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_SPINLOCK(kvm_lock);
  74 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
  75 LIST_HEAD(vm_list);
  76
  77 static cpumask_var_t cpus_hardware_enabled;
  78 static int kvm_usage_count = 0;
  79 static atomic_t hardware_enable_failed;
  80
  81 struct kmem_cache *kvm_vcpu_cache;
  82 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  83
  84 static __read_mostly struct preempt_ops kvm_preempt_ops;
  85
  86 struct dentry *kvm_debugfs_dir;
  87
  88 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  89                            unsigned long arg);
  90 #ifdef CONFIG_COMPAT
  91 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  92                                   unsigned long arg);
  93 #endif
  94 static int hardware_enable_all(void);
  95 static void hardware_disable_all(void);
  96
  97 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  98 static void update_memslots(struct kvm_memslots *slots,
  99                             struct kvm_memory_slot *new, u64 last_generation);
 100
 101 static void kvm_release_pfn_dirty(pfn_t pfn);
 102 static void mark_page_dirty_in_slot(struct kvm *kvm,
 103                                     struct kvm_memory_slot *memslot, gfn_t gfn);
 104
 105 __visible bool kvm_rebooting;
 106 EXPORT_SYMBOL_GPL(kvm_rebooting);
 107
 108 static bool largepages_enabled = true;
 109
 110 bool kvm_is_mmio_pfn(pfn_t pfn)
 111 {
 112         if (pfn_valid(pfn))
 113                 return PageReserved(pfn_to_page(pfn));
 114
 115         return true;
 116 }
 117
 118 /*
 119  * Switches to specified vcpu, until a matching vcpu_put()
 120  */
 121 int vcpu_load(struct kvm_vcpu *vcpu)
 122 {
 123         int cpu;
 124
 125         if (mutex_lock_killable(&vcpu->mutex))
 126                 return -EINTR;
 127         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 128                 /* The thread running this VCPU changed. */
 129                 struct pid *oldpid = vcpu->pid;
 130                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 131                 rcu_assign_pointer(vcpu->pid, newpid);
 132                 synchronize_rcu();
 133                 put_pid(oldpid);
 134         }
 135         cpu = get_cpu();
 136         preempt_notifier_register(&vcpu->preempt_notifier);
 137         kvm_arch_vcpu_load(vcpu, cpu);
 138         put_cpu();
 139         return 0;
 140 }
 141
 142 void vcpu_put(struct kvm_vcpu *vcpu)
 143 {
 144         preempt_disable();
 145         kvm_arch_vcpu_put(vcpu);
 146         preempt_notifier_unregister(&vcpu->preempt_notifier);
 147         preempt_enable();
 148         mutex_unlock(&vcpu->mutex);
 149 }
 150
 151 static void ack_flush(void *_completed)
 152 {
 153 }
 154
 155 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 156 {
 157         int i, cpu, me;
 158         cpumask_var_t cpus;
 159         bool called = true;
 160         struct kvm_vcpu *vcpu;
 161
 162         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 163
 164         me = get_cpu();
 165         kvm_for_each_vcpu(i, vcpu, kvm) {
 166                 kvm_make_request(req, vcpu);
 167                 cpu = vcpu->cpu;
 168
 169                 /* Set ->requests bit before we read ->mode */
 170                 smp_mb();
 171
 172                 if (cpus != NULL && cpu != -1 && cpu != me &&
 173                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 174                         cpumask_set_cpu(cpu, cpus);
 175         }
 176         if (unlikely(cpus == NULL))
 177                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 178         else if (!cpumask_empty(cpus))
 179                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 180         else
 181                 called = false;
 182         put_cpu();
 183         free_cpumask_var(cpus);
 184         return called;
 185 }
 186
 187 void kvm_flush_remote_tlbs(struct kvm *kvm)
 188 {
 189         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 190                 ++kvm->stat.remote_tlb_flush;
 191         kvm->tlbs_dirty = false;
 192 }
 193 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 194
 195 void kvm_reload_remote_mmus(struct kvm *kvm)
 196 {
 197         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 198 }
 199
 200 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 201 {
 202         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 203 }
 204
 205 void kvm_make_scan_ioapic_request(struct kvm *kvm)
 206 {
 207         make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
 208 }
 209
 210 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 211 {
 212         struct page *page;
 213         int r;
 214
 215         mutex_init(&vcpu->mutex);
 216         vcpu->cpu = -1;
 217         vcpu->kvm = kvm;
 218         vcpu->vcpu_id = id;
 219         vcpu->pid = NULL;
 220         init_waitqueue_head(&vcpu->wq);
 221         kvm_async_pf_vcpu_init(vcpu);
 222
 223         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 224         if (!page) {
 225                 r = -ENOMEM;
 226                 goto fail;
 227         }
 228         vcpu->run = page_address(page);
 229
 230         kvm_vcpu_set_in_spin_loop(vcpu, false);
 231         kvm_vcpu_set_dy_eligible(vcpu, false);
 232         vcpu->preempted = false;
 233
 234         r = kvm_arch_vcpu_init(vcpu);
 235         if (r < 0)
 236                 goto fail_free_run;
 237         return 0;
 238
 239 fail_free_run:
 240         free_page((unsigned long)vcpu->run);
 241 fail:
 242         return r;
 243 }
 244 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 245
 246 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 247 {
 248         put_pid(vcpu->pid);
 249         kvm_arch_vcpu_uninit(vcpu);
 250         free_page((unsigned long)vcpu->run);
 251 }
 252 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 253
 254 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 255 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 256 {
 257         return container_of(mn, struct kvm, mmu_notifier);
 258 }
 259
 260 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 261                                              struct mm_struct *mm,
 262                                              unsigned long address)
 263 {
 264         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 265         int need_tlb_flush, idx;
 266
 267         /*
 268          * When ->invalidate_page runs, the linux pte has been zapped
 269          * already but the page is still allocated until
 270          * ->invalidate_page returns. So if we increase the sequence
 271          * here the kvm page fault will notice if the spte can't be
 272          * established because the page is going to be freed. If
 273          * instead the kvm page fault establishes the spte before
 274          * ->invalidate_page runs, kvm_unmap_hva will release it
 275          * before returning.
 276          *
 277          * The sequence increase only need to be seen at spin_unlock
 278          * time, and not at spin_lock time.
 279          *
 280          * Increasing the sequence after the spin_unlock would be
 281          * unsafe because the kvm page fault could then establish the
 282          * pte after kvm_unmap_hva returned, without noticing the page
 283          * is going to be freed.
 284          */
 285         idx = srcu_read_lock(&kvm->srcu);
 286         spin_lock(&kvm->mmu_lock);
 287
 288         kvm->mmu_notifier_seq++;
 289         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 290         /* we've to flush the tlb before the pages can be freed */
 291         if (need_tlb_flush)
 292                 kvm_flush_remote_tlbs(kvm);
 293
 294         spin_unlock(&kvm->mmu_lock);
 295         srcu_read_unlock(&kvm->srcu, idx);
 296 }
 297
 298 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 299                                         struct mm_struct *mm,
 300                                         unsigned long address,
 301                                         pte_t pte)
 302 {
 303         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 304         int idx;
 305
 306         idx = srcu_read_lock(&kvm->srcu);
 307         spin_lock(&kvm->mmu_lock);
 308         kvm->mmu_notifier_seq++;
 309         kvm_set_spte_hva(kvm, address, pte);
 310         spin_unlock(&kvm->mmu_lock);
 311         srcu_read_unlock(&kvm->srcu, idx);
 312 }
 313
 314 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 315                                                     struct mm_struct *mm,
 316                                                     unsigned long start,
 317                                                     unsigned long end)
 318 {
 319         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 320         int need_tlb_flush = 0, idx;
 321
 322         idx = srcu_read_lock(&kvm->srcu);
 323         spin_lock(&kvm->mmu_lock);
 324         /*
 325          * The count increase must become visible at unlock time as no
 326          * spte can be established without taking the mmu_lock and
 327          * count is also read inside the mmu_lock critical section.
 328          */
 329         kvm->mmu_notifier_count++;
 330         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 331         need_tlb_flush |= kvm->tlbs_dirty;
 332         /* we've to flush the tlb before the pages can be freed */
 333         if (need_tlb_flush)
 334                 kvm_flush_remote_tlbs(kvm);
 335
 336         spin_unlock(&kvm->mmu_lock);
 337         srcu_read_unlock(&kvm->srcu, idx);
 338 }
 339
 340 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 341                                                   struct mm_struct *mm,
 342                                                   unsigned long start,
 343                                                   unsigned long end)
 344 {
 345         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 346
 347         spin_lock(&kvm->mmu_lock);
 348         /*
 349          * This sequence increase will notify the kvm page fault that
 350          * the page that is going to be mapped in the spte could have
 351          * been freed.
 352          */
 353         kvm->mmu_notifier_seq++;
 354         smp_wmb();
 355         /*
 356          * The above sequence increase must be visible before the
 357          * below count decrease, which is ensured by the smp_wmb above
 358          * in conjunction with the smp_rmb in mmu_notifier_retry().
 359          */
 360         kvm->mmu_notifier_count--;
 361         spin_unlock(&kvm->mmu_lock);
 362
 363         BUG_ON(kvm->mmu_notifier_count < 0);
 364 }
 365
 366 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 367                                               struct mm_struct *mm,
 368                                               unsigned long address)
 369 {
 370         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 371         int young, idx;
 372
 373         idx = srcu_read_lock(&kvm->srcu);
 374         spin_lock(&kvm->mmu_lock);
 375
 376         young = kvm_age_hva(kvm, address);
 377         if (young)
 378                 kvm_flush_remote_tlbs(kvm);
 379
 380         spin_unlock(&kvm->mmu_lock);
 381         srcu_read_unlock(&kvm->srcu, idx);
 382
 383         return young;
 384 }
 385
 386 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 387                                        struct mm_struct *mm,
 388                                        unsigned long address)
 389 {
 390         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 391         int young, idx;
 392
 393         idx = srcu_read_lock(&kvm->srcu);
 394         spin_lock(&kvm->mmu_lock);
 395         young = kvm_test_age_hva(kvm, address);
 396         spin_unlock(&kvm->mmu_lock);
 397         srcu_read_unlock(&kvm->srcu, idx);
 398
 399         return young;
 400 }
 401
 402 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 403                                      struct mm_struct *mm)
 404 {
 405         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 406         int idx;
 407
 408         idx = srcu_read_lock(&kvm->srcu);
 409         kvm_arch_flush_shadow_all(kvm);
 410         srcu_read_unlock(&kvm->srcu, idx);
 411 }
 412
 413 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 414         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 415         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 416         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 417         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 418         .test_young             = kvm_mmu_notifier_test_young,
 419         .change_pte             = kvm_mmu_notifier_change_pte,
 420         .release                = kvm_mmu_notifier_release,
 421 };
 422
 423 static int kvm_init_mmu_notifier(struct kvm *kvm)
 424 {
 425         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 426         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 427 }
 428
 429 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 430
 431 static int kvm_init_mmu_notifier(struct kvm *kvm)
 432 {
 433         return 0;
 434 }
 435
 436 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 437
 438 static void kvm_init_memslots_id(struct kvm *kvm)
 439 {
 440         int i;
 441         struct kvm_memslots *slots = kvm->memslots;
 442
 443         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 444                 slots->id_to_index[i] = slots->memslots[i].id = i;
 445 }
 446
 447 static struct kvm *kvm_create_vm(unsigned long type)
 448 {
 449         int r, i;
 450         struct kvm *kvm = kvm_arch_alloc_vm();
 451
 452         if (!kvm)
 453                 return ERR_PTR(-ENOMEM);
 454
 455         r = kvm_arch_init_vm(kvm, type);
 456         if (r)
 457                 goto out_err_nodisable;
 458
 459         r = hardware_enable_all();
 460         if (r)
 461                 goto out_err_nodisable;
 462
 463 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 464         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 465         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 466 #endif
 467
 468         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 469
 470         r = -ENOMEM;
 471         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 472         if (!kvm->memslots)
 473                 goto out_err_nosrcu;
 474         kvm_init_memslots_id(kvm);
 475         if (init_srcu_struct(&kvm->srcu))
 476                 goto out_err_nosrcu;
 477         for (i = 0; i < KVM_NR_BUSES; i++) {
 478                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 479                                         GFP_KERNEL);
 480                 if (!kvm->buses[i])
 481                         goto out_err;
 482         }
 483
 484         spin_lock_init(&kvm->mmu_lock);
 485         kvm->mm = current->mm;
 486         atomic_inc(&kvm->mm->mm_count);
 487         kvm_eventfd_init(kvm);
 488         mutex_init(&kvm->lock);
 489         mutex_init(&kvm->irq_lock);
 490         mutex_init(&kvm->slots_lock);
 491         atomic_set(&kvm->users_count, 1);
 492         INIT_LIST_HEAD(&kvm->devices);
 493
 494         r = kvm_init_mmu_notifier(kvm);
 495         if (r)
 496                 goto out_err;
 497
 498         spin_lock(&kvm_lock);
 499         list_add(&kvm->vm_list, &vm_list);
 500         spin_unlock(&kvm_lock);
 501
 502         return kvm;
 503
 504 out_err:
 505         cleanup_srcu_struct(&kvm->srcu);
 506 out_err_nosrcu:
 507         hardware_disable_all();
 508 out_err_nodisable:
 509         for (i = 0; i < KVM_NR_BUSES; i++)
 510                 kfree(kvm->buses[i]);
 511         kfree(kvm->memslots);
 512         kvm_arch_free_vm(kvm);
 513         return ERR_PTR(r);
 514 }
 515
 516 /*
 517  * Avoid using vmalloc for a small buffer.
 518  * Should not be used when the size is statically known.
 519  */
 520 void *kvm_kvzalloc(unsigned long size)
 521 {
 522         if (size > PAGE_SIZE)
 523                 return vzalloc(size);
 524         else
 525                 return kzalloc(size, GFP_KERNEL);
 526 }
 527
 528 void kvm_kvfree(const void *addr)
 529 {
 530         if (is_vmalloc_addr(addr))
 531                 vfree(addr);
 532         else
 533                 kfree(addr);
 534 }
 535
 536 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 537 {
 538         if (!memslot->dirty_bitmap)
 539                 return;
 540
 541         kvm_kvfree(memslot->dirty_bitmap);
 542         memslot->dirty_bitmap = NULL;
 543 }
 544
 545 /*
 546  * Free any memory in @free but not in @dont.
 547  */
 548 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
 549                                   struct kvm_memory_slot *dont)
 550 {
 551         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 552                 kvm_destroy_dirty_bitmap(free);
 553
 554         kvm_arch_free_memslot(kvm, free, dont);
 555
 556         free->npages = 0;
 557 }
 558
 559 static void kvm_free_physmem(struct kvm *kvm)
 560 {
 561         struct kvm_memslots *slots = kvm->memslots;
 562         struct kvm_memory_slot *memslot;
 563
 564         kvm_for_each_memslot(memslot, slots)
 565                 kvm_free_physmem_slot(kvm, memslot, NULL);
 566
 567         kfree(kvm->memslots);
 568 }
 569
 570 static void kvm_destroy_devices(struct kvm *kvm)
 571 {
 572         struct list_head *node, *tmp;
 573
 574         list_for_each_safe(node, tmp, &kvm->devices) {
 575                 struct kvm_device *dev =
 576                         list_entry(node, struct kvm_device, vm_node);
 577
 578                 list_del(node);
 579                 dev->ops->destroy(dev);
 580         }
 581 }
 582
 583 static void kvm_destroy_vm(struct kvm *kvm)
 584 {
 585         int i;
 586         struct mm_struct *mm = kvm->mm;
 587
 588         kvm_arch_sync_events(kvm);
 589         spin_lock(&kvm_lock);
 590         list_del(&kvm->vm_list);
 591         spin_unlock(&kvm_lock);
 592         kvm_free_irq_routing(kvm);
 593         for (i = 0; i < KVM_NR_BUSES; i++)
 594                 kvm_io_bus_destroy(kvm->buses[i]);
 595         kvm_coalesced_mmio_free(kvm);
 596 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 597         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 598 #else
 599         kvm_arch_flush_shadow_all(kvm);
 600 #endif
 601         kvm_arch_destroy_vm(kvm);
 602         kvm_destroy_devices(kvm);
 603         kvm_free_physmem(kvm);
 604         cleanup_srcu_struct(&kvm->srcu);
 605         kvm_arch_free_vm(kvm);
 606         hardware_disable_all();
 607         mmdrop(mm);
 608 }
 609
 610 void kvm_get_kvm(struct kvm *kvm)
 611 {
 612         atomic_inc(&kvm->users_count);
 613 }
 614 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 615
 616 void kvm_put_kvm(struct kvm *kvm)
 617 {
 618         if (atomic_dec_and_test(&kvm->users_count))
 619                 kvm_destroy_vm(kvm);
 620 }
 621 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 622
 623
 624 static int kvm_vm_release(struct inode *inode, struct file *filp)
 625 {
 626         struct kvm *kvm = filp->private_data;
 627
 628         kvm_irqfd_release(kvm);
 629
 630         kvm_put_kvm(kvm);
 631         return 0;
 632 }
 633
 634 /*
 635  * Allocation size is twice as large as the actual dirty bitmap size.
 636  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 637  */
 638 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 639 {
 640         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 641
 642         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 643         if (!memslot->dirty_bitmap)
 644                 return -ENOMEM;
 645
 646         return 0;
 647 }
 648
 649 static int cmp_memslot(const void *slot1, const void *slot2)
 650 {
 651         struct kvm_memory_slot *s1, *s2;
 652
 653         s1 = (struct kvm_memory_slot *)slot1;
 654         s2 = (struct kvm_memory_slot *)slot2;
 655
 656         if (s1->npages < s2->npages)
 657                 return 1;
 658         if (s1->npages > s2->npages)
 659                 return -1;
 660
 661         return 0;
 662 }
 663
 664 /*
 665  * Sort the memslots base on its size, so the larger slots
 666  * will get better fit.
 667  */
 668 static void sort_memslots(struct kvm_memslots *slots)
 669 {
 670         int i;
 671
 672         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 673               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 674
 675         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 676                 slots->id_to_index[slots->memslots[i].id] = i;
 677 }
 678
 679 static void update_memslots(struct kvm_memslots *slots,
 680                             struct kvm_memory_slot *new,
 681                             u64 last_generation)
 682 {
 683         if (new) {
 684                 int id = new->id;
 685                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 686                 unsigned long npages = old->npages;
 687
 688                 *old = *new;
 689                 if (new->npages != npages)
 690                         sort_memslots(slots);
 691         }
 692
 693         slots->generation = last_generation + 1;
 694 }
 695
 696 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 697 {
 698         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 699
 700 #ifdef KVM_CAP_READONLY_MEM
 701         valid_flags |= KVM_MEM_READONLY;
 702 #endif
 703
 704         if (mem->flags & ~valid_flags)
 705                 return -EINVAL;
 706
 707         return 0;
 708 }
 709
 710 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 711                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 712 {
 713         struct kvm_memslots *old_memslots = kvm->memslots;
 714
 715         update_memslots(slots, new, kvm->memslots->generation);
 716         rcu_assign_pointer(kvm->memslots, slots);
 717         synchronize_srcu_expedited(&kvm->srcu);
 718
 719         kvm_arch_memslots_updated(kvm);
 720
 721         return old_memslots;
 722 }
 723
 724 /*
 725  * Allocate some memory and give it an address in the guest physical address
 726  * space.
 727  *
 728  * Discontiguous memory is allowed, mostly for framebuffers.
 729  *
 730  * Must be called holding mmap_sem for write.
 731  */
 732 int __kvm_set_memory_region(struct kvm *kvm,
 733                             struct kvm_userspace_memory_region *mem)
 734 {
 735         int r;
 736         gfn_t base_gfn;
 737         unsigned long npages;
 738         struct kvm_memory_slot *slot;
 739         struct kvm_memory_slot old, new;
 740         struct kvm_memslots *slots = NULL, *old_memslots;
 741         enum kvm_mr_change change;
 742
 743         r = check_memory_region_flags(mem);
 744         if (r)
 745                 goto out;
 746
 747         r = -EINVAL;
 748         /* General sanity checks */
 749         if (mem->memory_size & (PAGE_SIZE - 1))
 750                 goto out;
 751         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 752                 goto out;
 753         /* We can read the guest memory with __xxx_user() later on. */
 754         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 755             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 756              !access_ok(VERIFY_WRITE,
 757                         (void __user *)(unsigned long)mem->userspace_addr,
 758                         mem->memory_size)))
 759                 goto out;
 760         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 761                 goto out;
 762         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 763                 goto out;
 764
 765         slot = id_to_memslot(kvm->memslots, mem->slot);
 766         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 767         npages = mem->memory_size >> PAGE_SHIFT;
 768
 769         r = -EINVAL;
 770         if (npages > KVM_MEM_MAX_NR_PAGES)
 771                 goto out;
 772
 773         if (!npages)
 774                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 775
 776         new = old = *slot;
 777
 778         new.id = mem->slot;
 779         new.base_gfn = base_gfn;
 780         new.npages = npages;
 781         new.flags = mem->flags;
 782
 783         r = -EINVAL;
 784         if (npages) {
 785                 if (!old.npages)
 786                         change = KVM_MR_CREATE;
 787                 else { /* Modify an existing slot. */
 788                         if ((mem->userspace_addr != old.userspace_addr) ||
 789                             (npages != old.npages) ||
 790                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 791                                 goto out;
 792
 793                         if (base_gfn != old.base_gfn)
 794                                 change = KVM_MR_MOVE;
 795                         else if (new.flags != old.flags)
 796                                 change = KVM_MR_FLAGS_ONLY;
 797                         else { /* Nothing to change. */
 798                                 r = 0;
 799                                 goto out;
 800                         }
 801                 }
 802         } else if (old.npages) {
 803                 change = KVM_MR_DELETE;
 804         } else /* Modify a non-existent slot: disallowed. */
 805                 goto out;
 806
 807         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 808                 /* Check for overlaps */
 809                 r = -EEXIST;
 810                 kvm_for_each_memslot(slot, kvm->memslots) {
 811                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 812                             (slot->id == mem->slot))
 813                                 continue;
 814                         if (!((base_gfn + npages <= slot->base_gfn) ||
 815                               (base_gfn >= slot->base_gfn + slot->npages)))
 816                                 goto out;
 817                 }
 818         }
 819
 820         /* Free page dirty bitmap if unneeded */
 821         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 822                 new.dirty_bitmap = NULL;
 823
 824         r = -ENOMEM;
 825         if (change == KVM_MR_CREATE) {
 826                 new.userspace_addr = mem->userspace_addr;
 827
 828                 if (kvm_arch_create_memslot(kvm, &new, npages))
 829                         goto out_free;
 830         }
 831
 832         /* Allocate page dirty bitmap if needed */
 833         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 834                 if (kvm_create_dirty_bitmap(&new) < 0)
 835                         goto out_free;
 836         }
 837
 838         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 839                 r = -ENOMEM;
 840                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 841                                 GFP_KERNEL);
 842                 if (!slots)
 843                         goto out_free;
 844                 slot = id_to_memslot(slots, mem->slot);
 845                 slot->flags |= KVM_MEMSLOT_INVALID;
 846
 847                 old_memslots = install_new_memslots(kvm, slots, NULL);
 848
 849                 /* slot was deleted or moved, clear iommu mapping */
 850                 kvm_iommu_unmap_pages(kvm, &old);
 851                 /* From this point no new shadow pages pointing to a deleted,
 852                  * or moved, memslot will be created.
 853                  *
 854                  * validation of sp->gfn happens in:
 855                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 856                  *      - kvm_is_visible_gfn (mmu_check_roots)
 857                  */
 858                 kvm_arch_flush_shadow_memslot(kvm, slot);
 859                 slots = old_memslots;
 860         }
 861
 862         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 863         if (r)
 864                 goto out_slots;
 865
 866         r = -ENOMEM;
 867         /*
 868          * We can re-use the old_memslots from above, the only difference
 869          * from the currently installed memslots is the invalid flag.  This
 870          * will get overwritten by update_memslots anyway.
 871          */
 872         if (!slots) {
 873                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 874                                 GFP_KERNEL);
 875                 if (!slots)
 876                         goto out_free;
 877         }
 878
 879         /* actual memory is freed via old in kvm_free_physmem_slot below */
 880         if (change == KVM_MR_DELETE) {
 881                 new.dirty_bitmap = NULL;
 882                 memset(&new.arch, 0, sizeof(new.arch));
 883         }
 884
 885         old_memslots = install_new_memslots(kvm, slots, &new);
 886
 887         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 888
 889         kvm_free_physmem_slot(kvm, &old, &new);
 890         kfree(old_memslots);
 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                 return r;
 904         }
 905
 906         return 0;
 907
 908 out_slots:
 909         kfree(slots);
 910 out_free:
 911         kvm_free_physmem_slot(kvm, &new, &old);
 912 out:
 913         return r;
 914 }
 915 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 916
 917 int kvm_set_memory_region(struct kvm *kvm,
 918                           struct kvm_userspace_memory_region *mem)
 919 {
 920         int r;
 921
 922         mutex_lock(&kvm->slots_lock);
 923         r = __kvm_set_memory_region(kvm, mem);
 924         mutex_unlock(&kvm->slots_lock);
 925         return r;
 926 }
 927 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 928
 929 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 930                                           struct kvm_userspace_memory_region *mem)
 931 {
 932         if (mem->slot >= KVM_USER_MEM_SLOTS)
 933                 return -EINVAL;
 934         return kvm_set_memory_region(kvm, mem);
 935 }
 936
 937 int kvm_get_dirty_log(struct kvm *kvm,
 938                         struct kvm_dirty_log *log, int *is_dirty)
 939 {
 940         struct kvm_memory_slot *memslot;
 941         int r, i;
 942         unsigned long n;
 943         unsigned long any = 0;
 944
 945         r = -EINVAL;
 946         if (log->slot >= KVM_USER_MEM_SLOTS)
 947                 goto out;
 948
 949         memslot = id_to_memslot(kvm->memslots, log->slot);
 950         r = -ENOENT;
 951         if (!memslot->dirty_bitmap)
 952                 goto out;
 953
 954         n = kvm_dirty_bitmap_bytes(memslot);
 955
 956         for (i = 0; !any && i < n/sizeof(long); ++i)
 957                 any = memslot->dirty_bitmap[i];
 958
 959         r = -EFAULT;
 960         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 961                 goto out;
 962
 963         if (any)
 964                 *is_dirty = 1;
 965
 966         r = 0;
 967 out:
 968         return r;
 969 }
 970 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
 971
 972 bool kvm_largepages_enabled(void)
 973 {
 974         return largepages_enabled;
 975 }
 976
 977 void kvm_disable_largepages(void)
 978 {
 979         largepages_enabled = false;
 980 }
 981 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 982
 983 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 984 {
 985         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 986 }
 987 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 988
 989 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 990 {
 991         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 992
 993         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
 994               memslot->flags & KVM_MEMSLOT_INVALID)
 995                 return 0;
 996
 997         return 1;
 998 }
 999 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1000
1001 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1002 {
1003         struct vm_area_struct *vma;
1004         unsigned long addr, size;
1005
1006         size = PAGE_SIZE;
1007
1008         addr = gfn_to_hva(kvm, gfn);
1009         if (kvm_is_error_hva(addr))
1010                 return PAGE_SIZE;
1011
1012         down_read(&current->mm->mmap_sem);
1013         vma = find_vma(current->mm, addr);
1014         if (!vma)
1015                 goto out;
1016
1017         size = vma_kernel_pagesize(vma);
1018
1019 out:
1020         up_read(&current->mm->mmap_sem);
1021
1022         return size;
1023 }
1024
1025 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1026 {
1027         return slot->flags & KVM_MEM_READONLY;
1028 }
1029
1030 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1031                                        gfn_t *nr_pages, bool write)
1032 {
1033         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1034                 return KVM_HVA_ERR_BAD;
1035
1036         if (memslot_is_readonly(slot) && write)
1037                 return KVM_HVA_ERR_RO_BAD;
1038
1039         if (nr_pages)
1040                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1041
1042         return __gfn_to_hva_memslot(slot, gfn);
1043 }
1044
1045 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1046                                      gfn_t *nr_pages)
1047 {
1048         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1049 }
1050
1051 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1052                                         gfn_t gfn)
1053 {
1054         return gfn_to_hva_many(slot, gfn, NULL);
1055 }
1056 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1057
1058 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1059 {
1060         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1061 }
1062 EXPORT_SYMBOL_GPL(gfn_to_hva);
1063
1064 /*
1065  * If writable is set to false, the hva returned by this function is only
1066  * allowed to be read.
1067  */
1068 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1069 {
1070         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1071         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1072
1073         if (!kvm_is_error_hva(hva) && writable)
1074                 *writable = !memslot_is_readonly(slot);
1075
1076         return hva;
1077 }
1078
1079 static int kvm_read_hva(void *data, void __user *hva, int len)
1080 {
1081         return __copy_from_user(data, hva, len);
1082 }
1083
1084 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1085 {
1086         return __copy_from_user_inatomic(data, hva, len);
1087 }
1088
1089 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1090         unsigned long start, int write, struct page **page)
1091 {
1092         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1093
1094         if (write)
1095                 flags |= FOLL_WRITE;
1096
1097         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1098 }
1099
1100 static inline int check_user_page_hwpoison(unsigned long addr)
1101 {
1102         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1103
1104         rc = __get_user_pages(current, current->mm, addr, 1,
1105                               flags, NULL, NULL, NULL);
1106         return rc == -EHWPOISON;
1107 }
1108
1109 /*
1110  * The atomic path to get the writable pfn which will be stored in @pfn,
1111  * true indicates success, otherwise false is returned.
1112  */
1113 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1114                             bool write_fault, bool *writable, pfn_t *pfn)
1115 {
1116         struct page *page[1];
1117         int npages;
1118
1119         if (!(async || atomic))
1120                 return false;
1121
1122         /*
1123          * Fast pin a writable pfn only if it is a write fault request
1124          * or the caller allows to map a writable pfn for a read fault
1125          * request.
1126          */
1127         if (!(write_fault || writable))
1128                 return false;
1129
1130         npages = __get_user_pages_fast(addr, 1, 1, page);
1131         if (npages == 1) {
1132                 *pfn = page_to_pfn(page[0]);
1133
1134                 if (writable)
1135                         *writable = true;
1136                 return true;
1137         }
1138
1139         return false;
1140 }
1141
1142 /*
1143  * The slow path to get the pfn of the specified host virtual address,
1144  * 1 indicates success, -errno is returned if error is detected.
1145  */
1146 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1147                            bool *writable, pfn_t *pfn)
1148 {
1149         struct page *page[1];
1150         int npages = 0;
1151
1152         might_sleep();
1153
1154         if (writable)
1155                 *writable = write_fault;
1156
1157         if (async) {
1158                 down_read(&current->mm->mmap_sem);
1159                 npages = get_user_page_nowait(current, current->mm,
1160                                               addr, write_fault, page);
1161                 up_read(&current->mm->mmap_sem);
1162         } else
1163                 npages = get_user_pages_fast(addr, 1, write_fault,
1164                                              page);
1165         if (npages != 1)
1166                 return npages;
1167
1168         /* map read fault as writable if possible */
1169         if (unlikely(!write_fault) && writable) {
1170                 struct page *wpage[1];
1171
1172                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1173                 if (npages == 1) {
1174                         *writable = true;
1175                         put_page(page[0]);
1176                         page[0] = wpage[0];
1177                 }
1178
1179                 npages = 1;
1180         }
1181         *pfn = page_to_pfn(page[0]);
1182         return npages;
1183 }
1184
1185 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1186 {
1187         if (unlikely(!(vma->vm_flags & VM_READ)))
1188                 return false;
1189
1190         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1191                 return false;
1192
1193         return true;
1194 }
1195
1196 /*
1197  * Pin guest page in memory and return its pfn.
1198  * @addr: host virtual address which maps memory to the guest
1199  * @atomic: whether this function can sleep
1200  * @async: whether this function need to wait IO complete if the
1201  *         host page is not in the memory
1202  * @write_fault: whether we should get a writable host page
1203  * @writable: whether it allows to map a writable host page for !@write_fault
1204  *
1205  * The function will map a writable host page for these two cases:
1206  * 1): @write_fault = true
1207  * 2): @write_fault = false && @writable, @writable will tell the caller
1208  *     whether the mapping is writable.
1209  */
1210 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1211                         bool write_fault, bool *writable)
1212 {
1213         struct vm_area_struct *vma;
1214         pfn_t pfn = 0;
1215         int npages;
1216
1217         /* we can do it either atomically or asynchronously, not both */
1218         BUG_ON(atomic && async);
1219
1220         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1221                 return pfn;
1222
1223         if (atomic)
1224                 return KVM_PFN_ERR_FAULT;
1225
1226         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1227         if (npages == 1)
1228                 return pfn;
1229
1230         down_read(&current->mm->mmap_sem);
1231         if (npages == -EHWPOISON ||
1232               (!async && check_user_page_hwpoison(addr))) {
1233                 pfn = KVM_PFN_ERR_HWPOISON;
1234                 goto exit;
1235         }
1236
1237         vma = find_vma_intersection(current->mm, addr, addr + 1);
1238
1239         if (vma == NULL)
1240                 pfn = KVM_PFN_ERR_FAULT;
1241         else if ((vma->vm_flags & VM_PFNMAP)) {
1242                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1243                         vma->vm_pgoff;
1244                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1245         } else {
1246                 if (async && vma_is_valid(vma, write_fault))
1247                         *async = true;
1248                 pfn = KVM_PFN_ERR_FAULT;
1249         }
1250 exit:
1251         up_read(&current->mm->mmap_sem);
1252         return pfn;
1253 }
1254
1255 static pfn_t
1256 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1257                      bool *async, bool write_fault, bool *writable)
1258 {
1259         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1260
1261         if (addr == KVM_HVA_ERR_RO_BAD)
1262                 return KVM_PFN_ERR_RO_FAULT;
1263
1264         if (kvm_is_error_hva(addr))
1265                 return KVM_PFN_NOSLOT;
1266
1267         /* Do not map writable pfn in the readonly memslot. */
1268         if (writable && memslot_is_readonly(slot)) {
1269                 *writable = false;
1270                 writable = NULL;
1271         }
1272
1273         return hva_to_pfn(addr, atomic, async, write_fault,
1274                           writable);
1275 }
1276
1277 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1278                           bool write_fault, bool *writable)
1279 {
1280         struct kvm_memory_slot *slot;
1281
1282         if (async)
1283                 *async = false;
1284
1285         slot = gfn_to_memslot(kvm, gfn);
1286
1287         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1288                                     writable);
1289 }
1290
1291 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1292 {
1293         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1294 }
1295 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1296
1297 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1298                        bool write_fault, bool *writable)
1299 {
1300         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1301 }
1302 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1303
1304 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1305 {
1306         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1307 }
1308 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1309
1310 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1311                       bool *writable)
1312 {
1313         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1314 }
1315 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1316
1317 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1318 {
1319         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1320 }
1321
1322 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1323 {
1324         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1325 }
1326 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1327
1328 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1329                                                                   int nr_pages)
1330 {
1331         unsigned long addr;
1332         gfn_t entry;
1333
1334         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1335         if (kvm_is_error_hva(addr))
1336                 return -1;
1337
1338         if (entry < nr_pages)
1339                 return 0;
1340
1341         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1342 }
1343 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1344
1345 static struct page *kvm_pfn_to_page(pfn_t pfn)
1346 {
1347         if (is_error_noslot_pfn(pfn))
1348                 return KVM_ERR_PTR_BAD_PAGE;
1349
1350         if (kvm_is_mmio_pfn(pfn)) {
1351                 WARN_ON(1);
1352                 return KVM_ERR_PTR_BAD_PAGE;
1353         }
1354
1355         return pfn_to_page(pfn);
1356 }
1357
1358 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1359 {
1360         pfn_t pfn;
1361
1362         pfn = gfn_to_pfn(kvm, gfn);
1363
1364         return kvm_pfn_to_page(pfn);
1365 }
1366
1367 EXPORT_SYMBOL_GPL(gfn_to_page);
1368
1369 void kvm_release_page_clean(struct page *page)
1370 {
1371         WARN_ON(is_error_page(page));
1372
1373         kvm_release_pfn_clean(page_to_pfn(page));
1374 }
1375 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1376
1377 void kvm_release_pfn_clean(pfn_t pfn)
1378 {
1379         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1380                 put_page(pfn_to_page(pfn));
1381 }
1382 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1383
1384 void kvm_release_page_dirty(struct page *page)
1385 {
1386         WARN_ON(is_error_page(page));
1387
1388         kvm_release_pfn_dirty(page_to_pfn(page));
1389 }
1390 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1391
1392 static void kvm_release_pfn_dirty(pfn_t pfn)
1393 {
1394         kvm_set_pfn_dirty(pfn);
1395         kvm_release_pfn_clean(pfn);
1396 }
1397
1398 void kvm_set_pfn_dirty(pfn_t pfn)
1399 {
1400         if (!kvm_is_mmio_pfn(pfn)) {
1401                 struct page *page = pfn_to_page(pfn);
1402                 if (!PageReserved(page))
1403                         SetPageDirty(page);
1404         }
1405 }
1406 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1407
1408 void kvm_set_pfn_accessed(pfn_t pfn)
1409 {
1410         if (!kvm_is_mmio_pfn(pfn))
1411                 mark_page_accessed(pfn_to_page(pfn));
1412 }
1413 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1414
1415 void kvm_get_pfn(pfn_t pfn)
1416 {
1417         if (!kvm_is_mmio_pfn(pfn))
1418                 get_page(pfn_to_page(pfn));
1419 }
1420 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1421
1422 static int next_segment(unsigned long len, int offset)
1423 {
1424         if (len > PAGE_SIZE - offset)
1425                 return PAGE_SIZE - offset;
1426         else
1427                 return len;
1428 }
1429
1430 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1431                         int len)
1432 {
1433         int r;
1434         unsigned long addr;
1435
1436         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1437         if (kvm_is_error_hva(addr))
1438                 return -EFAULT;
1439         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1440         if (r)
1441                 return -EFAULT;
1442         return 0;
1443 }
1444 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1445
1446 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1447 {
1448         gfn_t gfn = gpa >> PAGE_SHIFT;
1449         int seg;
1450         int offset = offset_in_page(gpa);
1451         int ret;
1452
1453         while ((seg = next_segment(len, offset)) != 0) {
1454                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1455                 if (ret < 0)
1456                         return ret;
1457                 offset = 0;
1458                 len -= seg;
1459                 data += seg;
1460                 ++gfn;
1461         }
1462         return 0;
1463 }
1464 EXPORT_SYMBOL_GPL(kvm_read_guest);
1465
1466 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1467                           unsigned long len)
1468 {
1469         int r;
1470         unsigned long addr;
1471         gfn_t gfn = gpa >> PAGE_SHIFT;
1472         int offset = offset_in_page(gpa);
1473
1474         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1475         if (kvm_is_error_hva(addr))
1476                 return -EFAULT;
1477         pagefault_disable();
1478         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1479         pagefault_enable();
1480         if (r)
1481                 return -EFAULT;
1482         return 0;
1483 }
1484 EXPORT_SYMBOL(kvm_read_guest_atomic);
1485
1486 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1487                          int offset, int len)
1488 {
1489         int r;
1490         unsigned long addr;
1491
1492         addr = gfn_to_hva(kvm, gfn);
1493         if (kvm_is_error_hva(addr))
1494                 return -EFAULT;
1495         r = __copy_to_user((void __user *)addr + offset, data, len);
1496         if (r)
1497                 return -EFAULT;
1498         mark_page_dirty(kvm, gfn);
1499         return 0;
1500 }
1501 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1502
1503 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1504                     unsigned long len)
1505 {
1506         gfn_t gfn = gpa >> PAGE_SHIFT;
1507         int seg;
1508         int offset = offset_in_page(gpa);
1509         int ret;
1510
1511         while ((seg = next_segment(len, offset)) != 0) {
1512                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1513                 if (ret < 0)
1514                         return ret;
1515                 offset = 0;
1516                 len -= seg;
1517                 data += seg;
1518                 ++gfn;
1519         }
1520         return 0;
1521 }
1522
1523 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1524                               gpa_t gpa, unsigned long len)
1525 {
1526         struct kvm_memslots *slots = kvm_memslots(kvm);
1527         int offset = offset_in_page(gpa);
1528         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1529         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1530         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1531         gfn_t nr_pages_avail;
1532
1533         ghc->gpa = gpa;
1534         ghc->generation = slots->generation;
1535         ghc->len = len;
1536         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1537         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1538         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1539                 ghc->hva += offset;
1540         } else {
1541                 /*
1542                  * If the requested region crosses two memslots, we still
1543                  * verify that the entire region is valid here.
1544                  */
1545                 while (start_gfn <= end_gfn) {
1546                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1547                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1548                                                    &nr_pages_avail);
1549                         if (kvm_is_error_hva(ghc->hva))
1550                                 return -EFAULT;
1551                         start_gfn += nr_pages_avail;
1552                 }
1553                 /* Use the slow path for cross page reads and writes. */
1554                 ghc->memslot = NULL;
1555         }
1556         return 0;
1557 }
1558 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1559
1560 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1561                            void *data, unsigned long len)
1562 {
1563         struct kvm_memslots *slots = kvm_memslots(kvm);
1564         int r;
1565
1566         BUG_ON(len > ghc->len);
1567
1568         if (slots->generation != ghc->generation)
1569                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1570
1571         if (unlikely(!ghc->memslot))
1572                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1573
1574         if (kvm_is_error_hva(ghc->hva))
1575                 return -EFAULT;
1576
1577         r = __copy_to_user((void __user *)ghc->hva, data, len);
1578         if (r)
1579                 return -EFAULT;
1580         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1581
1582         return 0;
1583 }
1584 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1585
1586 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1587                            void *data, unsigned long len)
1588 {
1589         struct kvm_memslots *slots = kvm_memslots(kvm);
1590         int r;
1591
1592         BUG_ON(len > ghc->len);
1593
1594         if (slots->generation != ghc->generation)
1595                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1596
1597         if (unlikely(!ghc->memslot))
1598                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1599
1600         if (kvm_is_error_hva(ghc->hva))
1601                 return -EFAULT;
1602
1603         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1604         if (r)
1605                 return -EFAULT;
1606
1607         return 0;
1608 }
1609 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1610
1611 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1612 {
1613         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1614
1615         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1616 }
1617 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1618
1619 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1620 {
1621         gfn_t gfn = gpa >> PAGE_SHIFT;
1622         int seg;
1623         int offset = offset_in_page(gpa);
1624         int ret;
1625
1626         while ((seg = next_segment(len, offset)) != 0) {
1627                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1628                 if (ret < 0)
1629                         return ret;
1630                 offset = 0;
1631                 len -= seg;
1632                 ++gfn;
1633         }
1634         return 0;
1635 }
1636 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1637
1638 static void mark_page_dirty_in_slot(struct kvm *kvm,
1639                                     struct kvm_memory_slot *memslot,
1640                                     gfn_t gfn)
1641 {
1642         if (memslot && memslot->dirty_bitmap) {
1643                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1644
1645                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1646         }
1647 }
1648
1649 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1650 {
1651         struct kvm_memory_slot *memslot;
1652
1653         memslot = gfn_to_memslot(kvm, gfn);
1654         mark_page_dirty_in_slot(kvm, memslot, gfn);
1655 }
1656 EXPORT_SYMBOL_GPL(mark_page_dirty);
1657
1658 /*
1659  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1660  */
1661 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1662 {
1663         DEFINE_WAIT(wait);
1664
1665         for (;;) {
1666                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1667
1668                 if (kvm_arch_vcpu_runnable(vcpu)) {
1669                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1670                         break;
1671                 }
1672                 if (kvm_cpu_has_pending_timer(vcpu))
1673                         break;
1674                 if (signal_pending(current))
1675                         break;
1676
1677                 schedule();
1678         }
1679
1680         finish_wait(&vcpu->wq, &wait);
1681 }
1682 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1683
1684 #ifndef CONFIG_S390
1685 /*
1686  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1687  */
1688 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1689 {
1690         int me;
1691         int cpu = vcpu->cpu;
1692         wait_queue_head_t *wqp;
1693
1694         wqp = kvm_arch_vcpu_wq(vcpu);
1695         if (waitqueue_active(wqp)) {
1696                 wake_up_interruptible(wqp);
1697                 ++vcpu->stat.halt_wakeup;
1698         }
1699
1700         me = get_cpu();
1701         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1702                 if (kvm_arch_vcpu_should_kick(vcpu))
1703                         smp_send_reschedule(cpu);
1704         put_cpu();
1705 }
1706 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1707 #endif /* !CONFIG_S390 */
1708
1709 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1710 {
1711         struct pid *pid;
1712         struct task_struct *task = NULL;
1713         bool ret = false;
1714
1715         rcu_read_lock();
1716         pid = rcu_dereference(target->pid);
1717         if (pid)
1718                 task = get_pid_task(target->pid, PIDTYPE_PID);
1719         rcu_read_unlock();
1720         if (!task)
1721                 return ret;
1722         if (task->flags & PF_VCPU) {
1723                 put_task_struct(task);
1724                 return ret;
1725         }
1726         ret = yield_to(task, 1);
1727         put_task_struct(task);
1728
1729         return ret;
1730 }
1731 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1732
1733 /*
1734  * Helper that checks whether a VCPU is eligible for directed yield.
1735  * Most eligible candidate to yield is decided by following heuristics:
1736  *
1737  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1738  *  (preempted lock holder), indicated by @in_spin_loop.
1739  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1740  *
1741  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1742  *  chance last time (mostly it has become eligible now since we have probably
1743  *  yielded to lockholder in last iteration. This is done by toggling
1744  *  @dy_eligible each time a VCPU checked for eligibility.)
1745  *
1746  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1747  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1748  *  burning. Giving priority for a potential lock-holder increases lock
1749  *  progress.
1750  *
1751  *  Since algorithm is based on heuristics, accessing another VCPU data without
1752  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1753  *  and continue with next VCPU and so on.
1754  */
1755 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1756 {
1757 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1758         bool eligible;
1759
1760         eligible = !vcpu->spin_loop.in_spin_loop ||
1761                         (vcpu->spin_loop.in_spin_loop &&
1762                          vcpu->spin_loop.dy_eligible);
1763
1764         if (vcpu->spin_loop.in_spin_loop)
1765                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1766
1767         return eligible;
1768 #else
1769         return true;
1770 #endif
1771 }
1772
1773 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1774 {
1775         struct kvm *kvm = me->kvm;
1776         struct kvm_vcpu *vcpu;
1777         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1778         int yielded = 0;
1779         int try = 3;
1780         int pass;
1781         int i;
1782
1783         kvm_vcpu_set_in_spin_loop(me, true);
1784         /*
1785          * We boost the priority of a VCPU that is runnable but not
1786          * currently running, because it got preempted by something
1787          * else and called schedule in __vcpu_run.  Hopefully that
1788          * VCPU is holding the lock that we need and will release it.
1789          * We approximate round-robin by starting at the last boosted VCPU.
1790          */
1791         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1792                 kvm_for_each_vcpu(i, vcpu, kvm) {
1793                         if (!pass && i <= last_boosted_vcpu) {
1794                                 i = last_boosted_vcpu;
1795                                 continue;
1796                         } else if (pass && i > last_boosted_vcpu)
1797                                 break;
1798                         if (!ACCESS_ONCE(vcpu->preempted))
1799                                 continue;
1800                         if (vcpu == me)
1801                                 continue;
1802                         if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1803                                 continue;
1804                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1805                                 continue;
1806
1807                         yielded = kvm_vcpu_yield_to(vcpu);
1808                         if (yielded > 0) {
1809                                 kvm->last_boosted_vcpu = i;
1810                                 break;
1811                         } else if (yielded < 0) {
1812                                 try--;
1813                                 if (!try)
1814                                         break;
1815                         }
1816                 }
1817         }
1818         kvm_vcpu_set_in_spin_loop(me, false);
1819
1820         /* Ensure vcpu is not eligible during next spinloop */
1821         kvm_vcpu_set_dy_eligible(me, false);
1822 }
1823 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1824
1825 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1826 {
1827         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1828         struct page *page;
1829
1830         if (vmf->pgoff == 0)
1831                 page = virt_to_page(vcpu->run);
1832 #ifdef CONFIG_X86
1833         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1834                 page = virt_to_page(vcpu->arch.pio_data);
1835 #endif
1836 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1837         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1838                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1839 #endif
1840         else
1841                 return kvm_arch_vcpu_fault(vcpu, vmf);
1842         get_page(page);
1843         vmf->page = page;
1844         return 0;
1845 }
1846
1847 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1848         .fault = kvm_vcpu_fault,
1849 };
1850
1851 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1852 {
1853         vma->vm_ops = &kvm_vcpu_vm_ops;
1854         return 0;
1855 }
1856
1857 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1858 {
1859         struct kvm_vcpu *vcpu = filp->private_data;
1860
1861         kvm_put_kvm(vcpu->kvm);
1862         return 0;
1863 }
1864
1865 static struct file_operations kvm_vcpu_fops = {
1866         .release        = kvm_vcpu_release,
1867         .unlocked_ioctl = kvm_vcpu_ioctl,
1868 #ifdef CONFIG_COMPAT
1869         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1870 #endif
1871         .mmap           = kvm_vcpu_mmap,
1872         .llseek         = noop_llseek,
1873 };
1874
1875 /*
1876  * Allocates an inode for the vcpu.
1877  */
1878 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1879 {
1880         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1881 }
1882
1883 /*
1884  * Creates some virtual cpus.  Good luck creating more than one.
1885  */
1886 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1887 {
1888         int r;
1889         struct kvm_vcpu *vcpu, *v;
1890
1891         if (id >= KVM_MAX_VCPUS)
1892                 return -EINVAL;
1893
1894         vcpu = kvm_arch_vcpu_create(kvm, id);
1895         if (IS_ERR(vcpu))
1896                 return PTR_ERR(vcpu);
1897
1898         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1899
1900         r = kvm_arch_vcpu_setup(vcpu);
1901         if (r)
1902                 goto vcpu_destroy;
1903
1904         mutex_lock(&kvm->lock);
1905         if (!kvm_vcpu_compatible(vcpu)) {
1906                 r = -EINVAL;
1907                 goto unlock_vcpu_destroy;
1908         }
1909         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1910                 r = -EINVAL;
1911                 goto unlock_vcpu_destroy;
1912         }
1913
1914         kvm_for_each_vcpu(r, v, kvm)
1915                 if (v->vcpu_id == id) {
1916                         r = -EEXIST;
1917                         goto unlock_vcpu_destroy;
1918                 }
1919
1920         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1921
1922         /* Now it's all set up, let userspace reach it */
1923         kvm_get_kvm(kvm);
1924         r = create_vcpu_fd(vcpu);
1925         if (r < 0) {
1926                 kvm_put_kvm(kvm);
1927                 goto unlock_vcpu_destroy;
1928         }
1929
1930         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1931         smp_wmb();
1932         atomic_inc(&kvm->online_vcpus);
1933
1934         mutex_unlock(&kvm->lock);
1935         kvm_arch_vcpu_postcreate(vcpu);
1936         return r;
1937
1938 unlock_vcpu_destroy:
1939         mutex_unlock(&kvm->lock);
1940 vcpu_destroy:
1941         kvm_arch_vcpu_destroy(vcpu);
1942         return r;
1943 }
1944
1945 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1946 {
1947         if (sigset) {
1948                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1949                 vcpu->sigset_active = 1;
1950                 vcpu->sigset = *sigset;
1951         } else
1952                 vcpu->sigset_active = 0;
1953         return 0;
1954 }
1955
1956 static long kvm_vcpu_ioctl(struct file *filp,
1957                            unsigned int ioctl, unsigned long arg)
1958 {
1959         struct kvm_vcpu *vcpu = filp->private_data;
1960         void __user *argp = (void __user *)arg;
1961         int r;
1962         struct kvm_fpu *fpu = NULL;
1963         struct kvm_sregs *kvm_sregs = NULL;
1964
1965         if (vcpu->kvm->mm != current->mm)
1966                 return -EIO;
1967
1968 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1969         /*
1970          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1971          * so vcpu_load() would break it.
1972          */
1973         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1974                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1975 #endif
1976
1977
1978         r = vcpu_load(vcpu);
1979         if (r)
1980                 return r;
1981         switch (ioctl) {
1982         case KVM_RUN:
1983                 r = -EINVAL;
1984                 if (arg)
1985                         goto out;
1986                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1987                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1988                 break;
1989         case KVM_GET_REGS: {
1990                 struct kvm_regs *kvm_regs;
1991
1992                 r = -ENOMEM;
1993                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1994                 if (!kvm_regs)
1995                         goto out;
1996                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1997                 if (r)
1998                         goto out_free1;
1999                 r = -EFAULT;
2000                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2001                         goto out_free1;
2002                 r = 0;
2003 out_free1:
2004                 kfree(kvm_regs);
2005                 break;
2006         }
2007         case KVM_SET_REGS: {
2008                 struct kvm_regs *kvm_regs;
2009
2010                 r = -ENOMEM;
2011                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2012                 if (IS_ERR(kvm_regs)) {
2013                         r = PTR_ERR(kvm_regs);
2014                         goto out;
2015                 }
2016                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2017                 kfree(kvm_regs);
2018                 break;
2019         }
2020         case KVM_GET_SREGS: {
2021                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2022                 r = -ENOMEM;
2023                 if (!kvm_sregs)
2024                         goto out;
2025                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2026                 if (r)
2027                         goto out;
2028                 r = -EFAULT;
2029                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2030                         goto out;
2031                 r = 0;
2032                 break;
2033         }
2034         case KVM_SET_SREGS: {
2035                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2036                 if (IS_ERR(kvm_sregs)) {
2037                         r = PTR_ERR(kvm_sregs);
2038                         kvm_sregs = NULL;
2039                         goto out;
2040                 }
2041                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2042                 break;
2043         }
2044         case KVM_GET_MP_STATE: {
2045                 struct kvm_mp_state mp_state;
2046
2047                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2048                 if (r)
2049                         goto out;
2050                 r = -EFAULT;
2051                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2052                         goto out;
2053                 r = 0;
2054                 break;
2055         }
2056         case KVM_SET_MP_STATE: {
2057                 struct kvm_mp_state mp_state;
2058
2059                 r = -EFAULT;
2060                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2061                         goto out;
2062                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2063                 break;
2064         }
2065         case KVM_TRANSLATE: {
2066                 struct kvm_translation tr;
2067
2068                 r = -EFAULT;
2069                 if (copy_from_user(&tr, argp, sizeof tr))
2070                         goto out;
2071                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2072                 if (r)
2073                         goto out;
2074                 r = -EFAULT;
2075                 if (copy_to_user(argp, &tr, sizeof tr))
2076                         goto out;
2077                 r = 0;
2078                 break;
2079         }
2080         case KVM_SET_GUEST_DEBUG: {
2081                 struct kvm_guest_debug dbg;
2082
2083                 r = -EFAULT;
2084                 if (copy_from_user(&dbg, argp, sizeof dbg))
2085                         goto out;
2086                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2087                 break;
2088         }
2089         case KVM_SET_SIGNAL_MASK: {
2090                 struct kvm_signal_mask __user *sigmask_arg = argp;
2091                 struct kvm_signal_mask kvm_sigmask;
2092                 sigset_t sigset, *p;
2093
2094                 p = NULL;
2095                 if (argp) {
2096                         r = -EFAULT;
2097                         if (copy_from_user(&kvm_sigmask, argp,
2098                                            sizeof kvm_sigmask))
2099                                 goto out;
2100                         r = -EINVAL;
2101                         if (kvm_sigmask.len != sizeof sigset)
2102                                 goto out;
2103                         r = -EFAULT;
2104                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2105                                            sizeof sigset))
2106                                 goto out;
2107                         p = &sigset;
2108                 }
2109                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2110                 break;
2111         }
2112         case KVM_GET_FPU: {
2113                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2114                 r = -ENOMEM;
2115                 if (!fpu)
2116                         goto out;
2117                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2118                 if (r)
2119                         goto out;
2120                 r = -EFAULT;
2121                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2122                         goto out;
2123                 r = 0;
2124                 break;
2125         }
2126         case KVM_SET_FPU: {
2127                 fpu = memdup_user(argp, sizeof(*fpu));
2128                 if (IS_ERR(fpu)) {
2129                         r = PTR_ERR(fpu);
2130                         fpu = NULL;
2131                         goto out;
2132                 }
2133                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2134                 break;
2135         }
2136         default:
2137                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2138         }
2139 out:
2140         vcpu_put(vcpu);
2141         kfree(fpu);
2142         kfree(kvm_sregs);
2143         return r;
2144 }
2145
2146 #ifdef CONFIG_COMPAT
2147 static long kvm_vcpu_compat_ioctl(struct file *filp,
2148                                   unsigned int ioctl, unsigned long arg)
2149 {
2150         struct kvm_vcpu *vcpu = filp->private_data;
2151         void __user *argp = compat_ptr(arg);
2152         int r;
2153
2154         if (vcpu->kvm->mm != current->mm)
2155                 return -EIO;
2156
2157         switch (ioctl) {
2158         case KVM_SET_SIGNAL_MASK: {
2159                 struct kvm_signal_mask __user *sigmask_arg = argp;
2160                 struct kvm_signal_mask kvm_sigmask;
2161                 compat_sigset_t csigset;
2162                 sigset_t sigset;
2163
2164                 if (argp) {
2165                         r = -EFAULT;
2166                         if (copy_from_user(&kvm_sigmask, argp,
2167                                            sizeof kvm_sigmask))
2168                                 goto out;
2169                         r = -EINVAL;
2170                         if (kvm_sigmask.len != sizeof csigset)
2171                                 goto out;
2172                         r = -EFAULT;
2173                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2174                                            sizeof csigset))
2175                                 goto out;
2176                         sigset_from_compat(&sigset, &csigset);
2177                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2178                 } else
2179                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2180                 break;
2181         }
2182         default:
2183                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2184         }
2185
2186 out:
2187         return r;
2188 }
2189 #endif
2190
2191 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2192                                  int (*accessor)(struct kvm_device *dev,
2193                                                  struct kvm_device_attr *attr),
2194                                  unsigned long arg)
2195 {
2196         struct kvm_device_attr attr;
2197
2198         if (!accessor)
2199                 return -EPERM;
2200
2201         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2202                 return -EFAULT;
2203
2204         return accessor(dev, &attr);
2205 }
2206
2207 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2208                              unsigned long arg)
2209 {
2210         struct kvm_device *dev = filp->private_data;
2211
2212         switch (ioctl) {
2213         case KVM_SET_DEVICE_ATTR:
2214                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2215         case KVM_GET_DEVICE_ATTR:
2216                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2217         case KVM_HAS_DEVICE_ATTR:
2218                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2219         default:
2220                 if (dev->ops->ioctl)
2221                         return dev->ops->ioctl(dev, ioctl, arg);
2222
2223                 return -ENOTTY;
2224         }
2225 }
2226
2227 static int kvm_device_release(struct inode *inode, struct file *filp)
2228 {
2229         struct kvm_device *dev = filp->private_data;
2230         struct kvm *kvm = dev->kvm;
2231
2232         kvm_put_kvm(kvm);
2233         return 0;
2234 }
2235
2236 static const struct file_operations kvm_device_fops = {
2237         .unlocked_ioctl = kvm_device_ioctl,
2238 #ifdef CONFIG_COMPAT
2239         .compat_ioctl = kvm_device_ioctl,
2240 #endif
2241         .release = kvm_device_release,
2242 };
2243
2244 struct kvm_device *kvm_device_from_filp(struct file *filp)
2245 {
2246         if (filp->f_op != &kvm_device_fops)
2247                 return NULL;
2248
2249         return filp->private_data;
2250 }
2251
2252 static int kvm_ioctl_create_device(struct kvm *kvm,
2253                                    struct kvm_create_device *cd)
2254 {
2255         struct kvm_device_ops *ops = NULL;
2256         struct kvm_device *dev;
2257         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2258         int ret;
2259
2260         switch (cd->type) {
2261 #ifdef CONFIG_KVM_MPIC
2262         case KVM_DEV_TYPE_FSL_MPIC_20:
2263         case KVM_DEV_TYPE_FSL_MPIC_42:
2264                 ops = &kvm_mpic_ops;
2265                 break;
2266 #endif
2267 #ifdef CONFIG_KVM_XICS
2268         case KVM_DEV_TYPE_XICS:
2269                 ops = &kvm_xics_ops;
2270                 break;
2271 #endif
2272 #ifdef CONFIG_KVM_VFIO
2273         case KVM_DEV_TYPE_VFIO:
2274                 ops = &kvm_vfio_ops;
2275                 break;
2276 #endif
2277 #ifdef CONFIG_KVM_ARM_VGIC
2278         case KVM_DEV_TYPE_ARM_VGIC_V2:
2279                 ops = &kvm_arm_vgic_v2_ops;
2280                 break;
2281 #endif
2282 #ifdef CONFIG_S390
2283         case KVM_DEV_TYPE_FLIC:
2284                 ops = &kvm_flic_ops;
2285                 break;
2286 #endif
2287         default:
2288                 return -ENODEV;
2289         }
2290
2291         if (test)
2292                 return 0;
2293
2294         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2295         if (!dev)
2296                 return -ENOMEM;
2297
2298         dev->ops = ops;
2299         dev->kvm = kvm;
2300
2301         ret = ops->create(dev, cd->type);
2302         if (ret < 0) {
2303                 kfree(dev);
2304                 return ret;
2305         }
2306
2307         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2308         if (ret < 0) {
2309                 ops->destroy(dev);
2310                 return ret;
2311         }
2312
2313         list_add(&dev->vm_node, &kvm->devices);
2314         kvm_get_kvm(kvm);
2315         cd->fd = ret;
2316         return 0;
2317 }
2318
2319 static long kvm_vm_ioctl(struct file *filp,
2320                            unsigned int ioctl, unsigned long arg)
2321 {
2322         struct kvm *kvm = filp->private_data;
2323         void __user *argp = (void __user *)arg;
2324         int r;
2325
2326         if (kvm->mm != current->mm)
2327                 return -EIO;
2328         switch (ioctl) {
2329         case KVM_CREATE_VCPU:
2330                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2331                 break;
2332         case KVM_SET_USER_MEMORY_REGION: {
2333                 struct kvm_userspace_memory_region kvm_userspace_mem;
2334
2335                 r = -EFAULT;
2336                 if (copy_from_user(&kvm_userspace_mem, argp,
2337                                                 sizeof kvm_userspace_mem))
2338                         goto out;
2339
2340                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2341                 break;
2342         }
2343         case KVM_GET_DIRTY_LOG: {
2344                 struct kvm_dirty_log log;
2345
2346                 r = -EFAULT;
2347                 if (copy_from_user(&log, argp, sizeof log))
2348                         goto out;
2349                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2350                 break;
2351         }
2352 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2353         case KVM_REGISTER_COALESCED_MMIO: {
2354                 struct kvm_coalesced_mmio_zone zone;
2355                 r = -EFAULT;
2356                 if (copy_from_user(&zone, argp, sizeof zone))
2357                         goto out;
2358                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2359                 break;
2360         }
2361         case KVM_UNREGISTER_COALESCED_MMIO: {
2362                 struct kvm_coalesced_mmio_zone zone;
2363                 r = -EFAULT;
2364                 if (copy_from_user(&zone, argp, sizeof zone))
2365                         goto out;
2366                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2367                 break;
2368         }
2369 #endif
2370         case KVM_IRQFD: {
2371                 struct kvm_irqfd data;
2372
2373                 r = -EFAULT;
2374                 if (copy_from_user(&data, argp, sizeof data))
2375                         goto out;
2376                 r = kvm_irqfd(kvm, &data);
2377                 break;
2378         }
2379         case KVM_IOEVENTFD: {
2380                 struct kvm_ioeventfd data;
2381
2382                 r = -EFAULT;
2383                 if (copy_from_user(&data, argp, sizeof data))
2384                         goto out;
2385                 r = kvm_ioeventfd(kvm, &data);
2386                 break;
2387         }
2388 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2389         case KVM_SET_BOOT_CPU_ID:
2390                 r = 0;
2391                 mutex_lock(&kvm->lock);
2392                 if (atomic_read(&kvm->online_vcpus) != 0)
2393                         r = -EBUSY;
2394                 else
2395                         kvm->bsp_vcpu_id = arg;
2396                 mutex_unlock(&kvm->lock);
2397                 break;
2398 #endif
2399 #ifdef CONFIG_HAVE_KVM_MSI
2400         case KVM_SIGNAL_MSI: {
2401                 struct kvm_msi msi;
2402
2403                 r = -EFAULT;
2404                 if (copy_from_user(&msi, argp, sizeof msi))
2405                         goto out;
2406                 r = kvm_send_userspace_msi(kvm, &msi);
2407                 break;
2408         }
2409 #endif
2410 #ifdef __KVM_HAVE_IRQ_LINE
2411         case KVM_IRQ_LINE_STATUS:
2412         case KVM_IRQ_LINE: {
2413                 struct kvm_irq_level irq_event;
2414
2415                 r = -EFAULT;
2416                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2417                         goto out;
2418
2419                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2420                                         ioctl == KVM_IRQ_LINE_STATUS);
2421                 if (r)
2422                         goto out;
2423
2424                 r = -EFAULT;
2425                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2426                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2427                                 goto out;
2428                 }
2429
2430                 r = 0;
2431                 break;
2432         }
2433 #endif
2434 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2435         case KVM_SET_GSI_ROUTING: {
2436                 struct kvm_irq_routing routing;
2437                 struct kvm_irq_routing __user *urouting;
2438                 struct kvm_irq_routing_entry *entries;
2439
2440                 r = -EFAULT;
2441                 if (copy_from_user(&routing, argp, sizeof(routing)))
2442                         goto out;
2443                 r = -EINVAL;
2444                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2445                         goto out;
2446                 if (routing.flags)
2447                         goto out;
2448                 r = -ENOMEM;
2449                 entries = vmalloc(routing.nr * sizeof(*entries));
2450                 if (!entries)
2451                         goto out;
2452                 r = -EFAULT;
2453                 urouting = argp;
2454                 if (copy_from_user(entries, urouting->entries,
2455                                    routing.nr * sizeof(*entries)))
2456                         goto out_free_irq_routing;
2457                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2458                                         routing.flags);
2459         out_free_irq_routing:
2460                 vfree(entries);
2461                 break;
2462         }
2463 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2464         case KVM_CREATE_DEVICE: {
2465                 struct kvm_create_device cd;
2466
2467                 r = -EFAULT;
2468                 if (copy_from_user(&cd, argp, sizeof(cd)))
2469                         goto out;
2470
2471                 r = kvm_ioctl_create_device(kvm, &cd);
2472                 if (r)
2473                         goto out;
2474
2475                 r = -EFAULT;
2476                 if (copy_to_user(argp, &cd, sizeof(cd)))
2477                         goto out;
2478
2479                 r = 0;
2480                 break;
2481         }
2482         default:
2483                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2484                 if (r == -ENOTTY)
2485                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2486         }
2487 out:
2488         return r;
2489 }
2490
2491 #ifdef CONFIG_COMPAT
2492 struct compat_kvm_dirty_log {
2493         __u32 slot;
2494         __u32 padding1;
2495         union {
2496                 compat_uptr_t dirty_bitmap; /* one bit per page */
2497                 __u64 padding2;
2498         };
2499 };
2500
2501 static long kvm_vm_compat_ioctl(struct file *filp,
2502                            unsigned int ioctl, unsigned long arg)
2503 {
2504         struct kvm *kvm = filp->private_data;
2505         int r;
2506
2507         if (kvm->mm != current->mm)
2508                 return -EIO;
2509         switch (ioctl) {
2510         case KVM_GET_DIRTY_LOG: {
2511                 struct compat_kvm_dirty_log compat_log;
2512                 struct kvm_dirty_log log;
2513
2514                 r = -EFAULT;
2515                 if (copy_from_user(&compat_log, (void __user *)arg,
2516                                    sizeof(compat_log)))
2517                         goto out;
2518                 log.slot         = compat_log.slot;
2519                 log.padding1     = compat_log.padding1;
2520                 log.padding2     = compat_log.padding2;
2521                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2522
2523                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2524                 break;
2525         }
2526         default:
2527                 r = kvm_vm_ioctl(filp, ioctl, arg);
2528         }
2529
2530 out:
2531         return r;
2532 }
2533 #endif
2534
2535 static struct file_operations kvm_vm_fops = {
2536         .release        = kvm_vm_release,
2537         .unlocked_ioctl = kvm_vm_ioctl,
2538 #ifdef CONFIG_COMPAT
2539         .compat_ioctl   = kvm_vm_compat_ioctl,
2540 #endif
2541         .llseek         = noop_llseek,
2542 };
2543
2544 static int kvm_dev_ioctl_create_vm(unsigned long type)
2545 {
2546         int r;
2547         struct kvm *kvm;
2548
2549         kvm = kvm_create_vm(type);
2550         if (IS_ERR(kvm))
2551                 return PTR_ERR(kvm);
2552 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2553         r = kvm_coalesced_mmio_init(kvm);
2554         if (r < 0) {
2555                 kvm_put_kvm(kvm);
2556                 return r;
2557         }
2558 #endif
2559         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2560         if (r < 0)
2561                 kvm_put_kvm(kvm);
2562
2563         return r;
2564 }
2565
2566 static long kvm_dev_ioctl_check_extension_generic(long arg)
2567 {
2568         switch (arg) {
2569         case KVM_CAP_USER_MEMORY:
2570         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2571         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2572 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2573         case KVM_CAP_SET_BOOT_CPU_ID:
2574 #endif
2575         case KVM_CAP_INTERNAL_ERROR_DATA:
2576 #ifdef CONFIG_HAVE_KVM_MSI
2577         case KVM_CAP_SIGNAL_MSI:
2578 #endif
2579 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2580         case KVM_CAP_IRQFD_RESAMPLE:
2581 #endif
2582                 return 1;
2583 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2584         case KVM_CAP_IRQ_ROUTING:
2585                 return KVM_MAX_IRQ_ROUTES;
2586 #endif
2587         default:
2588                 break;
2589         }
2590         return kvm_dev_ioctl_check_extension(arg);
2591 }
2592
2593 static long kvm_dev_ioctl(struct file *filp,
2594                           unsigned int ioctl, unsigned long arg)
2595 {
2596         long r = -EINVAL;
2597
2598         switch (ioctl) {
2599         case KVM_GET_API_VERSION:
2600                 r = -EINVAL;
2601                 if (arg)
2602                         goto out;
2603                 r = KVM_API_VERSION;
2604                 break;
2605         case KVM_CREATE_VM:
2606                 r = kvm_dev_ioctl_create_vm(arg);
2607                 break;
2608         case KVM_CHECK_EXTENSION:
2609                 r = kvm_dev_ioctl_check_extension_generic(arg);
2610                 break;
2611         case KVM_GET_VCPU_MMAP_SIZE:
2612                 r = -EINVAL;
2613                 if (arg)
2614                         goto out;
2615                 r = PAGE_SIZE;     /* struct kvm_run */
2616 #ifdef CONFIG_X86
2617                 r += PAGE_SIZE;    /* pio data page */
2618 #endif
2619 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2620                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2621 #endif
2622                 break;
2623         case KVM_TRACE_ENABLE:
2624         case KVM_TRACE_PAUSE:
2625         case KVM_TRACE_DISABLE:
2626                 r = -EOPNOTSUPP;
2627                 break;
2628         default:
2629                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2630         }
2631 out:
2632         return r;
2633 }
2634
2635 static struct file_operations kvm_chardev_ops = {
2636         .unlocked_ioctl = kvm_dev_ioctl,
2637         .compat_ioctl   = kvm_dev_ioctl,
2638         .llseek         = noop_llseek,
2639 };
2640
2641 static struct miscdevice kvm_dev = {
2642         KVM_MINOR,
2643         "kvm",
2644         &kvm_chardev_ops,
2645 };
2646
2647 static void hardware_enable_nolock(void *junk)
2648 {
2649         int cpu = raw_smp_processor_id();
2650         int r;
2651
2652         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2653                 return;
2654
2655         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2656
2657         r = kvm_arch_hardware_enable(NULL);
2658
2659         if (r) {
2660                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2661                 atomic_inc(&hardware_enable_failed);
2662                 printk(KERN_INFO "kvm: enabling virtualization on "
2663                                  "CPU%d failed\n", cpu);
2664         }
2665 }
2666
2667 static void hardware_enable(void)
2668 {
2669         raw_spin_lock(&kvm_count_lock);
2670         if (kvm_usage_count)
2671                 hardware_enable_nolock(NULL);
2672         raw_spin_unlock(&kvm_count_lock);
2673 }
2674
2675 static void hardware_disable_nolock(void *junk)
2676 {
2677         int cpu = raw_smp_processor_id();
2678
2679         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2680                 return;
2681         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2682         kvm_arch_hardware_disable(NULL);
2683 }
2684
2685 static void hardware_disable(void)
2686 {
2687         raw_spin_lock(&kvm_count_lock);
2688         if (kvm_usage_count)
2689                 hardware_disable_nolock(NULL);
2690         raw_spin_unlock(&kvm_count_lock);
2691 }
2692
2693 static void hardware_disable_all_nolock(void)
2694 {
2695         BUG_ON(!kvm_usage_count);
2696
2697         kvm_usage_count--;
2698         if (!kvm_usage_count)
2699                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2700 }
2701
2702 static void hardware_disable_all(void)
2703 {
2704         raw_spin_lock(&kvm_count_lock);
2705         hardware_disable_all_nolock();
2706         raw_spin_unlock(&kvm_count_lock);
2707 }
2708
2709 static int hardware_enable_all(void)
2710 {
2711         int r = 0;
2712
2713         raw_spin_lock(&kvm_count_lock);
2714
2715         kvm_usage_count++;
2716         if (kvm_usage_count == 1) {
2717                 atomic_set(&hardware_enable_failed, 0);
2718                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2719
2720                 if (atomic_read(&hardware_enable_failed)) {
2721                         hardware_disable_all_nolock();
2722                         r = -EBUSY;
2723                 }
2724         }
2725
2726         raw_spin_unlock(&kvm_count_lock);
2727
2728         return r;
2729 }
2730
2731 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2732                            void *v)
2733 {
2734         int cpu = (long)v;
2735
2736         val &= ~CPU_TASKS_FROZEN;
2737         switch (val) {
2738         case CPU_DYING:
2739                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2740                        cpu);
2741                 hardware_disable();
2742                 break;
2743         case CPU_STARTING:
2744                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2745                        cpu);
2746                 hardware_enable();
2747                 break;
2748         }
2749         return NOTIFY_OK;
2750 }
2751
2752 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2753                       void *v)
2754 {
2755         /*
2756          * Some (well, at least mine) BIOSes hang on reboot if
2757          * in vmx root mode.
2758          *
2759          * And Intel TXT required VMX off for all cpu when system shutdown.
2760          */
2761         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2762         kvm_rebooting = true;
2763         on_each_cpu(hardware_disable_nolock, NULL, 1);
2764         return NOTIFY_OK;
2765 }
2766
2767 static struct notifier_block kvm_reboot_notifier = {
2768         .notifier_call = kvm_reboot,
2769         .priority = 0,
2770 };
2771
2772 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2773 {
2774         int i;
2775
2776         for (i = 0; i < bus->dev_count; i++) {
2777                 struct kvm_io_device *pos = bus->range[i].dev;
2778
2779                 kvm_iodevice_destructor(pos);
2780         }
2781         kfree(bus);
2782 }
2783
2784 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2785                                  const struct kvm_io_range *r2)
2786 {
2787         if (r1->addr < r2->addr)
2788                 return -1;
2789         if (r1->addr + r1->len > r2->addr + r2->len)
2790                 return 1;
2791         return 0;
2792 }
2793
2794 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2795 {
2796         return kvm_io_bus_cmp(p1, p2);
2797 }
2798
2799 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2800                           gpa_t addr, int len)
2801 {
2802         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2803                 .addr = addr,
2804                 .len = len,
2805                 .dev = dev,
2806         };
2807
2808         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2809                 kvm_io_bus_sort_cmp, NULL);
2810
2811         return 0;
2812 }
2813
2814 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2815                              gpa_t addr, int len)
2816 {
2817         struct kvm_io_range *range, key;
2818         int off;
2819
2820         key = (struct kvm_io_range) {
2821                 .addr = addr,
2822                 .len = len,
2823         };
2824
2825         range = bsearch(&key, bus->range, bus->dev_count,
2826                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2827         if (range == NULL)
2828                 return -ENOENT;
2829
2830         off = range - bus->range;
2831
2832         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2833                 off--;
2834
2835         return off;
2836 }
2837
2838 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2839                               struct kvm_io_range *range, const void *val)
2840 {
2841         int idx;
2842
2843         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2844         if (idx < 0)
2845                 return -EOPNOTSUPP;
2846
2847         while (idx < bus->dev_count &&
2848                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2849                 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2850                                         range->len, val))
2851                         return idx;
2852                 idx++;
2853         }
2854
2855         return -EOPNOTSUPP;
2856 }
2857
2858 /* kvm_io_bus_write - called under kvm->slots_lock */
2859 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2860                      int len, const void *val)
2861 {
2862         struct kvm_io_bus *bus;
2863         struct kvm_io_range range;
2864         int r;
2865
2866         range = (struct kvm_io_range) {
2867                 .addr = addr,
2868                 .len = len,
2869         };
2870
2871         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2872         r = __kvm_io_bus_write(bus, &range, val);
2873         return r < 0 ? r : 0;
2874 }
2875
2876 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2877 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2878                             int len, const void *val, long cookie)
2879 {
2880         struct kvm_io_bus *bus;
2881         struct kvm_io_range range;
2882
2883         range = (struct kvm_io_range) {
2884                 .addr = addr,
2885                 .len = len,
2886         };
2887
2888         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2889
2890         /* First try the device referenced by cookie. */
2891         if ((cookie >= 0) && (cookie < bus->dev_count) &&
2892             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2893                 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2894                                         val))
2895                         return cookie;
2896
2897         /*
2898          * cookie contained garbage; fall back to search and return the
2899          * correct cookie value.
2900          */
2901         return __kvm_io_bus_write(bus, &range, val);
2902 }
2903
2904 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2905                              void *val)
2906 {
2907         int idx;
2908
2909         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2910         if (idx < 0)
2911                 return -EOPNOTSUPP;
2912
2913         while (idx < bus->dev_count &&
2914                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2915                 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2916                                        range->len, val))
2917                         return idx;
2918                 idx++;
2919         }
2920
2921         return -EOPNOTSUPP;
2922 }
2923
2924 /* kvm_io_bus_read - called under kvm->slots_lock */
2925 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2926                     int len, void *val)
2927 {
2928         struct kvm_io_bus *bus;
2929         struct kvm_io_range range;
2930         int r;
2931
2932         range = (struct kvm_io_range) {
2933                 .addr = addr,
2934                 .len = len,
2935         };
2936
2937         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2938         r = __kvm_io_bus_read(bus, &range, val);
2939         return r < 0 ? r : 0;
2940 }
2941
2942
2943 /* Caller must hold slots_lock. */
2944 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2945                             int len, struct kvm_io_device *dev)
2946 {
2947         struct kvm_io_bus *new_bus, *bus;
2948
2949         bus = kvm->buses[bus_idx];
2950         /* exclude ioeventfd which is limited by maximum fd */
2951         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2952                 return -ENOSPC;
2953
2954         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2955                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2956         if (!new_bus)
2957                 return -ENOMEM;
2958         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2959                sizeof(struct kvm_io_range)));
2960         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2961         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2962         synchronize_srcu_expedited(&kvm->srcu);
2963         kfree(bus);
2964
2965         return 0;
2966 }
2967
2968 /* Caller must hold slots_lock. */
2969 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2970                               struct kvm_io_device *dev)
2971 {
2972         int i, r;
2973         struct kvm_io_bus *new_bus, *bus;
2974
2975         bus = kvm->buses[bus_idx];
2976         r = -ENOENT;
2977         for (i = 0; i < bus->dev_count; i++)
2978                 if (bus->range[i].dev == dev) {
2979                         r = 0;
2980                         break;
2981                 }
2982
2983         if (r)
2984                 return r;
2985
2986         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2987                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2988         if (!new_bus)
2989                 return -ENOMEM;
2990
2991         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2992         new_bus->dev_count--;
2993         memcpy(new_bus->range + i, bus->range + i + 1,
2994                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2995
2996         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2997         synchronize_srcu_expedited(&kvm->srcu);
2998         kfree(bus);
2999         return r;
3000 }
3001
3002 static struct notifier_block kvm_cpu_notifier = {
3003         .notifier_call = kvm_cpu_hotplug,
3004 };
3005
3006 static int vm_stat_get(void *_offset, u64 *val)
3007 {
3008         unsigned offset = (long)_offset;
3009         struct kvm *kvm;
3010
3011         *val = 0;
3012         spin_lock(&kvm_lock);
3013         list_for_each_entry(kvm, &vm_list, vm_list)
3014                 *val += *(u32 *)((void *)kvm + offset);
3015         spin_unlock(&kvm_lock);
3016         return 0;
3017 }
3018
3019 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3020
3021 static int vcpu_stat_get(void *_offset, u64 *val)
3022 {
3023         unsigned offset = (long)_offset;
3024         struct kvm *kvm;
3025         struct kvm_vcpu *vcpu;
3026         int i;
3027
3028         *val = 0;
3029         spin_lock(&kvm_lock);
3030         list_for_each_entry(kvm, &vm_list, vm_list)
3031                 kvm_for_each_vcpu(i, vcpu, kvm)
3032                         *val += *(u32 *)((void *)vcpu + offset);
3033
3034         spin_unlock(&kvm_lock);
3035         return 0;
3036 }
3037
3038 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3039
3040 static const struct file_operations *stat_fops[] = {
3041         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3042         [KVM_STAT_VM]   = &vm_stat_fops,
3043 };
3044
3045 static int kvm_init_debug(void)
3046 {
3047         int r = -EEXIST;
3048         struct kvm_stats_debugfs_item *p;
3049
3050         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3051         if (kvm_debugfs_dir == NULL)
3052                 goto out;
3053
3054         for (p = debugfs_entries; p->name; ++p) {
3055                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3056                                                 (void *)(long)p->offset,
3057                                                 stat_fops[p->kind]);
3058                 if (p->dentry == NULL)
3059                         goto out_dir;
3060         }
3061
3062         return 0;
3063
3064 out_dir:
3065         debugfs_remove_recursive(kvm_debugfs_dir);
3066 out:
3067         return r;
3068 }
3069
3070 static void kvm_exit_debug(void)
3071 {
3072         struct kvm_stats_debugfs_item *p;
3073
3074         for (p = debugfs_entries; p->name; ++p)
3075                 debugfs_remove(p->dentry);
3076         debugfs_remove(kvm_debugfs_dir);
3077 }
3078
3079 static int kvm_suspend(void)
3080 {
3081         if (kvm_usage_count)
3082                 hardware_disable_nolock(NULL);
3083         return 0;
3084 }
3085
3086 static void kvm_resume(void)
3087 {
3088         if (kvm_usage_count) {
3089                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3090                 hardware_enable_nolock(NULL);
3091         }
3092 }
3093
3094 static struct syscore_ops kvm_syscore_ops = {
3095         .suspend = kvm_suspend,
3096         .resume = kvm_resume,
3097 };
3098
3099 static inline
3100 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3101 {
3102         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3103 }
3104
3105 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3106 {
3107         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3108         if (vcpu->preempted)
3109                 vcpu->preempted = false;
3110
3111         kvm_arch_vcpu_load(vcpu, cpu);
3112 }
3113
3114 static void kvm_sched_out(struct preempt_notifier *pn,
3115                           struct task_struct *next)
3116 {
3117         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3118
3119         if (current->state == TASK_RUNNING)
3120                 vcpu->preempted = true;
3121         kvm_arch_vcpu_put(vcpu);
3122 }
3123
3124 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3125                   struct module *module)
3126 {
3127         int r;
3128         int cpu;
3129
3130         r = kvm_arch_init(opaque);
3131         if (r)
3132                 goto out_fail;
3133
3134         /*
3135          * kvm_arch_init makes sure there's at most one caller
3136          * for architectures that support multiple implementations,
3137          * like intel and amd on x86.
3138          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3139          * conflicts in case kvm is already setup for another implementation.
3140          */
3141         r = kvm_irqfd_init();
3142         if (r)
3143                 goto out_irqfd;
3144
3145         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3146                 r = -ENOMEM;
3147                 goto out_free_0;
3148         }
3149
3150         r = kvm_arch_hardware_setup();
3151         if (r < 0)
3152                 goto out_free_0a;
3153
3154         for_each_online_cpu(cpu) {
3155                 smp_call_function_single(cpu,
3156                                 kvm_arch_check_processor_compat,
3157                                 &r, 1);
3158                 if (r < 0)
3159                         goto out_free_1;
3160         }
3161
3162         r = register_cpu_notifier(&kvm_cpu_notifier);
3163         if (r)
3164                 goto out_free_2;
3165         register_reboot_notifier(&kvm_reboot_notifier);
3166
3167         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3168         if (!vcpu_align)
3169                 vcpu_align = __alignof__(struct kvm_vcpu);
3170         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3171                                            0, NULL);
3172         if (!kvm_vcpu_cache) {
3173                 r = -ENOMEM;
3174                 goto out_free_3;
3175         }
3176
3177         r = kvm_async_pf_init();
3178         if (r)
3179                 goto out_free;
3180
3181         kvm_chardev_ops.owner = module;
3182         kvm_vm_fops.owner = module;
3183         kvm_vcpu_fops.owner = module;
3184
3185         r = misc_register(&kvm_dev);
3186         if (r) {
3187                 printk(KERN_ERR "kvm: misc device register failed\n");
3188                 goto out_unreg;
3189         }
3190
3191         register_syscore_ops(&kvm_syscore_ops);
3192
3193         kvm_preempt_ops.sched_in = kvm_sched_in;
3194         kvm_preempt_ops.sched_out = kvm_sched_out;
3195
3196         r = kvm_init_debug();
3197         if (r) {
3198                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3199                 goto out_undebugfs;
3200         }
3201
3202         return 0;
3203
3204 out_undebugfs:
3205         unregister_syscore_ops(&kvm_syscore_ops);
3206         misc_deregister(&kvm_dev);
3207 out_unreg:
3208         kvm_async_pf_deinit();
3209 out_free:
3210         kmem_cache_destroy(kvm_vcpu_cache);
3211 out_free_3:
3212         unregister_reboot_notifier(&kvm_reboot_notifier);
3213         unregister_cpu_notifier(&kvm_cpu_notifier);
3214 out_free_2:
3215 out_free_1:
3216         kvm_arch_hardware_unsetup();
3217 out_free_0a:
3218         free_cpumask_var(cpus_hardware_enabled);
3219 out_free_0:
3220         kvm_irqfd_exit();
3221 out_irqfd:
3222         kvm_arch_exit();
3223 out_fail:
3224         return r;
3225 }
3226 EXPORT_SYMBOL_GPL(kvm_init);
3227
3228 void kvm_exit(void)
3229 {
3230         kvm_exit_debug();
3231         misc_deregister(&kvm_dev);
3232         kmem_cache_destroy(kvm_vcpu_cache);
3233         kvm_async_pf_deinit();
3234         unregister_syscore_ops(&kvm_syscore_ops);
3235         unregister_reboot_notifier(&kvm_reboot_notifier);
3236         unregister_cpu_notifier(&kvm_cpu_notifier);
3237         on_each_cpu(hardware_disable_nolock, NULL, 1);
3238         kvm_arch_hardware_unsetup();
3239         kvm_arch_exit();
3240         kvm_irqfd_exit();
3241         free_cpumask_var(cpus_hardware_enabled);
3242 }
3243 EXPORT_SYMBOL_GPL(kvm_exit);