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 (pfn_valid(pfn)) {
 106                 int reserved;
 107                 struct page *tail = pfn_to_page(pfn);
 108                 struct page *head = compound_trans_head(tail);
 109                 reserved = PageReserved(head);
 110                 if (head != tail) {
 111                         /*
 112                          * "head" is not a dangling pointer
 113                          * (compound_trans_head takes care of that)
 114                          * but the hugepage may have been splitted
 115                          * from under us (and we may not hold a
 116                          * reference count on the head page so it can
 117                          * be reused before we run PageReferenced), so
 118                          * we've to check PageTail before returning
 119                          * what we just read.
 120                          */
 121                         smp_rmb();
 122                         if (PageTail(tail))
 123                                 return reserved;
 124                 }
 125                 return PageReserved(tail);
 126         }
 127
 128         return true;
 129 }
 130
 131 /*
 132  * Switches to specified vcpu, until a matching vcpu_put()
 133  */
 134 void vcpu_load(struct kvm_vcpu *vcpu)
 135 {
 136         int cpu;
 137
 138         mutex_lock(&vcpu->mutex);
 139         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 140                 /* The thread running this VCPU changed. */
 141                 struct pid *oldpid = vcpu->pid;
 142                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 143                 rcu_assign_pointer(vcpu->pid, newpid);
 144                 synchronize_rcu();
 145                 put_pid(oldpid);
 146         }
 147         cpu = get_cpu();
 148         preempt_notifier_register(&vcpu->preempt_notifier);
 149         kvm_arch_vcpu_load(vcpu, cpu);
 150         put_cpu();
 151 }
 152
 153 void vcpu_put(struct kvm_vcpu *vcpu)
 154 {
 155         preempt_disable();
 156         kvm_arch_vcpu_put(vcpu);
 157         preempt_notifier_unregister(&vcpu->preempt_notifier);
 158         preempt_enable();
 159         mutex_unlock(&vcpu->mutex);
 160 }
 161
 162 static void ack_flush(void *_completed)
 163 {
 164 }
 165
 166 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 167 {
 168         int i, cpu, me;
 169         cpumask_var_t cpus;
 170         bool called = true;
 171         struct kvm_vcpu *vcpu;
 172
 173         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 174
 175         me = get_cpu();
 176         kvm_for_each_vcpu(i, vcpu, kvm) {
 177                 kvm_make_request(req, vcpu);
 178                 cpu = vcpu->cpu;
 179
 180                 /* Set ->requests bit before we read ->mode */
 181                 smp_mb();
 182
 183                 if (cpus != NULL && cpu != -1 && cpu != me &&
 184                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 185                         cpumask_set_cpu(cpu, cpus);
 186         }
 187         if (unlikely(cpus == NULL))
 188                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 189         else if (!cpumask_empty(cpus))
 190                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 191         else
 192                 called = false;
 193         put_cpu();
 194         free_cpumask_var(cpus);
 195         return called;
 196 }
 197
 198 void kvm_flush_remote_tlbs(struct kvm *kvm)
 199 {
 200         long dirty_count = kvm->tlbs_dirty;
 201
 202         smp_mb();
 203         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 204                 ++kvm->stat.remote_tlb_flush;
 205         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 206 }
 207
 208 void kvm_reload_remote_mmus(struct kvm *kvm)
 209 {
 210         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 211 }
 212
 213 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 214 {
 215         struct page *page;
 216         int r;
 217
 218         mutex_init(&vcpu->mutex);
 219         vcpu->cpu = -1;
 220         vcpu->kvm = kvm;
 221         vcpu->vcpu_id = id;
 222         vcpu->pid = NULL;
 223         init_waitqueue_head(&vcpu->wq);
 224         kvm_async_pf_vcpu_init(vcpu);
 225
 226         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 227         if (!page) {
 228                 r = -ENOMEM;
 229                 goto fail;
 230         }
 231         vcpu->run = page_address(page);
 232
 233         kvm_vcpu_set_in_spin_loop(vcpu, false);
 234         kvm_vcpu_set_dy_eligible(vcpu, false);
 235
 236         r = kvm_arch_vcpu_init(vcpu);
 237         if (r < 0)
 238                 goto fail_free_run;
 239         return 0;
 240
 241 fail_free_run:
 242         free_page((unsigned long)vcpu->run);
 243 fail:
 244         return r;
 245 }
 246 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 247
 248 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 249 {
 250         put_pid(vcpu->pid);
 251         kvm_arch_vcpu_uninit(vcpu);
 252         free_page((unsigned long)vcpu->run);
 253 }
 254 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 255
 256 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 257 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 258 {
 259         return container_of(mn, struct kvm, mmu_notifier);
 260 }
 261
 262 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 263                                              struct mm_struct *mm,
 264                                              unsigned long address)
 265 {
 266         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 267         int need_tlb_flush, idx;
 268
 269         /*
 270          * When ->invalidate_page runs, the linux pte has been zapped
 271          * already but the page is still allocated until
 272          * ->invalidate_page returns. So if we increase the sequence
 273          * here the kvm page fault will notice if the spte can't be
 274          * established because the page is going to be freed. If
 275          * instead the kvm page fault establishes the spte before
 276          * ->invalidate_page runs, kvm_unmap_hva will release it
 277          * before returning.
 278          *
 279          * The sequence increase only need to be seen at spin_unlock
 280          * time, and not at spin_lock time.
 281          *
 282          * Increasing the sequence after the spin_unlock would be
 283          * unsafe because the kvm page fault could then establish the
 284          * pte after kvm_unmap_hva returned, without noticing the page
 285          * is going to be freed.
 286          */
 287         idx = srcu_read_lock(&kvm->srcu);
 288         spin_lock(&kvm->mmu_lock);
 289
 290         kvm->mmu_notifier_seq++;
 291         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 292         /* we've to flush the tlb before the pages can be freed */
 293         if (need_tlb_flush)
 294                 kvm_flush_remote_tlbs(kvm);
 295
 296         spin_unlock(&kvm->mmu_lock);
 297         srcu_read_unlock(&kvm->srcu, idx);
 298 }
 299
 300 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 301                                         struct mm_struct *mm,
 302                                         unsigned long address,
 303                                         pte_t pte)
 304 {
 305         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 306         int idx;
 307
 308         idx = srcu_read_lock(&kvm->srcu);
 309         spin_lock(&kvm->mmu_lock);
 310         kvm->mmu_notifier_seq++;
 311         kvm_set_spte_hva(kvm, address, pte);
 312         spin_unlock(&kvm->mmu_lock);
 313         srcu_read_unlock(&kvm->srcu, idx);
 314 }
 315
 316 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 317                                                     struct mm_struct *mm,
 318                                                     unsigned long start,
 319                                                     unsigned long end)
 320 {
 321         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 322         int need_tlb_flush = 0, idx;
 323
 324         idx = srcu_read_lock(&kvm->srcu);
 325         spin_lock(&kvm->mmu_lock);
 326         /*
 327          * The count increase must become visible at unlock time as no
 328          * spte can be established without taking the mmu_lock and
 329          * count is also read inside the mmu_lock critical section.
 330          */
 331         kvm->mmu_notifier_count++;
 332         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 333         need_tlb_flush |= kvm->tlbs_dirty;
 334         /* we've to flush the tlb before the pages can be freed */
 335         if (need_tlb_flush)
 336                 kvm_flush_remote_tlbs(kvm);
 337
 338         spin_unlock(&kvm->mmu_lock);
 339         srcu_read_unlock(&kvm->srcu, idx);
 340 }
 341
 342 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 343                                                   struct mm_struct *mm,
 344                                                   unsigned long start,
 345                                                   unsigned long end)
 346 {
 347         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 348
 349         spin_lock(&kvm->mmu_lock);
 350         /*
 351          * This sequence increase will notify the kvm page fault that
 352          * the page that is going to be mapped in the spte could have
 353          * been freed.
 354          */
 355         kvm->mmu_notifier_seq++;
 356         smp_wmb();
 357         /*
 358          * The above sequence increase must be visible before the
 359          * below count decrease, which is ensured by the smp_wmb above
 360          * in conjunction with the smp_rmb in mmu_notifier_retry().
 361          */
 362         kvm->mmu_notifier_count--;
 363         spin_unlock(&kvm->mmu_lock);
 364
 365         BUG_ON(kvm->mmu_notifier_count < 0);
 366 }
 367
 368 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 369                                               struct mm_struct *mm,
 370                                               unsigned long address)
 371 {
 372         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 373         int young, idx;
 374
 375         idx = srcu_read_lock(&kvm->srcu);
 376         spin_lock(&kvm->mmu_lock);
 377
 378         young = kvm_age_hva(kvm, address);
 379         if (young)
 380                 kvm_flush_remote_tlbs(kvm);
 381
 382         spin_unlock(&kvm->mmu_lock);
 383         srcu_read_unlock(&kvm->srcu, idx);
 384
 385         return young;
 386 }
 387
 388 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 389                                        struct mm_struct *mm,
 390                                        unsigned long address)
 391 {
 392         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 393         int young, idx;
 394
 395         idx = srcu_read_lock(&kvm->srcu);
 396         spin_lock(&kvm->mmu_lock);
 397         young = kvm_test_age_hva(kvm, address);
 398         spin_unlock(&kvm->mmu_lock);
 399         srcu_read_unlock(&kvm->srcu, idx);
 400
 401         return young;
 402 }
 403
 404 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 405                                      struct mm_struct *mm)
 406 {
 407         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 408         int idx;
 409
 410         idx = srcu_read_lock(&kvm->srcu);
 411         kvm_arch_flush_shadow(kvm);
 412         srcu_read_unlock(&kvm->srcu, idx);
 413 }
 414
 415 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 416         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 417         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 418         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 419         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 420         .test_young             = kvm_mmu_notifier_test_young,
 421         .change_pte             = kvm_mmu_notifier_change_pte,
 422         .release                = kvm_mmu_notifier_release,
 423 };
 424
 425 static int kvm_init_mmu_notifier(struct kvm *kvm)
 426 {
 427         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 428         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 429 }
 430
 431 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 432
 433 static int kvm_init_mmu_notifier(struct kvm *kvm)
 434 {
 435         return 0;
 436 }
 437
 438 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 439
 440 static void kvm_init_memslots_id(struct kvm *kvm)
 441 {
 442         int i;
 443         struct kvm_memslots *slots = kvm->memslots;
 444
 445         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 446                 slots->id_to_index[i] = slots->memslots[i].id = i;
 447 }
 448
 449 static struct kvm *kvm_create_vm(unsigned long type)
 450 {
 451         int r, i;
 452         struct kvm *kvm = kvm_arch_alloc_vm();
 453
 454         if (!kvm)
 455                 return ERR_PTR(-ENOMEM);
 456
 457         r = kvm_arch_init_vm(kvm, type);
 458         if (r)
 459                 goto out_err_nodisable;
 460
 461         r = hardware_enable_all();
 462         if (r)
 463                 goto out_err_nodisable;
 464
 465 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 466         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 467         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 468 #endif
 469
 470         r = -ENOMEM;
 471         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 472         if (!kvm->memslots)
 473                 goto out_err_nosrcu;
 474         kvm_init_memslots_id(kvm);
 475         if (init_srcu_struct(&kvm->srcu))
 476                 goto out_err_nosrcu;
 477         for (i = 0; i < KVM_NR_BUSES; i++) {
 478                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 479                                         GFP_KERNEL);
 480                 if (!kvm->buses[i])
 481                         goto out_err;
 482         }
 483
 484         spin_lock_init(&kvm->mmu_lock);
 485         kvm->mm = current->mm;
 486         atomic_inc(&kvm->mm->mm_count);
 487         kvm_eventfd_init(kvm);
 488         mutex_init(&kvm->lock);
 489         mutex_init(&kvm->irq_lock);
 490         mutex_init(&kvm->slots_lock);
 491         atomic_set(&kvm->users_count, 1);
 492
 493         r = kvm_init_mmu_notifier(kvm);
 494         if (r)
 495                 goto out_err;
 496
 497         raw_spin_lock(&kvm_lock);
 498         list_add(&kvm->vm_list, &vm_list);
 499         raw_spin_unlock(&kvm_lock);
 500
 501         return kvm;
 502
 503 out_err:
 504         cleanup_srcu_struct(&kvm->srcu);
 505 out_err_nosrcu:
 506         hardware_disable_all();
 507 out_err_nodisable:
 508         for (i = 0; i < KVM_NR_BUSES; i++)
 509                 kfree(kvm->buses[i]);
 510         kfree(kvm->memslots);
 511         kvm_arch_free_vm(kvm);
 512         return ERR_PTR(r);
 513 }
 514
 515 /*
 516  * Avoid using vmalloc for a small buffer.
 517  * Should not be used when the size is statically known.
 518  */
 519 void *kvm_kvzalloc(unsigned long size)
 520 {
 521         if (size > PAGE_SIZE)
 522                 return vzalloc(size);
 523         else
 524                 return kzalloc(size, GFP_KERNEL);
 525 }
 526
 527 void kvm_kvfree(const void *addr)
 528 {
 529         if (is_vmalloc_addr(addr))
 530                 vfree(addr);
 531         else
 532                 kfree(addr);
 533 }
 534
 535 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 536 {
 537         if (!memslot->dirty_bitmap)
 538                 return;
 539
 540         kvm_kvfree(memslot->dirty_bitmap);
 541         memslot->dirty_bitmap = NULL;
 542 }
 543
 544 /*
 545  * Free any memory in @free but not in @dont.
 546  */
 547 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 548                                   struct kvm_memory_slot *dont)
 549 {
 550         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 551                 kvm_destroy_dirty_bitmap(free);
 552
 553         kvm_arch_free_memslot(free, dont);
 554
 555         free->npages = 0;
 556 }
 557
 558 void kvm_free_physmem(struct kvm *kvm)
 559 {
 560         struct kvm_memslots *slots = kvm->memslots;
 561         struct kvm_memory_slot *memslot;
 562
 563         kvm_for_each_memslot(memslot, slots)
 564                 kvm_free_physmem_slot(memslot, NULL);
 565
 566         kfree(kvm->memslots);
 567 }
 568
 569 static void kvm_destroy_vm(struct kvm *kvm)
 570 {
 571         int i;
 572         struct mm_struct *mm = kvm->mm;
 573
 574         kvm_arch_sync_events(kvm);
 575         raw_spin_lock(&kvm_lock);
 576         list_del(&kvm->vm_list);
 577         raw_spin_unlock(&kvm_lock);
 578         kvm_free_irq_routing(kvm);
 579         for (i = 0; i < KVM_NR_BUSES; i++)
 580                 kvm_io_bus_destroy(kvm->buses[i]);
 581         kvm_coalesced_mmio_free(kvm);
 582 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 583         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 584 #else
 585         kvm_arch_flush_shadow(kvm);
 586 #endif
 587         kvm_arch_destroy_vm(kvm);
 588         kvm_free_physmem(kvm);
 589         cleanup_srcu_struct(&kvm->srcu);
 590         kvm_arch_free_vm(kvm);
 591         hardware_disable_all();
 592         mmdrop(mm);
 593 }
 594
 595 void kvm_get_kvm(struct kvm *kvm)
 596 {
 597         atomic_inc(&kvm->users_count);
 598 }
 599 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 600
 601 void kvm_put_kvm(struct kvm *kvm)
 602 {
 603         if (atomic_dec_and_test(&kvm->users_count))
 604                 kvm_destroy_vm(kvm);
 605 }
 606 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 607
 608
 609 static int kvm_vm_release(struct inode *inode, struct file *filp)
 610 {
 611         struct kvm *kvm = filp->private_data;
 612
 613         kvm_irqfd_release(kvm);
 614
 615         kvm_put_kvm(kvm);
 616         return 0;
 617 }
 618
 619 /*
 620  * Allocation size is twice as large as the actual dirty bitmap size.
 621  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 622  */
 623 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 624 {
 625 #ifndef CONFIG_S390
 626         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 627
 628         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 629         if (!memslot->dirty_bitmap)
 630                 return -ENOMEM;
 631
 632 #endif /* !CONFIG_S390 */
 633         return 0;
 634 }
 635
 636 static int cmp_memslot(const void *slot1, const void *slot2)
 637 {
 638         struct kvm_memory_slot *s1, *s2;
 639
 640         s1 = (struct kvm_memory_slot *)slot1;
 641         s2 = (struct kvm_memory_slot *)slot2;
 642
 643         if (s1->npages < s2->npages)
 644                 return 1;
 645         if (s1->npages > s2->npages)
 646                 return -1;
 647
 648         return 0;
 649 }
 650
 651 /*
 652  * Sort the memslots base on its size, so the larger slots
 653  * will get better fit.
 654  */
 655 static void sort_memslots(struct kvm_memslots *slots)
 656 {
 657         int i;
 658
 659         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 660               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 661
 662         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 663                 slots->id_to_index[slots->memslots[i].id] = i;
 664 }
 665
 666 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
 667 {
 668         if (new) {
 669                 int id = new->id;
 670                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 671                 unsigned long npages = old->npages;
 672
 673                 *old = *new;
 674                 if (new->npages != npages)
 675                         sort_memslots(slots);
 676         }
 677
 678         slots->generation++;
 679 }
 680
 681 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 682 {
 683         if (mem->flags & ~KVM_MEM_LOG_DIRTY_PAGES)
 684                 return -EINVAL;
 685
 686         return 0;
 687 }
 688
 689 /*
 690  * Allocate some memory and give it an address in the guest physical address
 691  * space.
 692  *
 693  * Discontiguous memory is allowed, mostly for framebuffers.
 694  *
 695  * Must be called holding mmap_sem for write.
 696  */
 697 int __kvm_set_memory_region(struct kvm *kvm,
 698                             struct kvm_userspace_memory_region *mem,
 699                             int user_alloc)
 700 {
 701         int r;
 702         gfn_t base_gfn;
 703         unsigned long npages;
 704         unsigned long i;
 705         struct kvm_memory_slot *memslot;
 706         struct kvm_memory_slot old, new;
 707         struct kvm_memslots *slots, *old_memslots;
 708
 709         r = check_memory_region_flags(mem);
 710         if (r)
 711                 goto out;
 712
 713         r = -EINVAL;
 714         /* General sanity checks */
 715         if (mem->memory_size & (PAGE_SIZE - 1))
 716                 goto out;
 717         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 718                 goto out;
 719         /* We can read the guest memory with __xxx_user() later on. */
 720         if (user_alloc &&
 721             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 722              !access_ok(VERIFY_WRITE,
 723                         (void __user *)(unsigned long)mem->userspace_addr,
 724                         mem->memory_size)))
 725                 goto out;
 726         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 727                 goto out;
 728         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 729                 goto out;
 730
 731         memslot = id_to_memslot(kvm->memslots, mem->slot);
 732         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 733         npages = mem->memory_size >> PAGE_SHIFT;
 734
 735         r = -EINVAL;
 736         if (npages > KVM_MEM_MAX_NR_PAGES)
 737                 goto out;
 738
 739         if (!npages)
 740                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 741
 742         new = old = *memslot;
 743
 744         new.id = mem->slot;
 745         new.base_gfn = base_gfn;
 746         new.npages = npages;
 747         new.flags = mem->flags;
 748
 749         /* Disallow changing a memory slot's size. */
 750         r = -EINVAL;
 751         if (npages && old.npages && npages != old.npages)
 752                 goto out_free;
 753
 754         /* Check for overlaps */
 755         r = -EEXIST;
 756         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 757                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 758
 759                 if (s == memslot || !s->npages)
 760                         continue;
 761                 if (!((base_gfn + npages <= s->base_gfn) ||
 762                       (base_gfn >= s->base_gfn + s->npages)))
 763                         goto out_free;
 764         }
 765
 766         /* Free page dirty bitmap if unneeded */
 767         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 768                 new.dirty_bitmap = NULL;
 769
 770         r = -ENOMEM;
 771
 772         /* Allocate if a slot is being created */
 773         if (npages && !old.npages) {
 774                 new.user_alloc = user_alloc;
 775                 new.userspace_addr = mem->userspace_addr;
 776
 777                 if (kvm_arch_create_memslot(&new, npages))
 778                         goto out_free;
 779         }
 780
 781         /* Allocate page dirty bitmap if needed */
 782         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 783                 if (kvm_create_dirty_bitmap(&new) < 0)
 784                         goto out_free;
 785                 /* destroy any largepage mappings for dirty tracking */
 786         }
 787
 788         if (!npages) {
 789                 struct kvm_memory_slot *slot;
 790
 791                 r = -ENOMEM;
 792                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 793                                 GFP_KERNEL);
 794                 if (!slots)
 795                         goto out_free;
 796                 slot = id_to_memslot(slots, mem->slot);
 797                 slot->flags |= KVM_MEMSLOT_INVALID;
 798
 799                 update_memslots(slots, NULL);
 800
 801                 old_memslots = kvm->memslots;
 802                 rcu_assign_pointer(kvm->memslots, slots);
 803                 synchronize_srcu_expedited(&kvm->srcu);
 804                 /* From this point no new shadow pages pointing to a deleted
 805                  * memslot will be created.
 806                  *
 807                  * validation of sp->gfn happens in:
 808                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 809                  *      - kvm_is_visible_gfn (mmu_check_roots)
 810                  */
 811                 kvm_arch_flush_shadow(kvm);
 812                 kfree(old_memslots);
 813         }
 814
 815         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 816         if (r)
 817                 goto out_free;
 818
 819         /* map/unmap the pages in iommu page table */
 820         if (npages) {
 821                 r = kvm_iommu_map_pages(kvm, &new);
 822                 if (r)
 823                         goto out_free;
 824         } else
 825                 kvm_iommu_unmap_pages(kvm, &old);
 826
 827         r = -ENOMEM;
 828         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 829                         GFP_KERNEL);
 830         if (!slots)
 831                 goto out_free;
 832
 833         /* actual memory is freed via old in kvm_free_physmem_slot below */
 834         if (!npages) {
 835                 new.dirty_bitmap = NULL;
 836                 memset(&new.arch, 0, sizeof(new.arch));
 837         }
 838
 839         update_memslots(slots, &new);
 840         old_memslots = kvm->memslots;
 841         rcu_assign_pointer(kvm->memslots, slots);
 842         synchronize_srcu_expedited(&kvm->srcu);
 843
 844         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 845
 846         /*
 847          * If the new memory slot is created, we need to clear all
 848          * mmio sptes.
 849          */
 850         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 851                 kvm_arch_flush_shadow(kvm);
 852
 853         kvm_free_physmem_slot(&old, &new);
 854         kfree(old_memslots);
 855
 856         return 0;
 857
 858 out_free:
 859         kvm_free_physmem_slot(&new, &old);
 860 out:
 861         return r;
 862
 863 }
 864 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 865
 866 int kvm_set_memory_region(struct kvm *kvm,
 867                           struct kvm_userspace_memory_region *mem,
 868                           int user_alloc)
 869 {
 870         int r;
 871
 872         mutex_lock(&kvm->slots_lock);
 873         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 874         mutex_unlock(&kvm->slots_lock);
 875         return r;
 876 }
 877 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 878
 879 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 880                                    struct
 881                                    kvm_userspace_memory_region *mem,
 882                                    int user_alloc)
 883 {
 884         if (mem->slot >= KVM_MEMORY_SLOTS)
 885                 return -EINVAL;
 886         return kvm_set_memory_region(kvm, mem, user_alloc);
 887 }
 888
 889 int kvm_get_dirty_log(struct kvm *kvm,
 890                         struct kvm_dirty_log *log, int *is_dirty)
 891 {
 892         struct kvm_memory_slot *memslot;
 893         int r, i;
 894         unsigned long n;
 895         unsigned long any = 0;
 896
 897         r = -EINVAL;
 898         if (log->slot >= KVM_MEMORY_SLOTS)
 899                 goto out;
 900
 901         memslot = id_to_memslot(kvm->memslots, log->slot);
 902         r = -ENOENT;
 903         if (!memslot->dirty_bitmap)
 904                 goto out;
 905
 906         n = kvm_dirty_bitmap_bytes(memslot);
 907
 908         for (i = 0; !any && i < n/sizeof(long); ++i)
 909                 any = memslot->dirty_bitmap[i];
 910
 911         r = -EFAULT;
 912         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 913                 goto out;
 914
 915         if (any)
 916                 *is_dirty = 1;
 917
 918         r = 0;
 919 out:
 920         return r;
 921 }
 922
 923 bool kvm_largepages_enabled(void)
 924 {
 925         return largepages_enabled;
 926 }
 927
 928 void kvm_disable_largepages(void)
 929 {
 930         largepages_enabled = false;
 931 }
 932 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 933
 934 static inline unsigned long bad_hva(void)
 935 {
 936         return PAGE_OFFSET;
 937 }
 938
 939 int kvm_is_error_hva(unsigned long addr)
 940 {
 941         return addr == bad_hva();
 942 }
 943 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 944
 945 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 946 {
 947         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 948 }
 949 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 950
 951 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 952 {
 953         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 954
 955         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
 956               memslot->flags & KVM_MEMSLOT_INVALID)
 957                 return 0;
 958
 959         return 1;
 960 }
 961 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 962
 963 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 964 {
 965         struct vm_area_struct *vma;
 966         unsigned long addr, size;
 967
 968         size = PAGE_SIZE;
 969
 970         addr = gfn_to_hva(kvm, gfn);
 971         if (kvm_is_error_hva(addr))
 972                 return PAGE_SIZE;
 973
 974         down_read(&current->mm->mmap_sem);
 975         vma = find_vma(current->mm, addr);
 976         if (!vma)
 977                 goto out;
 978
 979         size = vma_kernel_pagesize(vma);
 980
 981 out:
 982         up_read(&current->mm->mmap_sem);
 983
 984         return size;
 985 }
 986
 987 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
 988                                      gfn_t *nr_pages)
 989 {
 990         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 991                 return bad_hva();
 992
 993         if (nr_pages)
 994                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
 995
 996         return gfn_to_hva_memslot(slot, gfn);
 997 }
 998
 999 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1000 {
1001         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1002 }
1003 EXPORT_SYMBOL_GPL(gfn_to_hva);
1004
1005 /*
1006  * The hva returned by this function is only allowed to be read.
1007  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1008  */
1009 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1010 {
1011         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1012 }
1013
1014 static int kvm_read_hva(void *data, void __user *hva, int len)
1015 {
1016         return __copy_from_user(data, hva, len);
1017 }
1018
1019 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1020 {
1021         return __copy_from_user_inatomic(data, hva, len);
1022 }
1023
1024 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1025         unsigned long start, int write, struct page **page)
1026 {
1027         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1028
1029         if (write)
1030                 flags |= FOLL_WRITE;
1031
1032         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1033 }
1034
1035 static inline int check_user_page_hwpoison(unsigned long addr)
1036 {
1037         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1038
1039         rc = __get_user_pages(current, current->mm, addr, 1,
1040                               flags, NULL, NULL, NULL);
1041         return rc == -EHWPOISON;
1042 }
1043
1044 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1045                         bool write_fault, bool *writable)
1046 {
1047         struct page *page[1];
1048         int npages = 0;
1049         pfn_t pfn;
1050
1051         /* we can do it either atomically or asynchronously, not both */
1052         BUG_ON(atomic && async);
1053
1054         BUG_ON(!write_fault && !writable);
1055
1056         if (writable)
1057                 *writable = true;
1058
1059         if (atomic || async)
1060                 npages = __get_user_pages_fast(addr, 1, 1, page);
1061
1062         if (unlikely(npages != 1) && !atomic) {
1063                 might_sleep();
1064
1065                 if (writable)
1066                         *writable = write_fault;
1067
1068                 if (async) {
1069                         down_read(&current->mm->mmap_sem);
1070                         npages = get_user_page_nowait(current, current->mm,
1071                                                      addr, write_fault, page);
1072                         up_read(&current->mm->mmap_sem);
1073                 } else
1074                         npages = get_user_pages_fast(addr, 1, write_fault,
1075                                                      page);
1076
1077                 /* map read fault as writable if possible */
1078                 if (unlikely(!write_fault) && npages == 1) {
1079                         struct page *wpage[1];
1080
1081                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1082                         if (npages == 1) {
1083                                 *writable = true;
1084                                 put_page(page[0]);
1085                                 page[0] = wpage[0];
1086                         }
1087                         npages = 1;
1088                 }
1089         }
1090
1091         if (unlikely(npages != 1)) {
1092                 struct vm_area_struct *vma;
1093
1094                 if (atomic)
1095                         return KVM_PFN_ERR_FAULT;
1096
1097                 down_read(&current->mm->mmap_sem);
1098                 if (npages == -EHWPOISON ||
1099                         (!async && check_user_page_hwpoison(addr))) {
1100                         up_read(&current->mm->mmap_sem);
1101                         return KVM_PFN_ERR_HWPOISON;
1102                 }
1103
1104                 vma = find_vma_intersection(current->mm, addr, addr+1);
1105
1106                 if (vma == NULL)
1107                         pfn = KVM_PFN_ERR_FAULT;
1108                 else if ((vma->vm_flags & VM_PFNMAP)) {
1109                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1110                                 vma->vm_pgoff;
1111                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1112                 } else {
1113                         if (async && (vma->vm_flags & VM_WRITE))
1114                                 *async = true;
1115                         pfn = KVM_PFN_ERR_FAULT;
1116                 }
1117                 up_read(&current->mm->mmap_sem);
1118         } else
1119                 pfn = page_to_pfn(page[0]);
1120
1121         return pfn;
1122 }
1123
1124 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1125                           bool write_fault, bool *writable)
1126 {
1127         unsigned long addr;
1128
1129         if (async)
1130                 *async = false;
1131
1132         addr = gfn_to_hva(kvm, gfn);
1133         if (kvm_is_error_hva(addr))
1134                 return KVM_PFN_ERR_BAD;
1135
1136         return hva_to_pfn(addr, atomic, async, write_fault, writable);
1137 }
1138
1139 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1140 {
1141         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1142 }
1143 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1144
1145 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1146                        bool write_fault, bool *writable)
1147 {
1148         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1149 }
1150 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1151
1152 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1153 {
1154         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1155 }
1156 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1157
1158 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1159                       bool *writable)
1160 {
1161         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1162 }
1163 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1164
1165 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1166 {
1167         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1168         return hva_to_pfn(addr, false, NULL, true, NULL);
1169 }
1170
1171 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1172 {
1173         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1174
1175         return hva_to_pfn(addr, true, NULL, true, NULL);
1176 }
1177 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1178
1179 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1180                                                                   int nr_pages)
1181 {
1182         unsigned long addr;
1183         gfn_t entry;
1184
1185         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1186         if (kvm_is_error_hva(addr))
1187                 return -1;
1188
1189         if (entry < nr_pages)
1190                 return 0;
1191
1192         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1193 }
1194 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1195
1196 static struct page *kvm_pfn_to_page(pfn_t pfn)
1197 {
1198         if (is_error_pfn(pfn))
1199                 return KVM_ERR_PTR_BAD_PAGE;
1200
1201         if (kvm_is_mmio_pfn(pfn)) {
1202                 WARN_ON(1);
1203                 return KVM_ERR_PTR_BAD_PAGE;
1204         }
1205
1206         return pfn_to_page(pfn);
1207 }
1208
1209 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1210 {
1211         pfn_t pfn;
1212
1213         pfn = gfn_to_pfn(kvm, gfn);
1214
1215         return kvm_pfn_to_page(pfn);
1216 }
1217
1218 EXPORT_SYMBOL_GPL(gfn_to_page);
1219
1220 void kvm_release_page_clean(struct page *page)
1221 {
1222         WARN_ON(is_error_page(page));
1223
1224         kvm_release_pfn_clean(page_to_pfn(page));
1225 }
1226 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1227
1228 void kvm_release_pfn_clean(pfn_t pfn)
1229 {
1230         WARN_ON(is_error_pfn(pfn));
1231
1232         if (!kvm_is_mmio_pfn(pfn))
1233                 put_page(pfn_to_page(pfn));
1234 }
1235 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1236
1237 void kvm_release_page_dirty(struct page *page)
1238 {
1239         WARN_ON(is_error_page(page));
1240
1241         kvm_release_pfn_dirty(page_to_pfn(page));
1242 }
1243 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1244
1245 void kvm_release_pfn_dirty(pfn_t pfn)
1246 {
1247         kvm_set_pfn_dirty(pfn);
1248         kvm_release_pfn_clean(pfn);
1249 }
1250 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1251
1252 void kvm_set_page_dirty(struct page *page)
1253 {
1254         kvm_set_pfn_dirty(page_to_pfn(page));
1255 }
1256 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1257
1258 void kvm_set_pfn_dirty(pfn_t pfn)
1259 {
1260         if (!kvm_is_mmio_pfn(pfn)) {
1261                 struct page *page = pfn_to_page(pfn);
1262                 if (!PageReserved(page))
1263                         SetPageDirty(page);
1264         }
1265 }
1266 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1267
1268 void kvm_set_pfn_accessed(pfn_t pfn)
1269 {
1270         if (!kvm_is_mmio_pfn(pfn))
1271                 mark_page_accessed(pfn_to_page(pfn));
1272 }
1273 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1274
1275 void kvm_get_pfn(pfn_t pfn)
1276 {
1277         if (!kvm_is_mmio_pfn(pfn))
1278                 get_page(pfn_to_page(pfn));
1279 }
1280 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1281
1282 static int next_segment(unsigned long len, int offset)
1283 {
1284         if (len > PAGE_SIZE - offset)
1285                 return PAGE_SIZE - offset;
1286         else
1287                 return len;
1288 }
1289
1290 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1291                         int len)
1292 {
1293         int r;
1294         unsigned long addr;
1295
1296         addr = gfn_to_hva_read(kvm, gfn);
1297         if (kvm_is_error_hva(addr))
1298                 return -EFAULT;
1299         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1300         if (r)
1301                 return -EFAULT;
1302         return 0;
1303 }
1304 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1305
1306 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1307 {
1308         gfn_t gfn = gpa >> PAGE_SHIFT;
1309         int seg;
1310         int offset = offset_in_page(gpa);
1311         int ret;
1312
1313         while ((seg = next_segment(len, offset)) != 0) {
1314                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1315                 if (ret < 0)
1316                         return ret;
1317                 offset = 0;
1318                 len -= seg;
1319                 data += seg;
1320                 ++gfn;
1321         }
1322         return 0;
1323 }
1324 EXPORT_SYMBOL_GPL(kvm_read_guest);
1325
1326 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1327                           unsigned long len)
1328 {
1329         int r;
1330         unsigned long addr;
1331         gfn_t gfn = gpa >> PAGE_SHIFT;
1332         int offset = offset_in_page(gpa);
1333
1334         addr = gfn_to_hva_read(kvm, gfn);
1335         if (kvm_is_error_hva(addr))
1336                 return -EFAULT;
1337         pagefault_disable();
1338         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1339         pagefault_enable();
1340         if (r)
1341                 return -EFAULT;
1342         return 0;
1343 }
1344 EXPORT_SYMBOL(kvm_read_guest_atomic);
1345
1346 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1347                          int offset, int len)
1348 {
1349         int r;
1350         unsigned long addr;
1351
1352         addr = gfn_to_hva(kvm, gfn);
1353         if (kvm_is_error_hva(addr))
1354                 return -EFAULT;
1355         r = __copy_to_user((void __user *)addr + offset, data, len);
1356         if (r)
1357                 return -EFAULT;
1358         mark_page_dirty(kvm, gfn);
1359         return 0;
1360 }
1361 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1362
1363 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1364                     unsigned long len)
1365 {
1366         gfn_t gfn = gpa >> PAGE_SHIFT;
1367         int seg;
1368         int offset = offset_in_page(gpa);
1369         int ret;
1370
1371         while ((seg = next_segment(len, offset)) != 0) {
1372                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1373                 if (ret < 0)
1374                         return ret;
1375                 offset = 0;
1376                 len -= seg;
1377                 data += seg;
1378                 ++gfn;
1379         }
1380         return 0;
1381 }
1382
1383 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1384                               gpa_t gpa)
1385 {
1386         struct kvm_memslots *slots = kvm_memslots(kvm);
1387         int offset = offset_in_page(gpa);
1388         gfn_t gfn = gpa >> PAGE_SHIFT;
1389
1390         ghc->gpa = gpa;
1391         ghc->generation = slots->generation;
1392         ghc->memslot = gfn_to_memslot(kvm, gfn);
1393         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1394         if (!kvm_is_error_hva(ghc->hva))
1395                 ghc->hva += offset;
1396         else
1397                 return -EFAULT;
1398
1399         return 0;
1400 }
1401 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1402
1403 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1404                            void *data, unsigned long len)
1405 {
1406         struct kvm_memslots *slots = kvm_memslots(kvm);
1407         int r;
1408
1409         if (slots->generation != ghc->generation)
1410                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1411
1412         if (kvm_is_error_hva(ghc->hva))
1413                 return -EFAULT;
1414
1415         r = __copy_to_user((void __user *)ghc->hva, data, len);
1416         if (r)
1417                 return -EFAULT;
1418         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1419
1420         return 0;
1421 }
1422 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1423
1424 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1425                            void *data, unsigned long len)
1426 {
1427         struct kvm_memslots *slots = kvm_memslots(kvm);
1428         int r;
1429
1430         if (slots->generation != ghc->generation)
1431                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1432
1433         if (kvm_is_error_hva(ghc->hva))
1434                 return -EFAULT;
1435
1436         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1437         if (r)
1438                 return -EFAULT;
1439
1440         return 0;
1441 }
1442 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1443
1444 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1445 {
1446         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1447                                     offset, len);
1448 }
1449 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1450
1451 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1452 {
1453         gfn_t gfn = gpa >> PAGE_SHIFT;
1454         int seg;
1455         int offset = offset_in_page(gpa);
1456         int ret;
1457
1458         while ((seg = next_segment(len, offset)) != 0) {
1459                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1460                 if (ret < 0)
1461                         return ret;
1462                 offset = 0;
1463                 len -= seg;
1464                 ++gfn;
1465         }
1466         return 0;
1467 }
1468 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1469
1470 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1471                              gfn_t gfn)
1472 {
1473         if (memslot && memslot->dirty_bitmap) {
1474                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1475
1476                 /* TODO: introduce set_bit_le() and use it */
1477                 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1478         }
1479 }
1480
1481 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1482 {
1483         struct kvm_memory_slot *memslot;
1484
1485         memslot = gfn_to_memslot(kvm, gfn);
1486         mark_page_dirty_in_slot(kvm, memslot, gfn);
1487 }
1488
1489 /*
1490  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1491  */
1492 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1493 {
1494         DEFINE_WAIT(wait);
1495
1496         for (;;) {
1497                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1498
1499                 if (kvm_arch_vcpu_runnable(vcpu)) {
1500                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1501                         break;
1502                 }
1503                 if (kvm_cpu_has_pending_timer(vcpu))
1504                         break;
1505                 if (signal_pending(current))
1506                         break;
1507
1508                 schedule();
1509         }
1510
1511         finish_wait(&vcpu->wq, &wait);
1512 }
1513
1514 #ifndef CONFIG_S390
1515 /*
1516  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1517  */
1518 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1519 {
1520         int me;
1521         int cpu = vcpu->cpu;
1522         wait_queue_head_t *wqp;
1523
1524         wqp = kvm_arch_vcpu_wq(vcpu);
1525         if (waitqueue_active(wqp)) {
1526                 wake_up_interruptible(wqp);
1527                 ++vcpu->stat.halt_wakeup;
1528         }
1529
1530         me = get_cpu();
1531         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1532                 if (kvm_arch_vcpu_should_kick(vcpu))
1533                         smp_send_reschedule(cpu);
1534         put_cpu();
1535 }
1536 #endif /* !CONFIG_S390 */
1537
1538 void kvm_resched(struct kvm_vcpu *vcpu)
1539 {
1540         if (!need_resched())
1541                 return;
1542         cond_resched();
1543 }
1544 EXPORT_SYMBOL_GPL(kvm_resched);
1545
1546 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1547 {
1548         struct pid *pid;
1549         struct task_struct *task = NULL;
1550
1551         rcu_read_lock();
1552         pid = rcu_dereference(target->pid);
1553         if (pid)
1554                 task = get_pid_task(target->pid, PIDTYPE_PID);
1555         rcu_read_unlock();
1556         if (!task)
1557                 return false;
1558         if (task->flags & PF_VCPU) {
1559                 put_task_struct(task);
1560                 return false;
1561         }
1562         if (yield_to(task, 1)) {
1563                 put_task_struct(task);
1564                 return true;
1565         }
1566         put_task_struct(task);
1567         return false;
1568 }
1569 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1570
1571 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1572 /*
1573  * Helper that checks whether a VCPU is eligible for directed yield.
1574  * Most eligible candidate to yield is decided by following heuristics:
1575  *
1576  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1577  *  (preempted lock holder), indicated by @in_spin_loop.
1578  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1579  *
1580  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1581  *  chance last time (mostly it has become eligible now since we have probably
1582  *  yielded to lockholder in last iteration. This is done by toggling
1583  *  @dy_eligible each time a VCPU checked for eligibility.)
1584  *
1585  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1586  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1587  *  burning. Giving priority for a potential lock-holder increases lock
1588  *  progress.
1589  *
1590  *  Since algorithm is based on heuristics, accessing another VCPU data without
1591  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1592  *  and continue with next VCPU and so on.
1593  */
1594 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1595 {
1596         bool eligible;
1597
1598         eligible = !vcpu->spin_loop.in_spin_loop ||
1599                         (vcpu->spin_loop.in_spin_loop &&
1600                          vcpu->spin_loop.dy_eligible);
1601
1602         if (vcpu->spin_loop.in_spin_loop)
1603                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1604
1605         return eligible;
1606 }
1607 #endif
1608 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1609 {
1610         struct kvm *kvm = me->kvm;
1611         struct kvm_vcpu *vcpu;
1612         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1613         int yielded = 0;
1614         int pass;
1615         int i;
1616
1617         kvm_vcpu_set_in_spin_loop(me, true);
1618         /*
1619          * We boost the priority of a VCPU that is runnable but not
1620          * currently running, because it got preempted by something
1621          * else and called schedule in __vcpu_run.  Hopefully that
1622          * VCPU is holding the lock that we need and will release it.
1623          * We approximate round-robin by starting at the last boosted VCPU.
1624          */
1625         for (pass = 0; pass < 2 && !yielded; pass++) {
1626                 kvm_for_each_vcpu(i, vcpu, kvm) {
1627                         if (!pass && i <= last_boosted_vcpu) {
1628                                 i = last_boosted_vcpu;
1629                                 continue;
1630                         } else if (pass && i > last_boosted_vcpu)
1631                                 break;
1632                         if (vcpu == me)
1633                                 continue;
1634                         if (waitqueue_active(&vcpu->wq))
1635                                 continue;
1636                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1637                                 continue;
1638                         if (kvm_vcpu_yield_to(vcpu)) {
1639                                 kvm->last_boosted_vcpu = i;
1640                                 yielded = 1;
1641                                 break;
1642                         }
1643                 }
1644         }
1645         kvm_vcpu_set_in_spin_loop(me, false);
1646
1647         /* Ensure vcpu is not eligible during next spinloop */
1648         kvm_vcpu_set_dy_eligible(me, false);
1649 }
1650 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1651
1652 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1653 {
1654         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1655         struct page *page;
1656
1657         if (vmf->pgoff == 0)
1658                 page = virt_to_page(vcpu->run);
1659 #ifdef CONFIG_X86
1660         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1661                 page = virt_to_page(vcpu->arch.pio_data);
1662 #endif
1663 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1664         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1665                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1666 #endif
1667         else
1668                 return kvm_arch_vcpu_fault(vcpu, vmf);
1669         get_page(page);
1670         vmf->page = page;
1671         return 0;
1672 }
1673
1674 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1675         .fault = kvm_vcpu_fault,
1676 };
1677
1678 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1679 {
1680         vma->vm_ops = &kvm_vcpu_vm_ops;
1681         return 0;
1682 }
1683
1684 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1685 {
1686         struct kvm_vcpu *vcpu = filp->private_data;
1687
1688         kvm_put_kvm(vcpu->kvm);
1689         return 0;
1690 }
1691
1692 static struct file_operations kvm_vcpu_fops = {
1693         .release        = kvm_vcpu_release,
1694         .unlocked_ioctl = kvm_vcpu_ioctl,
1695 #ifdef CONFIG_COMPAT
1696         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1697 #endif
1698         .mmap           = kvm_vcpu_mmap,
1699         .llseek         = noop_llseek,
1700 };
1701
1702 /*
1703  * Allocates an inode for the vcpu.
1704  */
1705 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1706 {
1707         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1708 }
1709
1710 /*
1711  * Creates some virtual cpus.  Good luck creating more than one.
1712  */
1713 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1714 {
1715         int r;
1716         struct kvm_vcpu *vcpu, *v;
1717
1718         vcpu = kvm_arch_vcpu_create(kvm, id);
1719         if (IS_ERR(vcpu))
1720                 return PTR_ERR(vcpu);
1721
1722         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1723
1724         r = kvm_arch_vcpu_setup(vcpu);
1725         if (r)
1726                 goto vcpu_destroy;
1727
1728         mutex_lock(&kvm->lock);
1729         if (!kvm_vcpu_compatible(vcpu)) {
1730                 r = -EINVAL;
1731                 goto unlock_vcpu_destroy;
1732         }
1733         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1734                 r = -EINVAL;
1735                 goto unlock_vcpu_destroy;
1736         }
1737
1738         kvm_for_each_vcpu(r, v, kvm)
1739                 if (v->vcpu_id == id) {
1740                         r = -EEXIST;
1741                         goto unlock_vcpu_destroy;
1742                 }
1743
1744         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1745
1746         /* Now it's all set up, let userspace reach it */
1747         kvm_get_kvm(kvm);
1748         r = create_vcpu_fd(vcpu);
1749         if (r < 0) {
1750                 kvm_put_kvm(kvm);
1751                 goto unlock_vcpu_destroy;
1752         }
1753
1754         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1755         smp_wmb();
1756         atomic_inc(&kvm->online_vcpus);
1757
1758         mutex_unlock(&kvm->lock);
1759         return r;
1760
1761 unlock_vcpu_destroy:
1762         mutex_unlock(&kvm->lock);
1763 vcpu_destroy:
1764         kvm_arch_vcpu_destroy(vcpu);
1765         return r;
1766 }
1767
1768 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1769 {
1770         if (sigset) {
1771                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1772                 vcpu->sigset_active = 1;
1773                 vcpu->sigset = *sigset;
1774         } else
1775                 vcpu->sigset_active = 0;
1776         return 0;
1777 }
1778
1779 static long kvm_vcpu_ioctl(struct file *filp,
1780                            unsigned int ioctl, unsigned long arg)
1781 {
1782         struct kvm_vcpu *vcpu = filp->private_data;
1783         void __user *argp = (void __user *)arg;
1784         int r;
1785         struct kvm_fpu *fpu = NULL;
1786         struct kvm_sregs *kvm_sregs = NULL;
1787
1788         if (vcpu->kvm->mm != current->mm)
1789                 return -EIO;
1790
1791 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1792         /*
1793          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1794          * so vcpu_load() would break it.
1795          */
1796         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1797                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1798 #endif
1799
1800
1801         vcpu_load(vcpu);
1802         switch (ioctl) {
1803         case KVM_RUN:
1804                 r = -EINVAL;
1805                 if (arg)
1806                         goto out;
1807                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1808                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1809                 break;
1810         case KVM_GET_REGS: {
1811                 struct kvm_regs *kvm_regs;
1812
1813                 r = -ENOMEM;
1814                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1815                 if (!kvm_regs)
1816                         goto out;
1817                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1818                 if (r)
1819                         goto out_free1;
1820                 r = -EFAULT;
1821                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1822                         goto out_free1;
1823                 r = 0;
1824 out_free1:
1825                 kfree(kvm_regs);
1826                 break;
1827         }
1828         case KVM_SET_REGS: {
1829                 struct kvm_regs *kvm_regs;
1830
1831                 r = -ENOMEM;
1832                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1833                 if (IS_ERR(kvm_regs)) {
1834                         r = PTR_ERR(kvm_regs);
1835                         goto out;
1836                 }
1837                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1838                 if (r)
1839                         goto out_free2;
1840                 r = 0;
1841 out_free2:
1842                 kfree(kvm_regs);
1843                 break;
1844         }
1845         case KVM_GET_SREGS: {
1846                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1847                 r = -ENOMEM;
1848                 if (!kvm_sregs)
1849                         goto out;
1850                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1851                 if (r)
1852                         goto out;
1853                 r = -EFAULT;
1854                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1855                         goto out;
1856                 r = 0;
1857                 break;
1858         }
1859         case KVM_SET_SREGS: {
1860                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1861                 if (IS_ERR(kvm_sregs)) {
1862                         r = PTR_ERR(kvm_sregs);
1863                         goto out;
1864                 }
1865                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1866                 if (r)
1867                         goto out;
1868                 r = 0;
1869                 break;
1870         }
1871         case KVM_GET_MP_STATE: {
1872                 struct kvm_mp_state mp_state;
1873
1874                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1875                 if (r)
1876                         goto out;
1877                 r = -EFAULT;
1878                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1879                         goto out;
1880                 r = 0;
1881                 break;
1882         }
1883         case KVM_SET_MP_STATE: {
1884                 struct kvm_mp_state mp_state;
1885
1886                 r = -EFAULT;
1887                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1888                         goto out;
1889                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1890                 if (r)
1891                         goto out;
1892                 r = 0;
1893                 break;
1894         }
1895         case KVM_TRANSLATE: {
1896                 struct kvm_translation tr;
1897
1898                 r = -EFAULT;
1899                 if (copy_from_user(&tr, argp, sizeof tr))
1900                         goto out;
1901                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1902                 if (r)
1903                         goto out;
1904                 r = -EFAULT;
1905                 if (copy_to_user(argp, &tr, sizeof tr))
1906                         goto out;
1907                 r = 0;
1908                 break;
1909         }
1910         case KVM_SET_GUEST_DEBUG: {
1911                 struct kvm_guest_debug dbg;
1912
1913                 r = -EFAULT;
1914                 if (copy_from_user(&dbg, argp, sizeof dbg))
1915                         goto out;
1916                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1917                 if (r)
1918                         goto out;
1919                 r = 0;
1920                 break;
1921         }
1922         case KVM_SET_SIGNAL_MASK: {
1923                 struct kvm_signal_mask __user *sigmask_arg = argp;
1924                 struct kvm_signal_mask kvm_sigmask;
1925                 sigset_t sigset, *p;
1926
1927                 p = NULL;
1928                 if (argp) {
1929                         r = -EFAULT;
1930                         if (copy_from_user(&kvm_sigmask, argp,
1931                                            sizeof kvm_sigmask))
1932                                 goto out;
1933                         r = -EINVAL;
1934                         if (kvm_sigmask.len != sizeof sigset)
1935                                 goto out;
1936                         r = -EFAULT;
1937                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1938                                            sizeof sigset))
1939                                 goto out;
1940                         p = &sigset;
1941                 }
1942                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1943                 break;
1944         }
1945         case KVM_GET_FPU: {
1946                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1947                 r = -ENOMEM;
1948                 if (!fpu)
1949                         goto out;
1950                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1951                 if (r)
1952                         goto out;
1953                 r = -EFAULT;
1954                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1955                         goto out;
1956                 r = 0;
1957                 break;
1958         }
1959         case KVM_SET_FPU: {
1960                 fpu = memdup_user(argp, sizeof(*fpu));
1961                 if (IS_ERR(fpu)) {
1962                         r = PTR_ERR(fpu);
1963                         goto out;
1964                 }
1965                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1966                 if (r)
1967                         goto out;
1968                 r = 0;
1969                 break;
1970         }
1971         default:
1972                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1973         }
1974 out:
1975         vcpu_put(vcpu);
1976         kfree(fpu);
1977         kfree(kvm_sregs);
1978         return r;
1979 }
1980
1981 #ifdef CONFIG_COMPAT
1982 static long kvm_vcpu_compat_ioctl(struct file *filp,
1983                                   unsigned int ioctl, unsigned long arg)
1984 {
1985         struct kvm_vcpu *vcpu = filp->private_data;
1986         void __user *argp = compat_ptr(arg);
1987         int r;
1988
1989         if (vcpu->kvm->mm != current->mm)
1990                 return -EIO;
1991
1992         switch (ioctl) {
1993         case KVM_SET_SIGNAL_MASK: {
1994                 struct kvm_signal_mask __user *sigmask_arg = argp;
1995                 struct kvm_signal_mask kvm_sigmask;
1996                 compat_sigset_t csigset;
1997                 sigset_t sigset;
1998
1999                 if (argp) {
2000                         r = -EFAULT;
2001                         if (copy_from_user(&kvm_sigmask, argp,
2002                                            sizeof kvm_sigmask))
2003                                 goto out;
2004                         r = -EINVAL;
2005                         if (kvm_sigmask.len != sizeof csigset)
2006                                 goto out;
2007                         r = -EFAULT;
2008                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2009                                            sizeof csigset))
2010                                 goto out;
2011                 }
2012                 sigset_from_compat(&sigset, &csigset);
2013                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2014                 break;
2015         }
2016         default:
2017                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2018         }
2019
2020 out:
2021         return r;
2022 }
2023 #endif
2024
2025 static long kvm_vm_ioctl(struct file *filp,
2026                            unsigned int ioctl, unsigned long arg)
2027 {
2028         struct kvm *kvm = filp->private_data;
2029         void __user *argp = (void __user *)arg;
2030         int r;
2031
2032         if (kvm->mm != current->mm)
2033                 return -EIO;
2034         switch (ioctl) {
2035         case KVM_CREATE_VCPU:
2036                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2037                 if (r < 0)
2038                         goto out;
2039                 break;
2040         case KVM_SET_USER_MEMORY_REGION: {
2041                 struct kvm_userspace_memory_region kvm_userspace_mem;
2042
2043                 r = -EFAULT;
2044                 if (copy_from_user(&kvm_userspace_mem, argp,
2045                                                 sizeof kvm_userspace_mem))
2046                         goto out;
2047
2048                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2049                 if (r)
2050                         goto out;
2051                 break;
2052         }
2053         case KVM_GET_DIRTY_LOG: {
2054                 struct kvm_dirty_log log;
2055
2056                 r = -EFAULT;
2057                 if (copy_from_user(&log, argp, sizeof log))
2058                         goto out;
2059                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2060                 if (r)
2061                         goto out;
2062                 break;
2063         }
2064 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2065         case KVM_REGISTER_COALESCED_MMIO: {
2066                 struct kvm_coalesced_mmio_zone zone;
2067                 r = -EFAULT;
2068                 if (copy_from_user(&zone, argp, sizeof zone))
2069                         goto out;
2070                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2071                 if (r)
2072                         goto out;
2073                 r = 0;
2074                 break;
2075         }
2076         case KVM_UNREGISTER_COALESCED_MMIO: {
2077                 struct kvm_coalesced_mmio_zone zone;
2078                 r = -EFAULT;
2079                 if (copy_from_user(&zone, argp, sizeof zone))
2080                         goto out;
2081                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2082                 if (r)
2083                         goto out;
2084                 r = 0;
2085                 break;
2086         }
2087 #endif
2088         case KVM_IRQFD: {
2089                 struct kvm_irqfd data;
2090
2091                 r = -EFAULT;
2092                 if (copy_from_user(&data, argp, sizeof data))
2093                         goto out;
2094                 r = kvm_irqfd(kvm, &data);
2095                 break;
2096         }
2097         case KVM_IOEVENTFD: {
2098                 struct kvm_ioeventfd data;
2099
2100                 r = -EFAULT;
2101                 if (copy_from_user(&data, argp, sizeof data))
2102                         goto out;
2103                 r = kvm_ioeventfd(kvm, &data);
2104                 break;
2105         }
2106 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2107         case KVM_SET_BOOT_CPU_ID:
2108                 r = 0;
2109                 mutex_lock(&kvm->lock);
2110                 if (atomic_read(&kvm->online_vcpus) != 0)
2111                         r = -EBUSY;
2112                 else
2113                         kvm->bsp_vcpu_id = arg;
2114                 mutex_unlock(&kvm->lock);
2115                 break;
2116 #endif
2117 #ifdef CONFIG_HAVE_KVM_MSI
2118         case KVM_SIGNAL_MSI: {
2119                 struct kvm_msi msi;
2120
2121                 r = -EFAULT;
2122                 if (copy_from_user(&msi, argp, sizeof msi))
2123                         goto out;
2124                 r = kvm_send_userspace_msi(kvm, &msi);
2125                 break;
2126         }
2127 #endif
2128 #ifdef __KVM_HAVE_IRQ_LINE
2129         case KVM_IRQ_LINE_STATUS:
2130         case KVM_IRQ_LINE: {
2131                 struct kvm_irq_level irq_event;
2132
2133                 r = -EFAULT;
2134                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2135                         goto out;
2136
2137                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2138                 if (r)
2139                         goto out;
2140
2141                 r = -EFAULT;
2142                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2143                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2144                                 goto out;
2145                 }
2146
2147                 r = 0;
2148                 break;
2149         }
2150 #endif
2151         default:
2152                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2153                 if (r == -ENOTTY)
2154                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2155         }
2156 out:
2157         return r;
2158 }
2159
2160 #ifdef CONFIG_COMPAT
2161 struct compat_kvm_dirty_log {
2162         __u32 slot;
2163         __u32 padding1;
2164         union {
2165                 compat_uptr_t dirty_bitmap; /* one bit per page */
2166                 __u64 padding2;
2167         };
2168 };
2169
2170 static long kvm_vm_compat_ioctl(struct file *filp,
2171                            unsigned int ioctl, unsigned long arg)
2172 {
2173         struct kvm *kvm = filp->private_data;
2174         int r;
2175
2176         if (kvm->mm != current->mm)
2177                 return -EIO;
2178         switch (ioctl) {
2179         case KVM_GET_DIRTY_LOG: {
2180                 struct compat_kvm_dirty_log compat_log;
2181                 struct kvm_dirty_log log;
2182
2183                 r = -EFAULT;
2184                 if (copy_from_user(&compat_log, (void __user *)arg,
2185                                    sizeof(compat_log)))
2186                         goto out;
2187                 log.slot         = compat_log.slot;
2188                 log.padding1     = compat_log.padding1;
2189                 log.padding2     = compat_log.padding2;
2190                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2191
2192                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2193                 if (r)
2194                         goto out;
2195                 break;
2196         }
2197         default:
2198                 r = kvm_vm_ioctl(filp, ioctl, arg);
2199         }
2200
2201 out:
2202         return r;
2203 }
2204 #endif
2205
2206 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2207 {
2208         struct page *page[1];
2209         unsigned long addr;
2210         int npages;
2211         gfn_t gfn = vmf->pgoff;
2212         struct kvm *kvm = vma->vm_file->private_data;
2213
2214         addr = gfn_to_hva(kvm, gfn);
2215         if (kvm_is_error_hva(addr))
2216                 return VM_FAULT_SIGBUS;
2217
2218         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2219                                 NULL);
2220         if (unlikely(npages != 1))
2221                 return VM_FAULT_SIGBUS;
2222
2223         vmf->page = page[0];
2224         return 0;
2225 }
2226
2227 static const struct vm_operations_struct kvm_vm_vm_ops = {
2228         .fault = kvm_vm_fault,
2229 };
2230
2231 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2232 {
2233         vma->vm_ops = &kvm_vm_vm_ops;
2234         return 0;
2235 }
2236
2237 static struct file_operations kvm_vm_fops = {
2238         .release        = kvm_vm_release,
2239         .unlocked_ioctl = kvm_vm_ioctl,
2240 #ifdef CONFIG_COMPAT
2241         .compat_ioctl   = kvm_vm_compat_ioctl,
2242 #endif
2243         .mmap           = kvm_vm_mmap,
2244         .llseek         = noop_llseek,
2245 };
2246
2247 static int kvm_dev_ioctl_create_vm(unsigned long type)
2248 {
2249         int r;
2250         struct kvm *kvm;
2251
2252         kvm = kvm_create_vm(type);
2253         if (IS_ERR(kvm))
2254                 return PTR_ERR(kvm);
2255 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2256         r = kvm_coalesced_mmio_init(kvm);
2257         if (r < 0) {
2258                 kvm_put_kvm(kvm);
2259                 return r;
2260         }
2261 #endif
2262         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2263         if (r < 0)
2264                 kvm_put_kvm(kvm);
2265
2266         return r;
2267 }
2268
2269 static long kvm_dev_ioctl_check_extension_generic(long arg)
2270 {
2271         switch (arg) {
2272         case KVM_CAP_USER_MEMORY:
2273         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2274         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2275 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2276         case KVM_CAP_SET_BOOT_CPU_ID:
2277 #endif
2278         case KVM_CAP_INTERNAL_ERROR_DATA:
2279 #ifdef CONFIG_HAVE_KVM_MSI
2280         case KVM_CAP_SIGNAL_MSI:
2281 #endif
2282                 return 1;
2283 #ifdef KVM_CAP_IRQ_ROUTING
2284         case KVM_CAP_IRQ_ROUTING:
2285                 return KVM_MAX_IRQ_ROUTES;
2286 #endif
2287         default:
2288                 break;
2289         }
2290         return kvm_dev_ioctl_check_extension(arg);
2291 }
2292
2293 static long kvm_dev_ioctl(struct file *filp,
2294                           unsigned int ioctl, unsigned long arg)
2295 {
2296         long r = -EINVAL;
2297
2298         switch (ioctl) {
2299         case KVM_GET_API_VERSION:
2300                 r = -EINVAL;
2301                 if (arg)
2302                         goto out;
2303                 r = KVM_API_VERSION;
2304                 break;
2305         case KVM_CREATE_VM:
2306                 r = kvm_dev_ioctl_create_vm(arg);
2307                 break;
2308         case KVM_CHECK_EXTENSION:
2309                 r = kvm_dev_ioctl_check_extension_generic(arg);
2310                 break;
2311         case KVM_GET_VCPU_MMAP_SIZE:
2312                 r = -EINVAL;
2313                 if (arg)
2314                         goto out;
2315                 r = PAGE_SIZE;     /* struct kvm_run */
2316 #ifdef CONFIG_X86
2317                 r += PAGE_SIZE;    /* pio data page */
2318 #endif
2319 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2320                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2321 #endif
2322                 break;
2323         case KVM_TRACE_ENABLE:
2324         case KVM_TRACE_PAUSE:
2325         case KVM_TRACE_DISABLE:
2326                 r = -EOPNOTSUPP;
2327                 break;
2328         default:
2329                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2330         }
2331 out:
2332         return r;
2333 }
2334
2335 static struct file_operations kvm_chardev_ops = {
2336         .unlocked_ioctl = kvm_dev_ioctl,
2337         .compat_ioctl   = kvm_dev_ioctl,
2338         .llseek         = noop_llseek,
2339 };
2340
2341 static struct miscdevice kvm_dev = {
2342         KVM_MINOR,
2343         "kvm",
2344         &kvm_chardev_ops,
2345 };
2346
2347 static void hardware_enable_nolock(void *junk)
2348 {
2349         int cpu = raw_smp_processor_id();
2350         int r;
2351
2352         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2353                 return;
2354
2355         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2356
2357         r = kvm_arch_hardware_enable(NULL);
2358
2359         if (r) {
2360                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2361                 atomic_inc(&hardware_enable_failed);
2362                 printk(KERN_INFO "kvm: enabling virtualization on "
2363                                  "CPU%d failed\n", cpu);
2364         }
2365 }
2366
2367 static void hardware_enable(void *junk)
2368 {
2369         raw_spin_lock(&kvm_lock);
2370         hardware_enable_nolock(junk);
2371         raw_spin_unlock(&kvm_lock);
2372 }
2373
2374 static void hardware_disable_nolock(void *junk)
2375 {
2376         int cpu = raw_smp_processor_id();
2377
2378         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2379                 return;
2380         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2381         kvm_arch_hardware_disable(NULL);
2382 }
2383
2384 static void hardware_disable(void *junk)
2385 {
2386         raw_spin_lock(&kvm_lock);
2387         hardware_disable_nolock(junk);
2388         raw_spin_unlock(&kvm_lock);
2389 }
2390
2391 static void hardware_disable_all_nolock(void)
2392 {
2393         BUG_ON(!kvm_usage_count);
2394
2395         kvm_usage_count--;
2396         if (!kvm_usage_count)
2397                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2398 }
2399
2400 static void hardware_disable_all(void)
2401 {
2402         raw_spin_lock(&kvm_lock);
2403         hardware_disable_all_nolock();
2404         raw_spin_unlock(&kvm_lock);
2405 }
2406
2407 static int hardware_enable_all(void)
2408 {
2409         int r = 0;
2410
2411         raw_spin_lock(&kvm_lock);
2412
2413         kvm_usage_count++;
2414         if (kvm_usage_count == 1) {
2415                 atomic_set(&hardware_enable_failed, 0);
2416                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2417
2418                 if (atomic_read(&hardware_enable_failed)) {
2419                         hardware_disable_all_nolock();
2420                         r = -EBUSY;
2421                 }
2422         }
2423
2424         raw_spin_unlock(&kvm_lock);
2425
2426         return r;
2427 }
2428
2429 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2430                            void *v)
2431 {
2432         int cpu = (long)v;
2433
2434         if (!kvm_usage_count)
2435                 return NOTIFY_OK;
2436
2437         val &= ~CPU_TASKS_FROZEN;
2438         switch (val) {
2439         case CPU_DYING:
2440                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2441                        cpu);
2442                 hardware_disable(NULL);
2443                 break;
2444         case CPU_STARTING:
2445                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2446                        cpu);
2447                 hardware_enable(NULL);
2448                 break;
2449         }
2450         return NOTIFY_OK;
2451 }
2452
2453
2454 asmlinkage void kvm_spurious_fault(void)
2455 {
2456         /* Fault while not rebooting.  We want the trace. */
2457         BUG();
2458 }
2459 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2460
2461 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2462                       void *v)
2463 {
2464         /*
2465          * Some (well, at least mine) BIOSes hang on reboot if
2466          * in vmx root mode.
2467          *
2468          * And Intel TXT required VMX off for all cpu when system shutdown.
2469          */
2470         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2471         kvm_rebooting = true;
2472         on_each_cpu(hardware_disable_nolock, NULL, 1);
2473         return NOTIFY_OK;
2474 }
2475
2476 static struct notifier_block kvm_reboot_notifier = {
2477         .notifier_call = kvm_reboot,
2478         .priority = 0,
2479 };
2480
2481 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2482 {
2483         int i;
2484
2485         for (i = 0; i < bus->dev_count; i++) {
2486                 struct kvm_io_device *pos = bus->range[i].dev;
2487
2488                 kvm_iodevice_destructor(pos);
2489         }
2490         kfree(bus);
2491 }
2492
2493 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2494 {
2495         const struct kvm_io_range *r1 = p1;
2496         const struct kvm_io_range *r2 = p2;
2497
2498         if (r1->addr < r2->addr)
2499                 return -1;
2500         if (r1->addr + r1->len > r2->addr + r2->len)
2501                 return 1;
2502         return 0;
2503 }
2504
2505 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2506                           gpa_t addr, int len)
2507 {
2508         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2509                 .addr = addr,
2510                 .len = len,
2511                 .dev = dev,
2512         };
2513
2514         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2515                 kvm_io_bus_sort_cmp, NULL);
2516
2517         return 0;
2518 }
2519
2520 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2521                              gpa_t addr, int len)
2522 {
2523         struct kvm_io_range *range, key;
2524         int off;
2525
2526         key = (struct kvm_io_range) {
2527                 .addr = addr,
2528                 .len = len,
2529         };
2530
2531         range = bsearch(&key, bus->range, bus->dev_count,
2532                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2533         if (range == NULL)
2534                 return -ENOENT;
2535
2536         off = range - bus->range;
2537
2538         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2539                 off--;
2540
2541         return off;
2542 }
2543
2544 /* kvm_io_bus_write - called under kvm->slots_lock */
2545 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2546                      int len, const void *val)
2547 {
2548         int idx;
2549         struct kvm_io_bus *bus;
2550         struct kvm_io_range range;
2551
2552         range = (struct kvm_io_range) {
2553                 .addr = addr,
2554                 .len = len,
2555         };
2556
2557         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2558         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2559         if (idx < 0)
2560                 return -EOPNOTSUPP;
2561
2562         while (idx < bus->dev_count &&
2563                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2564                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2565                         return 0;
2566                 idx++;
2567         }
2568
2569         return -EOPNOTSUPP;
2570 }
2571
2572 /* kvm_io_bus_read - called under kvm->slots_lock */
2573 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2574                     int len, void *val)
2575 {
2576         int idx;
2577         struct kvm_io_bus *bus;
2578         struct kvm_io_range range;
2579
2580         range = (struct kvm_io_range) {
2581                 .addr = addr,
2582                 .len = len,
2583         };
2584
2585         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2586         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2587         if (idx < 0)
2588                 return -EOPNOTSUPP;
2589
2590         while (idx < bus->dev_count &&
2591                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2592                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2593                         return 0;
2594                 idx++;
2595         }
2596
2597         return -EOPNOTSUPP;
2598 }
2599
2600 /* Caller must hold slots_lock. */
2601 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2602                             int len, struct kvm_io_device *dev)
2603 {
2604         struct kvm_io_bus *new_bus, *bus;
2605
2606         bus = kvm->buses[bus_idx];
2607         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2608                 return -ENOSPC;
2609
2610         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2611                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2612         if (!new_bus)
2613                 return -ENOMEM;
2614         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2615                sizeof(struct kvm_io_range)));
2616         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2617         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2618         synchronize_srcu_expedited(&kvm->srcu);
2619         kfree(bus);
2620
2621         return 0;
2622 }
2623
2624 /* Caller must hold slots_lock. */
2625 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2626                               struct kvm_io_device *dev)
2627 {
2628         int i, r;
2629         struct kvm_io_bus *new_bus, *bus;
2630
2631         bus = kvm->buses[bus_idx];
2632         r = -ENOENT;
2633         for (i = 0; i < bus->dev_count; i++)
2634                 if (bus->range[i].dev == dev) {
2635                         r = 0;
2636                         break;
2637                 }
2638
2639         if (r)
2640                 return r;
2641
2642         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2643                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2644         if (!new_bus)
2645                 return -ENOMEM;
2646
2647         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2648         new_bus->dev_count--;
2649         memcpy(new_bus->range + i, bus->range + i + 1,
2650                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2651
2652         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2653         synchronize_srcu_expedited(&kvm->srcu);
2654         kfree(bus);
2655         return r;
2656 }
2657
2658 static struct notifier_block kvm_cpu_notifier = {
2659         .notifier_call = kvm_cpu_hotplug,
2660 };
2661
2662 static int vm_stat_get(void *_offset, u64 *val)
2663 {
2664         unsigned offset = (long)_offset;
2665         struct kvm *kvm;
2666
2667         *val = 0;
2668         raw_spin_lock(&kvm_lock);
2669         list_for_each_entry(kvm, &vm_list, vm_list)
2670                 *val += *(u32 *)((void *)kvm + offset);
2671         raw_spin_unlock(&kvm_lock);
2672         return 0;
2673 }
2674
2675 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2676
2677 static int vcpu_stat_get(void *_offset, u64 *val)
2678 {
2679         unsigned offset = (long)_offset;
2680         struct kvm *kvm;
2681         struct kvm_vcpu *vcpu;
2682         int i;
2683
2684         *val = 0;
2685         raw_spin_lock(&kvm_lock);
2686         list_for_each_entry(kvm, &vm_list, vm_list)
2687                 kvm_for_each_vcpu(i, vcpu, kvm)
2688                         *val += *(u32 *)((void *)vcpu + offset);
2689
2690         raw_spin_unlock(&kvm_lock);
2691         return 0;
2692 }
2693
2694 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2695
2696 static const struct file_operations *stat_fops[] = {
2697         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2698         [KVM_STAT_VM]   = &vm_stat_fops,
2699 };
2700
2701 static int kvm_init_debug(void)
2702 {
2703         int r = -EFAULT;
2704         struct kvm_stats_debugfs_item *p;
2705
2706         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2707         if (kvm_debugfs_dir == NULL)
2708                 goto out;
2709
2710         for (p = debugfs_entries; p->name; ++p) {
2711                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2712                                                 (void *)(long)p->offset,
2713                                                 stat_fops[p->kind]);
2714                 if (p->dentry == NULL)
2715                         goto out_dir;
2716         }
2717
2718         return 0;
2719
2720 out_dir:
2721         debugfs_remove_recursive(kvm_debugfs_dir);
2722 out:
2723         return r;
2724 }
2725
2726 static void kvm_exit_debug(void)
2727 {
2728         struct kvm_stats_debugfs_item *p;
2729
2730         for (p = debugfs_entries; p->name; ++p)
2731                 debugfs_remove(p->dentry);
2732         debugfs_remove(kvm_debugfs_dir);
2733 }
2734
2735 static int kvm_suspend(void)
2736 {
2737         if (kvm_usage_count)
2738                 hardware_disable_nolock(NULL);
2739         return 0;
2740 }
2741
2742 static void kvm_resume(void)
2743 {
2744         if (kvm_usage_count) {
2745                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2746                 hardware_enable_nolock(NULL);
2747         }
2748 }
2749
2750 static struct syscore_ops kvm_syscore_ops = {
2751         .suspend = kvm_suspend,
2752         .resume = kvm_resume,
2753 };
2754
2755 static inline
2756 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2757 {
2758         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2759 }
2760
2761 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2762 {
2763         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2764
2765         kvm_arch_vcpu_load(vcpu, cpu);
2766 }
2767
2768 static void kvm_sched_out(struct preempt_notifier *pn,
2769                           struct task_struct *next)
2770 {
2771         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2772
2773         kvm_arch_vcpu_put(vcpu);
2774 }
2775
2776 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2777                   struct module *module)
2778 {
2779         int r;
2780         int cpu;
2781
2782         r = kvm_arch_init(opaque);
2783         if (r)
2784                 goto out_fail;
2785
2786         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2787                 r = -ENOMEM;
2788                 goto out_free_0;
2789         }
2790
2791         r = kvm_arch_hardware_setup();
2792         if (r < 0)
2793                 goto out_free_0a;
2794
2795         for_each_online_cpu(cpu) {
2796                 smp_call_function_single(cpu,
2797                                 kvm_arch_check_processor_compat,
2798                                 &r, 1);
2799                 if (r < 0)
2800                         goto out_free_1;
2801         }
2802
2803         r = register_cpu_notifier(&kvm_cpu_notifier);
2804         if (r)
2805                 goto out_free_2;
2806         register_reboot_notifier(&kvm_reboot_notifier);
2807
2808         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2809         if (!vcpu_align)
2810                 vcpu_align = __alignof__(struct kvm_vcpu);
2811         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2812                                            0, NULL);
2813         if (!kvm_vcpu_cache) {
2814                 r = -ENOMEM;
2815                 goto out_free_3;
2816         }
2817
2818         r = kvm_async_pf_init();
2819         if (r)
2820                 goto out_free;
2821
2822         kvm_chardev_ops.owner = module;
2823         kvm_vm_fops.owner = module;
2824         kvm_vcpu_fops.owner = module;
2825
2826         r = misc_register(&kvm_dev);
2827         if (r) {
2828                 printk(KERN_ERR "kvm: misc device register failed\n");
2829                 goto out_unreg;
2830         }
2831
2832         register_syscore_ops(&kvm_syscore_ops);
2833
2834         kvm_preempt_ops.sched_in = kvm_sched_in;
2835         kvm_preempt_ops.sched_out = kvm_sched_out;
2836
2837         r = kvm_init_debug();
2838         if (r) {
2839                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2840                 goto out_undebugfs;
2841         }
2842
2843         return 0;
2844
2845 out_undebugfs:
2846         unregister_syscore_ops(&kvm_syscore_ops);
2847 out_unreg:
2848         kvm_async_pf_deinit();
2849 out_free:
2850         kmem_cache_destroy(kvm_vcpu_cache);
2851 out_free_3:
2852         unregister_reboot_notifier(&kvm_reboot_notifier);
2853         unregister_cpu_notifier(&kvm_cpu_notifier);
2854 out_free_2:
2855 out_free_1:
2856         kvm_arch_hardware_unsetup();
2857 out_free_0a:
2858         free_cpumask_var(cpus_hardware_enabled);
2859 out_free_0:
2860         kvm_arch_exit();
2861 out_fail:
2862         return r;
2863 }
2864 EXPORT_SYMBOL_GPL(kvm_init);
2865
2866 void kvm_exit(void)
2867 {
2868         kvm_exit_debug();
2869         misc_deregister(&kvm_dev);
2870         kmem_cache_destroy(kvm_vcpu_cache);
2871         kvm_async_pf_deinit();
2872         unregister_syscore_ops(&kvm_syscore_ops);
2873         unregister_reboot_notifier(&kvm_reboot_notifier);
2874         unregister_cpu_notifier(&kvm_cpu_notifier);
2875         on_each_cpu(hardware_disable_nolock, NULL, 1);
2876         kvm_arch_hardware_unsetup();
2877         kvm_arch_exit();
2878         free_cpumask_var(cpus_hardware_enabled);
2879 }
2880 EXPORT_SYMBOL_GPL(kvm_exit);