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 {
 732         int r;
 733         gfn_t base_gfn;
 734         unsigned long npages;
 735         struct kvm_memory_slot *slot;
 736         struct kvm_memory_slot old, new;
 737         struct kvm_memslots *slots = NULL, *old_memslots;
 738         enum kvm_mr_change change;
 739
 740         r = check_memory_region_flags(mem);
 741         if (r)
 742                 goto out;
 743
 744         r = -EINVAL;
 745         /* General sanity checks */
 746         if (mem->memory_size & (PAGE_SIZE - 1))
 747                 goto out;
 748         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 749                 goto out;
 750         /* We can read the guest memory with __xxx_user() later on. */
 751         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 752             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 753              !access_ok(VERIFY_WRITE,
 754                         (void __user *)(unsigned long)mem->userspace_addr,
 755                         mem->memory_size)))
 756                 goto out;
 757         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 758                 goto out;
 759         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 760                 goto out;
 761
 762         slot = id_to_memslot(kvm->memslots, mem->slot);
 763         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 764         npages = mem->memory_size >> PAGE_SHIFT;
 765
 766         r = -EINVAL;
 767         if (npages > KVM_MEM_MAX_NR_PAGES)
 768                 goto out;
 769
 770         if (!npages)
 771                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 772
 773         new = old = *slot;
 774
 775         new.id = mem->slot;
 776         new.base_gfn = base_gfn;
 777         new.npages = npages;
 778         new.flags = mem->flags;
 779
 780         r = -EINVAL;
 781         if (npages) {
 782                 if (!old.npages)
 783                         change = KVM_MR_CREATE;
 784                 else { /* Modify an existing slot. */
 785                         if ((mem->userspace_addr != old.userspace_addr) ||
 786                             (npages != old.npages) ||
 787                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 788                                 goto out;
 789
 790                         if (base_gfn != old.base_gfn)
 791                                 change = KVM_MR_MOVE;
 792                         else if (new.flags != old.flags)
 793                                 change = KVM_MR_FLAGS_ONLY;
 794                         else { /* Nothing to change. */
 795                                 r = 0;
 796                                 goto out;
 797                         }
 798                 }
 799         } else if (old.npages) {
 800                 change = KVM_MR_DELETE;
 801         } else /* Modify a non-existent slot: disallowed. */
 802                 goto out;
 803
 804         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 805                 /* Check for overlaps */
 806                 r = -EEXIST;
 807                 kvm_for_each_memslot(slot, kvm->memslots) {
 808                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 809                             (slot->id == mem->slot))
 810                                 continue;
 811                         if (!((base_gfn + npages <= slot->base_gfn) ||
 812                               (base_gfn >= slot->base_gfn + slot->npages)))
 813                                 goto out;
 814                 }
 815         }
 816
 817         /* Free page dirty bitmap if unneeded */
 818         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 819                 new.dirty_bitmap = NULL;
 820
 821         r = -ENOMEM;
 822         if (change == KVM_MR_CREATE) {
 823                 new.userspace_addr = mem->userspace_addr;
 824
 825                 if (kvm_arch_create_memslot(&new, npages))
 826                         goto out_free;
 827         }
 828
 829         /* Allocate page dirty bitmap if needed */
 830         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 831                 if (kvm_create_dirty_bitmap(&new) < 0)
 832                         goto out_free;
 833         }
 834
 835         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 836                 r = -ENOMEM;
 837                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 838                                 GFP_KERNEL);
 839                 if (!slots)
 840                         goto out_free;
 841                 slot = id_to_memslot(slots, mem->slot);
 842                 slot->flags |= KVM_MEMSLOT_INVALID;
 843
 844                 old_memslots = install_new_memslots(kvm, slots, NULL);
 845
 846                 /* slot was deleted or moved, clear iommu mapping */
 847                 kvm_iommu_unmap_pages(kvm, &old);
 848                 /* From this point no new shadow pages pointing to a deleted,
 849                  * or moved, memslot will be created.
 850                  *
 851                  * validation of sp->gfn happens in:
 852                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 853                  *      - kvm_is_visible_gfn (mmu_check_roots)
 854                  */
 855                 kvm_arch_flush_shadow_memslot(kvm, slot);
 856                 slots = old_memslots;
 857         }
 858
 859         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 860         if (r)
 861                 goto out_slots;
 862
 863         r = -ENOMEM;
 864         /*
 865          * We can re-use the old_memslots from above, the only difference
 866          * from the currently installed memslots is the invalid flag.  This
 867          * will get overwritten by update_memslots anyway.
 868          */
 869         if (!slots) {
 870                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 871                                 GFP_KERNEL);
 872                 if (!slots)
 873                         goto out_free;
 874         }
 875
 876         /*
 877          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 878          * un-mapped and re-mapped if their base changes.  Since base change
 879          * unmapping is handled above with slot deletion, mapping alone is
 880          * needed here.  Anything else the iommu might care about for existing
 881          * slots (size changes, userspace addr changes and read-only flag
 882          * changes) is disallowed above, so any other attribute changes getting
 883          * here can be skipped.
 884          */
 885         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 886                 r = kvm_iommu_map_pages(kvm, &new);
 887                 if (r)
 888                         goto out_slots;
 889         }
 890
 891         /* actual memory is freed via old in kvm_free_physmem_slot below */
 892         if (change == KVM_MR_DELETE) {
 893                 new.dirty_bitmap = NULL;
 894                 memset(&new.arch, 0, sizeof(new.arch));
 895         }
 896
 897         old_memslots = install_new_memslots(kvm, slots, &new);
 898
 899         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 900
 901         kvm_free_physmem_slot(&old, &new);
 902         kfree(old_memslots);
 903
 904         return 0;
 905
 906 out_slots:
 907         kfree(slots);
 908 out_free:
 909         kvm_free_physmem_slot(&new, &old);
 910 out:
 911         return r;
 912 }
 913 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 914
 915 int kvm_set_memory_region(struct kvm *kvm,
 916                           struct kvm_userspace_memory_region *mem)
 917 {
 918         int r;
 919
 920         mutex_lock(&kvm->slots_lock);
 921         r = __kvm_set_memory_region(kvm, mem);
 922         mutex_unlock(&kvm->slots_lock);
 923         return r;
 924 }
 925 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 926
 927 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 928                                    struct kvm_userspace_memory_region *mem)
 929 {
 930         if (mem->slot >= KVM_USER_MEM_SLOTS)
 931                 return -EINVAL;
 932         return kvm_set_memory_region(kvm, mem);
 933 }
 934
 935 int kvm_get_dirty_log(struct kvm *kvm,
 936                         struct kvm_dirty_log *log, int *is_dirty)
 937 {
 938         struct kvm_memory_slot *memslot;
 939         int r, i;
 940         unsigned long n;
 941         unsigned long any = 0;
 942
 943         r = -EINVAL;
 944         if (log->slot >= KVM_USER_MEM_SLOTS)
 945                 goto out;
 946
 947         memslot = id_to_memslot(kvm->memslots, log->slot);
 948         r = -ENOENT;
 949         if (!memslot->dirty_bitmap)
 950                 goto out;
 951
 952         n = kvm_dirty_bitmap_bytes(memslot);
 953
 954         for (i = 0; !any && i < n/sizeof(long); ++i)
 955                 any = memslot->dirty_bitmap[i];
 956
 957         r = -EFAULT;
 958         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 959                 goto out;
 960
 961         if (any)
 962                 *is_dirty = 1;
 963
 964         r = 0;
 965 out:
 966         return r;
 967 }
 968
 969 bool kvm_largepages_enabled(void)
 970 {
 971         return largepages_enabled;
 972 }
 973
 974 void kvm_disable_largepages(void)
 975 {
 976         largepages_enabled = false;
 977 }
 978 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 979
 980 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 981 {
 982         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 983 }
 984 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 985
 986 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 987 {
 988         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 989
 990         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
 991               memslot->flags & KVM_MEMSLOT_INVALID)
 992                 return 0;
 993
 994         return 1;
 995 }
 996 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 997
 998 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 999 {
1000         struct vm_area_struct *vma;
1001         unsigned long addr, size;
1002
1003         size = PAGE_SIZE;
1004
1005         addr = gfn_to_hva(kvm, gfn);
1006         if (kvm_is_error_hva(addr))
1007                 return PAGE_SIZE;
1008
1009         down_read(&current->mm->mmap_sem);
1010         vma = find_vma(current->mm, addr);
1011         if (!vma)
1012                 goto out;
1013
1014         size = vma_kernel_pagesize(vma);
1015
1016 out:
1017         up_read(&current->mm->mmap_sem);
1018
1019         return size;
1020 }
1021
1022 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1023 {
1024         return slot->flags & KVM_MEM_READONLY;
1025 }
1026
1027 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1028                                        gfn_t *nr_pages, bool write)
1029 {
1030         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1031                 return KVM_HVA_ERR_BAD;
1032
1033         if (memslot_is_readonly(slot) && write)
1034                 return KVM_HVA_ERR_RO_BAD;
1035
1036         if (nr_pages)
1037                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1038
1039         return __gfn_to_hva_memslot(slot, gfn);
1040 }
1041
1042 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1043                                      gfn_t *nr_pages)
1044 {
1045         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1046 }
1047
1048 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1049                                  gfn_t gfn)
1050 {
1051         return gfn_to_hva_many(slot, gfn, NULL);
1052 }
1053 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1054
1055 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1056 {
1057         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1058 }
1059 EXPORT_SYMBOL_GPL(gfn_to_hva);
1060
1061 /*
1062  * The hva returned by this function is only allowed to be read.
1063  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1064  */
1065 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1066 {
1067         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1068 }
1069
1070 static int kvm_read_hva(void *data, void __user *hva, int len)
1071 {
1072         return __copy_from_user(data, hva, len);
1073 }
1074
1075 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1076 {
1077         return __copy_from_user_inatomic(data, hva, len);
1078 }
1079
1080 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1081         unsigned long start, int write, struct page **page)
1082 {
1083         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1084
1085         if (write)
1086                 flags |= FOLL_WRITE;
1087
1088         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1089 }
1090
1091 static inline int check_user_page_hwpoison(unsigned long addr)
1092 {
1093         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1094
1095         rc = __get_user_pages(current, current->mm, addr, 1,
1096                               flags, NULL, NULL, NULL);
1097         return rc == -EHWPOISON;
1098 }
1099
1100 /*
1101  * The atomic path to get the writable pfn which will be stored in @pfn,
1102  * true indicates success, otherwise false is returned.
1103  */
1104 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1105                             bool write_fault, bool *writable, pfn_t *pfn)
1106 {
1107         struct page *page[1];
1108         int npages;
1109
1110         if (!(async || atomic))
1111                 return false;
1112
1113         /*
1114          * Fast pin a writable pfn only if it is a write fault request
1115          * or the caller allows to map a writable pfn for a read fault
1116          * request.
1117          */
1118         if (!(write_fault || writable))
1119                 return false;
1120
1121         npages = __get_user_pages_fast(addr, 1, 1, page);
1122         if (npages == 1) {
1123                 *pfn = page_to_pfn(page[0]);
1124
1125                 if (writable)
1126                         *writable = true;
1127                 return true;
1128         }
1129
1130         return false;
1131 }
1132
1133 /*
1134  * The slow path to get the pfn of the specified host virtual address,
1135  * 1 indicates success, -errno is returned if error is detected.
1136  */
1137 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1138                            bool *writable, pfn_t *pfn)
1139 {
1140         struct page *page[1];
1141         int npages = 0;
1142
1143         might_sleep();
1144
1145         if (writable)
1146                 *writable = write_fault;
1147
1148         if (async) {
1149                 down_read(&current->mm->mmap_sem);
1150                 npages = get_user_page_nowait(current, current->mm,
1151                                               addr, write_fault, page);
1152                 up_read(&current->mm->mmap_sem);
1153         } else
1154                 npages = get_user_pages_fast(addr, 1, write_fault,
1155                                              page);
1156         if (npages != 1)
1157                 return npages;
1158
1159         /* map read fault as writable if possible */
1160         if (unlikely(!write_fault) && writable) {
1161                 struct page *wpage[1];
1162
1163                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1164                 if (npages == 1) {
1165                         *writable = true;
1166                         put_page(page[0]);
1167                         page[0] = wpage[0];
1168                 }
1169
1170                 npages = 1;
1171         }
1172         *pfn = page_to_pfn(page[0]);
1173         return npages;
1174 }
1175
1176 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1177 {
1178         if (unlikely(!(vma->vm_flags & VM_READ)))
1179                 return false;
1180
1181         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1182                 return false;
1183
1184         return true;
1185 }
1186
1187 /*
1188  * Pin guest page in memory and return its pfn.
1189  * @addr: host virtual address which maps memory to the guest
1190  * @atomic: whether this function can sleep
1191  * @async: whether this function need to wait IO complete if the
1192  *         host page is not in the memory
1193  * @write_fault: whether we should get a writable host page
1194  * @writable: whether it allows to map a writable host page for !@write_fault
1195  *
1196  * The function will map a writable host page for these two cases:
1197  * 1): @write_fault = true
1198  * 2): @write_fault = false && @writable, @writable will tell the caller
1199  *     whether the mapping is writable.
1200  */
1201 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1202                         bool write_fault, bool *writable)
1203 {
1204         struct vm_area_struct *vma;
1205         pfn_t pfn = 0;
1206         int npages;
1207
1208         /* we can do it either atomically or asynchronously, not both */
1209         BUG_ON(atomic && async);
1210
1211         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1212                 return pfn;
1213
1214         if (atomic)
1215                 return KVM_PFN_ERR_FAULT;
1216
1217         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1218         if (npages == 1)
1219                 return pfn;
1220
1221         down_read(&current->mm->mmap_sem);
1222         if (npages == -EHWPOISON ||
1223               (!async && check_user_page_hwpoison(addr))) {
1224                 pfn = KVM_PFN_ERR_HWPOISON;
1225                 goto exit;
1226         }
1227
1228         vma = find_vma_intersection(current->mm, addr, addr + 1);
1229
1230         if (vma == NULL)
1231                 pfn = KVM_PFN_ERR_FAULT;
1232         else if ((vma->vm_flags & VM_PFNMAP)) {
1233                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1234                         vma->vm_pgoff;
1235                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1236         } else {
1237                 if (async && vma_is_valid(vma, write_fault))
1238                         *async = true;
1239                 pfn = KVM_PFN_ERR_FAULT;
1240         }
1241 exit:
1242         up_read(&current->mm->mmap_sem);
1243         return pfn;
1244 }
1245
1246 static pfn_t
1247 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1248                      bool *async, bool write_fault, bool *writable)
1249 {
1250         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1251
1252         if (addr == KVM_HVA_ERR_RO_BAD)
1253                 return KVM_PFN_ERR_RO_FAULT;
1254
1255         if (kvm_is_error_hva(addr))
1256                 return KVM_PFN_NOSLOT;
1257
1258         /* Do not map writable pfn in the readonly memslot. */
1259         if (writable && memslot_is_readonly(slot)) {
1260                 *writable = false;
1261                 writable = NULL;
1262         }
1263
1264         return hva_to_pfn(addr, atomic, async, write_fault,
1265                           writable);
1266 }
1267
1268 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1269                           bool write_fault, bool *writable)
1270 {
1271         struct kvm_memory_slot *slot;
1272
1273         if (async)
1274                 *async = false;
1275
1276         slot = gfn_to_memslot(kvm, gfn);
1277
1278         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1279                                     writable);
1280 }
1281
1282 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1283 {
1284         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1285 }
1286 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1287
1288 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1289                        bool write_fault, bool *writable)
1290 {
1291         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1292 }
1293 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1294
1295 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1296 {
1297         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1298 }
1299 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1300
1301 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1302                       bool *writable)
1303 {
1304         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1305 }
1306 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1307
1308 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1309 {
1310         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1311 }
1312
1313 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1314 {
1315         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1316 }
1317 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1318
1319 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1320                                                                   int nr_pages)
1321 {
1322         unsigned long addr;
1323         gfn_t entry;
1324
1325         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1326         if (kvm_is_error_hva(addr))
1327                 return -1;
1328
1329         if (entry < nr_pages)
1330                 return 0;
1331
1332         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1333 }
1334 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1335
1336 static struct page *kvm_pfn_to_page(pfn_t pfn)
1337 {
1338         if (is_error_noslot_pfn(pfn))
1339                 return KVM_ERR_PTR_BAD_PAGE;
1340
1341         if (kvm_is_mmio_pfn(pfn)) {
1342                 WARN_ON(1);
1343                 return KVM_ERR_PTR_BAD_PAGE;
1344         }
1345
1346         return pfn_to_page(pfn);
1347 }
1348
1349 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1350 {
1351         pfn_t pfn;
1352
1353         pfn = gfn_to_pfn(kvm, gfn);
1354
1355         return kvm_pfn_to_page(pfn);
1356 }
1357
1358 EXPORT_SYMBOL_GPL(gfn_to_page);
1359
1360 void kvm_release_page_clean(struct page *page)
1361 {
1362         WARN_ON(is_error_page(page));
1363
1364         kvm_release_pfn_clean(page_to_pfn(page));
1365 }
1366 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1367
1368 void kvm_release_pfn_clean(pfn_t pfn)
1369 {
1370         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1371                 put_page(pfn_to_page(pfn));
1372 }
1373 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1374
1375 void kvm_release_page_dirty(struct page *page)
1376 {
1377         WARN_ON(is_error_page(page));
1378
1379         kvm_release_pfn_dirty(page_to_pfn(page));
1380 }
1381 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1382
1383 void kvm_release_pfn_dirty(pfn_t pfn)
1384 {
1385         kvm_set_pfn_dirty(pfn);
1386         kvm_release_pfn_clean(pfn);
1387 }
1388 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1389
1390 void kvm_set_page_dirty(struct page *page)
1391 {
1392         kvm_set_pfn_dirty(page_to_pfn(page));
1393 }
1394 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1395
1396 void kvm_set_pfn_dirty(pfn_t pfn)
1397 {
1398         if (!kvm_is_mmio_pfn(pfn)) {
1399                 struct page *page = pfn_to_page(pfn);
1400                 if (!PageReserved(page))
1401                         SetPageDirty(page);
1402         }
1403 }
1404 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1405
1406 void kvm_set_pfn_accessed(pfn_t pfn)
1407 {
1408         if (!kvm_is_mmio_pfn(pfn))
1409                 mark_page_accessed(pfn_to_page(pfn));
1410 }
1411 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1412
1413 void kvm_get_pfn(pfn_t pfn)
1414 {
1415         if (!kvm_is_mmio_pfn(pfn))
1416                 get_page(pfn_to_page(pfn));
1417 }
1418 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1419
1420 static int next_segment(unsigned long len, int offset)
1421 {
1422         if (len > PAGE_SIZE - offset)
1423                 return PAGE_SIZE - offset;
1424         else
1425                 return len;
1426 }
1427
1428 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1429                         int len)
1430 {
1431         int r;
1432         unsigned long addr;
1433
1434         addr = gfn_to_hva_read(kvm, gfn);
1435         if (kvm_is_error_hva(addr))
1436                 return -EFAULT;
1437         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1438         if (r)
1439                 return -EFAULT;
1440         return 0;
1441 }
1442 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1443
1444 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1445 {
1446         gfn_t gfn = gpa >> PAGE_SHIFT;
1447         int seg;
1448         int offset = offset_in_page(gpa);
1449         int ret;
1450
1451         while ((seg = next_segment(len, offset)) != 0) {
1452                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1453                 if (ret < 0)
1454                         return ret;
1455                 offset = 0;
1456                 len -= seg;
1457                 data += seg;
1458                 ++gfn;
1459         }
1460         return 0;
1461 }
1462 EXPORT_SYMBOL_GPL(kvm_read_guest);
1463
1464 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1465                           unsigned long len)
1466 {
1467         int r;
1468         unsigned long addr;
1469         gfn_t gfn = gpa >> PAGE_SHIFT;
1470         int offset = offset_in_page(gpa);
1471
1472         addr = gfn_to_hva_read(kvm, gfn);
1473         if (kvm_is_error_hva(addr))
1474                 return -EFAULT;
1475         pagefault_disable();
1476         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1477         pagefault_enable();
1478         if (r)
1479                 return -EFAULT;
1480         return 0;
1481 }
1482 EXPORT_SYMBOL(kvm_read_guest_atomic);
1483
1484 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1485                          int offset, int len)
1486 {
1487         int r;
1488         unsigned long addr;
1489
1490         addr = gfn_to_hva(kvm, gfn);
1491         if (kvm_is_error_hva(addr))
1492                 return -EFAULT;
1493         r = __copy_to_user((void __user *)addr + offset, data, len);
1494         if (r)
1495                 return -EFAULT;
1496         mark_page_dirty(kvm, gfn);
1497         return 0;
1498 }
1499 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1500
1501 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1502                     unsigned long len)
1503 {
1504         gfn_t gfn = gpa >> PAGE_SHIFT;
1505         int seg;
1506         int offset = offset_in_page(gpa);
1507         int ret;
1508
1509         while ((seg = next_segment(len, offset)) != 0) {
1510                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1511                 if (ret < 0)
1512                         return ret;
1513                 offset = 0;
1514                 len -= seg;
1515                 data += seg;
1516                 ++gfn;
1517         }
1518         return 0;
1519 }
1520
1521 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1522                               gpa_t gpa)
1523 {
1524         struct kvm_memslots *slots = kvm_memslots(kvm);
1525         int offset = offset_in_page(gpa);
1526         gfn_t gfn = gpa >> PAGE_SHIFT;
1527
1528         ghc->gpa = gpa;
1529         ghc->generation = slots->generation;
1530         ghc->memslot = gfn_to_memslot(kvm, gfn);
1531         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1532         if (!kvm_is_error_hva(ghc->hva))
1533                 ghc->hva += offset;
1534         else
1535                 return -EFAULT;
1536
1537         return 0;
1538 }
1539 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1540
1541 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1542                            void *data, unsigned long len)
1543 {
1544         struct kvm_memslots *slots = kvm_memslots(kvm);
1545         int r;
1546
1547         if (slots->generation != ghc->generation)
1548                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1549
1550         if (kvm_is_error_hva(ghc->hva))
1551                 return -EFAULT;
1552
1553         r = __copy_to_user((void __user *)ghc->hva, data, len);
1554         if (r)
1555                 return -EFAULT;
1556         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1557
1558         return 0;
1559 }
1560 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1561
1562 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1563                            void *data, unsigned long len)
1564 {
1565         struct kvm_memslots *slots = kvm_memslots(kvm);
1566         int r;
1567
1568         if (slots->generation != ghc->generation)
1569                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1570
1571         if (kvm_is_error_hva(ghc->hva))
1572                 return -EFAULT;
1573
1574         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1575         if (r)
1576                 return -EFAULT;
1577
1578         return 0;
1579 }
1580 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1581
1582 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1583 {
1584         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1585                                     offset, len);
1586 }
1587 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1588
1589 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1590 {
1591         gfn_t gfn = gpa >> PAGE_SHIFT;
1592         int seg;
1593         int offset = offset_in_page(gpa);
1594         int ret;
1595
1596         while ((seg = next_segment(len, offset)) != 0) {
1597                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1598                 if (ret < 0)
1599                         return ret;
1600                 offset = 0;
1601                 len -= seg;
1602                 ++gfn;
1603         }
1604         return 0;
1605 }
1606 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1607
1608 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1609                              gfn_t gfn)
1610 {
1611         if (memslot && memslot->dirty_bitmap) {
1612                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1613
1614                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1615         }
1616 }
1617
1618 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1619 {
1620         struct kvm_memory_slot *memslot;
1621
1622         memslot = gfn_to_memslot(kvm, gfn);
1623         mark_page_dirty_in_slot(kvm, memslot, gfn);
1624 }
1625
1626 /*
1627  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1628  */
1629 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1630 {
1631         DEFINE_WAIT(wait);
1632
1633         for (;;) {
1634                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1635
1636                 if (kvm_arch_vcpu_runnable(vcpu)) {
1637                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1638                         break;
1639                 }
1640                 if (kvm_cpu_has_pending_timer(vcpu))
1641                         break;
1642                 if (signal_pending(current))
1643                         break;
1644
1645                 schedule();
1646         }
1647
1648         finish_wait(&vcpu->wq, &wait);
1649 }
1650
1651 #ifndef CONFIG_S390
1652 /*
1653  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1654  */
1655 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1656 {
1657         int me;
1658         int cpu = vcpu->cpu;
1659         wait_queue_head_t *wqp;
1660
1661         wqp = kvm_arch_vcpu_wq(vcpu);
1662         if (waitqueue_active(wqp)) {
1663                 wake_up_interruptible(wqp);
1664                 ++vcpu->stat.halt_wakeup;
1665         }
1666
1667         me = get_cpu();
1668         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1669                 if (kvm_arch_vcpu_should_kick(vcpu))
1670                         smp_send_reschedule(cpu);
1671         put_cpu();
1672 }
1673 #endif /* !CONFIG_S390 */
1674
1675 void kvm_resched(struct kvm_vcpu *vcpu)
1676 {
1677         if (!need_resched())
1678                 return;
1679         cond_resched();
1680 }
1681 EXPORT_SYMBOL_GPL(kvm_resched);
1682
1683 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1684 {
1685         struct pid *pid;
1686         struct task_struct *task = NULL;
1687         bool ret = false;
1688
1689         rcu_read_lock();
1690         pid = rcu_dereference(target->pid);
1691         if (pid)
1692                 task = get_pid_task(target->pid, PIDTYPE_PID);
1693         rcu_read_unlock();
1694         if (!task)
1695                 return ret;
1696         if (task->flags & PF_VCPU) {
1697                 put_task_struct(task);
1698                 return ret;
1699         }
1700         ret = yield_to(task, 1);
1701         put_task_struct(task);
1702
1703         return ret;
1704 }
1705 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1706
1707 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1708 /*
1709  * Helper that checks whether a VCPU is eligible for directed yield.
1710  * Most eligible candidate to yield is decided by following heuristics:
1711  *
1712  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1713  *  (preempted lock holder), indicated by @in_spin_loop.
1714  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1715  *
1716  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1717  *  chance last time (mostly it has become eligible now since we have probably
1718  *  yielded to lockholder in last iteration. This is done by toggling
1719  *  @dy_eligible each time a VCPU checked for eligibility.)
1720  *
1721  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1722  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1723  *  burning. Giving priority for a potential lock-holder increases lock
1724  *  progress.
1725  *
1726  *  Since algorithm is based on heuristics, accessing another VCPU data without
1727  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1728  *  and continue with next VCPU and so on.
1729  */
1730 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1731 {
1732         bool eligible;
1733
1734         eligible = !vcpu->spin_loop.in_spin_loop ||
1735                         (vcpu->spin_loop.in_spin_loop &&
1736                          vcpu->spin_loop.dy_eligible);
1737
1738         if (vcpu->spin_loop.in_spin_loop)
1739                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1740
1741         return eligible;
1742 }
1743 #endif
1744
1745 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1746 {
1747         struct kvm *kvm = me->kvm;
1748         struct kvm_vcpu *vcpu;
1749         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1750         int yielded = 0;
1751         int try = 3;
1752         int pass;
1753         int i;
1754
1755         kvm_vcpu_set_in_spin_loop(me, true);
1756         /*
1757          * We boost the priority of a VCPU that is runnable but not
1758          * currently running, because it got preempted by something
1759          * else and called schedule in __vcpu_run.  Hopefully that
1760          * VCPU is holding the lock that we need and will release it.
1761          * We approximate round-robin by starting at the last boosted VCPU.
1762          */
1763         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1764                 kvm_for_each_vcpu(i, vcpu, kvm) {
1765                         if (!pass && i <= last_boosted_vcpu) {
1766                                 i = last_boosted_vcpu;
1767                                 continue;
1768                         } else if (pass && i > last_boosted_vcpu)
1769                                 break;
1770                         if (vcpu == me)
1771                                 continue;
1772                         if (waitqueue_active(&vcpu->wq))
1773                                 continue;
1774                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1775                                 continue;
1776
1777                         yielded = kvm_vcpu_yield_to(vcpu);
1778                         if (yielded > 0) {
1779                                 kvm->last_boosted_vcpu = i;
1780                                 break;
1781                         } else if (yielded < 0) {
1782                                 try--;
1783                                 if (!try)
1784                                         break;
1785                         }
1786                 }
1787         }
1788         kvm_vcpu_set_in_spin_loop(me, false);
1789
1790         /* Ensure vcpu is not eligible during next spinloop */
1791         kvm_vcpu_set_dy_eligible(me, false);
1792 }
1793 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1794
1795 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1796 {
1797         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1798         struct page *page;
1799
1800         if (vmf->pgoff == 0)
1801                 page = virt_to_page(vcpu->run);
1802 #ifdef CONFIG_X86
1803         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1804                 page = virt_to_page(vcpu->arch.pio_data);
1805 #endif
1806 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1807         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1808                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1809 #endif
1810         else
1811                 return kvm_arch_vcpu_fault(vcpu, vmf);
1812         get_page(page);
1813         vmf->page = page;
1814         return 0;
1815 }
1816
1817 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1818         .fault = kvm_vcpu_fault,
1819 };
1820
1821 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1822 {
1823         vma->vm_ops = &kvm_vcpu_vm_ops;
1824         return 0;
1825 }
1826
1827 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1828 {
1829         struct kvm_vcpu *vcpu = filp->private_data;
1830
1831         kvm_put_kvm(vcpu->kvm);
1832         return 0;
1833 }
1834
1835 static struct file_operations kvm_vcpu_fops = {
1836         .release        = kvm_vcpu_release,
1837         .unlocked_ioctl = kvm_vcpu_ioctl,
1838 #ifdef CONFIG_COMPAT
1839         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1840 #endif
1841         .mmap           = kvm_vcpu_mmap,
1842         .llseek         = noop_llseek,
1843 };
1844
1845 /*
1846  * Allocates an inode for the vcpu.
1847  */
1848 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1849 {
1850         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1851 }
1852
1853 /*
1854  * Creates some virtual cpus.  Good luck creating more than one.
1855  */
1856 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1857 {
1858         int r;
1859         struct kvm_vcpu *vcpu, *v;
1860
1861         vcpu = kvm_arch_vcpu_create(kvm, id);
1862         if (IS_ERR(vcpu))
1863                 return PTR_ERR(vcpu);
1864
1865         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1866
1867         r = kvm_arch_vcpu_setup(vcpu);
1868         if (r)
1869                 goto vcpu_destroy;
1870
1871         mutex_lock(&kvm->lock);
1872         if (!kvm_vcpu_compatible(vcpu)) {
1873                 r = -EINVAL;
1874                 goto unlock_vcpu_destroy;
1875         }
1876         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1877                 r = -EINVAL;
1878                 goto unlock_vcpu_destroy;
1879         }
1880
1881         kvm_for_each_vcpu(r, v, kvm)
1882                 if (v->vcpu_id == id) {
1883                         r = -EEXIST;
1884                         goto unlock_vcpu_destroy;
1885                 }
1886
1887         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1888
1889         /* Now it's all set up, let userspace reach it */
1890         kvm_get_kvm(kvm);
1891         r = create_vcpu_fd(vcpu);
1892         if (r < 0) {
1893                 kvm_put_kvm(kvm);
1894                 goto unlock_vcpu_destroy;
1895         }
1896
1897         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1898         smp_wmb();
1899         atomic_inc(&kvm->online_vcpus);
1900
1901         mutex_unlock(&kvm->lock);
1902         kvm_arch_vcpu_postcreate(vcpu);
1903         return r;
1904
1905 unlock_vcpu_destroy:
1906         mutex_unlock(&kvm->lock);
1907 vcpu_destroy:
1908         kvm_arch_vcpu_destroy(vcpu);
1909         return r;
1910 }
1911
1912 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1913 {
1914         if (sigset) {
1915                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1916                 vcpu->sigset_active = 1;
1917                 vcpu->sigset = *sigset;
1918         } else
1919                 vcpu->sigset_active = 0;
1920         return 0;
1921 }
1922
1923 static long kvm_vcpu_ioctl(struct file *filp,
1924                            unsigned int ioctl, unsigned long arg)
1925 {
1926         struct kvm_vcpu *vcpu = filp->private_data;
1927         void __user *argp = (void __user *)arg;
1928         int r;
1929         struct kvm_fpu *fpu = NULL;
1930         struct kvm_sregs *kvm_sregs = NULL;
1931
1932         if (vcpu->kvm->mm != current->mm)
1933                 return -EIO;
1934
1935 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1936         /*
1937          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1938          * so vcpu_load() would break it.
1939          */
1940         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1941                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1942 #endif
1943
1944
1945         r = vcpu_load(vcpu);
1946         if (r)
1947                 return r;
1948         switch (ioctl) {
1949         case KVM_RUN:
1950                 r = -EINVAL;
1951                 if (arg)
1952                         goto out;
1953                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1954                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1955                 break;
1956         case KVM_GET_REGS: {
1957                 struct kvm_regs *kvm_regs;
1958
1959                 r = -ENOMEM;
1960                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1961                 if (!kvm_regs)
1962                         goto out;
1963                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1964                 if (r)
1965                         goto out_free1;
1966                 r = -EFAULT;
1967                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1968                         goto out_free1;
1969                 r = 0;
1970 out_free1:
1971                 kfree(kvm_regs);
1972                 break;
1973         }
1974         case KVM_SET_REGS: {
1975                 struct kvm_regs *kvm_regs;
1976
1977                 r = -ENOMEM;
1978                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1979                 if (IS_ERR(kvm_regs)) {
1980                         r = PTR_ERR(kvm_regs);
1981                         goto out;
1982                 }
1983                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1984                 kfree(kvm_regs);
1985                 break;
1986         }
1987         case KVM_GET_SREGS: {
1988                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1989                 r = -ENOMEM;
1990                 if (!kvm_sregs)
1991                         goto out;
1992                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1993                 if (r)
1994                         goto out;
1995                 r = -EFAULT;
1996                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1997                         goto out;
1998                 r = 0;
1999                 break;
2000         }
2001         case KVM_SET_SREGS: {
2002                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2003                 if (IS_ERR(kvm_sregs)) {
2004                         r = PTR_ERR(kvm_sregs);
2005                         kvm_sregs = NULL;
2006                         goto out;
2007                 }
2008                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2009                 break;
2010         }
2011         case KVM_GET_MP_STATE: {
2012                 struct kvm_mp_state mp_state;
2013
2014                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2015                 if (r)
2016                         goto out;
2017                 r = -EFAULT;
2018                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2019                         goto out;
2020                 r = 0;
2021                 break;
2022         }
2023         case KVM_SET_MP_STATE: {
2024                 struct kvm_mp_state mp_state;
2025
2026                 r = -EFAULT;
2027                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2028                         goto out;
2029                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2030                 break;
2031         }
2032         case KVM_TRANSLATE: {
2033                 struct kvm_translation tr;
2034
2035                 r = -EFAULT;
2036                 if (copy_from_user(&tr, argp, sizeof tr))
2037                         goto out;
2038                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2039                 if (r)
2040                         goto out;
2041                 r = -EFAULT;
2042                 if (copy_to_user(argp, &tr, sizeof tr))
2043                         goto out;
2044                 r = 0;
2045                 break;
2046         }
2047         case KVM_SET_GUEST_DEBUG: {
2048                 struct kvm_guest_debug dbg;
2049
2050                 r = -EFAULT;
2051                 if (copy_from_user(&dbg, argp, sizeof dbg))
2052                         goto out;
2053                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2054                 break;
2055         }
2056         case KVM_SET_SIGNAL_MASK: {
2057                 struct kvm_signal_mask __user *sigmask_arg = argp;
2058                 struct kvm_signal_mask kvm_sigmask;
2059                 sigset_t sigset, *p;
2060
2061                 p = NULL;
2062                 if (argp) {
2063                         r = -EFAULT;
2064                         if (copy_from_user(&kvm_sigmask, argp,
2065                                            sizeof kvm_sigmask))
2066                                 goto out;
2067                         r = -EINVAL;
2068                         if (kvm_sigmask.len != sizeof sigset)
2069                                 goto out;
2070                         r = -EFAULT;
2071                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2072                                            sizeof sigset))
2073                                 goto out;
2074                         p = &sigset;
2075                 }
2076                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2077                 break;
2078         }
2079         case KVM_GET_FPU: {
2080                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2081                 r = -ENOMEM;
2082                 if (!fpu)
2083                         goto out;
2084                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2085                 if (r)
2086                         goto out;
2087                 r = -EFAULT;
2088                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2089                         goto out;
2090                 r = 0;
2091                 break;
2092         }
2093         case KVM_SET_FPU: {
2094                 fpu = memdup_user(argp, sizeof(*fpu));
2095                 if (IS_ERR(fpu)) {
2096                         r = PTR_ERR(fpu);
2097                         fpu = NULL;
2098                         goto out;
2099                 }
2100                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2101                 break;
2102         }
2103         default:
2104                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2105         }
2106 out:
2107         vcpu_put(vcpu);
2108         kfree(fpu);
2109         kfree(kvm_sregs);
2110         return r;
2111 }
2112
2113 #ifdef CONFIG_COMPAT
2114 static long kvm_vcpu_compat_ioctl(struct file *filp,
2115                                   unsigned int ioctl, unsigned long arg)
2116 {
2117         struct kvm_vcpu *vcpu = filp->private_data;
2118         void __user *argp = compat_ptr(arg);
2119         int r;
2120
2121         if (vcpu->kvm->mm != current->mm)
2122                 return -EIO;
2123
2124         switch (ioctl) {
2125         case KVM_SET_SIGNAL_MASK: {
2126                 struct kvm_signal_mask __user *sigmask_arg = argp;
2127                 struct kvm_signal_mask kvm_sigmask;
2128                 compat_sigset_t csigset;
2129                 sigset_t sigset;
2130
2131                 if (argp) {
2132                         r = -EFAULT;
2133                         if (copy_from_user(&kvm_sigmask, argp,
2134                                            sizeof kvm_sigmask))
2135                                 goto out;
2136                         r = -EINVAL;
2137                         if (kvm_sigmask.len != sizeof csigset)
2138                                 goto out;
2139                         r = -EFAULT;
2140                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2141                                            sizeof csigset))
2142                                 goto out;
2143                         sigset_from_compat(&sigset, &csigset);
2144                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2145                 } else
2146                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2147                 break;
2148         }
2149         default:
2150                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2151         }
2152
2153 out:
2154         return r;
2155 }
2156 #endif
2157
2158 static long kvm_vm_ioctl(struct file *filp,
2159                            unsigned int ioctl, unsigned long arg)
2160 {
2161         struct kvm *kvm = filp->private_data;
2162         void __user *argp = (void __user *)arg;
2163         int r;
2164
2165         if (kvm->mm != current->mm)
2166                 return -EIO;
2167         switch (ioctl) {
2168         case KVM_CREATE_VCPU:
2169                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2170                 break;
2171         case KVM_SET_USER_MEMORY_REGION: {
2172                 struct kvm_userspace_memory_region kvm_userspace_mem;
2173
2174                 r = -EFAULT;
2175                 if (copy_from_user(&kvm_userspace_mem, argp,
2176                                                 sizeof kvm_userspace_mem))
2177                         goto out;
2178
2179                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2180                 break;
2181         }
2182         case KVM_GET_DIRTY_LOG: {
2183                 struct kvm_dirty_log log;
2184
2185                 r = -EFAULT;
2186                 if (copy_from_user(&log, argp, sizeof log))
2187                         goto out;
2188                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2189                 break;
2190         }
2191 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2192         case KVM_REGISTER_COALESCED_MMIO: {
2193                 struct kvm_coalesced_mmio_zone zone;
2194                 r = -EFAULT;
2195                 if (copy_from_user(&zone, argp, sizeof zone))
2196                         goto out;
2197                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2198                 break;
2199         }
2200         case KVM_UNREGISTER_COALESCED_MMIO: {
2201                 struct kvm_coalesced_mmio_zone zone;
2202                 r = -EFAULT;
2203                 if (copy_from_user(&zone, argp, sizeof zone))
2204                         goto out;
2205                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2206                 break;
2207         }
2208 #endif
2209         case KVM_IRQFD: {
2210                 struct kvm_irqfd data;
2211
2212                 r = -EFAULT;
2213                 if (copy_from_user(&data, argp, sizeof data))
2214                         goto out;
2215                 r = kvm_irqfd(kvm, &data);
2216                 break;
2217         }
2218         case KVM_IOEVENTFD: {
2219                 struct kvm_ioeventfd data;
2220
2221                 r = -EFAULT;
2222                 if (copy_from_user(&data, argp, sizeof data))
2223                         goto out;
2224                 r = kvm_ioeventfd(kvm, &data);
2225                 break;
2226         }
2227 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2228         case KVM_SET_BOOT_CPU_ID:
2229                 r = 0;
2230                 mutex_lock(&kvm->lock);
2231                 if (atomic_read(&kvm->online_vcpus) != 0)
2232                         r = -EBUSY;
2233                 else
2234                         kvm->bsp_vcpu_id = arg;
2235                 mutex_unlock(&kvm->lock);
2236                 break;
2237 #endif
2238 #ifdef CONFIG_HAVE_KVM_MSI
2239         case KVM_SIGNAL_MSI: {
2240                 struct kvm_msi msi;
2241
2242                 r = -EFAULT;
2243                 if (copy_from_user(&msi, argp, sizeof msi))
2244                         goto out;
2245                 r = kvm_send_userspace_msi(kvm, &msi);
2246                 break;
2247         }
2248 #endif
2249 #ifdef __KVM_HAVE_IRQ_LINE
2250         case KVM_IRQ_LINE_STATUS:
2251         case KVM_IRQ_LINE: {
2252                 struct kvm_irq_level irq_event;
2253
2254                 r = -EFAULT;
2255                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2256                         goto out;
2257
2258                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2259                 if (r)
2260                         goto out;
2261
2262                 r = -EFAULT;
2263                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2264                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2265                                 goto out;
2266                 }
2267
2268                 r = 0;
2269                 break;
2270         }
2271 #endif
2272         default:
2273                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2274                 if (r == -ENOTTY)
2275                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2276         }
2277 out:
2278         return r;
2279 }
2280
2281 #ifdef CONFIG_COMPAT
2282 struct compat_kvm_dirty_log {
2283         __u32 slot;
2284         __u32 padding1;
2285         union {
2286                 compat_uptr_t dirty_bitmap; /* one bit per page */
2287                 __u64 padding2;
2288         };
2289 };
2290
2291 static long kvm_vm_compat_ioctl(struct file *filp,
2292                            unsigned int ioctl, unsigned long arg)
2293 {
2294         struct kvm *kvm = filp->private_data;
2295         int r;
2296
2297         if (kvm->mm != current->mm)
2298                 return -EIO;
2299         switch (ioctl) {
2300         case KVM_GET_DIRTY_LOG: {
2301                 struct compat_kvm_dirty_log compat_log;
2302                 struct kvm_dirty_log log;
2303
2304                 r = -EFAULT;
2305                 if (copy_from_user(&compat_log, (void __user *)arg,
2306                                    sizeof(compat_log)))
2307                         goto out;
2308                 log.slot         = compat_log.slot;
2309                 log.padding1     = compat_log.padding1;
2310                 log.padding2     = compat_log.padding2;
2311                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2312
2313                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2314                 break;
2315         }
2316         default:
2317                 r = kvm_vm_ioctl(filp, ioctl, arg);
2318         }
2319
2320 out:
2321         return r;
2322 }
2323 #endif
2324
2325 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2326 {
2327         struct page *page[1];
2328         unsigned long addr;
2329         int npages;
2330         gfn_t gfn = vmf->pgoff;
2331         struct kvm *kvm = vma->vm_file->private_data;
2332
2333         addr = gfn_to_hva(kvm, gfn);
2334         if (kvm_is_error_hva(addr))
2335                 return VM_FAULT_SIGBUS;
2336
2337         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2338                                 NULL);
2339         if (unlikely(npages != 1))
2340                 return VM_FAULT_SIGBUS;
2341
2342         vmf->page = page[0];
2343         return 0;
2344 }
2345
2346 static const struct vm_operations_struct kvm_vm_vm_ops = {
2347         .fault = kvm_vm_fault,
2348 };
2349
2350 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2351 {
2352         vma->vm_ops = &kvm_vm_vm_ops;
2353         return 0;
2354 }
2355
2356 static struct file_operations kvm_vm_fops = {
2357         .release        = kvm_vm_release,
2358         .unlocked_ioctl = kvm_vm_ioctl,
2359 #ifdef CONFIG_COMPAT
2360         .compat_ioctl   = kvm_vm_compat_ioctl,
2361 #endif
2362         .mmap           = kvm_vm_mmap,
2363         .llseek         = noop_llseek,
2364 };
2365
2366 static int kvm_dev_ioctl_create_vm(unsigned long type)
2367 {
2368         int r;
2369         struct kvm *kvm;
2370
2371         kvm = kvm_create_vm(type);
2372         if (IS_ERR(kvm))
2373                 return PTR_ERR(kvm);
2374 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2375         r = kvm_coalesced_mmio_init(kvm);
2376         if (r < 0) {
2377                 kvm_put_kvm(kvm);
2378                 return r;
2379         }
2380 #endif
2381         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2382         if (r < 0)
2383                 kvm_put_kvm(kvm);
2384
2385         return r;
2386 }
2387
2388 static long kvm_dev_ioctl_check_extension_generic(long arg)
2389 {
2390         switch (arg) {
2391         case KVM_CAP_USER_MEMORY:
2392         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2393         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2394 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2395         case KVM_CAP_SET_BOOT_CPU_ID:
2396 #endif
2397         case KVM_CAP_INTERNAL_ERROR_DATA:
2398 #ifdef CONFIG_HAVE_KVM_MSI
2399         case KVM_CAP_SIGNAL_MSI:
2400 #endif
2401                 return 1;
2402 #ifdef KVM_CAP_IRQ_ROUTING
2403         case KVM_CAP_IRQ_ROUTING:
2404                 return KVM_MAX_IRQ_ROUTES;
2405 #endif
2406         default:
2407                 break;
2408         }
2409         return kvm_dev_ioctl_check_extension(arg);
2410 }
2411
2412 static long kvm_dev_ioctl(struct file *filp,
2413                           unsigned int ioctl, unsigned long arg)
2414 {
2415         long r = -EINVAL;
2416
2417         switch (ioctl) {
2418         case KVM_GET_API_VERSION:
2419                 r = -EINVAL;
2420                 if (arg)
2421                         goto out;
2422                 r = KVM_API_VERSION;
2423                 break;
2424         case KVM_CREATE_VM:
2425                 r = kvm_dev_ioctl_create_vm(arg);
2426                 break;
2427         case KVM_CHECK_EXTENSION:
2428                 r = kvm_dev_ioctl_check_extension_generic(arg);
2429                 break;
2430         case KVM_GET_VCPU_MMAP_SIZE:
2431                 r = -EINVAL;
2432                 if (arg)
2433                         goto out;
2434                 r = PAGE_SIZE;     /* struct kvm_run */
2435 #ifdef CONFIG_X86
2436                 r += PAGE_SIZE;    /* pio data page */
2437 #endif
2438 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2439                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2440 #endif
2441                 break;
2442         case KVM_TRACE_ENABLE:
2443         case KVM_TRACE_PAUSE:
2444         case KVM_TRACE_DISABLE:
2445                 r = -EOPNOTSUPP;
2446                 break;
2447         default:
2448                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2449         }
2450 out:
2451         return r;
2452 }
2453
2454 static struct file_operations kvm_chardev_ops = {
2455         .unlocked_ioctl = kvm_dev_ioctl,
2456         .compat_ioctl   = kvm_dev_ioctl,
2457         .llseek         = noop_llseek,
2458 };
2459
2460 static struct miscdevice kvm_dev = {
2461         KVM_MINOR,
2462         "kvm",
2463         &kvm_chardev_ops,
2464 };
2465
2466 static void hardware_enable_nolock(void *junk)
2467 {
2468         int cpu = raw_smp_processor_id();
2469         int r;
2470
2471         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2472                 return;
2473
2474         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2475
2476         r = kvm_arch_hardware_enable(NULL);
2477
2478         if (r) {
2479                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2480                 atomic_inc(&hardware_enable_failed);
2481                 printk(KERN_INFO "kvm: enabling virtualization on "
2482                                  "CPU%d failed\n", cpu);
2483         }
2484 }
2485
2486 static void hardware_enable(void *junk)
2487 {
2488         raw_spin_lock(&kvm_lock);
2489         hardware_enable_nolock(junk);
2490         raw_spin_unlock(&kvm_lock);
2491 }
2492
2493 static void hardware_disable_nolock(void *junk)
2494 {
2495         int cpu = raw_smp_processor_id();
2496
2497         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2498                 return;
2499         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2500         kvm_arch_hardware_disable(NULL);
2501 }
2502
2503 static void hardware_disable(void *junk)
2504 {
2505         raw_spin_lock(&kvm_lock);
2506         hardware_disable_nolock(junk);
2507         raw_spin_unlock(&kvm_lock);
2508 }
2509
2510 static void hardware_disable_all_nolock(void)
2511 {
2512         BUG_ON(!kvm_usage_count);
2513
2514         kvm_usage_count--;
2515         if (!kvm_usage_count)
2516                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2517 }
2518
2519 static void hardware_disable_all(void)
2520 {
2521         raw_spin_lock(&kvm_lock);
2522         hardware_disable_all_nolock();
2523         raw_spin_unlock(&kvm_lock);
2524 }
2525
2526 static int hardware_enable_all(void)
2527 {
2528         int r = 0;
2529
2530         raw_spin_lock(&kvm_lock);
2531
2532         kvm_usage_count++;
2533         if (kvm_usage_count == 1) {
2534                 atomic_set(&hardware_enable_failed, 0);
2535                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2536
2537                 if (atomic_read(&hardware_enable_failed)) {
2538                         hardware_disable_all_nolock();
2539                         r = -EBUSY;
2540                 }
2541         }
2542
2543         raw_spin_unlock(&kvm_lock);
2544
2545         return r;
2546 }
2547
2548 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2549                            void *v)
2550 {
2551         int cpu = (long)v;
2552
2553         if (!kvm_usage_count)
2554                 return NOTIFY_OK;
2555
2556         val &= ~CPU_TASKS_FROZEN;
2557         switch (val) {
2558         case CPU_DYING:
2559                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2560                        cpu);
2561                 hardware_disable(NULL);
2562                 break;
2563         case CPU_STARTING:
2564                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2565                        cpu);
2566                 hardware_enable(NULL);
2567                 break;
2568         }
2569         return NOTIFY_OK;
2570 }
2571
2572
2573 asmlinkage void kvm_spurious_fault(void)
2574 {
2575         /* Fault while not rebooting.  We want the trace. */
2576         BUG();
2577 }
2578 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2579
2580 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2581                       void *v)
2582 {
2583         /*
2584          * Some (well, at least mine) BIOSes hang on reboot if
2585          * in vmx root mode.
2586          *
2587          * And Intel TXT required VMX off for all cpu when system shutdown.
2588          */
2589         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2590         kvm_rebooting = true;
2591         on_each_cpu(hardware_disable_nolock, NULL, 1);
2592         return NOTIFY_OK;
2593 }
2594
2595 static struct notifier_block kvm_reboot_notifier = {
2596         .notifier_call = kvm_reboot,
2597         .priority = 0,
2598 };
2599
2600 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2601 {
2602         int i;
2603
2604         for (i = 0; i < bus->dev_count; i++) {
2605                 struct kvm_io_device *pos = bus->range[i].dev;
2606
2607                 kvm_iodevice_destructor(pos);
2608         }
2609         kfree(bus);
2610 }
2611
2612 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2613 {
2614         const struct kvm_io_range *r1 = p1;
2615         const struct kvm_io_range *r2 = p2;
2616
2617         if (r1->addr < r2->addr)
2618                 return -1;
2619         if (r1->addr + r1->len > r2->addr + r2->len)
2620                 return 1;
2621         return 0;
2622 }
2623
2624 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2625                           gpa_t addr, int len)
2626 {
2627         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2628                 .addr = addr,
2629                 .len = len,
2630                 .dev = dev,
2631         };
2632
2633         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2634                 kvm_io_bus_sort_cmp, NULL);
2635
2636         return 0;
2637 }
2638
2639 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2640                              gpa_t addr, int len)
2641 {
2642         struct kvm_io_range *range, key;
2643         int off;
2644
2645         key = (struct kvm_io_range) {
2646                 .addr = addr,
2647                 .len = len,
2648         };
2649
2650         range = bsearch(&key, bus->range, bus->dev_count,
2651                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2652         if (range == NULL)
2653                 return -ENOENT;
2654
2655         off = range - bus->range;
2656
2657         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2658                 off--;
2659
2660         return off;
2661 }
2662
2663 /* kvm_io_bus_write - called under kvm->slots_lock */
2664 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2665                      int len, const void *val)
2666 {
2667         int idx;
2668         struct kvm_io_bus *bus;
2669         struct kvm_io_range range;
2670
2671         range = (struct kvm_io_range) {
2672                 .addr = addr,
2673                 .len = len,
2674         };
2675
2676         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2677         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2678         if (idx < 0)
2679                 return -EOPNOTSUPP;
2680
2681         while (idx < bus->dev_count &&
2682                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2683                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2684                         return 0;
2685                 idx++;
2686         }
2687
2688         return -EOPNOTSUPP;
2689 }
2690
2691 /* kvm_io_bus_read - called under kvm->slots_lock */
2692 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2693                     int len, void *val)
2694 {
2695         int idx;
2696         struct kvm_io_bus *bus;
2697         struct kvm_io_range range;
2698
2699         range = (struct kvm_io_range) {
2700                 .addr = addr,
2701                 .len = len,
2702         };
2703
2704         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2705         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2706         if (idx < 0)
2707                 return -EOPNOTSUPP;
2708
2709         while (idx < bus->dev_count &&
2710                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2711                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2712                         return 0;
2713                 idx++;
2714         }
2715
2716         return -EOPNOTSUPP;
2717 }
2718
2719 /* Caller must hold slots_lock. */
2720 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2721                             int len, struct kvm_io_device *dev)
2722 {
2723         struct kvm_io_bus *new_bus, *bus;
2724
2725         bus = kvm->buses[bus_idx];
2726         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2727                 return -ENOSPC;
2728
2729         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2730                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2731         if (!new_bus)
2732                 return -ENOMEM;
2733         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2734                sizeof(struct kvm_io_range)));
2735         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2736         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2737         synchronize_srcu_expedited(&kvm->srcu);
2738         kfree(bus);
2739
2740         return 0;
2741 }
2742
2743 /* Caller must hold slots_lock. */
2744 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2745                               struct kvm_io_device *dev)
2746 {
2747         int i, r;
2748         struct kvm_io_bus *new_bus, *bus;
2749
2750         bus = kvm->buses[bus_idx];
2751         r = -ENOENT;
2752         for (i = 0; i < bus->dev_count; i++)
2753                 if (bus->range[i].dev == dev) {
2754                         r = 0;
2755                         break;
2756                 }
2757
2758         if (r)
2759                 return r;
2760
2761         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2762                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2763         if (!new_bus)
2764                 return -ENOMEM;
2765
2766         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2767         new_bus->dev_count--;
2768         memcpy(new_bus->range + i, bus->range + i + 1,
2769                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2770
2771         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2772         synchronize_srcu_expedited(&kvm->srcu);
2773         kfree(bus);
2774         return r;
2775 }
2776
2777 static struct notifier_block kvm_cpu_notifier = {
2778         .notifier_call = kvm_cpu_hotplug,
2779 };
2780
2781 static int vm_stat_get(void *_offset, u64 *val)
2782 {
2783         unsigned offset = (long)_offset;
2784         struct kvm *kvm;
2785
2786         *val = 0;
2787         raw_spin_lock(&kvm_lock);
2788         list_for_each_entry(kvm, &vm_list, vm_list)
2789                 *val += *(u32 *)((void *)kvm + offset);
2790         raw_spin_unlock(&kvm_lock);
2791         return 0;
2792 }
2793
2794 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2795
2796 static int vcpu_stat_get(void *_offset, u64 *val)
2797 {
2798         unsigned offset = (long)_offset;
2799         struct kvm *kvm;
2800         struct kvm_vcpu *vcpu;
2801         int i;
2802
2803         *val = 0;
2804         raw_spin_lock(&kvm_lock);
2805         list_for_each_entry(kvm, &vm_list, vm_list)
2806                 kvm_for_each_vcpu(i, vcpu, kvm)
2807                         *val += *(u32 *)((void *)vcpu + offset);
2808
2809         raw_spin_unlock(&kvm_lock);
2810         return 0;
2811 }
2812
2813 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2814
2815 static const struct file_operations *stat_fops[] = {
2816         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2817         [KVM_STAT_VM]   = &vm_stat_fops,
2818 };
2819
2820 static int kvm_init_debug(void)
2821 {
2822         int r = -EFAULT;
2823         struct kvm_stats_debugfs_item *p;
2824
2825         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2826         if (kvm_debugfs_dir == NULL)
2827                 goto out;
2828
2829         for (p = debugfs_entries; p->name; ++p) {
2830                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2831                                                 (void *)(long)p->offset,
2832                                                 stat_fops[p->kind]);
2833                 if (p->dentry == NULL)
2834                         goto out_dir;
2835         }
2836
2837         return 0;
2838
2839 out_dir:
2840         debugfs_remove_recursive(kvm_debugfs_dir);
2841 out:
2842         return r;
2843 }
2844
2845 static void kvm_exit_debug(void)
2846 {
2847         struct kvm_stats_debugfs_item *p;
2848
2849         for (p = debugfs_entries; p->name; ++p)
2850                 debugfs_remove(p->dentry);
2851         debugfs_remove(kvm_debugfs_dir);
2852 }
2853
2854 static int kvm_suspend(void)
2855 {
2856         if (kvm_usage_count)
2857                 hardware_disable_nolock(NULL);
2858         return 0;
2859 }
2860
2861 static void kvm_resume(void)
2862 {
2863         if (kvm_usage_count) {
2864                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2865                 hardware_enable_nolock(NULL);
2866         }
2867 }
2868
2869 static struct syscore_ops kvm_syscore_ops = {
2870         .suspend = kvm_suspend,
2871         .resume = kvm_resume,
2872 };
2873
2874 static inline
2875 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2876 {
2877         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2878 }
2879
2880 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2881 {
2882         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2883
2884         kvm_arch_vcpu_load(vcpu, cpu);
2885 }
2886
2887 static void kvm_sched_out(struct preempt_notifier *pn,
2888                           struct task_struct *next)
2889 {
2890         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2891
2892         kvm_arch_vcpu_put(vcpu);
2893 }
2894
2895 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2896                   struct module *module)
2897 {
2898         int r;
2899         int cpu;
2900
2901         r = kvm_arch_init(opaque);
2902         if (r)
2903                 goto out_fail;
2904
2905         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2906                 r = -ENOMEM;
2907                 goto out_free_0;
2908         }
2909
2910         r = kvm_arch_hardware_setup();
2911         if (r < 0)
2912                 goto out_free_0a;
2913
2914         for_each_online_cpu(cpu) {
2915                 smp_call_function_single(cpu,
2916                                 kvm_arch_check_processor_compat,
2917                                 &r, 1);
2918                 if (r < 0)
2919                         goto out_free_1;
2920         }
2921
2922         r = register_cpu_notifier(&kvm_cpu_notifier);
2923         if (r)
2924                 goto out_free_2;
2925         register_reboot_notifier(&kvm_reboot_notifier);
2926
2927         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2928         if (!vcpu_align)
2929                 vcpu_align = __alignof__(struct kvm_vcpu);
2930         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2931                                            0, NULL);
2932         if (!kvm_vcpu_cache) {
2933                 r = -ENOMEM;
2934                 goto out_free_3;
2935         }
2936
2937         r = kvm_async_pf_init();
2938         if (r)
2939                 goto out_free;
2940
2941         kvm_chardev_ops.owner = module;
2942         kvm_vm_fops.owner = module;
2943         kvm_vcpu_fops.owner = module;
2944
2945         r = misc_register(&kvm_dev);
2946         if (r) {
2947                 printk(KERN_ERR "kvm: misc device register failed\n");
2948                 goto out_unreg;
2949         }
2950
2951         register_syscore_ops(&kvm_syscore_ops);
2952
2953         kvm_preempt_ops.sched_in = kvm_sched_in;
2954         kvm_preempt_ops.sched_out = kvm_sched_out;
2955
2956         r = kvm_init_debug();
2957         if (r) {
2958                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2959                 goto out_undebugfs;
2960         }
2961
2962         return 0;
2963
2964 out_undebugfs:
2965         unregister_syscore_ops(&kvm_syscore_ops);
2966 out_unreg:
2967         kvm_async_pf_deinit();
2968 out_free:
2969         kmem_cache_destroy(kvm_vcpu_cache);
2970 out_free_3:
2971         unregister_reboot_notifier(&kvm_reboot_notifier);
2972         unregister_cpu_notifier(&kvm_cpu_notifier);
2973 out_free_2:
2974 out_free_1:
2975         kvm_arch_hardware_unsetup();
2976 out_free_0a:
2977         free_cpumask_var(cpus_hardware_enabled);
2978 out_free_0:
2979         kvm_arch_exit();
2980 out_fail:
2981         return r;
2982 }
2983 EXPORT_SYMBOL_GPL(kvm_init);
2984
2985 void kvm_exit(void)
2986 {
2987         kvm_exit_debug();
2988         misc_deregister(&kvm_dev);
2989         kmem_cache_destroy(kvm_vcpu_cache);
2990         kvm_async_pf_deinit();
2991         unregister_syscore_ops(&kvm_syscore_ops);
2992         unregister_reboot_notifier(&kvm_reboot_notifier);
2993         unregister_cpu_notifier(&kvm_cpu_notifier);
2994         on_each_cpu(hardware_disable_nolock, NULL, 1);
2995         kvm_arch_hardware_unsetup();
2996         kvm_arch_exit();
2997         free_cpumask_var(cpus_hardware_enabled);
2998 }
2999 EXPORT_SYMBOL_GPL(kvm_exit);