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 int vcpu_load(struct kvm_vcpu *vcpu)
 135 {
 136         int cpu;
 137
 138         if (mutex_lock_killable(&vcpu->mutex))
 139                 return -EINTR;
 140         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 141                 /* The thread running this VCPU changed. */
 142                 struct pid *oldpid = vcpu->pid;
 143                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 144                 rcu_assign_pointer(vcpu->pid, newpid);
 145                 synchronize_rcu();
 146                 put_pid(oldpid);
 147         }
 148         cpu = get_cpu();
 149         preempt_notifier_register(&vcpu->preempt_notifier);
 150         kvm_arch_vcpu_load(vcpu, cpu);
 151         put_cpu();
 152         return 0;
 153 }
 154
 155 void vcpu_put(struct kvm_vcpu *vcpu)
 156 {
 157         preempt_disable();
 158         kvm_arch_vcpu_put(vcpu);
 159         preempt_notifier_unregister(&vcpu->preempt_notifier);
 160         preempt_enable();
 161         mutex_unlock(&vcpu->mutex);
 162 }
 163
 164 static void ack_flush(void *_completed)
 165 {
 166 }
 167
 168 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 169 {
 170         int i, cpu, me;
 171         cpumask_var_t cpus;
 172         bool called = true;
 173         struct kvm_vcpu *vcpu;
 174
 175         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 176
 177         me = get_cpu();
 178         kvm_for_each_vcpu(i, vcpu, kvm) {
 179                 kvm_make_request(req, vcpu);
 180                 cpu = vcpu->cpu;
 181
 182                 /* Set ->requests bit before we read ->mode */
 183                 smp_mb();
 184
 185                 if (cpus != NULL && cpu != -1 && cpu != me &&
 186                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 187                         cpumask_set_cpu(cpu, cpus);
 188         }
 189         if (unlikely(cpus == NULL))
 190                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 191         else if (!cpumask_empty(cpus))
 192                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 193         else
 194                 called = false;
 195         put_cpu();
 196         free_cpumask_var(cpus);
 197         return called;
 198 }
 199
 200 void kvm_flush_remote_tlbs(struct kvm *kvm)
 201 {
 202         long dirty_count = kvm->tlbs_dirty;
 203
 204         smp_mb();
 205         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 206                 ++kvm->stat.remote_tlb_flush;
 207         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 208 }
 209
 210 void kvm_reload_remote_mmus(struct kvm *kvm)
 211 {
 212         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 213 }
 214
 215 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 216 {
 217         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 218 }
 219
 220 void kvm_make_update_eoibitmap_request(struct kvm *kvm)
 221 {
 222         make_all_cpus_request(kvm, KVM_REQ_EOIBITMAP);
 223 }
 224
 225 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 226 {
 227         struct page *page;
 228         int r;
 229
 230         mutex_init(&vcpu->mutex);
 231         vcpu->cpu = -1;
 232         vcpu->kvm = kvm;
 233         vcpu->vcpu_id = id;
 234         vcpu->pid = NULL;
 235         init_waitqueue_head(&vcpu->wq);
 236         kvm_async_pf_vcpu_init(vcpu);
 237
 238         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 239         if (!page) {
 240                 r = -ENOMEM;
 241                 goto fail;
 242         }
 243         vcpu->run = page_address(page);
 244
 245         kvm_vcpu_set_in_spin_loop(vcpu, false);
 246         kvm_vcpu_set_dy_eligible(vcpu, false);
 247
 248         r = kvm_arch_vcpu_init(vcpu);
 249         if (r < 0)
 250                 goto fail_free_run;
 251         return 0;
 252
 253 fail_free_run:
 254         free_page((unsigned long)vcpu->run);
 255 fail:
 256         return r;
 257 }
 258 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 259
 260 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 261 {
 262         put_pid(vcpu->pid);
 263         kvm_arch_vcpu_uninit(vcpu);
 264         free_page((unsigned long)vcpu->run);
 265 }
 266 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 267
 268 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 269 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 270 {
 271         return container_of(mn, struct kvm, mmu_notifier);
 272 }
 273
 274 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 275                                              struct mm_struct *mm,
 276                                              unsigned long address)
 277 {
 278         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 279         int need_tlb_flush, idx;
 280
 281         /*
 282          * When ->invalidate_page runs, the linux pte has been zapped
 283          * already but the page is still allocated until
 284          * ->invalidate_page returns. So if we increase the sequence
 285          * here the kvm page fault will notice if the spte can't be
 286          * established because the page is going to be freed. If
 287          * instead the kvm page fault establishes the spte before
 288          * ->invalidate_page runs, kvm_unmap_hva will release it
 289          * before returning.
 290          *
 291          * The sequence increase only need to be seen at spin_unlock
 292          * time, and not at spin_lock time.
 293          *
 294          * Increasing the sequence after the spin_unlock would be
 295          * unsafe because the kvm page fault could then establish the
 296          * pte after kvm_unmap_hva returned, without noticing the page
 297          * is going to be freed.
 298          */
 299         idx = srcu_read_lock(&kvm->srcu);
 300         spin_lock(&kvm->mmu_lock);
 301
 302         kvm->mmu_notifier_seq++;
 303         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 304         /* we've to flush the tlb before the pages can be freed */
 305         if (need_tlb_flush)
 306                 kvm_flush_remote_tlbs(kvm);
 307
 308         spin_unlock(&kvm->mmu_lock);
 309         srcu_read_unlock(&kvm->srcu, idx);
 310 }
 311
 312 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 313                                         struct mm_struct *mm,
 314                                         unsigned long address,
 315                                         pte_t pte)
 316 {
 317         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 318         int idx;
 319
 320         idx = srcu_read_lock(&kvm->srcu);
 321         spin_lock(&kvm->mmu_lock);
 322         kvm->mmu_notifier_seq++;
 323         kvm_set_spte_hva(kvm, address, pte);
 324         spin_unlock(&kvm->mmu_lock);
 325         srcu_read_unlock(&kvm->srcu, idx);
 326 }
 327
 328 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 329                                                     struct mm_struct *mm,
 330                                                     unsigned long start,
 331                                                     unsigned long end)
 332 {
 333         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 334         int need_tlb_flush = 0, idx;
 335
 336         idx = srcu_read_lock(&kvm->srcu);
 337         spin_lock(&kvm->mmu_lock);
 338         /*
 339          * The count increase must become visible at unlock time as no
 340          * spte can be established without taking the mmu_lock and
 341          * count is also read inside the mmu_lock critical section.
 342          */
 343         kvm->mmu_notifier_count++;
 344         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 345         need_tlb_flush |= kvm->tlbs_dirty;
 346         /* we've to flush the tlb before the pages can be freed */
 347         if (need_tlb_flush)
 348                 kvm_flush_remote_tlbs(kvm);
 349
 350         spin_unlock(&kvm->mmu_lock);
 351         srcu_read_unlock(&kvm->srcu, idx);
 352 }
 353
 354 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 355                                                   struct mm_struct *mm,
 356                                                   unsigned long start,
 357                                                   unsigned long end)
 358 {
 359         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 360
 361         spin_lock(&kvm->mmu_lock);
 362         /*
 363          * This sequence increase will notify the kvm page fault that
 364          * the page that is going to be mapped in the spte could have
 365          * been freed.
 366          */
 367         kvm->mmu_notifier_seq++;
 368         smp_wmb();
 369         /*
 370          * The above sequence increase must be visible before the
 371          * below count decrease, which is ensured by the smp_wmb above
 372          * in conjunction with the smp_rmb in mmu_notifier_retry().
 373          */
 374         kvm->mmu_notifier_count--;
 375         spin_unlock(&kvm->mmu_lock);
 376
 377         BUG_ON(kvm->mmu_notifier_count < 0);
 378 }
 379
 380 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 381                                               struct mm_struct *mm,
 382                                               unsigned long address)
 383 {
 384         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 385         int young, idx;
 386
 387         idx = srcu_read_lock(&kvm->srcu);
 388         spin_lock(&kvm->mmu_lock);
 389
 390         young = kvm_age_hva(kvm, address);
 391         if (young)
 392                 kvm_flush_remote_tlbs(kvm);
 393
 394         spin_unlock(&kvm->mmu_lock);
 395         srcu_read_unlock(&kvm->srcu, idx);
 396
 397         return young;
 398 }
 399
 400 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 401                                        struct mm_struct *mm,
 402                                        unsigned long address)
 403 {
 404         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 405         int young, idx;
 406
 407         idx = srcu_read_lock(&kvm->srcu);
 408         spin_lock(&kvm->mmu_lock);
 409         young = kvm_test_age_hva(kvm, address);
 410         spin_unlock(&kvm->mmu_lock);
 411         srcu_read_unlock(&kvm->srcu, idx);
 412
 413         return young;
 414 }
 415
 416 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 417                                      struct mm_struct *mm)
 418 {
 419         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 420         int idx;
 421
 422         idx = srcu_read_lock(&kvm->srcu);
 423         kvm_arch_flush_shadow_all(kvm);
 424         srcu_read_unlock(&kvm->srcu, idx);
 425 }
 426
 427 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 428         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 429         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 430         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 431         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 432         .test_young             = kvm_mmu_notifier_test_young,
 433         .change_pte             = kvm_mmu_notifier_change_pte,
 434         .release                = kvm_mmu_notifier_release,
 435 };
 436
 437 static int kvm_init_mmu_notifier(struct kvm *kvm)
 438 {
 439         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 440         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 441 }
 442
 443 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 444
 445 static int kvm_init_mmu_notifier(struct kvm *kvm)
 446 {
 447         return 0;
 448 }
 449
 450 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 451
 452 static void kvm_init_memslots_id(struct kvm *kvm)
 453 {
 454         int i;
 455         struct kvm_memslots *slots = kvm->memslots;
 456
 457         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 458                 slots->id_to_index[i] = slots->memslots[i].id = i;
 459 }
 460
 461 static struct kvm *kvm_create_vm(unsigned long type)
 462 {
 463         int r, i;
 464         struct kvm *kvm = kvm_arch_alloc_vm();
 465
 466         if (!kvm)
 467                 return ERR_PTR(-ENOMEM);
 468
 469         r = kvm_arch_init_vm(kvm, type);
 470         if (r)
 471                 goto out_err_nodisable;
 472
 473         r = hardware_enable_all();
 474         if (r)
 475                 goto out_err_nodisable;
 476
 477 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 478         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 479         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 480 #endif
 481
 482         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 483
 484         r = -ENOMEM;
 485         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 486         if (!kvm->memslots)
 487                 goto out_err_nosrcu;
 488         kvm_init_memslots_id(kvm);
 489         if (init_srcu_struct(&kvm->srcu))
 490                 goto out_err_nosrcu;
 491         for (i = 0; i < KVM_NR_BUSES; i++) {
 492                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 493                                         GFP_KERNEL);
 494                 if (!kvm->buses[i])
 495                         goto out_err;
 496         }
 497
 498         spin_lock_init(&kvm->mmu_lock);
 499         kvm->mm = current->mm;
 500         atomic_inc(&kvm->mm->mm_count);
 501         kvm_eventfd_init(kvm);
 502         mutex_init(&kvm->lock);
 503         mutex_init(&kvm->irq_lock);
 504         mutex_init(&kvm->slots_lock);
 505         atomic_set(&kvm->users_count, 1);
 506
 507         r = kvm_init_mmu_notifier(kvm);
 508         if (r)
 509                 goto out_err;
 510
 511         raw_spin_lock(&kvm_lock);
 512         list_add(&kvm->vm_list, &vm_list);
 513         raw_spin_unlock(&kvm_lock);
 514
 515         return kvm;
 516
 517 out_err:
 518         cleanup_srcu_struct(&kvm->srcu);
 519 out_err_nosrcu:
 520         hardware_disable_all();
 521 out_err_nodisable:
 522         for (i = 0; i < KVM_NR_BUSES; i++)
 523                 kfree(kvm->buses[i]);
 524         kfree(kvm->memslots);
 525         kvm_arch_free_vm(kvm);
 526         return ERR_PTR(r);
 527 }
 528
 529 /*
 530  * Avoid using vmalloc for a small buffer.
 531  * Should not be used when the size is statically known.
 532  */
 533 void *kvm_kvzalloc(unsigned long size)
 534 {
 535         if (size > PAGE_SIZE)
 536                 return vzalloc(size);
 537         else
 538                 return kzalloc(size, GFP_KERNEL);
 539 }
 540
 541 void kvm_kvfree(const void *addr)
 542 {
 543         if (is_vmalloc_addr(addr))
 544                 vfree(addr);
 545         else
 546                 kfree(addr);
 547 }
 548
 549 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 550 {
 551         if (!memslot->dirty_bitmap)
 552                 return;
 553
 554         kvm_kvfree(memslot->dirty_bitmap);
 555         memslot->dirty_bitmap = NULL;
 556 }
 557
 558 /*
 559  * Free any memory in @free but not in @dont.
 560  */
 561 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 562                                   struct kvm_memory_slot *dont)
 563 {
 564         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 565                 kvm_destroy_dirty_bitmap(free);
 566
 567         kvm_arch_free_memslot(free, dont);
 568
 569         free->npages = 0;
 570 }
 571
 572 void kvm_free_physmem(struct kvm *kvm)
 573 {
 574         struct kvm_memslots *slots = kvm->memslots;
 575         struct kvm_memory_slot *memslot;
 576
 577         kvm_for_each_memslot(memslot, slots)
 578                 kvm_free_physmem_slot(memslot, NULL);
 579
 580         kfree(kvm->memslots);
 581 }
 582
 583 static void kvm_destroy_vm(struct kvm *kvm)
 584 {
 585         int i;
 586         struct mm_struct *mm = kvm->mm;
 587
 588         kvm_arch_sync_events(kvm);
 589         raw_spin_lock(&kvm_lock);
 590         list_del(&kvm->vm_list);
 591         raw_spin_unlock(&kvm_lock);
 592         kvm_free_irq_routing(kvm);
 593         for (i = 0; i < KVM_NR_BUSES; i++)
 594                 kvm_io_bus_destroy(kvm->buses[i]);
 595         kvm_coalesced_mmio_free(kvm);
 596 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 597         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 598 #else
 599         kvm_arch_flush_shadow_all(kvm);
 600 #endif
 601         kvm_arch_destroy_vm(kvm);
 602         kvm_free_physmem(kvm);
 603         cleanup_srcu_struct(&kvm->srcu);
 604         kvm_arch_free_vm(kvm);
 605         hardware_disable_all();
 606         mmdrop(mm);
 607 }
 608
 609 void kvm_get_kvm(struct kvm *kvm)
 610 {
 611         atomic_inc(&kvm->users_count);
 612 }
 613 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 614
 615 void kvm_put_kvm(struct kvm *kvm)
 616 {
 617         if (atomic_dec_and_test(&kvm->users_count))
 618                 kvm_destroy_vm(kvm);
 619 }
 620 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 621
 622
 623 static int kvm_vm_release(struct inode *inode, struct file *filp)
 624 {
 625         struct kvm *kvm = filp->private_data;
 626
 627         kvm_irqfd_release(kvm);
 628
 629         kvm_put_kvm(kvm);
 630         return 0;
 631 }
 632
 633 /*
 634  * Allocation size is twice as large as the actual dirty bitmap size.
 635  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 636  */
 637 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 638 {
 639 #ifndef CONFIG_S390
 640         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 641
 642         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 643         if (!memslot->dirty_bitmap)
 644                 return -ENOMEM;
 645
 646 #endif /* !CONFIG_S390 */
 647         return 0;
 648 }
 649
 650 static int cmp_memslot(const void *slot1, const void *slot2)
 651 {
 652         struct kvm_memory_slot *s1, *s2;
 653
 654         s1 = (struct kvm_memory_slot *)slot1;
 655         s2 = (struct kvm_memory_slot *)slot2;
 656
 657         if (s1->npages < s2->npages)
 658                 return 1;
 659         if (s1->npages > s2->npages)
 660                 return -1;
 661
 662         return 0;
 663 }
 664
 665 /*
 666  * Sort the memslots base on its size, so the larger slots
 667  * will get better fit.
 668  */
 669 static void sort_memslots(struct kvm_memslots *slots)
 670 {
 671         int i;
 672
 673         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 674               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 675
 676         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 677                 slots->id_to_index[slots->memslots[i].id] = i;
 678 }
 679
 680 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
 681                      u64 last_generation)
 682 {
 683         if (new) {
 684                 int id = new->id;
 685                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 686                 unsigned long npages = old->npages;
 687
 688                 *old = *new;
 689                 if (new->npages != npages)
 690                         sort_memslots(slots);
 691         }
 692
 693         slots->generation = last_generation + 1;
 694 }
 695
 696 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 697 {
 698         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 699
 700 #ifdef KVM_CAP_READONLY_MEM
 701         valid_flags |= KVM_MEM_READONLY;
 702 #endif
 703
 704         if (mem->flags & ~valid_flags)
 705                 return -EINVAL;
 706
 707         return 0;
 708 }
 709
 710 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 711                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 712 {
 713         struct kvm_memslots *old_memslots = kvm->memslots;
 714
 715         update_memslots(slots, new, kvm->memslots->generation);
 716         rcu_assign_pointer(kvm->memslots, slots);
 717         synchronize_srcu_expedited(&kvm->srcu);
 718         return old_memslots;
 719 }
 720
 721 /*
 722  * Allocate some memory and give it an address in the guest physical address
 723  * space.
 724  *
 725  * Discontiguous memory is allowed, mostly for framebuffers.
 726  *
 727  * Must be called holding mmap_sem for write.
 728  */
 729 int __kvm_set_memory_region(struct kvm *kvm,
 730                             struct kvm_userspace_memory_region *mem,
 731                             bool user_alloc)
 732 {
 733         int r;
 734         gfn_t base_gfn;
 735         unsigned long npages;
 736         struct kvm_memory_slot *slot;
 737         struct kvm_memory_slot old, new;
 738         struct kvm_memslots *slots = NULL, *old_memslots;
 739         bool old_iommu_mapped;
 740
 741         r = check_memory_region_flags(mem);
 742         if (r)
 743                 goto out;
 744
 745         r = -EINVAL;
 746         /* General sanity checks */
 747         if (mem->memory_size & (PAGE_SIZE - 1))
 748                 goto out;
 749         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 750                 goto out;
 751         /* We can read the guest memory with __xxx_user() later on. */
 752         if (user_alloc &&
 753             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 754              !access_ok(VERIFY_WRITE,
 755                         (void __user *)(unsigned long)mem->userspace_addr,
 756                         mem->memory_size)))
 757                 goto out;
 758         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 759                 goto out;
 760         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 761                 goto out;
 762
 763         slot = id_to_memslot(kvm->memslots, mem->slot);
 764         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 765         npages = mem->memory_size >> PAGE_SHIFT;
 766
 767         r = -EINVAL;
 768         if (npages > KVM_MEM_MAX_NR_PAGES)
 769                 goto out;
 770
 771         if (!npages)
 772                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 773
 774         new = old = *slot;
 775
 776         new.id = mem->slot;
 777         new.base_gfn = base_gfn;
 778         new.npages = npages;
 779         new.flags = mem->flags;
 780
 781         old_iommu_mapped = old.npages;
 782
 783         /*
 784          * Disallow changing a memory slot's size or changing anything about
 785          * zero sized slots that doesn't involve making them non-zero.
 786          */
 787         r = -EINVAL;
 788         if (npages && old.npages && npages != old.npages)
 789                 goto out;
 790         if (!npages && !old.npages)
 791                 goto out;
 792
 793         if ((npages && !old.npages) || (base_gfn != old.base_gfn)) {
 794                 /* Check for overlaps */
 795                 r = -EEXIST;
 796                 kvm_for_each_memslot(slot, kvm->memslots) {
 797                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 798                             (slot->id == mem->slot))
 799                                 continue;
 800                         if (!((base_gfn + npages <= slot->base_gfn) ||
 801                               (base_gfn >= slot->base_gfn + slot->npages)))
 802                                 goto out;
 803                 }
 804         }
 805
 806         /* Free page dirty bitmap if unneeded */
 807         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 808                 new.dirty_bitmap = NULL;
 809
 810         r = -ENOMEM;
 811
 812         /*
 813          * Allocate if a slot is being created.  If modifying a slot,
 814          * the userspace_addr cannot change.
 815          */
 816         if (!old.npages) {
 817                 new.user_alloc = user_alloc;
 818                 new.userspace_addr = mem->userspace_addr;
 819
 820                 if (kvm_arch_create_memslot(&new, npages))
 821                         goto out_free;
 822         } else if (npages && mem->userspace_addr != old.userspace_addr) {
 823                 r = -EINVAL;
 824                 goto out_free;
 825         }
 826
 827         /* Allocate page dirty bitmap if needed */
 828         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 829                 if (kvm_create_dirty_bitmap(&new) < 0)
 830                         goto out_free;
 831         }
 832
 833         if (!npages || base_gfn != old.base_gfn) {
 834                 r = -ENOMEM;
 835                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 836                                 GFP_KERNEL);
 837                 if (!slots)
 838                         goto out_free;
 839                 slot = id_to_memslot(slots, mem->slot);
 840                 slot->flags |= KVM_MEMSLOT_INVALID;
 841
 842                 old_memslots = install_new_memslots(kvm, slots, NULL);
 843
 844                 /* slot was deleted or moved, clear iommu mapping */
 845                 kvm_iommu_unmap_pages(kvm, &old);
 846                 old_iommu_mapped = false;
 847                 /* From this point no new shadow pages pointing to a deleted,
 848                  * or moved, memslot will be created.
 849                  *
 850                  * validation of sp->gfn happens in:
 851                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 852                  *      - kvm_is_visible_gfn (mmu_check_roots)
 853                  */
 854                 kvm_arch_flush_shadow_memslot(kvm, slot);
 855                 slots = old_memslots;
 856         }
 857
 858         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 859         if (r)
 860                 goto out_slots;
 861
 862         r = -ENOMEM;
 863         /*
 864          * We can re-use the old_memslots from above, the only difference
 865          * from the currently installed memslots is the invalid flag.  This
 866          * will get overwritten by update_memslots anyway.
 867          */
 868         if (!slots) {
 869                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 870                                 GFP_KERNEL);
 871                 if (!slots)
 872                         goto out_free;
 873         }
 874
 875         /*
 876          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 877          * un-mapped and re-mapped if their base changes or if flags that the
 878          * iommu cares about change (read-only).  Base change unmapping is
 879          * handled above with slot deletion, so we only unmap incompatible
 880          * flags here.  Anything else the iommu might care about for existing
 881          * slots (size changes, userspace addr changes) is disallowed above,
 882          * so any other attribute changes getting here can be skipped.
 883          */
 884         if (npages) {
 885                 if (old_iommu_mapped &&
 886                     ((new.flags ^ old.flags) & KVM_MEM_READONLY)) {
 887                         kvm_iommu_unmap_pages(kvm, &old);
 888                         old_iommu_mapped = false;
 889                 }
 890
 891                 if (!old_iommu_mapped) {
 892                         r = kvm_iommu_map_pages(kvm, &new);
 893                         if (r)
 894                                 goto out_slots;
 895                 }
 896         }
 897
 898         /* actual memory is freed via old in kvm_free_physmem_slot below */
 899         if (!npages) {
 900                 new.dirty_bitmap = NULL;
 901                 memset(&new.arch, 0, sizeof(new.arch));
 902         }
 903
 904         old_memslots = install_new_memslots(kvm, slots, &new);
 905
 906         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 907
 908         kvm_free_physmem_slot(&old, &new);
 909         kfree(old_memslots);
 910
 911         return 0;
 912
 913 out_slots:
 914         kfree(slots);
 915 out_free:
 916         kvm_free_physmem_slot(&new, &old);
 917 out:
 918         return r;
 919 }
 920 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 921
 922 int kvm_set_memory_region(struct kvm *kvm,
 923                           struct kvm_userspace_memory_region *mem,
 924                           bool user_alloc)
 925 {
 926         int r;
 927
 928         mutex_lock(&kvm->slots_lock);
 929         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 930         mutex_unlock(&kvm->slots_lock);
 931         return r;
 932 }
 933 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 934
 935 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 936                                    struct
 937                                    kvm_userspace_memory_region *mem,
 938                                    bool user_alloc)
 939 {
 940         if (mem->slot >= KVM_USER_MEM_SLOTS)
 941                 return -EINVAL;
 942         return kvm_set_memory_region(kvm, mem, user_alloc);
 943 }
 944
 945 int kvm_get_dirty_log(struct kvm *kvm,
 946                         struct kvm_dirty_log *log, int *is_dirty)
 947 {
 948         struct kvm_memory_slot *memslot;
 949         int r, i;
 950         unsigned long n;
 951         unsigned long any = 0;
 952
 953         r = -EINVAL;
 954         if (log->slot >= KVM_USER_MEM_SLOTS)
 955                 goto out;
 956
 957         memslot = id_to_memslot(kvm->memslots, log->slot);
 958         r = -ENOENT;
 959         if (!memslot->dirty_bitmap)
 960                 goto out;
 961
 962         n = kvm_dirty_bitmap_bytes(memslot);
 963
 964         for (i = 0; !any && i < n/sizeof(long); ++i)
 965                 any = memslot->dirty_bitmap[i];
 966
 967         r = -EFAULT;
 968         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 969                 goto out;
 970
 971         if (any)
 972                 *is_dirty = 1;
 973
 974         r = 0;
 975 out:
 976         return r;
 977 }
 978
 979 bool kvm_largepages_enabled(void)
 980 {
 981         return largepages_enabled;
 982 }
 983
 984 void kvm_disable_largepages(void)
 985 {
 986         largepages_enabled = false;
 987 }
 988 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 989
 990 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 991 {
 992         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 993 }
 994 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 995
 996 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 997 {
 998         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 999
1000         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1001               memslot->flags & KVM_MEMSLOT_INVALID)
1002                 return 0;
1003
1004         return 1;
1005 }
1006 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1007
1008 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1009 {
1010         struct vm_area_struct *vma;
1011         unsigned long addr, size;
1012
1013         size = PAGE_SIZE;
1014
1015         addr = gfn_to_hva(kvm, gfn);
1016         if (kvm_is_error_hva(addr))
1017                 return PAGE_SIZE;
1018
1019         down_read(&current->mm->mmap_sem);
1020         vma = find_vma(current->mm, addr);
1021         if (!vma)
1022                 goto out;
1023
1024         size = vma_kernel_pagesize(vma);
1025
1026 out:
1027         up_read(&current->mm->mmap_sem);
1028
1029         return size;
1030 }
1031
1032 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1033 {
1034         return slot->flags & KVM_MEM_READONLY;
1035 }
1036
1037 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1038                                        gfn_t *nr_pages, bool write)
1039 {
1040         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1041                 return KVM_HVA_ERR_BAD;
1042
1043         if (memslot_is_readonly(slot) && write)
1044                 return KVM_HVA_ERR_RO_BAD;
1045
1046         if (nr_pages)
1047                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1048
1049         return __gfn_to_hva_memslot(slot, gfn);
1050 }
1051
1052 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1053                                      gfn_t *nr_pages)
1054 {
1055         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1056 }
1057
1058 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1059                                  gfn_t gfn)
1060 {
1061         return gfn_to_hva_many(slot, gfn, NULL);
1062 }
1063 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1064
1065 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1066 {
1067         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1068 }
1069 EXPORT_SYMBOL_GPL(gfn_to_hva);
1070
1071 /*
1072  * The hva returned by this function is only allowed to be read.
1073  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1074  */
1075 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1076 {
1077         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1078 }
1079
1080 static int kvm_read_hva(void *data, void __user *hva, int len)
1081 {
1082         return __copy_from_user(data, hva, len);
1083 }
1084
1085 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1086 {
1087         return __copy_from_user_inatomic(data, hva, len);
1088 }
1089
1090 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1091         unsigned long start, int write, struct page **page)
1092 {
1093         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1094
1095         if (write)
1096                 flags |= FOLL_WRITE;
1097
1098         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1099 }
1100
1101 static inline int check_user_page_hwpoison(unsigned long addr)
1102 {
1103         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1104
1105         rc = __get_user_pages(current, current->mm, addr, 1,
1106                               flags, NULL, NULL, NULL);
1107         return rc == -EHWPOISON;
1108 }
1109
1110 /*
1111  * The atomic path to get the writable pfn which will be stored in @pfn,
1112  * true indicates success, otherwise false is returned.
1113  */
1114 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1115                             bool write_fault, bool *writable, pfn_t *pfn)
1116 {
1117         struct page *page[1];
1118         int npages;
1119
1120         if (!(async || atomic))
1121                 return false;
1122
1123         /*
1124          * Fast pin a writable pfn only if it is a write fault request
1125          * or the caller allows to map a writable pfn for a read fault
1126          * request.
1127          */
1128         if (!(write_fault || writable))
1129                 return false;
1130
1131         npages = __get_user_pages_fast(addr, 1, 1, page);
1132         if (npages == 1) {
1133                 *pfn = page_to_pfn(page[0]);
1134
1135                 if (writable)
1136                         *writable = true;
1137                 return true;
1138         }
1139
1140         return false;
1141 }
1142
1143 /*
1144  * The slow path to get the pfn of the specified host virtual address,
1145  * 1 indicates success, -errno is returned if error is detected.
1146  */
1147 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1148                            bool *writable, pfn_t *pfn)
1149 {
1150         struct page *page[1];
1151         int npages = 0;
1152
1153         might_sleep();
1154
1155         if (writable)
1156                 *writable = write_fault;
1157
1158         if (async) {
1159                 down_read(&current->mm->mmap_sem);
1160                 npages = get_user_page_nowait(current, current->mm,
1161                                               addr, write_fault, page);
1162                 up_read(&current->mm->mmap_sem);
1163         } else
1164                 npages = get_user_pages_fast(addr, 1, write_fault,
1165                                              page);
1166         if (npages != 1)
1167                 return npages;
1168
1169         /* map read fault as writable if possible */
1170         if (unlikely(!write_fault) && writable) {
1171                 struct page *wpage[1];
1172
1173                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1174                 if (npages == 1) {
1175                         *writable = true;
1176                         put_page(page[0]);
1177                         page[0] = wpage[0];
1178                 }
1179
1180                 npages = 1;
1181         }
1182         *pfn = page_to_pfn(page[0]);
1183         return npages;
1184 }
1185
1186 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1187 {
1188         if (unlikely(!(vma->vm_flags & VM_READ)))
1189                 return false;
1190
1191         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1192                 return false;
1193
1194         return true;
1195 }
1196
1197 /*
1198  * Pin guest page in memory and return its pfn.
1199  * @addr: host virtual address which maps memory to the guest
1200  * @atomic: whether this function can sleep
1201  * @async: whether this function need to wait IO complete if the
1202  *         host page is not in the memory
1203  * @write_fault: whether we should get a writable host page
1204  * @writable: whether it allows to map a writable host page for !@write_fault
1205  *
1206  * The function will map a writable host page for these two cases:
1207  * 1): @write_fault = true
1208  * 2): @write_fault = false && @writable, @writable will tell the caller
1209  *     whether the mapping is writable.
1210  */
1211 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1212                         bool write_fault, bool *writable)
1213 {
1214         struct vm_area_struct *vma;
1215         pfn_t pfn = 0;
1216         int npages;
1217
1218         /* we can do it either atomically or asynchronously, not both */
1219         BUG_ON(atomic && async);
1220
1221         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1222                 return pfn;
1223
1224         if (atomic)
1225                 return KVM_PFN_ERR_FAULT;
1226
1227         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1228         if (npages == 1)
1229                 return pfn;
1230
1231         down_read(&current->mm->mmap_sem);
1232         if (npages == -EHWPOISON ||
1233               (!async && check_user_page_hwpoison(addr))) {
1234                 pfn = KVM_PFN_ERR_HWPOISON;
1235                 goto exit;
1236         }
1237
1238         vma = find_vma_intersection(current->mm, addr, addr + 1);
1239
1240         if (vma == NULL)
1241                 pfn = KVM_PFN_ERR_FAULT;
1242         else if ((vma->vm_flags & VM_PFNMAP)) {
1243                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1244                         vma->vm_pgoff;
1245                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1246         } else {
1247                 if (async && vma_is_valid(vma, write_fault))
1248                         *async = true;
1249                 pfn = KVM_PFN_ERR_FAULT;
1250         }
1251 exit:
1252         up_read(&current->mm->mmap_sem);
1253         return pfn;
1254 }
1255
1256 static pfn_t
1257 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1258                      bool *async, bool write_fault, bool *writable)
1259 {
1260         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1261
1262         if (addr == KVM_HVA_ERR_RO_BAD)
1263                 return KVM_PFN_ERR_RO_FAULT;
1264
1265         if (kvm_is_error_hva(addr))
1266                 return KVM_PFN_NOSLOT;
1267
1268         /* Do not map writable pfn in the readonly memslot. */
1269         if (writable && memslot_is_readonly(slot)) {
1270                 *writable = false;
1271                 writable = NULL;
1272         }
1273
1274         return hva_to_pfn(addr, atomic, async, write_fault,
1275                           writable);
1276 }
1277
1278 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1279                           bool write_fault, bool *writable)
1280 {
1281         struct kvm_memory_slot *slot;
1282
1283         if (async)
1284                 *async = false;
1285
1286         slot = gfn_to_memslot(kvm, gfn);
1287
1288         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1289                                     writable);
1290 }
1291
1292 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1293 {
1294         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1295 }
1296 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1297
1298 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1299                        bool write_fault, bool *writable)
1300 {
1301         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1302 }
1303 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1304
1305 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1306 {
1307         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1308 }
1309 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1310
1311 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1312                       bool *writable)
1313 {
1314         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1315 }
1316 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1317
1318 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1319 {
1320         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1321 }
1322
1323 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1324 {
1325         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1326 }
1327 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1328
1329 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1330                                                                   int nr_pages)
1331 {
1332         unsigned long addr;
1333         gfn_t entry;
1334
1335         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1336         if (kvm_is_error_hva(addr))
1337                 return -1;
1338
1339         if (entry < nr_pages)
1340                 return 0;
1341
1342         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1343 }
1344 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1345
1346 static struct page *kvm_pfn_to_page(pfn_t pfn)
1347 {
1348         if (is_error_noslot_pfn(pfn))
1349                 return KVM_ERR_PTR_BAD_PAGE;
1350
1351         if (kvm_is_mmio_pfn(pfn)) {
1352                 WARN_ON(1);
1353                 return KVM_ERR_PTR_BAD_PAGE;
1354         }
1355
1356         return pfn_to_page(pfn);
1357 }
1358
1359 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1360 {
1361         pfn_t pfn;
1362
1363         pfn = gfn_to_pfn(kvm, gfn);
1364
1365         return kvm_pfn_to_page(pfn);
1366 }
1367
1368 EXPORT_SYMBOL_GPL(gfn_to_page);
1369
1370 void kvm_release_page_clean(struct page *page)
1371 {
1372         WARN_ON(is_error_page(page));
1373
1374         kvm_release_pfn_clean(page_to_pfn(page));
1375 }
1376 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1377
1378 void kvm_release_pfn_clean(pfn_t pfn)
1379 {
1380         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1381                 put_page(pfn_to_page(pfn));
1382 }
1383 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1384
1385 void kvm_release_page_dirty(struct page *page)
1386 {
1387         WARN_ON(is_error_page(page));
1388
1389         kvm_release_pfn_dirty(page_to_pfn(page));
1390 }
1391 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1392
1393 void kvm_release_pfn_dirty(pfn_t pfn)
1394 {
1395         kvm_set_pfn_dirty(pfn);
1396         kvm_release_pfn_clean(pfn);
1397 }
1398 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1399
1400 void kvm_set_page_dirty(struct page *page)
1401 {
1402         kvm_set_pfn_dirty(page_to_pfn(page));
1403 }
1404 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1405
1406 void kvm_set_pfn_dirty(pfn_t pfn)
1407 {
1408         if (!kvm_is_mmio_pfn(pfn)) {
1409                 struct page *page = pfn_to_page(pfn);
1410                 if (!PageReserved(page))
1411                         SetPageDirty(page);
1412         }
1413 }
1414 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1415
1416 void kvm_set_pfn_accessed(pfn_t pfn)
1417 {
1418         if (!kvm_is_mmio_pfn(pfn))
1419                 mark_page_accessed(pfn_to_page(pfn));
1420 }
1421 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1422
1423 void kvm_get_pfn(pfn_t pfn)
1424 {
1425         if (!kvm_is_mmio_pfn(pfn))
1426                 get_page(pfn_to_page(pfn));
1427 }
1428 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1429
1430 static int next_segment(unsigned long len, int offset)
1431 {
1432         if (len > PAGE_SIZE - offset)
1433                 return PAGE_SIZE - offset;
1434         else
1435                 return len;
1436 }
1437
1438 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1439                         int len)
1440 {
1441         int r;
1442         unsigned long addr;
1443
1444         addr = gfn_to_hva_read(kvm, gfn);
1445         if (kvm_is_error_hva(addr))
1446                 return -EFAULT;
1447         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1448         if (r)
1449                 return -EFAULT;
1450         return 0;
1451 }
1452 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1453
1454 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1455 {
1456         gfn_t gfn = gpa >> PAGE_SHIFT;
1457         int seg;
1458         int offset = offset_in_page(gpa);
1459         int ret;
1460
1461         while ((seg = next_segment(len, offset)) != 0) {
1462                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1463                 if (ret < 0)
1464                         return ret;
1465                 offset = 0;
1466                 len -= seg;
1467                 data += seg;
1468                 ++gfn;
1469         }
1470         return 0;
1471 }
1472 EXPORT_SYMBOL_GPL(kvm_read_guest);
1473
1474 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1475                           unsigned long len)
1476 {
1477         int r;
1478         unsigned long addr;
1479         gfn_t gfn = gpa >> PAGE_SHIFT;
1480         int offset = offset_in_page(gpa);
1481
1482         addr = gfn_to_hva_read(kvm, gfn);
1483         if (kvm_is_error_hva(addr))
1484                 return -EFAULT;
1485         pagefault_disable();
1486         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1487         pagefault_enable();
1488         if (r)
1489                 return -EFAULT;
1490         return 0;
1491 }
1492 EXPORT_SYMBOL(kvm_read_guest_atomic);
1493
1494 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1495                          int offset, int len)
1496 {
1497         int r;
1498         unsigned long addr;
1499
1500         addr = gfn_to_hva(kvm, gfn);
1501         if (kvm_is_error_hva(addr))
1502                 return -EFAULT;
1503         r = __copy_to_user((void __user *)addr + offset, data, len);
1504         if (r)
1505                 return -EFAULT;
1506         mark_page_dirty(kvm, gfn);
1507         return 0;
1508 }
1509 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1510
1511 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1512                     unsigned long len)
1513 {
1514         gfn_t gfn = gpa >> PAGE_SHIFT;
1515         int seg;
1516         int offset = offset_in_page(gpa);
1517         int ret;
1518
1519         while ((seg = next_segment(len, offset)) != 0) {
1520                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1521                 if (ret < 0)
1522                         return ret;
1523                 offset = 0;
1524                 len -= seg;
1525                 data += seg;
1526                 ++gfn;
1527         }
1528         return 0;
1529 }
1530
1531 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1532                               gpa_t gpa)
1533 {
1534         struct kvm_memslots *slots = kvm_memslots(kvm);
1535         int offset = offset_in_page(gpa);
1536         gfn_t gfn = gpa >> PAGE_SHIFT;
1537
1538         ghc->gpa = gpa;
1539         ghc->generation = slots->generation;
1540         ghc->memslot = gfn_to_memslot(kvm, gfn);
1541         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1542         if (!kvm_is_error_hva(ghc->hva))
1543                 ghc->hva += offset;
1544         else
1545                 return -EFAULT;
1546
1547         return 0;
1548 }
1549 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1550
1551 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1552                            void *data, unsigned long len)
1553 {
1554         struct kvm_memslots *slots = kvm_memslots(kvm);
1555         int r;
1556
1557         if (slots->generation != ghc->generation)
1558                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1559
1560         if (kvm_is_error_hva(ghc->hva))
1561                 return -EFAULT;
1562
1563         r = __copy_to_user((void __user *)ghc->hva, data, len);
1564         if (r)
1565                 return -EFAULT;
1566         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1567
1568         return 0;
1569 }
1570 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1571
1572 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1573                            void *data, unsigned long len)
1574 {
1575         struct kvm_memslots *slots = kvm_memslots(kvm);
1576         int r;
1577
1578         if (slots->generation != ghc->generation)
1579                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1580
1581         if (kvm_is_error_hva(ghc->hva))
1582                 return -EFAULT;
1583
1584         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1585         if (r)
1586                 return -EFAULT;
1587
1588         return 0;
1589 }
1590 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1591
1592 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1593 {
1594         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1595                                     offset, len);
1596 }
1597 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1598
1599 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1600 {
1601         gfn_t gfn = gpa >> PAGE_SHIFT;
1602         int seg;
1603         int offset = offset_in_page(gpa);
1604         int ret;
1605
1606         while ((seg = next_segment(len, offset)) != 0) {
1607                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1608                 if (ret < 0)
1609                         return ret;
1610                 offset = 0;
1611                 len -= seg;
1612                 ++gfn;
1613         }
1614         return 0;
1615 }
1616 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1617
1618 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1619                              gfn_t gfn)
1620 {
1621         if (memslot && memslot->dirty_bitmap) {
1622                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1623
1624                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1625         }
1626 }
1627
1628 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1629 {
1630         struct kvm_memory_slot *memslot;
1631
1632         memslot = gfn_to_memslot(kvm, gfn);
1633         mark_page_dirty_in_slot(kvm, memslot, gfn);
1634 }
1635
1636 /*
1637  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1638  */
1639 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1640 {
1641         DEFINE_WAIT(wait);
1642
1643         for (;;) {
1644                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1645
1646                 if (kvm_arch_vcpu_runnable(vcpu)) {
1647                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1648                         break;
1649                 }
1650                 if (kvm_cpu_has_pending_timer(vcpu))
1651                         break;
1652                 if (signal_pending(current))
1653                         break;
1654
1655                 schedule();
1656         }
1657
1658         finish_wait(&vcpu->wq, &wait);
1659 }
1660
1661 #ifndef CONFIG_S390
1662 /*
1663  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1664  */
1665 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1666 {
1667         int me;
1668         int cpu = vcpu->cpu;
1669         wait_queue_head_t *wqp;
1670
1671         wqp = kvm_arch_vcpu_wq(vcpu);
1672         if (waitqueue_active(wqp)) {
1673                 wake_up_interruptible(wqp);
1674                 ++vcpu->stat.halt_wakeup;
1675         }
1676
1677         me = get_cpu();
1678         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1679                 if (kvm_arch_vcpu_should_kick(vcpu))
1680                         smp_send_reschedule(cpu);
1681         put_cpu();
1682 }
1683 #endif /* !CONFIG_S390 */
1684
1685 void kvm_resched(struct kvm_vcpu *vcpu)
1686 {
1687         if (!need_resched())
1688                 return;
1689         cond_resched();
1690 }
1691 EXPORT_SYMBOL_GPL(kvm_resched);
1692
1693 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1694 {
1695         struct pid *pid;
1696         struct task_struct *task = NULL;
1697
1698         rcu_read_lock();
1699         pid = rcu_dereference(target->pid);
1700         if (pid)
1701                 task = get_pid_task(target->pid, PIDTYPE_PID);
1702         rcu_read_unlock();
1703         if (!task)
1704                 return false;
1705         if (task->flags & PF_VCPU) {
1706                 put_task_struct(task);
1707                 return false;
1708         }
1709         if (yield_to(task, 1)) {
1710                 put_task_struct(task);
1711                 return true;
1712         }
1713         put_task_struct(task);
1714         return false;
1715 }
1716 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1717
1718 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1719 /*
1720  * Helper that checks whether a VCPU is eligible for directed yield.
1721  * Most eligible candidate to yield is decided by following heuristics:
1722  *
1723  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1724  *  (preempted lock holder), indicated by @in_spin_loop.
1725  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1726  *
1727  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1728  *  chance last time (mostly it has become eligible now since we have probably
1729  *  yielded to lockholder in last iteration. This is done by toggling
1730  *  @dy_eligible each time a VCPU checked for eligibility.)
1731  *
1732  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1733  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1734  *  burning. Giving priority for a potential lock-holder increases lock
1735  *  progress.
1736  *
1737  *  Since algorithm is based on heuristics, accessing another VCPU data without
1738  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1739  *  and continue with next VCPU and so on.
1740  */
1741 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1742 {
1743         bool eligible;
1744
1745         eligible = !vcpu->spin_loop.in_spin_loop ||
1746                         (vcpu->spin_loop.in_spin_loop &&
1747                          vcpu->spin_loop.dy_eligible);
1748
1749         if (vcpu->spin_loop.in_spin_loop)
1750                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1751
1752         return eligible;
1753 }
1754 #endif
1755 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1756 {
1757         struct kvm *kvm = me->kvm;
1758         struct kvm_vcpu *vcpu;
1759         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1760         int yielded = 0;
1761         int pass;
1762         int i;
1763
1764         kvm_vcpu_set_in_spin_loop(me, true);
1765         /*
1766          * We boost the priority of a VCPU that is runnable but not
1767          * currently running, because it got preempted by something
1768          * else and called schedule in __vcpu_run.  Hopefully that
1769          * VCPU is holding the lock that we need and will release it.
1770          * We approximate round-robin by starting at the last boosted VCPU.
1771          */
1772         for (pass = 0; pass < 2 && !yielded; pass++) {
1773                 kvm_for_each_vcpu(i, vcpu, kvm) {
1774                         if (!pass && i <= last_boosted_vcpu) {
1775                                 i = last_boosted_vcpu;
1776                                 continue;
1777                         } else if (pass && i > last_boosted_vcpu)
1778                                 break;
1779                         if (vcpu == me)
1780                                 continue;
1781                         if (waitqueue_active(&vcpu->wq))
1782                                 continue;
1783                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1784                                 continue;
1785                         if (kvm_vcpu_yield_to(vcpu)) {
1786                                 kvm->last_boosted_vcpu = i;
1787                                 yielded = 1;
1788                                 break;
1789                         }
1790                 }
1791         }
1792         kvm_vcpu_set_in_spin_loop(me, false);
1793
1794         /* Ensure vcpu is not eligible during next spinloop */
1795         kvm_vcpu_set_dy_eligible(me, false);
1796 }
1797 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1798
1799 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1800 {
1801         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1802         struct page *page;
1803
1804         if (vmf->pgoff == 0)
1805                 page = virt_to_page(vcpu->run);
1806 #ifdef CONFIG_X86
1807         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1808                 page = virt_to_page(vcpu->arch.pio_data);
1809 #endif
1810 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1811         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1812                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1813 #endif
1814         else
1815                 return kvm_arch_vcpu_fault(vcpu, vmf);
1816         get_page(page);
1817         vmf->page = page;
1818         return 0;
1819 }
1820
1821 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1822         .fault = kvm_vcpu_fault,
1823 };
1824
1825 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1826 {
1827         vma->vm_ops = &kvm_vcpu_vm_ops;
1828         return 0;
1829 }
1830
1831 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1832 {
1833         struct kvm_vcpu *vcpu = filp->private_data;
1834
1835         kvm_put_kvm(vcpu->kvm);
1836         return 0;
1837 }
1838
1839 static struct file_operations kvm_vcpu_fops = {
1840         .release        = kvm_vcpu_release,
1841         .unlocked_ioctl = kvm_vcpu_ioctl,
1842 #ifdef CONFIG_COMPAT
1843         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1844 #endif
1845         .mmap           = kvm_vcpu_mmap,
1846         .llseek         = noop_llseek,
1847 };
1848
1849 /*
1850  * Allocates an inode for the vcpu.
1851  */
1852 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1853 {
1854         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1855 }
1856
1857 /*
1858  * Creates some virtual cpus.  Good luck creating more than one.
1859  */
1860 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1861 {
1862         int r;
1863         struct kvm_vcpu *vcpu, *v;
1864
1865         vcpu = kvm_arch_vcpu_create(kvm, id);
1866         if (IS_ERR(vcpu))
1867                 return PTR_ERR(vcpu);
1868
1869         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1870
1871         r = kvm_arch_vcpu_setup(vcpu);
1872         if (r)
1873                 goto vcpu_destroy;
1874
1875         mutex_lock(&kvm->lock);
1876         if (!kvm_vcpu_compatible(vcpu)) {
1877                 r = -EINVAL;
1878                 goto unlock_vcpu_destroy;
1879         }
1880         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1881                 r = -EINVAL;
1882                 goto unlock_vcpu_destroy;
1883         }
1884
1885         kvm_for_each_vcpu(r, v, kvm)
1886                 if (v->vcpu_id == id) {
1887                         r = -EEXIST;
1888                         goto unlock_vcpu_destroy;
1889                 }
1890
1891         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1892
1893         /* Now it's all set up, let userspace reach it */
1894         kvm_get_kvm(kvm);
1895         r = create_vcpu_fd(vcpu);
1896         if (r < 0) {
1897                 kvm_put_kvm(kvm);
1898                 goto unlock_vcpu_destroy;
1899         }
1900
1901         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1902         smp_wmb();
1903         atomic_inc(&kvm->online_vcpus);
1904
1905         mutex_unlock(&kvm->lock);
1906         kvm_arch_vcpu_postcreate(vcpu);
1907         return r;
1908
1909 unlock_vcpu_destroy:
1910         mutex_unlock(&kvm->lock);
1911 vcpu_destroy:
1912         kvm_arch_vcpu_destroy(vcpu);
1913         return r;
1914 }
1915
1916 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1917 {
1918         if (sigset) {
1919                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1920                 vcpu->sigset_active = 1;
1921                 vcpu->sigset = *sigset;
1922         } else
1923                 vcpu->sigset_active = 0;
1924         return 0;
1925 }
1926
1927 static long kvm_vcpu_ioctl(struct file *filp,
1928                            unsigned int ioctl, unsigned long arg)
1929 {
1930         struct kvm_vcpu *vcpu = filp->private_data;
1931         void __user *argp = (void __user *)arg;
1932         int r;
1933         struct kvm_fpu *fpu = NULL;
1934         struct kvm_sregs *kvm_sregs = NULL;
1935
1936         if (vcpu->kvm->mm != current->mm)
1937                 return -EIO;
1938
1939 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1940         /*
1941          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1942          * so vcpu_load() would break it.
1943          */
1944         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1945                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1946 #endif
1947
1948
1949         r = vcpu_load(vcpu);
1950         if (r)
1951                 return r;
1952         switch (ioctl) {
1953         case KVM_RUN:
1954                 r = -EINVAL;
1955                 if (arg)
1956                         goto out;
1957                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1958                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1959                 break;
1960         case KVM_GET_REGS: {
1961                 struct kvm_regs *kvm_regs;
1962
1963                 r = -ENOMEM;
1964                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1965                 if (!kvm_regs)
1966                         goto out;
1967                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1968                 if (r)
1969                         goto out_free1;
1970                 r = -EFAULT;
1971                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1972                         goto out_free1;
1973                 r = 0;
1974 out_free1:
1975                 kfree(kvm_regs);
1976                 break;
1977         }
1978         case KVM_SET_REGS: {
1979                 struct kvm_regs *kvm_regs;
1980
1981                 r = -ENOMEM;
1982                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1983                 if (IS_ERR(kvm_regs)) {
1984                         r = PTR_ERR(kvm_regs);
1985                         goto out;
1986                 }
1987                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1988                 kfree(kvm_regs);
1989                 break;
1990         }
1991         case KVM_GET_SREGS: {
1992                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1993                 r = -ENOMEM;
1994                 if (!kvm_sregs)
1995                         goto out;
1996                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1997                 if (r)
1998                         goto out;
1999                 r = -EFAULT;
2000                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2001                         goto out;
2002                 r = 0;
2003                 break;
2004         }
2005         case KVM_SET_SREGS: {
2006                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2007                 if (IS_ERR(kvm_sregs)) {
2008                         r = PTR_ERR(kvm_sregs);
2009                         kvm_sregs = NULL;
2010                         goto out;
2011                 }
2012                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2013                 break;
2014         }
2015         case KVM_GET_MP_STATE: {
2016                 struct kvm_mp_state mp_state;
2017
2018                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2019                 if (r)
2020                         goto out;
2021                 r = -EFAULT;
2022                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2023                         goto out;
2024                 r = 0;
2025                 break;
2026         }
2027         case KVM_SET_MP_STATE: {
2028                 struct kvm_mp_state mp_state;
2029
2030                 r = -EFAULT;
2031                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2032                         goto out;
2033                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2034                 break;
2035         }
2036         case KVM_TRANSLATE: {
2037                 struct kvm_translation tr;
2038
2039                 r = -EFAULT;
2040                 if (copy_from_user(&tr, argp, sizeof tr))
2041                         goto out;
2042                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2043                 if (r)
2044                         goto out;
2045                 r = -EFAULT;
2046                 if (copy_to_user(argp, &tr, sizeof tr))
2047                         goto out;
2048                 r = 0;
2049                 break;
2050         }
2051         case KVM_SET_GUEST_DEBUG: {
2052                 struct kvm_guest_debug dbg;
2053
2054                 r = -EFAULT;
2055                 if (copy_from_user(&dbg, argp, sizeof dbg))
2056                         goto out;
2057                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2058                 break;
2059         }
2060         case KVM_SET_SIGNAL_MASK: {
2061                 struct kvm_signal_mask __user *sigmask_arg = argp;
2062                 struct kvm_signal_mask kvm_sigmask;
2063                 sigset_t sigset, *p;
2064
2065                 p = NULL;
2066                 if (argp) {
2067                         r = -EFAULT;
2068                         if (copy_from_user(&kvm_sigmask, argp,
2069                                            sizeof kvm_sigmask))
2070                                 goto out;
2071                         r = -EINVAL;
2072                         if (kvm_sigmask.len != sizeof sigset)
2073                                 goto out;
2074                         r = -EFAULT;
2075                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2076                                            sizeof sigset))
2077                                 goto out;
2078                         p = &sigset;
2079                 }
2080                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2081                 break;
2082         }
2083         case KVM_GET_FPU: {
2084                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2085                 r = -ENOMEM;
2086                 if (!fpu)
2087                         goto out;
2088                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2089                 if (r)
2090                         goto out;
2091                 r = -EFAULT;
2092                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2093                         goto out;
2094                 r = 0;
2095                 break;
2096         }
2097         case KVM_SET_FPU: {
2098                 fpu = memdup_user(argp, sizeof(*fpu));
2099                 if (IS_ERR(fpu)) {
2100                         r = PTR_ERR(fpu);
2101                         fpu = NULL;
2102                         goto out;
2103                 }
2104                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2105                 break;
2106         }
2107         default:
2108                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2109         }
2110 out:
2111         vcpu_put(vcpu);
2112         kfree(fpu);
2113         kfree(kvm_sregs);
2114         return r;
2115 }
2116
2117 #ifdef CONFIG_COMPAT
2118 static long kvm_vcpu_compat_ioctl(struct file *filp,
2119                                   unsigned int ioctl, unsigned long arg)
2120 {
2121         struct kvm_vcpu *vcpu = filp->private_data;
2122         void __user *argp = compat_ptr(arg);
2123         int r;
2124
2125         if (vcpu->kvm->mm != current->mm)
2126                 return -EIO;
2127
2128         switch (ioctl) {
2129         case KVM_SET_SIGNAL_MASK: {
2130                 struct kvm_signal_mask __user *sigmask_arg = argp;
2131                 struct kvm_signal_mask kvm_sigmask;
2132                 compat_sigset_t csigset;
2133                 sigset_t sigset;
2134
2135                 if (argp) {
2136                         r = -EFAULT;
2137                         if (copy_from_user(&kvm_sigmask, argp,
2138                                            sizeof kvm_sigmask))
2139                                 goto out;
2140                         r = -EINVAL;
2141                         if (kvm_sigmask.len != sizeof csigset)
2142                                 goto out;
2143                         r = -EFAULT;
2144                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2145                                            sizeof csigset))
2146                                 goto out;
2147                         sigset_from_compat(&sigset, &csigset);
2148                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2149                 } else
2150                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2151                 break;
2152         }
2153         default:
2154                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2155         }
2156
2157 out:
2158         return r;
2159 }
2160 #endif
2161
2162 static long kvm_vm_ioctl(struct file *filp,
2163                            unsigned int ioctl, unsigned long arg)
2164 {
2165         struct kvm *kvm = filp->private_data;
2166         void __user *argp = (void __user *)arg;
2167         int r;
2168
2169         if (kvm->mm != current->mm)
2170                 return -EIO;
2171         switch (ioctl) {
2172         case KVM_CREATE_VCPU:
2173                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2174                 break;
2175         case KVM_SET_USER_MEMORY_REGION: {
2176                 struct kvm_userspace_memory_region kvm_userspace_mem;
2177
2178                 r = -EFAULT;
2179                 if (copy_from_user(&kvm_userspace_mem, argp,
2180                                                 sizeof kvm_userspace_mem))
2181                         goto out;
2182
2183                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, true);
2184                 break;
2185         }
2186         case KVM_GET_DIRTY_LOG: {
2187                 struct kvm_dirty_log log;
2188
2189                 r = -EFAULT;
2190                 if (copy_from_user(&log, argp, sizeof log))
2191                         goto out;
2192                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2193                 break;
2194         }
2195 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2196         case KVM_REGISTER_COALESCED_MMIO: {
2197                 struct kvm_coalesced_mmio_zone zone;
2198                 r = -EFAULT;
2199                 if (copy_from_user(&zone, argp, sizeof zone))
2200                         goto out;
2201                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2202                 break;
2203         }
2204         case KVM_UNREGISTER_COALESCED_MMIO: {
2205                 struct kvm_coalesced_mmio_zone zone;
2206                 r = -EFAULT;
2207                 if (copy_from_user(&zone, argp, sizeof zone))
2208                         goto out;
2209                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2210                 break;
2211         }
2212 #endif
2213         case KVM_IRQFD: {
2214                 struct kvm_irqfd data;
2215
2216                 r = -EFAULT;
2217                 if (copy_from_user(&data, argp, sizeof data))
2218                         goto out;
2219                 r = kvm_irqfd(kvm, &data);
2220                 break;
2221         }
2222         case KVM_IOEVENTFD: {
2223                 struct kvm_ioeventfd data;
2224
2225                 r = -EFAULT;
2226                 if (copy_from_user(&data, argp, sizeof data))
2227                         goto out;
2228                 r = kvm_ioeventfd(kvm, &data);
2229                 break;
2230         }
2231 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2232         case KVM_SET_BOOT_CPU_ID:
2233                 r = 0;
2234                 mutex_lock(&kvm->lock);
2235                 if (atomic_read(&kvm->online_vcpus) != 0)
2236                         r = -EBUSY;
2237                 else
2238                         kvm->bsp_vcpu_id = arg;
2239                 mutex_unlock(&kvm->lock);
2240                 break;
2241 #endif
2242 #ifdef CONFIG_HAVE_KVM_MSI
2243         case KVM_SIGNAL_MSI: {
2244                 struct kvm_msi msi;
2245
2246                 r = -EFAULT;
2247                 if (copy_from_user(&msi, argp, sizeof msi))
2248                         goto out;
2249                 r = kvm_send_userspace_msi(kvm, &msi);
2250                 break;
2251         }
2252 #endif
2253 #ifdef __KVM_HAVE_IRQ_LINE
2254         case KVM_IRQ_LINE_STATUS:
2255         case KVM_IRQ_LINE: {
2256                 struct kvm_irq_level irq_event;
2257
2258                 r = -EFAULT;
2259                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2260                         goto out;
2261
2262                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2263                 if (r)
2264                         goto out;
2265
2266                 r = -EFAULT;
2267                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2268                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2269                                 goto out;
2270                 }
2271
2272                 r = 0;
2273                 break;
2274         }
2275 #endif
2276         default:
2277                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2278                 if (r == -ENOTTY)
2279                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2280         }
2281 out:
2282         return r;
2283 }
2284
2285 #ifdef CONFIG_COMPAT
2286 struct compat_kvm_dirty_log {
2287         __u32 slot;
2288         __u32 padding1;
2289         union {
2290                 compat_uptr_t dirty_bitmap; /* one bit per page */
2291                 __u64 padding2;
2292         };
2293 };
2294
2295 static long kvm_vm_compat_ioctl(struct file *filp,
2296                            unsigned int ioctl, unsigned long arg)
2297 {
2298         struct kvm *kvm = filp->private_data;
2299         int r;
2300
2301         if (kvm->mm != current->mm)
2302                 return -EIO;
2303         switch (ioctl) {
2304         case KVM_GET_DIRTY_LOG: {
2305                 struct compat_kvm_dirty_log compat_log;
2306                 struct kvm_dirty_log log;
2307
2308                 r = -EFAULT;
2309                 if (copy_from_user(&compat_log, (void __user *)arg,
2310                                    sizeof(compat_log)))
2311                         goto out;
2312                 log.slot         = compat_log.slot;
2313                 log.padding1     = compat_log.padding1;
2314                 log.padding2     = compat_log.padding2;
2315                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2316
2317                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2318                 break;
2319         }
2320         default:
2321                 r = kvm_vm_ioctl(filp, ioctl, arg);
2322         }
2323
2324 out:
2325         return r;
2326 }
2327 #endif
2328
2329 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2330 {
2331         struct page *page[1];
2332         unsigned long addr;
2333         int npages;
2334         gfn_t gfn = vmf->pgoff;
2335         struct kvm *kvm = vma->vm_file->private_data;
2336
2337         addr = gfn_to_hva(kvm, gfn);
2338         if (kvm_is_error_hva(addr))
2339                 return VM_FAULT_SIGBUS;
2340
2341         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2342                                 NULL);
2343         if (unlikely(npages != 1))
2344                 return VM_FAULT_SIGBUS;
2345
2346         vmf->page = page[0];
2347         return 0;
2348 }
2349
2350 static const struct vm_operations_struct kvm_vm_vm_ops = {
2351         .fault = kvm_vm_fault,
2352 };
2353
2354 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2355 {
2356         vma->vm_ops = &kvm_vm_vm_ops;
2357         return 0;
2358 }
2359
2360 static struct file_operations kvm_vm_fops = {
2361         .release        = kvm_vm_release,
2362         .unlocked_ioctl = kvm_vm_ioctl,
2363 #ifdef CONFIG_COMPAT
2364         .compat_ioctl   = kvm_vm_compat_ioctl,
2365 #endif
2366         .mmap           = kvm_vm_mmap,
2367         .llseek         = noop_llseek,
2368 };
2369
2370 static int kvm_dev_ioctl_create_vm(unsigned long type)
2371 {
2372         int r;
2373         struct kvm *kvm;
2374
2375         kvm = kvm_create_vm(type);
2376         if (IS_ERR(kvm))
2377                 return PTR_ERR(kvm);
2378 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2379         r = kvm_coalesced_mmio_init(kvm);
2380         if (r < 0) {
2381                 kvm_put_kvm(kvm);
2382                 return r;
2383         }
2384 #endif
2385         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2386         if (r < 0)
2387                 kvm_put_kvm(kvm);
2388
2389         return r;
2390 }
2391
2392 static long kvm_dev_ioctl_check_extension_generic(long arg)
2393 {
2394         switch (arg) {
2395         case KVM_CAP_USER_MEMORY:
2396         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2397         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2398 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2399         case KVM_CAP_SET_BOOT_CPU_ID:
2400 #endif
2401         case KVM_CAP_INTERNAL_ERROR_DATA:
2402 #ifdef CONFIG_HAVE_KVM_MSI
2403         case KVM_CAP_SIGNAL_MSI:
2404 #endif
2405                 return 1;
2406 #ifdef KVM_CAP_IRQ_ROUTING
2407         case KVM_CAP_IRQ_ROUTING:
2408                 return KVM_MAX_IRQ_ROUTES;
2409 #endif
2410         default:
2411                 break;
2412         }
2413         return kvm_dev_ioctl_check_extension(arg);
2414 }
2415
2416 static long kvm_dev_ioctl(struct file *filp,
2417                           unsigned int ioctl, unsigned long arg)
2418 {
2419         long r = -EINVAL;
2420
2421         switch (ioctl) {
2422         case KVM_GET_API_VERSION:
2423                 r = -EINVAL;
2424                 if (arg)
2425                         goto out;
2426                 r = KVM_API_VERSION;
2427                 break;
2428         case KVM_CREATE_VM:
2429                 r = kvm_dev_ioctl_create_vm(arg);
2430                 break;
2431         case KVM_CHECK_EXTENSION:
2432                 r = kvm_dev_ioctl_check_extension_generic(arg);
2433                 break;
2434         case KVM_GET_VCPU_MMAP_SIZE:
2435                 r = -EINVAL;
2436                 if (arg)
2437                         goto out;
2438                 r = PAGE_SIZE;     /* struct kvm_run */
2439 #ifdef CONFIG_X86
2440                 r += PAGE_SIZE;    /* pio data page */
2441 #endif
2442 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2443                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2444 #endif
2445                 break;
2446         case KVM_TRACE_ENABLE:
2447         case KVM_TRACE_PAUSE:
2448         case KVM_TRACE_DISABLE:
2449                 r = -EOPNOTSUPP;
2450                 break;
2451         default:
2452                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2453         }
2454 out:
2455         return r;
2456 }
2457
2458 static struct file_operations kvm_chardev_ops = {
2459         .unlocked_ioctl = kvm_dev_ioctl,
2460         .compat_ioctl   = kvm_dev_ioctl,
2461         .llseek         = noop_llseek,
2462 };
2463
2464 static struct miscdevice kvm_dev = {
2465         KVM_MINOR,
2466         "kvm",
2467         &kvm_chardev_ops,
2468 };
2469
2470 static void hardware_enable_nolock(void *junk)
2471 {
2472         int cpu = raw_smp_processor_id();
2473         int r;
2474
2475         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2476                 return;
2477
2478         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2479
2480         r = kvm_arch_hardware_enable(NULL);
2481
2482         if (r) {
2483                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2484                 atomic_inc(&hardware_enable_failed);
2485                 printk(KERN_INFO "kvm: enabling virtualization on "
2486                                  "CPU%d failed\n", cpu);
2487         }
2488 }
2489
2490 static void hardware_enable(void *junk)
2491 {
2492         raw_spin_lock(&kvm_lock);
2493         hardware_enable_nolock(junk);
2494         raw_spin_unlock(&kvm_lock);
2495 }
2496
2497 static void hardware_disable_nolock(void *junk)
2498 {
2499         int cpu = raw_smp_processor_id();
2500
2501         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2502                 return;
2503         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2504         kvm_arch_hardware_disable(NULL);
2505 }
2506
2507 static void hardware_disable(void *junk)
2508 {
2509         raw_spin_lock(&kvm_lock);
2510         hardware_disable_nolock(junk);
2511         raw_spin_unlock(&kvm_lock);
2512 }
2513
2514 static void hardware_disable_all_nolock(void)
2515 {
2516         BUG_ON(!kvm_usage_count);
2517
2518         kvm_usage_count--;
2519         if (!kvm_usage_count)
2520                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2521 }
2522
2523 static void hardware_disable_all(void)
2524 {
2525         raw_spin_lock(&kvm_lock);
2526         hardware_disable_all_nolock();
2527         raw_spin_unlock(&kvm_lock);
2528 }
2529
2530 static int hardware_enable_all(void)
2531 {
2532         int r = 0;
2533
2534         raw_spin_lock(&kvm_lock);
2535
2536         kvm_usage_count++;
2537         if (kvm_usage_count == 1) {
2538                 atomic_set(&hardware_enable_failed, 0);
2539                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2540
2541                 if (atomic_read(&hardware_enable_failed)) {
2542                         hardware_disable_all_nolock();
2543                         r = -EBUSY;
2544                 }
2545         }
2546
2547         raw_spin_unlock(&kvm_lock);
2548
2549         return r;
2550 }
2551
2552 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2553                            void *v)
2554 {
2555         int cpu = (long)v;
2556
2557         if (!kvm_usage_count)
2558                 return NOTIFY_OK;
2559
2560         val &= ~CPU_TASKS_FROZEN;
2561         switch (val) {
2562         case CPU_DYING:
2563                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2564                        cpu);
2565                 hardware_disable(NULL);
2566                 break;
2567         case CPU_STARTING:
2568                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2569                        cpu);
2570                 hardware_enable(NULL);
2571                 break;
2572         }
2573         return NOTIFY_OK;
2574 }
2575
2576
2577 asmlinkage void kvm_spurious_fault(void)
2578 {
2579         /* Fault while not rebooting.  We want the trace. */
2580         BUG();
2581 }
2582 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2583
2584 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2585                       void *v)
2586 {
2587         /*
2588          * Some (well, at least mine) BIOSes hang on reboot if
2589          * in vmx root mode.
2590          *
2591          * And Intel TXT required VMX off for all cpu when system shutdown.
2592          */
2593         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2594         kvm_rebooting = true;
2595         on_each_cpu(hardware_disable_nolock, NULL, 1);
2596         return NOTIFY_OK;
2597 }
2598
2599 static struct notifier_block kvm_reboot_notifier = {
2600         .notifier_call = kvm_reboot,
2601         .priority = 0,
2602 };
2603
2604 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2605 {
2606         int i;
2607
2608         for (i = 0; i < bus->dev_count; i++) {
2609                 struct kvm_io_device *pos = bus->range[i].dev;
2610
2611                 kvm_iodevice_destructor(pos);
2612         }
2613         kfree(bus);
2614 }
2615
2616 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2617 {
2618         const struct kvm_io_range *r1 = p1;
2619         const struct kvm_io_range *r2 = p2;
2620
2621         if (r1->addr < r2->addr)
2622                 return -1;
2623         if (r1->addr + r1->len > r2->addr + r2->len)
2624                 return 1;
2625         return 0;
2626 }
2627
2628 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2629                           gpa_t addr, int len)
2630 {
2631         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2632                 .addr = addr,
2633                 .len = len,
2634                 .dev = dev,
2635         };
2636
2637         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2638                 kvm_io_bus_sort_cmp, NULL);
2639
2640         return 0;
2641 }
2642
2643 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2644                              gpa_t addr, int len)
2645 {
2646         struct kvm_io_range *range, key;
2647         int off;
2648
2649         key = (struct kvm_io_range) {
2650                 .addr = addr,
2651                 .len = len,
2652         };
2653
2654         range = bsearch(&key, bus->range, bus->dev_count,
2655                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2656         if (range == NULL)
2657                 return -ENOENT;
2658
2659         off = range - bus->range;
2660
2661         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2662                 off--;
2663
2664         return off;
2665 }
2666
2667 /* kvm_io_bus_write - called under kvm->slots_lock */
2668 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2669                      int len, const void *val)
2670 {
2671         int idx;
2672         struct kvm_io_bus *bus;
2673         struct kvm_io_range range;
2674
2675         range = (struct kvm_io_range) {
2676                 .addr = addr,
2677                 .len = len,
2678         };
2679
2680         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2681         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2682         if (idx < 0)
2683                 return -EOPNOTSUPP;
2684
2685         while (idx < bus->dev_count &&
2686                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2687                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2688                         return 0;
2689                 idx++;
2690         }
2691
2692         return -EOPNOTSUPP;
2693 }
2694
2695 /* kvm_io_bus_read - called under kvm->slots_lock */
2696 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2697                     int len, void *val)
2698 {
2699         int idx;
2700         struct kvm_io_bus *bus;
2701         struct kvm_io_range range;
2702
2703         range = (struct kvm_io_range) {
2704                 .addr = addr,
2705                 .len = len,
2706         };
2707
2708         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2709         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2710         if (idx < 0)
2711                 return -EOPNOTSUPP;
2712
2713         while (idx < bus->dev_count &&
2714                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2715                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2716                         return 0;
2717                 idx++;
2718         }
2719
2720         return -EOPNOTSUPP;
2721 }
2722
2723 /* Caller must hold slots_lock. */
2724 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2725                             int len, struct kvm_io_device *dev)
2726 {
2727         struct kvm_io_bus *new_bus, *bus;
2728
2729         bus = kvm->buses[bus_idx];
2730         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2731                 return -ENOSPC;
2732
2733         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2734                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2735         if (!new_bus)
2736                 return -ENOMEM;
2737         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2738                sizeof(struct kvm_io_range)));
2739         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2740         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2741         synchronize_srcu_expedited(&kvm->srcu);
2742         kfree(bus);
2743
2744         return 0;
2745 }
2746
2747 /* Caller must hold slots_lock. */
2748 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2749                               struct kvm_io_device *dev)
2750 {
2751         int i, r;
2752         struct kvm_io_bus *new_bus, *bus;
2753
2754         bus = kvm->buses[bus_idx];
2755         r = -ENOENT;
2756         for (i = 0; i < bus->dev_count; i++)
2757                 if (bus->range[i].dev == dev) {
2758                         r = 0;
2759                         break;
2760                 }
2761
2762         if (r)
2763                 return r;
2764
2765         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2766                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2767         if (!new_bus)
2768                 return -ENOMEM;
2769
2770         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2771         new_bus->dev_count--;
2772         memcpy(new_bus->range + i, bus->range + i + 1,
2773                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2774
2775         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2776         synchronize_srcu_expedited(&kvm->srcu);
2777         kfree(bus);
2778         return r;
2779 }
2780
2781 static struct notifier_block kvm_cpu_notifier = {
2782         .notifier_call = kvm_cpu_hotplug,
2783 };
2784
2785 static int vm_stat_get(void *_offset, u64 *val)
2786 {
2787         unsigned offset = (long)_offset;
2788         struct kvm *kvm;
2789
2790         *val = 0;
2791         raw_spin_lock(&kvm_lock);
2792         list_for_each_entry(kvm, &vm_list, vm_list)
2793                 *val += *(u32 *)((void *)kvm + offset);
2794         raw_spin_unlock(&kvm_lock);
2795         return 0;
2796 }
2797
2798 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2799
2800 static int vcpu_stat_get(void *_offset, u64 *val)
2801 {
2802         unsigned offset = (long)_offset;
2803         struct kvm *kvm;
2804         struct kvm_vcpu *vcpu;
2805         int i;
2806
2807         *val = 0;
2808         raw_spin_lock(&kvm_lock);
2809         list_for_each_entry(kvm, &vm_list, vm_list)
2810                 kvm_for_each_vcpu(i, vcpu, kvm)
2811                         *val += *(u32 *)((void *)vcpu + offset);
2812
2813         raw_spin_unlock(&kvm_lock);
2814         return 0;
2815 }
2816
2817 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2818
2819 static const struct file_operations *stat_fops[] = {
2820         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2821         [KVM_STAT_VM]   = &vm_stat_fops,
2822 };
2823
2824 static int kvm_init_debug(void)
2825 {
2826         int r = -EFAULT;
2827         struct kvm_stats_debugfs_item *p;
2828
2829         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2830         if (kvm_debugfs_dir == NULL)
2831                 goto out;
2832
2833         for (p = debugfs_entries; p->name; ++p) {
2834                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2835                                                 (void *)(long)p->offset,
2836                                                 stat_fops[p->kind]);
2837                 if (p->dentry == NULL)
2838                         goto out_dir;
2839         }
2840
2841         return 0;
2842
2843 out_dir:
2844         debugfs_remove_recursive(kvm_debugfs_dir);
2845 out:
2846         return r;
2847 }
2848
2849 static void kvm_exit_debug(void)
2850 {
2851         struct kvm_stats_debugfs_item *p;
2852
2853         for (p = debugfs_entries; p->name; ++p)
2854                 debugfs_remove(p->dentry);
2855         debugfs_remove(kvm_debugfs_dir);
2856 }
2857
2858 static int kvm_suspend(void)
2859 {
2860         if (kvm_usage_count)
2861                 hardware_disable_nolock(NULL);
2862         return 0;
2863 }
2864
2865 static void kvm_resume(void)
2866 {
2867         if (kvm_usage_count) {
2868                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2869                 hardware_enable_nolock(NULL);
2870         }
2871 }
2872
2873 static struct syscore_ops kvm_syscore_ops = {
2874         .suspend = kvm_suspend,
2875         .resume = kvm_resume,
2876 };
2877
2878 static inline
2879 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2880 {
2881         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2882 }
2883
2884 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2885 {
2886         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2887
2888         kvm_arch_vcpu_load(vcpu, cpu);
2889 }
2890
2891 static void kvm_sched_out(struct preempt_notifier *pn,
2892                           struct task_struct *next)
2893 {
2894         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2895
2896         kvm_arch_vcpu_put(vcpu);
2897 }
2898
2899 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2900                   struct module *module)
2901 {
2902         int r;
2903         int cpu;
2904
2905         r = kvm_arch_init(opaque);
2906         if (r)
2907                 goto out_fail;
2908
2909         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2910                 r = -ENOMEM;
2911                 goto out_free_0;
2912         }
2913
2914         r = kvm_arch_hardware_setup();
2915         if (r < 0)
2916                 goto out_free_0a;
2917
2918         for_each_online_cpu(cpu) {
2919                 smp_call_function_single(cpu,
2920                                 kvm_arch_check_processor_compat,
2921                                 &r, 1);
2922                 if (r < 0)
2923                         goto out_free_1;
2924         }
2925
2926         r = register_cpu_notifier(&kvm_cpu_notifier);
2927         if (r)
2928                 goto out_free_2;
2929         register_reboot_notifier(&kvm_reboot_notifier);
2930
2931         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2932         if (!vcpu_align)
2933                 vcpu_align = __alignof__(struct kvm_vcpu);
2934         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2935                                            0, NULL);
2936         if (!kvm_vcpu_cache) {
2937                 r = -ENOMEM;
2938                 goto out_free_3;
2939         }
2940
2941         r = kvm_async_pf_init();
2942         if (r)
2943                 goto out_free;
2944
2945         kvm_chardev_ops.owner = module;
2946         kvm_vm_fops.owner = module;
2947         kvm_vcpu_fops.owner = module;
2948
2949         r = misc_register(&kvm_dev);
2950         if (r) {
2951                 printk(KERN_ERR "kvm: misc device register failed\n");
2952                 goto out_unreg;
2953         }
2954
2955         register_syscore_ops(&kvm_syscore_ops);
2956
2957         kvm_preempt_ops.sched_in = kvm_sched_in;
2958         kvm_preempt_ops.sched_out = kvm_sched_out;
2959
2960         r = kvm_init_debug();
2961         if (r) {
2962                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2963                 goto out_undebugfs;
2964         }
2965
2966         return 0;
2967
2968 out_undebugfs:
2969         unregister_syscore_ops(&kvm_syscore_ops);
2970 out_unreg:
2971         kvm_async_pf_deinit();
2972 out_free:
2973         kmem_cache_destroy(kvm_vcpu_cache);
2974 out_free_3:
2975         unregister_reboot_notifier(&kvm_reboot_notifier);
2976         unregister_cpu_notifier(&kvm_cpu_notifier);
2977 out_free_2:
2978 out_free_1:
2979         kvm_arch_hardware_unsetup();
2980 out_free_0a:
2981         free_cpumask_var(cpus_hardware_enabled);
2982 out_free_0:
2983         kvm_arch_exit();
2984 out_fail:
2985         return r;
2986 }
2987 EXPORT_SYMBOL_GPL(kvm_init);
2988
2989 void kvm_exit(void)
2990 {
2991         kvm_exit_debug();
2992         misc_deregister(&kvm_dev);
2993         kmem_cache_destroy(kvm_vcpu_cache);
2994         kvm_async_pf_deinit();
2995         unregister_syscore_ops(&kvm_syscore_ops);
2996         unregister_reboot_notifier(&kvm_reboot_notifier);
2997         unregister_cpu_notifier(&kvm_cpu_notifier);
2998         on_each_cpu(hardware_disable_nolock, NULL, 1);
2999         kvm_arch_hardware_unsetup();
3000         kvm_arch_exit();
3001         free_cpumask_var(cpus_hardware_enabled);
3002 }
3003 EXPORT_SYMBOL_GPL(kvm_exit);