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 #ifndef CONFIG_S390
 641         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 642
 643         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 644         if (!memslot->dirty_bitmap)
 645                 return -ENOMEM;
 646
 647 #endif /* !CONFIG_S390 */
 648         return 0;
 649 }
 650
 651 static int cmp_memslot(const void *slot1, const void *slot2)
 652 {
 653         struct kvm_memory_slot *s1, *s2;
 654
 655         s1 = (struct kvm_memory_slot *)slot1;
 656         s2 = (struct kvm_memory_slot *)slot2;
 657
 658         if (s1->npages < s2->npages)
 659                 return 1;
 660         if (s1->npages > s2->npages)
 661                 return -1;
 662
 663         return 0;
 664 }
 665
 666 /*
 667  * Sort the memslots base on its size, so the larger slots
 668  * will get better fit.
 669  */
 670 static void sort_memslots(struct kvm_memslots *slots)
 671 {
 672         int i;
 673
 674         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 675               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 676
 677         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 678                 slots->id_to_index[slots->memslots[i].id] = i;
 679 }
 680
 681 static void update_memslots(struct kvm_memslots *slots,
 682                             struct kvm_memory_slot *new,
 683                             u64 last_generation)
 684 {
 685         if (new) {
 686                 int id = new->id;
 687                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 688                 unsigned long npages = old->npages;
 689
 690                 *old = *new;
 691                 if (new->npages != npages)
 692                         sort_memslots(slots);
 693         }
 694
 695         slots->generation = last_generation + 1;
 696 }
 697
 698 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 699 {
 700         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 701
 702 #ifdef KVM_CAP_READONLY_MEM
 703         valid_flags |= KVM_MEM_READONLY;
 704 #endif
 705
 706         if (mem->flags & ~valid_flags)
 707                 return -EINVAL;
 708
 709         return 0;
 710 }
 711
 712 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 713                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 714 {
 715         struct kvm_memslots *old_memslots = kvm->memslots;
 716
 717         update_memslots(slots, new, kvm->memslots->generation);
 718         rcu_assign_pointer(kvm->memslots, slots);
 719         synchronize_srcu_expedited(&kvm->srcu);
 720
 721         kvm_arch_memslots_updated(kvm);
 722
 723         return old_memslots;
 724 }
 725
 726 /*
 727  * Allocate some memory and give it an address in the guest physical address
 728  * space.
 729  *
 730  * Discontiguous memory is allowed, mostly for framebuffers.
 731  *
 732  * Must be called holding mmap_sem for write.
 733  */
 734 int __kvm_set_memory_region(struct kvm *kvm,
 735                             struct kvm_userspace_memory_region *mem)
 736 {
 737         int r;
 738         gfn_t base_gfn;
 739         unsigned long npages;
 740         struct kvm_memory_slot *slot;
 741         struct kvm_memory_slot old, new;
 742         struct kvm_memslots *slots = NULL, *old_memslots;
 743         enum kvm_mr_change change;
 744
 745         r = check_memory_region_flags(mem);
 746         if (r)
 747                 goto out;
 748
 749         r = -EINVAL;
 750         /* General sanity checks */
 751         if (mem->memory_size & (PAGE_SIZE - 1))
 752                 goto out;
 753         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 754                 goto out;
 755         /* We can read the guest memory with __xxx_user() later on. */
 756         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 757             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 758              !access_ok(VERIFY_WRITE,
 759                         (void __user *)(unsigned long)mem->userspace_addr,
 760                         mem->memory_size)))
 761                 goto out;
 762         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 763                 goto out;
 764         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 765                 goto out;
 766
 767         slot = id_to_memslot(kvm->memslots, mem->slot);
 768         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 769         npages = mem->memory_size >> PAGE_SHIFT;
 770
 771         r = -EINVAL;
 772         if (npages > KVM_MEM_MAX_NR_PAGES)
 773                 goto out;
 774
 775         if (!npages)
 776                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 777
 778         new = old = *slot;
 779
 780         new.id = mem->slot;
 781         new.base_gfn = base_gfn;
 782         new.npages = npages;
 783         new.flags = mem->flags;
 784
 785         r = -EINVAL;
 786         if (npages) {
 787                 if (!old.npages)
 788                         change = KVM_MR_CREATE;
 789                 else { /* Modify an existing slot. */
 790                         if ((mem->userspace_addr != old.userspace_addr) ||
 791                             (npages != old.npages) ||
 792                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 793                                 goto out;
 794
 795                         if (base_gfn != old.base_gfn)
 796                                 change = KVM_MR_MOVE;
 797                         else if (new.flags != old.flags)
 798                                 change = KVM_MR_FLAGS_ONLY;
 799                         else { /* Nothing to change. */
 800                                 r = 0;
 801                                 goto out;
 802                         }
 803                 }
 804         } else if (old.npages) {
 805                 change = KVM_MR_DELETE;
 806         } else /* Modify a non-existent slot: disallowed. */
 807                 goto out;
 808
 809         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 810                 /* Check for overlaps */
 811                 r = -EEXIST;
 812                 kvm_for_each_memslot(slot, kvm->memslots) {
 813                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 814                             (slot->id == mem->slot))
 815                                 continue;
 816                         if (!((base_gfn + npages <= slot->base_gfn) ||
 817                               (base_gfn >= slot->base_gfn + slot->npages)))
 818                                 goto out;
 819                 }
 820         }
 821
 822         /* Free page dirty bitmap if unneeded */
 823         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 824                 new.dirty_bitmap = NULL;
 825
 826         r = -ENOMEM;
 827         if (change == KVM_MR_CREATE) {
 828                 new.userspace_addr = mem->userspace_addr;
 829
 830                 if (kvm_arch_create_memslot(kvm, &new, npages))
 831                         goto out_free;
 832         }
 833
 834         /* Allocate page dirty bitmap if needed */
 835         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 836                 if (kvm_create_dirty_bitmap(&new) < 0)
 837                         goto out_free;
 838         }
 839
 840         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 841                 r = -ENOMEM;
 842                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 843                                 GFP_KERNEL);
 844                 if (!slots)
 845                         goto out_free;
 846                 slot = id_to_memslot(slots, mem->slot);
 847                 slot->flags |= KVM_MEMSLOT_INVALID;
 848
 849                 old_memslots = install_new_memslots(kvm, slots, NULL);
 850
 851                 /* slot was deleted or moved, clear iommu mapping */
 852                 kvm_iommu_unmap_pages(kvm, &old);
 853                 /* From this point no new shadow pages pointing to a deleted,
 854                  * or moved, memslot will be created.
 855                  *
 856                  * validation of sp->gfn happens in:
 857                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 858                  *      - kvm_is_visible_gfn (mmu_check_roots)
 859                  */
 860                 kvm_arch_flush_shadow_memslot(kvm, slot);
 861                 slots = old_memslots;
 862         }
 863
 864         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 865         if (r)
 866                 goto out_slots;
 867
 868         r = -ENOMEM;
 869         /*
 870          * We can re-use the old_memslots from above, the only difference
 871          * from the currently installed memslots is the invalid flag.  This
 872          * will get overwritten by update_memslots anyway.
 873          */
 874         if (!slots) {
 875                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 876                                 GFP_KERNEL);
 877                 if (!slots)
 878                         goto out_free;
 879         }
 880
 881         /* actual memory is freed via old in kvm_free_physmem_slot below */
 882         if (change == KVM_MR_DELETE) {
 883                 new.dirty_bitmap = NULL;
 884                 memset(&new.arch, 0, sizeof(new.arch));
 885         }
 886
 887         old_memslots = install_new_memslots(kvm, slots, &new);
 888
 889         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 890
 891         kvm_free_physmem_slot(kvm, &old, &new);
 892         kfree(old_memslots);
 893
 894         /*
 895          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 896          * un-mapped and re-mapped if their base changes.  Since base change
 897          * unmapping is handled above with slot deletion, mapping alone is
 898          * needed here.  Anything else the iommu might care about for existing
 899          * slots (size changes, userspace addr changes and read-only flag
 900          * changes) is disallowed above, so any other attribute changes getting
 901          * here can be skipped.
 902          */
 903         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 904                 r = kvm_iommu_map_pages(kvm, &new);
 905                 return r;
 906         }
 907
 908         return 0;
 909
 910 out_slots:
 911         kfree(slots);
 912 out_free:
 913         kvm_free_physmem_slot(kvm, &new, &old);
 914 out:
 915         return r;
 916 }
 917 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 918
 919 int kvm_set_memory_region(struct kvm *kvm,
 920                           struct kvm_userspace_memory_region *mem)
 921 {
 922         int r;
 923
 924         mutex_lock(&kvm->slots_lock);
 925         r = __kvm_set_memory_region(kvm, mem);
 926         mutex_unlock(&kvm->slots_lock);
 927         return r;
 928 }
 929 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 930
 931 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 932                                           struct kvm_userspace_memory_region *mem)
 933 {
 934         if (mem->slot >= KVM_USER_MEM_SLOTS)
 935                 return -EINVAL;
 936         return kvm_set_memory_region(kvm, mem);
 937 }
 938
 939 int kvm_get_dirty_log(struct kvm *kvm,
 940                         struct kvm_dirty_log *log, int *is_dirty)
 941 {
 942         struct kvm_memory_slot *memslot;
 943         int r, i;
 944         unsigned long n;
 945         unsigned long any = 0;
 946
 947         r = -EINVAL;
 948         if (log->slot >= KVM_USER_MEM_SLOTS)
 949                 goto out;
 950
 951         memslot = id_to_memslot(kvm->memslots, log->slot);
 952         r = -ENOENT;
 953         if (!memslot->dirty_bitmap)
 954                 goto out;
 955
 956         n = kvm_dirty_bitmap_bytes(memslot);
 957
 958         for (i = 0; !any && i < n/sizeof(long); ++i)
 959                 any = memslot->dirty_bitmap[i];
 960
 961         r = -EFAULT;
 962         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 963                 goto out;
 964
 965         if (any)
 966                 *is_dirty = 1;
 967
 968         r = 0;
 969 out:
 970         return r;
 971 }
 972 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
 973
 974 bool kvm_largepages_enabled(void)
 975 {
 976         return largepages_enabled;
 977 }
 978
 979 void kvm_disable_largepages(void)
 980 {
 981         largepages_enabled = false;
 982 }
 983 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 984
 985 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 986 {
 987         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 988 }
 989 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 990
 991 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 992 {
 993         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 994
 995         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
 996               memslot->flags & KVM_MEMSLOT_INVALID)
 997                 return 0;
 998
 999         return 1;
1000 }
1001 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1002
1003 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1004 {
1005         struct vm_area_struct *vma;
1006         unsigned long addr, size;
1007
1008         size = PAGE_SIZE;
1009
1010         addr = gfn_to_hva(kvm, gfn);
1011         if (kvm_is_error_hva(addr))
1012                 return PAGE_SIZE;
1013
1014         down_read(&current->mm->mmap_sem);
1015         vma = find_vma(current->mm, addr);
1016         if (!vma)
1017                 goto out;
1018
1019         size = vma_kernel_pagesize(vma);
1020
1021 out:
1022         up_read(&current->mm->mmap_sem);
1023
1024         return size;
1025 }
1026
1027 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1028 {
1029         return slot->flags & KVM_MEM_READONLY;
1030 }
1031
1032 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1033                                        gfn_t *nr_pages, bool write)
1034 {
1035         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1036                 return KVM_HVA_ERR_BAD;
1037
1038         if (memslot_is_readonly(slot) && write)
1039                 return KVM_HVA_ERR_RO_BAD;
1040
1041         if (nr_pages)
1042                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1043
1044         return __gfn_to_hva_memslot(slot, gfn);
1045 }
1046
1047 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1048                                      gfn_t *nr_pages)
1049 {
1050         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1051 }
1052
1053 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1054                                         gfn_t gfn)
1055 {
1056         return gfn_to_hva_many(slot, gfn, NULL);
1057 }
1058 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1059
1060 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1061 {
1062         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1063 }
1064 EXPORT_SYMBOL_GPL(gfn_to_hva);
1065
1066 /*
1067  * If writable is set to false, the hva returned by this function is only
1068  * allowed to be read.
1069  */
1070 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1071 {
1072         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1073         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1074
1075         if (!kvm_is_error_hva(hva) && writable)
1076                 *writable = !memslot_is_readonly(slot);
1077
1078         return hva;
1079 }
1080
1081 static int kvm_read_hva(void *data, void __user *hva, int len)
1082 {
1083         return __copy_from_user(data, hva, len);
1084 }
1085
1086 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1087 {
1088         return __copy_from_user_inatomic(data, hva, len);
1089 }
1090
1091 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1092         unsigned long start, int write, struct page **page)
1093 {
1094         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1095
1096         if (write)
1097                 flags |= FOLL_WRITE;
1098
1099         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1100 }
1101
1102 static inline int check_user_page_hwpoison(unsigned long addr)
1103 {
1104         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1105
1106         rc = __get_user_pages(current, current->mm, addr, 1,
1107                               flags, NULL, NULL, NULL);
1108         return rc == -EHWPOISON;
1109 }
1110
1111 /*
1112  * The atomic path to get the writable pfn which will be stored in @pfn,
1113  * true indicates success, otherwise false is returned.
1114  */
1115 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1116                             bool write_fault, bool *writable, pfn_t *pfn)
1117 {
1118         struct page *page[1];
1119         int npages;
1120
1121         if (!(async || atomic))
1122                 return false;
1123
1124         /*
1125          * Fast pin a writable pfn only if it is a write fault request
1126          * or the caller allows to map a writable pfn for a read fault
1127          * request.
1128          */
1129         if (!(write_fault || writable))
1130                 return false;
1131
1132         npages = __get_user_pages_fast(addr, 1, 1, page);
1133         if (npages == 1) {
1134                 *pfn = page_to_pfn(page[0]);
1135
1136                 if (writable)
1137                         *writable = true;
1138                 return true;
1139         }
1140
1141         return false;
1142 }
1143
1144 /*
1145  * The slow path to get the pfn of the specified host virtual address,
1146  * 1 indicates success, -errno is returned if error is detected.
1147  */
1148 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1149                            bool *writable, pfn_t *pfn)
1150 {
1151         struct page *page[1];
1152         int npages = 0;
1153
1154         might_sleep();
1155
1156         if (writable)
1157                 *writable = write_fault;
1158
1159         if (async) {
1160                 down_read(&current->mm->mmap_sem);
1161                 npages = get_user_page_nowait(current, current->mm,
1162                                               addr, write_fault, page);
1163                 up_read(&current->mm->mmap_sem);
1164         } else
1165                 npages = get_user_pages_fast(addr, 1, write_fault,
1166                                              page);
1167         if (npages != 1)
1168                 return npages;
1169
1170         /* map read fault as writable if possible */
1171         if (unlikely(!write_fault) && writable) {
1172                 struct page *wpage[1];
1173
1174                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1175                 if (npages == 1) {
1176                         *writable = true;
1177                         put_page(page[0]);
1178                         page[0] = wpage[0];
1179                 }
1180
1181                 npages = 1;
1182         }
1183         *pfn = page_to_pfn(page[0]);
1184         return npages;
1185 }
1186
1187 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1188 {
1189         if (unlikely(!(vma->vm_flags & VM_READ)))
1190                 return false;
1191
1192         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1193                 return false;
1194
1195         return true;
1196 }
1197
1198 /*
1199  * Pin guest page in memory and return its pfn.
1200  * @addr: host virtual address which maps memory to the guest
1201  * @atomic: whether this function can sleep
1202  * @async: whether this function need to wait IO complete if the
1203  *         host page is not in the memory
1204  * @write_fault: whether we should get a writable host page
1205  * @writable: whether it allows to map a writable host page for !@write_fault
1206  *
1207  * The function will map a writable host page for these two cases:
1208  * 1): @write_fault = true
1209  * 2): @write_fault = false && @writable, @writable will tell the caller
1210  *     whether the mapping is writable.
1211  */
1212 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1213                         bool write_fault, bool *writable)
1214 {
1215         struct vm_area_struct *vma;
1216         pfn_t pfn = 0;
1217         int npages;
1218
1219         /* we can do it either atomically or asynchronously, not both */
1220         BUG_ON(atomic && async);
1221
1222         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1223                 return pfn;
1224
1225         if (atomic)
1226                 return KVM_PFN_ERR_FAULT;
1227
1228         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1229         if (npages == 1)
1230                 return pfn;
1231
1232         down_read(&current->mm->mmap_sem);
1233         if (npages == -EHWPOISON ||
1234               (!async && check_user_page_hwpoison(addr))) {
1235                 pfn = KVM_PFN_ERR_HWPOISON;
1236                 goto exit;
1237         }
1238
1239         vma = find_vma_intersection(current->mm, addr, addr + 1);
1240
1241         if (vma == NULL)
1242                 pfn = KVM_PFN_ERR_FAULT;
1243         else if ((vma->vm_flags & VM_PFNMAP)) {
1244                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1245                         vma->vm_pgoff;
1246                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1247         } else {
1248                 if (async && vma_is_valid(vma, write_fault))
1249                         *async = true;
1250                 pfn = KVM_PFN_ERR_FAULT;
1251         }
1252 exit:
1253         up_read(&current->mm->mmap_sem);
1254         return pfn;
1255 }
1256
1257 static pfn_t
1258 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1259                      bool *async, bool write_fault, bool *writable)
1260 {
1261         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1262
1263         if (addr == KVM_HVA_ERR_RO_BAD)
1264                 return KVM_PFN_ERR_RO_FAULT;
1265
1266         if (kvm_is_error_hva(addr))
1267                 return KVM_PFN_NOSLOT;
1268
1269         /* Do not map writable pfn in the readonly memslot. */
1270         if (writable && memslot_is_readonly(slot)) {
1271                 *writable = false;
1272                 writable = NULL;
1273         }
1274
1275         return hva_to_pfn(addr, atomic, async, write_fault,
1276                           writable);
1277 }
1278
1279 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1280                           bool write_fault, bool *writable)
1281 {
1282         struct kvm_memory_slot *slot;
1283
1284         if (async)
1285                 *async = false;
1286
1287         slot = gfn_to_memslot(kvm, gfn);
1288
1289         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1290                                     writable);
1291 }
1292
1293 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1294 {
1295         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1296 }
1297 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1298
1299 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1300                        bool write_fault, bool *writable)
1301 {
1302         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1303 }
1304 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1305
1306 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1307 {
1308         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1309 }
1310 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1311
1312 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1313                       bool *writable)
1314 {
1315         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1316 }
1317 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1318
1319 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1320 {
1321         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1322 }
1323
1324 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1325 {
1326         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1327 }
1328 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1329
1330 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1331                                                                   int nr_pages)
1332 {
1333         unsigned long addr;
1334         gfn_t entry;
1335
1336         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1337         if (kvm_is_error_hva(addr))
1338                 return -1;
1339
1340         if (entry < nr_pages)
1341                 return 0;
1342
1343         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1344 }
1345 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1346
1347 static struct page *kvm_pfn_to_page(pfn_t pfn)
1348 {
1349         if (is_error_noslot_pfn(pfn))
1350                 return KVM_ERR_PTR_BAD_PAGE;
1351
1352         if (kvm_is_mmio_pfn(pfn)) {
1353                 WARN_ON(1);
1354                 return KVM_ERR_PTR_BAD_PAGE;
1355         }
1356
1357         return pfn_to_page(pfn);
1358 }
1359
1360 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1361 {
1362         pfn_t pfn;
1363
1364         pfn = gfn_to_pfn(kvm, gfn);
1365
1366         return kvm_pfn_to_page(pfn);
1367 }
1368
1369 EXPORT_SYMBOL_GPL(gfn_to_page);
1370
1371 void kvm_release_page_clean(struct page *page)
1372 {
1373         WARN_ON(is_error_page(page));
1374
1375         kvm_release_pfn_clean(page_to_pfn(page));
1376 }
1377 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1378
1379 void kvm_release_pfn_clean(pfn_t pfn)
1380 {
1381         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1382                 put_page(pfn_to_page(pfn));
1383 }
1384 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1385
1386 void kvm_release_page_dirty(struct page *page)
1387 {
1388         WARN_ON(is_error_page(page));
1389
1390         kvm_release_pfn_dirty(page_to_pfn(page));
1391 }
1392 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1393
1394 static void kvm_release_pfn_dirty(pfn_t pfn)
1395 {
1396         kvm_set_pfn_dirty(pfn);
1397         kvm_release_pfn_clean(pfn);
1398 }
1399
1400 void kvm_set_pfn_dirty(pfn_t pfn)
1401 {
1402         if (!kvm_is_mmio_pfn(pfn)) {
1403                 struct page *page = pfn_to_page(pfn);
1404                 if (!PageReserved(page))
1405                         SetPageDirty(page);
1406         }
1407 }
1408 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1409
1410 void kvm_set_pfn_accessed(pfn_t pfn)
1411 {
1412         if (!kvm_is_mmio_pfn(pfn))
1413                 mark_page_accessed(pfn_to_page(pfn));
1414 }
1415 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1416
1417 void kvm_get_pfn(pfn_t pfn)
1418 {
1419         if (!kvm_is_mmio_pfn(pfn))
1420                 get_page(pfn_to_page(pfn));
1421 }
1422 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1423
1424 static int next_segment(unsigned long len, int offset)
1425 {
1426         if (len > PAGE_SIZE - offset)
1427                 return PAGE_SIZE - offset;
1428         else
1429                 return len;
1430 }
1431
1432 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1433                         int len)
1434 {
1435         int r;
1436         unsigned long addr;
1437
1438         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1439         if (kvm_is_error_hva(addr))
1440                 return -EFAULT;
1441         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1442         if (r)
1443                 return -EFAULT;
1444         return 0;
1445 }
1446 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1447
1448 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1449 {
1450         gfn_t gfn = gpa >> PAGE_SHIFT;
1451         int seg;
1452         int offset = offset_in_page(gpa);
1453         int ret;
1454
1455         while ((seg = next_segment(len, offset)) != 0) {
1456                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1457                 if (ret < 0)
1458                         return ret;
1459                 offset = 0;
1460                 len -= seg;
1461                 data += seg;
1462                 ++gfn;
1463         }
1464         return 0;
1465 }
1466 EXPORT_SYMBOL_GPL(kvm_read_guest);
1467
1468 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1469                           unsigned long len)
1470 {
1471         int r;
1472         unsigned long addr;
1473         gfn_t gfn = gpa >> PAGE_SHIFT;
1474         int offset = offset_in_page(gpa);
1475
1476         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1477         if (kvm_is_error_hva(addr))
1478                 return -EFAULT;
1479         pagefault_disable();
1480         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1481         pagefault_enable();
1482         if (r)
1483                 return -EFAULT;
1484         return 0;
1485 }
1486 EXPORT_SYMBOL(kvm_read_guest_atomic);
1487
1488 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1489                          int offset, int len)
1490 {
1491         int r;
1492         unsigned long addr;
1493
1494         addr = gfn_to_hva(kvm, gfn);
1495         if (kvm_is_error_hva(addr))
1496                 return -EFAULT;
1497         r = __copy_to_user((void __user *)addr + offset, data, len);
1498         if (r)
1499                 return -EFAULT;
1500         mark_page_dirty(kvm, gfn);
1501         return 0;
1502 }
1503 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1504
1505 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1506                     unsigned long len)
1507 {
1508         gfn_t gfn = gpa >> PAGE_SHIFT;
1509         int seg;
1510         int offset = offset_in_page(gpa);
1511         int ret;
1512
1513         while ((seg = next_segment(len, offset)) != 0) {
1514                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1515                 if (ret < 0)
1516                         return ret;
1517                 offset = 0;
1518                 len -= seg;
1519                 data += seg;
1520                 ++gfn;
1521         }
1522         return 0;
1523 }
1524
1525 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1526                               gpa_t gpa, unsigned long len)
1527 {
1528         struct kvm_memslots *slots = kvm_memslots(kvm);
1529         int offset = offset_in_page(gpa);
1530         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1531         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1532         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1533         gfn_t nr_pages_avail;
1534
1535         ghc->gpa = gpa;
1536         ghc->generation = slots->generation;
1537         ghc->len = len;
1538         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1539         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1540         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1541                 ghc->hva += offset;
1542         } else {
1543                 /*
1544                  * If the requested region crosses two memslots, we still
1545                  * verify that the entire region is valid here.
1546                  */
1547                 while (start_gfn <= end_gfn) {
1548                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1549                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1550                                                    &nr_pages_avail);
1551                         if (kvm_is_error_hva(ghc->hva))
1552                                 return -EFAULT;
1553                         start_gfn += nr_pages_avail;
1554                 }
1555                 /* Use the slow path for cross page reads and writes. */
1556                 ghc->memslot = NULL;
1557         }
1558         return 0;
1559 }
1560 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1561
1562 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1563                            void *data, unsigned long len)
1564 {
1565         struct kvm_memslots *slots = kvm_memslots(kvm);
1566         int r;
1567
1568         BUG_ON(len > ghc->len);
1569
1570         if (slots->generation != ghc->generation)
1571                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1572
1573         if (unlikely(!ghc->memslot))
1574                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1575
1576         if (kvm_is_error_hva(ghc->hva))
1577                 return -EFAULT;
1578
1579         r = __copy_to_user((void __user *)ghc->hva, data, len);
1580         if (r)
1581                 return -EFAULT;
1582         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1583
1584         return 0;
1585 }
1586 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1587
1588 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1589                            void *data, unsigned long len)
1590 {
1591         struct kvm_memslots *slots = kvm_memslots(kvm);
1592         int r;
1593
1594         BUG_ON(len > ghc->len);
1595
1596         if (slots->generation != ghc->generation)
1597                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1598
1599         if (unlikely(!ghc->memslot))
1600                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1601
1602         if (kvm_is_error_hva(ghc->hva))
1603                 return -EFAULT;
1604
1605         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1606         if (r)
1607                 return -EFAULT;
1608
1609         return 0;
1610 }
1611 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1612
1613 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1614 {
1615         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1616
1617         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1618 }
1619 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1620
1621 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1622 {
1623         gfn_t gfn = gpa >> PAGE_SHIFT;
1624         int seg;
1625         int offset = offset_in_page(gpa);
1626         int ret;
1627
1628         while ((seg = next_segment(len, offset)) != 0) {
1629                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1630                 if (ret < 0)
1631                         return ret;
1632                 offset = 0;
1633                 len -= seg;
1634                 ++gfn;
1635         }
1636         return 0;
1637 }
1638 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1639
1640 static void mark_page_dirty_in_slot(struct kvm *kvm,
1641                                     struct kvm_memory_slot *memslot,
1642                                     gfn_t gfn)
1643 {
1644         if (memslot && memslot->dirty_bitmap) {
1645                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1646
1647                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1648         }
1649 }
1650
1651 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1652 {
1653         struct kvm_memory_slot *memslot;
1654
1655         memslot = gfn_to_memslot(kvm, gfn);
1656         mark_page_dirty_in_slot(kvm, memslot, gfn);
1657 }
1658 EXPORT_SYMBOL_GPL(mark_page_dirty);
1659
1660 /*
1661  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1662  */
1663 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1664 {
1665         DEFINE_WAIT(wait);
1666
1667         for (;;) {
1668                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1669
1670                 if (kvm_arch_vcpu_runnable(vcpu)) {
1671                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1672                         break;
1673                 }
1674                 if (kvm_cpu_has_pending_timer(vcpu))
1675                         break;
1676                 if (signal_pending(current))
1677                         break;
1678
1679                 schedule();
1680         }
1681
1682         finish_wait(&vcpu->wq, &wait);
1683 }
1684 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1685
1686 #ifndef CONFIG_S390
1687 /*
1688  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1689  */
1690 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1691 {
1692         int me;
1693         int cpu = vcpu->cpu;
1694         wait_queue_head_t *wqp;
1695
1696         wqp = kvm_arch_vcpu_wq(vcpu);
1697         if (waitqueue_active(wqp)) {
1698                 wake_up_interruptible(wqp);
1699                 ++vcpu->stat.halt_wakeup;
1700         }
1701
1702         me = get_cpu();
1703         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1704                 if (kvm_arch_vcpu_should_kick(vcpu))
1705                         smp_send_reschedule(cpu);
1706         put_cpu();
1707 }
1708 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1709 #endif /* !CONFIG_S390 */
1710
1711 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1712 {
1713         struct pid *pid;
1714         struct task_struct *task = NULL;
1715         bool ret = false;
1716
1717         rcu_read_lock();
1718         pid = rcu_dereference(target->pid);
1719         if (pid)
1720                 task = get_pid_task(target->pid, PIDTYPE_PID);
1721         rcu_read_unlock();
1722         if (!task)
1723                 return ret;
1724         if (task->flags & PF_VCPU) {
1725                 put_task_struct(task);
1726                 return ret;
1727         }
1728         ret = yield_to(task, 1);
1729         put_task_struct(task);
1730
1731         return ret;
1732 }
1733 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1734
1735 /*
1736  * Helper that checks whether a VCPU is eligible for directed yield.
1737  * Most eligible candidate to yield is decided by following heuristics:
1738  *
1739  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1740  *  (preempted lock holder), indicated by @in_spin_loop.
1741  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1742  *
1743  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1744  *  chance last time (mostly it has become eligible now since we have probably
1745  *  yielded to lockholder in last iteration. This is done by toggling
1746  *  @dy_eligible each time a VCPU checked for eligibility.)
1747  *
1748  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1749  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1750  *  burning. Giving priority for a potential lock-holder increases lock
1751  *  progress.
1752  *
1753  *  Since algorithm is based on heuristics, accessing another VCPU data without
1754  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1755  *  and continue with next VCPU and so on.
1756  */
1757 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1758 {
1759 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1760         bool eligible;
1761
1762         eligible = !vcpu->spin_loop.in_spin_loop ||
1763                         (vcpu->spin_loop.in_spin_loop &&
1764                          vcpu->spin_loop.dy_eligible);
1765
1766         if (vcpu->spin_loop.in_spin_loop)
1767                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1768
1769         return eligible;
1770 #else
1771         return true;
1772 #endif
1773 }
1774
1775 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1776 {
1777         struct kvm *kvm = me->kvm;
1778         struct kvm_vcpu *vcpu;
1779         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1780         int yielded = 0;
1781         int try = 3;
1782         int pass;
1783         int i;
1784
1785         kvm_vcpu_set_in_spin_loop(me, true);
1786         /*
1787          * We boost the priority of a VCPU that is runnable but not
1788          * currently running, because it got preempted by something
1789          * else and called schedule in __vcpu_run.  Hopefully that
1790          * VCPU is holding the lock that we need and will release it.
1791          * We approximate round-robin by starting at the last boosted VCPU.
1792          */
1793         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1794                 kvm_for_each_vcpu(i, vcpu, kvm) {
1795                         if (!pass && i <= last_boosted_vcpu) {
1796                                 i = last_boosted_vcpu;
1797                                 continue;
1798                         } else if (pass && i > last_boosted_vcpu)
1799                                 break;
1800                         if (!ACCESS_ONCE(vcpu->preempted))
1801                                 continue;
1802                         if (vcpu == me)
1803                                 continue;
1804                         if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
1805                                 continue;
1806                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1807                                 continue;
1808
1809                         yielded = kvm_vcpu_yield_to(vcpu);
1810                         if (yielded > 0) {
1811                                 kvm->last_boosted_vcpu = i;
1812                                 break;
1813                         } else if (yielded < 0) {
1814                                 try--;
1815                                 if (!try)
1816                                         break;
1817                         }
1818                 }
1819         }
1820         kvm_vcpu_set_in_spin_loop(me, false);
1821
1822         /* Ensure vcpu is not eligible during next spinloop */
1823         kvm_vcpu_set_dy_eligible(me, false);
1824 }
1825 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1826
1827 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1828 {
1829         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1830         struct page *page;
1831
1832         if (vmf->pgoff == 0)
1833                 page = virt_to_page(vcpu->run);
1834 #ifdef CONFIG_X86
1835         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1836                 page = virt_to_page(vcpu->arch.pio_data);
1837 #endif
1838 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1839         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1840                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1841 #endif
1842         else
1843                 return kvm_arch_vcpu_fault(vcpu, vmf);
1844         get_page(page);
1845         vmf->page = page;
1846         return 0;
1847 }
1848
1849 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1850         .fault = kvm_vcpu_fault,
1851 };
1852
1853 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1854 {
1855         vma->vm_ops = &kvm_vcpu_vm_ops;
1856         return 0;
1857 }
1858
1859 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1860 {
1861         struct kvm_vcpu *vcpu = filp->private_data;
1862
1863         kvm_put_kvm(vcpu->kvm);
1864         return 0;
1865 }
1866
1867 static struct file_operations kvm_vcpu_fops = {
1868         .release        = kvm_vcpu_release,
1869         .unlocked_ioctl = kvm_vcpu_ioctl,
1870 #ifdef CONFIG_COMPAT
1871         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1872 #endif
1873         .mmap           = kvm_vcpu_mmap,
1874         .llseek         = noop_llseek,
1875 };
1876
1877 /*
1878  * Allocates an inode for the vcpu.
1879  */
1880 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1881 {
1882         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1883 }
1884
1885 /*
1886  * Creates some virtual cpus.  Good luck creating more than one.
1887  */
1888 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1889 {
1890         int r;
1891         struct kvm_vcpu *vcpu, *v;
1892
1893         if (id >= KVM_MAX_VCPUS)
1894                 return -EINVAL;
1895
1896         vcpu = kvm_arch_vcpu_create(kvm, id);
1897         if (IS_ERR(vcpu))
1898                 return PTR_ERR(vcpu);
1899
1900         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1901
1902         r = kvm_arch_vcpu_setup(vcpu);
1903         if (r)
1904                 goto vcpu_destroy;
1905
1906         mutex_lock(&kvm->lock);
1907         if (!kvm_vcpu_compatible(vcpu)) {
1908                 r = -EINVAL;
1909                 goto unlock_vcpu_destroy;
1910         }
1911         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1912                 r = -EINVAL;
1913                 goto unlock_vcpu_destroy;
1914         }
1915
1916         kvm_for_each_vcpu(r, v, kvm)
1917                 if (v->vcpu_id == id) {
1918                         r = -EEXIST;
1919                         goto unlock_vcpu_destroy;
1920                 }
1921
1922         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1923
1924         /* Now it's all set up, let userspace reach it */
1925         kvm_get_kvm(kvm);
1926         r = create_vcpu_fd(vcpu);
1927         if (r < 0) {
1928                 kvm_put_kvm(kvm);
1929                 goto unlock_vcpu_destroy;
1930         }
1931
1932         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1933         smp_wmb();
1934         atomic_inc(&kvm->online_vcpus);
1935
1936         mutex_unlock(&kvm->lock);
1937         kvm_arch_vcpu_postcreate(vcpu);
1938         return r;
1939
1940 unlock_vcpu_destroy:
1941         mutex_unlock(&kvm->lock);
1942 vcpu_destroy:
1943         kvm_arch_vcpu_destroy(vcpu);
1944         return r;
1945 }
1946
1947 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1948 {
1949         if (sigset) {
1950                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1951                 vcpu->sigset_active = 1;
1952                 vcpu->sigset = *sigset;
1953         } else
1954                 vcpu->sigset_active = 0;
1955         return 0;
1956 }
1957
1958 static long kvm_vcpu_ioctl(struct file *filp,
1959                            unsigned int ioctl, unsigned long arg)
1960 {
1961         struct kvm_vcpu *vcpu = filp->private_data;
1962         void __user *argp = (void __user *)arg;
1963         int r;
1964         struct kvm_fpu *fpu = NULL;
1965         struct kvm_sregs *kvm_sregs = NULL;
1966
1967         if (vcpu->kvm->mm != current->mm)
1968                 return -EIO;
1969
1970 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1971         /*
1972          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1973          * so vcpu_load() would break it.
1974          */
1975         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1976                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1977 #endif
1978
1979
1980         r = vcpu_load(vcpu);
1981         if (r)
1982                 return r;
1983         switch (ioctl) {
1984         case KVM_RUN:
1985                 r = -EINVAL;
1986                 if (arg)
1987                         goto out;
1988                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1989                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1990                 break;
1991         case KVM_GET_REGS: {
1992                 struct kvm_regs *kvm_regs;
1993
1994                 r = -ENOMEM;
1995                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1996                 if (!kvm_regs)
1997                         goto out;
1998                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1999                 if (r)
2000                         goto out_free1;
2001                 r = -EFAULT;
2002                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2003                         goto out_free1;
2004                 r = 0;
2005 out_free1:
2006                 kfree(kvm_regs);
2007                 break;
2008         }
2009         case KVM_SET_REGS: {
2010                 struct kvm_regs *kvm_regs;
2011
2012                 r = -ENOMEM;
2013                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2014                 if (IS_ERR(kvm_regs)) {
2015                         r = PTR_ERR(kvm_regs);
2016                         goto out;
2017                 }
2018                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2019                 kfree(kvm_regs);
2020                 break;
2021         }
2022         case KVM_GET_SREGS: {
2023                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2024                 r = -ENOMEM;
2025                 if (!kvm_sregs)
2026                         goto out;
2027                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2028                 if (r)
2029                         goto out;
2030                 r = -EFAULT;
2031                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2032                         goto out;
2033                 r = 0;
2034                 break;
2035         }
2036         case KVM_SET_SREGS: {
2037                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2038                 if (IS_ERR(kvm_sregs)) {
2039                         r = PTR_ERR(kvm_sregs);
2040                         kvm_sregs = NULL;
2041                         goto out;
2042                 }
2043                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2044                 break;
2045         }
2046         case KVM_GET_MP_STATE: {
2047                 struct kvm_mp_state mp_state;
2048
2049                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2050                 if (r)
2051                         goto out;
2052                 r = -EFAULT;
2053                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2054                         goto out;
2055                 r = 0;
2056                 break;
2057         }
2058         case KVM_SET_MP_STATE: {
2059                 struct kvm_mp_state mp_state;
2060
2061                 r = -EFAULT;
2062                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2063                         goto out;
2064                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2065                 break;
2066         }
2067         case KVM_TRANSLATE: {
2068                 struct kvm_translation tr;
2069
2070                 r = -EFAULT;
2071                 if (copy_from_user(&tr, argp, sizeof tr))
2072                         goto out;
2073                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2074                 if (r)
2075                         goto out;
2076                 r = -EFAULT;
2077                 if (copy_to_user(argp, &tr, sizeof tr))
2078                         goto out;
2079                 r = 0;
2080                 break;
2081         }
2082         case KVM_SET_GUEST_DEBUG: {
2083                 struct kvm_guest_debug dbg;
2084
2085                 r = -EFAULT;
2086                 if (copy_from_user(&dbg, argp, sizeof dbg))
2087                         goto out;
2088                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2089                 break;
2090         }
2091         case KVM_SET_SIGNAL_MASK: {
2092                 struct kvm_signal_mask __user *sigmask_arg = argp;
2093                 struct kvm_signal_mask kvm_sigmask;
2094                 sigset_t sigset, *p;
2095
2096                 p = NULL;
2097                 if (argp) {
2098                         r = -EFAULT;
2099                         if (copy_from_user(&kvm_sigmask, argp,
2100                                            sizeof kvm_sigmask))
2101                                 goto out;
2102                         r = -EINVAL;
2103                         if (kvm_sigmask.len != sizeof sigset)
2104                                 goto out;
2105                         r = -EFAULT;
2106                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2107                                            sizeof sigset))
2108                                 goto out;
2109                         p = &sigset;
2110                 }
2111                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2112                 break;
2113         }
2114         case KVM_GET_FPU: {
2115                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2116                 r = -ENOMEM;
2117                 if (!fpu)
2118                         goto out;
2119                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2120                 if (r)
2121                         goto out;
2122                 r = -EFAULT;
2123                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2124                         goto out;
2125                 r = 0;
2126                 break;
2127         }
2128         case KVM_SET_FPU: {
2129                 fpu = memdup_user(argp, sizeof(*fpu));
2130                 if (IS_ERR(fpu)) {
2131                         r = PTR_ERR(fpu);
2132                         fpu = NULL;
2133                         goto out;
2134                 }
2135                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2136                 break;
2137         }
2138         default:
2139                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2140         }
2141 out:
2142         vcpu_put(vcpu);
2143         kfree(fpu);
2144         kfree(kvm_sregs);
2145         return r;
2146 }
2147
2148 #ifdef CONFIG_COMPAT
2149 static long kvm_vcpu_compat_ioctl(struct file *filp,
2150                                   unsigned int ioctl, unsigned long arg)
2151 {
2152         struct kvm_vcpu *vcpu = filp->private_data;
2153         void __user *argp = compat_ptr(arg);
2154         int r;
2155
2156         if (vcpu->kvm->mm != current->mm)
2157                 return -EIO;
2158
2159         switch (ioctl) {
2160         case KVM_SET_SIGNAL_MASK: {
2161                 struct kvm_signal_mask __user *sigmask_arg = argp;
2162                 struct kvm_signal_mask kvm_sigmask;
2163                 compat_sigset_t csigset;
2164                 sigset_t sigset;
2165
2166                 if (argp) {
2167                         r = -EFAULT;
2168                         if (copy_from_user(&kvm_sigmask, argp,
2169                                            sizeof kvm_sigmask))
2170                                 goto out;
2171                         r = -EINVAL;
2172                         if (kvm_sigmask.len != sizeof csigset)
2173                                 goto out;
2174                         r = -EFAULT;
2175                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2176                                            sizeof csigset))
2177                                 goto out;
2178                         sigset_from_compat(&sigset, &csigset);
2179                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2180                 } else
2181                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2182                 break;
2183         }
2184         default:
2185                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2186         }
2187
2188 out:
2189         return r;
2190 }
2191 #endif
2192
2193 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2194                                  int (*accessor)(struct kvm_device *dev,
2195                                                  struct kvm_device_attr *attr),
2196                                  unsigned long arg)
2197 {
2198         struct kvm_device_attr attr;
2199
2200         if (!accessor)
2201                 return -EPERM;
2202
2203         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2204                 return -EFAULT;
2205
2206         return accessor(dev, &attr);
2207 }
2208
2209 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2210                              unsigned long arg)
2211 {
2212         struct kvm_device *dev = filp->private_data;
2213
2214         switch (ioctl) {
2215         case KVM_SET_DEVICE_ATTR:
2216                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2217         case KVM_GET_DEVICE_ATTR:
2218                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2219         case KVM_HAS_DEVICE_ATTR:
2220                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2221         default:
2222                 if (dev->ops->ioctl)
2223                         return dev->ops->ioctl(dev, ioctl, arg);
2224
2225                 return -ENOTTY;
2226         }
2227 }
2228
2229 static int kvm_device_release(struct inode *inode, struct file *filp)
2230 {
2231         struct kvm_device *dev = filp->private_data;
2232         struct kvm *kvm = dev->kvm;
2233
2234         kvm_put_kvm(kvm);
2235         return 0;
2236 }
2237
2238 static const struct file_operations kvm_device_fops = {
2239         .unlocked_ioctl = kvm_device_ioctl,
2240 #ifdef CONFIG_COMPAT
2241         .compat_ioctl = kvm_device_ioctl,
2242 #endif
2243         .release = kvm_device_release,
2244 };
2245
2246 struct kvm_device *kvm_device_from_filp(struct file *filp)
2247 {
2248         if (filp->f_op != &kvm_device_fops)
2249                 return NULL;
2250
2251         return filp->private_data;
2252 }
2253
2254 static int kvm_ioctl_create_device(struct kvm *kvm,
2255                                    struct kvm_create_device *cd)
2256 {
2257         struct kvm_device_ops *ops = NULL;
2258         struct kvm_device *dev;
2259         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2260         int ret;
2261
2262         switch (cd->type) {
2263 #ifdef CONFIG_KVM_MPIC
2264         case KVM_DEV_TYPE_FSL_MPIC_20:
2265         case KVM_DEV_TYPE_FSL_MPIC_42:
2266                 ops = &kvm_mpic_ops;
2267                 break;
2268 #endif
2269 #ifdef CONFIG_KVM_XICS
2270         case KVM_DEV_TYPE_XICS:
2271                 ops = &kvm_xics_ops;
2272                 break;
2273 #endif
2274 #ifdef CONFIG_KVM_VFIO
2275         case KVM_DEV_TYPE_VFIO:
2276                 ops = &kvm_vfio_ops;
2277                 break;
2278 #endif
2279 #ifdef CONFIG_KVM_ARM_VGIC
2280         case KVM_DEV_TYPE_ARM_VGIC_V2:
2281                 ops = &kvm_arm_vgic_v2_ops;
2282                 break;
2283 #endif
2284 #ifdef CONFIG_S390
2285         case KVM_DEV_TYPE_FLIC:
2286                 ops = &kvm_flic_ops;
2287                 break;
2288 #endif
2289         default:
2290                 return -ENODEV;
2291         }
2292
2293         if (test)
2294                 return 0;
2295
2296         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2297         if (!dev)
2298                 return -ENOMEM;
2299
2300         dev->ops = ops;
2301         dev->kvm = kvm;
2302
2303         ret = ops->create(dev, cd->type);
2304         if (ret < 0) {
2305                 kfree(dev);
2306                 return ret;
2307         }
2308
2309         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2310         if (ret < 0) {
2311                 ops->destroy(dev);
2312                 return ret;
2313         }
2314
2315         list_add(&dev->vm_node, &kvm->devices);
2316         kvm_get_kvm(kvm);
2317         cd->fd = ret;
2318         return 0;
2319 }
2320
2321 static long kvm_vm_ioctl(struct file *filp,
2322                            unsigned int ioctl, unsigned long arg)
2323 {
2324         struct kvm *kvm = filp->private_data;
2325         void __user *argp = (void __user *)arg;
2326         int r;
2327
2328         if (kvm->mm != current->mm)
2329                 return -EIO;
2330         switch (ioctl) {
2331         case KVM_CREATE_VCPU:
2332                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2333                 break;
2334         case KVM_SET_USER_MEMORY_REGION: {
2335                 struct kvm_userspace_memory_region kvm_userspace_mem;
2336
2337                 r = -EFAULT;
2338                 if (copy_from_user(&kvm_userspace_mem, argp,
2339                                                 sizeof kvm_userspace_mem))
2340                         goto out;
2341
2342                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2343                 break;
2344         }
2345         case KVM_GET_DIRTY_LOG: {
2346                 struct kvm_dirty_log log;
2347
2348                 r = -EFAULT;
2349                 if (copy_from_user(&log, argp, sizeof log))
2350                         goto out;
2351                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2352                 break;
2353         }
2354 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2355         case KVM_REGISTER_COALESCED_MMIO: {
2356                 struct kvm_coalesced_mmio_zone zone;
2357                 r = -EFAULT;
2358                 if (copy_from_user(&zone, argp, sizeof zone))
2359                         goto out;
2360                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2361                 break;
2362         }
2363         case KVM_UNREGISTER_COALESCED_MMIO: {
2364                 struct kvm_coalesced_mmio_zone zone;
2365                 r = -EFAULT;
2366                 if (copy_from_user(&zone, argp, sizeof zone))
2367                         goto out;
2368                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2369                 break;
2370         }
2371 #endif
2372         case KVM_IRQFD: {
2373                 struct kvm_irqfd data;
2374
2375                 r = -EFAULT;
2376                 if (copy_from_user(&data, argp, sizeof data))
2377                         goto out;
2378                 r = kvm_irqfd(kvm, &data);
2379                 break;
2380         }
2381         case KVM_IOEVENTFD: {
2382                 struct kvm_ioeventfd data;
2383
2384                 r = -EFAULT;
2385                 if (copy_from_user(&data, argp, sizeof data))
2386                         goto out;
2387                 r = kvm_ioeventfd(kvm, &data);
2388                 break;
2389         }
2390 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2391         case KVM_SET_BOOT_CPU_ID:
2392                 r = 0;
2393                 mutex_lock(&kvm->lock);
2394                 if (atomic_read(&kvm->online_vcpus) != 0)
2395                         r = -EBUSY;
2396                 else
2397                         kvm->bsp_vcpu_id = arg;
2398                 mutex_unlock(&kvm->lock);
2399                 break;
2400 #endif
2401 #ifdef CONFIG_HAVE_KVM_MSI
2402         case KVM_SIGNAL_MSI: {
2403                 struct kvm_msi msi;
2404
2405                 r = -EFAULT;
2406                 if (copy_from_user(&msi, argp, sizeof msi))
2407                         goto out;
2408                 r = kvm_send_userspace_msi(kvm, &msi);
2409                 break;
2410         }
2411 #endif
2412 #ifdef __KVM_HAVE_IRQ_LINE
2413         case KVM_IRQ_LINE_STATUS:
2414         case KVM_IRQ_LINE: {
2415                 struct kvm_irq_level irq_event;
2416
2417                 r = -EFAULT;
2418                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2419                         goto out;
2420
2421                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2422                                         ioctl == KVM_IRQ_LINE_STATUS);
2423                 if (r)
2424                         goto out;
2425
2426                 r = -EFAULT;
2427                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2428                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2429                                 goto out;
2430                 }
2431
2432                 r = 0;
2433                 break;
2434         }
2435 #endif
2436 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2437         case KVM_SET_GSI_ROUTING: {
2438                 struct kvm_irq_routing routing;
2439                 struct kvm_irq_routing __user *urouting;
2440                 struct kvm_irq_routing_entry *entries;
2441
2442                 r = -EFAULT;
2443                 if (copy_from_user(&routing, argp, sizeof(routing)))
2444                         goto out;
2445                 r = -EINVAL;
2446                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2447                         goto out;
2448                 if (routing.flags)
2449                         goto out;
2450                 r = -ENOMEM;
2451                 entries = vmalloc(routing.nr * sizeof(*entries));
2452                 if (!entries)
2453                         goto out;
2454                 r = -EFAULT;
2455                 urouting = argp;
2456                 if (copy_from_user(entries, urouting->entries,
2457                                    routing.nr * sizeof(*entries)))
2458                         goto out_free_irq_routing;
2459                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2460                                         routing.flags);
2461         out_free_irq_routing:
2462                 vfree(entries);
2463                 break;
2464         }
2465 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2466         case KVM_CREATE_DEVICE: {
2467                 struct kvm_create_device cd;
2468
2469                 r = -EFAULT;
2470                 if (copy_from_user(&cd, argp, sizeof(cd)))
2471                         goto out;
2472
2473                 r = kvm_ioctl_create_device(kvm, &cd);
2474                 if (r)
2475                         goto out;
2476
2477                 r = -EFAULT;
2478                 if (copy_to_user(argp, &cd, sizeof(cd)))
2479                         goto out;
2480
2481                 r = 0;
2482                 break;
2483         }
2484         default:
2485                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2486                 if (r == -ENOTTY)
2487                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2488         }
2489 out:
2490         return r;
2491 }
2492
2493 #ifdef CONFIG_COMPAT
2494 struct compat_kvm_dirty_log {
2495         __u32 slot;
2496         __u32 padding1;
2497         union {
2498                 compat_uptr_t dirty_bitmap; /* one bit per page */
2499                 __u64 padding2;
2500         };
2501 };
2502
2503 static long kvm_vm_compat_ioctl(struct file *filp,
2504                            unsigned int ioctl, unsigned long arg)
2505 {
2506         struct kvm *kvm = filp->private_data;
2507         int r;
2508
2509         if (kvm->mm != current->mm)
2510                 return -EIO;
2511         switch (ioctl) {
2512         case KVM_GET_DIRTY_LOG: {
2513                 struct compat_kvm_dirty_log compat_log;
2514                 struct kvm_dirty_log log;
2515
2516                 r = -EFAULT;
2517                 if (copy_from_user(&compat_log, (void __user *)arg,
2518                                    sizeof(compat_log)))
2519                         goto out;
2520                 log.slot         = compat_log.slot;
2521                 log.padding1     = compat_log.padding1;
2522                 log.padding2     = compat_log.padding2;
2523                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2524
2525                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2526                 break;
2527         }
2528         default:
2529                 r = kvm_vm_ioctl(filp, ioctl, arg);
2530         }
2531
2532 out:
2533         return r;
2534 }
2535 #endif
2536
2537 static struct file_operations kvm_vm_fops = {
2538         .release        = kvm_vm_release,
2539         .unlocked_ioctl = kvm_vm_ioctl,
2540 #ifdef CONFIG_COMPAT
2541         .compat_ioctl   = kvm_vm_compat_ioctl,
2542 #endif
2543         .llseek         = noop_llseek,
2544 };
2545
2546 static int kvm_dev_ioctl_create_vm(unsigned long type)
2547 {
2548         int r;
2549         struct kvm *kvm;
2550
2551         kvm = kvm_create_vm(type);
2552         if (IS_ERR(kvm))
2553                 return PTR_ERR(kvm);
2554 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2555         r = kvm_coalesced_mmio_init(kvm);
2556         if (r < 0) {
2557                 kvm_put_kvm(kvm);
2558                 return r;
2559         }
2560 #endif
2561         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2562         if (r < 0)
2563                 kvm_put_kvm(kvm);
2564
2565         return r;
2566 }
2567
2568 static long kvm_dev_ioctl_check_extension_generic(long arg)
2569 {
2570         switch (arg) {
2571         case KVM_CAP_USER_MEMORY:
2572         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2573         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2574 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2575         case KVM_CAP_SET_BOOT_CPU_ID:
2576 #endif
2577         case KVM_CAP_INTERNAL_ERROR_DATA:
2578 #ifdef CONFIG_HAVE_KVM_MSI
2579         case KVM_CAP_SIGNAL_MSI:
2580 #endif
2581 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2582         case KVM_CAP_IRQFD_RESAMPLE:
2583 #endif
2584                 return 1;
2585 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2586         case KVM_CAP_IRQ_ROUTING:
2587                 return KVM_MAX_IRQ_ROUTES;
2588 #endif
2589         default:
2590                 break;
2591         }
2592         return kvm_dev_ioctl_check_extension(arg);
2593 }
2594
2595 static long kvm_dev_ioctl(struct file *filp,
2596                           unsigned int ioctl, unsigned long arg)
2597 {
2598         long r = -EINVAL;
2599
2600         switch (ioctl) {
2601         case KVM_GET_API_VERSION:
2602                 r = -EINVAL;
2603                 if (arg)
2604                         goto out;
2605                 r = KVM_API_VERSION;
2606                 break;
2607         case KVM_CREATE_VM:
2608                 r = kvm_dev_ioctl_create_vm(arg);
2609                 break;
2610         case KVM_CHECK_EXTENSION:
2611                 r = kvm_dev_ioctl_check_extension_generic(arg);
2612                 break;
2613         case KVM_GET_VCPU_MMAP_SIZE:
2614                 r = -EINVAL;
2615                 if (arg)
2616                         goto out;
2617                 r = PAGE_SIZE;     /* struct kvm_run */
2618 #ifdef CONFIG_X86
2619                 r += PAGE_SIZE;    /* pio data page */
2620 #endif
2621 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2622                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2623 #endif
2624                 break;
2625         case KVM_TRACE_ENABLE:
2626         case KVM_TRACE_PAUSE:
2627         case KVM_TRACE_DISABLE:
2628                 r = -EOPNOTSUPP;
2629                 break;
2630         default:
2631                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2632         }
2633 out:
2634         return r;
2635 }
2636
2637 static struct file_operations kvm_chardev_ops = {
2638         .unlocked_ioctl = kvm_dev_ioctl,
2639         .compat_ioctl   = kvm_dev_ioctl,
2640         .llseek         = noop_llseek,
2641 };
2642
2643 static struct miscdevice kvm_dev = {
2644         KVM_MINOR,
2645         "kvm",
2646         &kvm_chardev_ops,
2647 };
2648
2649 static void hardware_enable_nolock(void *junk)
2650 {
2651         int cpu = raw_smp_processor_id();
2652         int r;
2653
2654         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2655                 return;
2656
2657         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2658
2659         r = kvm_arch_hardware_enable(NULL);
2660
2661         if (r) {
2662                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2663                 atomic_inc(&hardware_enable_failed);
2664                 printk(KERN_INFO "kvm: enabling virtualization on "
2665                                  "CPU%d failed\n", cpu);
2666         }
2667 }
2668
2669 static void hardware_enable(void)
2670 {
2671         raw_spin_lock(&kvm_count_lock);
2672         if (kvm_usage_count)
2673                 hardware_enable_nolock(NULL);
2674         raw_spin_unlock(&kvm_count_lock);
2675 }
2676
2677 static void hardware_disable_nolock(void *junk)
2678 {
2679         int cpu = raw_smp_processor_id();
2680
2681         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2682                 return;
2683         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2684         kvm_arch_hardware_disable(NULL);
2685 }
2686
2687 static void hardware_disable(void)
2688 {
2689         raw_spin_lock(&kvm_count_lock);
2690         if (kvm_usage_count)
2691                 hardware_disable_nolock(NULL);
2692         raw_spin_unlock(&kvm_count_lock);
2693 }
2694
2695 static void hardware_disable_all_nolock(void)
2696 {
2697         BUG_ON(!kvm_usage_count);
2698
2699         kvm_usage_count--;
2700         if (!kvm_usage_count)
2701                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2702 }
2703
2704 static void hardware_disable_all(void)
2705 {
2706         raw_spin_lock(&kvm_count_lock);
2707         hardware_disable_all_nolock();
2708         raw_spin_unlock(&kvm_count_lock);
2709 }
2710
2711 static int hardware_enable_all(void)
2712 {
2713         int r = 0;
2714
2715         raw_spin_lock(&kvm_count_lock);
2716
2717         kvm_usage_count++;
2718         if (kvm_usage_count == 1) {
2719                 atomic_set(&hardware_enable_failed, 0);
2720                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2721
2722                 if (atomic_read(&hardware_enable_failed)) {
2723                         hardware_disable_all_nolock();
2724                         r = -EBUSY;
2725                 }
2726         }
2727
2728         raw_spin_unlock(&kvm_count_lock);
2729
2730         return r;
2731 }
2732
2733 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2734                            void *v)
2735 {
2736         int cpu = (long)v;
2737
2738         val &= ~CPU_TASKS_FROZEN;
2739         switch (val) {
2740         case CPU_DYING:
2741                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2742                        cpu);
2743                 hardware_disable();
2744                 break;
2745         case CPU_STARTING:
2746                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2747                        cpu);
2748                 hardware_enable();
2749                 break;
2750         }
2751         return NOTIFY_OK;
2752 }
2753
2754 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2755                       void *v)
2756 {
2757         /*
2758          * Some (well, at least mine) BIOSes hang on reboot if
2759          * in vmx root mode.
2760          *
2761          * And Intel TXT required VMX off for all cpu when system shutdown.
2762          */
2763         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2764         kvm_rebooting = true;
2765         on_each_cpu(hardware_disable_nolock, NULL, 1);
2766         return NOTIFY_OK;
2767 }
2768
2769 static struct notifier_block kvm_reboot_notifier = {
2770         .notifier_call = kvm_reboot,
2771         .priority = 0,
2772 };
2773
2774 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2775 {
2776         int i;
2777
2778         for (i = 0; i < bus->dev_count; i++) {
2779                 struct kvm_io_device *pos = bus->range[i].dev;
2780
2781                 kvm_iodevice_destructor(pos);
2782         }
2783         kfree(bus);
2784 }
2785
2786 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2787                                  const struct kvm_io_range *r2)
2788 {
2789         if (r1->addr < r2->addr)
2790                 return -1;
2791         if (r1->addr + r1->len > r2->addr + r2->len)
2792                 return 1;
2793         return 0;
2794 }
2795
2796 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2797 {
2798         return kvm_io_bus_cmp(p1, p2);
2799 }
2800
2801 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2802                           gpa_t addr, int len)
2803 {
2804         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2805                 .addr = addr,
2806                 .len = len,
2807                 .dev = dev,
2808         };
2809
2810         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2811                 kvm_io_bus_sort_cmp, NULL);
2812
2813         return 0;
2814 }
2815
2816 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2817                              gpa_t addr, int len)
2818 {
2819         struct kvm_io_range *range, key;
2820         int off;
2821
2822         key = (struct kvm_io_range) {
2823                 .addr = addr,
2824                 .len = len,
2825         };
2826
2827         range = bsearch(&key, bus->range, bus->dev_count,
2828                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2829         if (range == NULL)
2830                 return -ENOENT;
2831
2832         off = range - bus->range;
2833
2834         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2835                 off--;
2836
2837         return off;
2838 }
2839
2840 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2841                               struct kvm_io_range *range, const void *val)
2842 {
2843         int idx;
2844
2845         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2846         if (idx < 0)
2847                 return -EOPNOTSUPP;
2848
2849         while (idx < bus->dev_count &&
2850                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2851                 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2852                                         range->len, val))
2853                         return idx;
2854                 idx++;
2855         }
2856
2857         return -EOPNOTSUPP;
2858 }
2859
2860 /* kvm_io_bus_write - called under kvm->slots_lock */
2861 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2862                      int len, const void *val)
2863 {
2864         struct kvm_io_bus *bus;
2865         struct kvm_io_range range;
2866         int r;
2867
2868         range = (struct kvm_io_range) {
2869                 .addr = addr,
2870                 .len = len,
2871         };
2872
2873         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2874         r = __kvm_io_bus_write(bus, &range, val);
2875         return r < 0 ? r : 0;
2876 }
2877
2878 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2879 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2880                             int len, const void *val, long cookie)
2881 {
2882         struct kvm_io_bus *bus;
2883         struct kvm_io_range range;
2884
2885         range = (struct kvm_io_range) {
2886                 .addr = addr,
2887                 .len = len,
2888         };
2889
2890         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2891
2892         /* First try the device referenced by cookie. */
2893         if ((cookie >= 0) && (cookie < bus->dev_count) &&
2894             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2895                 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2896                                         val))
2897                         return cookie;
2898
2899         /*
2900          * cookie contained garbage; fall back to search and return the
2901          * correct cookie value.
2902          */
2903         return __kvm_io_bus_write(bus, &range, val);
2904 }
2905
2906 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2907                              void *val)
2908 {
2909         int idx;
2910
2911         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2912         if (idx < 0)
2913                 return -EOPNOTSUPP;
2914
2915         while (idx < bus->dev_count &&
2916                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2917                 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2918                                        range->len, val))
2919                         return idx;
2920                 idx++;
2921         }
2922
2923         return -EOPNOTSUPP;
2924 }
2925
2926 /* kvm_io_bus_read - called under kvm->slots_lock */
2927 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2928                     int len, void *val)
2929 {
2930         struct kvm_io_bus *bus;
2931         struct kvm_io_range range;
2932         int r;
2933
2934         range = (struct kvm_io_range) {
2935                 .addr = addr,
2936                 .len = len,
2937         };
2938
2939         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2940         r = __kvm_io_bus_read(bus, &range, val);
2941         return r < 0 ? r : 0;
2942 }
2943
2944
2945 /* Caller must hold slots_lock. */
2946 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2947                             int len, struct kvm_io_device *dev)
2948 {
2949         struct kvm_io_bus *new_bus, *bus;
2950
2951         bus = kvm->buses[bus_idx];
2952         /* exclude ioeventfd which is limited by maximum fd */
2953         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2954                 return -ENOSPC;
2955
2956         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2957                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2958         if (!new_bus)
2959                 return -ENOMEM;
2960         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2961                sizeof(struct kvm_io_range)));
2962         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2963         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2964         synchronize_srcu_expedited(&kvm->srcu);
2965         kfree(bus);
2966
2967         return 0;
2968 }
2969
2970 /* Caller must hold slots_lock. */
2971 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2972                               struct kvm_io_device *dev)
2973 {
2974         int i, r;
2975         struct kvm_io_bus *new_bus, *bus;
2976
2977         bus = kvm->buses[bus_idx];
2978         r = -ENOENT;
2979         for (i = 0; i < bus->dev_count; i++)
2980                 if (bus->range[i].dev == dev) {
2981                         r = 0;
2982                         break;
2983                 }
2984
2985         if (r)
2986                 return r;
2987
2988         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2989                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2990         if (!new_bus)
2991                 return -ENOMEM;
2992
2993         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2994         new_bus->dev_count--;
2995         memcpy(new_bus->range + i, bus->range + i + 1,
2996                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2997
2998         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2999         synchronize_srcu_expedited(&kvm->srcu);
3000         kfree(bus);
3001         return r;
3002 }
3003
3004 static struct notifier_block kvm_cpu_notifier = {
3005         .notifier_call = kvm_cpu_hotplug,
3006 };
3007
3008 static int vm_stat_get(void *_offset, u64 *val)
3009 {
3010         unsigned offset = (long)_offset;
3011         struct kvm *kvm;
3012
3013         *val = 0;
3014         spin_lock(&kvm_lock);
3015         list_for_each_entry(kvm, &vm_list, vm_list)
3016                 *val += *(u32 *)((void *)kvm + offset);
3017         spin_unlock(&kvm_lock);
3018         return 0;
3019 }
3020
3021 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3022
3023 static int vcpu_stat_get(void *_offset, u64 *val)
3024 {
3025         unsigned offset = (long)_offset;
3026         struct kvm *kvm;
3027         struct kvm_vcpu *vcpu;
3028         int i;
3029
3030         *val = 0;
3031         spin_lock(&kvm_lock);
3032         list_for_each_entry(kvm, &vm_list, vm_list)
3033                 kvm_for_each_vcpu(i, vcpu, kvm)
3034                         *val += *(u32 *)((void *)vcpu + offset);
3035
3036         spin_unlock(&kvm_lock);
3037         return 0;
3038 }
3039
3040 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3041
3042 static const struct file_operations *stat_fops[] = {
3043         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3044         [KVM_STAT_VM]   = &vm_stat_fops,
3045 };
3046
3047 static int kvm_init_debug(void)
3048 {
3049         int r = -EEXIST;
3050         struct kvm_stats_debugfs_item *p;
3051
3052         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3053         if (kvm_debugfs_dir == NULL)
3054                 goto out;
3055
3056         for (p = debugfs_entries; p->name; ++p) {
3057                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3058                                                 (void *)(long)p->offset,
3059                                                 stat_fops[p->kind]);
3060                 if (p->dentry == NULL)
3061                         goto out_dir;
3062         }
3063
3064         return 0;
3065
3066 out_dir:
3067         debugfs_remove_recursive(kvm_debugfs_dir);
3068 out:
3069         return r;
3070 }
3071
3072 static void kvm_exit_debug(void)
3073 {
3074         struct kvm_stats_debugfs_item *p;
3075
3076         for (p = debugfs_entries; p->name; ++p)
3077                 debugfs_remove(p->dentry);
3078         debugfs_remove(kvm_debugfs_dir);
3079 }
3080
3081 static int kvm_suspend(void)
3082 {
3083         if (kvm_usage_count)
3084                 hardware_disable_nolock(NULL);
3085         return 0;
3086 }
3087
3088 static void kvm_resume(void)
3089 {
3090         if (kvm_usage_count) {
3091                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3092                 hardware_enable_nolock(NULL);
3093         }
3094 }
3095
3096 static struct syscore_ops kvm_syscore_ops = {
3097         .suspend = kvm_suspend,
3098         .resume = kvm_resume,
3099 };
3100
3101 static inline
3102 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3103 {
3104         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3105 }
3106
3107 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3108 {
3109         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3110         if (vcpu->preempted)
3111                 vcpu->preempted = false;
3112
3113         kvm_arch_vcpu_load(vcpu, cpu);
3114 }
3115
3116 static void kvm_sched_out(struct preempt_notifier *pn,
3117                           struct task_struct *next)
3118 {
3119         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3120
3121         if (current->state == TASK_RUNNING)
3122                 vcpu->preempted = true;
3123         kvm_arch_vcpu_put(vcpu);
3124 }
3125
3126 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3127                   struct module *module)
3128 {
3129         int r;
3130         int cpu;
3131
3132         r = kvm_arch_init(opaque);
3133         if (r)
3134                 goto out_fail;
3135
3136         /*
3137          * kvm_arch_init makes sure there's at most one caller
3138          * for architectures that support multiple implementations,
3139          * like intel and amd on x86.
3140          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3141          * conflicts in case kvm is already setup for another implementation.
3142          */
3143         r = kvm_irqfd_init();
3144         if (r)
3145                 goto out_irqfd;
3146
3147         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3148                 r = -ENOMEM;
3149                 goto out_free_0;
3150         }
3151
3152         r = kvm_arch_hardware_setup();
3153         if (r < 0)
3154                 goto out_free_0a;
3155
3156         for_each_online_cpu(cpu) {
3157                 smp_call_function_single(cpu,
3158                                 kvm_arch_check_processor_compat,
3159                                 &r, 1);
3160                 if (r < 0)
3161                         goto out_free_1;
3162         }
3163
3164         r = register_cpu_notifier(&kvm_cpu_notifier);
3165         if (r)
3166                 goto out_free_2;
3167         register_reboot_notifier(&kvm_reboot_notifier);
3168
3169         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3170         if (!vcpu_align)
3171                 vcpu_align = __alignof__(struct kvm_vcpu);
3172         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3173                                            0, NULL);
3174         if (!kvm_vcpu_cache) {
3175                 r = -ENOMEM;
3176                 goto out_free_3;
3177         }
3178
3179         r = kvm_async_pf_init();
3180         if (r)
3181                 goto out_free;
3182
3183         kvm_chardev_ops.owner = module;
3184         kvm_vm_fops.owner = module;
3185         kvm_vcpu_fops.owner = module;
3186
3187         r = misc_register(&kvm_dev);
3188         if (r) {
3189                 printk(KERN_ERR "kvm: misc device register failed\n");
3190                 goto out_unreg;
3191         }
3192
3193         register_syscore_ops(&kvm_syscore_ops);
3194
3195         kvm_preempt_ops.sched_in = kvm_sched_in;
3196         kvm_preempt_ops.sched_out = kvm_sched_out;
3197
3198         r = kvm_init_debug();
3199         if (r) {
3200                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3201                 goto out_undebugfs;
3202         }
3203
3204         return 0;
3205
3206 out_undebugfs:
3207         unregister_syscore_ops(&kvm_syscore_ops);
3208         misc_deregister(&kvm_dev);
3209 out_unreg:
3210         kvm_async_pf_deinit();
3211 out_free:
3212         kmem_cache_destroy(kvm_vcpu_cache);
3213 out_free_3:
3214         unregister_reboot_notifier(&kvm_reboot_notifier);
3215         unregister_cpu_notifier(&kvm_cpu_notifier);
3216 out_free_2:
3217 out_free_1:
3218         kvm_arch_hardware_unsetup();
3219 out_free_0a:
3220         free_cpumask_var(cpus_hardware_enabled);
3221 out_free_0:
3222         kvm_irqfd_exit();
3223 out_irqfd:
3224         kvm_arch_exit();
3225 out_fail:
3226         return r;
3227 }
3228 EXPORT_SYMBOL_GPL(kvm_init);
3229
3230 void kvm_exit(void)
3231 {
3232         kvm_exit_debug();
3233         misc_deregister(&kvm_dev);
3234         kmem_cache_destroy(kvm_vcpu_cache);
3235         kvm_async_pf_deinit();
3236         unregister_syscore_ops(&kvm_syscore_ops);
3237         unregister_reboot_notifier(&kvm_reboot_notifier);
3238         unregister_cpu_notifier(&kvm_cpu_notifier);
3239         on_each_cpu(hardware_disable_nolock, NULL, 1);
3240         kvm_arch_hardware_unsetup();
3241         kvm_arch_exit();
3242         kvm_irqfd_exit();
3243         free_cpumask_var(cpus_hardware_enabled);
3244 }
3245 EXPORT_SYMBOL_GPL(kvm_exit);