page **page)
1043 {
1044         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1045
1046         if (write)
1047                 flags |= FOLL_WRITE;
1048
1049         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1050 }
1051
1052 static inline int check_user_page_hwpoison(unsigned long addr)
1053 {
1054         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1055
1056         rc = __get_user_pages(current, current->mm, addr, 1,
1057                               flags, NULL, NULL, NULL);
1058         return rc == -EHWPOISON;
1059 }
1060
1061 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1062                         bool write_fault, bool *writable)
1063 {
1064         struct page *page[1];
1065         int npages = 0;
1066         pfn_t pfn;
1067
1068         /* we can do it either atomically or asynchronously, not both */
1069         BUG_ON(atomic && async);
1070
1071         BUG_ON(!write_fault && !writable);
1072
1073         if (writable)
1074                 *writable = true;
1075
1076         if (atomic || async)
1077                 npages = __get_user_pages_fast(addr, 1, 1, page);
1078
1079         if (unlikely(npages != 1) && !atomic) {
1080                 might_sleep();
1081
1082                 if (writable)
1083                         *writable = write_fault;
1084
1085                 if (async) {
1086                         down_read(&current->mm->mmap_sem);
1087                         npages = get_user_page_nowait(current, current->mm,
1088                                                      addr, write_fault, page);
1089                         up_read(&current->mm->mmap_sem);
1090                 } else
1091                         npages = get_user_pages_fast(addr, 1, write_fault,
1092                                                      page);
1093
1094                 /* map read fault as writable if possible */
1095                 if (unlikely(!write_fault) && npages == 1) {
1096                         struct page *wpage[1];
1097
1098                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1099                         if (npages == 1) {
1100                                 *writable = true;
1101                                 put_page(page[0]);
1102                                 page[0] = wpage[0];
1103                         }
1104                         npages = 1;
1105                 }
1106         }
1107
1108         if (unlikely(npages != 1)) {
1109                 struct vm_area_struct *vma;
1110
1111                 if (atomic)
1112                         return KVM_PFN_ERR_FAULT;
1113
1114                 down_read(&current->mm->mmap_sem);
1115                 if (npages == -EHWPOISON ||
1116                         (!async && check_user_page_hwpoison(addr))) {
1117                         up_read(&current->mm->mmap_sem);
1118                         return get_hwpoison_pfn();
1119                 }
1120
1121                 vma = find_vma_intersection(current->mm, addr, addr+1);
1122
1123                 if (vma == NULL)
1124                         pfn = KVM_PFN_ERR_FAULT;
1125                 else if ((vma->vm_flags & VM_PFNMAP)) {
1126                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1127                                 vma->vm_pgoff;
1128                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1129                 } else {
1130                         if (async && (vma->vm_flags & VM_WRITE))
1131                                 *async = true;
1132                         pfn = KVM_PFN_ERR_FAULT;
1133                 }
1134                 up_read(&current->mm->mmap_sem);
1135         } else
1136                 pfn = page_to_pfn(page[0]);
1137
1138         return pfn;
1139 }
1140
1141 pfn_t hva_to_pfn_atomic(unsigned long addr)
1142 {
1143         return hva_to_pfn(addr, true, NULL, true, NULL);
1144 }
1145 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1146
1147 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1148                           bool write_fault, bool *writable)
1149 {
1150         unsigned long addr;
1151
1152         if (async)
1153                 *async = false;
1154
1155         addr = gfn_to_hva(kvm, gfn);
1156         if (kvm_is_error_hva(addr))
1157                 return get_bad_pfn();
1158
1159         return hva_to_pfn(addr, atomic, async, write_fault, writable);
1160 }
1161
1162 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1163 {
1164         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1165 }
1166 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1167
1168 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1169                        bool write_fault, bool *writable)
1170 {
1171         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1172 }
1173 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1174
1175 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1176 {
1177         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1178 }
1179 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1180
1181 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1182                       bool *writable)
1183 {
1184         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1185 }
1186 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1187
1188 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1189 {
1190         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1191         return hva_to_pfn(addr, false, NULL, true, NULL);
1192 }
1193
1194 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1195                                                                   int nr_pages)
1196 {
1197         unsigned long addr;
1198         gfn_t entry;
1199
1200         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1201         if (kvm_is_error_hva(addr))
1202                 return -1;
1203
1204         if (entry < nr_pages)
1205                 return 0;
1206
1207         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1208 }
1209 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1210
1211 static struct page *kvm_pfn_to_page(pfn_t pfn)
1212 {
1213         WARN_ON(kvm_is_mmio_pfn(pfn));
1214
1215         if (is_error_pfn(pfn) || kvm_is_mmio_pfn(pfn))
1216                 return get_bad_page();
1217
1218         return pfn_to_page(pfn);
1219 }
1220
1221 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1222 {
1223         pfn_t pfn;
1224
1225         pfn = gfn_to_pfn(kvm, gfn);
1226
1227         return kvm_pfn_to_page(pfn);
1228 }
1229
1230 EXPORT_SYMBOL_GPL(gfn_to_page);
1231
1232 void kvm_release_page_clean(struct page *page)
1233 {
1234         if (!is_error_page(page))
1235                 kvm_release_pfn_clean(page_to_pfn(page));
1236 }
1237 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1238
1239 void kvm_release_pfn_clean(pfn_t pfn)
1240 {
1241         if (!is_error_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1242                 put_page(pfn_to_page(pfn));
1243 }
1244 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1245
1246 void kvm_release_page_dirty(struct page *page)
1247 {
1248         WARN_ON(is_error_page(page));
1249
1250         kvm_release_pfn_dirty(page_to_pfn(page));
1251 }
1252 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1253
1254 void kvm_release_pfn_dirty(pfn_t pfn)
1255 {
1256         kvm_set_pfn_dirty(pfn);
1257         kvm_release_pfn_clean(pfn);
1258 }
1259 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1260
1261 void kvm_set_page_dirty(struct page *page)
1262 {
1263         kvm_set_pfn_dirty(page_to_pfn(page));
1264 }
1265 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1266
1267 void kvm_set_pfn_dirty(pfn_t pfn)
1268 {
1269         if (!kvm_is_mmio_pfn(pfn)) {
1270                 struct page *page = pfn_to_page(pfn);
1271                 if (!PageReserved(page))
1272                         SetPageDirty(page);
1273         }
1274 }
1275 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1276
1277 void kvm_set_pfn_accessed(pfn_t pfn)
1278 {
1279         if (!kvm_is_mmio_pfn(pfn))
1280                 mark_page_accessed(pfn_to_page(pfn));
1281 }
1282 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1283
1284 void kvm_get_pfn(pfn_t pfn)
1285 {
1286         if (!kvm_is_mmio_pfn(pfn))
1287                 get_page(pfn_to_page(pfn));
1288 }
1289 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1290
1291 static int next_segment(unsigned long len, int offset)
1292 {
1293         if (len > PAGE_SIZE - offset)
1294                 return PAGE_SIZE - offset;
1295         else
1296                 return len;
1297 }
1298
1299 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1300                         int len)
1301 {
1302         int r;
1303         unsigned long addr;
1304
1305         addr = gfn_to_hva(kvm, gfn);
1306         if (kvm_is_error_hva(addr))
1307                 return -EFAULT;
1308         r = __copy_from_user(data, (void __user *)addr + offset, len);
1309         if (r)
1310                 return -EFAULT;
1311         return 0;
1312 }
1313 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1314
1315 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1316 {
1317         gfn_t gfn = gpa >> PAGE_SHIFT;
1318         int seg;
1319         int offset = offset_in_page(gpa);
1320         int ret;
1321
1322         while ((seg = next_segment(len, offset)) != 0) {
1323                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1324                 if (ret < 0)
1325                         return ret;
1326                 offset = 0;
1327                 len -= seg;
1328                 data += seg;
1329                 ++gfn;
1330         }
1331         return 0;
1332 }
1333 EXPORT_SYMBOL_GPL(kvm_read_guest);
1334
1335 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1336                           unsigned long len)
1337 {
1338         int r;
1339         unsigned long addr;
1340         gfn_t gfn = gpa >> PAGE_SHIFT;
1341         int offset = offset_in_page(gpa);
1342
1343         addr = gfn_to_hva(kvm, gfn);
1344         if (kvm_is_error_hva(addr))
1345                 return -EFAULT;
1346         pagefault_disable();
1347         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1348         pagefault_enable();
1349         if (r)
1350                 return -EFAULT;
1351         return 0;
1352 }
1353 EXPORT_SYMBOL(kvm_read_guest_atomic);
1354
1355 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1356                          int offset, int len)
1357 {
1358         int r;
1359         unsigned long addr;
1360
1361         addr = gfn_to_hva(kvm, gfn);
1362         if (kvm_is_error_hva(addr))
1363                 return -EFAULT;
1364         r = __copy_to_user((void __user *)addr + offset, data, len);
1365         if (r)
1366                 return -EFAULT;
1367         mark_page_dirty(kvm, gfn);
1368         return 0;
1369 }
1370 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1371
1372 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1373                     unsigned long len)
1374 {
1375         gfn_t gfn = gpa >> PAGE_SHIFT;
1376         int seg;
1377         int offset = offset_in_page(gpa);
1378         int ret;
1379
1380         while ((seg = next_segment(len, offset)) != 0) {
1381                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1382                 if (ret < 0)
1383                         return ret;
1384                 offset = 0;
1385                 len -= seg;
1386                 data += seg;
1387                 ++gfn;
1388         }
1389         return 0;
1390 }
1391
1392 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1393                               gpa_t gpa)
1394 {
1395         struct kvm_memslots *slots = kvm_memslots(kvm);
1396         int offset = offset_in_page(gpa);
1397         gfn_t gfn = gpa >> PAGE_SHIFT;
1398
1399         ghc->gpa = gpa;
1400         ghc->generation = slots->generation;
1401         ghc->memslot = gfn_to_memslot(kvm, gfn);
1402         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1403         if (!kvm_is_error_hva(ghc->hva))
1404                 ghc->hva += offset;
1405         else
1406                 return -EFAULT;
1407
1408         return 0;
1409 }
1410 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1411
1412 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1413                            void *data, unsigned long len)
1414 {
1415         struct kvm_memslots *slots = kvm_memslots(kvm);
1416         int r;
1417
1418         if (slots->generation != ghc->generation)
1419                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1420
1421         if (kvm_is_error_hva(ghc->hva))
1422                 return -EFAULT;
1423
1424         r = __copy_to_user((void __user *)ghc->hva, data, len);
1425         if (r)
1426                 return -EFAULT;
1427         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1428
1429         return 0;
1430 }
1431 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1432
1433 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1434                            void *data, unsigned long len)
1435 {
1436         struct kvm_memslots *slots = kvm_memslots(kvm);
1437         int r;
1438
1439         if (slots->generation != ghc->generation)
1440                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1441
1442         if (kvm_is_error_hva(ghc->hva))
1443                 return -EFAULT;
1444
1445         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1446         if (r)
1447                 return -EFAULT;
1448
1449         return 0;
1450 }
1451 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1452
1453 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1454 {
1455         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1456                                     offset, len);
1457 }
1458 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1459
1460 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1461 {
1462         gfn_t gfn = gpa >> PAGE_SHIFT;
1463         int seg;
1464         int offset = offset_in_page(gpa);
1465         int ret;
1466
1467         while ((seg = next_segment(len, offset)) != 0) {
1468                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1469                 if (ret < 0)
1470                         return ret;
1471                 offset = 0;
1472                 len -= seg;
1473                 ++gfn;
1474         }
1475         return 0;
1476 }
1477 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1478
1479 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1480                              gfn_t gfn)
1481 {
1482         if (memslot && memslot->dirty_bitmap) {
1483                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1484
1485                 /* TODO: introduce set_bit_le() and use it */
1486                 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1487         }
1488 }
1489
1490 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1491 {
1492         struct kvm_memory_slot *memslot;
1493
1494         memslot = gfn_to_memslot(kvm, gfn);
1495         mark_page_dirty_in_slot(kvm, memslot, gfn);
1496 }
1497
1498 /*
1499  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1500  */
1501 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1502 {
1503         DEFINE_WAIT(wait);
1504
1505         for (;;) {
1506                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1507
1508                 if (kvm_arch_vcpu_runnable(vcpu)) {
1509                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1510                         break;
1511                 }
1512                 if (kvm_cpu_has_pending_timer(vcpu))
1513                         break;
1514                 if (signal_pending(current))
1515                         break;
1516
1517                 schedule();
1518         }
1519
1520         finish_wait(&vcpu->wq, &wait);
1521 }
1522
1523 #ifndef CONFIG_S390
1524 /*
1525  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1526  */
1527 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1528 {
1529         int me;
1530         int cpu = vcpu->cpu;
1531         wait_queue_head_t *wqp;
1532
1533         wqp = kvm_arch_vcpu_wq(vcpu);
1534         if (waitqueue_active(wqp)) {
1535                 wake_up_interruptible(wqp);
1536                 ++vcpu->stat.halt_wakeup;
1537         }
1538
1539         me = get_cpu();
1540         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1541                 if (kvm_arch_vcpu_should_kick(vcpu))
1542                         smp_send_reschedule(cpu);
1543         put_cpu();
1544 }
1545 #endif /* !CONFIG_S390 */
1546
1547 void kvm_resched(struct kvm_vcpu *vcpu)
1548 {
1549         if (!need_resched())
1550                 return;
1551         cond_resched();
1552 }
1553 EXPORT_SYMBOL_GPL(kvm_resched);
1554
1555 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1556 {
1557         struct pid *pid;
1558         struct task_struct *task = NULL;
1559
1560         rcu_read_lock();
1561         pid = rcu_dereference(target->pid);
1562         if (pid)
1563                 task = get_pid_task(target->pid, PIDTYPE_PID);
1564         rcu_read_unlock();
1565         if (!task)
1566                 return false;
1567         if (task->flags & PF_VCPU) {
1568                 put_task_struct(task);
1569                 return false;
1570         }
1571         if (yield_to(task, 1)) {
1572                 put_task_struct(task);
1573                 return true;
1574         }
1575         put_task_struct(task);
1576         return false;
1577 }
1578 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1579
1580 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1581 /*
1582  * Helper that checks whether a VCPU is eligible for directed yield.
1583  * Most eligible candidate to yield is decided by following heuristics:
1584  *
1585  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1586  *  (preempted lock holder), indicated by @in_spin_loop.
1587  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1588  *
1589  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1590  *  chance last time (mostly it has become eligible now since we have probably
1591  *  yielded to lockholder in last iteration. This is done by toggling
1592  *  @dy_eligible each time a VCPU checked for eligibility.)
1593  *
1594  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1595  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1596  *  burning. Giving priority for a potential lock-holder increases lock
1597  *  progress.
1598  *
1599  *  Since algorithm is based on heuristics, accessing another VCPU data without
1600  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1601  *  and continue with next VCPU and so on.
1602  */
1603 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1604 {
1605         bool eligible;
1606
1607         eligible = !vcpu->spin_loop.in_spin_loop ||
1608                         (vcpu->spin_loop.in_spin_loop &&
1609                          vcpu->spin_loop.dy_eligible);
1610
1611         if (vcpu->spin_loop.in_spin_loop)
1612                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1613
1614         return eligible;
1615 }
1616 #endif
1617 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1618 {
1619         struct kvm *kvm = me->kvm;
1620         struct kvm_vcpu *vcpu;
1621         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1622         int yielded = 0;
1623         int pass;
1624         int i;
1625
1626         kvm_vcpu_set_in_spin_loop(me, true);
1627         /*
1628          * We boost the priority of a VCPU that is runnable but not
1629          * currently running, because it got preempted by something
1630          * else and called schedule in __vcpu_run.  Hopefully that
1631          * VCPU is holding the lock that we need and will release it.
1632          * We approximate round-robin by starting at the last boosted VCPU.
1633          */
1634         for (pass = 0; pass < 2 && !yielded; pass++) {
1635                 kvm_for_each_vcpu(i, vcpu, kvm) {
1636                         if (!pass && i <= last_boosted_vcpu) {
1637                                 i = last_boosted_vcpu;
1638                                 continue;
1639                         } else if (pass && i > last_boosted_vcpu)
1640                                 break;
1641                         if (vcpu == me)
1642                                 continue;
1643                         if (waitqueue_active(&vcpu->wq))
1644                                 continue;
1645                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1646                                 continue;
1647                         if (kvm_vcpu_yield_to(vcpu)) {
1648                                 kvm->last_boosted_vcpu = i;
1649                                 yielded = 1;
1650                                 break;
1651                         }
1652                 }
1653         }
1654         kvm_vcpu_set_in_spin_loop(me, false);
1655
1656         /* Ensure vcpu is not eligible during next spinloop */
1657         kvm_vcpu_set_dy_eligible(me, false);
1658 }
1659 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1660
1661 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1662 {
1663         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1664         struct page *page;
1665
1666         if (vmf->pgoff == 0)
1667                 page = virt_to_page(vcpu->run);
1668 #ifdef CONFIG_X86
1669         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1670                 page = virt_to_page(vcpu->arch.pio_data);
1671 #endif
1672 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1673         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1674                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1675 #endif
1676         else
1677                 return kvm_arch_vcpu_fault(vcpu, vmf);
1678         get_page(page);
1679         vmf->page = page;
1680         return 0;
1681 }
1682
1683 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1684         .fault = kvm_vcpu_fault,
1685 };
1686
1687 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1688 {
1689         vma->vm_ops = &kvm_vcpu_vm_ops;
1690         return 0;
1691 }
1692
1693 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1694 {
1695         struct kvm_vcpu *vcpu = filp->private_data;
1696
1697         kvm_put_kvm(vcpu->kvm);
1698         return 0;
1699 }
1700
1701 static struct file_operations kvm_vcpu_fops = {
1702         .release        = kvm_vcpu_release,
1703         .unlocked_ioctl = kvm_vcpu_ioctl,
1704 #ifdef CONFIG_COMPAT
1705         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1706 #endif
1707         .mmap           = kvm_vcpu_mmap,
1708         .llseek         = noop_llseek,
1709 };
1710
1711 /*
1712  * Allocates an inode for the vcpu.
1713  */
1714 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1715 {
1716         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1717 }
1718
1719 /*
1720  * Creates some virtual cpus.  Good luck creating more than one.
1721  */
1722 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1723 {
1724         int r;
1725         struct kvm_vcpu *vcpu, *v;
1726
1727         vcpu = kvm_arch_vcpu_create(kvm, id);
1728         if (IS_ERR(vcpu))
1729                 return PTR_ERR(vcpu);
1730
1731         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1732
1733         r = kvm_arch_vcpu_setup(vcpu);
1734         if (r)
1735                 goto vcpu_destroy;
1736
1737         mutex_lock(&kvm->lock);
1738         if (!kvm_vcpu_compatible(vcpu)) {
1739                 r = -EINVAL;
1740                 goto unlock_vcpu_destroy;
1741         }
1742         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1743                 r = -EINVAL;
1744                 goto unlock_vcpu_destroy;
1745         }
1746
1747         kvm_for_each_vcpu(r, v, kvm)
1748                 if (v->vcpu_id == id) {
1749                         r = -EEXIST;
1750                         goto unlock_vcpu_destroy;
1751                 }
1752
1753         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1754
1755         /* Now it's all set up, let userspace reach it */
1756         kvm_get_kvm(kvm);
1757         r = create_vcpu_fd(vcpu);
1758         if (r < 0) {
1759                 kvm_put_kvm(kvm);
1760                 goto unlock_vcpu_destroy;
1761         }
1762
1763         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1764         smp_wmb();
1765         atomic_inc(&kvm->online_vcpus);
1766
1767         mutex_unlock(&kvm->lock);
1768         return r;
1769
1770 unlock_vcpu_destroy:
1771         mutex_unlock(&kvm->lock);
1772 vcpu_destroy:
1773         kvm_arch_vcpu_destroy(vcpu);
1774         return r;
1775 }
1776
1777 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1778 {
1779         if (sigset) {
1780                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1781                 vcpu->sigset_active = 1;
1782                 vcpu->sigset = *sigset;
1783         } else
1784                 vcpu->sigset_active = 0;
1785         return 0;
1786 }
1787
1788 static long kvm_vcpu_ioctl(struct file *filp,
1789                            unsigned int ioctl, unsigned long arg)
1790 {
1791         struct kvm_vcpu *vcpu = filp->private_data;
1792         void __user *argp = (void __user *)arg;
1793         int r;
1794         struct kvm_fpu *fpu = NULL;
1795         struct kvm_sregs *kvm_sregs = NULL;
1796
1797         if (vcpu->kvm->mm != current->mm)
1798                 return -EIO;
1799
1800 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1801         /*
1802          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1803          * so vcpu_load() would break it.
1804          */
1805         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1806                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1807 #endif
1808
1809
1810         vcpu_load(vcpu);
1811         switch (ioctl) {
1812         case KVM_RUN:
1813                 r = -EINVAL;
1814                 if (arg)
1815                         goto out;
1816                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1817                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1818                 break;
1819         case KVM_GET_REGS: {
1820                 struct kvm_regs *kvm_regs;
1821
1822                 r = -ENOMEM;
1823                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1824                 if (!kvm_regs)
1825                         goto out;
1826                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1827                 if (r)
1828                         goto out_free1;
1829                 r = -EFAULT;
1830                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1831                         goto out_free1;
1832                 r = 0;
1833 out_free1:
1834                 kfree(kvm_regs);
1835                 break;
1836         }
1837         case KVM_SET_REGS: {
1838                 struct kvm_regs *kvm_regs;
1839
1840                 r = -ENOMEM;
1841                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1842                 if (IS_ERR(kvm_regs)) {
1843                         r = PTR_ERR(kvm_regs);
1844                         goto out;
1845                 }
1846                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1847                 if (r)
1848                         goto out_free2;
1849                 r = 0;
1850 out_free2:
1851                 kfree(kvm_regs);
1852                 break;
1853         }
1854         case KVM_GET_SREGS: {
1855                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1856                 r = -ENOMEM;
1857                 if (!kvm_sregs)
1858                         goto out;
1859                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1860                 if (r)
1861                         goto out;
1862                 r = -EFAULT;
1863                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1864                         goto out;
1865                 r = 0;
1866                 break;
1867         }
1868         case KVM_SET_SREGS: {
1869                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1870                 if (IS_ERR(kvm_sregs)) {
1871                         r = PTR_ERR(kvm_sregs);
1872                         goto out;
1873                 }
1874                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1875                 if (r)
1876                         goto out;
1877                 r = 0;
1878                 break;
1879         }
1880         case KVM_GET_MP_STATE: {
1881                 struct kvm_mp_state mp_state;
1882
1883                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1884                 if (r)
1885                         goto out;
1886                 r = -EFAULT;
1887                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1888                         goto out;
1889                 r = 0;
1890                 break;
1891         }
1892         case KVM_SET_MP_STATE: {
1893                 struct kvm_mp_state mp_state;
1894
1895                 r = -EFAULT;
1896                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1897                         goto out;
1898                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1899                 if (r)
1900                         goto out;
1901                 r = 0;
1902                 break;
1903         }
1904         case KVM_TRANSLATE: {
1905                 struct kvm_translation tr;
1906
1907                 r = -EFAULT;
1908                 if (copy_from_user(&tr, argp, sizeof tr))
1909                         goto out;
1910                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1911                 if (r)
1912                         goto out;
1913                 r = -EFAULT;
1914                 if (copy_to_user(argp, &tr, sizeof tr))
1915                         goto out;
1916                 r = 0;
1917                 break;
1918         }
1919         case KVM_SET_GUEST_DEBUG: {
1920                 struct kvm_guest_debug dbg;
1921
1922                 r = -EFAULT;
1923                 if (copy_from_user(&dbg, argp, sizeof dbg))
1924                         goto out;
1925                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1926                 if (r)
1927                         goto out;
1928                 r = 0;
1929                 break;
1930         }
1931         case KVM_SET_SIGNAL_MASK: {
1932                 struct kvm_signal_mask __user *sigmask_arg = argp;
1933                 struct kvm_signal_mask kvm_sigmask;
1934                 sigset_t sigset, *p;
1935
1936                 p = NULL;
1937                 if (argp) {
1938                         r = -EFAULT;
1939                         if (copy_from_user(&kvm_sigmask, argp,
1940                                            sizeof kvm_sigmask))
1941                                 goto out;
1942                         r = -EINVAL;
1943                         if (kvm_sigmask.len != sizeof sigset)
1944                                 goto out;
1945                         r = -EFAULT;
1946                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1947                                            sizeof sigset))
1948                                 goto out;
1949                         p = &sigset;
1950                 }
1951                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1952                 break;
1953         }
1954         case KVM_GET_FPU: {
1955                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1956                 r = -ENOMEM;
1957                 if (!fpu)
1958                         goto out;
1959                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1960                 if (r)
1961                         goto out;
1962                 r = -EFAULT;
1963                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1964                         goto out;
1965                 r = 0;
1966                 break;
1967         }
1968         case KVM_SET_FPU: {
1969                 fpu = memdup_user(argp, sizeof(*fpu));
1970                 if (IS_ERR(fpu)) {
1971                         r = PTR_ERR(fpu);
1972                         goto out;
1973                 }
1974                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1975                 if (r)
1976                         goto out;
1977                 r = 0;
1978                 break;
1979         }
1980         default:
1981                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1982         }
1983 out:
1984         vcpu_put(vcpu);
1985         kfree(fpu);
1986         kfree(kvm_sregs);
1987         return r;
1988 }
1989
1990 #ifdef CONFIG_COMPAT
1991 static long kvm_vcpu_compat_ioctl(struct file *filp,
1992                                   unsigned int ioctl, unsigned long arg)
1993 {
1994         struct kvm_vcpu *vcpu = filp->private_data;
1995         void __user *argp = compat_ptr(arg);
1996         int r;
1997
1998         if (vcpu->kvm->mm != current->mm)
1999                 return -EIO;
2000
2001         switch (ioctl) {
2002         case KVM_SET_SIGNAL_MASK: {
2003                 struct kvm_signal_mask __user *sigmask_arg = argp;
2004                 struct kvm_signal_mask kvm_sigmask;
2005                 compat_sigset_t csigset;
2006                 sigset_t sigset;
2007
2008                 if (argp) {
2009                         r = -EFAULT;
2010                         if (copy_from_user(&kvm_sigmask, argp,
2011                                            sizeof kvm_sigmask))
2012                                 goto out;
2013                         r = -EINVAL;
2014                         if (kvm_sigmask.len != sizeof csigset)
2015                                 goto out;
2016                         r = -EFAULT;
2017                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2018                                            sizeof csigset))
2019                                 goto out;
2020                 }
2021                 sigset_from_compat(&sigset, &csigset);
2022                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2023                 break;
2024         }
2025         default:
2026                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2027         }
2028
2029 out:
2030         return r;
2031 }
2032 #endif
2033
2034 static long kvm_vm_ioctl(struct file *filp,
2035                            unsigned int ioctl, unsigned long arg)
2036 {
2037         struct kvm *kvm = filp->private_data;
2038         void __user *argp = (void __user *)arg;
2039         int r;
2040
2041         if (kvm->mm != current->mm)
2042                 return -EIO;
2043         switch (ioctl) {
2044         case KVM_CREATE_VCPU:
2045                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2046                 if (r < 0)
2047                         goto out;
2048                 break;
2049         case KVM_SET_USER_MEMORY_REGION: {
2050                 struct kvm_userspace_memory_region kvm_userspace_mem;
2051
2052                 r = -EFAULT;
2053                 if (copy_from_user(&kvm_userspace_mem, argp,
2054                                                 sizeof kvm_userspace_mem))
2055                         goto out;
2056
2057                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2058                 if (r)
2059                         goto out;
2060                 break;
2061         }
2062         case KVM_GET_DIRTY_LOG: {
2063                 struct kvm_dirty_log log;
2064
2065                 r = -EFAULT;
2066                 if (copy_from_user(&log, argp, sizeof log))
2067                         goto out;
2068                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2069                 if (r)
2070                         goto out;
2071                 break;
2072         }
2073 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2074         case KVM_REGISTER_COALESCED_MMIO: {
2075                 struct kvm_coalesced_mmio_zone zone;
2076                 r = -EFAULT;
2077                 if (copy_from_user(&zone, argp, sizeof zone))
2078                         goto out;
2079                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2080                 if (r)
2081                         goto out;
2082                 r = 0;
2083                 break;
2084         }
2085         case KVM_UNREGISTER_COALESCED_MMIO: {
2086                 struct kvm_coalesced_mmio_zone zone;
2087                 r = -EFAULT;
2088                 if (copy_from_user(&zone, argp, sizeof zone))
2089                         goto out;
2090                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2091                 if (r)
2092                         goto out;
2093                 r = 0;
2094                 break;
2095         }
2096 #endif
2097         case KVM_IRQFD: {
2098                 struct kvm_irqfd data;
2099
2100                 r = -EFAULT;
2101                 if (copy_from_user(&data, argp, sizeof data))
2102                         goto out;
2103                 r = kvm_irqfd(kvm, &data);
2104                 break;
2105         }
2106         case KVM_IOEVENTFD: {
2107                 struct kvm_ioeventfd data;
2108
2109                 r = -EFAULT;
2110                 if (copy_from_user(&data, argp, sizeof data))
2111                         goto out;
2112                 r = kvm_ioeventfd(kvm, &data);
2113                 break;
2114         }
2115 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2116         case KVM_SET_BOOT_CPU_ID:
2117                 r = 0;
2118                 mutex_lock(&kvm->lock);
2119                 if (atomic_read(&kvm->online_vcpus) != 0)
2120                         r = -EBUSY;
2121                 else
2122                         kvm->bsp_vcpu_id = arg;
2123                 mutex_unlock(&kvm->lock);
2124                 break;
2125 #endif
2126 #ifdef CONFIG_HAVE_KVM_MSI
2127         case KVM_SIGNAL_MSI: {
2128                 struct kvm_msi msi;
2129
2130                 r = -EFAULT;
2131                 if (copy_from_user(&msi, argp, sizeof msi))
2132                         goto out;
2133                 r = kvm_send_userspace_msi(kvm, &msi);
2134                 break;
2135         }
2136 #endif
2137 #ifdef __KVM_HAVE_IRQ_LINE
2138         case KVM_IRQ_LINE_STATUS:
2139         case KVM_IRQ_LINE: {
2140                 struct kvm_irq_level irq_event;
2141
2142                 r = -EFAULT;
2143                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2144                         goto out;
2145
2146                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2147                 if (r)
2148                         goto out;
2149
2150                 r = -EFAULT;
2151                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2152                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2153                                 goto out;
2154                 }
2155
2156                 r = 0;
2157                 break;
2158         }
2159 #endif
2160         default:
2161                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2162                 if (r == -ENOTTY)
2163                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2164         }
2165 out:
2166         return r;
2167 }
2168
2169 #ifdef CONFIG_COMPAT
2170 struct compat_kvm_dirty_log {
2171         __u32 slot;
2172         __u32 padding1;
2173         union {
2174                 compat_uptr_t dirty_bitmap; /* one bit per page */
2175                 __u64 padding2;
2176         };
2177 };
2178
2179 static long kvm_vm_compat_ioctl(struct file *filp,
2180                            unsigned int ioctl, unsigned long arg)
2181 {
2182         struct kvm *kvm = filp->private_data;
2183         int r;
2184
2185         if (kvm->mm != current->mm)
2186                 return -EIO;
2187         switch (ioctl) {
2188         case KVM_GET_DIRTY_LOG: {
2189                 struct compat_kvm_dirty_log compat_log;
2190                 struct kvm_dirty_log log;
2191
2192                 r = -EFAULT;
2193                 if (copy_from_user(&compat_log, (void __user *)arg,
2194                                    sizeof(compat_log)))
2195                         goto out;
2196                 log.slot         = compat_log.slot;
2197                 log.padding1     = compat_log.padding1;
2198                 log.padding2     = compat_log.padding2;
2199                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2200
2201                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2202                 if (r)
2203                         goto out;
2204                 break;
2205         }
2206         default:
2207                 r = kvm_vm_ioctl(filp, ioctl, arg);
2208         }
2209
2210 out:
2211         return r;
2212 }
2213 #endif
2214
2215 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2216 {
2217         struct page *page[1];
2218         unsigned long addr;
2219         int npages;
2220         gfn_t gfn = vmf->pgoff;
2221         struct kvm *kvm = vma->vm_file->private_data;
2222
2223         addr = gfn_to_hva(kvm, gfn);
2224         if (kvm_is_error_hva(addr))
2225                 return VM_FAULT_SIGBUS;
2226
2227         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2228                                 NULL);
2229         if (unlikely(npages != 1))
2230                 return VM_FAULT_SIGBUS;
2231
2232         vmf->page = page[0];
2233         return 0;
2234 }
2235
2236 static const struct vm_operations_struct kvm_vm_vm_ops = {
2237         .fault = kvm_vm_fault,
2238 };
2239
2240 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2241 {
2242         vma->vm_ops = &kvm_vm_vm_ops;
2243         return 0;
2244 }
2245
2246 static struct file_operations kvm_vm_fops = {
2247         .release        = kvm_vm_release,
2248         .unlocked_ioctl = kvm_vm_ioctl,
2249 #ifdef CONFIG_COMPAT
2250         .compat_ioctl   = kvm_vm_compat_ioctl,
2251 #endif
2252         .mmap           = kvm_vm_mmap,
2253         .llseek         = noop_llseek,
2254 };
2255
2256 static int kvm_dev_ioctl_create_vm(unsigned long type)
2257 {
2258         int r;
2259         struct kvm *kvm;
2260
2261         kvm = kvm_create_vm(type);
2262         if (IS_ERR(kvm))
2263                 return PTR_ERR(kvm);
2264 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2265         r = kvm_coalesced_mmio_init(kvm);
2266         if (r < 0) {
2267                 kvm_put_kvm(kvm);
2268                 return r;
2269         }
2270 #endif
2271         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2272         if (r < 0)
2273                 kvm_put_kvm(kvm);
2274
2275         return r;
2276 }
2277
2278 static long kvm_dev_ioctl_check_extension_generic(long arg)
2279 {
2280         switch (arg) {
2281         case KVM_CAP_USER_MEMORY:
2282         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2283         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2284 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2285         case KVM_CAP_SET_BOOT_CPU_ID:
2286 #endif
2287         case KVM_CAP_INTERNAL_ERROR_DATA:
2288 #ifdef CONFIG_HAVE_KVM_MSI
2289         case KVM_CAP_SIGNAL_MSI:
2290 #endif
2291                 return 1;
2292 #ifdef KVM_CAP_IRQ_ROUTING
2293         case KVM_CAP_IRQ_ROUTING:
2294                 return KVM_MAX_IRQ_ROUTES;
2295 #endif
2296         default:
2297                 break;
2298         }
2299         return kvm_dev_ioctl_check_extension(arg);
2300 }
2301
2302 static long kvm_dev_ioctl(struct file *filp,
2303                           unsigned int ioctl, unsigned long arg)
2304 {
2305         long r = -EINVAL;
2306
2307         switch (ioctl) {
2308         case KVM_GET_API_VERSION:
2309                 r = -EINVAL;
2310                 if (arg)
2311                         goto out;
2312                 r = KVM_API_VERSION;
2313                 break;
2314         case KVM_CREATE_VM:
2315                 r = kvm_dev_ioctl_create_vm(arg);
2316                 break;
2317         case KVM_CHECK_EXTENSION:
2318                 r = kvm_dev_ioctl_check_extension_generic(arg);
2319                 break;
2320         case KVM_GET_VCPU_MMAP_SIZE:
2321                 r = -EINVAL;
2322                 if (arg)
2323                         goto out;
2324                 r = PAGE_SIZE;     /* struct kvm_run */
2325 #ifdef CONFIG_X86
2326                 r += PAGE_SIZE;    /* pio data page */
2327 #endif
2328 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2329                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2330 #endif
2331                 break;
2332         case KVM_TRACE_ENABLE:
2333         case KVM_TRACE_PAUSE:
2334         case KVM_TRACE_DISABLE:
2335                 r = -EOPNOTSUPP;
2336                 break;
2337         default:
2338                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2339         }
2340 out:
2341         return r;
2342 }
2343
2344 static struct file_operations kvm_chardev_ops = {
2345         .unlocked_ioctl = kvm_dev_ioctl,
2346         .compat_ioctl   = kvm_dev_ioctl,
2347         .llseek         = noop_llseek,
2348 };
2349
2350 static struct miscdevice kvm_dev = {
2351         KVM_MINOR,
2352         "kvm",
2353         &kvm_chardev_ops,
2354 };
2355
2356 static void hardware_enable_nolock(void *junk)
2357 {
2358         int cpu = raw_smp_processor_id();
2359         int r;
2360
2361         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2362                 return;
2363
2364         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2365
2366         r = kvm_arch_hardware_enable(NULL);
2367
2368         if (r) {
2369                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2370                 atomic_inc(&hardware_enable_failed);
2371                 printk(KERN_INFO "kvm: enabling virtualization on "
2372                                  "CPU%d failed\n", cpu);
2373         }
2374 }
2375
2376 static void hardware_enable(void *junk)
2377 {
2378         raw_spin_lock(&kvm_lock);
2379         hardware_enable_nolock(junk);
2380         raw_spin_unlock(&kvm_lock);
2381 }
2382
2383 static void hardware_disable_nolock(void *junk)
2384 {
2385         int cpu = raw_smp_processor_id();
2386
2387         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2388                 return;
2389         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2390         kvm_arch_hardware_disable(NULL);
2391 }
2392
2393 static void hardware_disable(void *junk)
2394 {
2395         raw_spin_lock(&kvm_lock);
2396         hardware_disable_nolock(junk);
2397         raw_spin_unlock(&kvm_lock);
2398 }
2399
2400 static void hardware_disable_all_nolock(void)
2401 {
2402         BUG_ON(!kvm_usage_count);
2403
2404         kvm_usage_count--;
2405         if (!kvm_usage_count)
2406                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2407 }
2408
2409 static void hardware_disable_all(void)
2410 {
2411         raw_spin_lock(&kvm_lock);
2412         hardware_disable_all_nolock();
2413         raw_spin_unlock(&kvm_lock);
2414 }
2415
2416 static int hardware_enable_all(void)
2417 {
2418         int r = 0;
2419
2420         raw_spin_lock(&kvm_lock);
2421
2422         kvm_usage_count++;
2423         if (kvm_usage_count == 1) {
2424                 atomic_set(&hardware_enable_failed, 0);
2425                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2426
2427                 if (atomic_read(&hardware_enable_failed)) {
2428                         hardware_disable_all_nolock();
2429                         r = -EBUSY;
2430                 }
2431         }
2432
2433         raw_spin_unlock(&kvm_lock);
2434
2435         return r;
2436 }
2437
2438 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2439                            void *v)
2440 {
2441         int cpu = (long)v;
2442
2443         if (!kvm_usage_count)
2444                 return NOTIFY_OK;
2445
2446         val &= ~CPU_TASKS_FROZEN;
2447         switch (val) {
2448         case CPU_DYING:
2449                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2450                        cpu);
2451                 hardware_disable(NULL);
2452                 break;
2453         case CPU_STARTING:
2454                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2455                        cpu);
2456                 hardware_enable(NULL);
2457                 break;
2458         }
2459         return NOTIFY_OK;
2460 }
2461
2462
2463 asmlinkage void kvm_spurious_fault(void)
2464 {
2465         /* Fault while not rebooting.  We want the trace. */
2466         BUG();
2467 }
2468 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2469
2470 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2471                       void *v)
2472 {
2473         /*
2474          * Some (well, at least mine) BIOSes hang on reboot if
2475          * in vmx root mode.
2476          *
2477          * And Intel TXT required VMX off for all cpu when system shutdown.
2478          */
2479         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2480         kvm_rebooting = true;
2481         on_each_cpu(hardware_disable_nolock, NULL, 1);
2482         return NOTIFY_OK;
2483 }
2484
2485 static struct notifier_block kvm_reboot_notifier = {
2486         .notifier_call = kvm_reboot,
2487         .priority = 0,
2488 };
2489
2490 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2491 {
2492         int i;
2493
2494         for (i = 0; i < bus->dev_count; i++) {
2495                 struct kvm_io_device *pos = bus->range[i].dev;
2496
2497                 kvm_iodevice_destructor(pos);
2498         }
2499         kfree(bus);
2500 }
2501
2502 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2503 {
2504         const struct kvm_io_range *r1 = p1;
2505         const struct kvm_io_range *r2 = p2;
2506
2507         if (r1->addr < r2->addr)
2508                 return -1;
2509         if (r1->addr + r1->len > r2->addr + r2->len)
2510                 return 1;
2511         return 0;
2512 }
2513
2514 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2515                           gpa_t addr, int len)
2516 {
2517         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2518                 .addr = addr,
2519                 .len = len,
2520                 .dev = dev,
2521         };
2522
2523         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2524                 kvm_io_bus_sort_cmp, NULL);
2525
2526         return 0;
2527 }
2528
2529 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2530                              gpa_t addr, int len)
2531 {
2532         struct kvm_io_range *range, key;
2533         int off;
2534
2535         key = (struct kvm_io_range) {
2536                 .addr = addr,
2537                 .len = len,
2538         };
2539
2540         range = bsearch(&key, bus->range, bus->dev_count,
2541                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2542         if (range == NULL)
2543                 return -ENOENT;
2544
2545         off = range - bus->range;
2546
2547         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2548                 off--;
2549
2550         return off;
2551 }
2552
2553 /* kvm_io_bus_write - called under kvm->slots_lock */
2554 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2555                      int len, const void *val)
2556 {
2557         int idx;
2558         struct kvm_io_bus *bus;
2559         struct kvm_io_range range;
2560
2561         range = (struct kvm_io_range) {
2562                 .addr = addr,
2563                 .len = len,
2564         };
2565
2566         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2567         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2568         if (idx < 0)
2569                 return -EOPNOTSUPP;
2570
2571         while (idx < bus->dev_count &&
2572                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2573                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2574                         return 0;
2575                 idx++;
2576         }
2577
2578         return -EOPNOTSUPP;
2579 }
2580
2581 /* kvm_io_bus_read - called under kvm->slots_lock */
2582 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2583                     int len, void *val)
2584 {
2585         int idx;
2586         struct kvm_io_bus *bus;
2587         struct kvm_io_range range;
2588
2589         range = (struct kvm_io_range) {
2590                 .addr = addr,
2591                 .len = len,
2592         };
2593
2594         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2595         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2596         if (idx < 0)
2597                 return -EOPNOTSUPP;
2598
2599         while (idx < bus->dev_count &&
2600                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2601                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2602                         return 0;
2603                 idx++;
2604         }
2605
2606         return -EOPNOTSUPP;
2607 }
2608
2609 /* Caller must hold slots_lock. */
2610 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2611                             int len, struct kvm_io_device *dev)
2612 {
2613         struct kvm_io_bus *new_bus, *bus;
2614
2615         bus = kvm->buses[bus_idx];
2616         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2617                 return -ENOSPC;
2618
2619         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2620                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2621         if (!new_bus)
2622                 return -ENOMEM;
2623         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2624                sizeof(struct kvm_io_range)));
2625         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2626         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2627         synchronize_srcu_expedited(&kvm->srcu);
2628         kfree(bus);
2629
2630         return 0;
2631 }
2632
2633 /* Caller must hold slots_lock. */
2634 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2635                               struct kvm_io_device *dev)
2636 {
2637         int i, r;
2638         struct kvm_io_bus *new_bus, *bus;
2639
2640         bus = kvm->buses[bus_idx];
2641         r = -ENOENT;
2642         for (i = 0; i < bus->dev_count; i++)
2643                 if (bus->range[i].dev == dev) {
2644                         r = 0;
2645                         break;
2646                 }
2647
2648         if (r)
2649                 return r;
2650
2651         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2652                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2653         if (!new_bus)
2654                 return -ENOMEM;
2655
2656         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2657         new_bus->dev_count--;
2658         memcpy(new_bus->range + i, bus->range + i + 1,
2659                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2660
2661         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2662         synchronize_srcu_expedited(&kvm->srcu);
2663         kfree(bus);
2664         return r;
2665 }
2666
2667 static struct notifier_block kvm_cpu_notifier = {
2668         .notifier_call = kvm_cpu_hotplug,
2669 };
2670
2671 static int vm_stat_get(void *_offset, u64 *val)
2672 {
2673         unsigned offset = (long)_offset;
2674         struct kvm *kvm;
2675
2676         *val = 0;
2677         raw_spin_lock(&kvm_lock);
2678         list_for_each_entry(kvm, &vm_list, vm_list)
2679                 *val += *(u32 *)((void *)kvm + offset);
2680         raw_spin_unlock(&kvm_lock);
2681         return 0;
2682 }
2683
2684 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2685
2686 static int vcpu_stat_get(void *_offset, u64 *val)
2687 {
2688         unsigned offset = (long)_offset;
2689         struct kvm *kvm;
2690         struct kvm_vcpu *vcpu;
2691         int i;
2692
2693         *val = 0;
2694         raw_spin_lock(&kvm_lock);
2695         list_for_each_entry(kvm, &vm_list, vm_list)
2696                 kvm_for_each_vcpu(i, vcpu, kvm)
2697                         *val += *(u32 *)((void *)vcpu + offset);
2698
2699         raw_spin_unlock(&kvm_lock);
2700         return 0;
2701 }
2702
2703 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2704
2705 static const struct file_operations *stat_fops[] = {
2706         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2707         [KVM_STAT_VM]   = &vm_stat_fops,
2708 };
2709
2710 static int kvm_init_debug(void)
2711 {
2712         int r = -EFAULT;
2713         struct kvm_stats_debugfs_item *p;
2714
2715         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2716         if (kvm_debugfs_dir == NULL)
2717                 goto out;
2718
2719         for (p = debugfs_entries; p->name; ++p) {
2720                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2721                                                 (void *)(long)p->offset,
2722                                                 stat_fops[p->kind]);
2723                 if (p->dentry == NULL)
2724                         goto out_dir;
2725         }
2726
2727         return 0;
2728
2729 out_dir:
2730         debugfs_remove_recursive(kvm_debugfs_dir);
2731 out:
2732         return r;
2733 }
2734
2735 static void kvm_exit_debug(void)
2736 {
2737         struct kvm_stats_debugfs_item *p;
2738
2739         for (p = debugfs_entries; p->name; ++p)
2740                 debugfs_remove(p->dentry);
2741         debugfs_remove(kvm_debugfs_dir);
2742 }
2743
2744 static int kvm_suspend(void)
2745 {
2746         if (kvm_usage_count)
2747                 hardware_disable_nolock(NULL);
2748         return 0;
2749 }
2750
2751 static void kvm_resume(void)
2752 {
2753         if (kvm_usage_count) {
2754                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2755                 hardware_enable_nolock(NULL);
2756         }
2757 }
2758
2759 static struct syscore_ops kvm_syscore_ops = {
2760         .suspend = kvm_suspend,
2761         .resume = kvm_resume,
2762 };
2763
2764 static inline
2765 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2766 {
2767         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2768 }
2769
2770 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2771 {
2772         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2773
2774         kvm_arch_vcpu_load(vcpu, cpu);
2775 }
2776
2777 static void kvm_sched_out(struct preempt_notifier *pn,
2778                           struct task_struct *next)
2779 {
2780         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2781
2782         kvm_arch_vcpu_put(vcpu);
2783 }
2784
2785 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2786                   struct module *module)
2787 {
2788         int r;
2789         int cpu;
2790
2791         r = kvm_arch_init(opaque);
2792         if (r)
2793                 goto out_fail;
2794
2795         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2796                 r = -ENOMEM;
2797                 goto out_free_0;
2798         }
2799
2800         r = kvm_arch_hardware_setup();
2801         if (r < 0)
2802                 goto out_free_0a;
2803
2804         for_each_online_cpu(cpu) {
2805                 smp_call_function_single(cpu,
2806                                 kvm_arch_check_processor_compat,
2807                                 &r, 1);
2808                 if (r < 0)
2809                         goto out_free_1;
2810         }
2811
2812         r = register_cpu_notifier(&kvm_cpu_notifier);
2813         if (r)
2814                 goto out_free_2;
2815         register_reboot_notifier(&kvm_reboot_notifier);
2816
2817         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2818         if (!vcpu_align)
2819                 vcpu_align = __alignof__(struct kvm_vcpu);
2820         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2821                                            0, NULL);
2822         if (!kvm_vcpu_cache) {
2823                 r = -ENOMEM;
2824                 goto out_free_3;
2825         }
2826
2827         r = kvm_async_pf_init();
2828         if (r)
2829                 goto out_free;
2830
2831         kvm_chardev_ops.owner = module;
2832         kvm_vm_fops.owner = module;
2833         kvm_vcpu_fops.owner = module;
2834
2835         r = misc_register(&kvm_dev);
2836         if (r) {
2837                 printk(KERN_ERR "kvm: misc device register failed\n");
2838                 goto out_unreg;
2839         }
2840
2841         register_syscore_ops(&kvm_syscore_ops);
2842
2843         kvm_preempt_ops.sched_in = kvm_sched_in;
2844         kvm_preempt_ops.sched_out = kvm_sched_out;
2845
2846         r = kvm_init_debug();
2847         if (r) {
2848                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2849                 goto out_undebugfs;
2850         }
2851
2852         return 0;
2853
2854 out_undebugfs:
2855         unregister_syscore_ops(&kvm_syscore_ops);
2856 out_unreg:
2857         kvm_async_pf_deinit();
2858 out_free:
2859         kmem_cache_destroy(kvm_vcpu_cache);
2860 out_free_3:
2861         unregister_reboot_notifier(&kvm_reboot_notifier);
2862         unregister_cpu_notifier(&kvm_cpu_notifier);
2863 out_free_2:
2864 out_free_1:
2865         kvm_arch_hardware_unsetup();
2866 out_free_0a:
2867         free_cpumask_var(cpus_hardware_enabled);
2868 out_free_0:
2869         kvm_arch_exit();
2870 out_fail:
2871         return r;
2872 }
2873 EXPORT_SYMBOL_GPL(kvm_init);
2874
2875 void kvm_exit(void)
2876 {
2877         kvm_exit_debug();
2878         misc_deregister(&kvm_dev);
2879         kmem_cache_destroy(kvm_vcpu_cache);
2880         kvm_async_pf_deinit();
2881         unregister_syscore_ops(&kvm_syscore_ops);
2882         unregister_reboot_notifier(&kvm_reboot_notifier);
2883         unregister_cpu_notifier(&kvm_cpu_notifier);
2884         on_each_cpu(hardware_disable_nolock, NULL, 1);
2885         kvm_arch_hardware_unsetup();
2886         kvm_arch_exit();
2887         free_cpumask_var(cpus_hardware_enabled);
2888 }
2889 EXPORT_SYMBOL_GPL(kvm_exit);