virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/syscore_ops.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/pgtable.h>
  57
  58 #include "coalesced_mmio.h"
  59 #include "async_pf.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/kvm.h>
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 /*
  68  * Ordering of locks:
  69  *
  70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  71  */
  72
  73 DEFINE_SPINLOCK(kvm_lock);
  74 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
  75 LIST_HEAD(vm_list);
  76
  77 static cpumask_var_t cpus_hardware_enabled;
  78 static int kvm_usage_count = 0;
  79 static atomic_t hardware_enable_failed;
  80
  81 struct kmem_cache *kvm_vcpu_cache;
  82 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  83
  84 static __read_mostly struct preempt_ops kvm_preempt_ops;
  85
  86 struct dentry *kvm_debugfs_dir;
  87
  88 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  89                            unsigned long arg);
  90 #ifdef CONFIG_COMPAT
  91 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  92                                   unsigned long arg);
  93 #endif
  94 static int hardware_enable_all(void);
  95 static void hardware_disable_all(void);
  96
  97 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  98
  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 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 191
 192 void kvm_reload_remote_mmus(struct kvm *kvm)
 193 {
 194         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 195 }
 196
 197 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 198 {
 199         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 200 }
 201
 202 void kvm_make_scan_ioapic_request(struct kvm *kvm)
 203 {
 204         make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
 205 }
 206
 207 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 208 {
 209         struct page *page;
 210         int r;
 211
 212         mutex_init(&vcpu->mutex);
 213         vcpu->cpu = -1;
 214         vcpu->kvm = kvm;
 215         vcpu->vcpu_id = id;
 216         vcpu->pid = NULL;
 217         init_waitqueue_head(&vcpu->wq);
 218         kvm_async_pf_vcpu_init(vcpu);
 219
 220         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 221         if (!page) {
 222                 r = -ENOMEM;
 223                 goto fail;
 224         }
 225         vcpu->run = page_address(page);
 226
 227         kvm_vcpu_set_in_spin_loop(vcpu, false);
 228         kvm_vcpu_set_dy_eligible(vcpu, false);
 229         vcpu->preempted = false;
 230
 231         r = kvm_arch_vcpu_init(vcpu);
 232         if (r < 0)
 233                 goto fail_free_run;
 234         return 0;
 235
 236 fail_free_run:
 237         free_page((unsigned long)vcpu->run);
 238 fail:
 239         return r;
 240 }
 241 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 242
 243 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 244 {
 245         put_pid(vcpu->pid);
 246         kvm_arch_vcpu_uninit(vcpu);
 247         free_page((unsigned long)vcpu->run);
 248 }
 249 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 250
 251 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 252 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 253 {
 254         return container_of(mn, struct kvm, mmu_notifier);
 255 }
 256
 257 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 258                                              struct mm_struct *mm,
 259                                              unsigned long address)
 260 {
 261         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 262         int need_tlb_flush, idx;
 263
 264         /*
 265          * When ->invalidate_page runs, the linux pte has been zapped
 266          * already but the page is still allocated until
 267          * ->invalidate_page returns. So if we increase the sequence
 268          * here the kvm page fault will notice if the spte can't be
 269          * established because the page is going to be freed. If
 270          * instead the kvm page fault establishes the spte before
 271          * ->invalidate_page runs, kvm_unmap_hva will release it
 272          * before returning.
 273          *
 274          * The sequence increase only need to be seen at spin_unlock
 275          * time, and not at spin_lock time.
 276          *
 277          * Increasing the sequence after the spin_unlock would be
 278          * unsafe because the kvm page fault could then establish the
 279          * pte after kvm_unmap_hva returned, without noticing the page
 280          * is going to be freed.
 281          */
 282         idx = srcu_read_lock(&kvm->srcu);
 283         spin_lock(&kvm->mmu_lock);
 284
 285         kvm->mmu_notifier_seq++;
 286         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 287         /* we've to flush the tlb before the pages can be freed */
 288         if (need_tlb_flush)
 289                 kvm_flush_remote_tlbs(kvm);
 290
 291         spin_unlock(&kvm->mmu_lock);
 292         srcu_read_unlock(&kvm->srcu, idx);
 293 }
 294
 295 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 296                                         struct mm_struct *mm,
 297                                         unsigned long address,
 298                                         pte_t pte)
 299 {
 300         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 301         int idx;
 302
 303         idx = srcu_read_lock(&kvm->srcu);
 304         spin_lock(&kvm->mmu_lock);
 305         kvm->mmu_notifier_seq++;
 306         kvm_set_spte_hva(kvm, address, pte);
 307         spin_unlock(&kvm->mmu_lock);
 308         srcu_read_unlock(&kvm->srcu, idx);
 309 }
 310
 311 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 312                                                     struct mm_struct *mm,
 313                                                     unsigned long start,
 314                                                     unsigned long end)
 315 {
 316         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 317         int need_tlb_flush = 0, idx;
 318
 319         idx = srcu_read_lock(&kvm->srcu);
 320         spin_lock(&kvm->mmu_lock);
 321         /*
 322          * The count increase must become visible at unlock time as no
 323          * spte can be established without taking the mmu_lock and
 324          * count is also read inside the mmu_lock critical section.
 325          */
 326         kvm->mmu_notifier_count++;
 327         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 328         need_tlb_flush |= kvm->tlbs_dirty;
 329         /* we've to flush the tlb before the pages can be freed */
 330         if (need_tlb_flush)
 331                 kvm_flush_remote_tlbs(kvm);
 332
 333         spin_unlock(&kvm->mmu_lock);
 334         srcu_read_unlock(&kvm->srcu, idx);
 335 }
 336
 337 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 338                                                   struct mm_struct *mm,
 339                                                   unsigned long start,
 340                                                   unsigned long end)
 341 {
 342         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 343
 344         spin_lock(&kvm->mmu_lock);
 345         /*
 346          * This sequence increase will notify the kvm page fault that
 347          * the page that is going to be mapped in the spte could have
 348          * been freed.
 349          */
 350         kvm->mmu_notifier_seq++;
 351         smp_wmb();
 352         /*
 353          * The above sequence increase must be visible before the
 354          * below count decrease, which is ensured by the smp_wmb above
 355          * in conjunction with the smp_rmb in mmu_notifier_retry().
 356          */
 357         kvm->mmu_notifier_count--;
 358         spin_unlock(&kvm->mmu_lock);
 359
 360         BUG_ON(kvm->mmu_notifier_count < 0);
 361 }
 362
 363 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 364                                               struct mm_struct *mm,
 365                                               unsigned long address)
 366 {
 367         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 368         int young, idx;
 369
 370         idx = srcu_read_lock(&kvm->srcu);
 371         spin_lock(&kvm->mmu_lock);
 372
 373         young = kvm_age_hva(kvm, address);
 374         if (young)
 375                 kvm_flush_remote_tlbs(kvm);
 376
 377         spin_unlock(&kvm->mmu_lock);
 378         srcu_read_unlock(&kvm->srcu, idx);
 379
 380         return young;
 381 }
 382
 383 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 384                                        struct mm_struct *mm,
 385                                        unsigned long address)
 386 {
 387         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 388         int young, idx;
 389
 390         idx = srcu_read_lock(&kvm->srcu);
 391         spin_lock(&kvm->mmu_lock);
 392         young = kvm_test_age_hva(kvm, address);
 393         spin_unlock(&kvm->mmu_lock);
 394         srcu_read_unlock(&kvm->srcu, idx);
 395
 396         return young;
 397 }
 398
 399 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 400                                      struct mm_struct *mm)
 401 {
 402         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 403         int idx;
 404
 405         idx = srcu_read_lock(&kvm->srcu);
 406         kvm_arch_flush_shadow_all(kvm);
 407         srcu_read_unlock(&kvm->srcu, idx);
 408 }
 409
 410 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 411         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 412         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 413         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 414         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 415         .test_young             = kvm_mmu_notifier_test_young,
 416         .change_pte             = kvm_mmu_notifier_change_pte,
 417         .release                = kvm_mmu_notifier_release,
 418 };
 419
 420 static int kvm_init_mmu_notifier(struct kvm *kvm)
 421 {
 422         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 423         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 424 }
 425
 426 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 427
 428 static int kvm_init_mmu_notifier(struct kvm *kvm)
 429 {
 430         return 0;
 431 }
 432
 433 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 434
 435 static void kvm_init_memslots_id(struct kvm *kvm)
 436 {
 437         int i;
 438         struct kvm_memslots *slots = kvm->memslots;
 439
 440         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 441                 slots->id_to_index[i] = slots->memslots[i].id = i;
 442 }
 443
 444 static struct kvm *kvm_create_vm(unsigned long type)
 445 {
 446         int r, i;
 447         struct kvm *kvm = kvm_arch_alloc_vm();
 448
 449         if (!kvm)
 450                 return ERR_PTR(-ENOMEM);
 451
 452         r = kvm_arch_init_vm(kvm, type);
 453         if (r)
 454                 goto out_err_nodisable;
 455
 456         r = hardware_enable_all();
 457         if (r)
 458                 goto out_err_nodisable;
 459
 460 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 461         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 462         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 463 #endif
 464
 465         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 466
 467         r = -ENOMEM;
 468         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 469         if (!kvm->memslots)
 470                 goto out_err_nosrcu;
 471         kvm_init_memslots_id(kvm);
 472         if (init_srcu_struct(&kvm->srcu))
 473                 goto out_err_nosrcu;
 474         for (i = 0; i < KVM_NR_BUSES; i++) {
 475                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 476                                         GFP_KERNEL);
 477                 if (!kvm->buses[i])
 478                         goto out_err;
 479         }
 480
 481         spin_lock_init(&kvm->mmu_lock);
 482         kvm->mm = current->mm;
 483         atomic_inc(&kvm->mm->mm_count);
 484         kvm_eventfd_init(kvm);
 485         mutex_init(&kvm->lock);
 486         mutex_init(&kvm->irq_lock);
 487         mutex_init(&kvm->slots_lock);
 488         atomic_set(&kvm->users_count, 1);
 489         INIT_LIST_HEAD(&kvm->devices);
 490
 491         r = kvm_init_mmu_notifier(kvm);
 492         if (r)
 493                 goto out_err;
 494
 495         spin_lock(&kvm_lock);
 496         list_add(&kvm->vm_list, &vm_list);
 497         spin_unlock(&kvm_lock);
 498
 499         return kvm;
 500
 501 out_err:
 502         cleanup_srcu_struct(&kvm->srcu);
 503 out_err_nosrcu:
 504         hardware_disable_all();
 505 out_err_nodisable:
 506         for (i = 0; i < KVM_NR_BUSES; i++)
 507                 kfree(kvm->buses[i]);
 508         kfree(kvm->memslots);
 509         kvm_arch_free_vm(kvm);
 510         return ERR_PTR(r);
 511 }
 512
 513 /*
 514  * Avoid using vmalloc for a small buffer.
 515  * Should not be used when the size is statically known.
 516  */
 517 void *kvm_kvzalloc(unsigned long size)
 518 {
 519         if (size > PAGE_SIZE)
 520                 return vzalloc(size);
 521         else
 522                 return kzalloc(size, GFP_KERNEL);
 523 }
 524
 525 void kvm_kvfree(const void *addr)
 526 {
 527         if (is_vmalloc_addr(addr))
 528                 vfree(addr);
 529         else
 530                 kfree(addr);
 531 }
 532
 533 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 534 {
 535         if (!memslot->dirty_bitmap)
 536                 return;
 537
 538         kvm_kvfree(memslot->dirty_bitmap);
 539         memslot->dirty_bitmap = NULL;
 540 }
 541
 542 /*
 543  * Free any memory in @free but not in @dont.
 544  */
 545 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
 546                                   struct kvm_memory_slot *dont)
 547 {
 548         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 549                 kvm_destroy_dirty_bitmap(free);
 550
 551         kvm_arch_free_memslot(kvm, free, dont);
 552
 553         free->npages = 0;
 554 }
 555
 556 void kvm_free_physmem(struct kvm *kvm)
 557 {
 558         struct kvm_memslots *slots = kvm->memslots;
 559         struct kvm_memory_slot *memslot;
 560
 561         kvm_for_each_memslot(memslot, slots)
 562                 kvm_free_physmem_slot(kvm, memslot, NULL);
 563
 564         kfree(kvm->memslots);
 565 }
 566
 567 static void kvm_destroy_devices(struct kvm *kvm)
 568 {
 569         struct list_head *node, *tmp;
 570
 571         list_for_each_safe(node, tmp, &kvm->devices) {
 572                 struct kvm_device *dev =
 573                         list_entry(node, struct kvm_device, vm_node);
 574
 575                 list_del(node);
 576                 dev->ops->destroy(dev);
 577         }
 578 }
 579
 580 static void kvm_destroy_vm(struct kvm *kvm)
 581 {
 582         int i;
 583         struct mm_struct *mm = kvm->mm;
 584
 585         kvm_arch_sync_events(kvm);
 586         spin_lock(&kvm_lock);
 587         list_del(&kvm->vm_list);
 588         spin_unlock(&kvm_lock);
 589         kvm_free_irq_routing(kvm);
 590         for (i = 0; i < KVM_NR_BUSES; i++)
 591                 kvm_io_bus_destroy(kvm->buses[i]);
 592         kvm_coalesced_mmio_free(kvm);
 593 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 594         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 595 #else
 596         kvm_arch_flush_shadow_all(kvm);
 597 #endif
 598         kvm_arch_destroy_vm(kvm);
 599         kvm_destroy_devices(kvm);
 600         kvm_free_physmem(kvm);
 601         cleanup_srcu_struct(&kvm->srcu);
 602         kvm_arch_free_vm(kvm);
 603         hardware_disable_all();
 604         mmdrop(mm);
 605 }
 606
 607 void kvm_get_kvm(struct kvm *kvm)
 608 {
 609         atomic_inc(&kvm->users_count);
 610 }
 611 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 612
 613 void kvm_put_kvm(struct kvm *kvm)
 614 {
 615         if (atomic_dec_and_test(&kvm->users_count))
 616                 kvm_destroy_vm(kvm);
 617 }
 618 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 619
 620
 621 static int kvm_vm_release(struct inode *inode, struct file *filp)
 622 {
 623         struct kvm *kvm = filp->private_data;
 624
 625         kvm_irqfd_release(kvm);
 626
 627         kvm_put_kvm(kvm);
 628         return 0;
 629 }
 630
 631 /*
 632  * Allocation size is twice as large as the actual dirty bitmap size.
 633  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 634  */
 635 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 636 {
 637 #ifndef CONFIG_S390
 638         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 639
 640         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 641         if (!memslot->dirty_bitmap)
 642                 return -ENOMEM;
 643
 644 #endif /* !CONFIG_S390 */
 645         return 0;
 646 }
 647
 648 static int cmp_memslot(const void *slot1, const void *slot2)
 649 {
 650         struct kvm_memory_slot *s1, *s2;
 651
 652         s1 = (struct kvm_memory_slot *)slot1;
 653         s2 = (struct kvm_memory_slot *)slot2;
 654
 655         if (s1->npages < s2->npages)
 656                 return 1;
 657         if (s1->npages > s2->npages)
 658                 return -1;
 659
 660         return 0;
 661 }
 662
 663 /*
 664  * Sort the memslots base on its size, so the larger slots
 665  * will get better fit.
 666  */
 667 static void sort_memslots(struct kvm_memslots *slots)
 668 {
 669         int i;
 670
 671         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 672               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 673
 674         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 675                 slots->id_to_index[slots->memslots[i].id] = i;
 676 }
 677
 678 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
 679                      u64 last_generation)
 680 {
 681         if (new) {
 682                 int id = new->id;
 683                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 684                 unsigned long npages = old->npages;
 685
 686                 *old = *new;
 687                 if (new->npages != npages)
 688                         sort_memslots(slots);
 689         }
 690
 691         slots->generation = last_generation + 1;
 692 }
 693
 694 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 695 {
 696         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 697
 698 #ifdef KVM_CAP_READONLY_MEM
 699         valid_flags |= KVM_MEM_READONLY;
 700 #endif
 701
 702         if (mem->flags & ~valid_flags)
 703                 return -EINVAL;
 704
 705         return 0;
 706 }
 707
 708 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 709                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 710 {
 711         struct kvm_memslots *old_memslots = kvm->memslots;
 712
 713         update_memslots(slots, new, kvm->memslots->generation);
 714         rcu_assign_pointer(kvm->memslots, slots);
 715         synchronize_srcu_expedited(&kvm->srcu);
 716
 717         kvm_arch_memslots_updated(kvm);
 718
 719         return old_memslots;
 720 }
 721
 722 /*
 723  * Allocate some memory and give it an address in the guest physical address
 724  * space.
 725  *
 726  * Discontiguous memory is allowed, mostly for framebuffers.
 727  *
 728  * Must be called holding mmap_sem for write.
 729  */
 730 int __kvm_set_memory_region(struct kvm *kvm,
 731                             struct kvm_userspace_memory_region *mem)
 732 {
 733         int r;
 734         gfn_t base_gfn;
 735         unsigned long npages;
 736         struct kvm_memory_slot *slot;
 737         struct kvm_memory_slot old, new;
 738         struct kvm_memslots *slots = NULL, *old_memslots;
 739         enum kvm_mr_change change;
 740
 741         r = check_memory_region_flags(mem);
 742         if (r)
 743                 goto out;
 744
 745         r = -EINVAL;
 746         /* General sanity checks */
 747         if (mem->memory_size & (PAGE_SIZE - 1))
 748                 goto out;
 749         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 750                 goto out;
 751         /* We can read the guest memory with __xxx_user() later on. */
 752         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 753             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 754              !access_ok(VERIFY_WRITE,
 755                         (void __user *)(unsigned long)mem->userspace_addr,
 756                         mem->memory_size)))
 757                 goto out;
 758         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 759                 goto out;
 760         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 761                 goto out;
 762
 763         slot = id_to_memslot(kvm->memslots, mem->slot);
 764         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 765         npages = mem->memory_size >> PAGE_SHIFT;
 766
 767         r = -EINVAL;
 768         if (npages > KVM_MEM_MAX_NR_PAGES)
 769                 goto out;
 770
 771         if (!npages)
 772                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 773
 774         new = old = *slot;
 775
 776         new.id = mem->slot;
 777         new.base_gfn = base_gfn;
 778         new.npages = npages;
 779         new.flags = mem->flags;
 780
 781         r = -EINVAL;
 782         if (npages) {
 783                 if (!old.npages)
 784                         change = KVM_MR_CREATE;
 785                 else { /* Modify an existing slot. */
 786                         if ((mem->userspace_addr != old.userspace_addr) ||
 787                             (npages != old.npages) ||
 788                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 789                                 goto out;
 790
 791                         if (base_gfn != old.base_gfn)
 792                                 change = KVM_MR_MOVE;
 793                         else if (new.flags != old.flags)
 794                                 change = KVM_MR_FLAGS_ONLY;
 795                         else { /* Nothing to change. */
 796                                 r = 0;
 797                                 goto out;
 798                         }
 799                 }
 800         } else if (old.npages) {
 801                 change = KVM_MR_DELETE;
 802         } else /* Modify a non-existent slot: disallowed. */
 803                 goto out;
 804
 805         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 806                 /* Check for overlaps */
 807                 r = -EEXIST;
 808                 kvm_for_each_memslot(slot, kvm->memslots) {
 809                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 810                             (slot->id == mem->slot))
 811                                 continue;
 812                         if (!((base_gfn + npages <= slot->base_gfn) ||
 813                               (base_gfn >= slot->base_gfn + slot->npages)))
 814                                 goto out;
 815                 }
 816         }
 817
 818         /* Free page dirty bitmap if unneeded */
 819         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 820                 new.dirty_bitmap = NULL;
 821
 822         r = -ENOMEM;
 823         if (change == KVM_MR_CREATE) {
 824                 new.userspace_addr = mem->userspace_addr;
 825
 826                 if (kvm_arch_create_memslot(kvm, &new, npages))
 827                         goto out_free;
 828         }
 829
 830         /* Allocate page dirty bitmap if needed */
 831         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 832                 if (kvm_create_dirty_bitmap(&new) < 0)
 833                         goto out_free;
 834         }
 835
 836         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 837                 r = -ENOMEM;
 838                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 839                                 GFP_KERNEL);
 840                 if (!slots)
 841                         goto out_free;
 842                 slot = id_to_memslot(slots, mem->slot);
 843                 slot->flags |= KVM_MEMSLOT_INVALID;
 844
 845                 old_memslots = install_new_memslots(kvm, slots, NULL);
 846
 847                 /* slot was deleted or moved, clear iommu mapping */
 848                 kvm_iommu_unmap_pages(kvm, &old);
 849                 /* From this point no new shadow pages pointing to a deleted,
 850                  * or moved, memslot will be created.
 851                  *
 852                  * validation of sp->gfn happens in:
 853                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 854                  *      - kvm_is_visible_gfn (mmu_check_roots)
 855                  */
 856                 kvm_arch_flush_shadow_memslot(kvm, slot);
 857                 slots = old_memslots;
 858         }
 859
 860         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 861         if (r)
 862                 goto out_slots;
 863
 864         r = -ENOMEM;
 865         /*
 866          * We can re-use the old_memslots from above, the only difference
 867          * from the currently installed memslots is the invalid flag.  This
 868          * will get overwritten by update_memslots anyway.
 869          */
 870         if (!slots) {
 871                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 872                                 GFP_KERNEL);
 873                 if (!slots)
 874                         goto out_free;
 875         }
 876
 877         /* actual memory is freed via old in kvm_free_physmem_slot below */
 878         if (change == KVM_MR_DELETE) {
 879                 new.dirty_bitmap = NULL;
 880                 memset(&new.arch, 0, sizeof(new.arch));
 881         }
 882
 883         old_memslots = install_new_memslots(kvm, slots, &new);
 884
 885         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 886
 887         kvm_free_physmem_slot(kvm, &old, &new);
 888         kfree(old_memslots);
 889
 890         /*
 891          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 892          * un-mapped and re-mapped if their base changes.  Since base change
 893          * unmapping is handled above with slot deletion, mapping alone is
 894          * needed here.  Anything else the iommu might care about for existing
 895          * slots (size changes, userspace addr changes and read-only flag
 896          * changes) is disallowed above, so any other attribute changes getting
 897          * here can be skipped.
 898          */
 899         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 900                 r = kvm_iommu_map_pages(kvm, &new);
 901                 return r;
 902         }
 903
 904         return 0;
 905
 906 out_slots:
 907         kfree(slots);
 908 out_free:
 909         kvm_free_physmem_slot(kvm, &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 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
 969
 970 bool kvm_largepages_enabled(void)
 971 {
 972         return largepages_enabled;
 973 }
 974
 975 void kvm_disable_largepages(void)
 976 {
 977         largepages_enabled = false;
 978 }
 979 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 980
 981 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 982 {
 983         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 984 }
 985 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 986
 987 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 988 {
 989         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 990
 991         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
 992               memslot->flags & KVM_MEMSLOT_INVALID)
 993                 return 0;
 994
 995         return 1;
 996 }
 997 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 998
 999 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1000 {
1001         struct vm_area_struct *vma;
1002         unsigned long addr, size;
1003
1004         size = PAGE_SIZE;
1005
1006         addr = gfn_to_hva(kvm, gfn);
1007         if (kvm_is_error_hva(addr))
1008                 return PAGE_SIZE;
1009
1010         down_read(&current->mm->mmap_sem);
1011         vma = find_vma(current->mm, addr);
1012         if (!vma)
1013                 goto out;
1014
1015         size = vma_kernel_pagesize(vma);
1016
1017 out:
1018         up_read(&current->mm->mmap_sem);
1019
1020         return size;
1021 }
1022
1023 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1024 {
1025         return slot->flags & KVM_MEM_READONLY;
1026 }
1027
1028 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1029                                        gfn_t *nr_pages, bool write)
1030 {
1031         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1032                 return KVM_HVA_ERR_BAD;
1033
1034         if (memslot_is_readonly(slot) && write)
1035                 return KVM_HVA_ERR_RO_BAD;
1036
1037         if (nr_pages)
1038                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1039
1040         return __gfn_to_hva_memslot(slot, gfn);
1041 }
1042
1043 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1044                                      gfn_t *nr_pages)
1045 {
1046         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1047 }
1048
1049 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1050                                  gfn_t gfn)
1051 {
1052         return gfn_to_hva_many(slot, gfn, NULL);
1053 }
1054 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1055
1056 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1057 {
1058         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1059 }
1060 EXPORT_SYMBOL_GPL(gfn_to_hva);
1061
1062 /*
1063  * If writable is set to false, the hva returned by this function is only
1064  * allowed to be read.
1065  */
1066 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1067 {
1068         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1069         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1070
1071         if (!kvm_is_error_hva(hva) && writable)
1072                 *writable = !memslot_is_readonly(slot);
1073
1074         return hva;
1075 }
1076
1077 static int kvm_read_hva(void *data, void __user *hva, int len)
1078 {
1079         return __copy_from_user(data, hva, len);
1080 }
1081
1082 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1083 {
1084         return __copy_from_user_inatomic(data, hva, len);
1085 }
1086
1087 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1088         unsigned long start, int write, struct page **page)
1089 {
1090         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1091
1092         if (write)
1093                 flags |= FOLL_WRITE;
1094
1095         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1096 }
1097
1098 static inline int check_user_page_hwpoison(unsigned long addr)
1099 {
1100         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1101
1102         rc = __get_user_pages(current, current->mm, addr, 1,
1103                               flags, NULL, NULL, NULL);
1104         return rc == -EHWPOISON;
1105 }
1106
1107 /*
1108  * The atomic path to get the writable pfn which will be stored in @pfn,
1109  * true indicates success, otherwise false is returned.
1110  */
1111 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1112                             bool write_fault, bool *writable, pfn_t *pfn)
1113 {
1114         struct page *page[1];
1115         int npages;
1116
1117         if (!(async || atomic))
1118                 return false;
1119
1120         /*
1121          * Fast pin a writable pfn only if it is a write fault request
1122          * or the caller allows to map a writable pfn for a read fault
1123          * request.
1124          */
1125         if (!(write_fault || writable))
1126                 return false;
1127
1128         npages = __get_user_pages_fast(addr, 1, 1, page);
1129         if (npages == 1) {
1130                 *pfn = page_to_pfn(page[0]);
1131
1132                 if (writable)
1133                         *writable = true;
1134                 return true;
1135         }
1136
1137         return false;
1138 }
1139
1140 /*
1141  * The slow path to get the pfn of the specified host virtual address,
1142  * 1 indicates success, -errno is returned if error is detected.
1143  */
1144 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1145                            bool *writable, pfn_t *pfn)
1146 {
1147         struct page *page[1];
1148         int npages = 0;
1149
1150         might_sleep();
1151
1152         if (writable)
1153                 *writable = write_fault;
1154
1155         if (async) {
1156                 down_read(&current->mm->mmap_sem);
1157                 npages = get_user_page_nowait(current, current->mm,
1158                                               addr, write_fault, page);
1159                 up_read(&current->mm->mmap_sem);
1160         } else
1161                 npages = get_user_pages_fast(addr, 1, write_fault,
1162                                              page);
1163         if (npages != 1)
1164                 return npages;
1165
1166         /* map read fault as writable if possible */
1167         if (unlikely(!write_fault) && writable) {
1168                 struct page *wpage[1];
1169
1170                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1171                 if (npages == 1) {
1172                         *writable = true;
1173                         put_page(page[0]);
1174                         page[0] = wpage[0];
1175                 }
1176
1177                 npages = 1;
1178         }
1179         *pfn = page_to_pfn(page[0]);
1180         return npages;
1181 }
1182
1183 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1184 {
1185         if (unlikely(!(vma->vm_flags & VM_READ)))
1186                 return false;
1187
1188         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1189                 return false;
1190
1191         return true;
1192 }
1193
1194 /*
1195  * Pin guest page in memory and return its pfn.
1196  * @addr: host virtual address which maps memory to the guest
1197  * @atomic: whether this function can sleep
1198  * @async: whether this function need to wait IO complete if the
1199  *         host page is not in the memory
1200  * @write_fault: whether we should get a writable host page
1201  * @writable: whether it allows to map a writable host page for !@write_fault
1202  *
1203  * The function will map a writable host page for these two cases:
1204  * 1): @write_fault = true
1205  * 2): @write_fault = false && @writable, @writable will tell the caller
1206  *     whether the mapping is writable.
1207  */
1208 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1209                         bool write_fault, bool *writable)
1210 {
1211         struct vm_area_struct *vma;
1212         pfn_t pfn = 0;
1213         int npages;
1214
1215         /* we can do it either atomically or asynchronously, not both */
1216         BUG_ON(atomic && async);
1217
1218         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1219                 return pfn;
1220
1221         if (atomic)
1222                 return KVM_PFN_ERR_FAULT;
1223
1224         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1225         if (npages == 1)
1226                 return pfn;
1227
1228         down_read(&current->mm->mmap_sem);
1229         if (npages == -EHWPOISON ||
1230               (!async && check_user_page_hwpoison(addr))) {
1231                 pfn = KVM_PFN_ERR_HWPOISON;
1232                 goto exit;
1233         }
1234
1235         vma = find_vma_intersection(current->mm, addr, addr + 1);
1236
1237         if (vma == NULL)
1238                 pfn = KVM_PFN_ERR_FAULT;
1239         else if ((vma->vm_flags & VM_PFNMAP)) {
1240                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1241                         vma->vm_pgoff;
1242                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1243         } else {
1244                 if (async && vma_is_valid(vma, write_fault))
1245                         *async = true;
1246                 pfn = KVM_PFN_ERR_FAULT;
1247         }
1248 exit:
1249         up_read(&current->mm->mmap_sem);
1250         return pfn;
1251 }
1252
1253 static pfn_t
1254 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1255                      bool *async, bool write_fault, bool *writable)
1256 {
1257         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1258
1259         if (addr == KVM_HVA_ERR_RO_BAD)
1260                 return KVM_PFN_ERR_RO_FAULT;
1261
1262         if (kvm_is_error_hva(addr))
1263                 return KVM_PFN_NOSLOT;
1264
1265         /* Do not map writable pfn in the readonly memslot. */
1266         if (writable && memslot_is_readonly(slot)) {
1267                 *writable = false;
1268                 writable = NULL;
1269         }
1270
1271         return hva_to_pfn(addr, atomic, async, write_fault,
1272                           writable);
1273 }
1274
1275 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1276                           bool write_fault, bool *writable)
1277 {
1278         struct kvm_memory_slot *slot;
1279
1280         if (async)
1281                 *async = false;
1282
1283         slot = gfn_to_memslot(kvm, gfn);
1284
1285         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1286                                     writable);
1287 }
1288
1289 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1290 {
1291         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1292 }
1293 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1294
1295 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1296                        bool write_fault, bool *writable)
1297 {
1298         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1299 }
1300 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1301
1302 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1303 {
1304         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1305 }
1306 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1307
1308 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1309                       bool *writable)
1310 {
1311         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1312 }
1313 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1314
1315 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1316 {
1317         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1318 }
1319
1320 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1321 {
1322         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1323 }
1324 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1325
1326 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1327                                                                   int nr_pages)
1328 {
1329         unsigned long addr;
1330         gfn_t entry;
1331
1332         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1333         if (kvm_is_error_hva(addr))
1334                 return -1;
1335
1336         if (entry < nr_pages)
1337                 return 0;
1338
1339         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1340 }
1341 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1342
1343 static struct page *kvm_pfn_to_page(pfn_t pfn)
1344 {
1345         if (is_error_noslot_pfn(pfn))
1346                 return KVM_ERR_PTR_BAD_PAGE;
1347
1348         if (kvm_is_mmio_pfn(pfn)) {
1349                 WARN_ON(1);
1350                 return KVM_ERR_PTR_BAD_PAGE;
1351         }
1352
1353         return pfn_to_page(pfn);
1354 }
1355
1356 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1357 {
1358         pfn_t pfn;
1359
1360         pfn = gfn_to_pfn(kvm, gfn);
1361
1362         return kvm_pfn_to_page(pfn);
1363 }
1364
1365 EXPORT_SYMBOL_GPL(gfn_to_page);
1366
1367 void kvm_release_page_clean(struct page *page)
1368 {
1369         WARN_ON(is_error_page(page));
1370
1371         kvm_release_pfn_clean(page_to_pfn(page));
1372 }
1373 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1374
1375 void kvm_release_pfn_clean(pfn_t pfn)
1376 {
1377         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1378                 put_page(pfn_to_page(pfn));
1379 }
1380 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1381
1382 void kvm_release_page_dirty(struct page *page)
1383 {
1384         WARN_ON(is_error_page(page));
1385
1386         kvm_release_pfn_dirty(page_to_pfn(page));
1387 }
1388 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1389
1390 void kvm_release_pfn_dirty(pfn_t pfn)
1391 {
1392         kvm_set_pfn_dirty(pfn);
1393         kvm_release_pfn_clean(pfn);
1394 }
1395 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1396
1397 void kvm_set_page_dirty(struct page *page)
1398 {
1399         kvm_set_pfn_dirty(page_to_pfn(page));
1400 }
1401 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1402
1403 void kvm_set_pfn_dirty(pfn_t pfn)
1404 {
1405         if (!kvm_is_mmio_pfn(pfn)) {
1406                 struct page *page = pfn_to_page(pfn);
1407                 if (!PageReserved(page))
1408                         SetPageDirty(page);
1409         }
1410 }
1411 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1412
1413 void kvm_set_pfn_accessed(pfn_t pfn)
1414 {
1415         if (!kvm_is_mmio_pfn(pfn))
1416                 mark_page_accessed(pfn_to_page(pfn));
1417 }
1418 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1419
1420 void kvm_get_pfn(pfn_t pfn)
1421 {
1422         if (!kvm_is_mmio_pfn(pfn))
1423                 get_page(pfn_to_page(pfn));
1424 }
1425 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1426
1427 static int next_segment(unsigned long len, int offset)
1428 {
1429         if (len > PAGE_SIZE - offset)
1430                 return PAGE_SIZE - offset;
1431         else
1432                 return len;
1433 }
1434
1435 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1436                         int len)
1437 {
1438         int r;
1439         unsigned long addr;
1440
1441         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1442         if (kvm_is_error_hva(addr))
1443                 return -EFAULT;
1444         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1445         if (r)
1446                 return -EFAULT;
1447         return 0;
1448 }
1449 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1450
1451 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1452 {
1453         gfn_t gfn = gpa >> PAGE_SHIFT;
1454         int seg;
1455         int offset = offset_in_page(gpa);
1456         int ret;
1457
1458         while ((seg = next_segment(len, offset)) != 0) {
1459                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1460                 if (ret < 0)
1461                         return ret;
1462                 offset = 0;
1463                 len -= seg;
1464                 data += seg;
1465                 ++gfn;
1466         }
1467         return 0;
1468 }
1469 EXPORT_SYMBOL_GPL(kvm_read_guest);
1470
1471 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1472                           unsigned long len)
1473 {
1474         int r;
1475         unsigned long addr;
1476         gfn_t gfn = gpa >> PAGE_SHIFT;
1477         int offset = offset_in_page(gpa);
1478
1479         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1480         if (kvm_is_error_hva(addr))
1481                 return -EFAULT;
1482         pagefault_disable();
1483         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1484         pagefault_enable();
1485         if (r)
1486                 return -EFAULT;
1487         return 0;
1488 }
1489 EXPORT_SYMBOL(kvm_read_guest_atomic);
1490
1491 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1492                          int offset, int len)
1493 {
1494         int r;
1495         unsigned long addr;
1496
1497         addr = gfn_to_hva(kvm, gfn);
1498         if (kvm_is_error_hva(addr))
1499                 return -EFAULT;
1500         r = __copy_to_user((void __user *)addr + offset, data, len);
1501         if (r)
1502                 return -EFAULT;
1503         mark_page_dirty(kvm, gfn);
1504         return 0;
1505 }
1506 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1507
1508 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1509                     unsigned long len)
1510 {
1511         gfn_t gfn = gpa >> PAGE_SHIFT;
1512         int seg;
1513         int offset = offset_in_page(gpa);
1514         int ret;
1515
1516         while ((seg = next_segment(len, offset)) != 0) {
1517                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1518                 if (ret < 0)
1519                         return ret;
1520                 offset = 0;
1521                 len -= seg;
1522                 data += seg;
1523                 ++gfn;
1524         }
1525         return 0;
1526 }
1527
1528 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1529                               gpa_t gpa, unsigned long len)
1530 {
1531         struct kvm_memslots *slots = kvm_memslots(kvm);
1532         int offset = offset_in_page(gpa);
1533         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1534         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1535         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1536         gfn_t nr_pages_avail;
1537
1538         ghc->gpa = gpa;
1539         ghc->generation = slots->generation;
1540         ghc->len = len;
1541         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1542         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1543         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1544                 ghc->hva += offset;
1545         } else {
1546                 /*
1547                  * If the requested region crosses two memslots, we still
1548                  * verify that the entire region is valid here.
1549                  */
1550                 while (start_gfn <= end_gfn) {
1551                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1552                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1553                                                    &nr_pages_avail);
1554                         if (kvm_is_error_hva(ghc->hva))
1555                                 return -EFAULT;
1556                         start_gfn += nr_pages_avail;
1557                 }
1558                 /* Use the slow path for cross page reads and writes. */
1559                 ghc->memslot = NULL;
1560         }
1561         return 0;
1562 }
1563 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1564
1565 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1566                            void *data, unsigned long len)
1567 {
1568         struct kvm_memslots *slots = kvm_memslots(kvm);
1569         int r;
1570
1571         BUG_ON(len > ghc->len);
1572
1573         if (slots->generation != ghc->generation)
1574                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1575
1576         if (unlikely(!ghc->memslot))
1577                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1578
1579         if (kvm_is_error_hva(ghc->hva))
1580                 return -EFAULT;
1581
1582         r = __copy_to_user((void __user *)ghc->hva, data, len);
1583         if (r)
1584                 return -EFAULT;
1585         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1586
1587         return 0;
1588 }
1589 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1590
1591 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1592                            void *data, unsigned long len)
1593 {
1594         struct kvm_memslots *slots = kvm_memslots(kvm);
1595         int r;
1596
1597         BUG_ON(len > ghc->len);
1598
1599         if (slots->generation != ghc->generation)
1600                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1601
1602         if (unlikely(!ghc->memslot))
1603                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1604
1605         if (kvm_is_error_hva(ghc->hva))
1606                 return -EFAULT;
1607
1608         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1609         if (r)
1610                 return -EFAULT;
1611
1612         return 0;
1613 }
1614 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1615
1616 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1617 {
1618         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1619
1620         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1621 }
1622 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1623
1624 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1625 {
1626         gfn_t gfn = gpa >> PAGE_SHIFT;
1627         int seg;
1628         int offset = offset_in_page(gpa);
1629         int ret;
1630
1631         while ((seg = next_segment(len, offset)) != 0) {
1632                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1633                 if (ret < 0)
1634                         return ret;
1635                 offset = 0;
1636                 len -= seg;
1637                 ++gfn;
1638         }
1639         return 0;
1640 }
1641 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1642
1643 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1644                              gfn_t gfn)
1645 {
1646         if (memslot && memslot->dirty_bitmap) {
1647                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1648
1649                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1650         }
1651 }
1652
1653 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1654 {
1655         struct kvm_memory_slot *memslot;
1656
1657         memslot = gfn_to_memslot(kvm, gfn);
1658         mark_page_dirty_in_slot(kvm, memslot, gfn);
1659 }
1660 EXPORT_SYMBOL_GPL(mark_page_dirty);
1661
1662 /*
1663  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1664  */
1665 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1666 {
1667         DEFINE_WAIT(wait);
1668
1669         for (;;) {
1670                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1671
1672                 if (kvm_arch_vcpu_runnable(vcpu)) {
1673                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1674                         break;
1675                 }
1676                 if (kvm_cpu_has_pending_timer(vcpu))
1677                         break;
1678                 if (signal_pending(current))
1679                         break;
1680
1681                 schedule();
1682         }
1683
1684         finish_wait(&vcpu->wq, &wait);
1685 }
1686 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
1687
1688 #ifndef CONFIG_S390
1689 /*
1690  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1691  */
1692 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1693 {
1694         int me;
1695         int cpu = vcpu->cpu;
1696         wait_queue_head_t *wqp;
1697
1698         wqp = kvm_arch_vcpu_wq(vcpu);
1699         if (waitqueue_active(wqp)) {
1700                 wake_up_interruptible(wqp);
1701                 ++vcpu->stat.halt_wakeup;
1702         }
1703
1704         me = get_cpu();
1705         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1706                 if (kvm_arch_vcpu_should_kick(vcpu))
1707                         smp_send_reschedule(cpu);
1708         put_cpu();
1709 }
1710 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1711 #endif /* !CONFIG_S390 */
1712
1713 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1714 {
1715         struct pid *pid;
1716         struct task_struct *task = NULL;
1717         bool ret = false;
1718
1719         rcu_read_lock();
1720         pid = rcu_dereference(target->pid);
1721         if (pid)
1722                 task = get_pid_task(target->pid, PIDTYPE_PID);
1723         rcu_read_unlock();
1724         if (!task)
1725                 return ret;
1726         if (task->flags & PF_VCPU) {
1727                 put_task_struct(task);
1728                 return ret;
1729         }
1730         ret = yield_to(task, 1);
1731         put_task_struct(task);
1732
1733         return ret;
1734 }
1735 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1736
1737 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1738 /*
1739  * Helper that checks whether a VCPU is eligible for directed yield.
1740  * Most eligible candidate to yield is decided by following heuristics:
1741  *
1742  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1743  *  (preempted lock holder), indicated by @in_spin_loop.
1744  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1745  *
1746  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1747  *  chance last time (mostly it has become eligible now since we have probably
1748  *  yielded to lockholder in last iteration. This is done by toggling
1749  *  @dy_eligible each time a VCPU checked for eligibility.)
1750  *
1751  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1752  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1753  *  burning. Giving priority for a potential lock-holder increases lock
1754  *  progress.
1755  *
1756  *  Since algorithm is based on heuristics, accessing another VCPU data without
1757  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1758  *  and continue with next VCPU and so on.
1759  */
1760 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1761 {
1762         bool eligible;
1763
1764         eligible = !vcpu->spin_loop.in_spin_loop ||
1765                         (vcpu->spin_loop.in_spin_loop &&
1766                          vcpu->spin_loop.dy_eligible);
1767
1768         if (vcpu->spin_loop.in_spin_loop)
1769                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1770
1771         return eligible;
1772 }
1773 #endif
1774
1775 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1776 {
1777         struct kvm *kvm = me->kvm;
1778         struct kvm_vcpu *vcpu;
1779         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1780         int yielded = 0;
1781         int try = 3;
1782         int pass;
1783         int i;
1784
1785         kvm_vcpu_set_in_spin_loop(me, true);
1786         /*
1787          * We boost the priority of a VCPU that is runnable but not
1788          * currently running, because it got preempted by something
1789          * else and called schedule in __vcpu_run.  Hopefully that
1790          * VCPU is holding the lock that we need and will release it.
1791          * We approximate round-robin by starting at the last boosted VCPU.
1792          */
1793         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1794                 kvm_for_each_vcpu(i, vcpu, kvm) {
1795                         if (!pass && i <= last_boosted_vcpu) {
1796                                 i = last_boosted_vcpu;
1797                                 continue;
1798                         } else if (pass && i > last_boosted_vcpu)
1799                                 break;
1800                         if (!ACCESS_ONCE(vcpu->preempted))
1801                                 continue;
1802                         if (vcpu == me)
1803                                 continue;
1804                         if (waitqueue_active(&vcpu->wq))
1805                                 continue;
1806                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1807                                 continue;
1808
1809                         yielded = kvm_vcpu_yield_to(vcpu);
1810                         if (yielded > 0) {
1811                                 kvm->last_boosted_vcpu = i;
1812                                 break;
1813                         } else if (yielded < 0) {
1814                                 try--;
1815                                 if (!try)
1816                                         break;
1817                         }
1818                 }
1819         }
1820         kvm_vcpu_set_in_spin_loop(me, false);
1821
1822         /* Ensure vcpu is not eligible during next spinloop */
1823         kvm_vcpu_set_dy_eligible(me, false);
1824 }
1825 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1826
1827 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1828 {
1829         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1830         struct page *page;
1831
1832         if (vmf->pgoff == 0)
1833                 page = virt_to_page(vcpu->run);
1834 #ifdef CONFIG_X86
1835         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1836                 page = virt_to_page(vcpu->arch.pio_data);
1837 #endif
1838 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1839         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1840                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1841 #endif
1842         else
1843                 return kvm_arch_vcpu_fault(vcpu, vmf);
1844         get_page(page);
1845         vmf->page = page;
1846         return 0;
1847 }
1848
1849 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1850         .fault = kvm_vcpu_fault,
1851 };
1852
1853 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1854 {
1855         vma->vm_ops = &kvm_vcpu_vm_ops;
1856         return 0;
1857 }
1858
1859 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1860 {
1861         struct kvm_vcpu *vcpu = filp->private_data;
1862
1863         kvm_put_kvm(vcpu->kvm);
1864         return 0;
1865 }
1866
1867 static struct file_operations kvm_vcpu_fops = {
1868         .release        = kvm_vcpu_release,
1869         .unlocked_ioctl = kvm_vcpu_ioctl,
1870 #ifdef CONFIG_COMPAT
1871         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1872 #endif
1873         .mmap           = kvm_vcpu_mmap,
1874         .llseek         = noop_llseek,
1875 };
1876
1877 /*
1878  * Allocates an inode for the vcpu.
1879  */
1880 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1881 {
1882         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1883 }
1884
1885 /*
1886  * Creates some virtual cpus.  Good luck creating more than one.
1887  */
1888 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1889 {
1890         int r;
1891         struct kvm_vcpu *vcpu, *v;
1892
1893         vcpu = kvm_arch_vcpu_create(kvm, id);
1894         if (IS_ERR(vcpu))
1895                 return PTR_ERR(vcpu);
1896
1897         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1898
1899         r = kvm_arch_vcpu_setup(vcpu);
1900         if (r)
1901                 goto vcpu_destroy;
1902
1903         mutex_lock(&kvm->lock);
1904         if (!kvm_vcpu_compatible(vcpu)) {
1905                 r = -EINVAL;
1906                 goto unlock_vcpu_destroy;
1907         }
1908         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1909                 r = -EINVAL;
1910                 goto unlock_vcpu_destroy;
1911         }
1912
1913         kvm_for_each_vcpu(r, v, kvm)
1914                 if (v->vcpu_id == id) {
1915                         r = -EEXIST;
1916                         goto unlock_vcpu_destroy;
1917                 }
1918
1919         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1920
1921         /* Now it's all set up, let userspace reach it */
1922         kvm_get_kvm(kvm);
1923         r = create_vcpu_fd(vcpu);
1924         if (r < 0) {
1925                 kvm_put_kvm(kvm);
1926                 goto unlock_vcpu_destroy;
1927         }
1928
1929         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1930         smp_wmb();
1931         atomic_inc(&kvm->online_vcpus);
1932
1933         mutex_unlock(&kvm->lock);
1934         kvm_arch_vcpu_postcreate(vcpu);
1935         return r;
1936
1937 unlock_vcpu_destroy:
1938         mutex_unlock(&kvm->lock);
1939 vcpu_destroy:
1940         kvm_arch_vcpu_destroy(vcpu);
1941         return r;
1942 }
1943
1944 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1945 {
1946         if (sigset) {
1947                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1948                 vcpu->sigset_active = 1;
1949                 vcpu->sigset = *sigset;
1950         } else
1951                 vcpu->sigset_active = 0;
1952         return 0;
1953 }
1954
1955 static long kvm_vcpu_ioctl(struct file *filp,
1956                            unsigned int ioctl, unsigned long arg)
1957 {
1958         struct kvm_vcpu *vcpu = filp->private_data;
1959         void __user *argp = (void __user *)arg;
1960         int r;
1961         struct kvm_fpu *fpu = NULL;
1962         struct kvm_sregs *kvm_sregs = NULL;
1963
1964         if (vcpu->kvm->mm != current->mm)
1965                 return -EIO;
1966
1967 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1968         /*
1969          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1970          * so vcpu_load() would break it.
1971          */
1972         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1973                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1974 #endif
1975
1976
1977         r = vcpu_load(vcpu);
1978         if (r)
1979                 return r;
1980         switch (ioctl) {
1981         case KVM_RUN:
1982                 r = -EINVAL;
1983                 if (arg)
1984                         goto out;
1985                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1986                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1987                 break;
1988         case KVM_GET_REGS: {
1989                 struct kvm_regs *kvm_regs;
1990
1991                 r = -ENOMEM;
1992                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1993                 if (!kvm_regs)
1994                         goto out;
1995                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1996                 if (r)
1997                         goto out_free1;
1998                 r = -EFAULT;
1999                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2000                         goto out_free1;
2001                 r = 0;
2002 out_free1:
2003                 kfree(kvm_regs);
2004                 break;
2005         }
2006         case KVM_SET_REGS: {
2007                 struct kvm_regs *kvm_regs;
2008
2009                 r = -ENOMEM;
2010                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2011                 if (IS_ERR(kvm_regs)) {
2012                         r = PTR_ERR(kvm_regs);
2013                         goto out;
2014                 }
2015                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2016                 kfree(kvm_regs);
2017                 break;
2018         }
2019         case KVM_GET_SREGS: {
2020                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2021                 r = -ENOMEM;
2022                 if (!kvm_sregs)
2023                         goto out;
2024                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2025                 if (r)
2026                         goto out;
2027                 r = -EFAULT;
2028                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2029                         goto out;
2030                 r = 0;
2031                 break;
2032         }
2033         case KVM_SET_SREGS: {
2034                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2035                 if (IS_ERR(kvm_sregs)) {
2036                         r = PTR_ERR(kvm_sregs);
2037                         kvm_sregs = NULL;
2038                         goto out;
2039                 }
2040                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2041                 break;
2042         }
2043         case KVM_GET_MP_STATE: {
2044                 struct kvm_mp_state mp_state;
2045
2046                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2047                 if (r)
2048                         goto out;
2049                 r = -EFAULT;
2050                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2051                         goto out;
2052                 r = 0;
2053                 break;
2054         }
2055         case KVM_SET_MP_STATE: {
2056                 struct kvm_mp_state mp_state;
2057
2058                 r = -EFAULT;
2059                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2060                         goto out;
2061                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2062                 break;
2063         }
2064         case KVM_TRANSLATE: {
2065                 struct kvm_translation tr;
2066
2067                 r = -EFAULT;
2068                 if (copy_from_user(&tr, argp, sizeof tr))
2069                         goto out;
2070                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2071                 if (r)
2072                         goto out;
2073                 r = -EFAULT;
2074                 if (copy_to_user(argp, &tr, sizeof tr))
2075                         goto out;
2076                 r = 0;
2077                 break;
2078         }
2079         case KVM_SET_GUEST_DEBUG: {
2080                 struct kvm_guest_debug dbg;
2081
2082                 r = -EFAULT;
2083                 if (copy_from_user(&dbg, argp, sizeof dbg))
2084                         goto out;
2085                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2086                 break;
2087         }
2088         case KVM_SET_SIGNAL_MASK: {
2089                 struct kvm_signal_mask __user *sigmask_arg = argp;
2090                 struct kvm_signal_mask kvm_sigmask;
2091                 sigset_t sigset, *p;
2092
2093                 p = NULL;
2094                 if (argp) {
2095                         r = -EFAULT;
2096                         if (copy_from_user(&kvm_sigmask, argp,
2097                                            sizeof kvm_sigmask))
2098                                 goto out;
2099                         r = -EINVAL;
2100                         if (kvm_sigmask.len != sizeof sigset)
2101                                 goto out;
2102                         r = -EFAULT;
2103                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2104                                            sizeof sigset))
2105                                 goto out;
2106                         p = &sigset;
2107                 }
2108                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2109                 break;
2110         }
2111         case KVM_GET_FPU: {
2112                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2113                 r = -ENOMEM;
2114                 if (!fpu)
2115                         goto out;
2116                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2117                 if (r)
2118                         goto out;
2119                 r = -EFAULT;
2120                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2121                         goto out;
2122                 r = 0;
2123                 break;
2124         }
2125         case KVM_SET_FPU: {
2126                 fpu = memdup_user(argp, sizeof(*fpu));
2127                 if (IS_ERR(fpu)) {
2128                         r = PTR_ERR(fpu);
2129                         fpu = NULL;
2130                         goto out;
2131                 }
2132                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2133                 break;
2134         }
2135         default:
2136                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2137         }
2138 out:
2139         vcpu_put(vcpu);
2140         kfree(fpu);
2141         kfree(kvm_sregs);
2142         return r;
2143 }
2144
2145 #ifdef CONFIG_COMPAT
2146 static long kvm_vcpu_compat_ioctl(struct file *filp,
2147                                   unsigned int ioctl, unsigned long arg)
2148 {
2149         struct kvm_vcpu *vcpu = filp->private_data;
2150         void __user *argp = compat_ptr(arg);
2151         int r;
2152
2153         if (vcpu->kvm->mm != current->mm)
2154                 return -EIO;
2155
2156         switch (ioctl) {
2157         case KVM_SET_SIGNAL_MASK: {
2158                 struct kvm_signal_mask __user *sigmask_arg = argp;
2159                 struct kvm_signal_mask kvm_sigmask;
2160                 compat_sigset_t csigset;
2161                 sigset_t sigset;
2162
2163                 if (argp) {
2164                         r = -EFAULT;
2165                         if (copy_from_user(&kvm_sigmask, argp,
2166                                            sizeof kvm_sigmask))
2167                                 goto out;
2168                         r = -EINVAL;
2169                         if (kvm_sigmask.len != sizeof csigset)
2170                                 goto out;
2171                         r = -EFAULT;
2172                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2173                                            sizeof csigset))
2174                                 goto out;
2175                         sigset_from_compat(&sigset, &csigset);
2176                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2177                 } else
2178                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2179                 break;
2180         }
2181         default:
2182                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2183         }
2184
2185 out:
2186         return r;
2187 }
2188 #endif
2189
2190 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2191                                  int (*accessor)(struct kvm_device *dev,
2192                                                  struct kvm_device_attr *attr),
2193                                  unsigned long arg)
2194 {
2195         struct kvm_device_attr attr;
2196
2197         if (!accessor)
2198                 return -EPERM;
2199
2200         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2201                 return -EFAULT;
2202
2203         return accessor(dev, &attr);
2204 }
2205
2206 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2207                              unsigned long arg)
2208 {
2209         struct kvm_device *dev = filp->private_data;
2210
2211         switch (ioctl) {
2212         case KVM_SET_DEVICE_ATTR:
2213                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2214         case KVM_GET_DEVICE_ATTR:
2215                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2216         case KVM_HAS_DEVICE_ATTR:
2217                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2218         default:
2219                 if (dev->ops->ioctl)
2220                         return dev->ops->ioctl(dev, ioctl, arg);
2221
2222                 return -ENOTTY;
2223         }
2224 }
2225
2226 static int kvm_device_release(struct inode *inode, struct file *filp)
2227 {
2228         struct kvm_device *dev = filp->private_data;
2229         struct kvm *kvm = dev->kvm;
2230
2231         kvm_put_kvm(kvm);
2232         return 0;
2233 }
2234
2235 static const struct file_operations kvm_device_fops = {
2236         .unlocked_ioctl = kvm_device_ioctl,
2237 #ifdef CONFIG_COMPAT
2238         .compat_ioctl = kvm_device_ioctl,
2239 #endif
2240         .release = kvm_device_release,
2241 };
2242
2243 struct kvm_device *kvm_device_from_filp(struct file *filp)
2244 {
2245         if (filp->f_op != &kvm_device_fops)
2246                 return NULL;
2247
2248         return filp->private_data;
2249 }
2250
2251 static int kvm_ioctl_create_device(struct kvm *kvm,
2252                                    struct kvm_create_device *cd)
2253 {
2254         struct kvm_device_ops *ops = NULL;
2255         struct kvm_device *dev;
2256         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2257         int ret;
2258
2259         switch (cd->type) {
2260 #ifdef CONFIG_KVM_MPIC
2261         case KVM_DEV_TYPE_FSL_MPIC_20:
2262         case KVM_DEV_TYPE_FSL_MPIC_42:
2263                 ops = &kvm_mpic_ops;
2264                 break;
2265 #endif
2266 #ifdef CONFIG_KVM_XICS
2267         case KVM_DEV_TYPE_XICS:
2268                 ops = &kvm_xics_ops;
2269                 break;
2270 #endif
2271 #ifdef CONFIG_KVM_VFIO
2272         case KVM_DEV_TYPE_VFIO:
2273                 ops = &kvm_vfio_ops;
2274                 break;
2275 #endif
2276 #ifdef CONFIG_KVM_ARM_VGIC
2277         case KVM_DEV_TYPE_ARM_VGIC_V2:
2278                 ops = &kvm_arm_vgic_v2_ops;
2279                 break;
2280 #endif
2281         default:
2282                 return -ENODEV;
2283         }
2284
2285         if (test)
2286                 return 0;
2287
2288         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2289         if (!dev)
2290                 return -ENOMEM;
2291
2292         dev->ops = ops;
2293         dev->kvm = kvm;
2294
2295         ret = ops->create(dev, cd->type);
2296         if (ret < 0) {
2297                 kfree(dev);
2298                 return ret;
2299         }
2300
2301         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2302         if (ret < 0) {
2303                 ops->destroy(dev);
2304                 return ret;
2305         }
2306
2307         list_add(&dev->vm_node, &kvm->devices);
2308         kvm_get_kvm(kvm);
2309         cd->fd = ret;
2310         return 0;
2311 }
2312
2313 static long kvm_vm_ioctl(struct file *filp,
2314                            unsigned int ioctl, unsigned long arg)
2315 {
2316         struct kvm *kvm = filp->private_data;
2317         void __user *argp = (void __user *)arg;
2318         int r;
2319
2320         if (kvm->mm != current->mm)
2321                 return -EIO;
2322         switch (ioctl) {
2323         case KVM_CREATE_VCPU:
2324                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2325                 break;
2326         case KVM_SET_USER_MEMORY_REGION: {
2327                 struct kvm_userspace_memory_region kvm_userspace_mem;
2328
2329                 r = -EFAULT;
2330                 if (copy_from_user(&kvm_userspace_mem, argp,
2331                                                 sizeof kvm_userspace_mem))
2332                         goto out;
2333
2334                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2335                 break;
2336         }
2337         case KVM_GET_DIRTY_LOG: {
2338                 struct kvm_dirty_log log;
2339
2340                 r = -EFAULT;
2341                 if (copy_from_user(&log, argp, sizeof log))
2342                         goto out;
2343                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2344                 break;
2345         }
2346 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2347         case KVM_REGISTER_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_register_coalesced_mmio(kvm, &zone);
2353                 break;
2354         }
2355         case KVM_UNREGISTER_COALESCED_MMIO: {
2356                 struct kvm_coalesced_mmio_zone zone;
2357                 r = -EFAULT;
2358                 if (copy_from_user(&zone, argp, sizeof zone))
2359                         goto out;
2360                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2361                 break;
2362         }
2363 #endif
2364         case KVM_IRQFD: {
2365                 struct kvm_irqfd data;
2366
2367                 r = -EFAULT;
2368                 if (copy_from_user(&data, argp, sizeof data))
2369                         goto out;
2370                 r = kvm_irqfd(kvm, &data);
2371                 break;
2372         }
2373         case KVM_IOEVENTFD: {
2374                 struct kvm_ioeventfd data;
2375
2376                 r = -EFAULT;
2377                 if (copy_from_user(&data, argp, sizeof data))
2378                         goto out;
2379                 r = kvm_ioeventfd(kvm, &data);
2380                 break;
2381         }
2382 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2383         case KVM_SET_BOOT_CPU_ID:
2384                 r = 0;
2385                 mutex_lock(&kvm->lock);
2386                 if (atomic_read(&kvm->online_vcpus) != 0)
2387                         r = -EBUSY;
2388                 else
2389                         kvm->bsp_vcpu_id = arg;
2390                 mutex_unlock(&kvm->lock);
2391                 break;
2392 #endif
2393 #ifdef CONFIG_HAVE_KVM_MSI
2394         case KVM_SIGNAL_MSI: {
2395                 struct kvm_msi msi;
2396
2397                 r = -EFAULT;
2398                 if (copy_from_user(&msi, argp, sizeof msi))
2399                         goto out;
2400                 r = kvm_send_userspace_msi(kvm, &msi);
2401                 break;
2402         }
2403 #endif
2404 #ifdef __KVM_HAVE_IRQ_LINE
2405         case KVM_IRQ_LINE_STATUS:
2406         case KVM_IRQ_LINE: {
2407                 struct kvm_irq_level irq_event;
2408
2409                 r = -EFAULT;
2410                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2411                         goto out;
2412
2413                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2414                                         ioctl == KVM_IRQ_LINE_STATUS);
2415                 if (r)
2416                         goto out;
2417
2418                 r = -EFAULT;
2419                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2420                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2421                                 goto out;
2422                 }
2423
2424                 r = 0;
2425                 break;
2426         }
2427 #endif
2428 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2429         case KVM_SET_GSI_ROUTING: {
2430                 struct kvm_irq_routing routing;
2431                 struct kvm_irq_routing __user *urouting;
2432                 struct kvm_irq_routing_entry *entries;
2433
2434                 r = -EFAULT;
2435                 if (copy_from_user(&routing, argp, sizeof(routing)))
2436                         goto out;
2437                 r = -EINVAL;
2438                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2439                         goto out;
2440                 if (routing.flags)
2441                         goto out;
2442                 r = -ENOMEM;
2443                 entries = vmalloc(routing.nr * sizeof(*entries));
2444                 if (!entries)
2445                         goto out;
2446                 r = -EFAULT;
2447                 urouting = argp;
2448                 if (copy_from_user(entries, urouting->entries,
2449                                    routing.nr * sizeof(*entries)))
2450                         goto out_free_irq_routing;
2451                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2452                                         routing.flags);
2453         out_free_irq_routing:
2454                 vfree(entries);
2455                 break;
2456         }
2457 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2458         case KVM_CREATE_DEVICE: {
2459                 struct kvm_create_device cd;
2460
2461                 r = -EFAULT;
2462                 if (copy_from_user(&cd, argp, sizeof(cd)))
2463                         goto out;
2464
2465                 r = kvm_ioctl_create_device(kvm, &cd);
2466                 if (r)
2467                         goto out;
2468
2469                 r = -EFAULT;
2470                 if (copy_to_user(argp, &cd, sizeof(cd)))
2471                         goto out;
2472
2473                 r = 0;
2474                 break;
2475         }
2476         default:
2477                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2478                 if (r == -ENOTTY)
2479                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2480         }
2481 out:
2482         return r;
2483 }
2484
2485 #ifdef CONFIG_COMPAT
2486 struct compat_kvm_dirty_log {
2487         __u32 slot;
2488         __u32 padding1;
2489         union {
2490                 compat_uptr_t dirty_bitmap; /* one bit per page */
2491                 __u64 padding2;
2492         };
2493 };
2494
2495 static long kvm_vm_compat_ioctl(struct file *filp,
2496                            unsigned int ioctl, unsigned long arg)
2497 {
2498         struct kvm *kvm = filp->private_data;
2499         int r;
2500
2501         if (kvm->mm != current->mm)
2502                 return -EIO;
2503         switch (ioctl) {
2504         case KVM_GET_DIRTY_LOG: {
2505                 struct compat_kvm_dirty_log compat_log;
2506                 struct kvm_dirty_log log;
2507
2508                 r = -EFAULT;
2509                 if (copy_from_user(&compat_log, (void __user *)arg,
2510                                    sizeof(compat_log)))
2511                         goto out;
2512                 log.slot         = compat_log.slot;
2513                 log.padding1     = compat_log.padding1;
2514                 log.padding2     = compat_log.padding2;
2515                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2516
2517                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2518                 break;
2519         }
2520         default:
2521                 r = kvm_vm_ioctl(filp, ioctl, arg);
2522         }
2523
2524 out:
2525         return r;
2526 }
2527 #endif
2528
2529 static struct file_operations kvm_vm_fops = {
2530         .release        = kvm_vm_release,
2531         .unlocked_ioctl = kvm_vm_ioctl,
2532 #ifdef CONFIG_COMPAT
2533         .compat_ioctl   = kvm_vm_compat_ioctl,
2534 #endif
2535         .llseek         = noop_llseek,
2536 };
2537
2538 static int kvm_dev_ioctl_create_vm(unsigned long type)
2539 {
2540         int r;
2541         struct kvm *kvm;
2542
2543         kvm = kvm_create_vm(type);
2544         if (IS_ERR(kvm))
2545                 return PTR_ERR(kvm);
2546 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2547         r = kvm_coalesced_mmio_init(kvm);
2548         if (r < 0) {
2549                 kvm_put_kvm(kvm);
2550                 return r;
2551         }
2552 #endif
2553         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2554         if (r < 0)
2555                 kvm_put_kvm(kvm);
2556
2557         return r;
2558 }
2559
2560 static long kvm_dev_ioctl_check_extension_generic(long arg)
2561 {
2562         switch (arg) {
2563         case KVM_CAP_USER_MEMORY:
2564         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2565         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2566 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2567         case KVM_CAP_SET_BOOT_CPU_ID:
2568 #endif
2569         case KVM_CAP_INTERNAL_ERROR_DATA:
2570 #ifdef CONFIG_HAVE_KVM_MSI
2571         case KVM_CAP_SIGNAL_MSI:
2572 #endif
2573 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2574         case KVM_CAP_IRQFD_RESAMPLE:
2575 #endif
2576                 return 1;
2577 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2578         case KVM_CAP_IRQ_ROUTING:
2579                 return KVM_MAX_IRQ_ROUTES;
2580 #endif
2581         default:
2582                 break;
2583         }
2584         return kvm_dev_ioctl_check_extension(arg);
2585 }
2586
2587 static long kvm_dev_ioctl(struct file *filp,
2588                           unsigned int ioctl, unsigned long arg)
2589 {
2590         long r = -EINVAL;
2591
2592         switch (ioctl) {
2593         case KVM_GET_API_VERSION:
2594                 r = -EINVAL;
2595                 if (arg)
2596                         goto out;
2597                 r = KVM_API_VERSION;
2598                 break;
2599         case KVM_CREATE_VM:
2600                 r = kvm_dev_ioctl_create_vm(arg);
2601                 break;
2602         case KVM_CHECK_EXTENSION:
2603                 r = kvm_dev_ioctl_check_extension_generic(arg);
2604                 break;
2605         case KVM_GET_VCPU_MMAP_SIZE:
2606                 r = -EINVAL;
2607                 if (arg)
2608                         goto out;
2609                 r = PAGE_SIZE;     /* struct kvm_run */
2610 #ifdef CONFIG_X86
2611                 r += PAGE_SIZE;    /* pio data page */
2612 #endif
2613 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2614                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2615 #endif
2616                 break;
2617         case KVM_TRACE_ENABLE:
2618         case KVM_TRACE_PAUSE:
2619         case KVM_TRACE_DISABLE:
2620                 r = -EOPNOTSUPP;
2621                 break;
2622         default:
2623                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2624         }
2625 out:
2626         return r;
2627 }
2628
2629 static struct file_operations kvm_chardev_ops = {
2630         .unlocked_ioctl = kvm_dev_ioctl,
2631         .compat_ioctl   = kvm_dev_ioctl,
2632         .llseek         = noop_llseek,
2633 };
2634
2635 static struct miscdevice kvm_dev = {
2636         KVM_MINOR,
2637         "kvm",
2638         &kvm_chardev_ops,
2639 };
2640
2641 static void hardware_enable_nolock(void *junk)
2642 {
2643         int cpu = raw_smp_processor_id();
2644         int r;
2645
2646         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2647                 return;
2648
2649         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2650
2651         r = kvm_arch_hardware_enable(NULL);
2652
2653         if (r) {
2654                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2655                 atomic_inc(&hardware_enable_failed);
2656                 printk(KERN_INFO "kvm: enabling virtualization on "
2657                                  "CPU%d failed\n", cpu);
2658         }
2659 }
2660
2661 static void hardware_enable(void)
2662 {
2663         raw_spin_lock(&kvm_count_lock);
2664         if (kvm_usage_count)
2665                 hardware_enable_nolock(NULL);
2666         raw_spin_unlock(&kvm_count_lock);
2667 }
2668
2669 static void hardware_disable_nolock(void *junk)
2670 {
2671         int cpu = raw_smp_processor_id();
2672
2673         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2674                 return;
2675         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2676         kvm_arch_hardware_disable(NULL);
2677 }
2678
2679 static void hardware_disable(void)
2680 {
2681         raw_spin_lock(&kvm_count_lock);
2682         if (kvm_usage_count)
2683                 hardware_disable_nolock(NULL);
2684         raw_spin_unlock(&kvm_count_lock);
2685 }
2686
2687 static void hardware_disable_all_nolock(void)
2688 {
2689         BUG_ON(!kvm_usage_count);
2690
2691         kvm_usage_count--;
2692         if (!kvm_usage_count)
2693                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2694 }
2695
2696 static void hardware_disable_all(void)
2697 {
2698         raw_spin_lock(&kvm_count_lock);
2699         hardware_disable_all_nolock();
2700         raw_spin_unlock(&kvm_count_lock);
2701 }
2702
2703 static int hardware_enable_all(void)
2704 {
2705         int r = 0;
2706
2707         raw_spin_lock(&kvm_count_lock);
2708
2709         kvm_usage_count++;
2710         if (kvm_usage_count == 1) {
2711                 atomic_set(&hardware_enable_failed, 0);
2712                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2713
2714                 if (atomic_read(&hardware_enable_failed)) {
2715                         hardware_disable_all_nolock();
2716                         r = -EBUSY;
2717                 }
2718         }
2719
2720         raw_spin_unlock(&kvm_count_lock);
2721
2722         return r;
2723 }
2724
2725 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2726                            void *v)
2727 {
2728         int cpu = (long)v;
2729
2730         val &= ~CPU_TASKS_FROZEN;
2731         switch (val) {
2732         case CPU_DYING:
2733                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2734                        cpu);
2735                 hardware_disable();
2736                 break;
2737         case CPU_STARTING:
2738                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2739                        cpu);
2740                 hardware_enable();
2741                 break;
2742         }
2743         return NOTIFY_OK;
2744 }
2745
2746 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2747                       void *v)
2748 {
2749         /*
2750          * Some (well, at least mine) BIOSes hang on reboot if
2751          * in vmx root mode.
2752          *
2753          * And Intel TXT required VMX off for all cpu when system shutdown.
2754          */
2755         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2756         kvm_rebooting = true;
2757         on_each_cpu(hardware_disable_nolock, NULL, 1);
2758         return NOTIFY_OK;
2759 }
2760
2761 static struct notifier_block kvm_reboot_notifier = {
2762         .notifier_call = kvm_reboot,
2763         .priority = 0,
2764 };
2765
2766 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2767 {
2768         int i;
2769
2770         for (i = 0; i < bus->dev_count; i++) {
2771                 struct kvm_io_device *pos = bus->range[i].dev;
2772
2773                 kvm_iodevice_destructor(pos);
2774         }
2775         kfree(bus);
2776 }
2777
2778 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
2779                                  const struct kvm_io_range *r2)
2780 {
2781         if (r1->addr < r2->addr)
2782                 return -1;
2783         if (r1->addr + r1->len > r2->addr + r2->len)
2784                 return 1;
2785         return 0;
2786 }
2787
2788 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2789 {
2790         return kvm_io_bus_cmp(p1, p2);
2791 }
2792
2793 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2794                           gpa_t addr, int len)
2795 {
2796         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2797                 .addr = addr,
2798                 .len = len,
2799                 .dev = dev,
2800         };
2801
2802         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2803                 kvm_io_bus_sort_cmp, NULL);
2804
2805         return 0;
2806 }
2807
2808 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2809                              gpa_t addr, int len)
2810 {
2811         struct kvm_io_range *range, key;
2812         int off;
2813
2814         key = (struct kvm_io_range) {
2815                 .addr = addr,
2816                 .len = len,
2817         };
2818
2819         range = bsearch(&key, bus->range, bus->dev_count,
2820                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2821         if (range == NULL)
2822                 return -ENOENT;
2823
2824         off = range - bus->range;
2825
2826         while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
2827                 off--;
2828
2829         return off;
2830 }
2831
2832 static int __kvm_io_bus_write(struct kvm_io_bus *bus,
2833                               struct kvm_io_range *range, const void *val)
2834 {
2835         int idx;
2836
2837         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2838         if (idx < 0)
2839                 return -EOPNOTSUPP;
2840
2841         while (idx < bus->dev_count &&
2842                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2843                 if (!kvm_iodevice_write(bus->range[idx].dev, range->addr,
2844                                         range->len, val))
2845                         return idx;
2846                 idx++;
2847         }
2848
2849         return -EOPNOTSUPP;
2850 }
2851
2852 /* kvm_io_bus_write - called under kvm->slots_lock */
2853 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2854                      int len, const void *val)
2855 {
2856         struct kvm_io_bus *bus;
2857         struct kvm_io_range range;
2858         int r;
2859
2860         range = (struct kvm_io_range) {
2861                 .addr = addr,
2862                 .len = len,
2863         };
2864
2865         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2866         r = __kvm_io_bus_write(bus, &range, val);
2867         return r < 0 ? r : 0;
2868 }
2869
2870 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
2871 int kvm_io_bus_write_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2872                             int len, const void *val, long cookie)
2873 {
2874         struct kvm_io_bus *bus;
2875         struct kvm_io_range range;
2876
2877         range = (struct kvm_io_range) {
2878                 .addr = addr,
2879                 .len = len,
2880         };
2881
2882         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2883
2884         /* First try the device referenced by cookie. */
2885         if ((cookie >= 0) && (cookie < bus->dev_count) &&
2886             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2887                 if (!kvm_iodevice_write(bus->range[cookie].dev, addr, len,
2888                                         val))
2889                         return cookie;
2890
2891         /*
2892          * cookie contained garbage; fall back to search and return the
2893          * correct cookie value.
2894          */
2895         return __kvm_io_bus_write(bus, &range, val);
2896 }
2897
2898 static int __kvm_io_bus_read(struct kvm_io_bus *bus, struct kvm_io_range *range,
2899                              void *val)
2900 {
2901         int idx;
2902
2903         idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
2904         if (idx < 0)
2905                 return -EOPNOTSUPP;
2906
2907         while (idx < bus->dev_count &&
2908                 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
2909                 if (!kvm_iodevice_read(bus->range[idx].dev, range->addr,
2910                                        range->len, val))
2911                         return idx;
2912                 idx++;
2913         }
2914
2915         return -EOPNOTSUPP;
2916 }
2917
2918 /* kvm_io_bus_read - called under kvm->slots_lock */
2919 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2920                     int len, void *val)
2921 {
2922         struct kvm_io_bus *bus;
2923         struct kvm_io_range range;
2924         int r;
2925
2926         range = (struct kvm_io_range) {
2927                 .addr = addr,
2928                 .len = len,
2929         };
2930
2931         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2932         r = __kvm_io_bus_read(bus, &range, val);
2933         return r < 0 ? r : 0;
2934 }
2935
2936 /* kvm_io_bus_read_cookie - called under kvm->slots_lock */
2937 int kvm_io_bus_read_cookie(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2938                            int len, void *val, long cookie)
2939 {
2940         struct kvm_io_bus *bus;
2941         struct kvm_io_range range;
2942
2943         range = (struct kvm_io_range) {
2944                 .addr = addr,
2945                 .len = len,
2946         };
2947
2948         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2949
2950         /* First try the device referenced by cookie. */
2951         if ((cookie >= 0) && (cookie < bus->dev_count) &&
2952             (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
2953                 if (!kvm_iodevice_read(bus->range[cookie].dev, addr, len,
2954                                        val))
2955                         return cookie;
2956
2957         /*
2958          * cookie contained garbage; fall back to search and return the
2959          * correct cookie value.
2960          */
2961         return __kvm_io_bus_read(bus, &range, val);
2962 }
2963
2964 /* Caller must hold slots_lock. */
2965 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2966                             int len, struct kvm_io_device *dev)
2967 {
2968         struct kvm_io_bus *new_bus, *bus;
2969
2970         bus = kvm->buses[bus_idx];
2971         /* exclude ioeventfd which is limited by maximum fd */
2972         if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
2973                 return -ENOSPC;
2974
2975         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2976                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2977         if (!new_bus)
2978                 return -ENOMEM;
2979         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2980                sizeof(struct kvm_io_range)));
2981         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2982         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2983         synchronize_srcu_expedited(&kvm->srcu);
2984         kfree(bus);
2985
2986         return 0;
2987 }
2988
2989 /* Caller must hold slots_lock. */
2990 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2991                               struct kvm_io_device *dev)
2992 {
2993         int i, r;
2994         struct kvm_io_bus *new_bus, *bus;
2995
2996         bus = kvm->buses[bus_idx];
2997         r = -ENOENT;
2998         for (i = 0; i < bus->dev_count; i++)
2999                 if (bus->range[i].dev == dev) {
3000                         r = 0;
3001                         break;
3002                 }
3003
3004         if (r)
3005                 return r;
3006
3007         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3008                           sizeof(struct kvm_io_range)), GFP_KERNEL);
3009         if (!new_bus)
3010                 return -ENOMEM;
3011
3012         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3013         new_bus->dev_count--;
3014         memcpy(new_bus->range + i, bus->range + i + 1,
3015                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3016
3017         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3018         synchronize_srcu_expedited(&kvm->srcu);
3019         kfree(bus);
3020         return r;
3021 }
3022
3023 static struct notifier_block kvm_cpu_notifier = {
3024         .notifier_call = kvm_cpu_hotplug,
3025 };
3026
3027 static int vm_stat_get(void *_offset, u64 *val)
3028 {
3029         unsigned offset = (long)_offset;
3030         struct kvm *kvm;
3031
3032         *val = 0;
3033         spin_lock(&kvm_lock);
3034         list_for_each_entry(kvm, &vm_list, vm_list)
3035                 *val += *(u32 *)((void *)kvm + offset);
3036         spin_unlock(&kvm_lock);
3037         return 0;
3038 }
3039
3040 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3041
3042 static int vcpu_stat_get(void *_offset, u64 *val)
3043 {
3044         unsigned offset = (long)_offset;
3045         struct kvm *kvm;
3046         struct kvm_vcpu *vcpu;
3047         int i;
3048
3049         *val = 0;
3050         spin_lock(&kvm_lock);
3051         list_for_each_entry(kvm, &vm_list, vm_list)
3052                 kvm_for_each_vcpu(i, vcpu, kvm)
3053                         *val += *(u32 *)((void *)vcpu + offset);
3054
3055         spin_unlock(&kvm_lock);
3056         return 0;
3057 }
3058
3059 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3060
3061 static const struct file_operations *stat_fops[] = {
3062         [KVM_STAT_VCPU] = &vcpu_stat_fops,
3063         [KVM_STAT_VM]   = &vm_stat_fops,
3064 };
3065
3066 static int kvm_init_debug(void)
3067 {
3068         int r = -EEXIST;
3069         struct kvm_stats_debugfs_item *p;
3070
3071         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3072         if (kvm_debugfs_dir == NULL)
3073                 goto out;
3074
3075         for (p = debugfs_entries; p->name; ++p) {
3076                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3077                                                 (void *)(long)p->offset,
3078                                                 stat_fops[p->kind]);
3079                 if (p->dentry == NULL)
3080                         goto out_dir;
3081         }
3082
3083         return 0;
3084
3085 out_dir:
3086         debugfs_remove_recursive(kvm_debugfs_dir);
3087 out:
3088         return r;
3089 }
3090
3091 static void kvm_exit_debug(void)
3092 {
3093         struct kvm_stats_debugfs_item *p;
3094
3095         for (p = debugfs_entries; p->name; ++p)
3096                 debugfs_remove(p->dentry);
3097         debugfs_remove(kvm_debugfs_dir);
3098 }
3099
3100 static int kvm_suspend(void)
3101 {
3102         if (kvm_usage_count)
3103                 hardware_disable_nolock(NULL);
3104         return 0;
3105 }
3106
3107 static void kvm_resume(void)
3108 {
3109         if (kvm_usage_count) {
3110                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3111                 hardware_enable_nolock(NULL);
3112         }
3113 }
3114
3115 static struct syscore_ops kvm_syscore_ops = {
3116         .suspend = kvm_suspend,
3117         .resume = kvm_resume,
3118 };
3119
3120 static inline
3121 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3122 {
3123         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3124 }
3125
3126 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3127 {
3128         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3129         if (vcpu->preempted)
3130                 vcpu->preempted = false;
3131
3132         kvm_arch_vcpu_load(vcpu, cpu);
3133 }
3134
3135 static void kvm_sched_out(struct preempt_notifier *pn,
3136                           struct task_struct *next)
3137 {
3138         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3139
3140         if (current->state == TASK_RUNNING)
3141                 vcpu->preempted = true;
3142         kvm_arch_vcpu_put(vcpu);
3143 }
3144
3145 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3146                   struct module *module)
3147 {
3148         int r;
3149         int cpu;
3150
3151         r = kvm_arch_init(opaque);
3152         if (r)
3153                 goto out_fail;
3154
3155         /*
3156          * kvm_arch_init makes sure there's at most one caller
3157          * for architectures that support multiple implementations,
3158          * like intel and amd on x86.
3159          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3160          * conflicts in case kvm is already setup for another implementation.
3161          */
3162         r = kvm_irqfd_init();
3163         if (r)
3164                 goto out_irqfd;
3165
3166         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3167                 r = -ENOMEM;
3168                 goto out_free_0;
3169         }
3170
3171         r = kvm_arch_hardware_setup();
3172         if (r < 0)
3173                 goto out_free_0a;
3174
3175         for_each_online_cpu(cpu) {
3176                 smp_call_function_single(cpu,
3177                                 kvm_arch_check_processor_compat,
3178                                 &r, 1);
3179                 if (r < 0)
3180                         goto out_free_1;
3181         }
3182
3183         r = register_cpu_notifier(&kvm_cpu_notifier);
3184         if (r)
3185                 goto out_free_2;
3186         register_reboot_notifier(&kvm_reboot_notifier);
3187
3188         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3189         if (!vcpu_align)
3190                 vcpu_align = __alignof__(struct kvm_vcpu);
3191         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3192                                            0, NULL);
3193         if (!kvm_vcpu_cache) {
3194                 r = -ENOMEM;
3195                 goto out_free_3;
3196         }
3197
3198         r = kvm_async_pf_init();
3199         if (r)
3200                 goto out_free;
3201
3202         kvm_chardev_ops.owner = module;
3203         kvm_vm_fops.owner = module;
3204         kvm_vcpu_fops.owner = module;
3205
3206         r = misc_register(&kvm_dev);
3207         if (r) {
3208                 printk(KERN_ERR "kvm: misc device register failed\n");
3209                 goto out_unreg;
3210         }
3211
3212         register_syscore_ops(&kvm_syscore_ops);
3213
3214         kvm_preempt_ops.sched_in = kvm_sched_in;
3215         kvm_preempt_ops.sched_out = kvm_sched_out;
3216
3217         r = kvm_init_debug();
3218         if (r) {
3219                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3220                 goto out_undebugfs;
3221         }
3222
3223         return 0;
3224
3225 out_undebugfs:
3226         unregister_syscore_ops(&kvm_syscore_ops);
3227         misc_deregister(&kvm_dev);
3228 out_unreg:
3229         kvm_async_pf_deinit();
3230 out_free:
3231         kmem_cache_destroy(kvm_vcpu_cache);
3232 out_free_3:
3233         unregister_reboot_notifier(&kvm_reboot_notifier);
3234         unregister_cpu_notifier(&kvm_cpu_notifier);
3235 out_free_2:
3236 out_free_1:
3237         kvm_arch_hardware_unsetup();
3238 out_free_0a:
3239         free_cpumask_var(cpus_hardware_enabled);
3240 out_free_0:
3241         kvm_irqfd_exit();
3242 out_irqfd:
3243         kvm_arch_exit();
3244 out_fail:
3245         return r;
3246 }
3247 EXPORT_SYMBOL_GPL(kvm_init);
3248
3249 void kvm_exit(void)
3250 {
3251         kvm_exit_debug();
3252         misc_deregister(&kvm_dev);
3253         kmem_cache_destroy(kvm_vcpu_cache);
3254         kvm_async_pf_deinit();
3255         unregister_syscore_ops(&kvm_syscore_ops);
3256         unregister_reboot_notifier(&kvm_reboot_notifier);
3257         unregister_cpu_notifier(&kvm_cpu_notifier);
3258         on_each_cpu(hardware_disable_nolock, NULL, 1);
3259         kvm_arch_hardware_unsetup();
3260         kvm_arch_exit();
3261         kvm_irqfd_exit();
3262         free_cpumask_var(cpus_hardware_enabled);
3263 }
3264 EXPORT_SYMBOL_GPL(kvm_exit);