virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/syscore_ops.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/pgtable.h>
  57
  58 #include "coalesced_mmio.h"
  59 #include "async_pf.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/kvm.h>
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 /*
  68  * Ordering of locks:
  69  *
  70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  71  */
  72
  73 DEFINE_RAW_SPINLOCK(kvm_lock);
  74 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
  99 bool kvm_rebooting;
 100 EXPORT_SYMBOL_GPL(kvm_rebooting);
 101
 102 static bool largepages_enabled = true;
 103
 104 bool kvm_is_mmio_pfn(pfn_t pfn)
 105 {
 106         if (pfn_valid(pfn))
 107                 return PageReserved(pfn_to_page(pfn));
 108
 109         return true;
 110 }
 111
 112 /*
 113  * Switches to specified vcpu, until a matching vcpu_put()
 114  */
 115 int vcpu_load(struct kvm_vcpu *vcpu)
 116 {
 117         int cpu;
 118
 119         if (mutex_lock_killable(&vcpu->mutex))
 120                 return -EINTR;
 121         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 122                 /* The thread running this VCPU changed. */
 123                 struct pid *oldpid = vcpu->pid;
 124                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 125                 rcu_assign_pointer(vcpu->pid, newpid);
 126                 synchronize_rcu();
 127                 put_pid(oldpid);
 128         }
 129         cpu = get_cpu();
 130         preempt_notifier_register(&vcpu->preempt_notifier);
 131         kvm_arch_vcpu_load(vcpu, cpu);
 132         put_cpu();
 133         return 0;
 134 }
 135
 136 void vcpu_put(struct kvm_vcpu *vcpu)
 137 {
 138         preempt_disable();
 139         kvm_arch_vcpu_put(vcpu);
 140         preempt_notifier_unregister(&vcpu->preempt_notifier);
 141         preempt_enable();
 142         mutex_unlock(&vcpu->mutex);
 143 }
 144
 145 static void ack_flush(void *_completed)
 146 {
 147 }
 148
 149 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 150 {
 151         int i, cpu, me;
 152         cpumask_var_t cpus;
 153         bool called = true;
 154         struct kvm_vcpu *vcpu;
 155
 156         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 157
 158         me = get_cpu();
 159         kvm_for_each_vcpu(i, vcpu, kvm) {
 160                 kvm_make_request(req, vcpu);
 161                 cpu = vcpu->cpu;
 162
 163                 /* Set ->requests bit before we read ->mode */
 164                 smp_mb();
 165
 166                 if (cpus != NULL && cpu != -1 && cpu != me &&
 167                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 168                         cpumask_set_cpu(cpu, cpus);
 169         }
 170         if (unlikely(cpus == NULL))
 171                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 172         else if (!cpumask_empty(cpus))
 173                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 174         else
 175                 called = false;
 176         put_cpu();
 177         free_cpumask_var(cpus);
 178         return called;
 179 }
 180
 181 void kvm_flush_remote_tlbs(struct kvm *kvm)
 182 {
 183         long dirty_count = kvm->tlbs_dirty;
 184
 185         smp_mb();
 186         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 187                 ++kvm->stat.remote_tlb_flush;
 188         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 189 }
 190
 191 void kvm_reload_remote_mmus(struct kvm *kvm)
 192 {
 193         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 194 }
 195
 196 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 197 {
 198         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 199 }
 200
 201 void kvm_make_scan_ioapic_request(struct kvm *kvm)
 202 {
 203         make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
 204 }
 205
 206 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 207 {
 208         struct page *page;
 209         int r;
 210
 211         mutex_init(&vcpu->mutex);
 212         vcpu->cpu = -1;
 213         vcpu->kvm = kvm;
 214         vcpu->vcpu_id = id;
 215         vcpu->pid = NULL;
 216         init_waitqueue_head(&vcpu->wq);
 217         kvm_async_pf_vcpu_init(vcpu);
 218
 219         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 220         if (!page) {
 221                 r = -ENOMEM;
 222                 goto fail;
 223         }
 224         vcpu->run = page_address(page);
 225
 226         kvm_vcpu_set_in_spin_loop(vcpu, false);
 227         kvm_vcpu_set_dy_eligible(vcpu, false);
 228         vcpu->preempted = false;
 229
 230         r = kvm_arch_vcpu_init(vcpu);
 231         if (r < 0)
 232                 goto fail_free_run;
 233         return 0;
 234
 235 fail_free_run:
 236         free_page((unsigned long)vcpu->run);
 237 fail:
 238         return r;
 239 }
 240 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 241
 242 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 243 {
 244         put_pid(vcpu->pid);
 245         kvm_arch_vcpu_uninit(vcpu);
 246         free_page((unsigned long)vcpu->run);
 247 }
 248 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 249
 250 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 251 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 252 {
 253         return container_of(mn, struct kvm, mmu_notifier);
 254 }
 255
 256 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 257                                              struct mm_struct *mm,
 258                                              unsigned long address)
 259 {
 260         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 261         int need_tlb_flush, idx;
 262
 263         /*
 264          * When ->invalidate_page runs, the linux pte has been zapped
 265          * already but the page is still allocated until
 266          * ->invalidate_page returns. So if we increase the sequence
 267          * here the kvm page fault will notice if the spte can't be
 268          * established because the page is going to be freed. If
 269          * instead the kvm page fault establishes the spte before
 270          * ->invalidate_page runs, kvm_unmap_hva will release it
 271          * before returning.
 272          *
 273          * The sequence increase only need to be seen at spin_unlock
 274          * time, and not at spin_lock time.
 275          *
 276          * Increasing the sequence after the spin_unlock would be
 277          * unsafe because the kvm page fault could then establish the
 278          * pte after kvm_unmap_hva returned, without noticing the page
 279          * is going to be freed.
 280          */
 281         idx = srcu_read_lock(&kvm->srcu);
 282         spin_lock(&kvm->mmu_lock);
 283
 284         kvm->mmu_notifier_seq++;
 285         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 286         /* we've to flush the tlb before the pages can be freed */
 287         if (need_tlb_flush)
 288                 kvm_flush_remote_tlbs(kvm);
 289
 290         spin_unlock(&kvm->mmu_lock);
 291         srcu_read_unlock(&kvm->srcu, idx);
 292 }
 293
 294 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 295                                         struct mm_struct *mm,
 296                                         unsigned long address,
 297                                         pte_t pte)
 298 {
 299         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 300         int idx;
 301
 302         idx = srcu_read_lock(&kvm->srcu);
 303         spin_lock(&kvm->mmu_lock);
 304         kvm->mmu_notifier_seq++;
 305         kvm_set_spte_hva(kvm, address, pte);
 306         spin_unlock(&kvm->mmu_lock);
 307         srcu_read_unlock(&kvm->srcu, idx);
 308 }
 309
 310 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 311                                                     struct mm_struct *mm,
 312                                                     unsigned long start,
 313                                                     unsigned long end)
 314 {
 315         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 316         int need_tlb_flush = 0, idx;
 317
 318         idx = srcu_read_lock(&kvm->srcu);
 319         spin_lock(&kvm->mmu_lock);
 320         /*
 321          * The count increase must become visible at unlock time as no
 322          * spte can be established without taking the mmu_lock and
 323          * count is also read inside the mmu_lock critical section.
 324          */
 325         kvm->mmu_notifier_count++;
 326         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 327         need_tlb_flush |= kvm->tlbs_dirty;
 328         /* we've to flush the tlb before the pages can be freed */
 329         if (need_tlb_flush)
 330                 kvm_flush_remote_tlbs(kvm);
 331
 332         spin_unlock(&kvm->mmu_lock);
 333         srcu_read_unlock(&kvm->srcu, idx);
 334 }
 335
 336 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 337                                                   struct mm_struct *mm,
 338                                                   unsigned long start,
 339                                                   unsigned long end)
 340 {
 341         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 342
 343         spin_lock(&kvm->mmu_lock);
 344         /*
 345          * This sequence increase will notify the kvm page fault that
 346          * the page that is going to be mapped in the spte could have
 347          * been freed.
 348          */
 349         kvm->mmu_notifier_seq++;
 350         smp_wmb();
 351         /*
 352          * The above sequence increase must be visible before the
 353          * below count decrease, which is ensured by the smp_wmb above
 354          * in conjunction with the smp_rmb in mmu_notifier_retry().
 355          */
 356         kvm->mmu_notifier_count--;
 357         spin_unlock(&kvm->mmu_lock);
 358
 359         BUG_ON(kvm->mmu_notifier_count < 0);
 360 }
 361
 362 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 363                                               struct mm_struct *mm,
 364                                               unsigned long address)
 365 {
 366         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 367         int young, idx;
 368
 369         idx = srcu_read_lock(&kvm->srcu);
 370         spin_lock(&kvm->mmu_lock);
 371
 372         young = kvm_age_hva(kvm, address);
 373         if (young)
 374                 kvm_flush_remote_tlbs(kvm);
 375
 376         spin_unlock(&kvm->mmu_lock);
 377         srcu_read_unlock(&kvm->srcu, idx);
 378
 379         return young;
 380 }
 381
 382 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 383                                        struct mm_struct *mm,
 384                                        unsigned long address)
 385 {
 386         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 387         int young, idx;
 388
 389         idx = srcu_read_lock(&kvm->srcu);
 390         spin_lock(&kvm->mmu_lock);
 391         young = kvm_test_age_hva(kvm, address);
 392         spin_unlock(&kvm->mmu_lock);
 393         srcu_read_unlock(&kvm->srcu, idx);
 394
 395         return young;
 396 }
 397
 398 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 399                                      struct mm_struct *mm)
 400 {
 401         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 402         int idx;
 403
 404         idx = srcu_read_lock(&kvm->srcu);
 405         kvm_arch_flush_shadow_all(kvm);
 406         srcu_read_unlock(&kvm->srcu, idx);
 407 }
 408
 409 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 410         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 411         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 412         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 413         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 414         .test_young             = kvm_mmu_notifier_test_young,
 415         .change_pte             = kvm_mmu_notifier_change_pte,
 416         .release                = kvm_mmu_notifier_release,
 417 };
 418
 419 static int kvm_init_mmu_notifier(struct kvm *kvm)
 420 {
 421         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 422         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 423 }
 424
 425 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 426
 427 static int kvm_init_mmu_notifier(struct kvm *kvm)
 428 {
 429         return 0;
 430 }
 431
 432 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 433
 434 static void kvm_init_memslots_id(struct kvm *kvm)
 435 {
 436         int i;
 437         struct kvm_memslots *slots = kvm->memslots;
 438
 439         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 440                 slots->id_to_index[i] = slots->memslots[i].id = i;
 441 }
 442
 443 static struct kvm *kvm_create_vm(unsigned long type)
 444 {
 445         int r, i;
 446         struct kvm *kvm = kvm_arch_alloc_vm();
 447
 448         if (!kvm)
 449                 return ERR_PTR(-ENOMEM);
 450
 451         r = kvm_arch_init_vm(kvm, type);
 452         if (r)
 453                 goto out_err_nodisable;
 454
 455         r = hardware_enable_all();
 456         if (r)
 457                 goto out_err_nodisable;
 458
 459 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 460         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 461         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 462 #endif
 463
 464         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 465
 466         r = -ENOMEM;
 467         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 468         if (!kvm->memslots)
 469                 goto out_err_nosrcu;
 470         kvm_init_memslots_id(kvm);
 471         if (init_srcu_struct(&kvm->srcu))
 472                 goto out_err_nosrcu;
 473         for (i = 0; i < KVM_NR_BUSES; i++) {
 474                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 475                                         GFP_KERNEL);
 476                 if (!kvm->buses[i])
 477                         goto out_err;
 478         }
 479
 480         spin_lock_init(&kvm->mmu_lock);
 481         kvm->mm = current->mm;
 482         atomic_inc(&kvm->mm->mm_count);
 483         kvm_eventfd_init(kvm);
 484         mutex_init(&kvm->lock);
 485         mutex_init(&kvm->irq_lock);
 486         mutex_init(&kvm->slots_lock);
 487         atomic_set(&kvm->users_count, 1);
 488         INIT_LIST_HEAD(&kvm->devices);
 489
 490         r = kvm_init_mmu_notifier(kvm);
 491         if (r)
 492                 goto out_err;
 493
 494         raw_spin_lock(&kvm_lock);
 495         list_add(&kvm->vm_list, &vm_list);
 496         raw_spin_unlock(&kvm_lock);
 497
 498         return kvm;
 499
 500 out_err:
 501         cleanup_srcu_struct(&kvm->srcu);
 502 out_err_nosrcu:
 503         hardware_disable_all();
 504 out_err_nodisable:
 505         for (i = 0; i < KVM_NR_BUSES; i++)
 506                 kfree(kvm->buses[i]);
 507         kfree(kvm->memslots);
 508         kvm_arch_free_vm(kvm);
 509         return ERR_PTR(r);
 510 }
 511
 512 /*
 513  * Avoid using vmalloc for a small buffer.
 514  * Should not be used when the size is statically known.
 515  */
 516 void *kvm_kvzalloc(unsigned long size)
 517 {
 518         if (size > PAGE_SIZE)
 519                 return vzalloc(size);
 520         else
 521                 return kzalloc(size, GFP_KERNEL);
 522 }
 523
 524 void kvm_kvfree(const void *addr)
 525 {
 526         if (is_vmalloc_addr(addr))
 527                 vfree(addr);
 528         else
 529                 kfree(addr);
 530 }
 531
 532 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 533 {
 534         if (!memslot->dirty_bitmap)
 535                 return;
 536
 537         kvm_kvfree(memslot->dirty_bitmap);
 538         memslot->dirty_bitmap = NULL;
 539 }
 540
 541 /*
 542  * Free any memory in @free but not in @dont.
 543  */
 544 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 545                                   struct kvm_memory_slot *dont)
 546 {
 547         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 548                 kvm_destroy_dirty_bitmap(free);
 549
 550         kvm_arch_free_memslot(free, dont);
 551
 552         free->npages = 0;
 553 }
 554
 555 void kvm_free_physmem(struct kvm *kvm)
 556 {
 557         struct kvm_memslots *slots = kvm->memslots;
 558         struct kvm_memory_slot *memslot;
 559
 560         kvm_for_each_memslot(memslot, slots)
 561                 kvm_free_physmem_slot(memslot, NULL);
 562
 563         kfree(kvm->memslots);
 564 }
 565
 566 static void kvm_destroy_devices(struct kvm *kvm)
 567 {
 568         struct list_head *node, *tmp;
 569
 570         list_for_each_safe(node, tmp, &kvm->devices) {
 571                 struct kvm_device *dev =
 572                         list_entry(node, struct kvm_device, vm_node);
 573
 574                 list_del(node);
 575                 dev->ops->destroy(dev);
 576         }
 577 }
 578
 579 static void kvm_destroy_vm(struct kvm *kvm)
 580 {
 581         int i;
 582         struct mm_struct *mm = kvm->mm;
 583
 584         kvm_arch_sync_events(kvm);
 585         raw_spin_lock(&kvm_lock);
 586         list_del(&kvm->vm_list);
 587         raw_spin_unlock(&kvm_lock);
 588         kvm_free_irq_routing(kvm);
 589         for (i = 0; i < KVM_NR_BUSES; i++)
 590                 kvm_io_bus_destroy(kvm->buses[i]);
 591         kvm_coalesced_mmio_free(kvm);
 592 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 593         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 594 #else
 595         kvm_arch_flush_shadow_all(kvm);
 596 #endif
 597         kvm_arch_destroy_vm(kvm);
 598         kvm_destroy_devices(kvm);
 599         kvm_free_physmem(kvm);
 600         cleanup_srcu_struct(&kvm->srcu);
 601         kvm_arch_free_vm(kvm);
 602         hardware_disable_all();
 603         mmdrop(mm);
 604 }
 605
 606 void kvm_get_kvm(struct kvm *kvm)
 607 {
 608         atomic_inc(&kvm->users_count);
 609 }
 610 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 611
 612 void kvm_put_kvm(struct kvm *kvm)
 613 {
 614         if (atomic_dec_and_test(&kvm->users_count))
 615                 kvm_destroy_vm(kvm);
 616 }
 617 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 618
 619
 620 static int kvm_vm_release(struct inode *inode, struct file *filp)
 621 {
 622         struct kvm *kvm = filp->private_data;
 623
 624         kvm_irqfd_release(kvm);
 625
 626         kvm_put_kvm(kvm);
 627         return 0;
 628 }
 629
 630 /*
 631  * Allocation size is twice as large as the actual dirty bitmap size.
 632  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 633  */
 634 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 635 {
 636 #ifndef CONFIG_S390
 637         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 638
 639         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 640         if (!memslot->dirty_bitmap)
 641                 return -ENOMEM;
 642
 643 #endif /* !CONFIG_S390 */
 644         return 0;
 645 }
 646
 647 static int cmp_memslot(const void *slot1, const void *slot2)
 648 {
 649         struct kvm_memory_slot *s1, *s2;
 650
 651         s1 = (struct kvm_memory_slot *)slot1;
 652         s2 = (struct kvm_memory_slot *)slot2;
 653
 654         if (s1->npages < s2->npages)
 655                 return 1;
 656         if (s1->npages > s2->npages)
 657                 return -1;
 658
 659         return 0;
 660 }
 661
 662 /*
 663  * Sort the memslots base on its size, so the larger slots
 664  * will get better fit.
 665  */
 666 static void sort_memslots(struct kvm_memslots *slots)
 667 {
 668         int i;
 669
 670         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 671               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 672
 673         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 674                 slots->id_to_index[slots->memslots[i].id] = i;
 675 }
 676
 677 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
 678                      u64 last_generation)
 679 {
 680         if (new) {
 681                 int id = new->id;
 682                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 683                 unsigned long npages = old->npages;
 684
 685                 *old = *new;
 686                 if (new->npages != npages)
 687                         sort_memslots(slots);
 688         }
 689
 690         slots->generation = last_generation + 1;
 691 }
 692
 693 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 694 {
 695         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 696
 697 #ifdef KVM_CAP_READONLY_MEM
 698         valid_flags |= KVM_MEM_READONLY;
 699 #endif
 700
 701         if (mem->flags & ~valid_flags)
 702                 return -EINVAL;
 703
 704         return 0;
 705 }
 706
 707 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 708                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 709 {
 710         struct kvm_memslots *old_memslots = kvm->memslots;
 711
 712         update_memslots(slots, new, kvm->memslots->generation);
 713         rcu_assign_pointer(kvm->memslots, slots);
 714         synchronize_srcu_expedited(&kvm->srcu);
 715
 716         kvm_arch_memslots_updated(kvm);
 717
 718         return old_memslots;
 719 }
 720
 721 /*
 722  * Allocate some memory and give it an address in the guest physical address
 723  * space.
 724  *
 725  * Discontiguous memory is allowed, mostly for framebuffers.
 726  *
 727  * Must be called holding mmap_sem for write.
 728  */
 729 int __kvm_set_memory_region(struct kvm *kvm,
 730                             struct kvm_userspace_memory_region *mem)
 731 {
 732         int r;
 733         gfn_t base_gfn;
 734         unsigned long npages;
 735         struct kvm_memory_slot *slot;
 736         struct kvm_memory_slot old, new;
 737         struct kvm_memslots *slots = NULL, *old_memslots;
 738         enum kvm_mr_change change;
 739
 740         r = check_memory_region_flags(mem);
 741         if (r)
 742                 goto out;
 743
 744         r = -EINVAL;
 745         /* General sanity checks */
 746         if (mem->memory_size & (PAGE_SIZE - 1))
 747                 goto out;
 748         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 749                 goto out;
 750         /* We can read the guest memory with __xxx_user() later on. */
 751         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 752             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 753              !access_ok(VERIFY_WRITE,
 754                         (void __user *)(unsigned long)mem->userspace_addr,
 755                         mem->memory_size)))
 756                 goto out;
 757         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 758                 goto out;
 759         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 760                 goto out;
 761
 762         slot = id_to_memslot(kvm->memslots, mem->slot);
 763         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 764         npages = mem->memory_size >> PAGE_SHIFT;
 765
 766         r = -EINVAL;
 767         if (npages > KVM_MEM_MAX_NR_PAGES)
 768                 goto out;
 769
 770         if (!npages)
 771                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 772
 773         new = old = *slot;
 774
 775         new.id = mem->slot;
 776         new.base_gfn = base_gfn;
 777         new.npages = npages;
 778         new.flags = mem->flags;
 779
 780         r = -EINVAL;
 781         if (npages) {
 782                 if (!old.npages)
 783                         change = KVM_MR_CREATE;
 784                 else { /* Modify an existing slot. */
 785                         if ((mem->userspace_addr != old.userspace_addr) ||
 786                             (npages != old.npages) ||
 787                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 788                                 goto out;
 789
 790                         if (base_gfn != old.base_gfn)
 791                                 change = KVM_MR_MOVE;
 792                         else if (new.flags != old.flags)
 793                                 change = KVM_MR_FLAGS_ONLY;
 794                         else { /* Nothing to change. */
 795                                 r = 0;
 796                                 goto out;
 797                         }
 798                 }
 799         } else if (old.npages) {
 800                 change = KVM_MR_DELETE;
 801         } else /* Modify a non-existent slot: disallowed. */
 802                 goto out;
 803
 804         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 805                 /* Check for overlaps */
 806                 r = -EEXIST;
 807                 kvm_for_each_memslot(slot, kvm->memslots) {
 808                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 809                             (slot->id == mem->slot))
 810                                 continue;
 811                         if (!((base_gfn + npages <= slot->base_gfn) ||
 812                               (base_gfn >= slot->base_gfn + slot->npages)))
 813                                 goto out;
 814                 }
 815         }
 816
 817         /* Free page dirty bitmap if unneeded */
 818         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 819                 new.dirty_bitmap = NULL;
 820
 821         r = -ENOMEM;
 822         if (change == KVM_MR_CREATE) {
 823                 new.userspace_addr = mem->userspace_addr;
 824
 825                 if (kvm_arch_create_memslot(&new, npages))
 826                         goto out_free;
 827         }
 828
 829         /* Allocate page dirty bitmap if needed */
 830         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 831                 if (kvm_create_dirty_bitmap(&new) < 0)
 832                         goto out_free;
 833         }
 834
 835         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 836                 r = -ENOMEM;
 837                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 838                                 GFP_KERNEL);
 839                 if (!slots)
 840                         goto out_free;
 841                 slot = id_to_memslot(slots, mem->slot);
 842                 slot->flags |= KVM_MEMSLOT_INVALID;
 843
 844                 old_memslots = install_new_memslots(kvm, slots, NULL);
 845
 846                 /* slot was deleted or moved, clear iommu mapping */
 847                 kvm_iommu_unmap_pages(kvm, &old);
 848                 /* From this point no new shadow pages pointing to a deleted,
 849                  * or moved, memslot will be created.
 850                  *
 851                  * validation of sp->gfn happens in:
 852                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 853                  *      - kvm_is_visible_gfn (mmu_check_roots)
 854                  */
 855                 kvm_arch_flush_shadow_memslot(kvm, slot);
 856                 slots = old_memslots;
 857         }
 858
 859         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 860         if (r)
 861                 goto out_slots;
 862
 863         r = -ENOMEM;
 864         /*
 865          * We can re-use the old_memslots from above, the only difference
 866          * from the currently installed memslots is the invalid flag.  This
 867          * will get overwritten by update_memslots anyway.
 868          */
 869         if (!slots) {
 870                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 871                                 GFP_KERNEL);
 872                 if (!slots)
 873                         goto out_free;
 874         }
 875
 876         /*
 877          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 878          * un-mapped and re-mapped if their base changes.  Since base change
 879          * unmapping is handled above with slot deletion, mapping alone is
 880          * needed here.  Anything else the iommu might care about for existing
 881          * slots (size changes, userspace addr changes and read-only flag
 882          * changes) is disallowed above, so any other attribute changes getting
 883          * here can be skipped.
 884          */
 885         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 886                 r = kvm_iommu_map_pages(kvm, &new);
 887                 if (r)
 888                         goto out_slots;
 889         }
 890
 891         /* actual memory is freed via old in kvm_free_physmem_slot below */
 892         if (change == KVM_MR_DELETE) {
 893                 new.dirty_bitmap = NULL;
 894                 memset(&new.arch, 0, sizeof(new.arch));
 895         }
 896
 897         old_memslots = install_new_memslots(kvm, slots, &new);
 898
 899         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 900
 901         kvm_free_physmem_slot(&old, &new);
 902         kfree(old_memslots);
 903
 904         return 0;
 905
 906 out_slots:
 907         kfree(slots);
 908 out_free:
 909         kvm_free_physmem_slot(&new, &old);
 910 out:
 911         return r;
 912 }
 913 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 914
 915 int kvm_set_memory_region(struct kvm *kvm,
 916                           struct kvm_userspace_memory_region *mem)
 917 {
 918         int r;
 919
 920         mutex_lock(&kvm->slots_lock);
 921         r = __kvm_set_memory_region(kvm, mem);
 922         mutex_unlock(&kvm->slots_lock);
 923         return r;
 924 }
 925 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 926
 927 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 928                                    struct kvm_userspace_memory_region *mem)
 929 {
 930         if (mem->slot >= KVM_USER_MEM_SLOTS)
 931                 return -EINVAL;
 932         return kvm_set_memory_region(kvm, mem);
 933 }
 934
 935 int kvm_get_dirty_log(struct kvm *kvm,
 936                         struct kvm_dirty_log *log, int *is_dirty)
 937 {
 938         struct kvm_memory_slot *memslot;
 939         int r, i;
 940         unsigned long n;
 941         unsigned long any = 0;
 942
 943         r = -EINVAL;
 944         if (log->slot >= KVM_USER_MEM_SLOTS)
 945                 goto out;
 946
 947         memslot = id_to_memslot(kvm->memslots, log->slot);
 948         r = -ENOENT;
 949         if (!memslot->dirty_bitmap)
 950                 goto out;
 951
 952         n = kvm_dirty_bitmap_bytes(memslot);
 953
 954         for (i = 0; !any && i < n/sizeof(long); ++i)
 955                 any = memslot->dirty_bitmap[i];
 956
 957         r = -EFAULT;
 958         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 959                 goto out;
 960
 961         if (any)
 962                 *is_dirty = 1;
 963
 964         r = 0;
 965 out:
 966         return r;
 967 }
 968
 969 bool kvm_largepages_enabled(void)
 970 {
 971         return largepages_enabled;
 972 }
 973
 974 void kvm_disable_largepages(void)
 975 {
 976         largepages_enabled = false;
 977 }
 978 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 979
 980 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 981 {
 982         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 983 }
 984 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 985
 986 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 987 {
 988         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 989
 990         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
 991               memslot->flags & KVM_MEMSLOT_INVALID)
 992                 return 0;
 993
 994         return 1;
 995 }
 996 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 997
 998 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 999 {
1000         struct vm_area_struct *vma;
1001         unsigned long addr, size;
1002
1003         size = PAGE_SIZE;
1004
1005         addr = gfn_to_hva(kvm, gfn);
1006         if (kvm_is_error_hva(addr))
1007                 return PAGE_SIZE;
1008
1009         down_read(&current->mm->mmap_sem);
1010         vma = find_vma(current->mm, addr);
1011         if (!vma)
1012                 goto out;
1013
1014         size = vma_kernel_pagesize(vma);
1015
1016 out:
1017         up_read(&current->mm->mmap_sem);
1018
1019         return size;
1020 }
1021
1022 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1023 {
1024         return slot->flags & KVM_MEM_READONLY;
1025 }
1026
1027 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1028                                        gfn_t *nr_pages, bool write)
1029 {
1030         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1031                 return KVM_HVA_ERR_BAD;
1032
1033         if (memslot_is_readonly(slot) && write)
1034                 return KVM_HVA_ERR_RO_BAD;
1035
1036         if (nr_pages)
1037                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1038
1039         return __gfn_to_hva_memslot(slot, gfn);
1040 }
1041
1042 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1043                                      gfn_t *nr_pages)
1044 {
1045         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1046 }
1047
1048 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1049                                  gfn_t gfn)
1050 {
1051         return gfn_to_hva_many(slot, gfn, NULL);
1052 }
1053 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1054
1055 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1056 {
1057         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1058 }
1059 EXPORT_SYMBOL_GPL(gfn_to_hva);
1060
1061 /*
1062  * If writable is set to false, the hva returned by this function is only
1063  * allowed to be read.
1064  */
1065 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1066 {
1067         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1068         if (writable)
1069                 *writable = !memslot_is_readonly(slot);
1070
1071         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1072 }
1073
1074 static int kvm_read_hva(void *data, void __user *hva, int len)
1075 {
1076         return __copy_from_user(data, hva, len);
1077 }
1078
1079 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1080 {
1081         return __copy_from_user_inatomic(data, hva, len);
1082 }
1083
1084 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1085         unsigned long start, int write, struct page **page)
1086 {
1087         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1088
1089         if (write)
1090                 flags |= FOLL_WRITE;
1091
1092         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1093 }
1094
1095 static inline int check_user_page_hwpoison(unsigned long addr)
1096 {
1097         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1098
1099         rc = __get_user_pages(current, current->mm, addr, 1,
1100                               flags, NULL, NULL, NULL);
1101         return rc == -EHWPOISON;
1102 }
1103
1104 /*
1105  * The atomic path to get the writable pfn which will be stored in @pfn,
1106  * true indicates success, otherwise false is returned.
1107  */
1108 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1109                             bool write_fault, bool *writable, pfn_t *pfn)
1110 {
1111         struct page *page[1];
1112         int npages;
1113
1114         if (!(async || atomic))
1115                 return false;
1116
1117         /*
1118          * Fast pin a writable pfn only if it is a write fault request
1119          * or the caller allows to map a writable pfn for a read fault
1120          * request.
1121          */
1122         if (!(write_fault || writable))
1123                 return false;
1124
1125         npages = __get_user_pages_fast(addr, 1, 1, page);
1126         if (npages == 1) {
1127                 *pfn = page_to_pfn(page[0]);
1128
1129                 if (writable)
1130                         *writable = true;
1131                 return true;
1132         }
1133
1134         return false;
1135 }
1136
1137 /*
1138  * The slow path to get the pfn of the specified host virtual address,
1139  * 1 indicates success, -errno is returned if error is detected.
1140  */
1141 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1142                            bool *writable, pfn_t *pfn)
1143 {
1144         struct page *page[1];
1145         int npages = 0;
1146
1147         might_sleep();
1148
1149         if (writable)
1150                 *writable = write_fault;
1151
1152         if (async) {
1153                 down_read(&current->mm->mmap_sem);
1154                 npages = get_user_page_nowait(current, current->mm,
1155                                               addr, write_fault, page);
1156                 up_read(&current->mm->mmap_sem);
1157         } else
1158                 npages = get_user_pages_fast(addr, 1, write_fault,
1159                                              page);
1160         if (npages != 1)
1161                 return npages;
1162
1163         /* map read fault as writable if possible */
1164         if (unlikely(!write_fault) && writable) {
1165                 struct page *wpage[1];
1166
1167                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1168                 if (npages == 1) {
1169                         *writable = true;
1170                         put_page(page[0]);
1171                         page[0] = wpage[0];
1172                 }
1173
1174                 npages = 1;
1175         }
1176         *pfn = page_to_pfn(page[0]);
1177         return npages;
1178 }
1179
1180 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1181 {
1182         if (unlikely(!(vma->vm_flags & VM_READ)))
1183                 return false;
1184
1185         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1186                 return false;
1187
1188         return true;
1189 }
1190
1191 /*
1192  * Pin guest page in memory and return its pfn.
1193  * @addr: host virtual address which maps memory to the guest
1194  * @atomic: whether this function can sleep
1195  * @async: whether this function need to wait IO complete if the
1196  *         host page is not in the memory
1197  * @write_fault: whether we should get a writable host page
1198  * @writable: whether it allows to map a writable host page for !@write_fault
1199  *
1200  * The function will map a writable host page for these two cases:
1201  * 1): @write_fault = true
1202  * 2): @write_fault = false && @writable, @writable will tell the caller
1203  *     whether the mapping is writable.
1204  */
1205 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1206                         bool write_fault, bool *writable)
1207 {
1208         struct vm_area_struct *vma;
1209         pfn_t pfn = 0;
1210         int npages;
1211
1212         /* we can do it either atomically or asynchronously, not both */
1213         BUG_ON(atomic && async);
1214
1215         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1216                 return pfn;
1217
1218         if (atomic)
1219                 return KVM_PFN_ERR_FAULT;
1220
1221         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1222         if (npages == 1)
1223                 return pfn;
1224
1225         down_read(&current->mm->mmap_sem);
1226         if (npages == -EHWPOISON ||
1227               (!async && check_user_page_hwpoison(addr))) {
1228                 pfn = KVM_PFN_ERR_HWPOISON;
1229                 goto exit;
1230         }
1231
1232         vma = find_vma_intersection(current->mm, addr, addr + 1);
1233
1234         if (vma == NULL)
1235                 pfn = KVM_PFN_ERR_FAULT;
1236         else if ((vma->vm_flags & VM_PFNMAP)) {
1237                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1238                         vma->vm_pgoff;
1239                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1240         } else {
1241                 if (async && vma_is_valid(vma, write_fault))
1242                         *async = true;
1243                 pfn = KVM_PFN_ERR_FAULT;
1244         }
1245 exit:
1246         up_read(&current->mm->mmap_sem);
1247         return pfn;
1248 }
1249
1250 static pfn_t
1251 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1252                      bool *async, bool write_fault, bool *writable)
1253 {
1254         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1255
1256         if (addr == KVM_HVA_ERR_RO_BAD)
1257                 return KVM_PFN_ERR_RO_FAULT;
1258
1259         if (kvm_is_error_hva(addr))
1260                 return KVM_PFN_NOSLOT;
1261
1262         /* Do not map writable pfn in the readonly memslot. */
1263         if (writable && memslot_is_readonly(slot)) {
1264                 *writable = false;
1265                 writable = NULL;
1266         }
1267
1268         return hva_to_pfn(addr, atomic, async, write_fault,
1269                           writable);
1270 }
1271
1272 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1273                           bool write_fault, bool *writable)
1274 {
1275         struct kvm_memory_slot *slot;
1276
1277         if (async)
1278                 *async = false;
1279
1280         slot = gfn_to_memslot(kvm, gfn);
1281
1282         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1283                                     writable);
1284 }
1285
1286 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1287 {
1288         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1289 }
1290 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1291
1292 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1293                        bool write_fault, bool *writable)
1294 {
1295         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1296 }
1297 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1298
1299 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1300 {
1301         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1302 }
1303 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1304
1305 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1306                       bool *writable)
1307 {
1308         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1309 }
1310 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1311
1312 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1313 {
1314         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1315 }
1316
1317 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1318 {
1319         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1320 }
1321 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1322
1323 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1324                                                                   int nr_pages)
1325 {
1326         unsigned long addr;
1327         gfn_t entry;
1328
1329         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1330         if (kvm_is_error_hva(addr))
1331                 return -1;
1332
1333         if (entry < nr_pages)
1334                 return 0;
1335
1336         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1337 }
1338 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1339
1340 static struct page *kvm_pfn_to_page(pfn_t pfn)
1341 {
1342         if (is_error_noslot_pfn(pfn))
1343                 return KVM_ERR_PTR_BAD_PAGE;
1344
1345         if (kvm_is_mmio_pfn(pfn)) {
1346                 WARN_ON(1);
1347                 return KVM_ERR_PTR_BAD_PAGE;
1348         }
1349
1350         return pfn_to_page(pfn);
1351 }
1352
1353 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1354 {
1355         pfn_t pfn;
1356
1357         pfn = gfn_to_pfn(kvm, gfn);
1358
1359         return kvm_pfn_to_page(pfn);
1360 }
1361
1362 EXPORT_SYMBOL_GPL(gfn_to_page);
1363
1364 void kvm_release_page_clean(struct page *page)
1365 {
1366         WARN_ON(is_error_page(page));
1367
1368         kvm_release_pfn_clean(page_to_pfn(page));
1369 }
1370 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1371
1372 void kvm_release_pfn_clean(pfn_t pfn)
1373 {
1374         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1375                 put_page(pfn_to_page(pfn));
1376 }
1377 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1378
1379 void kvm_release_page_dirty(struct page *page)
1380 {
1381         WARN_ON(is_error_page(page));
1382
1383         kvm_release_pfn_dirty(page_to_pfn(page));
1384 }
1385 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1386
1387 void kvm_release_pfn_dirty(pfn_t pfn)
1388 {
1389         kvm_set_pfn_dirty(pfn);
1390         kvm_release_pfn_clean(pfn);
1391 }
1392 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1393
1394 void kvm_set_page_dirty(struct page *page)
1395 {
1396         kvm_set_pfn_dirty(page_to_pfn(page));
1397 }
1398 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
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         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1616                                     offset, len);
1617 }
1618 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1619
1620 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1621 {
1622         gfn_t gfn = gpa >> PAGE_SHIFT;
1623         int seg;
1624         int offset = offset_in_page(gpa);
1625         int ret;
1626
1627         while ((seg = next_segment(len, offset)) != 0) {
1628                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1629                 if (ret < 0)
1630                         return ret;
1631                 offset = 0;
1632                 len -= seg;
1633                 ++gfn;
1634         }
1635         return 0;
1636 }
1637 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1638
1639 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1640                              gfn_t gfn)
1641 {
1642         if (memslot && memslot->dirty_bitmap) {
1643                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1644
1645                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1646         }
1647 }
1648
1649 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1650 {
1651         struct kvm_memory_slot *memslot;
1652
1653         memslot = gfn_to_memslot(kvm, gfn);
1654         mark_page_dirty_in_slot(kvm, memslot, gfn);
1655 }
1656
1657 /*
1658  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1659  */
1660 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1661 {
1662         DEFINE_WAIT(wait);
1663
1664         for (;;) {
1665                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1666
1667                 if (kvm_arch_vcpu_runnable(vcpu)) {
1668                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1669                         break;
1670                 }
1671                 if (kvm_cpu_has_pending_timer(vcpu))
1672                         break;
1673                 if (signal_pending(current))
1674                         break;
1675
1676                 schedule();
1677         }
1678
1679         finish_wait(&vcpu->wq, &wait);
1680 }
1681
1682 #ifndef CONFIG_S390
1683 /*
1684  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1685  */
1686 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1687 {
1688         int me;
1689         int cpu = vcpu->cpu;
1690         wait_queue_head_t *wqp;
1691
1692         wqp = kvm_arch_vcpu_wq(vcpu);
1693         if (waitqueue_active(wqp)) {
1694                 wake_up_interruptible(wqp);
1695                 ++vcpu->stat.halt_wakeup;
1696         }
1697
1698         me = get_cpu();
1699         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1700                 if (kvm_arch_vcpu_should_kick(vcpu))
1701                         smp_send_reschedule(cpu);
1702         put_cpu();
1703 }
1704 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1705 #endif /* !CONFIG_S390 */
1706
1707 void kvm_resched(struct kvm_vcpu *vcpu)
1708 {
1709         if (!need_resched())
1710                 return;
1711         cond_resched();
1712 }
1713 EXPORT_SYMBOL_GPL(kvm_resched);
1714
1715 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1716 {
1717         struct pid *pid;
1718         struct task_struct *task = NULL;
1719         bool ret = false;
1720
1721         rcu_read_lock();
1722         pid = rcu_dereference(target->pid);
1723         if (pid)
1724                 task = get_pid_task(target->pid, PIDTYPE_PID);
1725         rcu_read_unlock();
1726         if (!task)
1727                 return ret;
1728         if (task->flags & PF_VCPU) {
1729                 put_task_struct(task);
1730                 return ret;
1731         }
1732         ret = yield_to(task, 1);
1733         put_task_struct(task);
1734
1735         return ret;
1736 }
1737 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1738
1739 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1740 /*
1741  * Helper that checks whether a VCPU is eligible for directed yield.
1742  * Most eligible candidate to yield is decided by following heuristics:
1743  *
1744  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1745  *  (preempted lock holder), indicated by @in_spin_loop.
1746  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1747  *
1748  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1749  *  chance last time (mostly it has become eligible now since we have probably
1750  *  yielded to lockholder in last iteration. This is done by toggling
1751  *  @dy_eligible each time a VCPU checked for eligibility.)
1752  *
1753  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1754  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1755  *  burning. Giving priority for a potential lock-holder increases lock
1756  *  progress.
1757  *
1758  *  Since algorithm is based on heuristics, accessing another VCPU data without
1759  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1760  *  and continue with next VCPU and so on.
1761  */
1762 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1763 {
1764         bool eligible;
1765
1766         eligible = !vcpu->spin_loop.in_spin_loop ||
1767                         (vcpu->spin_loop.in_spin_loop &&
1768                          vcpu->spin_loop.dy_eligible);
1769
1770         if (vcpu->spin_loop.in_spin_loop)
1771                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1772
1773         return eligible;
1774 }
1775 #endif
1776
1777 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1778 {
1779         struct kvm *kvm = me->kvm;
1780         struct kvm_vcpu *vcpu;
1781         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1782         int yielded = 0;
1783         int try = 3;
1784         int pass;
1785         int i;
1786
1787         kvm_vcpu_set_in_spin_loop(me, true);
1788         /*
1789          * We boost the priority of a VCPU that is runnable but not
1790          * currently running, because it got preempted by something
1791          * else and called schedule in __vcpu_run.  Hopefully that
1792          * VCPU is holding the lock that we need and will release it.
1793          * We approximate round-robin by starting at the last boosted VCPU.
1794          */
1795         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1796                 kvm_for_each_vcpu(i, vcpu, kvm) {
1797                         if (!pass && i <= last_boosted_vcpu) {
1798                                 i = last_boosted_vcpu;
1799                                 continue;
1800                         } else if (pass && i > last_boosted_vcpu)
1801                                 break;
1802                         if (!ACCESS_ONCE(vcpu->preempted))
1803                                 continue;
1804                         if (vcpu == me)
1805                                 continue;
1806                         if (waitqueue_active(&vcpu->wq))
1807                                 continue;
1808                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1809                                 continue;
1810
1811                         yielded = kvm_vcpu_yield_to(vcpu);
1812                         if (yielded > 0) {
1813                                 kvm->last_boosted_vcpu = i;
1814                                 break;
1815                         } else if (yielded < 0) {
1816                                 try--;
1817                                 if (!try)
1818                                         break;
1819                         }
1820                 }
1821         }
1822         kvm_vcpu_set_in_spin_loop(me, false);
1823
1824         /* Ensure vcpu is not eligible during next spinloop */
1825         kvm_vcpu_set_dy_eligible(me, false);
1826 }
1827 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1828
1829 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1830 {
1831         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1832         struct page *page;
1833
1834         if (vmf->pgoff == 0)
1835                 page = virt_to_page(vcpu->run);
1836 #ifdef CONFIG_X86
1837         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1838                 page = virt_to_page(vcpu->arch.pio_data);
1839 #endif
1840 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1841         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1842                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1843 #endif
1844         else
1845                 return kvm_arch_vcpu_fault(vcpu, vmf);
1846         get_page(page);
1847         vmf->page = page;
1848         return 0;
1849 }
1850
1851 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1852         .fault = kvm_vcpu_fault,
1853 };
1854
1855 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1856 {
1857         vma->vm_ops = &kvm_vcpu_vm_ops;
1858         return 0;
1859 }
1860
1861 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1862 {
1863         struct kvm_vcpu *vcpu = filp->private_data;
1864
1865         kvm_put_kvm(vcpu->kvm);
1866         return 0;
1867 }
1868
1869 static struct file_operations kvm_vcpu_fops = {
1870         .release        = kvm_vcpu_release,
1871         .unlocked_ioctl = kvm_vcpu_ioctl,
1872 #ifdef CONFIG_COMPAT
1873         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1874 #endif
1875         .mmap           = kvm_vcpu_mmap,
1876         .llseek         = noop_llseek,
1877 };
1878
1879 /*
1880  * Allocates an inode for the vcpu.
1881  */
1882 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1883 {
1884         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1885 }
1886
1887 /*
1888  * Creates some virtual cpus.  Good luck creating more than one.
1889  */
1890 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1891 {
1892         int r;
1893         struct kvm_vcpu *vcpu, *v;
1894
1895         vcpu = kvm_arch_vcpu_create(kvm, id);
1896         if (IS_ERR(vcpu))
1897                 return PTR_ERR(vcpu);
1898
1899         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1900
1901         r = kvm_arch_vcpu_setup(vcpu);
1902         if (r)
1903                 goto vcpu_destroy;
1904
1905         mutex_lock(&kvm->lock);
1906         if (!kvm_vcpu_compatible(vcpu)) {
1907                 r = -EINVAL;
1908                 goto unlock_vcpu_destroy;
1909         }
1910         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1911                 r = -EINVAL;
1912                 goto unlock_vcpu_destroy;
1913         }
1914
1915         kvm_for_each_vcpu(r, v, kvm)
1916                 if (v->vcpu_id == id) {
1917                         r = -EEXIST;
1918                         goto unlock_vcpu_destroy;
1919                 }
1920
1921         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1922
1923         /* Now it's all set up, let userspace reach it */
1924         kvm_get_kvm(kvm);
1925         r = create_vcpu_fd(vcpu);
1926         if (r < 0) {
1927                 kvm_put_kvm(kvm);
1928                 goto unlock_vcpu_destroy;
1929         }
1930
1931         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1932         smp_wmb();
1933         atomic_inc(&kvm->online_vcpus);
1934
1935         mutex_unlock(&kvm->lock);
1936         kvm_arch_vcpu_postcreate(vcpu);
1937         return r;
1938
1939 unlock_vcpu_destroy:
1940         mutex_unlock(&kvm->lock);
1941 vcpu_destroy:
1942         kvm_arch_vcpu_destroy(vcpu);
1943         return r;
1944 }
1945
1946 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1947 {
1948         if (sigset) {
1949                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1950                 vcpu->sigset_active = 1;
1951                 vcpu->sigset = *sigset;
1952         } else
1953                 vcpu->sigset_active = 0;
1954         return 0;
1955 }
1956
1957 static long kvm_vcpu_ioctl(struct file *filp,
1958                            unsigned int ioctl, unsigned long arg)
1959 {
1960         struct kvm_vcpu *vcpu = filp->private_data;
1961         void __user *argp = (void __user *)arg;
1962         int r;
1963         struct kvm_fpu *fpu = NULL;
1964         struct kvm_sregs *kvm_sregs = NULL;
1965
1966         if (vcpu->kvm->mm != current->mm)
1967                 return -EIO;
1968
1969 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1970         /*
1971          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1972          * so vcpu_load() would break it.
1973          */
1974         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1975                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1976 #endif
1977
1978
1979         r = vcpu_load(vcpu);
1980         if (r)
1981                 return r;
1982         switch (ioctl) {
1983         case KVM_RUN:
1984                 r = -EINVAL;
1985                 if (arg)
1986                         goto out;
1987                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1988                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1989                 break;
1990         case KVM_GET_REGS: {
1991                 struct kvm_regs *kvm_regs;
1992
1993                 r = -ENOMEM;
1994                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1995                 if (!kvm_regs)
1996                         goto out;
1997                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1998                 if (r)
1999                         goto out_free1;
2000                 r = -EFAULT;
2001                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2002                         goto out_free1;
2003                 r = 0;
2004 out_free1:
2005                 kfree(kvm_regs);
2006                 break;
2007         }
2008         case KVM_SET_REGS: {
2009                 struct kvm_regs *kvm_regs;
2010
2011                 r = -ENOMEM;
2012                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2013                 if (IS_ERR(kvm_regs)) {
2014                         r = PTR_ERR(kvm_regs);
2015                         goto out;
2016                 }
2017                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2018                 kfree(kvm_regs);
2019                 break;
2020         }
2021         case KVM_GET_SREGS: {
2022                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2023                 r = -ENOMEM;
2024                 if (!kvm_sregs)
2025                         goto out;
2026                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2027                 if (r)
2028                         goto out;
2029                 r = -EFAULT;
2030                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2031                         goto out;
2032                 r = 0;
2033                 break;
2034         }
2035         case KVM_SET_SREGS: {
2036                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2037                 if (IS_ERR(kvm_sregs)) {
2038                         r = PTR_ERR(kvm_sregs);
2039                         kvm_sregs = NULL;
2040                         goto out;
2041                 }
2042                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2043                 break;
2044         }
2045         case KVM_GET_MP_STATE: {
2046                 struct kvm_mp_state mp_state;
2047
2048                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2049                 if (r)
2050                         goto out;
2051                 r = -EFAULT;
2052                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2053                         goto out;
2054                 r = 0;
2055                 break;
2056         }
2057         case KVM_SET_MP_STATE: {
2058                 struct kvm_mp_state mp_state;
2059
2060                 r = -EFAULT;
2061                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2062                         goto out;
2063                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2064                 break;
2065         }
2066         case KVM_TRANSLATE: {
2067                 struct kvm_translation tr;
2068
2069                 r = -EFAULT;
2070                 if (copy_from_user(&tr, argp, sizeof tr))
2071                         goto out;
2072                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2073                 if (r)
2074                         goto out;
2075                 r = -EFAULT;
2076                 if (copy_to_user(argp, &tr, sizeof tr))
2077                         goto out;
2078                 r = 0;
2079                 break;
2080         }
2081         case KVM_SET_GUEST_DEBUG: {
2082                 struct kvm_guest_debug dbg;
2083
2084                 r = -EFAULT;
2085                 if (copy_from_user(&dbg, argp, sizeof dbg))
2086                         goto out;
2087                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2088                 break;
2089         }
2090         case KVM_SET_SIGNAL_MASK: {
2091                 struct kvm_signal_mask __user *sigmask_arg = argp;
2092                 struct kvm_signal_mask kvm_sigmask;
2093                 sigset_t sigset, *p;
2094
2095                 p = NULL;
2096                 if (argp) {
2097                         r = -EFAULT;
2098                         if (copy_from_user(&kvm_sigmask, argp,
2099                                            sizeof kvm_sigmask))
2100                                 goto out;
2101                         r = -EINVAL;
2102                         if (kvm_sigmask.len != sizeof sigset)
2103                                 goto out;
2104                         r = -EFAULT;
2105                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2106                                            sizeof sigset))
2107                                 goto out;
2108                         p = &sigset;
2109                 }
2110                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2111                 break;
2112         }
2113         case KVM_GET_FPU: {
2114                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2115                 r = -ENOMEM;
2116                 if (!fpu)
2117                         goto out;
2118                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2119                 if (r)
2120                         goto out;
2121                 r = -EFAULT;
2122                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2123                         goto out;
2124                 r = 0;
2125                 break;
2126         }
2127         case KVM_SET_FPU: {
2128                 fpu = memdup_user(argp, sizeof(*fpu));
2129                 if (IS_ERR(fpu)) {
2130                         r = PTR_ERR(fpu);
2131                         fpu = NULL;
2132                         goto out;
2133                 }
2134                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2135                 break;
2136         }
2137         default:
2138                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2139         }
2140 out:
2141         vcpu_put(vcpu);
2142         kfree(fpu);
2143         kfree(kvm_sregs);
2144         return r;
2145 }
2146
2147 #ifdef CONFIG_COMPAT
2148 static long kvm_vcpu_compat_ioctl(struct file *filp,
2149                                   unsigned int ioctl, unsigned long arg)
2150 {
2151         struct kvm_vcpu *vcpu = filp->private_data;
2152         void __user *argp = compat_ptr(arg);
2153         int r;
2154
2155         if (vcpu->kvm->mm != current->mm)
2156                 return -EIO;
2157
2158         switch (ioctl) {
2159         case KVM_SET_SIGNAL_MASK: {
2160                 struct kvm_signal_mask __user *sigmask_arg = argp;
2161                 struct kvm_signal_mask kvm_sigmask;
2162                 compat_sigset_t csigset;
2163                 sigset_t sigset;
2164
2165                 if (argp) {
2166                         r = -EFAULT;
2167                         if (copy_from_user(&kvm_sigmask, argp,
2168                                            sizeof kvm_sigmask))
2169                                 goto out;
2170                         r = -EINVAL;
2171                         if (kvm_sigmask.len != sizeof csigset)
2172                                 goto out;
2173                         r = -EFAULT;
2174                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2175                                            sizeof csigset))
2176                                 goto out;
2177                         sigset_from_compat(&sigset, &csigset);
2178                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2179                 } else
2180                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2181                 break;
2182         }
2183         default:
2184                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2185         }
2186
2187 out:
2188         return r;
2189 }
2190 #endif
2191
2192 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2193                                  int (*accessor)(struct kvm_device *dev,
2194                                                  struct kvm_device_attr *attr),
2195                                  unsigned long arg)
2196 {
2197         struct kvm_device_attr attr;
2198
2199         if (!accessor)
2200                 return -EPERM;
2201
2202         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2203                 return -EFAULT;
2204
2205         return accessor(dev, &attr);
2206 }
2207
2208 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2209                              unsigned long arg)
2210 {
2211         struct kvm_device *dev = filp->private_data;
2212
2213         switch (ioctl) {
2214         case KVM_SET_DEVICE_ATTR:
2215                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2216         case KVM_GET_DEVICE_ATTR:
2217                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2218         case KVM_HAS_DEVICE_ATTR:
2219                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2220         default:
2221                 if (dev->ops->ioctl)
2222                         return dev->ops->ioctl(dev, ioctl, arg);
2223
2224                 return -ENOTTY;
2225         }
2226 }
2227
2228 static int kvm_device_release(struct inode *inode, struct file *filp)
2229 {
2230         struct kvm_device *dev = filp->private_data;
2231         struct kvm *kvm = dev->kvm;
2232
2233         kvm_put_kvm(kvm);
2234         return 0;
2235 }
2236
2237 static const struct file_operations kvm_device_fops = {
2238         .unlocked_ioctl = kvm_device_ioctl,
2239 #ifdef CONFIG_COMPAT
2240         .compat_ioctl = kvm_device_ioctl,
2241 #endif
2242         .release = kvm_device_release,
2243 };
2244
2245 struct kvm_device *kvm_device_from_filp(struct file *filp)
2246 {
2247         if (filp->f_op != &kvm_device_fops)
2248                 return NULL;
2249
2250         return filp->private_data;
2251 }
2252
2253 static int kvm_ioctl_create_device(struct kvm *kvm,
2254                                    struct kvm_create_device *cd)
2255 {
2256         struct kvm_device_ops *ops = NULL;
2257         struct kvm_device *dev;
2258         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2259         int ret;
2260
2261         switch (cd->type) {
2262 #ifdef CONFIG_KVM_MPIC
2263         case KVM_DEV_TYPE_FSL_MPIC_20:
2264         case KVM_DEV_TYPE_FSL_MPIC_42:
2265                 ops = &kvm_mpic_ops;
2266                 break;
2267 #endif
2268 #ifdef CONFIG_KVM_XICS
2269         case KVM_DEV_TYPE_XICS:
2270                 ops = &kvm_xics_ops;
2271                 break;
2272 #endif
2273         default:
2274                 return -ENODEV;
2275         }
2276
2277         if (test)
2278                 return 0;
2279
2280         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2281         if (!dev)
2282                 return -ENOMEM;
2283
2284         dev->ops = ops;
2285         dev->kvm = kvm;
2286
2287         ret = ops->create(dev, cd->type);
2288         if (ret < 0) {
2289                 kfree(dev);
2290                 return ret;
2291         }
2292
2293         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2294         if (ret < 0) {
2295                 ops->destroy(dev);
2296                 return ret;
2297         }
2298
2299         list_add(&dev->vm_node, &kvm->devices);
2300         kvm_get_kvm(kvm);
2301         cd->fd = ret;
2302         return 0;
2303 }
2304
2305 static long kvm_vm_ioctl(struct file *filp,
2306                            unsigned int ioctl, unsigned long arg)
2307 {
2308         struct kvm *kvm = filp->private_data;
2309         void __user *argp = (void __user *)arg;
2310         int r;
2311
2312         if (kvm->mm != current->mm)
2313                 return -EIO;
2314         switch (ioctl) {
2315         case KVM_CREATE_VCPU:
2316                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2317                 break;
2318         case KVM_SET_USER_MEMORY_REGION: {
2319                 struct kvm_userspace_memory_region kvm_userspace_mem;
2320
2321                 r = -EFAULT;
2322                 if (copy_from_user(&kvm_userspace_mem, argp,
2323                                                 sizeof kvm_userspace_mem))
2324                         goto out;
2325
2326                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2327                 break;
2328         }
2329         case KVM_GET_DIRTY_LOG: {
2330                 struct kvm_dirty_log log;
2331
2332                 r = -EFAULT;
2333                 if (copy_from_user(&log, argp, sizeof log))
2334                         goto out;
2335                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2336                 break;
2337         }
2338 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2339         case KVM_REGISTER_COALESCED_MMIO: {
2340                 struct kvm_coalesced_mmio_zone zone;
2341                 r = -EFAULT;
2342                 if (copy_from_user(&zone, argp, sizeof zone))
2343                         goto out;
2344                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2345                 break;
2346         }
2347         case KVM_UNREGISTER_COALESCED_MMIO: {
2348                 struct kvm_coalesced_mmio_zone zone;
2349                 r = -EFAULT;
2350                 if (copy_from_user(&zone, argp, sizeof zone))
2351                         goto out;
2352                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2353                 break;
2354         }
2355 #endif
2356         case KVM_IRQFD: {
2357                 struct kvm_irqfd data;
2358
2359                 r = -EFAULT;
2360                 if (copy_from_user(&data, argp, sizeof data))
2361                         goto out;
2362                 r = kvm_irqfd(kvm, &data);
2363                 break;
2364         }
2365         case KVM_IOEVENTFD: {
2366                 struct kvm_ioeventfd data;
2367
2368                 r = -EFAULT;
2369                 if (copy_from_user(&data, argp, sizeof data))
2370                         goto out;
2371                 r = kvm_ioeventfd(kvm, &data);
2372                 break;
2373         }
2374 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2375         case KVM_SET_BOOT_CPU_ID:
2376                 r = 0;
2377                 mutex_lock(&kvm->lock);
2378                 if (atomic_read(&kvm->online_vcpus) != 0)
2379                         r = -EBUSY;
2380                 else
2381                         kvm->bsp_vcpu_id = arg;
2382                 mutex_unlock(&kvm->lock);
2383                 break;
2384 #endif
2385 #ifdef CONFIG_HAVE_KVM_MSI
2386         case KVM_SIGNAL_MSI: {
2387                 struct kvm_msi msi;
2388
2389                 r = -EFAULT;
2390                 if (copy_from_user(&msi, argp, sizeof msi))
2391                         goto out;
2392                 r = kvm_send_userspace_msi(kvm, &msi);
2393                 break;
2394         }
2395 #endif
2396 #ifdef __KVM_HAVE_IRQ_LINE
2397         case KVM_IRQ_LINE_STATUS:
2398         case KVM_IRQ_LINE: {
2399                 struct kvm_irq_level irq_event;
2400
2401                 r = -EFAULT;
2402                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2403                         goto out;
2404
2405                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2406                                         ioctl == KVM_IRQ_LINE_STATUS);
2407                 if (r)
2408                         goto out;
2409
2410                 r = -EFAULT;
2411                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2412                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2413                                 goto out;
2414                 }
2415
2416                 r = 0;
2417                 break;
2418         }
2419 #endif
2420 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2421         case KVM_SET_GSI_ROUTING: {
2422                 struct kvm_irq_routing routing;
2423                 struct kvm_irq_routing __user *urouting;
2424                 struct kvm_irq_routing_entry *entries;
2425
2426                 r = -EFAULT;
2427                 if (copy_from_user(&routing, argp, sizeof(routing)))
2428                         goto out;
2429                 r = -EINVAL;
2430                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2431                         goto out;
2432                 if (routing.flags)
2433                         goto out;
2434                 r = -ENOMEM;
2435                 entries = vmalloc(routing.nr * sizeof(*entries));
2436                 if (!entries)
2437                         goto out;
2438                 r = -EFAULT;
2439                 urouting = argp;
2440                 if (copy_from_user(entries, urouting->entries,
2441                                    routing.nr * sizeof(*entries)))
2442                         goto out_free_irq_routing;
2443                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2444                                         routing.flags);
2445         out_free_irq_routing:
2446                 vfree(entries);
2447                 break;
2448         }
2449 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2450         case KVM_CREATE_DEVICE: {
2451                 struct kvm_create_device cd;
2452
2453                 r = -EFAULT;
2454                 if (copy_from_user(&cd, argp, sizeof(cd)))
2455                         goto out;
2456
2457                 r = kvm_ioctl_create_device(kvm, &cd);
2458                 if (r)
2459                         goto out;
2460
2461                 r = -EFAULT;
2462                 if (copy_to_user(argp, &cd, sizeof(cd)))
2463                         goto out;
2464
2465                 r = 0;
2466                 break;
2467         }
2468         default:
2469                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2470                 if (r == -ENOTTY)
2471                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2472         }
2473 out:
2474         return r;
2475 }
2476
2477 #ifdef CONFIG_COMPAT
2478 struct compat_kvm_dirty_log {
2479         __u32 slot;
2480         __u32 padding1;
2481         union {
2482                 compat_uptr_t dirty_bitmap; /* one bit per page */
2483                 __u64 padding2;
2484         };
2485 };
2486
2487 static long kvm_vm_compat_ioctl(struct file *filp,
2488                            unsigned int ioctl, unsigned long arg)
2489 {
2490         struct kvm *kvm = filp->private_data;
2491         int r;
2492
2493         if (kvm->mm != current->mm)
2494                 return -EIO;
2495         switch (ioctl) {
2496         case KVM_GET_DIRTY_LOG: {
2497                 struct compat_kvm_dirty_log compat_log;
2498                 struct kvm_dirty_log log;
2499
2500                 r = -EFAULT;
2501                 if (copy_from_user(&compat_log, (void __user *)arg,
2502                                    sizeof(compat_log)))
2503                         goto out;
2504                 log.slot         = compat_log.slot;
2505                 log.padding1     = compat_log.padding1;
2506                 log.padding2     = compat_log.padding2;
2507                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2508
2509                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2510                 break;
2511         }
2512         default:
2513                 r = kvm_vm_ioctl(filp, ioctl, arg);
2514         }
2515
2516 out:
2517         return r;
2518 }
2519 #endif
2520
2521 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2522 {
2523         struct page *page[1];
2524         unsigned long addr;
2525         int npages;
2526         gfn_t gfn = vmf->pgoff;
2527         struct kvm *kvm = vma->vm_file->private_data;
2528
2529         addr = gfn_to_hva(kvm, gfn);
2530         if (kvm_is_error_hva(addr))
2531                 return VM_FAULT_SIGBUS;
2532
2533         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2534                                 NULL);
2535         if (unlikely(npages != 1))
2536                 return VM_FAULT_SIGBUS;
2537
2538         vmf->page = page[0];
2539         return 0;
2540 }
2541
2542 static const struct vm_operations_struct kvm_vm_vm_ops = {
2543         .fault = kvm_vm_fault,
2544 };
2545
2546 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2547 {
2548         vma->vm_ops = &kvm_vm_vm_ops;
2549         return 0;
2550 }
2551
2552 static struct file_operations kvm_vm_fops = {
2553         .release        = kvm_vm_release,
2554         .unlocked_ioctl = kvm_vm_ioctl,
2555 #ifdef CONFIG_COMPAT
2556         .compat_ioctl   = kvm_vm_compat_ioctl,
2557 #endif
2558         .mmap           = kvm_vm_mmap,
2559         .llseek         = noop_llseek,
2560 };
2561
2562 static int kvm_dev_ioctl_create_vm(unsigned long type)
2563 {
2564         int r;
2565         struct kvm *kvm;
2566
2567         kvm = kvm_create_vm(type);
2568         if (IS_ERR(kvm))
2569                 return PTR_ERR(kvm);
2570 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2571         r = kvm_coalesced_mmio_init(kvm);
2572         if (r < 0) {
2573                 kvm_put_kvm(kvm);
2574                 return r;
2575         }
2576 #endif
2577         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2578         if (r < 0)
2579                 kvm_put_kvm(kvm);
2580
2581         return r;
2582 }
2583
2584 static long kvm_dev_ioctl_check_extension_generic(long arg)
2585 {
2586         switch (arg) {
2587         case KVM_CAP_USER_MEMORY:
2588         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2589         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2590 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2591         case KVM_CAP_SET_BOOT_CPU_ID:
2592 #endif
2593         case KVM_CAP_INTERNAL_ERROR_DATA:
2594 #ifdef CONFIG_HAVE_KVM_MSI
2595         case KVM_CAP_SIGNAL_MSI:
2596 #endif
2597 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2598         case KVM_CAP_IRQFD_RESAMPLE:
2599 #endif
2600                 return 1;
2601 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2602         case KVM_CAP_IRQ_ROUTING:
2603                 return KVM_MAX_IRQ_ROUTES;
2604 #endif
2605         default:
2606                 break;
2607         }
2608         return kvm_dev_ioctl_check_extension(arg);
2609 }
2610
2611 static long kvm_dev_ioctl(struct file *filp,
2612                           unsigned int ioctl, unsigned long arg)
2613 {
2614         long r = -EINVAL;
2615
2616         switch (ioctl) {
2617         case KVM_GET_API_VERSION:
2618                 r = -EINVAL;
2619                 if (arg)
2620                         goto out;
2621                 r = KVM_API_VERSION;
2622                 break;
2623         case KVM_CREATE_VM:
2624                 r = kvm_dev_ioctl_create_vm(arg);
2625                 break;
2626         case KVM_CHECK_EXTENSION:
2627                 r = kvm_dev_ioctl_check_extension_generic(arg);
2628                 break;
2629         case KVM_GET_VCPU_MMAP_SIZE:
2630                 r = -EINVAL;
2631                 if (arg)
2632                         goto out;
2633                 r = PAGE_SIZE;     /* struct kvm_run */
2634 #ifdef CONFIG_X86
2635                 r += PAGE_SIZE;    /* pio data page */
2636 #endif
2637 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2638                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2639 #endif
2640                 break;
2641         case KVM_TRACE_ENABLE:
2642         case KVM_TRACE_PAUSE:
2643         case KVM_TRACE_DISABLE:
2644                 r = -EOPNOTSUPP;
2645                 break;
2646         default:
2647                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2648         }
2649 out:
2650         return r;
2651 }
2652
2653 static struct file_operations kvm_chardev_ops = {
2654         .unlocked_ioctl = kvm_dev_ioctl,
2655         .compat_ioctl   = kvm_dev_ioctl,
2656         .llseek         = noop_llseek,
2657 };
2658
2659 static struct miscdevice kvm_dev = {
2660         KVM_MINOR,
2661         "kvm",
2662         &kvm_chardev_ops,
2663 };
2664
2665 static void hardware_enable_nolock(void *junk)
2666 {
2667         int cpu = raw_smp_processor_id();
2668         int r;
2669
2670         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2671                 return;
2672
2673         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2674
2675         r = kvm_arch_hardware_enable(NULL);
2676
2677         if (r) {
2678                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2679                 atomic_inc(&hardware_enable_failed);
2680                 printk(KERN_INFO "kvm: enabling virtualization on "
2681                                  "CPU%d failed\n", cpu);
2682         }
2683 }
2684
2685 static void hardware_enable(void)
2686 {
2687         raw_spin_lock(&kvm_count_lock);
2688         if (kvm_usage_count)
2689                 hardware_enable_nolock(NULL);
2690         raw_spin_unlock(&kvm_count_lock);
2691 }
2692
2693 static void hardware_disable_nolock(void *junk)
2694 {
2695         int cpu = raw_smp_processor_id();
2696
2697         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2698                 return;
2699         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2700         kvm_arch_hardware_disable(NULL);
2701 }
2702
2703 static void hardware_disable(void)
2704 {
2705         raw_spin_lock(&kvm_count_lock);
2706         if (kvm_usage_count)
2707                 hardware_disable_nolock(NULL);
2708         raw_spin_unlock(&kvm_count_lock);
2709 }
2710
2711 static void hardware_disable_all_nolock(void)
2712 {
2713         BUG_ON(!kvm_usage_count);
2714
2715         kvm_usage_count--;
2716         if (!kvm_usage_count)
2717                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2718 }
2719
2720 static void hardware_disable_all(void)
2721 {
2722         raw_spin_lock(&kvm_count_lock);
2723         hardware_disable_all_nolock();
2724         raw_spin_unlock(&kvm_count_lock);
2725 }
2726
2727 static int hardware_enable_all(void)
2728 {
2729         int r = 0;
2730
2731         raw_spin_lock(&kvm_count_lock);
2732
2733         kvm_usage_count++;
2734         if (kvm_usage_count == 1) {
2735                 atomic_set(&hardware_enable_failed, 0);
2736                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2737
2738                 if (atomic_read(&hardware_enable_failed)) {
2739                         hardware_disable_all_nolock();
2740                         r = -EBUSY;
2741                 }
2742         }
2743
2744         raw_spin_unlock(&kvm_count_lock);
2745
2746         return r;
2747 }
2748
2749 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2750                            void *v)
2751 {
2752         int cpu = (long)v;
2753
2754         val &= ~CPU_TASKS_FROZEN;
2755         switch (val) {
2756         case CPU_DYING:
2757                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2758                        cpu);
2759                 hardware_disable();
2760                 break;
2761         case CPU_STARTING:
2762                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2763                        cpu);
2764                 hardware_enable();
2765                 break;
2766         }
2767         return NOTIFY_OK;
2768 }
2769
2770 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2771                       void *v)
2772 {
2773         /*
2774          * Some (well, at least mine) BIOSes hang on reboot if
2775          * in vmx root mode.
2776          *
2777          * And Intel TXT required VMX off for all cpu when system shutdown.
2778          */
2779         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2780         kvm_rebooting = true;
2781         on_each_cpu(hardware_disable_nolock, NULL, 1);
2782         return NOTIFY_OK;
2783 }
2784
2785 static struct notifier_block kvm_reboot_notifier = {
2786         .notifier_call = kvm_reboot,
2787         .priority = 0,
2788 };
2789
2790 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2791 {
2792         int i;
2793
2794         for (i = 0; i < bus->dev_count; i++) {
2795                 struct kvm_io_device *pos = bus->range[i].dev;
2796
2797                 kvm_iodevice_destructor(pos);
2798         }
2799         kfree(bus);
2800 }
2801
2802 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2803 {
2804         const struct kvm_io_range *r1 = p1;
2805         const struct kvm_io_range *r2 = p2;
2806
2807         if (r1->addr < r2->addr)
2808                 return -1;
2809         if (r1->addr + r1->len > r2->addr + r2->len)
2810                 return 1;
2811         return 0;
2812 }
2813
2814 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2815                           gpa_t addr, int len)
2816 {
2817         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2818                 .addr = addr,
2819                 .len = len,
2820                 .dev = dev,
2821         };
2822
2823         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2824                 kvm_io_bus_sort_cmp, NULL);
2825
2826         return 0;
2827 }
2828
2829 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2830                              gpa_t addr, int len)
2831 {
2832         struct kvm_io_range *range, key;
2833         int off;
2834
2835         key = (struct kvm_io_range) {
2836                 .addr = addr,
2837                 .len = len,
2838         };
2839
2840         range = bsearch(&key, bus->range, bus->dev_count,
2841                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2842         if (range == NULL)
2843                 return -ENOENT;
2844
2845         off = range - bus->range;
2846
2847         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2848                 off--;
2849
2850         return off;
2851 }
2852
2853 /* kvm_io_bus_write - called under kvm->slots_lock */
2854 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2855                      int len, const void *val)
2856 {
2857         int idx;
2858         struct kvm_io_bus *bus;
2859         struct kvm_io_range range;
2860
2861         range = (struct kvm_io_range) {
2862                 .addr = addr,
2863                 .len = len,
2864         };
2865
2866         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2867         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2868         if (idx < 0)
2869                 return -EOPNOTSUPP;
2870
2871         while (idx < bus->dev_count &&
2872                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2873                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2874                         return 0;
2875                 idx++;
2876         }
2877
2878         return -EOPNOTSUPP;
2879 }
2880
2881 /* kvm_io_bus_read - called under kvm->slots_lock */
2882 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2883                     int len, void *val)
2884 {
2885         int idx;
2886         struct kvm_io_bus *bus;
2887         struct kvm_io_range range;
2888
2889         range = (struct kvm_io_range) {
2890                 .addr = addr,
2891                 .len = len,
2892         };
2893
2894         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2895         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2896         if (idx < 0)
2897                 return -EOPNOTSUPP;
2898
2899         while (idx < bus->dev_count &&
2900                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2901                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2902                         return 0;
2903                 idx++;
2904         }
2905
2906         return -EOPNOTSUPP;
2907 }
2908
2909 /* Caller must hold slots_lock. */
2910 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2911                             int len, struct kvm_io_device *dev)
2912 {
2913         struct kvm_io_bus *new_bus, *bus;
2914
2915         bus = kvm->buses[bus_idx];
2916         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2917                 return -ENOSPC;
2918
2919         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2920                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2921         if (!new_bus)
2922                 return -ENOMEM;
2923         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2924                sizeof(struct kvm_io_range)));
2925         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2926         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2927         synchronize_srcu_expedited(&kvm->srcu);
2928         kfree(bus);
2929
2930         return 0;
2931 }
2932
2933 /* Caller must hold slots_lock. */
2934 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2935                               struct kvm_io_device *dev)
2936 {
2937         int i, r;
2938         struct kvm_io_bus *new_bus, *bus;
2939
2940         bus = kvm->buses[bus_idx];
2941         r = -ENOENT;
2942         for (i = 0; i < bus->dev_count; i++)
2943                 if (bus->range[i].dev == dev) {
2944                         r = 0;
2945                         break;
2946                 }
2947
2948         if (r)
2949                 return r;
2950
2951         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2952                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2953         if (!new_bus)
2954                 return -ENOMEM;
2955
2956         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2957         new_bus->dev_count--;
2958         memcpy(new_bus->range + i, bus->range + i + 1,
2959                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2960
2961         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2962         synchronize_srcu_expedited(&kvm->srcu);
2963         kfree(bus);
2964         return r;
2965 }
2966
2967 static struct notifier_block kvm_cpu_notifier = {
2968         .notifier_call = kvm_cpu_hotplug,
2969 };
2970
2971 static int vm_stat_get(void *_offset, u64 *val)
2972 {
2973         unsigned offset = (long)_offset;
2974         struct kvm *kvm;
2975
2976         *val = 0;
2977         raw_spin_lock(&kvm_lock);
2978         list_for_each_entry(kvm, &vm_list, vm_list)
2979                 *val += *(u32 *)((void *)kvm + offset);
2980         raw_spin_unlock(&kvm_lock);
2981         return 0;
2982 }
2983
2984 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2985
2986 static int vcpu_stat_get(void *_offset, u64 *val)
2987 {
2988         unsigned offset = (long)_offset;
2989         struct kvm *kvm;
2990         struct kvm_vcpu *vcpu;
2991         int i;
2992
2993         *val = 0;
2994         raw_spin_lock(&kvm_lock);
2995         list_for_each_entry(kvm, &vm_list, vm_list)
2996                 kvm_for_each_vcpu(i, vcpu, kvm)
2997                         *val += *(u32 *)((void *)vcpu + offset);
2998
2999         raw_spin_unlock(&kvm_lock);
3000         return 0;
3001 }
3002
3003 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3004
3005 static const struct file_operations *stat_fops[] = {
3006         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3007         [KVM_STAT_VM]   = &vm_stat_fops,
3008 };
3009
3010 static int kvm_init_debug(void)
3011 {
3012         int r = -EFAULT;
3013         struct kvm_stats_debugfs_item *p;
3014
3015         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3016         if (kvm_debugfs_dir == NULL)
3017                 goto out;
3018
3019         for (p = debugfs_entries; p->name; ++p) {
3020                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3021                                                 (void *)(long)p->offset,
3022                                                 stat_fops[p->kind]);
3023                 if (p->dentry == NULL)
3024                         goto out_dir;
3025         }
3026
3027         return 0;
3028
3029 out_dir:
3030         debugfs_remove_recursive(kvm_debugfs_dir);
3031 out:
3032         return r;
3033 }
3034
3035 static void kvm_exit_debug(void)
3036 {
3037         struct kvm_stats_debugfs_item *p;
3038
3039         for (p = debugfs_entries; p->name; ++p)
3040                 debugfs_remove(p->dentry);
3041         debugfs_remove(kvm_debugfs_dir);
3042 }
3043
3044 static int kvm_suspend(void)
3045 {
3046         if (kvm_usage_count)
3047                 hardware_disable_nolock(NULL);
3048         return 0;
3049 }
3050
3051 static void kvm_resume(void)
3052 {
3053         if (kvm_usage_count) {
3054                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3055                 hardware_enable_nolock(NULL);
3056         }
3057 }
3058
3059 static struct syscore_ops kvm_syscore_ops = {
3060         .suspend = kvm_suspend,
3061         .resume = kvm_resume,
3062 };
3063
3064 static inline
3065 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3066 {
3067         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3068 }
3069
3070 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3071 {
3072         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3073         if (vcpu->preempted)
3074                 vcpu->preempted = false;
3075
3076         kvm_arch_vcpu_load(vcpu, cpu);
3077 }
3078
3079 static void kvm_sched_out(struct preempt_notifier *pn,
3080                           struct task_struct *next)
3081 {
3082         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3083
3084         if (current->state == TASK_RUNNING)
3085                 vcpu->preempted = true;
3086         kvm_arch_vcpu_put(vcpu);
3087 }
3088
3089 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3090                   struct module *module)
3091 {
3092         int r;
3093         int cpu;
3094
3095         r = kvm_arch_init(opaque);
3096         if (r)
3097                 goto out_fail;
3098
3099         /*
3100          * kvm_arch_init makes sure there's at most one caller
3101          * for architectures that support multiple implementations,
3102          * like intel and amd on x86.
3103          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3104          * conflicts in case kvm is already setup for another implementation.
3105          */
3106         r = kvm_irqfd_init();
3107         if (r)
3108                 goto out_irqfd;
3109
3110         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3111                 r = -ENOMEM;
3112                 goto out_free_0;
3113         }
3114
3115         r = kvm_arch_hardware_setup();
3116         if (r < 0)
3117                 goto out_free_0a;
3118
3119         for_each_online_cpu(cpu) {
3120                 smp_call_function_single(cpu,
3121                                 kvm_arch_check_processor_compat,
3122                                 &r, 1);
3123                 if (r < 0)
3124                         goto out_free_1;
3125         }
3126
3127         r = register_cpu_notifier(&kvm_cpu_notifier);
3128         if (r)
3129                 goto out_free_2;
3130         register_reboot_notifier(&kvm_reboot_notifier);
3131
3132         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3133         if (!vcpu_align)
3134                 vcpu_align = __alignof__(struct kvm_vcpu);
3135         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3136                                            0, NULL);
3137         if (!kvm_vcpu_cache) {
3138                 r = -ENOMEM;
3139                 goto out_free_3;
3140         }
3141
3142         r = kvm_async_pf_init();
3143         if (r)
3144                 goto out_free;
3145
3146         kvm_chardev_ops.owner = module;
3147         kvm_vm_fops.owner = module;
3148         kvm_vcpu_fops.owner = module;
3149
3150         r = misc_register(&kvm_dev);
3151         if (r) {
3152                 printk(KERN_ERR "kvm: misc device register failed\n");
3153                 goto out_unreg;
3154         }
3155
3156         register_syscore_ops(&kvm_syscore_ops);
3157
3158         kvm_preempt_ops.sched_in = kvm_sched_in;
3159         kvm_preempt_ops.sched_out = kvm_sched_out;
3160
3161         r = kvm_init_debug();
3162         if (r) {
3163                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3164                 goto out_undebugfs;
3165         }
3166
3167         return 0;
3168
3169 out_undebugfs:
3170         unregister_syscore_ops(&kvm_syscore_ops);
3171         misc_deregister(&kvm_dev);
3172 out_unreg:
3173         kvm_async_pf_deinit();
3174 out_free:
3175         kmem_cache_destroy(kvm_vcpu_cache);
3176 out_free_3:
3177         unregister_reboot_notifier(&kvm_reboot_notifier);
3178         unregister_cpu_notifier(&kvm_cpu_notifier);
3179 out_free_2:
3180 out_free_1:
3181         kvm_arch_hardware_unsetup();
3182 out_free_0a:
3183         free_cpumask_var(cpus_hardware_enabled);
3184 out_free_0:
3185         kvm_irqfd_exit();
3186 out_irqfd:
3187         kvm_arch_exit();
3188 out_fail:
3189         return r;
3190 }
3191 EXPORT_SYMBOL_GPL(kvm_init);
3192
3193 void kvm_exit(void)
3194 {
3195         kvm_exit_debug();
3196         misc_deregister(&kvm_dev);
3197         kmem_cache_destroy(kvm_vcpu_cache);
3198         kvm_async_pf_deinit();
3199         unregister_syscore_ops(&kvm_syscore_ops);
3200         unregister_reboot_notifier(&kvm_reboot_notifier);
3201         unregister_cpu_notifier(&kvm_cpu_notifier);
3202         on_each_cpu(hardware_disable_nolock, NULL, 1);
3203         kvm_arch_hardware_unsetup();
3204         kvm_arch_exit();
3205         kvm_irqfd_exit();
3206         free_cpumask_var(cpus_hardware_enabled);
3207 }
3208 EXPORT_SYMBOL_GPL(kvm_exit);