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_scan_ioapic_request(struct kvm *kvm)
 221 {
 222         make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
 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         vcpu->preempted = false;
 248
 249         r = kvm_arch_vcpu_init(vcpu);
 250         if (r < 0)
 251                 goto fail_free_run;
 252         return 0;
 253
 254 fail_free_run:
 255         free_page((unsigned long)vcpu->run);
 256 fail:
 257         return r;
 258 }
 259 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 260
 261 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 262 {
 263         put_pid(vcpu->pid);
 264         kvm_arch_vcpu_uninit(vcpu);
 265         free_page((unsigned long)vcpu->run);
 266 }
 267 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 268
 269 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 270 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 271 {
 272         return container_of(mn, struct kvm, mmu_notifier);
 273 }
 274
 275 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 276                                              struct mm_struct *mm,
 277                                              unsigned long address)
 278 {
 279         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 280         int need_tlb_flush, idx;
 281
 282         /*
 283          * When ->invalidate_page runs, the linux pte has been zapped
 284          * already but the page is still allocated until
 285          * ->invalidate_page returns. So if we increase the sequence
 286          * here the kvm page fault will notice if the spte can't be
 287          * established because the page is going to be freed. If
 288          * instead the kvm page fault establishes the spte before
 289          * ->invalidate_page runs, kvm_unmap_hva will release it
 290          * before returning.
 291          *
 292          * The sequence increase only need to be seen at spin_unlock
 293          * time, and not at spin_lock time.
 294          *
 295          * Increasing the sequence after the spin_unlock would be
 296          * unsafe because the kvm page fault could then establish the
 297          * pte after kvm_unmap_hva returned, without noticing the page
 298          * is going to be freed.
 299          */
 300         idx = srcu_read_lock(&kvm->srcu);
 301         spin_lock(&kvm->mmu_lock);
 302
 303         kvm->mmu_notifier_seq++;
 304         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 305         /* we've to flush the tlb before the pages can be freed */
 306         if (need_tlb_flush)
 307                 kvm_flush_remote_tlbs(kvm);
 308
 309         spin_unlock(&kvm->mmu_lock);
 310         srcu_read_unlock(&kvm->srcu, idx);
 311 }
 312
 313 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 314                                         struct mm_struct *mm,
 315                                         unsigned long address,
 316                                         pte_t pte)
 317 {
 318         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 319         int idx;
 320
 321         idx = srcu_read_lock(&kvm->srcu);
 322         spin_lock(&kvm->mmu_lock);
 323         kvm->mmu_notifier_seq++;
 324         kvm_set_spte_hva(kvm, address, pte);
 325         spin_unlock(&kvm->mmu_lock);
 326         srcu_read_unlock(&kvm->srcu, idx);
 327 }
 328
 329 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 330                                                     struct mm_struct *mm,
 331                                                     unsigned long start,
 332                                                     unsigned long end)
 333 {
 334         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 335         int need_tlb_flush = 0, idx;
 336
 337         idx = srcu_read_lock(&kvm->srcu);
 338         spin_lock(&kvm->mmu_lock);
 339         /*
 340          * The count increase must become visible at unlock time as no
 341          * spte can be established without taking the mmu_lock and
 342          * count is also read inside the mmu_lock critical section.
 343          */
 344         kvm->mmu_notifier_count++;
 345         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 346         need_tlb_flush |= kvm->tlbs_dirty;
 347         /* we've to flush the tlb before the pages can be freed */
 348         if (need_tlb_flush)
 349                 kvm_flush_remote_tlbs(kvm);
 350
 351         spin_unlock(&kvm->mmu_lock);
 352         srcu_read_unlock(&kvm->srcu, idx);
 353 }
 354
 355 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 356                                                   struct mm_struct *mm,
 357                                                   unsigned long start,
 358                                                   unsigned long end)
 359 {
 360         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 361
 362         spin_lock(&kvm->mmu_lock);
 363         /*
 364          * This sequence increase will notify the kvm page fault that
 365          * the page that is going to be mapped in the spte could have
 366          * been freed.
 367          */
 368         kvm->mmu_notifier_seq++;
 369         smp_wmb();
 370         /*
 371          * The above sequence increase must be visible before the
 372          * below count decrease, which is ensured by the smp_wmb above
 373          * in conjunction with the smp_rmb in mmu_notifier_retry().
 374          */
 375         kvm->mmu_notifier_count--;
 376         spin_unlock(&kvm->mmu_lock);
 377
 378         BUG_ON(kvm->mmu_notifier_count < 0);
 379 }
 380
 381 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 382                                               struct mm_struct *mm,
 383                                               unsigned long address)
 384 {
 385         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 386         int young, idx;
 387
 388         idx = srcu_read_lock(&kvm->srcu);
 389         spin_lock(&kvm->mmu_lock);
 390
 391         young = kvm_age_hva(kvm, address);
 392         if (young)
 393                 kvm_flush_remote_tlbs(kvm);
 394
 395         spin_unlock(&kvm->mmu_lock);
 396         srcu_read_unlock(&kvm->srcu, idx);
 397
 398         return young;
 399 }
 400
 401 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 402                                        struct mm_struct *mm,
 403                                        unsigned long address)
 404 {
 405         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 406         int young, idx;
 407
 408         idx = srcu_read_lock(&kvm->srcu);
 409         spin_lock(&kvm->mmu_lock);
 410         young = kvm_test_age_hva(kvm, address);
 411         spin_unlock(&kvm->mmu_lock);
 412         srcu_read_unlock(&kvm->srcu, idx);
 413
 414         return young;
 415 }
 416
 417 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 418                                      struct mm_struct *mm)
 419 {
 420         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 421         int idx;
 422
 423         idx = srcu_read_lock(&kvm->srcu);
 424         kvm_arch_flush_shadow_all(kvm);
 425         srcu_read_unlock(&kvm->srcu, idx);
 426 }
 427
 428 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 429         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 430         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 431         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 432         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 433         .test_young             = kvm_mmu_notifier_test_young,
 434         .change_pte             = kvm_mmu_notifier_change_pte,
 435         .release                = kvm_mmu_notifier_release,
 436 };
 437
 438 static int kvm_init_mmu_notifier(struct kvm *kvm)
 439 {
 440         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 441         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 442 }
 443
 444 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 445
 446 static int kvm_init_mmu_notifier(struct kvm *kvm)
 447 {
 448         return 0;
 449 }
 450
 451 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 452
 453 static void kvm_init_memslots_id(struct kvm *kvm)
 454 {
 455         int i;
 456         struct kvm_memslots *slots = kvm->memslots;
 457
 458         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 459                 slots->id_to_index[i] = slots->memslots[i].id = i;
 460 }
 461
 462 static struct kvm *kvm_create_vm(unsigned long type)
 463 {
 464         int r, i;
 465         struct kvm *kvm = kvm_arch_alloc_vm();
 466
 467         if (!kvm)
 468                 return ERR_PTR(-ENOMEM);
 469
 470         r = kvm_arch_init_vm(kvm, type);
 471         if (r)
 472                 goto out_err_nodisable;
 473
 474         r = hardware_enable_all();
 475         if (r)
 476                 goto out_err_nodisable;
 477
 478 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 479         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 480         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 481 #endif
 482
 483         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 484
 485         r = -ENOMEM;
 486         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 487         if (!kvm->memslots)
 488                 goto out_err_nosrcu;
 489         kvm_init_memslots_id(kvm);
 490         if (init_srcu_struct(&kvm->srcu))
 491                 goto out_err_nosrcu;
 492         for (i = 0; i < KVM_NR_BUSES; i++) {
 493                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 494                                         GFP_KERNEL);
 495                 if (!kvm->buses[i])
 496                         goto out_err;
 497         }
 498
 499         spin_lock_init(&kvm->mmu_lock);
 500         kvm->mm = current->mm;
 501         atomic_inc(&kvm->mm->mm_count);
 502         kvm_eventfd_init(kvm);
 503         mutex_init(&kvm->lock);
 504         mutex_init(&kvm->irq_lock);
 505         mutex_init(&kvm->slots_lock);
 506         atomic_set(&kvm->users_count, 1);
 507         INIT_LIST_HEAD(&kvm->devices);
 508
 509         r = kvm_init_mmu_notifier(kvm);
 510         if (r)
 511                 goto out_err;
 512
 513         raw_spin_lock(&kvm_lock);
 514         list_add(&kvm->vm_list, &vm_list);
 515         raw_spin_unlock(&kvm_lock);
 516
 517         return kvm;
 518
 519 out_err:
 520         cleanup_srcu_struct(&kvm->srcu);
 521 out_err_nosrcu:
 522         hardware_disable_all();
 523 out_err_nodisable:
 524         for (i = 0; i < KVM_NR_BUSES; i++)
 525                 kfree(kvm->buses[i]);
 526         kfree(kvm->memslots);
 527         kvm_arch_free_vm(kvm);
 528         return ERR_PTR(r);
 529 }
 530
 531 /*
 532  * Avoid using vmalloc for a small buffer.
 533  * Should not be used when the size is statically known.
 534  */
 535 void *kvm_kvzalloc(unsigned long size)
 536 {
 537         if (size > PAGE_SIZE)
 538                 return vzalloc(size);
 539         else
 540                 return kzalloc(size, GFP_KERNEL);
 541 }
 542
 543 void kvm_kvfree(const void *addr)
 544 {
 545         if (is_vmalloc_addr(addr))
 546                 vfree(addr);
 547         else
 548                 kfree(addr);
 549 }
 550
 551 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 552 {
 553         if (!memslot->dirty_bitmap)
 554                 return;
 555
 556         kvm_kvfree(memslot->dirty_bitmap);
 557         memslot->dirty_bitmap = NULL;
 558 }
 559
 560 /*
 561  * Free any memory in @free but not in @dont.
 562  */
 563 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 564                                   struct kvm_memory_slot *dont)
 565 {
 566         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 567                 kvm_destroy_dirty_bitmap(free);
 568
 569         kvm_arch_free_memslot(free, dont);
 570
 571         free->npages = 0;
 572 }
 573
 574 void kvm_free_physmem(struct kvm *kvm)
 575 {
 576         struct kvm_memslots *slots = kvm->memslots;
 577         struct kvm_memory_slot *memslot;
 578
 579         kvm_for_each_memslot(memslot, slots)
 580                 kvm_free_physmem_slot(memslot, NULL);
 581
 582         kfree(kvm->memslots);
 583 }
 584
 585 static void kvm_destroy_devices(struct kvm *kvm)
 586 {
 587         struct list_head *node, *tmp;
 588
 589         list_for_each_safe(node, tmp, &kvm->devices) {
 590                 struct kvm_device *dev =
 591                         list_entry(node, struct kvm_device, vm_node);
 592
 593                 list_del(node);
 594                 dev->ops->destroy(dev);
 595         }
 596 }
 597
 598 static void kvm_destroy_vm(struct kvm *kvm)
 599 {
 600         int i;
 601         struct mm_struct *mm = kvm->mm;
 602
 603         kvm_arch_sync_events(kvm);
 604         raw_spin_lock(&kvm_lock);
 605         list_del(&kvm->vm_list);
 606         raw_spin_unlock(&kvm_lock);
 607         kvm_free_irq_routing(kvm);
 608         for (i = 0; i < KVM_NR_BUSES; i++)
 609                 kvm_io_bus_destroy(kvm->buses[i]);
 610         kvm_coalesced_mmio_free(kvm);
 611 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 612         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 613 #else
 614         kvm_arch_flush_shadow_all(kvm);
 615 #endif
 616         kvm_arch_destroy_vm(kvm);
 617         kvm_destroy_devices(kvm);
 618         kvm_free_physmem(kvm);
 619         cleanup_srcu_struct(&kvm->srcu);
 620         kvm_arch_free_vm(kvm);
 621         hardware_disable_all();
 622         mmdrop(mm);
 623 }
 624
 625 void kvm_get_kvm(struct kvm *kvm)
 626 {
 627         atomic_inc(&kvm->users_count);
 628 }
 629 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 630
 631 void kvm_put_kvm(struct kvm *kvm)
 632 {
 633         if (atomic_dec_and_test(&kvm->users_count))
 634                 kvm_destroy_vm(kvm);
 635 }
 636 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 637
 638
 639 static int kvm_vm_release(struct inode *inode, struct file *filp)
 640 {
 641         struct kvm *kvm = filp->private_data;
 642
 643         kvm_irqfd_release(kvm);
 644
 645         kvm_put_kvm(kvm);
 646         return 0;
 647 }
 648
 649 /*
 650  * Allocation size is twice as large as the actual dirty bitmap size.
 651  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 652  */
 653 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 654 {
 655 #ifndef CONFIG_S390
 656         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 657
 658         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 659         if (!memslot->dirty_bitmap)
 660                 return -ENOMEM;
 661
 662 #endif /* !CONFIG_S390 */
 663         return 0;
 664 }
 665
 666 static int cmp_memslot(const void *slot1, const void *slot2)
 667 {
 668         struct kvm_memory_slot *s1, *s2;
 669
 670         s1 = (struct kvm_memory_slot *)slot1;
 671         s2 = (struct kvm_memory_slot *)slot2;
 672
 673         if (s1->npages < s2->npages)
 674                 return 1;
 675         if (s1->npages > s2->npages)
 676                 return -1;
 677
 678         return 0;
 679 }
 680
 681 /*
 682  * Sort the memslots base on its size, so the larger slots
 683  * will get better fit.
 684  */
 685 static void sort_memslots(struct kvm_memslots *slots)
 686 {
 687         int i;
 688
 689         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 690               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 691
 692         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 693                 slots->id_to_index[slots->memslots[i].id] = i;
 694 }
 695
 696 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
 697                      u64 last_generation)
 698 {
 699         if (new) {
 700                 int id = new->id;
 701                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 702                 unsigned long npages = old->npages;
 703
 704                 *old = *new;
 705                 if (new->npages != npages)
 706                         sort_memslots(slots);
 707         }
 708
 709         slots->generation = last_generation + 1;
 710 }
 711
 712 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 713 {
 714         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 715
 716 #ifdef KVM_CAP_READONLY_MEM
 717         valid_flags |= KVM_MEM_READONLY;
 718 #endif
 719
 720         if (mem->flags & ~valid_flags)
 721                 return -EINVAL;
 722
 723         return 0;
 724 }
 725
 726 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 727                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 728 {
 729         struct kvm_memslots *old_memslots = kvm->memslots;
 730
 731         update_memslots(slots, new, kvm->memslots->generation);
 732         rcu_assign_pointer(kvm->memslots, slots);
 733         synchronize_srcu_expedited(&kvm->srcu);
 734         return old_memslots;
 735 }
 736
 737 /*
 738  * Allocate some memory and give it an address in the guest physical address
 739  * space.
 740  *
 741  * Discontiguous memory is allowed, mostly for framebuffers.
 742  *
 743  * Must be called holding mmap_sem for write.
 744  */
 745 int __kvm_set_memory_region(struct kvm *kvm,
 746                             struct kvm_userspace_memory_region *mem)
 747 {
 748         int r;
 749         gfn_t base_gfn;
 750         unsigned long npages;
 751         struct kvm_memory_slot *slot;
 752         struct kvm_memory_slot old, new;
 753         struct kvm_memslots *slots = NULL, *old_memslots;
 754         enum kvm_mr_change change;
 755
 756         r = check_memory_region_flags(mem);
 757         if (r)
 758                 goto out;
 759
 760         r = -EINVAL;
 761         /* General sanity checks */
 762         if (mem->memory_size & (PAGE_SIZE - 1))
 763                 goto out;
 764         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 765                 goto out;
 766         /* We can read the guest memory with __xxx_user() later on. */
 767         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 768             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 769              !access_ok(VERIFY_WRITE,
 770                         (void __user *)(unsigned long)mem->userspace_addr,
 771                         mem->memory_size)))
 772                 goto out;
 773         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 774                 goto out;
 775         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 776                 goto out;
 777
 778         slot = id_to_memslot(kvm->memslots, mem->slot);
 779         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 780         npages = mem->memory_size >> PAGE_SHIFT;
 781
 782         r = -EINVAL;
 783         if (npages > KVM_MEM_MAX_NR_PAGES)
 784                 goto out;
 785
 786         if (!npages)
 787                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 788
 789         new = old = *slot;
 790
 791         new.id = mem->slot;
 792         new.base_gfn = base_gfn;
 793         new.npages = npages;
 794         new.flags = mem->flags;
 795
 796         r = -EINVAL;
 797         if (npages) {
 798                 if (!old.npages)
 799                         change = KVM_MR_CREATE;
 800                 else { /* Modify an existing slot. */
 801                         if ((mem->userspace_addr != old.userspace_addr) ||
 802                             (npages != old.npages) ||
 803                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 804                                 goto out;
 805
 806                         if (base_gfn != old.base_gfn)
 807                                 change = KVM_MR_MOVE;
 808                         else if (new.flags != old.flags)
 809                                 change = KVM_MR_FLAGS_ONLY;
 810                         else { /* Nothing to change. */
 811                                 r = 0;
 812                                 goto out;
 813                         }
 814                 }
 815         } else if (old.npages) {
 816                 change = KVM_MR_DELETE;
 817         } else /* Modify a non-existent slot: disallowed. */
 818                 goto out;
 819
 820         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 821                 /* Check for overlaps */
 822                 r = -EEXIST;
 823                 kvm_for_each_memslot(slot, kvm->memslots) {
 824                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 825                             (slot->id == mem->slot))
 826                                 continue;
 827                         if (!((base_gfn + npages <= slot->base_gfn) ||
 828                               (base_gfn >= slot->base_gfn + slot->npages)))
 829                                 goto out;
 830                 }
 831         }
 832
 833         /* Free page dirty bitmap if unneeded */
 834         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 835                 new.dirty_bitmap = NULL;
 836
 837         r = -ENOMEM;
 838         if (change == KVM_MR_CREATE) {
 839                 new.userspace_addr = mem->userspace_addr;
 840
 841                 if (kvm_arch_create_memslot(&new, npages))
 842                         goto out_free;
 843         }
 844
 845         /* Allocate page dirty bitmap if needed */
 846         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 847                 if (kvm_create_dirty_bitmap(&new) < 0)
 848                         goto out_free;
 849         }
 850
 851         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 852                 r = -ENOMEM;
 853                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 854                                 GFP_KERNEL);
 855                 if (!slots)
 856                         goto out_free;
 857                 slot = id_to_memslot(slots, mem->slot);
 858                 slot->flags |= KVM_MEMSLOT_INVALID;
 859
 860                 old_memslots = install_new_memslots(kvm, slots, NULL);
 861
 862                 /* slot was deleted or moved, clear iommu mapping */
 863                 kvm_iommu_unmap_pages(kvm, &old);
 864                 /* From this point no new shadow pages pointing to a deleted,
 865                  * or moved, memslot will be created.
 866                  *
 867                  * validation of sp->gfn happens in:
 868                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 869                  *      - kvm_is_visible_gfn (mmu_check_roots)
 870                  */
 871                 kvm_arch_flush_shadow_memslot(kvm, slot);
 872                 slots = old_memslots;
 873         }
 874
 875         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 876         if (r)
 877                 goto out_slots;
 878
 879         r = -ENOMEM;
 880         /*
 881          * We can re-use the old_memslots from above, the only difference
 882          * from the currently installed memslots is the invalid flag.  This
 883          * will get overwritten by update_memslots anyway.
 884          */
 885         if (!slots) {
 886                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 887                                 GFP_KERNEL);
 888                 if (!slots)
 889                         goto out_free;
 890         }
 891
 892         /*
 893          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 894          * un-mapped and re-mapped if their base changes.  Since base change
 895          * unmapping is handled above with slot deletion, mapping alone is
 896          * needed here.  Anything else the iommu might care about for existing
 897          * slots (size changes, userspace addr changes and read-only flag
 898          * changes) is disallowed above, so any other attribute changes getting
 899          * here can be skipped.
 900          */
 901         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 902                 r = kvm_iommu_map_pages(kvm, &new);
 903                 if (r)
 904                         goto out_slots;
 905         }
 906
 907         /* actual memory is freed via old in kvm_free_physmem_slot below */
 908         if (change == KVM_MR_DELETE) {
 909                 new.dirty_bitmap = NULL;
 910                 memset(&new.arch, 0, sizeof(new.arch));
 911         }
 912
 913         old_memslots = install_new_memslots(kvm, slots, &new);
 914
 915         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 916
 917         kvm_free_physmem_slot(&old, &new);
 918         kfree(old_memslots);
 919
 920         return 0;
 921
 922 out_slots:
 923         kfree(slots);
 924 out_free:
 925         kvm_free_physmem_slot(&new, &old);
 926 out:
 927         return r;
 928 }
 929 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 930
 931 int kvm_set_memory_region(struct kvm *kvm,
 932                           struct kvm_userspace_memory_region *mem)
 933 {
 934         int r;
 935
 936         mutex_lock(&kvm->slots_lock);
 937         r = __kvm_set_memory_region(kvm, mem);
 938         mutex_unlock(&kvm->slots_lock);
 939         return r;
 940 }
 941 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 942
 943 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 944                                    struct kvm_userspace_memory_region *mem)
 945 {
 946         if (mem->slot >= KVM_USER_MEM_SLOTS)
 947                 return -EINVAL;
 948         return kvm_set_memory_region(kvm, mem);
 949 }
 950
 951 int kvm_get_dirty_log(struct kvm *kvm,
 952                         struct kvm_dirty_log *log, int *is_dirty)
 953 {
 954         struct kvm_memory_slot *memslot;
 955         int r, i;
 956         unsigned long n;
 957         unsigned long any = 0;
 958
 959         r = -EINVAL;
 960         if (log->slot >= KVM_USER_MEM_SLOTS)
 961                 goto out;
 962
 963         memslot = id_to_memslot(kvm->memslots, log->slot);
 964         r = -ENOENT;
 965         if (!memslot->dirty_bitmap)
 966                 goto out;
 967
 968         n = kvm_dirty_bitmap_bytes(memslot);
 969
 970         for (i = 0; !any && i < n/sizeof(long); ++i)
 971                 any = memslot->dirty_bitmap[i];
 972
 973         r = -EFAULT;
 974         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 975                 goto out;
 976
 977         if (any)
 978                 *is_dirty = 1;
 979
 980         r = 0;
 981 out:
 982         return r;
 983 }
 984
 985 bool kvm_largepages_enabled(void)
 986 {
 987         return largepages_enabled;
 988 }
 989
 990 void kvm_disable_largepages(void)
 991 {
 992         largepages_enabled = false;
 993 }
 994 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 995
 996 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 997 {
 998         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 999 }
1000 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1001
1002 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1003 {
1004         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1005
1006         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1007               memslot->flags & KVM_MEMSLOT_INVALID)
1008                 return 0;
1009
1010         return 1;
1011 }
1012 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1013
1014 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1015 {
1016         struct vm_area_struct *vma;
1017         unsigned long addr, size;
1018
1019         size = PAGE_SIZE;
1020
1021         addr = gfn_to_hva(kvm, gfn);
1022         if (kvm_is_error_hva(addr))
1023                 return PAGE_SIZE;
1024
1025         down_read(&current->mm->mmap_sem);
1026         vma = find_vma(current->mm, addr);
1027         if (!vma)
1028                 goto out;
1029
1030         size = vma_kernel_pagesize(vma);
1031
1032 out:
1033         up_read(&current->mm->mmap_sem);
1034
1035         return size;
1036 }
1037
1038 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1039 {
1040         return slot->flags & KVM_MEM_READONLY;
1041 }
1042
1043 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1044                                        gfn_t *nr_pages, bool write)
1045 {
1046         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1047                 return KVM_HVA_ERR_BAD;
1048
1049         if (memslot_is_readonly(slot) && write)
1050                 return KVM_HVA_ERR_RO_BAD;
1051
1052         if (nr_pages)
1053                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1054
1055         return __gfn_to_hva_memslot(slot, gfn);
1056 }
1057
1058 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1059                                      gfn_t *nr_pages)
1060 {
1061         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1062 }
1063
1064 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1065                                  gfn_t gfn)
1066 {
1067         return gfn_to_hva_many(slot, gfn, NULL);
1068 }
1069 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1070
1071 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1072 {
1073         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1074 }
1075 EXPORT_SYMBOL_GPL(gfn_to_hva);
1076
1077 /*
1078  * The hva returned by this function is only allowed to be read.
1079  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1080  */
1081 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1082 {
1083         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1084 }
1085
1086 static int kvm_read_hva(void *data, void __user *hva, int len)
1087 {
1088         return __copy_from_user(data, hva, len);
1089 }
1090
1091 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1092 {
1093         return __copy_from_user_inatomic(data, hva, len);
1094 }
1095
1096 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1097         unsigned long start, int write, struct page **page)
1098 {
1099         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1100
1101         if (write)
1102                 flags |= FOLL_WRITE;
1103
1104         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1105 }
1106
1107 static inline int check_user_page_hwpoison(unsigned long addr)
1108 {
1109         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1110
1111         rc = __get_user_pages(current, current->mm, addr, 1,
1112                               flags, NULL, NULL, NULL);
1113         return rc == -EHWPOISON;
1114 }
1115
1116 /*
1117  * The atomic path to get the writable pfn which will be stored in @pfn,
1118  * true indicates success, otherwise false is returned.
1119  */
1120 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1121                             bool write_fault, bool *writable, pfn_t *pfn)
1122 {
1123         struct page *page[1];
1124         int npages;
1125
1126         if (!(async || atomic))
1127                 return false;
1128
1129         /*
1130          * Fast pin a writable pfn only if it is a write fault request
1131          * or the caller allows to map a writable pfn for a read fault
1132          * request.
1133          */
1134         if (!(write_fault || writable))
1135                 return false;
1136
1137         npages = __get_user_pages_fast(addr, 1, 1, page);
1138         if (npages == 1) {
1139                 *pfn = page_to_pfn(page[0]);
1140
1141                 if (writable)
1142                         *writable = true;
1143                 return true;
1144         }
1145
1146         return false;
1147 }
1148
1149 /*
1150  * The slow path to get the pfn of the specified host virtual address,
1151  * 1 indicates success, -errno is returned if error is detected.
1152  */
1153 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1154                            bool *writable, pfn_t *pfn)
1155 {
1156         struct page *page[1];
1157         int npages = 0;
1158
1159         might_sleep();
1160
1161         if (writable)
1162                 *writable = write_fault;
1163
1164         if (async) {
1165                 down_read(&current->mm->mmap_sem);
1166                 npages = get_user_page_nowait(current, current->mm,
1167                                               addr, write_fault, page);
1168                 up_read(&current->mm->mmap_sem);
1169         } else
1170                 npages = get_user_pages_fast(addr, 1, write_fault,
1171                                              page);
1172         if (npages != 1)
1173                 return npages;
1174
1175         /* map read fault as writable if possible */
1176         if (unlikely(!write_fault) && writable) {
1177                 struct page *wpage[1];
1178
1179                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1180                 if (npages == 1) {
1181                         *writable = true;
1182                         put_page(page[0]);
1183                         page[0] = wpage[0];
1184                 }
1185
1186                 npages = 1;
1187         }
1188         *pfn = page_to_pfn(page[0]);
1189         return npages;
1190 }
1191
1192 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1193 {
1194         if (unlikely(!(vma->vm_flags & VM_READ)))
1195                 return false;
1196
1197         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1198                 return false;
1199
1200         return true;
1201 }
1202
1203 /*
1204  * Pin guest page in memory and return its pfn.
1205  * @addr: host virtual address which maps memory to the guest
1206  * @atomic: whether this function can sleep
1207  * @async: whether this function need to wait IO complete if the
1208  *         host page is not in the memory
1209  * @write_fault: whether we should get a writable host page
1210  * @writable: whether it allows to map a writable host page for !@write_fault
1211  *
1212  * The function will map a writable host page for these two cases:
1213  * 1): @write_fault = true
1214  * 2): @write_fault = false && @writable, @writable will tell the caller
1215  *     whether the mapping is writable.
1216  */
1217 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1218                         bool write_fault, bool *writable)
1219 {
1220         struct vm_area_struct *vma;
1221         pfn_t pfn = 0;
1222         int npages;
1223
1224         /* we can do it either atomically or asynchronously, not both */
1225         BUG_ON(atomic && async);
1226
1227         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1228                 return pfn;
1229
1230         if (atomic)
1231                 return KVM_PFN_ERR_FAULT;
1232
1233         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1234         if (npages == 1)
1235                 return pfn;
1236
1237         down_read(&current->mm->mmap_sem);
1238         if (npages == -EHWPOISON ||
1239               (!async && check_user_page_hwpoison(addr))) {
1240                 pfn = KVM_PFN_ERR_HWPOISON;
1241                 goto exit;
1242         }
1243
1244         vma = find_vma_intersection(current->mm, addr, addr + 1);
1245
1246         if (vma == NULL)
1247                 pfn = KVM_PFN_ERR_FAULT;
1248         else if ((vma->vm_flags & VM_PFNMAP)) {
1249                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1250                         vma->vm_pgoff;
1251                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1252         } else {
1253                 if (async && vma_is_valid(vma, write_fault))
1254                         *async = true;
1255                 pfn = KVM_PFN_ERR_FAULT;
1256         }
1257 exit:
1258         up_read(&current->mm->mmap_sem);
1259         return pfn;
1260 }
1261
1262 static pfn_t
1263 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1264                      bool *async, bool write_fault, bool *writable)
1265 {
1266         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1267
1268         if (addr == KVM_HVA_ERR_RO_BAD)
1269                 return KVM_PFN_ERR_RO_FAULT;
1270
1271         if (kvm_is_error_hva(addr))
1272                 return KVM_PFN_NOSLOT;
1273
1274         /* Do not map writable pfn in the readonly memslot. */
1275         if (writable && memslot_is_readonly(slot)) {
1276                 *writable = false;
1277                 writable = NULL;
1278         }
1279
1280         return hva_to_pfn(addr, atomic, async, write_fault,
1281                           writable);
1282 }
1283
1284 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1285                           bool write_fault, bool *writable)
1286 {
1287         struct kvm_memory_slot *slot;
1288
1289         if (async)
1290                 *async = false;
1291
1292         slot = gfn_to_memslot(kvm, gfn);
1293
1294         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1295                                     writable);
1296 }
1297
1298 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1299 {
1300         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1301 }
1302 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1303
1304 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1305                        bool write_fault, bool *writable)
1306 {
1307         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1308 }
1309 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1310
1311 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1312 {
1313         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1314 }
1315 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1316
1317 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1318                       bool *writable)
1319 {
1320         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1321 }
1322 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1323
1324 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1325 {
1326         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1327 }
1328
1329 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1330 {
1331         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1332 }
1333 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1334
1335 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1336                                                                   int nr_pages)
1337 {
1338         unsigned long addr;
1339         gfn_t entry;
1340
1341         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1342         if (kvm_is_error_hva(addr))
1343                 return -1;
1344
1345         if (entry < nr_pages)
1346                 return 0;
1347
1348         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1349 }
1350 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1351
1352 static struct page *kvm_pfn_to_page(pfn_t pfn)
1353 {
1354         if (is_error_noslot_pfn(pfn))
1355                 return KVM_ERR_PTR_BAD_PAGE;
1356
1357         if (kvm_is_mmio_pfn(pfn)) {
1358                 WARN_ON(1);
1359                 return KVM_ERR_PTR_BAD_PAGE;
1360         }
1361
1362         return pfn_to_page(pfn);
1363 }
1364
1365 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1366 {
1367         pfn_t pfn;
1368
1369         pfn = gfn_to_pfn(kvm, gfn);
1370
1371         return kvm_pfn_to_page(pfn);
1372 }
1373
1374 EXPORT_SYMBOL_GPL(gfn_to_page);
1375
1376 void kvm_release_page_clean(struct page *page)
1377 {
1378         WARN_ON(is_error_page(page));
1379
1380         kvm_release_pfn_clean(page_to_pfn(page));
1381 }
1382 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1383
1384 void kvm_release_pfn_clean(pfn_t pfn)
1385 {
1386         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1387                 put_page(pfn_to_page(pfn));
1388 }
1389 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1390
1391 void kvm_release_page_dirty(struct page *page)
1392 {
1393         WARN_ON(is_error_page(page));
1394
1395         kvm_release_pfn_dirty(page_to_pfn(page));
1396 }
1397 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1398
1399 void kvm_release_pfn_dirty(pfn_t pfn)
1400 {
1401         kvm_set_pfn_dirty(pfn);
1402         kvm_release_pfn_clean(pfn);
1403 }
1404 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1405
1406 void kvm_set_page_dirty(struct page *page)
1407 {
1408         kvm_set_pfn_dirty(page_to_pfn(page));
1409 }
1410 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1411
1412 void kvm_set_pfn_dirty(pfn_t pfn)
1413 {
1414         if (!kvm_is_mmio_pfn(pfn)) {
1415                 struct page *page = pfn_to_page(pfn);
1416                 if (!PageReserved(page))
1417                         SetPageDirty(page);
1418         }
1419 }
1420 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1421
1422 void kvm_set_pfn_accessed(pfn_t pfn)
1423 {
1424         if (!kvm_is_mmio_pfn(pfn))
1425                 mark_page_accessed(pfn_to_page(pfn));
1426 }
1427 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1428
1429 void kvm_get_pfn(pfn_t pfn)
1430 {
1431         if (!kvm_is_mmio_pfn(pfn))
1432                 get_page(pfn_to_page(pfn));
1433 }
1434 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1435
1436 static int next_segment(unsigned long len, int offset)
1437 {
1438         if (len > PAGE_SIZE - offset)
1439                 return PAGE_SIZE - offset;
1440         else
1441                 return len;
1442 }
1443
1444 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1445                         int len)
1446 {
1447         int r;
1448         unsigned long addr;
1449
1450         addr = gfn_to_hva_read(kvm, gfn);
1451         if (kvm_is_error_hva(addr))
1452                 return -EFAULT;
1453         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1454         if (r)
1455                 return -EFAULT;
1456         return 0;
1457 }
1458 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1459
1460 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1461 {
1462         gfn_t gfn = gpa >> PAGE_SHIFT;
1463         int seg;
1464         int offset = offset_in_page(gpa);
1465         int ret;
1466
1467         while ((seg = next_segment(len, offset)) != 0) {
1468                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1469                 if (ret < 0)
1470                         return ret;
1471                 offset = 0;
1472                 len -= seg;
1473                 data += seg;
1474                 ++gfn;
1475         }
1476         return 0;
1477 }
1478 EXPORT_SYMBOL_GPL(kvm_read_guest);
1479
1480 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1481                           unsigned long len)
1482 {
1483         int r;
1484         unsigned long addr;
1485         gfn_t gfn = gpa >> PAGE_SHIFT;
1486         int offset = offset_in_page(gpa);
1487
1488         addr = gfn_to_hva_read(kvm, gfn);
1489         if (kvm_is_error_hva(addr))
1490                 return -EFAULT;
1491         pagefault_disable();
1492         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1493         pagefault_enable();
1494         if (r)
1495                 return -EFAULT;
1496         return 0;
1497 }
1498 EXPORT_SYMBOL(kvm_read_guest_atomic);
1499
1500 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1501                          int offset, int len)
1502 {
1503         int r;
1504         unsigned long addr;
1505
1506         addr = gfn_to_hva(kvm, gfn);
1507         if (kvm_is_error_hva(addr))
1508                 return -EFAULT;
1509         r = __copy_to_user((void __user *)addr + offset, data, len);
1510         if (r)
1511                 return -EFAULT;
1512         mark_page_dirty(kvm, gfn);
1513         return 0;
1514 }
1515 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1516
1517 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1518                     unsigned long len)
1519 {
1520         gfn_t gfn = gpa >> PAGE_SHIFT;
1521         int seg;
1522         int offset = offset_in_page(gpa);
1523         int ret;
1524
1525         while ((seg = next_segment(len, offset)) != 0) {
1526                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1527                 if (ret < 0)
1528                         return ret;
1529                 offset = 0;
1530                 len -= seg;
1531                 data += seg;
1532                 ++gfn;
1533         }
1534         return 0;
1535 }
1536
1537 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1538                               gpa_t gpa)
1539 {
1540         struct kvm_memslots *slots = kvm_memslots(kvm);
1541         int offset = offset_in_page(gpa);
1542         gfn_t gfn = gpa >> PAGE_SHIFT;
1543
1544         ghc->gpa = gpa;
1545         ghc->generation = slots->generation;
1546         ghc->memslot = gfn_to_memslot(kvm, gfn);
1547         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1548         if (!kvm_is_error_hva(ghc->hva))
1549                 ghc->hva += offset;
1550         else
1551                 return -EFAULT;
1552
1553         return 0;
1554 }
1555 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1556
1557 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1558                            void *data, unsigned long len)
1559 {
1560         struct kvm_memslots *slots = kvm_memslots(kvm);
1561         int r;
1562
1563         if (slots->generation != ghc->generation)
1564                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1565
1566         if (kvm_is_error_hva(ghc->hva))
1567                 return -EFAULT;
1568
1569         r = __copy_to_user((void __user *)ghc->hva, data, len);
1570         if (r)
1571                 return -EFAULT;
1572         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1573
1574         return 0;
1575 }
1576 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1577
1578 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1579                            void *data, unsigned long len)
1580 {
1581         struct kvm_memslots *slots = kvm_memslots(kvm);
1582         int r;
1583
1584         if (slots->generation != ghc->generation)
1585                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1586
1587         if (kvm_is_error_hva(ghc->hva))
1588                 return -EFAULT;
1589
1590         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1591         if (r)
1592                 return -EFAULT;
1593
1594         return 0;
1595 }
1596 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1597
1598 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1599 {
1600         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1601                                     offset, len);
1602 }
1603 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1604
1605 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1606 {
1607         gfn_t gfn = gpa >> PAGE_SHIFT;
1608         int seg;
1609         int offset = offset_in_page(gpa);
1610         int ret;
1611
1612         while ((seg = next_segment(len, offset)) != 0) {
1613                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1614                 if (ret < 0)
1615                         return ret;
1616                 offset = 0;
1617                 len -= seg;
1618                 ++gfn;
1619         }
1620         return 0;
1621 }
1622 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1623
1624 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1625                              gfn_t gfn)
1626 {
1627         if (memslot && memslot->dirty_bitmap) {
1628                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1629
1630                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1631         }
1632 }
1633
1634 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1635 {
1636         struct kvm_memory_slot *memslot;
1637
1638         memslot = gfn_to_memslot(kvm, gfn);
1639         mark_page_dirty_in_slot(kvm, memslot, gfn);
1640 }
1641
1642 /*
1643  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1644  */
1645 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1646 {
1647         DEFINE_WAIT(wait);
1648
1649         for (;;) {
1650                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1651
1652                 if (kvm_arch_vcpu_runnable(vcpu)) {
1653                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1654                         break;
1655                 }
1656                 if (kvm_cpu_has_pending_timer(vcpu))
1657                         break;
1658                 if (signal_pending(current))
1659                         break;
1660
1661                 schedule();
1662         }
1663
1664         finish_wait(&vcpu->wq, &wait);
1665 }
1666
1667 #ifndef CONFIG_S390
1668 /*
1669  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1670  */
1671 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1672 {
1673         int me;
1674         int cpu = vcpu->cpu;
1675         wait_queue_head_t *wqp;
1676
1677         wqp = kvm_arch_vcpu_wq(vcpu);
1678         if (waitqueue_active(wqp)) {
1679                 wake_up_interruptible(wqp);
1680                 ++vcpu->stat.halt_wakeup;
1681         }
1682
1683         me = get_cpu();
1684         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1685                 if (kvm_arch_vcpu_should_kick(vcpu))
1686                         smp_send_reschedule(cpu);
1687         put_cpu();
1688 }
1689 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1690 #endif /* !CONFIG_S390 */
1691
1692 void kvm_resched(struct kvm_vcpu *vcpu)
1693 {
1694         if (!need_resched())
1695                 return;
1696         cond_resched();
1697 }
1698 EXPORT_SYMBOL_GPL(kvm_resched);
1699
1700 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1701 {
1702         struct pid *pid;
1703         struct task_struct *task = NULL;
1704         bool ret = false;
1705
1706         rcu_read_lock();
1707         pid = rcu_dereference(target->pid);
1708         if (pid)
1709                 task = get_pid_task(target->pid, PIDTYPE_PID);
1710         rcu_read_unlock();
1711         if (!task)
1712                 return ret;
1713         if (task->flags & PF_VCPU) {
1714                 put_task_struct(task);
1715                 return ret;
1716         }
1717         ret = yield_to(task, 1);
1718         put_task_struct(task);
1719
1720         return ret;
1721 }
1722 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1723
1724 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1725 /*
1726  * Helper that checks whether a VCPU is eligible for directed yield.
1727  * Most eligible candidate to yield is decided by following heuristics:
1728  *
1729  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1730  *  (preempted lock holder), indicated by @in_spin_loop.
1731  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1732  *
1733  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1734  *  chance last time (mostly it has become eligible now since we have probably
1735  *  yielded to lockholder in last iteration. This is done by toggling
1736  *  @dy_eligible each time a VCPU checked for eligibility.)
1737  *
1738  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1739  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1740  *  burning. Giving priority for a potential lock-holder increases lock
1741  *  progress.
1742  *
1743  *  Since algorithm is based on heuristics, accessing another VCPU data without
1744  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1745  *  and continue with next VCPU and so on.
1746  */
1747 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1748 {
1749         bool eligible;
1750
1751         eligible = !vcpu->spin_loop.in_spin_loop ||
1752                         (vcpu->spin_loop.in_spin_loop &&
1753                          vcpu->spin_loop.dy_eligible);
1754
1755         if (vcpu->spin_loop.in_spin_loop)
1756                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1757
1758         return eligible;
1759 }
1760 #endif
1761
1762 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1763 {
1764         struct kvm *kvm = me->kvm;
1765         struct kvm_vcpu *vcpu;
1766         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1767         int yielded = 0;
1768         int try = 3;
1769         int pass;
1770         int i;
1771
1772         kvm_vcpu_set_in_spin_loop(me, true);
1773         /*
1774          * We boost the priority of a VCPU that is runnable but not
1775          * currently running, because it got preempted by something
1776          * else and called schedule in __vcpu_run.  Hopefully that
1777          * VCPU is holding the lock that we need and will release it.
1778          * We approximate round-robin by starting at the last boosted VCPU.
1779          */
1780         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1781                 kvm_for_each_vcpu(i, vcpu, kvm) {
1782                         if (!pass && i <= last_boosted_vcpu) {
1783                                 i = last_boosted_vcpu;
1784                                 continue;
1785                         } else if (pass && i > last_boosted_vcpu)
1786                                 break;
1787                         if (!ACCESS_ONCE(vcpu->preempted))
1788                                 continue;
1789                         if (vcpu == me)
1790                                 continue;
1791                         if (waitqueue_active(&vcpu->wq))
1792                                 continue;
1793                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1794                                 continue;
1795
1796                         yielded = kvm_vcpu_yield_to(vcpu);
1797                         if (yielded > 0) {
1798                                 kvm->last_boosted_vcpu = i;
1799                                 break;
1800                         } else if (yielded < 0) {
1801                                 try--;
1802                                 if (!try)
1803                                         break;
1804                         }
1805                 }
1806         }
1807         kvm_vcpu_set_in_spin_loop(me, false);
1808
1809         /* Ensure vcpu is not eligible during next spinloop */
1810         kvm_vcpu_set_dy_eligible(me, false);
1811 }
1812 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1813
1814 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1815 {
1816         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1817         struct page *page;
1818
1819         if (vmf->pgoff == 0)
1820                 page = virt_to_page(vcpu->run);
1821 #ifdef CONFIG_X86
1822         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1823                 page = virt_to_page(vcpu->arch.pio_data);
1824 #endif
1825 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1826         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1827                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1828 #endif
1829         else
1830                 return kvm_arch_vcpu_fault(vcpu, vmf);
1831         get_page(page);
1832         vmf->page = page;
1833         return 0;
1834 }
1835
1836 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1837         .fault = kvm_vcpu_fault,
1838 };
1839
1840 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1841 {
1842         vma->vm_ops = &kvm_vcpu_vm_ops;
1843         return 0;
1844 }
1845
1846 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1847 {
1848         struct kvm_vcpu *vcpu = filp->private_data;
1849
1850         kvm_put_kvm(vcpu->kvm);
1851         return 0;
1852 }
1853
1854 static struct file_operations kvm_vcpu_fops = {
1855         .release        = kvm_vcpu_release,
1856         .unlocked_ioctl = kvm_vcpu_ioctl,
1857 #ifdef CONFIG_COMPAT
1858         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1859 #endif
1860         .mmap           = kvm_vcpu_mmap,
1861         .llseek         = noop_llseek,
1862 };
1863
1864 /*
1865  * Allocates an inode for the vcpu.
1866  */
1867 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1868 {
1869         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1870 }
1871
1872 /*
1873  * Creates some virtual cpus.  Good luck creating more than one.
1874  */
1875 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1876 {
1877         int r;
1878         struct kvm_vcpu *vcpu, *v;
1879
1880         vcpu = kvm_arch_vcpu_create(kvm, id);
1881         if (IS_ERR(vcpu))
1882                 return PTR_ERR(vcpu);
1883
1884         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1885
1886         r = kvm_arch_vcpu_setup(vcpu);
1887         if (r)
1888                 goto vcpu_destroy;
1889
1890         mutex_lock(&kvm->lock);
1891         if (!kvm_vcpu_compatible(vcpu)) {
1892                 r = -EINVAL;
1893                 goto unlock_vcpu_destroy;
1894         }
1895         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1896                 r = -EINVAL;
1897                 goto unlock_vcpu_destroy;
1898         }
1899
1900         kvm_for_each_vcpu(r, v, kvm)
1901                 if (v->vcpu_id == id) {
1902                         r = -EEXIST;
1903                         goto unlock_vcpu_destroy;
1904                 }
1905
1906         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1907
1908         /* Now it's all set up, let userspace reach it */
1909         kvm_get_kvm(kvm);
1910         r = create_vcpu_fd(vcpu);
1911         if (r < 0) {
1912                 kvm_put_kvm(kvm);
1913                 goto unlock_vcpu_destroy;
1914         }
1915
1916         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1917         smp_wmb();
1918         atomic_inc(&kvm->online_vcpus);
1919
1920         mutex_unlock(&kvm->lock);
1921         kvm_arch_vcpu_postcreate(vcpu);
1922         return r;
1923
1924 unlock_vcpu_destroy:
1925         mutex_unlock(&kvm->lock);
1926 vcpu_destroy:
1927         kvm_arch_vcpu_destroy(vcpu);
1928         return r;
1929 }
1930
1931 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1932 {
1933         if (sigset) {
1934                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1935                 vcpu->sigset_active = 1;
1936                 vcpu->sigset = *sigset;
1937         } else
1938                 vcpu->sigset_active = 0;
1939         return 0;
1940 }
1941
1942 static long kvm_vcpu_ioctl(struct file *filp,
1943                            unsigned int ioctl, unsigned long arg)
1944 {
1945         struct kvm_vcpu *vcpu = filp->private_data;
1946         void __user *argp = (void __user *)arg;
1947         int r;
1948         struct kvm_fpu *fpu = NULL;
1949         struct kvm_sregs *kvm_sregs = NULL;
1950
1951         if (vcpu->kvm->mm != current->mm)
1952                 return -EIO;
1953
1954 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1955         /*
1956          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1957          * so vcpu_load() would break it.
1958          */
1959         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1960                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1961 #endif
1962
1963
1964         r = vcpu_load(vcpu);
1965         if (r)
1966                 return r;
1967         switch (ioctl) {
1968         case KVM_RUN:
1969                 r = -EINVAL;
1970                 if (arg)
1971                         goto out;
1972                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1973                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1974                 break;
1975         case KVM_GET_REGS: {
1976                 struct kvm_regs *kvm_regs;
1977
1978                 r = -ENOMEM;
1979                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1980                 if (!kvm_regs)
1981                         goto out;
1982                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1983                 if (r)
1984                         goto out_free1;
1985                 r = -EFAULT;
1986                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1987                         goto out_free1;
1988                 r = 0;
1989 out_free1:
1990                 kfree(kvm_regs);
1991                 break;
1992         }
1993         case KVM_SET_REGS: {
1994                 struct kvm_regs *kvm_regs;
1995
1996                 r = -ENOMEM;
1997                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1998                 if (IS_ERR(kvm_regs)) {
1999                         r = PTR_ERR(kvm_regs);
2000                         goto out;
2001                 }
2002                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2003                 kfree(kvm_regs);
2004                 break;
2005         }
2006         case KVM_GET_SREGS: {
2007                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2008                 r = -ENOMEM;
2009                 if (!kvm_sregs)
2010                         goto out;
2011                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2012                 if (r)
2013                         goto out;
2014                 r = -EFAULT;
2015                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2016                         goto out;
2017                 r = 0;
2018                 break;
2019         }
2020         case KVM_SET_SREGS: {
2021                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2022                 if (IS_ERR(kvm_sregs)) {
2023                         r = PTR_ERR(kvm_sregs);
2024                         kvm_sregs = NULL;
2025                         goto out;
2026                 }
2027                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2028                 break;
2029         }
2030         case KVM_GET_MP_STATE: {
2031                 struct kvm_mp_state mp_state;
2032
2033                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2034                 if (r)
2035                         goto out;
2036                 r = -EFAULT;
2037                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2038                         goto out;
2039                 r = 0;
2040                 break;
2041         }
2042         case KVM_SET_MP_STATE: {
2043                 struct kvm_mp_state mp_state;
2044
2045                 r = -EFAULT;
2046                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2047                         goto out;
2048                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2049                 break;
2050         }
2051         case KVM_TRANSLATE: {
2052                 struct kvm_translation tr;
2053
2054                 r = -EFAULT;
2055                 if (copy_from_user(&tr, argp, sizeof tr))
2056                         goto out;
2057                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2058                 if (r)
2059                         goto out;
2060                 r = -EFAULT;
2061                 if (copy_to_user(argp, &tr, sizeof tr))
2062                         goto out;
2063                 r = 0;
2064                 break;
2065         }
2066         case KVM_SET_GUEST_DEBUG: {
2067                 struct kvm_guest_debug dbg;
2068
2069                 r = -EFAULT;
2070                 if (copy_from_user(&dbg, argp, sizeof dbg))
2071                         goto out;
2072                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2073                 break;
2074         }
2075         case KVM_SET_SIGNAL_MASK: {
2076                 struct kvm_signal_mask __user *sigmask_arg = argp;
2077                 struct kvm_signal_mask kvm_sigmask;
2078                 sigset_t sigset, *p;
2079
2080                 p = NULL;
2081                 if (argp) {
2082                         r = -EFAULT;
2083                         if (copy_from_user(&kvm_sigmask, argp,
2084                                            sizeof kvm_sigmask))
2085                                 goto out;
2086                         r = -EINVAL;
2087                         if (kvm_sigmask.len != sizeof sigset)
2088                                 goto out;
2089                         r = -EFAULT;
2090                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2091                                            sizeof sigset))
2092                                 goto out;
2093                         p = &sigset;
2094                 }
2095                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2096                 break;
2097         }
2098         case KVM_GET_FPU: {
2099                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2100                 r = -ENOMEM;
2101                 if (!fpu)
2102                         goto out;
2103                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2104                 if (r)
2105                         goto out;
2106                 r = -EFAULT;
2107                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2108                         goto out;
2109                 r = 0;
2110                 break;
2111         }
2112         case KVM_SET_FPU: {
2113                 fpu = memdup_user(argp, sizeof(*fpu));
2114                 if (IS_ERR(fpu)) {
2115                         r = PTR_ERR(fpu);
2116                         fpu = NULL;
2117                         goto out;
2118                 }
2119                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2120                 break;
2121         }
2122         default:
2123                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2124         }
2125 out:
2126         vcpu_put(vcpu);
2127         kfree(fpu);
2128         kfree(kvm_sregs);
2129         return r;
2130 }
2131
2132 #ifdef CONFIG_COMPAT
2133 static long kvm_vcpu_compat_ioctl(struct file *filp,
2134                                   unsigned int ioctl, unsigned long arg)
2135 {
2136         struct kvm_vcpu *vcpu = filp->private_data;
2137         void __user *argp = compat_ptr(arg);
2138         int r;
2139
2140         if (vcpu->kvm->mm != current->mm)
2141                 return -EIO;
2142
2143         switch (ioctl) {
2144         case KVM_SET_SIGNAL_MASK: {
2145                 struct kvm_signal_mask __user *sigmask_arg = argp;
2146                 struct kvm_signal_mask kvm_sigmask;
2147                 compat_sigset_t csigset;
2148                 sigset_t sigset;
2149
2150                 if (argp) {
2151                         r = -EFAULT;
2152                         if (copy_from_user(&kvm_sigmask, argp,
2153                                            sizeof kvm_sigmask))
2154                                 goto out;
2155                         r = -EINVAL;
2156                         if (kvm_sigmask.len != sizeof csigset)
2157                                 goto out;
2158                         r = -EFAULT;
2159                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2160                                            sizeof csigset))
2161                                 goto out;
2162                         sigset_from_compat(&sigset, &csigset);
2163                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2164                 } else
2165                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2166                 break;
2167         }
2168         default:
2169                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2170         }
2171
2172 out:
2173         return r;
2174 }
2175 #endif
2176
2177 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2178                                  int (*accessor)(struct kvm_device *dev,
2179                                                  struct kvm_device_attr *attr),
2180                                  unsigned long arg)
2181 {
2182         struct kvm_device_attr attr;
2183
2184         if (!accessor)
2185                 return -EPERM;
2186
2187         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2188                 return -EFAULT;
2189
2190         return accessor(dev, &attr);
2191 }
2192
2193 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2194                              unsigned long arg)
2195 {
2196         struct kvm_device *dev = filp->private_data;
2197
2198         switch (ioctl) {
2199         case KVM_SET_DEVICE_ATTR:
2200                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2201         case KVM_GET_DEVICE_ATTR:
2202                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2203         case KVM_HAS_DEVICE_ATTR:
2204                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2205         default:
2206                 if (dev->ops->ioctl)
2207                         return dev->ops->ioctl(dev, ioctl, arg);
2208
2209                 return -ENOTTY;
2210         }
2211 }
2212
2213 static int kvm_device_release(struct inode *inode, struct file *filp)
2214 {
2215         struct kvm_device *dev = filp->private_data;
2216         struct kvm *kvm = dev->kvm;
2217
2218         kvm_put_kvm(kvm);
2219         return 0;
2220 }
2221
2222 static const struct file_operations kvm_device_fops = {
2223         .unlocked_ioctl = kvm_device_ioctl,
2224         .release = kvm_device_release,
2225 };
2226
2227 struct kvm_device *kvm_device_from_filp(struct file *filp)
2228 {
2229         if (filp->f_op != &kvm_device_fops)
2230                 return NULL;
2231
2232         return filp->private_data;
2233 }
2234
2235 static int kvm_ioctl_create_device(struct kvm *kvm,
2236                                    struct kvm_create_device *cd)
2237 {
2238         struct kvm_device_ops *ops = NULL;
2239         struct kvm_device *dev;
2240         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2241         int ret;
2242
2243         switch (cd->type) {
2244 #ifdef CONFIG_KVM_MPIC
2245         case KVM_DEV_TYPE_FSL_MPIC_20:
2246         case KVM_DEV_TYPE_FSL_MPIC_42:
2247                 ops = &kvm_mpic_ops;
2248                 break;
2249 #endif
2250 #ifdef CONFIG_KVM_XICS
2251         case KVM_DEV_TYPE_XICS:
2252                 ops = &kvm_xics_ops;
2253                 break;
2254 #endif
2255         default:
2256                 return -ENODEV;
2257         }
2258
2259         if (test)
2260                 return 0;
2261
2262         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2263         if (!dev)
2264                 return -ENOMEM;
2265
2266         dev->ops = ops;
2267         dev->kvm = kvm;
2268
2269         ret = ops->create(dev, cd->type);
2270         if (ret < 0) {
2271                 kfree(dev);
2272                 return ret;
2273         }
2274
2275         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR);
2276         if (ret < 0) {
2277                 ops->destroy(dev);
2278                 return ret;
2279         }
2280
2281         list_add(&dev->vm_node, &kvm->devices);
2282         kvm_get_kvm(kvm);
2283         cd->fd = ret;
2284         return 0;
2285 }
2286
2287 static long kvm_vm_ioctl(struct file *filp,
2288                            unsigned int ioctl, unsigned long arg)
2289 {
2290         struct kvm *kvm = filp->private_data;
2291         void __user *argp = (void __user *)arg;
2292         int r;
2293
2294         if (kvm->mm != current->mm)
2295                 return -EIO;
2296         switch (ioctl) {
2297         case KVM_CREATE_VCPU:
2298                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2299                 break;
2300         case KVM_SET_USER_MEMORY_REGION: {
2301                 struct kvm_userspace_memory_region kvm_userspace_mem;
2302
2303                 r = -EFAULT;
2304                 if (copy_from_user(&kvm_userspace_mem, argp,
2305                                                 sizeof kvm_userspace_mem))
2306                         goto out;
2307
2308                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2309                 break;
2310         }
2311         case KVM_GET_DIRTY_LOG: {
2312                 struct kvm_dirty_log log;
2313
2314                 r = -EFAULT;
2315                 if (copy_from_user(&log, argp, sizeof log))
2316                         goto out;
2317                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2318                 break;
2319         }
2320 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2321         case KVM_REGISTER_COALESCED_MMIO: {
2322                 struct kvm_coalesced_mmio_zone zone;
2323                 r = -EFAULT;
2324                 if (copy_from_user(&zone, argp, sizeof zone))
2325                         goto out;
2326                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2327                 break;
2328         }
2329         case KVM_UNREGISTER_COALESCED_MMIO: {
2330                 struct kvm_coalesced_mmio_zone zone;
2331                 r = -EFAULT;
2332                 if (copy_from_user(&zone, argp, sizeof zone))
2333                         goto out;
2334                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2335                 break;
2336         }
2337 #endif
2338         case KVM_IRQFD: {
2339                 struct kvm_irqfd data;
2340
2341                 r = -EFAULT;
2342                 if (copy_from_user(&data, argp, sizeof data))
2343                         goto out;
2344                 r = kvm_irqfd(kvm, &data);
2345                 break;
2346         }
2347         case KVM_IOEVENTFD: {
2348                 struct kvm_ioeventfd data;
2349
2350                 r = -EFAULT;
2351                 if (copy_from_user(&data, argp, sizeof data))
2352                         goto out;
2353                 r = kvm_ioeventfd(kvm, &data);
2354                 break;
2355         }
2356 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2357         case KVM_SET_BOOT_CPU_ID:
2358                 r = 0;
2359                 mutex_lock(&kvm->lock);
2360                 if (atomic_read(&kvm->online_vcpus) != 0)
2361                         r = -EBUSY;
2362                 else
2363                         kvm->bsp_vcpu_id = arg;
2364                 mutex_unlock(&kvm->lock);
2365                 break;
2366 #endif
2367 #ifdef CONFIG_HAVE_KVM_MSI
2368         case KVM_SIGNAL_MSI: {
2369                 struct kvm_msi msi;
2370
2371                 r = -EFAULT;
2372                 if (copy_from_user(&msi, argp, sizeof msi))
2373                         goto out;
2374                 r = kvm_send_userspace_msi(kvm, &msi);
2375                 break;
2376         }
2377 #endif
2378 #ifdef __KVM_HAVE_IRQ_LINE
2379         case KVM_IRQ_LINE_STATUS:
2380         case KVM_IRQ_LINE: {
2381                 struct kvm_irq_level irq_event;
2382
2383                 r = -EFAULT;
2384                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2385                         goto out;
2386
2387                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2388                                         ioctl == KVM_IRQ_LINE_STATUS);
2389                 if (r)
2390                         goto out;
2391
2392                 r = -EFAULT;
2393                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2394                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2395                                 goto out;
2396                 }
2397
2398                 r = 0;
2399                 break;
2400         }
2401 #endif
2402 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2403         case KVM_SET_GSI_ROUTING: {
2404                 struct kvm_irq_routing routing;
2405                 struct kvm_irq_routing __user *urouting;
2406                 struct kvm_irq_routing_entry *entries;
2407
2408                 r = -EFAULT;
2409                 if (copy_from_user(&routing, argp, sizeof(routing)))
2410                         goto out;
2411                 r = -EINVAL;
2412                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2413                         goto out;
2414                 if (routing.flags)
2415                         goto out;
2416                 r = -ENOMEM;
2417                 entries = vmalloc(routing.nr * sizeof(*entries));
2418                 if (!entries)
2419                         goto out;
2420                 r = -EFAULT;
2421                 urouting = argp;
2422                 if (copy_from_user(entries, urouting->entries,
2423                                    routing.nr * sizeof(*entries)))
2424                         goto out_free_irq_routing;
2425                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2426                                         routing.flags);
2427         out_free_irq_routing:
2428                 vfree(entries);
2429                 break;
2430         }
2431 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2432         case KVM_CREATE_DEVICE: {
2433                 struct kvm_create_device cd;
2434
2435                 r = -EFAULT;
2436                 if (copy_from_user(&cd, argp, sizeof(cd)))
2437                         goto out;
2438
2439                 r = kvm_ioctl_create_device(kvm, &cd);
2440                 if (r)
2441                         goto out;
2442
2443                 r = -EFAULT;
2444                 if (copy_to_user(argp, &cd, sizeof(cd)))
2445                         goto out;
2446
2447                 r = 0;
2448                 break;
2449         }
2450         default:
2451                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2452                 if (r == -ENOTTY)
2453                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2454         }
2455 out:
2456         return r;
2457 }
2458
2459 #ifdef CONFIG_COMPAT
2460 struct compat_kvm_dirty_log {
2461         __u32 slot;
2462         __u32 padding1;
2463         union {
2464                 compat_uptr_t dirty_bitmap; /* one bit per page */
2465                 __u64 padding2;
2466         };
2467 };
2468
2469 static long kvm_vm_compat_ioctl(struct file *filp,
2470                            unsigned int ioctl, unsigned long arg)
2471 {
2472         struct kvm *kvm = filp->private_data;
2473         int r;
2474
2475         if (kvm->mm != current->mm)
2476                 return -EIO;
2477         switch (ioctl) {
2478         case KVM_GET_DIRTY_LOG: {
2479                 struct compat_kvm_dirty_log compat_log;
2480                 struct kvm_dirty_log log;
2481
2482                 r = -EFAULT;
2483                 if (copy_from_user(&compat_log, (void __user *)arg,
2484                                    sizeof(compat_log)))
2485                         goto out;
2486                 log.slot         = compat_log.slot;
2487                 log.padding1     = compat_log.padding1;
2488                 log.padding2     = compat_log.padding2;
2489                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2490
2491                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2492                 break;
2493         }
2494         default:
2495                 r = kvm_vm_ioctl(filp, ioctl, arg);
2496         }
2497
2498 out:
2499         return r;
2500 }
2501 #endif
2502
2503 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2504 {
2505         struct page *page[1];
2506         unsigned long addr;
2507         int npages;
2508         gfn_t gfn = vmf->pgoff;
2509         struct kvm *kvm = vma->vm_file->private_data;
2510
2511         addr = gfn_to_hva(kvm, gfn);
2512         if (kvm_is_error_hva(addr))
2513                 return VM_FAULT_SIGBUS;
2514
2515         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2516                                 NULL);
2517         if (unlikely(npages != 1))
2518                 return VM_FAULT_SIGBUS;
2519
2520         vmf->page = page[0];
2521         return 0;
2522 }
2523
2524 static const struct vm_operations_struct kvm_vm_vm_ops = {
2525         .fault = kvm_vm_fault,
2526 };
2527
2528 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2529 {
2530         vma->vm_ops = &kvm_vm_vm_ops;
2531         return 0;
2532 }
2533
2534 static struct file_operations kvm_vm_fops = {
2535         .release        = kvm_vm_release,
2536         .unlocked_ioctl = kvm_vm_ioctl,
2537 #ifdef CONFIG_COMPAT
2538         .compat_ioctl   = kvm_vm_compat_ioctl,
2539 #endif
2540         .mmap           = kvm_vm_mmap,
2541         .llseek         = noop_llseek,
2542 };
2543
2544 static int kvm_dev_ioctl_create_vm(unsigned long type)
2545 {
2546         int r;
2547         struct kvm *kvm;
2548
2549         kvm = kvm_create_vm(type);
2550         if (IS_ERR(kvm))
2551                 return PTR_ERR(kvm);
2552 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2553         r = kvm_coalesced_mmio_init(kvm);
2554         if (r < 0) {
2555                 kvm_put_kvm(kvm);
2556                 return r;
2557         }
2558 #endif
2559         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2560         if (r < 0)
2561                 kvm_put_kvm(kvm);
2562
2563         return r;
2564 }
2565
2566 static long kvm_dev_ioctl_check_extension_generic(long arg)
2567 {
2568         switch (arg) {
2569         case KVM_CAP_USER_MEMORY:
2570         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2571         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2572 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2573         case KVM_CAP_SET_BOOT_CPU_ID:
2574 #endif
2575         case KVM_CAP_INTERNAL_ERROR_DATA:
2576 #ifdef CONFIG_HAVE_KVM_MSI
2577         case KVM_CAP_SIGNAL_MSI:
2578 #endif
2579 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2580         case KVM_CAP_IRQFD_RESAMPLE:
2581 #endif
2582                 return 1;
2583 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2584         case KVM_CAP_IRQ_ROUTING:
2585                 return KVM_MAX_IRQ_ROUTES;
2586 #endif
2587         default:
2588                 break;
2589         }
2590         return kvm_dev_ioctl_check_extension(arg);
2591 }
2592
2593 static long kvm_dev_ioctl(struct file *filp,
2594                           unsigned int ioctl, unsigned long arg)
2595 {
2596         long r = -EINVAL;
2597
2598         switch (ioctl) {
2599         case KVM_GET_API_VERSION:
2600                 r = -EINVAL;
2601                 if (arg)
2602                         goto out;
2603                 r = KVM_API_VERSION;
2604                 break;
2605         case KVM_CREATE_VM:
2606                 r = kvm_dev_ioctl_create_vm(arg);
2607                 break;
2608         case KVM_CHECK_EXTENSION:
2609                 r = kvm_dev_ioctl_check_extension_generic(arg);
2610                 break;
2611         case KVM_GET_VCPU_MMAP_SIZE:
2612                 r = -EINVAL;
2613                 if (arg)
2614                         goto out;
2615                 r = PAGE_SIZE;     /* struct kvm_run */
2616 #ifdef CONFIG_X86
2617                 r += PAGE_SIZE;    /* pio data page */
2618 #endif
2619 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2620                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2621 #endif
2622                 break;
2623         case KVM_TRACE_ENABLE:
2624         case KVM_TRACE_PAUSE:
2625         case KVM_TRACE_DISABLE:
2626                 r = -EOPNOTSUPP;
2627                 break;
2628         default:
2629                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2630         }
2631 out:
2632         return r;
2633 }
2634
2635 static struct file_operations kvm_chardev_ops = {
2636         .unlocked_ioctl = kvm_dev_ioctl,
2637         .compat_ioctl   = kvm_dev_ioctl,
2638         .llseek         = noop_llseek,
2639 };
2640
2641 static struct miscdevice kvm_dev = {
2642         KVM_MINOR,
2643         "kvm",
2644         &kvm_chardev_ops,
2645 };
2646
2647 static void hardware_enable_nolock(void *junk)
2648 {
2649         int cpu = raw_smp_processor_id();
2650         int r;
2651
2652         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2653                 return;
2654
2655         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2656
2657         r = kvm_arch_hardware_enable(NULL);
2658
2659         if (r) {
2660                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2661                 atomic_inc(&hardware_enable_failed);
2662                 printk(KERN_INFO "kvm: enabling virtualization on "
2663                                  "CPU%d failed\n", cpu);
2664         }
2665 }
2666
2667 static void hardware_enable(void *junk)
2668 {
2669         raw_spin_lock(&kvm_lock);
2670         hardware_enable_nolock(junk);
2671         raw_spin_unlock(&kvm_lock);
2672 }
2673
2674 static void hardware_disable_nolock(void *junk)
2675 {
2676         int cpu = raw_smp_processor_id();
2677
2678         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2679                 return;
2680         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2681         kvm_arch_hardware_disable(NULL);
2682 }
2683
2684 static void hardware_disable(void *junk)
2685 {
2686         raw_spin_lock(&kvm_lock);
2687         hardware_disable_nolock(junk);
2688         raw_spin_unlock(&kvm_lock);
2689 }
2690
2691 static void hardware_disable_all_nolock(void)
2692 {
2693         BUG_ON(!kvm_usage_count);
2694
2695         kvm_usage_count--;
2696         if (!kvm_usage_count)
2697                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2698 }
2699
2700 static void hardware_disable_all(void)
2701 {
2702         raw_spin_lock(&kvm_lock);
2703         hardware_disable_all_nolock();
2704         raw_spin_unlock(&kvm_lock);
2705 }
2706
2707 static int hardware_enable_all(void)
2708 {
2709         int r = 0;
2710
2711         raw_spin_lock(&kvm_lock);
2712
2713         kvm_usage_count++;
2714         if (kvm_usage_count == 1) {
2715                 atomic_set(&hardware_enable_failed, 0);
2716                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2717
2718                 if (atomic_read(&hardware_enable_failed)) {
2719                         hardware_disable_all_nolock();
2720                         r = -EBUSY;
2721                 }
2722         }
2723
2724         raw_spin_unlock(&kvm_lock);
2725
2726         return r;
2727 }
2728
2729 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2730                            void *v)
2731 {
2732         int cpu = (long)v;
2733
2734         if (!kvm_usage_count)
2735                 return NOTIFY_OK;
2736
2737         val &= ~CPU_TASKS_FROZEN;
2738         switch (val) {
2739         case CPU_DYING:
2740                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2741                        cpu);
2742                 hardware_disable(NULL);
2743                 break;
2744         case CPU_STARTING:
2745                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2746                        cpu);
2747                 hardware_enable(NULL);
2748                 break;
2749         }
2750         return NOTIFY_OK;
2751 }
2752
2753 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2754                       void *v)
2755 {
2756         /*
2757          * Some (well, at least mine) BIOSes hang on reboot if
2758          * in vmx root mode.
2759          *
2760          * And Intel TXT required VMX off for all cpu when system shutdown.
2761          */
2762         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2763         kvm_rebooting = true;
2764         on_each_cpu(hardware_disable_nolock, NULL, 1);
2765         return NOTIFY_OK;
2766 }
2767
2768 static struct notifier_block kvm_reboot_notifier = {
2769         .notifier_call = kvm_reboot,
2770         .priority = 0,
2771 };
2772
2773 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2774 {
2775         int i;
2776
2777         for (i = 0; i < bus->dev_count; i++) {
2778                 struct kvm_io_device *pos = bus->range[i].dev;
2779
2780                 kvm_iodevice_destructor(pos);
2781         }
2782         kfree(bus);
2783 }
2784
2785 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2786 {
2787         const struct kvm_io_range *r1 = p1;
2788         const struct kvm_io_range *r2 = p2;
2789
2790         if (r1->addr < r2->addr)
2791                 return -1;
2792         if (r1->addr + r1->len > r2->addr + r2->len)
2793                 return 1;
2794         return 0;
2795 }
2796
2797 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2798                           gpa_t addr, int len)
2799 {
2800         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2801                 .addr = addr,
2802                 .len = len,
2803                 .dev = dev,
2804         };
2805
2806         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2807                 kvm_io_bus_sort_cmp, NULL);
2808
2809         return 0;
2810 }
2811
2812 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2813                              gpa_t addr, int len)
2814 {
2815         struct kvm_io_range *range, key;
2816         int off;
2817
2818         key = (struct kvm_io_range) {
2819                 .addr = addr,
2820                 .len = len,
2821         };
2822
2823         range = bsearch(&key, bus->range, bus->dev_count,
2824                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2825         if (range == NULL)
2826                 return -ENOENT;
2827
2828         off = range - bus->range;
2829
2830         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2831                 off--;
2832
2833         return off;
2834 }
2835
2836 /* kvm_io_bus_write - called under kvm->slots_lock */
2837 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2838                      int len, const void *val)
2839 {
2840         int idx;
2841         struct kvm_io_bus *bus;
2842         struct kvm_io_range range;
2843
2844         range = (struct kvm_io_range) {
2845                 .addr = addr,
2846                 .len = len,
2847         };
2848
2849         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2850         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2851         if (idx < 0)
2852                 return -EOPNOTSUPP;
2853
2854         while (idx < bus->dev_count &&
2855                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2856                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2857                         return 0;
2858                 idx++;
2859         }
2860
2861         return -EOPNOTSUPP;
2862 }
2863
2864 /* kvm_io_bus_read - called under kvm->slots_lock */
2865 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2866                     int len, void *val)
2867 {
2868         int idx;
2869         struct kvm_io_bus *bus;
2870         struct kvm_io_range range;
2871
2872         range = (struct kvm_io_range) {
2873                 .addr = addr,
2874                 .len = len,
2875         };
2876
2877         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2878         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2879         if (idx < 0)
2880                 return -EOPNOTSUPP;
2881
2882         while (idx < bus->dev_count &&
2883                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2884                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2885                         return 0;
2886                 idx++;
2887         }
2888
2889         return -EOPNOTSUPP;
2890 }
2891
2892 /* Caller must hold slots_lock. */
2893 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2894                             int len, struct kvm_io_device *dev)
2895 {
2896         struct kvm_io_bus *new_bus, *bus;
2897
2898         bus = kvm->buses[bus_idx];
2899         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2900                 return -ENOSPC;
2901
2902         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2903                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2904         if (!new_bus)
2905                 return -ENOMEM;
2906         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2907                sizeof(struct kvm_io_range)));
2908         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2909         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2910         synchronize_srcu_expedited(&kvm->srcu);
2911         kfree(bus);
2912
2913         return 0;
2914 }
2915
2916 /* Caller must hold slots_lock. */
2917 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2918                               struct kvm_io_device *dev)
2919 {
2920         int i, r;
2921         struct kvm_io_bus *new_bus, *bus;
2922
2923         bus = kvm->buses[bus_idx];
2924         r = -ENOENT;
2925         for (i = 0; i < bus->dev_count; i++)
2926                 if (bus->range[i].dev == dev) {
2927                         r = 0;
2928                         break;
2929                 }
2930
2931         if (r)
2932                 return r;
2933
2934         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2935                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2936         if (!new_bus)
2937                 return -ENOMEM;
2938
2939         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2940         new_bus->dev_count--;
2941         memcpy(new_bus->range + i, bus->range + i + 1,
2942                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2943
2944         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2945         synchronize_srcu_expedited(&kvm->srcu);
2946         kfree(bus);
2947         return r;
2948 }
2949
2950 static struct notifier_block kvm_cpu_notifier = {
2951         .notifier_call = kvm_cpu_hotplug,
2952 };
2953
2954 static int vm_stat_get(void *_offset, u64 *val)
2955 {
2956         unsigned offset = (long)_offset;
2957         struct kvm *kvm;
2958
2959         *val = 0;
2960         raw_spin_lock(&kvm_lock);
2961         list_for_each_entry(kvm, &vm_list, vm_list)
2962                 *val += *(u32 *)((void *)kvm + offset);
2963         raw_spin_unlock(&kvm_lock);
2964         return 0;
2965 }
2966
2967 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2968
2969 static int vcpu_stat_get(void *_offset, u64 *val)
2970 {
2971         unsigned offset = (long)_offset;
2972         struct kvm *kvm;
2973         struct kvm_vcpu *vcpu;
2974         int i;
2975
2976         *val = 0;
2977         raw_spin_lock(&kvm_lock);
2978         list_for_each_entry(kvm, &vm_list, vm_list)
2979                 kvm_for_each_vcpu(i, vcpu, kvm)
2980                         *val += *(u32 *)((void *)vcpu + offset);
2981
2982         raw_spin_unlock(&kvm_lock);
2983         return 0;
2984 }
2985
2986 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2987
2988 static const struct file_operations *stat_fops[] = {
2989         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2990         [KVM_STAT_VM]   = &vm_stat_fops,
2991 };
2992
2993 static int kvm_init_debug(void)
2994 {
2995         int r = -EFAULT;
2996         struct kvm_stats_debugfs_item *p;
2997
2998         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2999         if (kvm_debugfs_dir == NULL)
3000                 goto out;
3001
3002         for (p = debugfs_entries; p->name; ++p) {
3003                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3004                                                 (void *)(long)p->offset,
3005                                                 stat_fops[p->kind]);
3006                 if (p->dentry == NULL)
3007                         goto out_dir;
3008         }
3009
3010         return 0;
3011
3012 out_dir:
3013         debugfs_remove_recursive(kvm_debugfs_dir);
3014 out:
3015         return r;
3016 }
3017
3018 static void kvm_exit_debug(void)
3019 {
3020         struct kvm_stats_debugfs_item *p;
3021
3022         for (p = debugfs_entries; p->name; ++p)
3023                 debugfs_remove(p->dentry);
3024         debugfs_remove(kvm_debugfs_dir);
3025 }
3026
3027 static int kvm_suspend(void)
3028 {
3029         if (kvm_usage_count)
3030                 hardware_disable_nolock(NULL);
3031         return 0;
3032 }
3033
3034 static void kvm_resume(void)
3035 {
3036         if (kvm_usage_count) {
3037                 WARN_ON(raw_spin_is_locked(&kvm_lock));
3038                 hardware_enable_nolock(NULL);
3039         }
3040 }
3041
3042 static struct syscore_ops kvm_syscore_ops = {
3043         .suspend = kvm_suspend,
3044         .resume = kvm_resume,
3045 };
3046
3047 static inline
3048 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3049 {
3050         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3051 }
3052
3053 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3054 {
3055         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3056         if (vcpu->preempted)
3057                 vcpu->preempted = false;
3058
3059         kvm_arch_vcpu_load(vcpu, cpu);
3060 }
3061
3062 static void kvm_sched_out(struct preempt_notifier *pn,
3063                           struct task_struct *next)
3064 {
3065         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3066
3067         if (current->state == TASK_RUNNING)
3068                 vcpu->preempted = true;
3069         kvm_arch_vcpu_put(vcpu);
3070 }
3071
3072 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3073                   struct module *module)
3074 {
3075         int r;
3076         int cpu;
3077
3078         r = kvm_irqfd_init();
3079         if (r)
3080                 goto out_irqfd;
3081         r = kvm_arch_init(opaque);
3082         if (r)
3083                 goto out_fail;
3084
3085         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3086                 r = -ENOMEM;
3087                 goto out_free_0;
3088         }
3089
3090         r = kvm_arch_hardware_setup();
3091         if (r < 0)
3092                 goto out_free_0a;
3093
3094         for_each_online_cpu(cpu) {
3095                 smp_call_function_single(cpu,
3096                                 kvm_arch_check_processor_compat,
3097                                 &r, 1);
3098                 if (r < 0)
3099                         goto out_free_1;
3100         }
3101
3102         r = register_cpu_notifier(&kvm_cpu_notifier);
3103         if (r)
3104                 goto out_free_2;
3105         register_reboot_notifier(&kvm_reboot_notifier);
3106
3107         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3108         if (!vcpu_align)
3109                 vcpu_align = __alignof__(struct kvm_vcpu);
3110         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3111                                            0, NULL);
3112         if (!kvm_vcpu_cache) {
3113                 r = -ENOMEM;
3114                 goto out_free_3;
3115         }
3116
3117         r = kvm_async_pf_init();
3118         if (r)
3119                 goto out_free;
3120
3121         kvm_chardev_ops.owner = module;
3122         kvm_vm_fops.owner = module;
3123         kvm_vcpu_fops.owner = module;
3124
3125         r = misc_register(&kvm_dev);
3126         if (r) {
3127                 printk(KERN_ERR "kvm: misc device register failed\n");
3128                 goto out_unreg;
3129         }
3130
3131         register_syscore_ops(&kvm_syscore_ops);
3132
3133         kvm_preempt_ops.sched_in = kvm_sched_in;
3134         kvm_preempt_ops.sched_out = kvm_sched_out;
3135
3136         r = kvm_init_debug();
3137         if (r) {
3138                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3139                 goto out_undebugfs;
3140         }
3141
3142         return 0;
3143
3144 out_undebugfs:
3145         unregister_syscore_ops(&kvm_syscore_ops);
3146 out_unreg:
3147         kvm_async_pf_deinit();
3148 out_free:
3149         kmem_cache_destroy(kvm_vcpu_cache);
3150 out_free_3:
3151         unregister_reboot_notifier(&kvm_reboot_notifier);
3152         unregister_cpu_notifier(&kvm_cpu_notifier);
3153 out_free_2:
3154 out_free_1:
3155         kvm_arch_hardware_unsetup();
3156 out_free_0a:
3157         free_cpumask_var(cpus_hardware_enabled);
3158 out_free_0:
3159         kvm_arch_exit();
3160 out_fail:
3161         kvm_irqfd_exit();
3162 out_irqfd:
3163         return r;
3164 }
3165 EXPORT_SYMBOL_GPL(kvm_init);
3166
3167 void kvm_exit(void)
3168 {
3169         kvm_exit_debug();
3170         misc_deregister(&kvm_dev);
3171         kmem_cache_destroy(kvm_vcpu_cache);
3172         kvm_async_pf_deinit();
3173         unregister_syscore_ops(&kvm_syscore_ops);
3174         unregister_reboot_notifier(&kvm_reboot_notifier);
3175         unregister_cpu_notifier(&kvm_cpu_notifier);
3176         on_each_cpu(hardware_disable_nolock, NULL, 1);
3177         kvm_arch_hardware_unsetup();
3178         kvm_arch_exit();
3179         kvm_irqfd_exit();
3180         free_cpumask_var(cpus_hardware_enabled);
3181 }
3182 EXPORT_SYMBOL_GPL(kvm_exit);