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         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 684
 685 #ifdef KVM_CAP_READONLY_MEM
 686         valid_flags |= KVM_MEM_READONLY;
 687 #endif
 688
 689         if (mem->flags & ~valid_flags)
 690                 return -EINVAL;
 691
 692         return 0;
 693 }
 694
 695 /*
 696  * Allocate some memory and give it an address in the guest physical address
 697  * space.
 698  *
 699  * Discontiguous memory is allowed, mostly for framebuffers.
 700  *
 701  * Must be called holding mmap_sem for write.
 702  */
 703 int __kvm_set_memory_region(struct kvm *kvm,
 704                             struct kvm_userspace_memory_region *mem,
 705                             int user_alloc)
 706 {
 707         int r;
 708         gfn_t base_gfn;
 709         unsigned long npages;
 710         unsigned long i;
 711         struct kvm_memory_slot *memslot;
 712         struct kvm_memory_slot old, new;
 713         struct kvm_memslots *slots, *old_memslots;
 714
 715         r = check_memory_region_flags(mem);
 716         if (r)
 717                 goto out;
 718
 719         r = -EINVAL;
 720         /* General sanity checks */
 721         if (mem->memory_size & (PAGE_SIZE - 1))
 722                 goto out;
 723         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 724                 goto out;
 725         /* We can read the guest memory with __xxx_user() later on. */
 726         if (user_alloc &&
 727             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 728              !access_ok(VERIFY_WRITE,
 729                         (void __user *)(unsigned long)mem->userspace_addr,
 730                         mem->memory_size)))
 731                 goto out;
 732         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 733                 goto out;
 734         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 735                 goto out;
 736
 737         memslot = id_to_memslot(kvm->memslots, mem->slot);
 738         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 739         npages = mem->memory_size >> PAGE_SHIFT;
 740
 741         r = -EINVAL;
 742         if (npages > KVM_MEM_MAX_NR_PAGES)
 743                 goto out;
 744
 745         if (!npages)
 746                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 747
 748         new = old = *memslot;
 749
 750         new.id = mem->slot;
 751         new.base_gfn = base_gfn;
 752         new.npages = npages;
 753         new.flags = mem->flags;
 754
 755         /* Disallow changing a memory slot's size. */
 756         r = -EINVAL;
 757         if (npages && old.npages && npages != old.npages)
 758                 goto out_free;
 759
 760         /* Check for overlaps */
 761         r = -EEXIST;
 762         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 763                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 764
 765                 if (s == memslot || !s->npages)
 766                         continue;
 767                 if (!((base_gfn + npages <= s->base_gfn) ||
 768                       (base_gfn >= s->base_gfn + s->npages)))
 769                         goto out_free;
 770         }
 771
 772         /* Free page dirty bitmap if unneeded */
 773         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 774                 new.dirty_bitmap = NULL;
 775
 776         r = -ENOMEM;
 777
 778         /* Allocate if a slot is being created */
 779         if (npages && !old.npages) {
 780                 new.user_alloc = user_alloc;
 781                 new.userspace_addr = mem->userspace_addr;
 782
 783                 if (kvm_arch_create_memslot(&new, npages))
 784                         goto out_free;
 785         }
 786
 787         /* Allocate page dirty bitmap if needed */
 788         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 789                 if (kvm_create_dirty_bitmap(&new) < 0)
 790                         goto out_free;
 791                 /* destroy any largepage mappings for dirty tracking */
 792         }
 793
 794         if (!npages) {
 795                 struct kvm_memory_slot *slot;
 796
 797                 r = -ENOMEM;
 798                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 799                                 GFP_KERNEL);
 800                 if (!slots)
 801                         goto out_free;
 802                 slot = id_to_memslot(slots, mem->slot);
 803                 slot->flags |= KVM_MEMSLOT_INVALID;
 804
 805                 update_memslots(slots, NULL);
 806
 807                 old_memslots = kvm->memslots;
 808                 rcu_assign_pointer(kvm->memslots, slots);
 809                 synchronize_srcu_expedited(&kvm->srcu);
 810                 /* From this point no new shadow pages pointing to a deleted
 811                  * memslot will be created.
 812                  *
 813                  * validation of sp->gfn happens in:
 814                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 815                  *      - kvm_is_visible_gfn (mmu_check_roots)
 816                  */
 817                 kvm_arch_flush_shadow(kvm);
 818                 kfree(old_memslots);
 819         }
 820
 821         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 822         if (r)
 823                 goto out_free;
 824
 825         /* map/unmap the pages in iommu page table */
 826         if (npages) {
 827                 r = kvm_iommu_map_pages(kvm, &new);
 828                 if (r)
 829                         goto out_free;
 830         } else
 831                 kvm_iommu_unmap_pages(kvm, &old);
 832
 833         r = -ENOMEM;
 834         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 835                         GFP_KERNEL);
 836         if (!slots)
 837                 goto out_free;
 838
 839         /* actual memory is freed via old in kvm_free_physmem_slot below */
 840         if (!npages) {
 841                 new.dirty_bitmap = NULL;
 842                 memset(&new.arch, 0, sizeof(new.arch));
 843         }
 844
 845         update_memslots(slots, &new);
 846         old_memslots = kvm->memslots;
 847         rcu_assign_pointer(kvm->memslots, slots);
 848         synchronize_srcu_expedited(&kvm->srcu);
 849
 850         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 851
 852         /*
 853          * If the new memory slot is created, we need to clear all
 854          * mmio sptes.
 855          */
 856         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 857                 kvm_arch_flush_shadow(kvm);
 858
 859         kvm_free_physmem_slot(&old, &new);
 860         kfree(old_memslots);
 861
 862         return 0;
 863
 864 out_free:
 865         kvm_free_physmem_slot(&new, &old);
 866 out:
 867         return r;
 868
 869 }
 870 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 871
 872 int kvm_set_memory_region(struct kvm *kvm,
 873                           struct kvm_userspace_memory_region *mem,
 874                           int user_alloc)
 875 {
 876         int r;
 877
 878         mutex_lock(&kvm->slots_lock);
 879         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 880         mutex_unlock(&kvm->slots_lock);
 881         return r;
 882 }
 883 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 884
 885 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 886                                    struct
 887                                    kvm_userspace_memory_region *mem,
 888                                    int user_alloc)
 889 {
 890         if (mem->slot >= KVM_MEMORY_SLOTS)
 891                 return -EINVAL;
 892         return kvm_set_memory_region(kvm, mem, user_alloc);
 893 }
 894
 895 int kvm_get_dirty_log(struct kvm *kvm,
 896                         struct kvm_dirty_log *log, int *is_dirty)
 897 {
 898         struct kvm_memory_slot *memslot;
 899         int r, i;
 900         unsigned long n;
 901         unsigned long any = 0;
 902
 903         r = -EINVAL;
 904         if (log->slot >= KVM_MEMORY_SLOTS)
 905                 goto out;
 906
 907         memslot = id_to_memslot(kvm->memslots, log->slot);
 908         r = -ENOENT;
 909         if (!memslot->dirty_bitmap)
 910                 goto out;
 911
 912         n = kvm_dirty_bitmap_bytes(memslot);
 913
 914         for (i = 0; !any && i < n/sizeof(long); ++i)
 915                 any = memslot->dirty_bitmap[i];
 916
 917         r = -EFAULT;
 918         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 919                 goto out;
 920
 921         if (any)
 922                 *is_dirty = 1;
 923
 924         r = 0;
 925 out:
 926         return r;
 927 }
 928
 929 bool kvm_largepages_enabled(void)
 930 {
 931         return largepages_enabled;
 932 }
 933
 934 void kvm_disable_largepages(void)
 935 {
 936         largepages_enabled = false;
 937 }
 938 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 939
 940 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 941 {
 942         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 943 }
 944 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 945
 946 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 947 {
 948         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 949
 950         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
 951               memslot->flags & KVM_MEMSLOT_INVALID)
 952                 return 0;
 953
 954         return 1;
 955 }
 956 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 957
 958 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 959 {
 960         struct vm_area_struct *vma;
 961         unsigned long addr, size;
 962
 963         size = PAGE_SIZE;
 964
 965         addr = gfn_to_hva(kvm, gfn);
 966         if (kvm_is_error_hva(addr))
 967                 return PAGE_SIZE;
 968
 969         down_read(&current->mm->mmap_sem);
 970         vma = find_vma(current->mm, addr);
 971         if (!vma)
 972                 goto out;
 973
 974         size = vma_kernel_pagesize(vma);
 975
 976 out:
 977         up_read(&current->mm->mmap_sem);
 978
 979         return size;
 980 }
 981
 982 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
 983 {
 984         return slot->flags & KVM_MEM_READONLY;
 985 }
 986
 987 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
 988                                        gfn_t *nr_pages, bool write)
 989 {
 990         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 991                 return KVM_HVA_ERR_BAD;
 992
 993         if (memslot_is_readonly(slot) && write)
 994                 return KVM_HVA_ERR_RO_BAD;
 995
 996         if (nr_pages)
 997                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
 998
 999         return __gfn_to_hva_memslot(slot, gfn);
1000 }
1001
1002 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1003                                      gfn_t *nr_pages)
1004 {
1005         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1006 }
1007
1008 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1009                                  gfn_t gfn)
1010 {
1011         return gfn_to_hva_many(slot, gfn, NULL);
1012 }
1013 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1014
1015 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1016 {
1017         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1018 }
1019 EXPORT_SYMBOL_GPL(gfn_to_hva);
1020
1021 /*
1022  * The hva returned by this function is only allowed to be read.
1023  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1024  */
1025 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1026 {
1027         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1028 }
1029
1030 static int kvm_read_hva(void *data, void __user *hva, int len)
1031 {
1032         return __copy_from_user(data, hva, len);
1033 }
1034
1035 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1036 {
1037         return __copy_from_user_inatomic(data, hva, len);
1038 }
1039
1040 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1041         unsigned long start, int write, struct page **page)
1042 {
1043         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1044
1045         if (write)
1046                 flags |= FOLL_WRITE;
1047
1048         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1049 }
1050
1051 static inline int check_user_page_hwpoison(unsigned long addr)
1052 {
1053         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1054
1055         rc = __get_user_pages(current, current->mm, addr, 1,
1056                               flags, NULL, NULL, NULL);
1057         return rc == -EHWPOISON;
1058 }
1059
1060 /*
1061  * The atomic path to get the writable pfn which will be stored in @pfn,
1062  * true indicates success, otherwise false is returned.
1063  */
1064 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1065                             bool write_fault, bool *writable, pfn_t *pfn)
1066 {
1067         struct page *page[1];
1068         int npages;
1069
1070         if (!(async || atomic))
1071                 return false;
1072
1073         /*
1074          * Fast pin a writable pfn only if it is a write fault request
1075          * or the caller allows to map a writable pfn for a read fault
1076          * request.
1077          */
1078         if (!(write_fault || writable))
1079                 return false;
1080
1081         npages = __get_user_pages_fast(addr, 1, 1, page);
1082         if (npages == 1) {
1083                 *pfn = page_to_pfn(page[0]);
1084
1085                 if (writable)
1086                         *writable = true;
1087                 return true;
1088         }
1089
1090         return false;
1091 }
1092
1093 /*
1094  * The slow path to get the pfn of the specified host virtual address,
1095  * 1 indicates success, -errno is returned if error is detected.
1096  */
1097 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1098                            bool *writable, pfn_t *pfn)
1099 {
1100         struct page *page[1];
1101         int npages = 0;
1102
1103         might_sleep();
1104
1105         if (writable)
1106                 *writable = write_fault;
1107
1108         if (async) {
1109                 down_read(&current->mm->mmap_sem);
1110                 npages = get_user_page_nowait(current, current->mm,
1111                                               addr, write_fault, page);
1112                 up_read(&current->mm->mmap_sem);
1113         } else
1114                 npages = get_user_pages_fast(addr, 1, write_fault,
1115                                              page);
1116         if (npages != 1)
1117                 return npages;
1118
1119         /* map read fault as writable if possible */
1120         if (unlikely(!write_fault) && writable) {
1121                 struct page *wpage[1];
1122
1123                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1124                 if (npages == 1) {
1125                         *writable = true;
1126                         put_page(page[0]);
1127                         page[0] = wpage[0];
1128                 }
1129
1130                 npages = 1;
1131         }
1132         *pfn = page_to_pfn(page[0]);
1133         return npages;
1134 }
1135
1136 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1137 {
1138         if (unlikely(!(vma->vm_flags & VM_READ)))
1139                 return false;
1140
1141         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1142                 return false;
1143
1144         return true;
1145 }
1146
1147 /*
1148  * Pin guest page in memory and return its pfn.
1149  * @addr: host virtual address which maps memory to the guest
1150  * @atomic: whether this function can sleep
1151  * @async: whether this function need to wait IO complete if the
1152  *         host page is not in the memory
1153  * @write_fault: whether we should get a writable host page
1154  * @writable: whether it allows to map a writable host page for !@write_fault
1155  *
1156  * The function will map a writable host page for these two cases:
1157  * 1): @write_fault = true
1158  * 2): @write_fault = false && @writable, @writable will tell the caller
1159  *     whether the mapping is writable.
1160  */
1161 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1162                         bool write_fault, bool *writable)
1163 {
1164         struct vm_area_struct *vma;
1165         pfn_t pfn = 0;
1166         int npages;
1167
1168         /* we can do it either atomically or asynchronously, not both */
1169         BUG_ON(atomic && async);
1170
1171         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1172                 return pfn;
1173
1174         if (atomic)
1175                 return KVM_PFN_ERR_FAULT;
1176
1177         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1178         if (npages == 1)
1179                 return pfn;
1180
1181         down_read(&current->mm->mmap_sem);
1182         if (npages == -EHWPOISON ||
1183               (!async && check_user_page_hwpoison(addr))) {
1184                 pfn = KVM_PFN_ERR_HWPOISON;
1185                 goto exit;
1186         }
1187
1188         vma = find_vma_intersection(current->mm, addr, addr + 1);
1189
1190         if (vma == NULL)
1191                 pfn = KVM_PFN_ERR_FAULT;
1192         else if ((vma->vm_flags & VM_PFNMAP)) {
1193                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1194                         vma->vm_pgoff;
1195                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1196         } else {
1197                 if (async && vma_is_valid(vma, write_fault))
1198                         *async = true;
1199                 pfn = KVM_PFN_ERR_FAULT;
1200         }
1201 exit:
1202         up_read(&current->mm->mmap_sem);
1203         return pfn;
1204 }
1205
1206 static pfn_t
1207 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1208                      bool *async, bool write_fault, bool *writable)
1209 {
1210         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1211
1212         if (addr == KVM_HVA_ERR_RO_BAD)
1213                 return KVM_PFN_ERR_RO_FAULT;
1214
1215         if (kvm_is_error_hva(addr))
1216                 return KVM_PFN_ERR_BAD;
1217
1218         /* Do not map writable pfn in the readonly memslot. */
1219         if (writable && memslot_is_readonly(slot)) {
1220                 *writable = false;
1221                 writable = NULL;
1222         }
1223
1224         return hva_to_pfn(addr, atomic, async, write_fault,
1225                           writable);
1226 }
1227
1228 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1229                           bool write_fault, bool *writable)
1230 {
1231         struct kvm_memory_slot *slot;
1232
1233         if (async)
1234                 *async = false;
1235
1236         slot = gfn_to_memslot(kvm, gfn);
1237
1238         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1239                                     writable);
1240 }
1241
1242 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1243 {
1244         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1245 }
1246 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1247
1248 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1249                        bool write_fault, bool *writable)
1250 {
1251         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1252 }
1253 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1254
1255 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1256 {
1257         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1258 }
1259 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1260
1261 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1262                       bool *writable)
1263 {
1264         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1265 }
1266 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1267
1268 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1269 {
1270         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1271 }
1272
1273 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1274 {
1275         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1276 }
1277 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1278
1279 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1280                                                                   int nr_pages)
1281 {
1282         unsigned long addr;
1283         gfn_t entry;
1284
1285         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1286         if (kvm_is_error_hva(addr))
1287                 return -1;
1288
1289         if (entry < nr_pages)
1290                 return 0;
1291
1292         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1293 }
1294 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1295
1296 static struct page *kvm_pfn_to_page(pfn_t pfn)
1297 {
1298         if (is_error_pfn(pfn))
1299                 return KVM_ERR_PTR_BAD_PAGE;
1300
1301         if (kvm_is_mmio_pfn(pfn)) {
1302                 WARN_ON(1);
1303                 return KVM_ERR_PTR_BAD_PAGE;
1304         }
1305
1306         return pfn_to_page(pfn);
1307 }
1308
1309 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1310 {
1311         pfn_t pfn;
1312
1313         pfn = gfn_to_pfn(kvm, gfn);
1314
1315         return kvm_pfn_to_page(pfn);
1316 }
1317
1318 EXPORT_SYMBOL_GPL(gfn_to_page);
1319
1320 void kvm_release_page_clean(struct page *page)
1321 {
1322         WARN_ON(is_error_page(page));
1323
1324         kvm_release_pfn_clean(page_to_pfn(page));
1325 }
1326 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1327
1328 void kvm_release_pfn_clean(pfn_t pfn)
1329 {
1330         WARN_ON(is_error_pfn(pfn));
1331
1332         if (!kvm_is_mmio_pfn(pfn))
1333                 put_page(pfn_to_page(pfn));
1334 }
1335 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1336
1337 void kvm_release_page_dirty(struct page *page)
1338 {
1339         WARN_ON(is_error_page(page));
1340
1341         kvm_release_pfn_dirty(page_to_pfn(page));
1342 }
1343 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1344
1345 void kvm_release_pfn_dirty(pfn_t pfn)
1346 {
1347         kvm_set_pfn_dirty(pfn);
1348         kvm_release_pfn_clean(pfn);
1349 }
1350 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1351
1352 void kvm_set_page_dirty(struct page *page)
1353 {
1354         kvm_set_pfn_dirty(page_to_pfn(page));
1355 }
1356 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1357
1358 void kvm_set_pfn_dirty(pfn_t pfn)
1359 {
1360         if (!kvm_is_mmio_pfn(pfn)) {
1361                 struct page *page = pfn_to_page(pfn);
1362                 if (!PageReserved(page))
1363                         SetPageDirty(page);
1364         }
1365 }
1366 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1367
1368 void kvm_set_pfn_accessed(pfn_t pfn)
1369 {
1370         if (!kvm_is_mmio_pfn(pfn))
1371                 mark_page_accessed(pfn_to_page(pfn));
1372 }
1373 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1374
1375 void kvm_get_pfn(pfn_t pfn)
1376 {
1377         if (!kvm_is_mmio_pfn(pfn))
1378                 get_page(pfn_to_page(pfn));
1379 }
1380 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1381
1382 static int next_segment(unsigned long len, int offset)
1383 {
1384         if (len > PAGE_SIZE - offset)
1385                 return PAGE_SIZE - offset;
1386         else
1387                 return len;
1388 }
1389
1390 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1391                         int len)
1392 {
1393         int r;
1394         unsigned long addr;
1395
1396         addr = gfn_to_hva_read(kvm, gfn);
1397         if (kvm_is_error_hva(addr))
1398                 return -EFAULT;
1399         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1400         if (r)
1401                 return -EFAULT;
1402         return 0;
1403 }
1404 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1405
1406 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1407 {
1408         gfn_t gfn = gpa >> PAGE_SHIFT;
1409         int seg;
1410         int offset = offset_in_page(gpa);
1411         int ret;
1412
1413         while ((seg = next_segment(len, offset)) != 0) {
1414                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1415                 if (ret < 0)
1416                         return ret;
1417                 offset = 0;
1418                 len -= seg;
1419                 data += seg;
1420                 ++gfn;
1421         }
1422         return 0;
1423 }
1424 EXPORT_SYMBOL_GPL(kvm_read_guest);
1425
1426 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1427                           unsigned long len)
1428 {
1429         int r;
1430         unsigned long addr;
1431         gfn_t gfn = gpa >> PAGE_SHIFT;
1432         int offset = offset_in_page(gpa);
1433
1434         addr = gfn_to_hva_read(kvm, gfn);
1435         if (kvm_is_error_hva(addr))
1436                 return -EFAULT;
1437         pagefault_disable();
1438         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1439         pagefault_enable();
1440         if (r)
1441                 return -EFAULT;
1442         return 0;
1443 }
1444 EXPORT_SYMBOL(kvm_read_guest_atomic);
1445
1446 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1447                          int offset, int len)
1448 {
1449         int r;
1450         unsigned long addr;
1451
1452         addr = gfn_to_hva(kvm, gfn);
1453         if (kvm_is_error_hva(addr))
1454                 return -EFAULT;
1455         r = __copy_to_user((void __user *)addr + offset, data, len);
1456         if (r)
1457                 return -EFAULT;
1458         mark_page_dirty(kvm, gfn);
1459         return 0;
1460 }
1461 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1462
1463 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1464                     unsigned long len)
1465 {
1466         gfn_t gfn = gpa >> PAGE_SHIFT;
1467         int seg;
1468         int offset = offset_in_page(gpa);
1469         int ret;
1470
1471         while ((seg = next_segment(len, offset)) != 0) {
1472                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1473                 if (ret < 0)
1474                         return ret;
1475                 offset = 0;
1476                 len -= seg;
1477                 data += seg;
1478                 ++gfn;
1479         }
1480         return 0;
1481 }
1482
1483 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1484                               gpa_t gpa)
1485 {
1486         struct kvm_memslots *slots = kvm_memslots(kvm);
1487         int offset = offset_in_page(gpa);
1488         gfn_t gfn = gpa >> PAGE_SHIFT;
1489
1490         ghc->gpa = gpa;
1491         ghc->generation = slots->generation;
1492         ghc->memslot = gfn_to_memslot(kvm, gfn);
1493         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1494         if (!kvm_is_error_hva(ghc->hva))
1495                 ghc->hva += offset;
1496         else
1497                 return -EFAULT;
1498
1499         return 0;
1500 }
1501 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1502
1503 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1504                            void *data, unsigned long len)
1505 {
1506         struct kvm_memslots *slots = kvm_memslots(kvm);
1507         int r;
1508
1509         if (slots->generation != ghc->generation)
1510                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1511
1512         if (kvm_is_error_hva(ghc->hva))
1513                 return -EFAULT;
1514
1515         r = __copy_to_user((void __user *)ghc->hva, data, len);
1516         if (r)
1517                 return -EFAULT;
1518         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1519
1520         return 0;
1521 }
1522 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1523
1524 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1525                            void *data, unsigned long len)
1526 {
1527         struct kvm_memslots *slots = kvm_memslots(kvm);
1528         int r;
1529
1530         if (slots->generation != ghc->generation)
1531                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1532
1533         if (kvm_is_error_hva(ghc->hva))
1534                 return -EFAULT;
1535
1536         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1537         if (r)
1538                 return -EFAULT;
1539
1540         return 0;
1541 }
1542 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1543
1544 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1545 {
1546         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1547                                     offset, len);
1548 }
1549 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1550
1551 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1552 {
1553         gfn_t gfn = gpa >> PAGE_SHIFT;
1554         int seg;
1555         int offset = offset_in_page(gpa);
1556         int ret;
1557
1558         while ((seg = next_segment(len, offset)) != 0) {
1559                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1560                 if (ret < 0)
1561                         return ret;
1562                 offset = 0;
1563                 len -= seg;
1564                 ++gfn;
1565         }
1566         return 0;
1567 }
1568 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1569
1570 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1571                              gfn_t gfn)
1572 {
1573         if (memslot && memslot->dirty_bitmap) {
1574                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1575
1576                 /* TODO: introduce set_bit_le() and use it */
1577                 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1578         }
1579 }
1580
1581 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1582 {
1583         struct kvm_memory_slot *memslot;
1584
1585         memslot = gfn_to_memslot(kvm, gfn);
1586         mark_page_dirty_in_slot(kvm, memslot, gfn);
1587 }
1588
1589 /*
1590  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1591  */
1592 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1593 {
1594         DEFINE_WAIT(wait);
1595
1596         for (;;) {
1597                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1598
1599                 if (kvm_arch_vcpu_runnable(vcpu)) {
1600                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1601                         break;
1602                 }
1603                 if (kvm_cpu_has_pending_timer(vcpu))
1604                         break;
1605                 if (signal_pending(current))
1606                         break;
1607
1608                 schedule();
1609         }
1610
1611         finish_wait(&vcpu->wq, &wait);
1612 }
1613
1614 #ifndef CONFIG_S390
1615 /*
1616  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1617  */
1618 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1619 {
1620         int me;
1621         int cpu = vcpu->cpu;
1622         wait_queue_head_t *wqp;
1623
1624         wqp = kvm_arch_vcpu_wq(vcpu);
1625         if (waitqueue_active(wqp)) {
1626                 wake_up_interruptible(wqp);
1627                 ++vcpu->stat.halt_wakeup;
1628         }
1629
1630         me = get_cpu();
1631         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1632                 if (kvm_arch_vcpu_should_kick(vcpu))
1633                         smp_send_reschedule(cpu);
1634         put_cpu();
1635 }
1636 #endif /* !CONFIG_S390 */
1637
1638 void kvm_resched(struct kvm_vcpu *vcpu)
1639 {
1640         if (!need_resched())
1641                 return;
1642         cond_resched();
1643 }
1644 EXPORT_SYMBOL_GPL(kvm_resched);
1645
1646 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1647 {
1648         struct pid *pid;
1649         struct task_struct *task = NULL;
1650
1651         rcu_read_lock();
1652         pid = rcu_dereference(target->pid);
1653         if (pid)
1654                 task = get_pid_task(target->pid, PIDTYPE_PID);
1655         rcu_read_unlock();
1656         if (!task)
1657                 return false;
1658         if (task->flags & PF_VCPU) {
1659                 put_task_struct(task);
1660                 return false;
1661         }
1662         if (yield_to(task, 1)) {
1663                 put_task_struct(task);
1664                 return true;
1665         }
1666         put_task_struct(task);
1667         return false;
1668 }
1669 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1670
1671 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1672 /*
1673  * Helper that checks whether a VCPU is eligible for directed yield.
1674  * Most eligible candidate to yield is decided by following heuristics:
1675  *
1676  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1677  *  (preempted lock holder), indicated by @in_spin_loop.
1678  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1679  *
1680  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1681  *  chance last time (mostly it has become eligible now since we have probably
1682  *  yielded to lockholder in last iteration. This is done by toggling
1683  *  @dy_eligible each time a VCPU checked for eligibility.)
1684  *
1685  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1686  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1687  *  burning. Giving priority for a potential lock-holder increases lock
1688  *  progress.
1689  *
1690  *  Since algorithm is based on heuristics, accessing another VCPU data without
1691  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1692  *  and continue with next VCPU and so on.
1693  */
1694 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1695 {
1696         bool eligible;
1697
1698         eligible = !vcpu->spin_loop.in_spin_loop ||
1699                         (vcpu->spin_loop.in_spin_loop &&
1700                          vcpu->spin_loop.dy_eligible);
1701
1702         if (vcpu->spin_loop.in_spin_loop)
1703                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1704
1705         return eligible;
1706 }
1707 #endif
1708 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1709 {
1710         struct kvm *kvm = me->kvm;
1711         struct kvm_vcpu *vcpu;
1712         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1713         int yielded = 0;
1714         int pass;
1715         int i;
1716
1717         kvm_vcpu_set_in_spin_loop(me, true);
1718         /*
1719          * We boost the priority of a VCPU that is runnable but not
1720          * currently running, because it got preempted by something
1721          * else and called schedule in __vcpu_run.  Hopefully that
1722          * VCPU is holding the lock that we need and will release it.
1723          * We approximate round-robin by starting at the last boosted VCPU.
1724          */
1725         for (pass = 0; pass < 2 && !yielded; pass++) {
1726                 kvm_for_each_vcpu(i, vcpu, kvm) {
1727                         if (!pass && i <= last_boosted_vcpu) {
1728                                 i = last_boosted_vcpu;
1729                                 continue;
1730                         } else if (pass && i > last_boosted_vcpu)
1731                                 break;
1732                         if (vcpu == me)
1733                                 continue;
1734                         if (waitqueue_active(&vcpu->wq))
1735                                 continue;
1736                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1737                                 continue;
1738                         if (kvm_vcpu_yield_to(vcpu)) {
1739                                 kvm->last_boosted_vcpu = i;
1740                                 yielded = 1;
1741                                 break;
1742                         }
1743                 }
1744         }
1745         kvm_vcpu_set_in_spin_loop(me, false);
1746
1747         /* Ensure vcpu is not eligible during next spinloop */
1748         kvm_vcpu_set_dy_eligible(me, false);
1749 }
1750 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1751
1752 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1753 {
1754         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1755         struct page *page;
1756
1757         if (vmf->pgoff == 0)
1758                 page = virt_to_page(vcpu->run);
1759 #ifdef CONFIG_X86
1760         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1761                 page = virt_to_page(vcpu->arch.pio_data);
1762 #endif
1763 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1764         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1765                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1766 #endif
1767         else
1768                 return kvm_arch_vcpu_fault(vcpu, vmf);
1769         get_page(page);
1770         vmf->page = page;
1771         return 0;
1772 }
1773
1774 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1775         .fault = kvm_vcpu_fault,
1776 };
1777
1778 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1779 {
1780         vma->vm_ops = &kvm_vcpu_vm_ops;
1781         return 0;
1782 }
1783
1784 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1785 {
1786         struct kvm_vcpu *vcpu = filp->private_data;
1787
1788         kvm_put_kvm(vcpu->kvm);
1789         return 0;
1790 }
1791
1792 static struct file_operations kvm_vcpu_fops = {
1793         .release        = kvm_vcpu_release,
1794         .unlocked_ioctl = kvm_vcpu_ioctl,
1795 #ifdef CONFIG_COMPAT
1796         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1797 #endif
1798         .mmap           = kvm_vcpu_mmap,
1799         .llseek         = noop_llseek,
1800 };
1801
1802 /*
1803  * Allocates an inode for the vcpu.
1804  */
1805 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1806 {
1807         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1808 }
1809
1810 /*
1811  * Creates some virtual cpus.  Good luck creating more than one.
1812  */
1813 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1814 {
1815         int r;
1816         struct kvm_vcpu *vcpu, *v;
1817
1818         vcpu = kvm_arch_vcpu_create(kvm, id);
1819         if (IS_ERR(vcpu))
1820                 return PTR_ERR(vcpu);
1821
1822         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1823
1824         r = kvm_arch_vcpu_setup(vcpu);
1825         if (r)
1826                 goto vcpu_destroy;
1827
1828         mutex_lock(&kvm->lock);
1829         if (!kvm_vcpu_compatible(vcpu)) {
1830                 r = -EINVAL;
1831                 goto unlock_vcpu_destroy;
1832         }
1833         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1834                 r = -EINVAL;
1835                 goto unlock_vcpu_destroy;
1836         }
1837
1838         kvm_for_each_vcpu(r, v, kvm)
1839                 if (v->vcpu_id == id) {
1840                         r = -EEXIST;
1841                         goto unlock_vcpu_destroy;
1842                 }
1843
1844         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1845
1846         /* Now it's all set up, let userspace reach it */
1847         kvm_get_kvm(kvm);
1848         r = create_vcpu_fd(vcpu);
1849         if (r < 0) {
1850                 kvm_put_kvm(kvm);
1851                 goto unlock_vcpu_destroy;
1852         }
1853
1854         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1855         smp_wmb();
1856         atomic_inc(&kvm->online_vcpus);
1857
1858         mutex_unlock(&kvm->lock);
1859         return r;
1860
1861 unlock_vcpu_destroy:
1862         mutex_unlock(&kvm->lock);
1863 vcpu_destroy:
1864         kvm_arch_vcpu_destroy(vcpu);
1865         return r;
1866 }
1867
1868 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1869 {
1870         if (sigset) {
1871                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1872                 vcpu->sigset_active = 1;
1873                 vcpu->sigset = *sigset;
1874         } else
1875                 vcpu->sigset_active = 0;
1876         return 0;
1877 }
1878
1879 static long kvm_vcpu_ioctl(struct file *filp,
1880                            unsigned int ioctl, unsigned long arg)
1881 {
1882         struct kvm_vcpu *vcpu = filp->private_data;
1883         void __user *argp = (void __user *)arg;
1884         int r;
1885         struct kvm_fpu *fpu = NULL;
1886         struct kvm_sregs *kvm_sregs = NULL;
1887
1888         if (vcpu->kvm->mm != current->mm)
1889                 return -EIO;
1890
1891 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1892         /*
1893          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1894          * so vcpu_load() would break it.
1895          */
1896         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1897                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1898 #endif
1899
1900
1901         vcpu_load(vcpu);
1902         switch (ioctl) {
1903         case KVM_RUN:
1904                 r = -EINVAL;
1905                 if (arg)
1906                         goto out;
1907                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1908                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1909                 break;
1910         case KVM_GET_REGS: {
1911                 struct kvm_regs *kvm_regs;
1912
1913                 r = -ENOMEM;
1914                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1915                 if (!kvm_regs)
1916                         goto out;
1917                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1918                 if (r)
1919                         goto out_free1;
1920                 r = -EFAULT;
1921                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1922                         goto out_free1;
1923                 r = 0;
1924 out_free1:
1925                 kfree(kvm_regs);
1926                 break;
1927         }
1928         case KVM_SET_REGS: {
1929                 struct kvm_regs *kvm_regs;
1930
1931                 r = -ENOMEM;
1932                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1933                 if (IS_ERR(kvm_regs)) {
1934                         r = PTR_ERR(kvm_regs);
1935                         goto out;
1936                 }
1937                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1938                 if (r)
1939                         goto out_free2;
1940                 r = 0;
1941 out_free2:
1942                 kfree(kvm_regs);
1943                 break;
1944         }
1945         case KVM_GET_SREGS: {
1946                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1947                 r = -ENOMEM;
1948                 if (!kvm_sregs)
1949                         goto out;
1950                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1951                 if (r)
1952                         goto out;
1953                 r = -EFAULT;
1954                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1955                         goto out;
1956                 r = 0;
1957                 break;
1958         }
1959         case KVM_SET_SREGS: {
1960                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1961                 if (IS_ERR(kvm_sregs)) {
1962                         r = PTR_ERR(kvm_sregs);
1963                         goto out;
1964                 }
1965                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1966                 if (r)
1967                         goto out;
1968                 r = 0;
1969                 break;
1970         }
1971         case KVM_GET_MP_STATE: {
1972                 struct kvm_mp_state mp_state;
1973
1974                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1975                 if (r)
1976                         goto out;
1977                 r = -EFAULT;
1978                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1979                         goto out;
1980                 r = 0;
1981                 break;
1982         }
1983         case KVM_SET_MP_STATE: {
1984                 struct kvm_mp_state mp_state;
1985
1986                 r = -EFAULT;
1987                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1988                         goto out;
1989                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1990                 if (r)
1991                         goto out;
1992                 r = 0;
1993                 break;
1994         }
1995         case KVM_TRANSLATE: {
1996                 struct kvm_translation tr;
1997
1998                 r = -EFAULT;
1999                 if (copy_from_user(&tr, argp, sizeof tr))
2000                         goto out;
2001                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2002                 if (r)
2003                         goto out;
2004                 r = -EFAULT;
2005                 if (copy_to_user(argp, &tr, sizeof tr))
2006                         goto out;
2007                 r = 0;
2008                 break;
2009         }
2010         case KVM_SET_GUEST_DEBUG: {
2011                 struct kvm_guest_debug dbg;
2012
2013                 r = -EFAULT;
2014                 if (copy_from_user(&dbg, argp, sizeof dbg))
2015                         goto out;
2016                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2017                 if (r)
2018                         goto out;
2019                 r = 0;
2020                 break;
2021         }
2022         case KVM_SET_SIGNAL_MASK: {
2023                 struct kvm_signal_mask __user *sigmask_arg = argp;
2024                 struct kvm_signal_mask kvm_sigmask;
2025                 sigset_t sigset, *p;
2026
2027                 p = NULL;
2028                 if (argp) {
2029                         r = -EFAULT;
2030                         if (copy_from_user(&kvm_sigmask, argp,
2031                                            sizeof kvm_sigmask))
2032                                 goto out;
2033                         r = -EINVAL;
2034                         if (kvm_sigmask.len != sizeof sigset)
2035                                 goto out;
2036                         r = -EFAULT;
2037                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2038                                            sizeof sigset))
2039                                 goto out;
2040                         p = &sigset;
2041                 }
2042                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2043                 break;
2044         }
2045         case KVM_GET_FPU: {
2046                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2047                 r = -ENOMEM;
2048                 if (!fpu)
2049                         goto out;
2050                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2051                 if (r)
2052                         goto out;
2053                 r = -EFAULT;
2054                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2055                         goto out;
2056                 r = 0;
2057                 break;
2058         }
2059         case KVM_SET_FPU: {
2060                 fpu = memdup_user(argp, sizeof(*fpu));
2061                 if (IS_ERR(fpu)) {
2062                         r = PTR_ERR(fpu);
2063                         goto out;
2064                 }
2065                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2066                 if (r)
2067                         goto out;
2068                 r = 0;
2069                 break;
2070         }
2071         default:
2072                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2073         }
2074 out:
2075         vcpu_put(vcpu);
2076         kfree(fpu);
2077         kfree(kvm_sregs);
2078         return r;
2079 }
2080
2081 #ifdef CONFIG_COMPAT
2082 static long kvm_vcpu_compat_ioctl(struct file *filp,
2083                                   unsigned int ioctl, unsigned long arg)
2084 {
2085         struct kvm_vcpu *vcpu = filp->private_data;
2086         void __user *argp = compat_ptr(arg);
2087         int r;
2088
2089         if (vcpu->kvm->mm != current->mm)
2090                 return -EIO;
2091
2092         switch (ioctl) {
2093         case KVM_SET_SIGNAL_MASK: {
2094                 struct kvm_signal_mask __user *sigmask_arg = argp;
2095                 struct kvm_signal_mask kvm_sigmask;
2096                 compat_sigset_t csigset;
2097                 sigset_t sigset;
2098
2099                 if (argp) {
2100                         r = -EFAULT;
2101                         if (copy_from_user(&kvm_sigmask, argp,
2102                                            sizeof kvm_sigmask))
2103                                 goto out;
2104                         r = -EINVAL;
2105                         if (kvm_sigmask.len != sizeof csigset)
2106                                 goto out;
2107                         r = -EFAULT;
2108                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2109                                            sizeof csigset))
2110                                 goto out;
2111                 }
2112                 sigset_from_compat(&sigset, &csigset);
2113                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2114                 break;
2115         }
2116         default:
2117                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2118         }
2119
2120 out:
2121         return r;
2122 }
2123 #endif
2124
2125 static long kvm_vm_ioctl(struct file *filp,
2126                            unsigned int ioctl, unsigned long arg)
2127 {
2128         struct kvm *kvm = filp->private_data;
2129         void __user *argp = (void __user *)arg;
2130         int r;
2131
2132         if (kvm->mm != current->mm)
2133                 return -EIO;
2134         switch (ioctl) {
2135         case KVM_CREATE_VCPU:
2136                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2137                 if (r < 0)
2138                         goto out;
2139                 break;
2140         case KVM_SET_USER_MEMORY_REGION: {
2141                 struct kvm_userspace_memory_region kvm_userspace_mem;
2142
2143                 r = -EFAULT;
2144                 if (copy_from_user(&kvm_userspace_mem, argp,
2145                                                 sizeof kvm_userspace_mem))
2146                         goto out;
2147
2148                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2149                 if (r)
2150                         goto out;
2151                 break;
2152         }
2153         case KVM_GET_DIRTY_LOG: {
2154                 struct kvm_dirty_log log;
2155
2156                 r = -EFAULT;
2157                 if (copy_from_user(&log, argp, sizeof log))
2158                         goto out;
2159                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2160                 if (r)
2161                         goto out;
2162                 break;
2163         }
2164 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2165         case KVM_REGISTER_COALESCED_MMIO: {
2166                 struct kvm_coalesced_mmio_zone zone;
2167                 r = -EFAULT;
2168                 if (copy_from_user(&zone, argp, sizeof zone))
2169                         goto out;
2170                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2171                 if (r)
2172                         goto out;
2173                 r = 0;
2174                 break;
2175         }
2176         case KVM_UNREGISTER_COALESCED_MMIO: {
2177                 struct kvm_coalesced_mmio_zone zone;
2178                 r = -EFAULT;
2179                 if (copy_from_user(&zone, argp, sizeof zone))
2180                         goto out;
2181                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2182                 if (r)
2183                         goto out;
2184                 r = 0;
2185                 break;
2186         }
2187 #endif
2188         case KVM_IRQFD: {
2189                 struct kvm_irqfd data;
2190
2191                 r = -EFAULT;
2192                 if (copy_from_user(&data, argp, sizeof data))
2193                         goto out;
2194                 r = kvm_irqfd(kvm, &data);
2195                 break;
2196         }
2197         case KVM_IOEVENTFD: {
2198                 struct kvm_ioeventfd data;
2199
2200                 r = -EFAULT;
2201                 if (copy_from_user(&data, argp, sizeof data))
2202                         goto out;
2203                 r = kvm_ioeventfd(kvm, &data);
2204                 break;
2205         }
2206 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2207         case KVM_SET_BOOT_CPU_ID:
2208                 r = 0;
2209                 mutex_lock(&kvm->lock);
2210                 if (atomic_read(&kvm->online_vcpus) != 0)
2211                         r = -EBUSY;
2212                 else
2213                         kvm->bsp_vcpu_id = arg;
2214                 mutex_unlock(&kvm->lock);
2215                 break;
2216 #endif
2217 #ifdef CONFIG_HAVE_KVM_MSI
2218         case KVM_SIGNAL_MSI: {
2219                 struct kvm_msi msi;
2220
2221                 r = -EFAULT;
2222                 if (copy_from_user(&msi, argp, sizeof msi))
2223                         goto out;
2224                 r = kvm_send_userspace_msi(kvm, &msi);
2225                 break;
2226         }
2227 #endif
2228 #ifdef __KVM_HAVE_IRQ_LINE
2229         case KVM_IRQ_LINE_STATUS:
2230         case KVM_IRQ_LINE: {
2231                 struct kvm_irq_level irq_event;
2232
2233                 r = -EFAULT;
2234                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2235                         goto out;
2236
2237                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2238                 if (r)
2239                         goto out;
2240
2241                 r = -EFAULT;
2242                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2243                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2244                                 goto out;
2245                 }
2246
2247                 r = 0;
2248                 break;
2249         }
2250 #endif
2251         default:
2252                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2253                 if (r == -ENOTTY)
2254                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2255         }
2256 out:
2257         return r;
2258 }
2259
2260 #ifdef CONFIG_COMPAT
2261 struct compat_kvm_dirty_log {
2262         __u32 slot;
2263         __u32 padding1;
2264         union {
2265                 compat_uptr_t dirty_bitmap; /* one bit per page */
2266                 __u64 padding2;
2267         };
2268 };
2269
2270 static long kvm_vm_compat_ioctl(struct file *filp,
2271                            unsigned int ioctl, unsigned long arg)
2272 {
2273         struct kvm *kvm = filp->private_data;
2274         int r;
2275
2276         if (kvm->mm != current->mm)
2277                 return -EIO;
2278         switch (ioctl) {
2279         case KVM_GET_DIRTY_LOG: {
2280                 struct compat_kvm_dirty_log compat_log;
2281                 struct kvm_dirty_log log;
2282
2283                 r = -EFAULT;
2284                 if (copy_from_user(&compat_log, (void __user *)arg,
2285                                    sizeof(compat_log)))
2286                         goto out;
2287                 log.slot         = compat_log.slot;
2288                 log.padding1     = compat_log.padding1;
2289                 log.padding2     = compat_log.padding2;
2290                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2291
2292                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2293                 if (r)
2294                         goto out;
2295                 break;
2296         }
2297         default:
2298                 r = kvm_vm_ioctl(filp, ioctl, arg);
2299         }
2300
2301 out:
2302         return r;
2303 }
2304 #endif
2305
2306 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2307 {
2308         struct page *page[1];
2309         unsigned long addr;
2310         int npages;
2311         gfn_t gfn = vmf->pgoff;
2312         struct kvm *kvm = vma->vm_file->private_data;
2313
2314         addr = gfn_to_hva(kvm, gfn);
2315         if (kvm_is_error_hva(addr))
2316                 return VM_FAULT_SIGBUS;
2317
2318         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2319                                 NULL);
2320         if (unlikely(npages != 1))
2321                 return VM_FAULT_SIGBUS;
2322
2323         vmf->page = page[0];
2324         return 0;
2325 }
2326
2327 static const struct vm_operations_struct kvm_vm_vm_ops = {
2328         .fault = kvm_vm_fault,
2329 };
2330
2331 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2332 {
2333         vma->vm_ops = &kvm_vm_vm_ops;
2334         return 0;
2335 }
2336
2337 static struct file_operations kvm_vm_fops = {
2338         .release        = kvm_vm_release,
2339         .unlocked_ioctl = kvm_vm_ioctl,
2340 #ifdef CONFIG_COMPAT
2341         .compat_ioctl   = kvm_vm_compat_ioctl,
2342 #endif
2343         .mmap           = kvm_vm_mmap,
2344         .llseek         = noop_llseek,
2345 };
2346
2347 static int kvm_dev_ioctl_create_vm(unsigned long type)
2348 {
2349         int r;
2350         struct kvm *kvm;
2351
2352         kvm = kvm_create_vm(type);
2353         if (IS_ERR(kvm))
2354                 return PTR_ERR(kvm);
2355 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2356         r = kvm_coalesced_mmio_init(kvm);
2357         if (r < 0) {
2358                 kvm_put_kvm(kvm);
2359                 return r;
2360         }
2361 #endif
2362         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2363         if (r < 0)
2364                 kvm_put_kvm(kvm);
2365
2366         return r;
2367 }
2368
2369 static long kvm_dev_ioctl_check_extension_generic(long arg)
2370 {
2371         switch (arg) {
2372         case KVM_CAP_USER_MEMORY:
2373         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2374         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2375 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2376         case KVM_CAP_SET_BOOT_CPU_ID:
2377 #endif
2378         case KVM_CAP_INTERNAL_ERROR_DATA:
2379 #ifdef CONFIG_HAVE_KVM_MSI
2380         case KVM_CAP_SIGNAL_MSI:
2381 #endif
2382                 return 1;
2383 #ifdef KVM_CAP_IRQ_ROUTING
2384         case KVM_CAP_IRQ_ROUTING:
2385                 return KVM_MAX_IRQ_ROUTES;
2386 #endif
2387         default:
2388                 break;
2389         }
2390         return kvm_dev_ioctl_check_extension(arg);
2391 }
2392
2393 static long kvm_dev_ioctl(struct file *filp,
2394                           unsigned int ioctl, unsigned long arg)
2395 {
2396         long r = -EINVAL;
2397
2398         switch (ioctl) {
2399         case KVM_GET_API_VERSION:
2400                 r = -EINVAL;
2401                 if (arg)
2402                         goto out;
2403                 r = KVM_API_VERSION;
2404                 break;
2405         case KVM_CREATE_VM:
2406                 r = kvm_dev_ioctl_create_vm(arg);
2407                 break;
2408         case KVM_CHECK_EXTENSION:
2409                 r = kvm_dev_ioctl_check_extension_generic(arg);
2410                 break;
2411         case KVM_GET_VCPU_MMAP_SIZE:
2412                 r = -EINVAL;
2413                 if (arg)
2414                         goto out;
2415                 r = PAGE_SIZE;     /* struct kvm_run */
2416 #ifdef CONFIG_X86
2417                 r += PAGE_SIZE;    /* pio data page */
2418 #endif
2419 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2420                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2421 #endif
2422                 break;
2423         case KVM_TRACE_ENABLE:
2424         case KVM_TRACE_PAUSE:
2425         case KVM_TRACE_DISABLE:
2426                 r = -EOPNOTSUPP;
2427                 break;
2428         default:
2429                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2430         }
2431 out:
2432         return r;
2433 }
2434
2435 static struct file_operations kvm_chardev_ops = {
2436         .unlocked_ioctl = kvm_dev_ioctl,
2437         .compat_ioctl   = kvm_dev_ioctl,
2438         .llseek         = noop_llseek,
2439 };
2440
2441 static struct miscdevice kvm_dev = {
2442         KVM_MINOR,
2443         "kvm",
2444         &kvm_chardev_ops,
2445 };
2446
2447 static void hardware_enable_nolock(void *junk)
2448 {
2449         int cpu = raw_smp_processor_id();
2450         int r;
2451
2452         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2453                 return;
2454
2455         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2456
2457         r = kvm_arch_hardware_enable(NULL);
2458
2459         if (r) {
2460                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2461                 atomic_inc(&hardware_enable_failed);
2462                 printk(KERN_INFO "kvm: enabling virtualization on "
2463                                  "CPU%d failed\n", cpu);
2464         }
2465 }
2466
2467 static void hardware_enable(void *junk)
2468 {
2469         raw_spin_lock(&kvm_lock);
2470         hardware_enable_nolock(junk);
2471         raw_spin_unlock(&kvm_lock);
2472 }
2473
2474 static void hardware_disable_nolock(void *junk)
2475 {
2476         int cpu = raw_smp_processor_id();
2477
2478         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2479                 return;
2480         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2481         kvm_arch_hardware_disable(NULL);
2482 }
2483
2484 static void hardware_disable(void *junk)
2485 {
2486         raw_spin_lock(&kvm_lock);
2487         hardware_disable_nolock(junk);
2488         raw_spin_unlock(&kvm_lock);
2489 }
2490
2491 static void hardware_disable_all_nolock(void)
2492 {
2493         BUG_ON(!kvm_usage_count);
2494
2495         kvm_usage_count--;
2496         if (!kvm_usage_count)
2497                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2498 }
2499
2500 static void hardware_disable_all(void)
2501 {
2502         raw_spin_lock(&kvm_lock);
2503         hardware_disable_all_nolock();
2504         raw_spin_unlock(&kvm_lock);
2505 }
2506
2507 static int hardware_enable_all(void)
2508 {
2509         int r = 0;
2510
2511         raw_spin_lock(&kvm_lock);
2512
2513         kvm_usage_count++;
2514         if (kvm_usage_count == 1) {
2515                 atomic_set(&hardware_enable_failed, 0);
2516                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2517
2518                 if (atomic_read(&hardware_enable_failed)) {
2519                         hardware_disable_all_nolock();
2520                         r = -EBUSY;
2521                 }
2522         }
2523
2524         raw_spin_unlock(&kvm_lock);
2525
2526         return r;
2527 }
2528
2529 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2530                            void *v)
2531 {
2532         int cpu = (long)v;
2533
2534         if (!kvm_usage_count)
2535                 return NOTIFY_OK;
2536
2537         val &= ~CPU_TASKS_FROZEN;
2538         switch (val) {
2539         case CPU_DYING:
2540                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2541                        cpu);
2542                 hardware_disable(NULL);
2543                 break;
2544         case CPU_STARTING:
2545                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2546                        cpu);
2547                 hardware_enable(NULL);
2548                 break;
2549         }
2550         return NOTIFY_OK;
2551 }
2552
2553
2554 asmlinkage void kvm_spurious_fault(void)
2555 {
2556         /* Fault while not rebooting.  We want the trace. */
2557         BUG();
2558 }
2559 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2560
2561 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2562                       void *v)
2563 {
2564         /*
2565          * Some (well, at least mine) BIOSes hang on reboot if
2566          * in vmx root mode.
2567          *
2568          * And Intel TXT required VMX off for all cpu when system shutdown.
2569          */
2570         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2571         kvm_rebooting = true;
2572         on_each_cpu(hardware_disable_nolock, NULL, 1);
2573         return NOTIFY_OK;
2574 }
2575
2576 static struct notifier_block kvm_reboot_notifier = {
2577         .notifier_call = kvm_reboot,
2578         .priority = 0,
2579 };
2580
2581 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2582 {
2583         int i;
2584
2585         for (i = 0; i < bus->dev_count; i++) {
2586                 struct kvm_io_device *pos = bus->range[i].dev;
2587
2588                 kvm_iodevice_destructor(pos);
2589         }
2590         kfree(bus);
2591 }
2592
2593 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2594 {
2595         const struct kvm_io_range *r1 = p1;
2596         const struct kvm_io_range *r2 = p2;
2597
2598         if (r1->addr < r2->addr)
2599                 return -1;
2600         if (r1->addr + r1->len > r2->addr + r2->len)
2601                 return 1;
2602         return 0;
2603 }
2604
2605 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2606                           gpa_t addr, int len)
2607 {
2608         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2609                 .addr = addr,
2610                 .len = len,
2611                 .dev = dev,
2612         };
2613
2614         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2615                 kvm_io_bus_sort_cmp, NULL);
2616
2617         return 0;
2618 }
2619
2620 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2621                              gpa_t addr, int len)
2622 {
2623         struct kvm_io_range *range, key;
2624         int off;
2625
2626         key = (struct kvm_io_range) {
2627                 .addr = addr,
2628                 .len = len,
2629         };
2630
2631         range = bsearch(&key, bus->range, bus->dev_count,
2632                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2633         if (range == NULL)
2634                 return -ENOENT;
2635
2636         off = range - bus->range;
2637
2638         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2639                 off--;
2640
2641         return off;
2642 }
2643
2644 /* kvm_io_bus_write - called under kvm->slots_lock */
2645 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2646                      int len, const void *val)
2647 {
2648         int idx;
2649         struct kvm_io_bus *bus;
2650         struct kvm_io_range range;
2651
2652         range = (struct kvm_io_range) {
2653                 .addr = addr,
2654                 .len = len,
2655         };
2656
2657         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2658         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2659         if (idx < 0)
2660                 return -EOPNOTSUPP;
2661
2662         while (idx < bus->dev_count &&
2663                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2664                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2665                         return 0;
2666                 idx++;
2667         }
2668
2669         return -EOPNOTSUPP;
2670 }
2671
2672 /* kvm_io_bus_read - called under kvm->slots_lock */
2673 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2674                     int len, void *val)
2675 {
2676         int idx;
2677         struct kvm_io_bus *bus;
2678         struct kvm_io_range range;
2679
2680         range = (struct kvm_io_range) {
2681                 .addr = addr,
2682                 .len = len,
2683         };
2684
2685         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2686         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2687         if (idx < 0)
2688                 return -EOPNOTSUPP;
2689
2690         while (idx < bus->dev_count &&
2691                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2692                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2693                         return 0;
2694                 idx++;
2695         }
2696
2697         return -EOPNOTSUPP;
2698 }
2699
2700 /* Caller must hold slots_lock. */
2701 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2702                             int len, struct kvm_io_device *dev)
2703 {
2704         struct kvm_io_bus *new_bus, *bus;
2705
2706         bus = kvm->buses[bus_idx];
2707         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2708                 return -ENOSPC;
2709
2710         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2711                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2712         if (!new_bus)
2713                 return -ENOMEM;
2714         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2715                sizeof(struct kvm_io_range)));
2716         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2717         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2718         synchronize_srcu_expedited(&kvm->srcu);
2719         kfree(bus);
2720
2721         return 0;
2722 }
2723
2724 /* Caller must hold slots_lock. */
2725 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2726                               struct kvm_io_device *dev)
2727 {
2728         int i, r;
2729         struct kvm_io_bus *new_bus, *bus;
2730
2731         bus = kvm->buses[bus_idx];
2732         r = -ENOENT;
2733         for (i = 0; i < bus->dev_count; i++)
2734                 if (bus->range[i].dev == dev) {
2735                         r = 0;
2736                         break;
2737                 }
2738
2739         if (r)
2740                 return r;
2741
2742         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2743                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2744         if (!new_bus)
2745                 return -ENOMEM;
2746
2747         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2748         new_bus->dev_count--;
2749         memcpy(new_bus->range + i, bus->range + i + 1,
2750                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2751
2752         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2753         synchronize_srcu_expedited(&kvm->srcu);
2754         kfree(bus);
2755         return r;
2756 }
2757
2758 static struct notifier_block kvm_cpu_notifier = {
2759         .notifier_call = kvm_cpu_hotplug,
2760 };
2761
2762 static int vm_stat_get(void *_offset, u64 *val)
2763 {
2764         unsigned offset = (long)_offset;
2765         struct kvm *kvm;
2766
2767         *val = 0;
2768         raw_spin_lock(&kvm_lock);
2769         list_for_each_entry(kvm, &vm_list, vm_list)
2770                 *val += *(u32 *)((void *)kvm + offset);
2771         raw_spin_unlock(&kvm_lock);
2772         return 0;
2773 }
2774
2775 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2776
2777 static int vcpu_stat_get(void *_offset, u64 *val)
2778 {
2779         unsigned offset = (long)_offset;
2780         struct kvm *kvm;
2781         struct kvm_vcpu *vcpu;
2782         int i;
2783
2784         *val = 0;
2785         raw_spin_lock(&kvm_lock);
2786         list_for_each_entry(kvm, &vm_list, vm_list)
2787                 kvm_for_each_vcpu(i, vcpu, kvm)
2788                         *val += *(u32 *)((void *)vcpu + offset);
2789
2790         raw_spin_unlock(&kvm_lock);
2791         return 0;
2792 }
2793
2794 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2795
2796 static const struct file_operations *stat_fops[] = {
2797         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2798         [KVM_STAT_VM]   = &vm_stat_fops,
2799 };
2800
2801 static int kvm_init_debug(void)
2802 {
2803         int r = -EFAULT;
2804         struct kvm_stats_debugfs_item *p;
2805
2806         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2807         if (kvm_debugfs_dir == NULL)
2808                 goto out;
2809
2810         for (p = debugfs_entries; p->name; ++p) {
2811                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2812                                                 (void *)(long)p->offset,
2813                                                 stat_fops[p->kind]);
2814                 if (p->dentry == NULL)
2815                         goto out_dir;
2816         }
2817
2818         return 0;
2819
2820 out_dir:
2821         debugfs_remove_recursive(kvm_debugfs_dir);
2822 out:
2823         return r;
2824 }
2825
2826 static void kvm_exit_debug(void)
2827 {
2828         struct kvm_stats_debugfs_item *p;
2829
2830         for (p = debugfs_entries; p->name; ++p)
2831                 debugfs_remove(p->dentry);
2832         debugfs_remove(kvm_debugfs_dir);
2833 }
2834
2835 static int kvm_suspend(void)
2836 {
2837         if (kvm_usage_count)
2838                 hardware_disable_nolock(NULL);
2839         return 0;
2840 }
2841
2842 static void kvm_resume(void)
2843 {
2844         if (kvm_usage_count) {
2845                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2846                 hardware_enable_nolock(NULL);
2847         }
2848 }
2849
2850 static struct syscore_ops kvm_syscore_ops = {
2851         .suspend = kvm_suspend,
2852         .resume = kvm_resume,
2853 };
2854
2855 static inline
2856 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2857 {
2858         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2859 }
2860
2861 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2862 {
2863         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2864
2865         kvm_arch_vcpu_load(vcpu, cpu);
2866 }
2867
2868 static void kvm_sched_out(struct preempt_notifier *pn,
2869                           struct task_struct *next)
2870 {
2871         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2872
2873         kvm_arch_vcpu_put(vcpu);
2874 }
2875
2876 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2877                   struct module *module)
2878 {
2879         int r;
2880         int cpu;
2881
2882         r = kvm_arch_init(opaque);
2883         if (r)
2884                 goto out_fail;
2885
2886         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2887                 r = -ENOMEM;
2888                 goto out_free_0;
2889         }
2890
2891         r = kvm_arch_hardware_setup();
2892         if (r < 0)
2893                 goto out_free_0a;
2894
2895         for_each_online_cpu(cpu) {
2896                 smp_call_function_single(cpu,
2897                                 kvm_arch_check_processor_compat,
2898                                 &r, 1);
2899                 if (r < 0)
2900                         goto out_free_1;
2901         }
2902
2903         r = register_cpu_notifier(&kvm_cpu_notifier);
2904         if (r)
2905                 goto out_free_2;
2906         register_reboot_notifier(&kvm_reboot_notifier);
2907
2908         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2909         if (!vcpu_align)
2910                 vcpu_align = __alignof__(struct kvm_vcpu);
2911         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2912                                            0, NULL);
2913         if (!kvm_vcpu_cache) {
2914                 r = -ENOMEM;
2915                 goto out_free_3;
2916         }
2917
2918         r = kvm_async_pf_init();
2919         if (r)
2920                 goto out_free;
2921
2922         kvm_chardev_ops.owner = module;
2923         kvm_vm_fops.owner = module;
2924         kvm_vcpu_fops.owner = module;
2925
2926         r = misc_register(&kvm_dev);
2927         if (r) {
2928                 printk(KERN_ERR "kvm: misc device register failed\n");
2929                 goto out_unreg;
2930         }
2931
2932         register_syscore_ops(&kvm_syscore_ops);
2933
2934         kvm_preempt_ops.sched_in = kvm_sched_in;
2935         kvm_preempt_ops.sched_out = kvm_sched_out;
2936
2937         r = kvm_init_debug();
2938         if (r) {
2939                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2940                 goto out_undebugfs;
2941         }
2942
2943         return 0;
2944
2945 out_undebugfs:
2946         unregister_syscore_ops(&kvm_syscore_ops);
2947 out_unreg:
2948         kvm_async_pf_deinit();
2949 out_free:
2950         kmem_cache_destroy(kvm_vcpu_cache);
2951 out_free_3:
2952         unregister_reboot_notifier(&kvm_reboot_notifier);
2953         unregister_cpu_notifier(&kvm_cpu_notifier);
2954 out_free_2:
2955 out_free_1:
2956         kvm_arch_hardware_unsetup();
2957 out_free_0a:
2958         free_cpumask_var(cpus_hardware_enabled);
2959 out_free_0:
2960         kvm_arch_exit();
2961 out_fail:
2962         return r;
2963 }
2964 EXPORT_SYMBOL_GPL(kvm_init);
2965
2966 void kvm_exit(void)
2967 {
2968         kvm_exit_debug();
2969         misc_deregister(&kvm_dev);
2970         kmem_cache_destroy(kvm_vcpu_cache);
2971         kvm_async_pf_deinit();
2972         unregister_syscore_ops(&kvm_syscore_ops);
2973         unregister_reboot_notifier(&kvm_reboot_notifier);
2974         unregister_cpu_notifier(&kvm_cpu_notifier);
2975         on_each_cpu(hardware_disable_nolock, NULL, 1);
2976         kvm_arch_hardware_unsetup();
2977         kvm_arch_exit();
2978         free_cpumask_var(cpus_hardware_enabled);
2979 }
2980 EXPORT_SYMBOL_GPL(kvm_exit);