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_head(tail);
 109                 reserved = PageReserved(head);
 110                 if (head != tail) {
 111                         /*
 112                          * "head" is not a dangling pointer
 113                          * (compound_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, unsigned long len)
1539 {
1540         struct kvm_memslots *slots = kvm_memslots(kvm);
1541         int offset = offset_in_page(gpa);
1542         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1543         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1544         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1545         gfn_t nr_pages_avail;
1546
1547         ghc->gpa = gpa;
1548         ghc->generation = slots->generation;
1549         ghc->len = len;
1550         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1551         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1552         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1553                 ghc->hva += offset;
1554         } else {
1555                 /*
1556                  * If the requested region crosses two memslots, we still
1557                  * verify that the entire region is valid here.
1558                  */
1559                 while (start_gfn <= end_gfn) {
1560                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1561                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1562                                                    &nr_pages_avail);
1563                         if (kvm_is_error_hva(ghc->hva))
1564                                 return -EFAULT;
1565                         start_gfn += nr_pages_avail;
1566                 }
1567                 /* Use the slow path for cross page reads and writes. */
1568                 ghc->memslot = NULL;
1569         }
1570         return 0;
1571 }
1572 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1573
1574 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1575                            void *data, unsigned long len)
1576 {
1577         struct kvm_memslots *slots = kvm_memslots(kvm);
1578         int r;
1579
1580         BUG_ON(len > ghc->len);
1581
1582         if (slots->generation != ghc->generation)
1583                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1584
1585         if (unlikely(!ghc->memslot))
1586                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1587
1588         if (kvm_is_error_hva(ghc->hva))
1589                 return -EFAULT;
1590
1591         r = __copy_to_user((void __user *)ghc->hva, data, len);
1592         if (r)
1593                 return -EFAULT;
1594         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1595
1596         return 0;
1597 }
1598 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1599
1600 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1601                            void *data, unsigned long len)
1602 {
1603         struct kvm_memslots *slots = kvm_memslots(kvm);
1604         int r;
1605
1606         BUG_ON(len > ghc->len);
1607
1608         if (slots->generation != ghc->generation)
1609                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1610
1611         if (unlikely(!ghc->memslot))
1612                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1613
1614         if (kvm_is_error_hva(ghc->hva))
1615                 return -EFAULT;
1616
1617         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1618         if (r)
1619                 return -EFAULT;
1620
1621         return 0;
1622 }
1623 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1624
1625 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1626 {
1627         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1628                                     offset, len);
1629 }
1630 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1631
1632 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1633 {
1634         gfn_t gfn = gpa >> PAGE_SHIFT;
1635         int seg;
1636         int offset = offset_in_page(gpa);
1637         int ret;
1638
1639         while ((seg = next_segment(len, offset)) != 0) {
1640                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1641                 if (ret < 0)
1642                         return ret;
1643                 offset = 0;
1644                 len -= seg;
1645                 ++gfn;
1646         }
1647         return 0;
1648 }
1649 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1650
1651 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1652                              gfn_t gfn)
1653 {
1654         if (memslot && memslot->dirty_bitmap) {
1655                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1656
1657                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1658         }
1659 }
1660
1661 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1662 {
1663         struct kvm_memory_slot *memslot;
1664
1665         memslot = gfn_to_memslot(kvm, gfn);
1666         mark_page_dirty_in_slot(kvm, memslot, gfn);
1667 }
1668
1669 /*
1670  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1671  */
1672 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1673 {
1674         DEFINE_WAIT(wait);
1675
1676         for (;;) {
1677                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1678
1679                 if (kvm_arch_vcpu_runnable(vcpu)) {
1680                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1681                         break;
1682                 }
1683                 if (kvm_cpu_has_pending_timer(vcpu))
1684                         break;
1685                 if (signal_pending(current))
1686                         break;
1687
1688                 schedule();
1689         }
1690
1691         finish_wait(&vcpu->wq, &wait);
1692 }
1693
1694 #ifndef CONFIG_S390
1695 /*
1696  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1697  */
1698 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1699 {
1700         int me;
1701         int cpu = vcpu->cpu;
1702         wait_queue_head_t *wqp;
1703
1704         wqp = kvm_arch_vcpu_wq(vcpu);
1705         if (waitqueue_active(wqp)) {
1706                 wake_up_interruptible(wqp);
1707                 ++vcpu->stat.halt_wakeup;
1708         }
1709
1710         me = get_cpu();
1711         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1712                 if (kvm_arch_vcpu_should_kick(vcpu))
1713                         smp_send_reschedule(cpu);
1714         put_cpu();
1715 }
1716 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1717 #endif /* !CONFIG_S390 */
1718
1719 void kvm_resched(struct kvm_vcpu *vcpu)
1720 {
1721         if (!need_resched())
1722                 return;
1723         cond_resched();
1724 }
1725 EXPORT_SYMBOL_GPL(kvm_resched);
1726
1727 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1728 {
1729         struct pid *pid;
1730         struct task_struct *task = NULL;
1731         bool ret = false;
1732
1733         rcu_read_lock();
1734         pid = rcu_dereference(target->pid);
1735         if (pid)
1736                 task = get_pid_task(target->pid, PIDTYPE_PID);
1737         rcu_read_unlock();
1738         if (!task)
1739                 return ret;
1740         if (task->flags & PF_VCPU) {
1741                 put_task_struct(task);
1742                 return ret;
1743         }
1744         ret = yield_to(task, 1);
1745         put_task_struct(task);
1746
1747         return ret;
1748 }
1749 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1750
1751 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1752 /*
1753  * Helper that checks whether a VCPU is eligible for directed yield.
1754  * Most eligible candidate to yield is decided by following heuristics:
1755  *
1756  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1757  *  (preempted lock holder), indicated by @in_spin_loop.
1758  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1759  *
1760  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1761  *  chance last time (mostly it has become eligible now since we have probably
1762  *  yielded to lockholder in last iteration. This is done by toggling
1763  *  @dy_eligible each time a VCPU checked for eligibility.)
1764  *
1765  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1766  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1767  *  burning. Giving priority for a potential lock-holder increases lock
1768  *  progress.
1769  *
1770  *  Since algorithm is based on heuristics, accessing another VCPU data without
1771  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1772  *  and continue with next VCPU and so on.
1773  */
1774 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1775 {
1776         bool eligible;
1777
1778         eligible = !vcpu->spin_loop.in_spin_loop ||
1779                         (vcpu->spin_loop.in_spin_loop &&
1780                          vcpu->spin_loop.dy_eligible);
1781
1782         if (vcpu->spin_loop.in_spin_loop)
1783                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1784
1785         return eligible;
1786 }
1787 #endif
1788
1789 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1790 {
1791         struct kvm *kvm = me->kvm;
1792         struct kvm_vcpu *vcpu;
1793         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1794         int yielded = 0;
1795         int try = 3;
1796         int pass;
1797         int i;
1798
1799         kvm_vcpu_set_in_spin_loop(me, true);
1800         /*
1801          * We boost the priority of a VCPU that is runnable but not
1802          * currently running, because it got preempted by something
1803          * else and called schedule in __vcpu_run.  Hopefully that
1804          * VCPU is holding the lock that we need and will release it.
1805          * We approximate round-robin by starting at the last boosted VCPU.
1806          */
1807         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1808                 kvm_for_each_vcpu(i, vcpu, kvm) {
1809                         if (!pass && i <= last_boosted_vcpu) {
1810                                 i = last_boosted_vcpu;
1811                                 continue;
1812                         } else if (pass && i > last_boosted_vcpu)
1813                                 break;
1814                         if (!ACCESS_ONCE(vcpu->preempted))
1815                                 continue;
1816                         if (vcpu == me)
1817                                 continue;
1818                         if (waitqueue_active(&vcpu->wq))
1819                                 continue;
1820                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1821                                 continue;
1822
1823                         yielded = kvm_vcpu_yield_to(vcpu);
1824                         if (yielded > 0) {
1825                                 kvm->last_boosted_vcpu = i;
1826                                 break;
1827                         } else if (yielded < 0) {
1828                                 try--;
1829                                 if (!try)
1830                                         break;
1831                         }
1832                 }
1833         }
1834         kvm_vcpu_set_in_spin_loop(me, false);
1835
1836         /* Ensure vcpu is not eligible during next spinloop */
1837         kvm_vcpu_set_dy_eligible(me, false);
1838 }
1839 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1840
1841 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1842 {
1843         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1844         struct page *page;
1845
1846         if (vmf->pgoff == 0)
1847                 page = virt_to_page(vcpu->run);
1848 #ifdef CONFIG_X86
1849         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1850                 page = virt_to_page(vcpu->arch.pio_data);
1851 #endif
1852 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1853         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1854                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1855 #endif
1856         else
1857                 return kvm_arch_vcpu_fault(vcpu, vmf);
1858         get_page(page);
1859         vmf->page = page;
1860         return 0;
1861 }
1862
1863 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1864         .fault = kvm_vcpu_fault,
1865 };
1866
1867 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1868 {
1869         vma->vm_ops = &kvm_vcpu_vm_ops;
1870         return 0;
1871 }
1872
1873 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1874 {
1875         struct kvm_vcpu *vcpu = filp->private_data;
1876
1877         kvm_put_kvm(vcpu->kvm);
1878         return 0;
1879 }
1880
1881 static struct file_operations kvm_vcpu_fops = {
1882         .release        = kvm_vcpu_release,
1883         .unlocked_ioctl = kvm_vcpu_ioctl,
1884 #ifdef CONFIG_COMPAT
1885         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1886 #endif
1887         .mmap           = kvm_vcpu_mmap,
1888         .llseek         = noop_llseek,
1889 };
1890
1891 /*
1892  * Allocates an inode for the vcpu.
1893  */
1894 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1895 {
1896         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1897 }
1898
1899 /*
1900  * Creates some virtual cpus.  Good luck creating more than one.
1901  */
1902 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1903 {
1904         int r;
1905         struct kvm_vcpu *vcpu, *v;
1906
1907         if (id >= KVM_MAX_VCPUS)
1908                 return -EINVAL;
1909
1910         vcpu = kvm_arch_vcpu_create(kvm, id);
1911         if (IS_ERR(vcpu))
1912                 return PTR_ERR(vcpu);
1913
1914         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1915
1916         r = kvm_arch_vcpu_setup(vcpu);
1917         if (r)
1918                 goto vcpu_destroy;
1919
1920         mutex_lock(&kvm->lock);
1921         if (!kvm_vcpu_compatible(vcpu)) {
1922                 r = -EINVAL;
1923                 goto unlock_vcpu_destroy;
1924         }
1925         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1926                 r = -EINVAL;
1927                 goto unlock_vcpu_destroy;
1928         }
1929
1930         kvm_for_each_vcpu(r, v, kvm)
1931                 if (v->vcpu_id == id) {
1932                         r = -EEXIST;
1933                         goto unlock_vcpu_destroy;
1934                 }
1935
1936         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1937
1938         /* Now it's all set up, let userspace reach it */
1939         kvm_get_kvm(kvm);
1940         r = create_vcpu_fd(vcpu);
1941         if (r < 0) {
1942                 kvm_put_kvm(kvm);
1943                 goto unlock_vcpu_destroy;
1944         }
1945
1946         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1947         smp_wmb();
1948         atomic_inc(&kvm->online_vcpus);
1949
1950         mutex_unlock(&kvm->lock);
1951         kvm_arch_vcpu_postcreate(vcpu);
1952         return r;
1953
1954 unlock_vcpu_destroy:
1955         mutex_unlock(&kvm->lock);
1956 vcpu_destroy:
1957         kvm_arch_vcpu_destroy(vcpu);
1958         return r;
1959 }
1960
1961 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1962 {
1963         if (sigset) {
1964                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1965                 vcpu->sigset_active = 1;
1966                 vcpu->sigset = *sigset;
1967         } else
1968                 vcpu->sigset_active = 0;
1969         return 0;
1970 }
1971
1972 static long kvm_vcpu_ioctl(struct file *filp,
1973                            unsigned int ioctl, unsigned long arg)
1974 {
1975         struct kvm_vcpu *vcpu = filp->private_data;
1976         void __user *argp = (void __user *)arg;
1977         int r;
1978         struct kvm_fpu *fpu = NULL;
1979         struct kvm_sregs *kvm_sregs = NULL;
1980
1981         if (vcpu->kvm->mm != current->mm)
1982                 return -EIO;
1983
1984 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1985         /*
1986          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1987          * so vcpu_load() would break it.
1988          */
1989         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1990                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1991 #endif
1992
1993
1994         r = vcpu_load(vcpu);
1995         if (r)
1996                 return r;
1997         switch (ioctl) {
1998         case KVM_RUN:
1999                 r = -EINVAL;
2000                 if (arg)
2001                         goto out;
2002                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2003                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2004                 break;
2005         case KVM_GET_REGS: {
2006                 struct kvm_regs *kvm_regs;
2007
2008                 r = -ENOMEM;
2009                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2010                 if (!kvm_regs)
2011                         goto out;
2012                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2013                 if (r)
2014                         goto out_free1;
2015                 r = -EFAULT;
2016                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2017                         goto out_free1;
2018                 r = 0;
2019 out_free1:
2020                 kfree(kvm_regs);
2021                 break;
2022         }
2023         case KVM_SET_REGS: {
2024                 struct kvm_regs *kvm_regs;
2025
2026                 r = -ENOMEM;
2027                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2028                 if (IS_ERR(kvm_regs)) {
2029                         r = PTR_ERR(kvm_regs);
2030                         goto out;
2031                 }
2032                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2033                 kfree(kvm_regs);
2034                 break;
2035         }
2036         case KVM_GET_SREGS: {
2037                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2038                 r = -ENOMEM;
2039                 if (!kvm_sregs)
2040                         goto out;
2041                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2042                 if (r)
2043                         goto out;
2044                 r = -EFAULT;
2045                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2046                         goto out;
2047                 r = 0;
2048                 break;
2049         }
2050         case KVM_SET_SREGS: {
2051                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2052                 if (IS_ERR(kvm_sregs)) {
2053                         r = PTR_ERR(kvm_sregs);
2054                         kvm_sregs = NULL;
2055                         goto out;
2056                 }
2057                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2058                 break;
2059         }
2060         case KVM_GET_MP_STATE: {
2061                 struct kvm_mp_state mp_state;
2062
2063                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2064                 if (r)
2065                         goto out;
2066                 r = -EFAULT;
2067                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2068                         goto out;
2069                 r = 0;
2070                 break;
2071         }
2072         case KVM_SET_MP_STATE: {
2073                 struct kvm_mp_state mp_state;
2074
2075                 r = -EFAULT;
2076                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2077                         goto out;
2078                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2079                 break;
2080         }
2081         case KVM_TRANSLATE: {
2082                 struct kvm_translation tr;
2083
2084                 r = -EFAULT;
2085                 if (copy_from_user(&tr, argp, sizeof tr))
2086                         goto out;
2087                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2088                 if (r)
2089                         goto out;
2090                 r = -EFAULT;
2091                 if (copy_to_user(argp, &tr, sizeof tr))
2092                         goto out;
2093                 r = 0;
2094                 break;
2095         }
2096         case KVM_SET_GUEST_DEBUG: {
2097                 struct kvm_guest_debug dbg;
2098
2099                 r = -EFAULT;
2100                 if (copy_from_user(&dbg, argp, sizeof dbg))
2101                         goto out;
2102                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2103                 break;
2104         }
2105         case KVM_SET_SIGNAL_MASK: {
2106                 struct kvm_signal_mask __user *sigmask_arg = argp;
2107                 struct kvm_signal_mask kvm_sigmask;
2108                 sigset_t sigset, *p;
2109
2110                 p = NULL;
2111                 if (argp) {
2112                         r = -EFAULT;
2113                         if (copy_from_user(&kvm_sigmask, argp,
2114                                            sizeof kvm_sigmask))
2115                                 goto out;
2116                         r = -EINVAL;
2117                         if (kvm_sigmask.len != sizeof sigset)
2118                                 goto out;
2119                         r = -EFAULT;
2120                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2121                                            sizeof sigset))
2122                                 goto out;
2123                         p = &sigset;
2124                 }
2125                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2126                 break;
2127         }
2128         case KVM_GET_FPU: {
2129                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2130                 r = -ENOMEM;
2131                 if (!fpu)
2132                         goto out;
2133                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2134                 if (r)
2135                         goto out;
2136                 r = -EFAULT;
2137                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2138                         goto out;
2139                 r = 0;
2140                 break;
2141         }
2142         case KVM_SET_FPU: {
2143                 fpu = memdup_user(argp, sizeof(*fpu));
2144                 if (IS_ERR(fpu)) {
2145                         r = PTR_ERR(fpu);
2146                         fpu = NULL;
2147                         goto out;
2148                 }
2149                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2150                 break;
2151         }
2152         default:
2153                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2154         }
2155 out:
2156         vcpu_put(vcpu);
2157         kfree(fpu);
2158         kfree(kvm_sregs);
2159         return r;
2160 }
2161
2162 #ifdef CONFIG_COMPAT
2163 static long kvm_vcpu_compat_ioctl(struct file *filp,
2164                                   unsigned int ioctl, unsigned long arg)
2165 {
2166         struct kvm_vcpu *vcpu = filp->private_data;
2167         void __user *argp = compat_ptr(arg);
2168         int r;
2169
2170         if (vcpu->kvm->mm != current->mm)
2171                 return -EIO;
2172
2173         switch (ioctl) {
2174         case KVM_SET_SIGNAL_MASK: {
2175                 struct kvm_signal_mask __user *sigmask_arg = argp;
2176                 struct kvm_signal_mask kvm_sigmask;
2177                 compat_sigset_t csigset;
2178                 sigset_t sigset;
2179
2180                 if (argp) {
2181                         r = -EFAULT;
2182                         if (copy_from_user(&kvm_sigmask, argp,
2183                                            sizeof kvm_sigmask))
2184                                 goto out;
2185                         r = -EINVAL;
2186                         if (kvm_sigmask.len != sizeof csigset)
2187                                 goto out;
2188                         r = -EFAULT;
2189                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2190                                            sizeof csigset))
2191                                 goto out;
2192                         sigset_from_compat(&sigset, &csigset);
2193                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2194                 } else
2195                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2196                 break;
2197         }
2198         default:
2199                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2200         }
2201
2202 out:
2203         return r;
2204 }
2205 #endif
2206
2207 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2208                                  int (*accessor)(struct kvm_device *dev,
2209                                                  struct kvm_device_attr *attr),
2210                                  unsigned long arg)
2211 {
2212         struct kvm_device_attr attr;
2213
2214         if (!accessor)
2215                 return -EPERM;
2216
2217         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2218                 return -EFAULT;
2219
2220         return accessor(dev, &attr);
2221 }
2222
2223 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2224                              unsigned long arg)
2225 {
2226         struct kvm_device *dev = filp->private_data;
2227
2228         switch (ioctl) {
2229         case KVM_SET_DEVICE_ATTR:
2230                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2231         case KVM_GET_DEVICE_ATTR:
2232                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2233         case KVM_HAS_DEVICE_ATTR:
2234                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2235         default:
2236                 if (dev->ops->ioctl)
2237                         return dev->ops->ioctl(dev, ioctl, arg);
2238
2239                 return -ENOTTY;
2240         }
2241 }
2242
2243 static int kvm_device_release(struct inode *inode, struct file *filp)
2244 {
2245         struct kvm_device *dev = filp->private_data;
2246         struct kvm *kvm = dev->kvm;
2247
2248         kvm_put_kvm(kvm);
2249         return 0;
2250 }
2251
2252 static const struct file_operations kvm_device_fops = {
2253         .unlocked_ioctl = kvm_device_ioctl,
2254 #ifdef CONFIG_COMPAT
2255         .compat_ioctl = kvm_device_ioctl,
2256 #endif
2257         .release = kvm_device_release,
2258 };
2259
2260 struct kvm_device *kvm_device_from_filp(struct file *filp)
2261 {
2262         if (filp->f_op != &kvm_device_fops)
2263                 return NULL;
2264
2265         return filp->private_data;
2266 }
2267
2268 static int kvm_ioctl_create_device(struct kvm *kvm,
2269                                    struct kvm_create_device *cd)
2270 {
2271         struct kvm_device_ops *ops = NULL;
2272         struct kvm_device *dev;
2273         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2274         int ret;
2275
2276         switch (cd->type) {
2277 #ifdef CONFIG_KVM_MPIC
2278         case KVM_DEV_TYPE_FSL_MPIC_20:
2279         case KVM_DEV_TYPE_FSL_MPIC_42:
2280                 ops = &kvm_mpic_ops;
2281                 break;
2282 #endif
2283 #ifdef CONFIG_KVM_XICS
2284         case KVM_DEV_TYPE_XICS:
2285                 ops = &kvm_xics_ops;
2286                 break;
2287 #endif
2288         default:
2289                 return -ENODEV;
2290         }
2291
2292         if (test)
2293                 return 0;
2294
2295         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2296         if (!dev)
2297                 return -ENOMEM;
2298
2299         dev->ops = ops;
2300         dev->kvm = kvm;
2301
2302         ret = ops->create(dev, cd->type);
2303         if (ret < 0) {
2304                 kfree(dev);
2305                 return ret;
2306         }
2307
2308         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR);
2309         if (ret < 0) {
2310                 ops->destroy(dev);
2311                 return ret;
2312         }
2313
2314         list_add(&dev->vm_node, &kvm->devices);
2315         kvm_get_kvm(kvm);
2316         cd->fd = ret;
2317         return 0;
2318 }
2319
2320 static long kvm_vm_ioctl(struct file *filp,
2321                            unsigned int ioctl, unsigned long arg)
2322 {
2323         struct kvm *kvm = filp->private_data;
2324         void __user *argp = (void __user *)arg;
2325         int r;
2326
2327         if (kvm->mm != current->mm)
2328                 return -EIO;
2329         switch (ioctl) {
2330         case KVM_CREATE_VCPU:
2331                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2332                 break;
2333         case KVM_SET_USER_MEMORY_REGION: {
2334                 struct kvm_userspace_memory_region kvm_userspace_mem;
2335
2336                 r = -EFAULT;
2337                 if (copy_from_user(&kvm_userspace_mem, argp,
2338                                                 sizeof kvm_userspace_mem))
2339                         goto out;
2340
2341                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2342                 break;
2343         }
2344         case KVM_GET_DIRTY_LOG: {
2345                 struct kvm_dirty_log log;
2346
2347                 r = -EFAULT;
2348                 if (copy_from_user(&log, argp, sizeof log))
2349                         goto out;
2350                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2351                 break;
2352         }
2353 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2354         case KVM_REGISTER_COALESCED_MMIO: {
2355                 struct kvm_coalesced_mmio_zone zone;
2356                 r = -EFAULT;
2357                 if (copy_from_user(&zone, argp, sizeof zone))
2358                         goto out;
2359                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2360                 break;
2361         }
2362         case KVM_UNREGISTER_COALESCED_MMIO: {
2363                 struct kvm_coalesced_mmio_zone zone;
2364                 r = -EFAULT;
2365                 if (copy_from_user(&zone, argp, sizeof zone))
2366                         goto out;
2367                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2368                 break;
2369         }
2370 #endif
2371         case KVM_IRQFD: {
2372                 struct kvm_irqfd data;
2373
2374                 r = -EFAULT;
2375                 if (copy_from_user(&data, argp, sizeof data))
2376                         goto out;
2377                 r = kvm_irqfd(kvm, &data);
2378                 break;
2379         }
2380         case KVM_IOEVENTFD: {
2381                 struct kvm_ioeventfd data;
2382
2383                 r = -EFAULT;
2384                 if (copy_from_user(&data, argp, sizeof data))
2385                         goto out;
2386                 r = kvm_ioeventfd(kvm, &data);
2387                 break;
2388         }
2389 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2390         case KVM_SET_BOOT_CPU_ID:
2391                 r = 0;
2392                 mutex_lock(&kvm->lock);
2393                 if (atomic_read(&kvm->online_vcpus) != 0)
2394                         r = -EBUSY;
2395                 else
2396                         kvm->bsp_vcpu_id = arg;
2397                 mutex_unlock(&kvm->lock);
2398                 break;
2399 #endif
2400 #ifdef CONFIG_HAVE_KVM_MSI
2401         case KVM_SIGNAL_MSI: {
2402                 struct kvm_msi msi;
2403
2404                 r = -EFAULT;
2405                 if (copy_from_user(&msi, argp, sizeof msi))
2406                         goto out;
2407                 r = kvm_send_userspace_msi(kvm, &msi);
2408                 break;
2409         }
2410 #endif
2411 #ifdef __KVM_HAVE_IRQ_LINE
2412         case KVM_IRQ_LINE_STATUS:
2413         case KVM_IRQ_LINE: {
2414                 struct kvm_irq_level irq_event;
2415
2416                 r = -EFAULT;
2417                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2418                         goto out;
2419
2420                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2421                                         ioctl == KVM_IRQ_LINE_STATUS);
2422                 if (r)
2423                         goto out;
2424
2425                 r = -EFAULT;
2426                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2427                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2428                                 goto out;
2429                 }
2430
2431                 r = 0;
2432                 break;
2433         }
2434 #endif
2435 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2436         case KVM_SET_GSI_ROUTING: {
2437                 struct kvm_irq_routing routing;
2438                 struct kvm_irq_routing __user *urouting;
2439                 struct kvm_irq_routing_entry *entries;
2440
2441                 r = -EFAULT;
2442                 if (copy_from_user(&routing, argp, sizeof(routing)))
2443                         goto out;
2444                 r = -EINVAL;
2445                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2446                         goto out;
2447                 if (routing.flags)
2448                         goto out;
2449                 r = -ENOMEM;
2450                 entries = vmalloc(routing.nr * sizeof(*entries));
2451                 if (!entries)
2452                         goto out;
2453                 r = -EFAULT;
2454                 urouting = argp;
2455                 if (copy_from_user(entries, urouting->entries,
2456                                    routing.nr * sizeof(*entries)))
2457                         goto out_free_irq_routing;
2458                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2459                                         routing.flags);
2460         out_free_irq_routing:
2461                 vfree(entries);
2462                 break;
2463         }
2464 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2465         case KVM_CREATE_DEVICE: {
2466                 struct kvm_create_device cd;
2467
2468                 r = -EFAULT;
2469                 if (copy_from_user(&cd, argp, sizeof(cd)))
2470                         goto out;
2471
2472                 r = kvm_ioctl_create_device(kvm, &cd);
2473                 if (r)
2474                         goto out;
2475
2476                 r = -EFAULT;
2477                 if (copy_to_user(argp, &cd, sizeof(cd)))
2478                         goto out;
2479
2480                 r = 0;
2481                 break;
2482         }
2483         default:
2484                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2485                 if (r == -ENOTTY)
2486                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2487         }
2488 out:
2489         return r;
2490 }
2491
2492 #ifdef CONFIG_COMPAT
2493 struct compat_kvm_dirty_log {
2494         __u32 slot;
2495         __u32 padding1;
2496         union {
2497                 compat_uptr_t dirty_bitmap; /* one bit per page */
2498                 __u64 padding2;
2499         };
2500 };
2501
2502 static long kvm_vm_compat_ioctl(struct file *filp,
2503                            unsigned int ioctl, unsigned long arg)
2504 {
2505         struct kvm *kvm = filp->private_data;
2506         int r;
2507
2508         if (kvm->mm != current->mm)
2509                 return -EIO;
2510         switch (ioctl) {
2511         case KVM_GET_DIRTY_LOG: {
2512                 struct compat_kvm_dirty_log compat_log;
2513                 struct kvm_dirty_log log;
2514
2515                 r = -EFAULT;
2516                 if (copy_from_user(&compat_log, (void __user *)arg,
2517                                    sizeof(compat_log)))
2518                         goto out;
2519                 log.slot         = compat_log.slot;
2520                 log.padding1     = compat_log.padding1;
2521                 log.padding2     = compat_log.padding2;
2522                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2523
2524                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2525                 break;
2526         }
2527         default:
2528                 r = kvm_vm_ioctl(filp, ioctl, arg);
2529         }
2530
2531 out:
2532         return r;
2533 }
2534 #endif
2535
2536 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2537 {
2538         struct page *page[1];
2539         unsigned long addr;
2540         int npages;
2541         gfn_t gfn = vmf->pgoff;
2542         struct kvm *kvm = vma->vm_file->private_data;
2543
2544         addr = gfn_to_hva(kvm, gfn);
2545         if (kvm_is_error_hva(addr))
2546                 return VM_FAULT_SIGBUS;
2547
2548         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2549                                 NULL);
2550         if (unlikely(npages != 1))
2551                 return VM_FAULT_SIGBUS;
2552
2553         vmf->page = page[0];
2554         return 0;
2555 }
2556
2557 static const struct vm_operations_struct kvm_vm_vm_ops = {
2558         .fault = kvm_vm_fault,
2559 };
2560
2561 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2562 {
2563         vma->vm_ops = &kvm_vm_vm_ops;
2564         return 0;
2565 }
2566
2567 static struct file_operations kvm_vm_fops = {
2568         .release        = kvm_vm_release,
2569         .unlocked_ioctl = kvm_vm_ioctl,
2570 #ifdef CONFIG_COMPAT
2571         .compat_ioctl   = kvm_vm_compat_ioctl,
2572 #endif
2573         .mmap           = kvm_vm_mmap,
2574         .llseek         = noop_llseek,
2575 };
2576
2577 static int kvm_dev_ioctl_create_vm(unsigned long type)
2578 {
2579         int r;
2580         struct kvm *kvm;
2581
2582         kvm = kvm_create_vm(type);
2583         if (IS_ERR(kvm))
2584                 return PTR_ERR(kvm);
2585 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2586         r = kvm_coalesced_mmio_init(kvm);
2587         if (r < 0) {
2588                 kvm_put_kvm(kvm);
2589                 return r;
2590         }
2591 #endif
2592         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2593         if (r < 0)
2594                 kvm_put_kvm(kvm);
2595
2596         return r;
2597 }
2598
2599 static long kvm_dev_ioctl_check_extension_generic(long arg)
2600 {
2601         switch (arg) {
2602         case KVM_CAP_USER_MEMORY:
2603         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2604         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2605 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2606         case KVM_CAP_SET_BOOT_CPU_ID:
2607 #endif
2608         case KVM_CAP_INTERNAL_ERROR_DATA:
2609 #ifdef CONFIG_HAVE_KVM_MSI
2610         case KVM_CAP_SIGNAL_MSI:
2611 #endif
2612 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2613         case KVM_CAP_IRQFD_RESAMPLE:
2614 #endif
2615                 return 1;
2616 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2617         case KVM_CAP_IRQ_ROUTING:
2618                 return KVM_MAX_IRQ_ROUTES;
2619 #endif
2620         default:
2621                 break;
2622         }
2623         return kvm_dev_ioctl_check_extension(arg);
2624 }
2625
2626 static long kvm_dev_ioctl(struct file *filp,
2627                           unsigned int ioctl, unsigned long arg)
2628 {
2629         long r = -EINVAL;
2630
2631         switch (ioctl) {
2632         case KVM_GET_API_VERSION:
2633                 r = -EINVAL;
2634                 if (arg)
2635                         goto out;
2636                 r = KVM_API_VERSION;
2637                 break;
2638         case KVM_CREATE_VM:
2639                 r = kvm_dev_ioctl_create_vm(arg);
2640                 break;
2641         case KVM_CHECK_EXTENSION:
2642                 r = kvm_dev_ioctl_check_extension_generic(arg);
2643                 break;
2644         case KVM_GET_VCPU_MMAP_SIZE:
2645                 r = -EINVAL;
2646                 if (arg)
2647                         goto out;
2648                 r = PAGE_SIZE;     /* struct kvm_run */
2649 #ifdef CONFIG_X86
2650                 r += PAGE_SIZE;    /* pio data page */
2651 #endif
2652 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2653                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2654 #endif
2655                 break;
2656         case KVM_TRACE_ENABLE:
2657         case KVM_TRACE_PAUSE:
2658         case KVM_TRACE_DISABLE:
2659                 r = -EOPNOTSUPP;
2660                 break;
2661         default:
2662                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2663         }
2664 out:
2665         return r;
2666 }
2667
2668 static struct file_operations kvm_chardev_ops = {
2669         .unlocked_ioctl = kvm_dev_ioctl,
2670         .compat_ioctl   = kvm_dev_ioctl,
2671         .llseek         = noop_llseek,
2672 };
2673
2674 static struct miscdevice kvm_dev = {
2675         KVM_MINOR,
2676         "kvm",
2677         &kvm_chardev_ops,
2678 };
2679
2680 static void hardware_enable_nolock(void *junk)
2681 {
2682         int cpu = raw_smp_processor_id();
2683         int r;
2684
2685         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2686                 return;
2687
2688         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2689
2690         r = kvm_arch_hardware_enable(NULL);
2691
2692         if (r) {
2693                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2694                 atomic_inc(&hardware_enable_failed);
2695                 printk(KERN_INFO "kvm: enabling virtualization on "
2696                                  "CPU%d failed\n", cpu);
2697         }
2698 }
2699
2700 static void hardware_enable(void *junk)
2701 {
2702         raw_spin_lock(&kvm_lock);
2703         hardware_enable_nolock(junk);
2704         raw_spin_unlock(&kvm_lock);
2705 }
2706
2707 static void hardware_disable_nolock(void *junk)
2708 {
2709         int cpu = raw_smp_processor_id();
2710
2711         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2712                 return;
2713         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2714         kvm_arch_hardware_disable(NULL);
2715 }
2716
2717 static void hardware_disable(void *junk)
2718 {
2719         raw_spin_lock(&kvm_lock);
2720         hardware_disable_nolock(junk);
2721         raw_spin_unlock(&kvm_lock);
2722 }
2723
2724 static void hardware_disable_all_nolock(void)
2725 {
2726         BUG_ON(!kvm_usage_count);
2727
2728         kvm_usage_count--;
2729         if (!kvm_usage_count)
2730                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2731 }
2732
2733 static void hardware_disable_all(void)
2734 {
2735         raw_spin_lock(&kvm_lock);
2736         hardware_disable_all_nolock();
2737         raw_spin_unlock(&kvm_lock);
2738 }
2739
2740 static int hardware_enable_all(void)
2741 {
2742         int r = 0;
2743
2744         raw_spin_lock(&kvm_lock);
2745
2746         kvm_usage_count++;
2747         if (kvm_usage_count == 1) {
2748                 atomic_set(&hardware_enable_failed, 0);
2749                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2750
2751                 if (atomic_read(&hardware_enable_failed)) {
2752                         hardware_disable_all_nolock();
2753                         r = -EBUSY;
2754                 }
2755         }
2756
2757         raw_spin_unlock(&kvm_lock);
2758
2759         return r;
2760 }
2761
2762 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2763                            void *v)
2764 {
2765         int cpu = (long)v;
2766
2767         if (!kvm_usage_count)
2768                 return NOTIFY_OK;
2769
2770         val &= ~CPU_TASKS_FROZEN;
2771         switch (val) {
2772         case CPU_DYING:
2773                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2774                        cpu);
2775                 hardware_disable(NULL);
2776                 break;
2777         case CPU_STARTING:
2778                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2779                        cpu);
2780                 hardware_enable(NULL);
2781                 break;
2782         }
2783         return NOTIFY_OK;
2784 }
2785
2786 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2787                       void *v)
2788 {
2789         /*
2790          * Some (well, at least mine) BIOSes hang on reboot if
2791          * in vmx root mode.
2792          *
2793          * And Intel TXT required VMX off for all cpu when system shutdown.
2794          */
2795         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2796         kvm_rebooting = true;
2797         on_each_cpu(hardware_disable_nolock, NULL, 1);
2798         return NOTIFY_OK;
2799 }
2800
2801 static struct notifier_block kvm_reboot_notifier = {
2802         .notifier_call = kvm_reboot,
2803         .priority = 0,
2804 };
2805
2806 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2807 {
2808         int i;
2809
2810         for (i = 0; i < bus->dev_count; i++) {
2811                 struct kvm_io_device *pos = bus->range[i].dev;
2812
2813                 kvm_iodevice_destructor(pos);
2814         }
2815         kfree(bus);
2816 }
2817
2818 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2819 {
2820         const struct kvm_io_range *r1 = p1;
2821         const struct kvm_io_range *r2 = p2;
2822
2823         if (r1->addr < r2->addr)
2824                 return -1;
2825         if (r1->addr + r1->len > r2->addr + r2->len)
2826                 return 1;
2827         return 0;
2828 }
2829
2830 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2831                           gpa_t addr, int len)
2832 {
2833         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2834                 .addr = addr,
2835                 .len = len,
2836                 .dev = dev,
2837         };
2838
2839         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2840                 kvm_io_bus_sort_cmp, NULL);
2841
2842         return 0;
2843 }
2844
2845 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2846                              gpa_t addr, int len)
2847 {
2848         struct kvm_io_range *range, key;
2849         int off;
2850
2851         key = (struct kvm_io_range) {
2852                 .addr = addr,
2853                 .len = len,
2854         };
2855
2856         range = bsearch(&key, bus->range, bus->dev_count,
2857                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2858         if (range == NULL)
2859                 return -ENOENT;
2860
2861         off = range - bus->range;
2862
2863         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2864                 off--;
2865
2866         return off;
2867 }
2868
2869 /* kvm_io_bus_write - called under kvm->slots_lock */
2870 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2871                      int len, const void *val)
2872 {
2873         int idx;
2874         struct kvm_io_bus *bus;
2875         struct kvm_io_range range;
2876
2877         range = (struct kvm_io_range) {
2878                 .addr = addr,
2879                 .len = len,
2880         };
2881
2882         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2883         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2884         if (idx < 0)
2885                 return -EOPNOTSUPP;
2886
2887         while (idx < bus->dev_count &&
2888                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2889                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2890                         return 0;
2891                 idx++;
2892         }
2893
2894         return -EOPNOTSUPP;
2895 }
2896
2897 /* kvm_io_bus_read - called under kvm->slots_lock */
2898 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2899                     int len, void *val)
2900 {
2901         int idx;
2902         struct kvm_io_bus *bus;
2903         struct kvm_io_range range;
2904
2905         range = (struct kvm_io_range) {
2906                 .addr = addr,
2907                 .len = len,
2908         };
2909
2910         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2911         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2912         if (idx < 0)
2913                 return -EOPNOTSUPP;
2914
2915         while (idx < bus->dev_count &&
2916                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2917                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2918                         return 0;
2919                 idx++;
2920         }
2921
2922         return -EOPNOTSUPP;
2923 }
2924
2925 /* Caller must hold slots_lock. */
2926 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2927                             int len, struct kvm_io_device *dev)
2928 {
2929         struct kvm_io_bus *new_bus, *bus;
2930
2931         bus = kvm->buses[bus_idx];
2932         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2933                 return -ENOSPC;
2934
2935         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2936                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2937         if (!new_bus)
2938                 return -ENOMEM;
2939         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2940                sizeof(struct kvm_io_range)));
2941         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2942         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2943         synchronize_srcu_expedited(&kvm->srcu);
2944         kfree(bus);
2945
2946         return 0;
2947 }
2948
2949 /* Caller must hold slots_lock. */
2950 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2951                               struct kvm_io_device *dev)
2952 {
2953         int i, r;
2954         struct kvm_io_bus *new_bus, *bus;
2955
2956         bus = kvm->buses[bus_idx];
2957         r = -ENOENT;
2958         for (i = 0; i < bus->dev_count; i++)
2959                 if (bus->range[i].dev == dev) {
2960                         r = 0;
2961                         break;
2962                 }
2963
2964         if (r)
2965                 return r;
2966
2967         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2968                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2969         if (!new_bus)
2970                 return -ENOMEM;
2971
2972         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2973         new_bus->dev_count--;
2974         memcpy(new_bus->range + i, bus->range + i + 1,
2975                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2976
2977         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2978         synchronize_srcu_expedited(&kvm->srcu);
2979         kfree(bus);
2980         return r;
2981 }
2982
2983 static struct notifier_block kvm_cpu_notifier = {
2984         .notifier_call = kvm_cpu_hotplug,
2985 };
2986
2987 static int vm_stat_get(void *_offset, u64 *val)
2988 {
2989         unsigned offset = (long)_offset;
2990         struct kvm *kvm;
2991
2992         *val = 0;
2993         raw_spin_lock(&kvm_lock);
2994         list_for_each_entry(kvm, &vm_list, vm_list)
2995                 *val += *(u32 *)((void *)kvm + offset);
2996         raw_spin_unlock(&kvm_lock);
2997         return 0;
2998 }
2999
3000 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3001
3002 static int vcpu_stat_get(void *_offset, u64 *val)
3003 {
3004         unsigned offset = (long)_offset;
3005         struct kvm *kvm;
3006         struct kvm_vcpu *vcpu;
3007         int i;
3008
3009         *val = 0;
3010         raw_spin_lock(&kvm_lock);
3011         list_for_each_entry(kvm, &vm_list, vm_list)
3012                 kvm_for_each_vcpu(i, vcpu, kvm)
3013                         *val += *(u32 *)((void *)vcpu + offset);
3014
3015         raw_spin_unlock(&kvm_lock);
3016         return 0;
3017 }
3018
3019 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3020
3021 static const struct file_operations *stat_fops[] = {
3022         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3023         [KVM_STAT_VM]   = &vm_stat_fops,
3024 };
3025
3026 static int kvm_init_debug(void)
3027 {
3028         int r = -EFAULT;
3029         struct kvm_stats_debugfs_item *p;
3030
3031         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3032         if (kvm_debugfs_dir == NULL)
3033                 goto out;
3034
3035         for (p = debugfs_entries; p->name; ++p) {
3036                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3037                                                 (void *)(long)p->offset,
3038                                                 stat_fops[p->kind]);
3039                 if (p->dentry == NULL)
3040                         goto out_dir;
3041         }
3042
3043         return 0;
3044
3045 out_dir:
3046         debugfs_remove_recursive(kvm_debugfs_dir);
3047 out:
3048         return r;
3049 }
3050
3051 static void kvm_exit_debug(void)
3052 {
3053         struct kvm_stats_debugfs_item *p;
3054
3055         for (p = debugfs_entries; p->name; ++p)
3056                 debugfs_remove(p->dentry);
3057         debugfs_remove(kvm_debugfs_dir);
3058 }
3059
3060 static int kvm_suspend(void)
3061 {
3062         if (kvm_usage_count)
3063                 hardware_disable_nolock(NULL);
3064         return 0;
3065 }
3066
3067 static void kvm_resume(void)
3068 {
3069         if (kvm_usage_count) {
3070                 WARN_ON(raw_spin_is_locked(&kvm_lock));
3071                 hardware_enable_nolock(NULL);
3072         }
3073 }
3074
3075 static struct syscore_ops kvm_syscore_ops = {
3076         .suspend = kvm_suspend,
3077         .resume = kvm_resume,
3078 };
3079
3080 static inline
3081 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3082 {
3083         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3084 }
3085
3086 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3087 {
3088         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3089         if (vcpu->preempted)
3090                 vcpu->preempted = false;
3091
3092         kvm_arch_vcpu_load(vcpu, cpu);
3093 }
3094
3095 static void kvm_sched_out(struct preempt_notifier *pn,
3096                           struct task_struct *next)
3097 {
3098         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3099
3100         if (current->state == TASK_RUNNING)
3101                 vcpu->preempted = true;
3102         kvm_arch_vcpu_put(vcpu);
3103 }
3104
3105 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3106                   struct module *module)
3107 {
3108         int r;
3109         int cpu;
3110
3111         r = kvm_arch_init(opaque);
3112         if (r)
3113                 goto out_fail;
3114
3115         /*
3116          * kvm_arch_init makes sure there's at most one caller
3117          * for architectures that support multiple implementations,
3118          * like intel and amd on x86.
3119          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3120          * conflicts in case kvm is already setup for another implementation.
3121          */
3122         r = kvm_irqfd_init();
3123         if (r)
3124                 goto out_irqfd;
3125
3126         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3127                 r = -ENOMEM;
3128                 goto out_free_0;
3129         }
3130
3131         r = kvm_arch_hardware_setup();
3132         if (r < 0)
3133                 goto out_free_0a;
3134
3135         for_each_online_cpu(cpu) {
3136                 smp_call_function_single(cpu,
3137                                 kvm_arch_check_processor_compat,
3138                                 &r, 1);
3139                 if (r < 0)
3140                         goto out_free_1;
3141         }
3142
3143         r = register_cpu_notifier(&kvm_cpu_notifier);
3144         if (r)
3145                 goto out_free_2;
3146         register_reboot_notifier(&kvm_reboot_notifier);
3147
3148         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3149         if (!vcpu_align)
3150                 vcpu_align = __alignof__(struct kvm_vcpu);
3151         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3152                                            0, NULL);
3153         if (!kvm_vcpu_cache) {
3154                 r = -ENOMEM;
3155                 goto out_free_3;
3156         }
3157
3158         r = kvm_async_pf_init();
3159         if (r)
3160                 goto out_free;
3161
3162         kvm_chardev_ops.owner = module;
3163         kvm_vm_fops.owner = module;
3164         kvm_vcpu_fops.owner = module;
3165
3166         r = misc_register(&kvm_dev);
3167         if (r) {
3168                 printk(KERN_ERR "kvm: misc device register failed\n");
3169                 goto out_unreg;
3170         }
3171
3172         register_syscore_ops(&kvm_syscore_ops);
3173
3174         kvm_preempt_ops.sched_in = kvm_sched_in;
3175         kvm_preempt_ops.sched_out = kvm_sched_out;
3176
3177         r = kvm_init_debug();
3178         if (r) {
3179                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3180                 goto out_undebugfs;
3181         }
3182
3183         return 0;
3184
3185 out_undebugfs:
3186         unregister_syscore_ops(&kvm_syscore_ops);
3187 out_unreg:
3188         kvm_async_pf_deinit();
3189 out_free:
3190         kmem_cache_destroy(kvm_vcpu_cache);
3191 out_free_3:
3192         unregister_reboot_notifier(&kvm_reboot_notifier);
3193         unregister_cpu_notifier(&kvm_cpu_notifier);
3194 out_free_2:
3195 out_free_1:
3196         kvm_arch_hardware_unsetup();
3197 out_free_0a:
3198         free_cpumask_var(cpus_hardware_enabled);
3199 out_free_0:
3200         kvm_irqfd_exit();
3201 out_irqfd:
3202         kvm_arch_exit();
3203 out_fail:
3204         return r;
3205 }
3206 EXPORT_SYMBOL_GPL(kvm_init);
3207
3208 void kvm_exit(void)
3209 {
3210         kvm_exit_debug();
3211         misc_deregister(&kvm_dev);
3212         kmem_cache_destroy(kvm_vcpu_cache);
3213         kvm_async_pf_deinit();
3214         unregister_syscore_ops(&kvm_syscore_ops);
3215         unregister_reboot_notifier(&kvm_reboot_notifier);
3216         unregister_cpu_notifier(&kvm_cpu_notifier);
3217         on_each_cpu(hardware_disable_nolock, NULL, 1);
3218         kvm_arch_hardware_unsetup();
3219         kvm_arch_exit();
3220         kvm_irqfd_exit();
3221         free_cpumask_var(cpus_hardware_enabled);
3222 }
3223 EXPORT_SYMBOL_GPL(kvm_exit);