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
 191 void kvm_reload_remote_mmus(struct kvm *kvm)
 192 {
 193         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 194 }
 195
 196 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 197 {
 198         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 199 }
 200
 201 void kvm_make_scan_ioapic_request(struct kvm *kvm)
 202 {
 203         make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
 204 }
 205
 206 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 207 {
 208         struct page *page;
 209         int r;
 210
 211         mutex_init(&vcpu->mutex);
 212         vcpu->cpu = -1;
 213         vcpu->kvm = kvm;
 214         vcpu->vcpu_id = id;
 215         vcpu->pid = NULL;
 216         init_waitqueue_head(&vcpu->wq);
 217         kvm_async_pf_vcpu_init(vcpu);
 218
 219         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 220         if (!page) {
 221                 r = -ENOMEM;
 222                 goto fail;
 223         }
 224         vcpu->run = page_address(page);
 225
 226         kvm_vcpu_set_in_spin_loop(vcpu, false);
 227         kvm_vcpu_set_dy_eligible(vcpu, false);
 228         vcpu->preempted = false;
 229
 230         r = kvm_arch_vcpu_init(vcpu);
 231         if (r < 0)
 232                 goto fail_free_run;
 233         return 0;
 234
 235 fail_free_run:
 236         free_page((unsigned long)vcpu->run);
 237 fail:
 238         return r;
 239 }
 240 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 241
 242 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 243 {
 244         put_pid(vcpu->pid);
 245         kvm_arch_vcpu_uninit(vcpu);
 246         free_page((unsigned long)vcpu->run);
 247 }
 248 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 249
 250 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 251 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 252 {
 253         return container_of(mn, struct kvm, mmu_notifier);
 254 }
 255
 256 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 257                                              struct mm_struct *mm,
 258                                              unsigned long address)
 259 {
 260         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 261         int need_tlb_flush, idx;
 262
 263         /*
 264          * When ->invalidate_page runs, the linux pte has been zapped
 265          * already but the page is still allocated until
 266          * ->invalidate_page returns. So if we increase the sequence
 267          * here the kvm page fault will notice if the spte can't be
 268          * established because the page is going to be freed. If
 269          * instead the kvm page fault establishes the spte before
 270          * ->invalidate_page runs, kvm_unmap_hva will release it
 271          * before returning.
 272          *
 273          * The sequence increase only need to be seen at spin_unlock
 274          * time, and not at spin_lock time.
 275          *
 276          * Increasing the sequence after the spin_unlock would be
 277          * unsafe because the kvm page fault could then establish the
 278          * pte after kvm_unmap_hva returned, without noticing the page
 279          * is going to be freed.
 280          */
 281         idx = srcu_read_lock(&kvm->srcu);
 282         spin_lock(&kvm->mmu_lock);
 283
 284         kvm->mmu_notifier_seq++;
 285         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 286         /* we've to flush the tlb before the pages can be freed */
 287         if (need_tlb_flush)
 288                 kvm_flush_remote_tlbs(kvm);
 289
 290         spin_unlock(&kvm->mmu_lock);
 291         srcu_read_unlock(&kvm->srcu, idx);
 292 }
 293
 294 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 295                                         struct mm_struct *mm,
 296                                         unsigned long address,
 297                                         pte_t pte)
 298 {
 299         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 300         int idx;
 301
 302         idx = srcu_read_lock(&kvm->srcu);
 303         spin_lock(&kvm->mmu_lock);
 304         kvm->mmu_notifier_seq++;
 305         kvm_set_spte_hva(kvm, address, pte);
 306         spin_unlock(&kvm->mmu_lock);
 307         srcu_read_unlock(&kvm->srcu, idx);
 308 }
 309
 310 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 311                                                     struct mm_struct *mm,
 312                                                     unsigned long start,
 313                                                     unsigned long end)
 314 {
 315         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 316         int need_tlb_flush = 0, idx;
 317
 318         idx = srcu_read_lock(&kvm->srcu);
 319         spin_lock(&kvm->mmu_lock);
 320         /*
 321          * The count increase must become visible at unlock time as no
 322          * spte can be established without taking the mmu_lock and
 323          * count is also read inside the mmu_lock critical section.
 324          */
 325         kvm->mmu_notifier_count++;
 326         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 327         need_tlb_flush |= kvm->tlbs_dirty;
 328         /* we've to flush the tlb before the pages can be freed */
 329         if (need_tlb_flush)
 330                 kvm_flush_remote_tlbs(kvm);
 331
 332         spin_unlock(&kvm->mmu_lock);
 333         srcu_read_unlock(&kvm->srcu, idx);
 334 }
 335
 336 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 337                                                   struct mm_struct *mm,
 338                                                   unsigned long start,
 339                                                   unsigned long end)
 340 {
 341         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 342
 343         spin_lock(&kvm->mmu_lock);
 344         /*
 345          * This sequence increase will notify the kvm page fault that
 346          * the page that is going to be mapped in the spte could have
 347          * been freed.
 348          */
 349         kvm->mmu_notifier_seq++;
 350         smp_wmb();
 351         /*
 352          * The above sequence increase must be visible before the
 353          * below count decrease, which is ensured by the smp_wmb above
 354          * in conjunction with the smp_rmb in mmu_notifier_retry().
 355          */
 356         kvm->mmu_notifier_count--;
 357         spin_unlock(&kvm->mmu_lock);
 358
 359         BUG_ON(kvm->mmu_notifier_count < 0);
 360 }
 361
 362 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 363                                               struct mm_struct *mm,
 364                                               unsigned long address)
 365 {
 366         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 367         int young, idx;
 368
 369         idx = srcu_read_lock(&kvm->srcu);
 370         spin_lock(&kvm->mmu_lock);
 371
 372         young = kvm_age_hva(kvm, address);
 373         if (young)
 374                 kvm_flush_remote_tlbs(kvm);
 375
 376         spin_unlock(&kvm->mmu_lock);
 377         srcu_read_unlock(&kvm->srcu, idx);
 378
 379         return young;
 380 }
 381
 382 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 383                                        struct mm_struct *mm,
 384                                        unsigned long address)
 385 {
 386         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 387         int young, idx;
 388
 389         idx = srcu_read_lock(&kvm->srcu);
 390         spin_lock(&kvm->mmu_lock);
 391         young = kvm_test_age_hva(kvm, address);
 392         spin_unlock(&kvm->mmu_lock);
 393         srcu_read_unlock(&kvm->srcu, idx);
 394
 395         return young;
 396 }
 397
 398 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 399                                      struct mm_struct *mm)
 400 {
 401         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 402         int idx;
 403
 404         idx = srcu_read_lock(&kvm->srcu);
 405         kvm_arch_flush_shadow_all(kvm);
 406         srcu_read_unlock(&kvm->srcu, idx);
 407 }
 408
 409 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 410         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 411         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 412         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 413         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 414         .test_young             = kvm_mmu_notifier_test_young,
 415         .change_pte             = kvm_mmu_notifier_change_pte,
 416         .release                = kvm_mmu_notifier_release,
 417 };
 418
 419 static int kvm_init_mmu_notifier(struct kvm *kvm)
 420 {
 421         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 422         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 423 }
 424
 425 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 426
 427 static int kvm_init_mmu_notifier(struct kvm *kvm)
 428 {
 429         return 0;
 430 }
 431
 432 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 433
 434 static void kvm_init_memslots_id(struct kvm *kvm)
 435 {
 436         int i;
 437         struct kvm_memslots *slots = kvm->memslots;
 438
 439         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 440                 slots->id_to_index[i] = slots->memslots[i].id = i;
 441 }
 442
 443 static struct kvm *kvm_create_vm(unsigned long type)
 444 {
 445         int r, i;
 446         struct kvm *kvm = kvm_arch_alloc_vm();
 447
 448         if (!kvm)
 449                 return ERR_PTR(-ENOMEM);
 450
 451         r = kvm_arch_init_vm(kvm, type);
 452         if (r)
 453                 goto out_err_nodisable;
 454
 455         r = hardware_enable_all();
 456         if (r)
 457                 goto out_err_nodisable;
 458
 459 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 460         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 461         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 462 #endif
 463
 464         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 465
 466         r = -ENOMEM;
 467         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 468         if (!kvm->memslots)
 469                 goto out_err_nosrcu;
 470         kvm_init_memslots_id(kvm);
 471         if (init_srcu_struct(&kvm->srcu))
 472                 goto out_err_nosrcu;
 473         for (i = 0; i < KVM_NR_BUSES; i++) {
 474                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 475                                         GFP_KERNEL);
 476                 if (!kvm->buses[i])
 477                         goto out_err;
 478         }
 479
 480         spin_lock_init(&kvm->mmu_lock);
 481         kvm->mm = current->mm;
 482         atomic_inc(&kvm->mm->mm_count);
 483         kvm_eventfd_init(kvm);
 484         mutex_init(&kvm->lock);
 485         mutex_init(&kvm->irq_lock);
 486         mutex_init(&kvm->slots_lock);
 487         atomic_set(&kvm->users_count, 1);
 488         INIT_LIST_HEAD(&kvm->devices);
 489
 490         r = kvm_init_mmu_notifier(kvm);
 491         if (r)
 492                 goto out_err;
 493
 494         spin_lock(&kvm_lock);
 495         list_add(&kvm->vm_list, &vm_list);
 496         spin_unlock(&kvm_lock);
 497
 498         return kvm;
 499
 500 out_err:
 501         cleanup_srcu_struct(&kvm->srcu);
 502 out_err_nosrcu:
 503         hardware_disable_all();
 504 out_err_nodisable:
 505         for (i = 0; i < KVM_NR_BUSES; i++)
 506                 kfree(kvm->buses[i]);
 507         kfree(kvm->memslots);
 508         kvm_arch_free_vm(kvm);
 509         return ERR_PTR(r);
 510 }
 511
 512 /*
 513  * Avoid using vmalloc for a small buffer.
 514  * Should not be used when the size is statically known.
 515  */
 516 void *kvm_kvzalloc(unsigned long size)
 517 {
 518         if (size > PAGE_SIZE)
 519                 return vzalloc(size);
 520         else
 521                 return kzalloc(size, GFP_KERNEL);
 522 }
 523
 524 void kvm_kvfree(const void *addr)
 525 {
 526         if (is_vmalloc_addr(addr))
 527                 vfree(addr);
 528         else
 529                 kfree(addr);
 530 }
 531
 532 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 533 {
 534         if (!memslot->dirty_bitmap)
 535                 return;
 536
 537         kvm_kvfree(memslot->dirty_bitmap);
 538         memslot->dirty_bitmap = NULL;
 539 }
 540
 541 /*
 542  * Free any memory in @free but not in @dont.
 543  */
 544 static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
 545                                   struct kvm_memory_slot *dont)
 546 {
 547         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 548                 kvm_destroy_dirty_bitmap(free);
 549
 550         kvm_arch_free_memslot(kvm, free, dont);
 551
 552         free->npages = 0;
 553 }
 554
 555 void kvm_free_physmem(struct kvm *kvm)
 556 {
 557         struct kvm_memslots *slots = kvm->memslots;
 558         struct kvm_memory_slot *memslot;
 559
 560         kvm_for_each_memslot(memslot, slots)
 561                 kvm_free_physmem_slot(kvm, memslot, NULL);
 562
 563         kfree(kvm->memslots);
 564 }
 565
 566 static void kvm_destroy_devices(struct kvm *kvm)
 567 {
 568         struct list_head *node, *tmp;
 569
 570         list_for_each_safe(node, tmp, &kvm->devices) {
 571                 struct kvm_device *dev =
 572                         list_entry(node, struct kvm_device, vm_node);
 573
 574                 list_del(node);
 575                 dev->ops->destroy(dev);
 576         }
 577 }
 578
 579 static void kvm_destroy_vm(struct kvm *kvm)
 580 {
 581         int i;
 582         struct mm_struct *mm = kvm->mm;
 583
 584         kvm_arch_sync_events(kvm);
 585         spin_lock(&kvm_lock);
 586         list_del(&kvm->vm_list);
 587         spin_unlock(&kvm_lock);
 588         kvm_free_irq_routing(kvm);
 589         for (i = 0; i < KVM_NR_BUSES; i++)
 590                 kvm_io_bus_destroy(kvm->buses[i]);
 591         kvm_coalesced_mmio_free(kvm);
 592 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 593         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 594 #else
 595         kvm_arch_flush_shadow_all(kvm);
 596 #endif
 597         kvm_arch_destroy_vm(kvm);
 598         kvm_destroy_devices(kvm);
 599         kvm_free_physmem(kvm);
 600         cleanup_srcu_struct(&kvm->srcu);
 601         kvm_arch_free_vm(kvm);
 602         hardware_disable_all();
 603         mmdrop(mm);
 604 }
 605
 606 void kvm_get_kvm(struct kvm *kvm)
 607 {
 608         atomic_inc(&kvm->users_count);
 609 }
 610 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 611
 612 void kvm_put_kvm(struct kvm *kvm)
 613 {
 614         if (atomic_dec_and_test(&kvm->users_count))
 615                 kvm_destroy_vm(kvm);
 616 }
 617 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 618
 619
 620 static int kvm_vm_release(struct inode *inode, struct file *filp)
 621 {
 622         struct kvm *kvm = filp->private_data;
 623
 624         kvm_irqfd_release(kvm);
 625
 626         kvm_put_kvm(kvm);
 627         return 0;
 628 }
 629
 630 /*
 631  * Allocation size is twice as large as the actual dirty bitmap size.
 632  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 633  */
 634 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 635 {
 636 #ifndef CONFIG_S390
 637         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 638
 639         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 640         if (!memslot->dirty_bitmap)
 641                 return -ENOMEM;
 642
 643 #endif /* !CONFIG_S390 */
 644         return 0;
 645 }
 646
 647 static int cmp_memslot(const void *slot1, const void *slot2)
 648 {
 649         struct kvm_memory_slot *s1, *s2;
 650
 651         s1 = (struct kvm_memory_slot *)slot1;
 652         s2 = (struct kvm_memory_slot *)slot2;
 653
 654         if (s1->npages < s2->npages)
 655                 return 1;
 656         if (s1->npages > s2->npages)
 657                 return -1;
 658
 659         return 0;
 660 }
 661
 662 /*
 663  * Sort the memslots base on its size, so the larger slots
 664  * will get better fit.
 665  */
 666 static void sort_memslots(struct kvm_memslots *slots)
 667 {
 668         int i;
 669
 670         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 671               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 672
 673         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 674                 slots->id_to_index[slots->memslots[i].id] = i;
 675 }
 676
 677 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
 678                      u64 last_generation)
 679 {
 680         if (new) {
 681                 int id = new->id;
 682                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 683                 unsigned long npages = old->npages;
 684
 685                 *old = *new;
 686                 if (new->npages != npages)
 687                         sort_memslots(slots);
 688         }
 689
 690         slots->generation = last_generation + 1;
 691 }
 692
 693 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 694 {
 695         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 696
 697 #ifdef KVM_CAP_READONLY_MEM
 698         valid_flags |= KVM_MEM_READONLY;
 699 #endif
 700
 701         if (mem->flags & ~valid_flags)
 702                 return -EINVAL;
 703
 704         return 0;
 705 }
 706
 707 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 708                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 709 {
 710         struct kvm_memslots *old_memslots = kvm->memslots;
 711
 712         update_memslots(slots, new, kvm->memslots->generation);
 713         rcu_assign_pointer(kvm->memslots, slots);
 714         synchronize_srcu_expedited(&kvm->srcu);
 715
 716         kvm_arch_memslots_updated(kvm);
 717
 718         return old_memslots;
 719 }
 720
 721 /*
 722  * Allocate some memory and give it an address in the guest physical address
 723  * space.
 724  *
 725  * Discontiguous memory is allowed, mostly for framebuffers.
 726  *
 727  * Must be called holding mmap_sem for write.
 728  */
 729 int __kvm_set_memory_region(struct kvm *kvm,
 730                             struct kvm_userspace_memory_region *mem)
 731 {
 732         int r;
 733         gfn_t base_gfn;
 734         unsigned long npages;
 735         struct kvm_memory_slot *slot;
 736         struct kvm_memory_slot old, new;
 737         struct kvm_memslots *slots = NULL, *old_memslots;
 738         enum kvm_mr_change change;
 739
 740         r = check_memory_region_flags(mem);
 741         if (r)
 742                 goto out;
 743
 744         r = -EINVAL;
 745         /* General sanity checks */
 746         if (mem->memory_size & (PAGE_SIZE - 1))
 747                 goto out;
 748         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 749                 goto out;
 750         /* We can read the guest memory with __xxx_user() later on. */
 751         if ((mem->slot < KVM_USER_MEM_SLOTS) &&
 752             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 753              !access_ok(VERIFY_WRITE,
 754                         (void __user *)(unsigned long)mem->userspace_addr,
 755                         mem->memory_size)))
 756                 goto out;
 757         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 758                 goto out;
 759         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 760                 goto out;
 761
 762         slot = id_to_memslot(kvm->memslots, mem->slot);
 763         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 764         npages = mem->memory_size >> PAGE_SHIFT;
 765
 766         r = -EINVAL;
 767         if (npages > KVM_MEM_MAX_NR_PAGES)
 768                 goto out;
 769
 770         if (!npages)
 771                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 772
 773         new = old = *slot;
 774
 775         new.id = mem->slot;
 776         new.base_gfn = base_gfn;
 777         new.npages = npages;
 778         new.flags = mem->flags;
 779
 780         r = -EINVAL;
 781         if (npages) {
 782                 if (!old.npages)
 783                         change = KVM_MR_CREATE;
 784                 else { /* Modify an existing slot. */
 785                         if ((mem->userspace_addr != old.userspace_addr) ||
 786                             (npages != old.npages) ||
 787                             ((new.flags ^ old.flags) & KVM_MEM_READONLY))
 788                                 goto out;
 789
 790                         if (base_gfn != old.base_gfn)
 791                                 change = KVM_MR_MOVE;
 792                         else if (new.flags != old.flags)
 793                                 change = KVM_MR_FLAGS_ONLY;
 794                         else { /* Nothing to change. */
 795                                 r = 0;
 796                                 goto out;
 797                         }
 798                 }
 799         } else if (old.npages) {
 800                 change = KVM_MR_DELETE;
 801         } else /* Modify a non-existent slot: disallowed. */
 802                 goto out;
 803
 804         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 805                 /* Check for overlaps */
 806                 r = -EEXIST;
 807                 kvm_for_each_memslot(slot, kvm->memslots) {
 808                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 809                             (slot->id == mem->slot))
 810                                 continue;
 811                         if (!((base_gfn + npages <= slot->base_gfn) ||
 812                               (base_gfn >= slot->base_gfn + slot->npages)))
 813                                 goto out;
 814                 }
 815         }
 816
 817         /* Free page dirty bitmap if unneeded */
 818         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 819                 new.dirty_bitmap = NULL;
 820
 821         r = -ENOMEM;
 822         if (change == KVM_MR_CREATE) {
 823                 new.userspace_addr = mem->userspace_addr;
 824
 825                 if (kvm_arch_create_memslot(kvm, &new, npages))
 826                         goto out_free;
 827         }
 828
 829         /* Allocate page dirty bitmap if needed */
 830         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 831                 if (kvm_create_dirty_bitmap(&new) < 0)
 832                         goto out_free;
 833         }
 834
 835         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 836                 r = -ENOMEM;
 837                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 838                                 GFP_KERNEL);
 839                 if (!slots)
 840                         goto out_free;
 841                 slot = id_to_memslot(slots, mem->slot);
 842                 slot->flags |= KVM_MEMSLOT_INVALID;
 843
 844                 old_memslots = install_new_memslots(kvm, slots, NULL);
 845
 846                 /* slot was deleted or moved, clear iommu mapping */
 847                 kvm_iommu_unmap_pages(kvm, &old);
 848                 /* From this point no new shadow pages pointing to a deleted,
 849                  * or moved, memslot will be created.
 850                  *
 851                  * validation of sp->gfn happens in:
 852                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 853                  *      - kvm_is_visible_gfn (mmu_check_roots)
 854                  */
 855                 kvm_arch_flush_shadow_memslot(kvm, slot);
 856                 slots = old_memslots;
 857         }
 858
 859         r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
 860         if (r)
 861                 goto out_slots;
 862
 863         r = -ENOMEM;
 864         /*
 865          * We can re-use the old_memslots from above, the only difference
 866          * from the currently installed memslots is the invalid flag.  This
 867          * will get overwritten by update_memslots anyway.
 868          */
 869         if (!slots) {
 870                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 871                                 GFP_KERNEL);
 872                 if (!slots)
 873                         goto out_free;
 874         }
 875
 876         /* actual memory is freed via old in kvm_free_physmem_slot below */
 877         if (change == KVM_MR_DELETE) {
 878                 new.dirty_bitmap = NULL;
 879                 memset(&new.arch, 0, sizeof(new.arch));
 880         }
 881
 882         old_memslots = install_new_memslots(kvm, slots, &new);
 883
 884         kvm_arch_commit_memory_region(kvm, mem, &old, change);
 885
 886         kvm_free_physmem_slot(kvm, &old, &new);
 887         kfree(old_memslots);
 888
 889         /*
 890          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 891          * un-mapped and re-mapped if their base changes.  Since base change
 892          * unmapping is handled above with slot deletion, mapping alone is
 893          * needed here.  Anything else the iommu might care about for existing
 894          * slots (size changes, userspace addr changes and read-only flag
 895          * changes) is disallowed above, so any other attribute changes getting
 896          * here can be skipped.
 897          */
 898         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 899                 r = kvm_iommu_map_pages(kvm, &new);
 900                 return r;
 901         }
 902
 903         return 0;
 904
 905 out_slots:
 906         kfree(slots);
 907 out_free:
 908         kvm_free_physmem_slot(kvm, &new, &old);
 909 out:
 910         return r;
 911 }
 912 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 913
 914 int kvm_set_memory_region(struct kvm *kvm,
 915                           struct kvm_userspace_memory_region *mem)
 916 {
 917         int r;
 918
 919         mutex_lock(&kvm->slots_lock);
 920         r = __kvm_set_memory_region(kvm, mem);
 921         mutex_unlock(&kvm->slots_lock);
 922         return r;
 923 }
 924 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 925
 926 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 927                                    struct kvm_userspace_memory_region *mem)
 928 {
 929         if (mem->slot >= KVM_USER_MEM_SLOTS)
 930                 return -EINVAL;
 931         return kvm_set_memory_region(kvm, mem);
 932 }
 933
 934 int kvm_get_dirty_log(struct kvm *kvm,
 935                         struct kvm_dirty_log *log, int *is_dirty)
 936 {
 937         struct kvm_memory_slot *memslot;
 938         int r, i;
 939         unsigned long n;
 940         unsigned long any = 0;
 941
 942         r = -EINVAL;
 943         if (log->slot >= KVM_USER_MEM_SLOTS)
 944                 goto out;
 945
 946         memslot = id_to_memslot(kvm->memslots, log->slot);
 947         r = -ENOENT;
 948         if (!memslot->dirty_bitmap)
 949                 goto out;
 950
 951         n = kvm_dirty_bitmap_bytes(memslot);
 952
 953         for (i = 0; !any && i < n/sizeof(long); ++i)
 954                 any = memslot->dirty_bitmap[i];
 955
 956         r = -EFAULT;
 957         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 958                 goto out;
 959
 960         if (any)
 961                 *is_dirty = 1;
 962
 963         r = 0;
 964 out:
 965         return r;
 966 }
 967
 968 bool kvm_largepages_enabled(void)
 969 {
 970         return largepages_enabled;
 971 }
 972
 973 void kvm_disable_largepages(void)
 974 {
 975         largepages_enabled = false;
 976 }
 977 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 978
 979 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 980 {
 981         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 982 }
 983 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 984
 985 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 986 {
 987         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 988
 989         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
 990               memslot->flags & KVM_MEMSLOT_INVALID)
 991                 return 0;
 992
 993         return 1;
 994 }
 995 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 996
 997 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 998 {
 999         struct vm_area_struct *vma;
1000         unsigned long addr, size;
1001
1002         size = PAGE_SIZE;
1003
1004         addr = gfn_to_hva(kvm, gfn);
1005         if (kvm_is_error_hva(addr))
1006                 return PAGE_SIZE;
1007
1008         down_read(&current->mm->mmap_sem);
1009         vma = find_vma(current->mm, addr);
1010         if (!vma)
1011                 goto out;
1012
1013         size = vma_kernel_pagesize(vma);
1014
1015 out:
1016         up_read(&current->mm->mmap_sem);
1017
1018         return size;
1019 }
1020
1021 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1022 {
1023         return slot->flags & KVM_MEM_READONLY;
1024 }
1025
1026 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1027                                        gfn_t *nr_pages, bool write)
1028 {
1029         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1030                 return KVM_HVA_ERR_BAD;
1031
1032         if (memslot_is_readonly(slot) && write)
1033                 return KVM_HVA_ERR_RO_BAD;
1034
1035         if (nr_pages)
1036                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1037
1038         return __gfn_to_hva_memslot(slot, gfn);
1039 }
1040
1041 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1042                                      gfn_t *nr_pages)
1043 {
1044         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1045 }
1046
1047 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1048                                  gfn_t gfn)
1049 {
1050         return gfn_to_hva_many(slot, gfn, NULL);
1051 }
1052 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1053
1054 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1055 {
1056         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1057 }
1058 EXPORT_SYMBOL_GPL(gfn_to_hva);
1059
1060 /*
1061  * If writable is set to false, the hva returned by this function is only
1062  * allowed to be read.
1063  */
1064 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1065 {
1066         struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1067         unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1068
1069         if (!kvm_is_error_hva(hva) && writable)
1070                 *writable = !memslot_is_readonly(slot);
1071
1072         return hva;
1073 }
1074
1075 static int kvm_read_hva(void *data, void __user *hva, int len)
1076 {
1077         return __copy_from_user(data, hva, len);
1078 }
1079
1080 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1081 {
1082         return __copy_from_user_inatomic(data, hva, len);
1083 }
1084
1085 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1086         unsigned long start, int write, struct page **page)
1087 {
1088         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1089
1090         if (write)
1091                 flags |= FOLL_WRITE;
1092
1093         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1094 }
1095
1096 static inline int check_user_page_hwpoison(unsigned long addr)
1097 {
1098         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1099
1100         rc = __get_user_pages(current, current->mm, addr, 1,
1101                               flags, NULL, NULL, NULL);
1102         return rc == -EHWPOISON;
1103 }
1104
1105 /*
1106  * The atomic path to get the writable pfn which will be stored in @pfn,
1107  * true indicates success, otherwise false is returned.
1108  */
1109 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1110                             bool write_fault, bool *writable, pfn_t *pfn)
1111 {
1112         struct page *page[1];
1113         int npages;
1114
1115         if (!(async || atomic))
1116                 return false;
1117
1118         /*
1119          * Fast pin a writable pfn only if it is a write fault request
1120          * or the caller allows to map a writable pfn for a read fault
1121          * request.
1122          */
1123         if (!(write_fault || writable))
1124                 return false;
1125
1126         npages = __get_user_pages_fast(addr, 1, 1, page);
1127         if (npages == 1) {
1128                 *pfn = page_to_pfn(page[0]);
1129
1130                 if (writable)
1131                         *writable = true;
1132                 return true;
1133         }
1134
1135         return false;
1136 }
1137
1138 /*
1139  * The slow path to get the pfn of the specified host virtual address,
1140  * 1 indicates success, -errno is returned if error is detected.
1141  */
1142 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1143                            bool *writable, pfn_t *pfn)
1144 {
1145         struct page *page[1];
1146         int npages = 0;
1147
1148         might_sleep();
1149
1150         if (writable)
1151                 *writable = write_fault;
1152
1153         if (async) {
1154                 down_read(&current->mm->mmap_sem);
1155                 npages = get_user_page_nowait(current, current->mm,
1156                                               addr, write_fault, page);
1157                 up_read(&current->mm->mmap_sem);
1158         } else
1159                 npages = get_user_pages_fast(addr, 1, write_fault,
1160                                              page);
1161         if (npages != 1)
1162                 return npages;
1163
1164         /* map read fault as writable if possible */
1165         if (unlikely(!write_fault) && writable) {
1166                 struct page *wpage[1];
1167
1168                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1169                 if (npages == 1) {
1170                         *writable = true;
1171                         put_page(page[0]);
1172                         page[0] = wpage[0];
1173                 }
1174
1175                 npages = 1;
1176         }
1177         *pfn = page_to_pfn(page[0]);
1178         return npages;
1179 }
1180
1181 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1182 {
1183         if (unlikely(!(vma->vm_flags & VM_READ)))
1184                 return false;
1185
1186         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1187                 return false;
1188
1189         return true;
1190 }
1191
1192 /*
1193  * Pin guest page in memory and return its pfn.
1194  * @addr: host virtual address which maps memory to the guest
1195  * @atomic: whether this function can sleep
1196  * @async: whether this function need to wait IO complete if the
1197  *         host page is not in the memory
1198  * @write_fault: whether we should get a writable host page
1199  * @writable: whether it allows to map a writable host page for !@write_fault
1200  *
1201  * The function will map a writable host page for these two cases:
1202  * 1): @write_fault = true
1203  * 2): @write_fault = false && @writable, @writable will tell the caller
1204  *     whether the mapping is writable.
1205  */
1206 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1207                         bool write_fault, bool *writable)
1208 {
1209         struct vm_area_struct *vma;
1210         pfn_t pfn = 0;
1211         int npages;
1212
1213         /* we can do it either atomically or asynchronously, not both */
1214         BUG_ON(atomic && async);
1215
1216         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1217                 return pfn;
1218
1219         if (atomic)
1220                 return KVM_PFN_ERR_FAULT;
1221
1222         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1223         if (npages == 1)
1224                 return pfn;
1225
1226         down_read(&current->mm->mmap_sem);
1227         if (npages == -EHWPOISON ||
1228               (!async && check_user_page_hwpoison(addr))) {
1229                 pfn = KVM_PFN_ERR_HWPOISON;
1230                 goto exit;
1231         }
1232
1233         vma = find_vma_intersection(current->mm, addr, addr + 1);
1234
1235         if (vma == NULL)
1236                 pfn = KVM_PFN_ERR_FAULT;
1237         else if ((vma->vm_flags & VM_PFNMAP)) {
1238                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1239                         vma->vm_pgoff;
1240                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1241         } else {
1242                 if (async && vma_is_valid(vma, write_fault))
1243                         *async = true;
1244                 pfn = KVM_PFN_ERR_FAULT;
1245         }
1246 exit:
1247         up_read(&current->mm->mmap_sem);
1248         return pfn;
1249 }
1250
1251 static pfn_t
1252 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1253                      bool *async, bool write_fault, bool *writable)
1254 {
1255         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1256
1257         if (addr == KVM_HVA_ERR_RO_BAD)
1258                 return KVM_PFN_ERR_RO_FAULT;
1259
1260         if (kvm_is_error_hva(addr))
1261                 return KVM_PFN_NOSLOT;
1262
1263         /* Do not map writable pfn in the readonly memslot. */
1264         if (writable && memslot_is_readonly(slot)) {
1265                 *writable = false;
1266                 writable = NULL;
1267         }
1268
1269         return hva_to_pfn(addr, atomic, async, write_fault,
1270                           writable);
1271 }
1272
1273 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1274                           bool write_fault, bool *writable)
1275 {
1276         struct kvm_memory_slot *slot;
1277
1278         if (async)
1279                 *async = false;
1280
1281         slot = gfn_to_memslot(kvm, gfn);
1282
1283         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1284                                     writable);
1285 }
1286
1287 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1288 {
1289         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1290 }
1291 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1292
1293 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1294                        bool write_fault, bool *writable)
1295 {
1296         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1297 }
1298 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1299
1300 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1301 {
1302         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1303 }
1304 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1305
1306 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1307                       bool *writable)
1308 {
1309         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1310 }
1311 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1312
1313 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1314 {
1315         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1316 }
1317
1318 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1319 {
1320         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1321 }
1322 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1323
1324 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1325                                                                   int nr_pages)
1326 {
1327         unsigned long addr;
1328         gfn_t entry;
1329
1330         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1331         if (kvm_is_error_hva(addr))
1332                 return -1;
1333
1334         if (entry < nr_pages)
1335                 return 0;
1336
1337         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1338 }
1339 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1340
1341 static struct page *kvm_pfn_to_page(pfn_t pfn)
1342 {
1343         if (is_error_noslot_pfn(pfn))
1344                 return KVM_ERR_PTR_BAD_PAGE;
1345
1346         if (kvm_is_mmio_pfn(pfn)) {
1347                 WARN_ON(1);
1348                 return KVM_ERR_PTR_BAD_PAGE;
1349         }
1350
1351         return pfn_to_page(pfn);
1352 }
1353
1354 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1355 {
1356         pfn_t pfn;
1357
1358         pfn = gfn_to_pfn(kvm, gfn);
1359
1360         return kvm_pfn_to_page(pfn);
1361 }
1362
1363 EXPORT_SYMBOL_GPL(gfn_to_page);
1364
1365 void kvm_release_page_clean(struct page *page)
1366 {
1367         WARN_ON(is_error_page(page));
1368
1369         kvm_release_pfn_clean(page_to_pfn(page));
1370 }
1371 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1372
1373 void kvm_release_pfn_clean(pfn_t pfn)
1374 {
1375         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1376                 put_page(pfn_to_page(pfn));
1377 }
1378 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1379
1380 void kvm_release_page_dirty(struct page *page)
1381 {
1382         WARN_ON(is_error_page(page));
1383
1384         kvm_release_pfn_dirty(page_to_pfn(page));
1385 }
1386 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1387
1388 void kvm_release_pfn_dirty(pfn_t pfn)
1389 {
1390         kvm_set_pfn_dirty(pfn);
1391         kvm_release_pfn_clean(pfn);
1392 }
1393 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1394
1395 void kvm_set_page_dirty(struct page *page)
1396 {
1397         kvm_set_pfn_dirty(page_to_pfn(page));
1398 }
1399 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1400
1401 void kvm_set_pfn_dirty(pfn_t pfn)
1402 {
1403         if (!kvm_is_mmio_pfn(pfn)) {
1404                 struct page *page = pfn_to_page(pfn);
1405                 if (!PageReserved(page))
1406                         SetPageDirty(page);
1407         }
1408 }
1409 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1410
1411 void kvm_set_pfn_accessed(pfn_t pfn)
1412 {
1413         if (!kvm_is_mmio_pfn(pfn))
1414                 mark_page_accessed(pfn_to_page(pfn));
1415 }
1416 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1417
1418 void kvm_get_pfn(pfn_t pfn)
1419 {
1420         if (!kvm_is_mmio_pfn(pfn))
1421                 get_page(pfn_to_page(pfn));
1422 }
1423 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1424
1425 static int next_segment(unsigned long len, int offset)
1426 {
1427         if (len > PAGE_SIZE - offset)
1428                 return PAGE_SIZE - offset;
1429         else
1430                 return len;
1431 }
1432
1433 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1434                         int len)
1435 {
1436         int r;
1437         unsigned long addr;
1438
1439         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1440         if (kvm_is_error_hva(addr))
1441                 return -EFAULT;
1442         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1443         if (r)
1444                 return -EFAULT;
1445         return 0;
1446 }
1447 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1448
1449 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1450 {
1451         gfn_t gfn = gpa >> PAGE_SHIFT;
1452         int seg;
1453         int offset = offset_in_page(gpa);
1454         int ret;
1455
1456         while ((seg = next_segment(len, offset)) != 0) {
1457                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1458                 if (ret < 0)
1459                         return ret;
1460                 offset = 0;
1461                 len -= seg;
1462                 data += seg;
1463                 ++gfn;
1464         }
1465         return 0;
1466 }
1467 EXPORT_SYMBOL_GPL(kvm_read_guest);
1468
1469 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1470                           unsigned long len)
1471 {
1472         int r;
1473         unsigned long addr;
1474         gfn_t gfn = gpa >> PAGE_SHIFT;
1475         int offset = offset_in_page(gpa);
1476
1477         addr = gfn_to_hva_prot(kvm, gfn, NULL);
1478         if (kvm_is_error_hva(addr))
1479                 return -EFAULT;
1480         pagefault_disable();
1481         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1482         pagefault_enable();
1483         if (r)
1484                 return -EFAULT;
1485         return 0;
1486 }
1487 EXPORT_SYMBOL(kvm_read_guest_atomic);
1488
1489 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1490                          int offset, int len)
1491 {
1492         int r;
1493         unsigned long addr;
1494
1495         addr = gfn_to_hva(kvm, gfn);
1496         if (kvm_is_error_hva(addr))
1497                 return -EFAULT;
1498         r = __copy_to_user((void __user *)addr + offset, data, len);
1499         if (r)
1500                 return -EFAULT;
1501         mark_page_dirty(kvm, gfn);
1502         return 0;
1503 }
1504 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1505
1506 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1507                     unsigned long len)
1508 {
1509         gfn_t gfn = gpa >> PAGE_SHIFT;
1510         int seg;
1511         int offset = offset_in_page(gpa);
1512         int ret;
1513
1514         while ((seg = next_segment(len, offset)) != 0) {
1515                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1516                 if (ret < 0)
1517                         return ret;
1518                 offset = 0;
1519                 len -= seg;
1520                 data += seg;
1521                 ++gfn;
1522         }
1523         return 0;
1524 }
1525
1526 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1527                               gpa_t gpa, unsigned long len)
1528 {
1529         struct kvm_memslots *slots = kvm_memslots(kvm);
1530         int offset = offset_in_page(gpa);
1531         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1532         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1533         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1534         gfn_t nr_pages_avail;
1535
1536         ghc->gpa = gpa;
1537         ghc->generation = slots->generation;
1538         ghc->len = len;
1539         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1540         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1541         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1542                 ghc->hva += offset;
1543         } else {
1544                 /*
1545                  * If the requested region crosses two memslots, we still
1546                  * verify that the entire region is valid here.
1547                  */
1548                 while (start_gfn <= end_gfn) {
1549                         ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1550                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1551                                                    &nr_pages_avail);
1552                         if (kvm_is_error_hva(ghc->hva))
1553                                 return -EFAULT;
1554                         start_gfn += nr_pages_avail;
1555                 }
1556                 /* Use the slow path for cross page reads and writes. */
1557                 ghc->memslot = NULL;
1558         }
1559         return 0;
1560 }
1561 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1562
1563 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1564                            void *data, unsigned long len)
1565 {
1566         struct kvm_memslots *slots = kvm_memslots(kvm);
1567         int r;
1568
1569         BUG_ON(len > ghc->len);
1570
1571         if (slots->generation != ghc->generation)
1572                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1573
1574         if (unlikely(!ghc->memslot))
1575                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1576
1577         if (kvm_is_error_hva(ghc->hva))
1578                 return -EFAULT;
1579
1580         r = __copy_to_user((void __user *)ghc->hva, data, len);
1581         if (r)
1582                 return -EFAULT;
1583         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1584
1585         return 0;
1586 }
1587 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1588
1589 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1590                            void *data, unsigned long len)
1591 {
1592         struct kvm_memslots *slots = kvm_memslots(kvm);
1593         int r;
1594
1595         BUG_ON(len > ghc->len);
1596
1597         if (slots->generation != ghc->generation)
1598                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1599
1600         if (unlikely(!ghc->memslot))
1601                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1602
1603         if (kvm_is_error_hva(ghc->hva))
1604                 return -EFAULT;
1605
1606         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1607         if (r)
1608                 return -EFAULT;
1609
1610         return 0;
1611 }
1612 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1613
1614 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1615 {
1616         const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1617
1618         return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1619 }
1620 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1621
1622 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1623 {
1624         gfn_t gfn = gpa >> PAGE_SHIFT;
1625         int seg;
1626         int offset = offset_in_page(gpa);
1627         int ret;
1628
1629         while ((seg = next_segment(len, offset)) != 0) {
1630                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1631                 if (ret < 0)
1632                         return ret;
1633                 offset = 0;
1634                 len -= seg;
1635                 ++gfn;
1636         }
1637         return 0;
1638 }
1639 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1640
1641 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1642                              gfn_t gfn)
1643 {
1644         if (memslot && memslot->dirty_bitmap) {
1645                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1646
1647                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1648         }
1649 }
1650
1651 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1652 {
1653         struct kvm_memory_slot *memslot;
1654
1655         memslot = gfn_to_memslot(kvm, gfn);
1656         mark_page_dirty_in_slot(kvm, memslot, gfn);
1657 }
1658
1659 /*
1660  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1661  */
1662 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1663 {
1664         DEFINE_WAIT(wait);
1665
1666         for (;;) {
1667                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1668
1669                 if (kvm_arch_vcpu_runnable(vcpu)) {
1670                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1671                         break;
1672                 }
1673                 if (kvm_cpu_has_pending_timer(vcpu))
1674                         break;
1675                 if (signal_pending(current))
1676                         break;
1677
1678                 schedule();
1679         }
1680
1681         finish_wait(&vcpu->wq, &wait);
1682 }
1683
1684 #ifndef CONFIG_S390
1685 /*
1686  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1687  */
1688 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1689 {
1690         int me;
1691         int cpu = vcpu->cpu;
1692         wait_queue_head_t *wqp;
1693
1694         wqp = kvm_arch_vcpu_wq(vcpu);
1695         if (waitqueue_active(wqp)) {
1696                 wake_up_interruptible(wqp);
1697                 ++vcpu->stat.halt_wakeup;
1698         }
1699
1700         me = get_cpu();
1701         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1702                 if (kvm_arch_vcpu_should_kick(vcpu))
1703                         smp_send_reschedule(cpu);
1704         put_cpu();
1705 }
1706 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
1707 #endif /* !CONFIG_S390 */
1708
1709 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1710 {
1711         struct pid *pid;
1712         struct task_struct *task = NULL;
1713         bool ret = false;
1714
1715         rcu_read_lock();
1716         pid = rcu_dereference(target->pid);
1717         if (pid)
1718                 task = get_pid_task(target->pid, PIDTYPE_PID);
1719         rcu_read_unlock();
1720         if (!task)
1721                 return ret;
1722         if (task->flags & PF_VCPU) {
1723                 put_task_struct(task);
1724                 return ret;
1725         }
1726         ret = yield_to(task, 1);
1727         put_task_struct(task);
1728
1729         return ret;
1730 }
1731 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1732
1733 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1734 /*
1735  * Helper that checks whether a VCPU is eligible for directed yield.
1736  * Most eligible candidate to yield is decided by following heuristics:
1737  *
1738  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1739  *  (preempted lock holder), indicated by @in_spin_loop.
1740  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1741  *
1742  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1743  *  chance last time (mostly it has become eligible now since we have probably
1744  *  yielded to lockholder in last iteration. This is done by toggling
1745  *  @dy_eligible each time a VCPU checked for eligibility.)
1746  *
1747  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1748  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1749  *  burning. Giving priority for a potential lock-holder increases lock
1750  *  progress.
1751  *
1752  *  Since algorithm is based on heuristics, accessing another VCPU data without
1753  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1754  *  and continue with next VCPU and so on.
1755  */
1756 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1757 {
1758         bool eligible;
1759
1760         eligible = !vcpu->spin_loop.in_spin_loop ||
1761                         (vcpu->spin_loop.in_spin_loop &&
1762                          vcpu->spin_loop.dy_eligible);
1763
1764         if (vcpu->spin_loop.in_spin_loop)
1765                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1766
1767         return eligible;
1768 }
1769 #endif
1770
1771 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1772 {
1773         struct kvm *kvm = me->kvm;
1774         struct kvm_vcpu *vcpu;
1775         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1776         int yielded = 0;
1777         int try = 3;
1778         int pass;
1779         int i;
1780
1781         kvm_vcpu_set_in_spin_loop(me, true);
1782         /*
1783          * We boost the priority of a VCPU that is runnable but not
1784          * currently running, because it got preempted by something
1785          * else and called schedule in __vcpu_run.  Hopefully that
1786          * VCPU is holding the lock that we need and will release it.
1787          * We approximate round-robin by starting at the last boosted VCPU.
1788          */
1789         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1790                 kvm_for_each_vcpu(i, vcpu, kvm) {
1791                         if (!pass && i <= last_boosted_vcpu) {
1792                                 i = last_boosted_vcpu;
1793                                 continue;
1794                         } else if (pass && i > last_boosted_vcpu)
1795                                 break;
1796                         if (!ACCESS_ONCE(vcpu->preempted))
1797                                 continue;
1798                         if (vcpu == me)
1799                                 continue;
1800                         if (waitqueue_active(&vcpu->wq))
1801                                 continue;
1802                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1803                                 continue;
1804
1805                         yielded = kvm_vcpu_yield_to(vcpu);
1806                         if (yielded > 0) {
1807                                 kvm->last_boosted_vcpu = i;
1808                                 break;
1809                         } else if (yielded < 0) {
1810                                 try--;
1811                                 if (!try)
1812                                         break;
1813                         }
1814                 }
1815         }
1816         kvm_vcpu_set_in_spin_loop(me, false);
1817
1818         /* Ensure vcpu is not eligible during next spinloop */
1819         kvm_vcpu_set_dy_eligible(me, false);
1820 }
1821 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1822
1823 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1824 {
1825         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1826         struct page *page;
1827
1828         if (vmf->pgoff == 0)
1829                 page = virt_to_page(vcpu->run);
1830 #ifdef CONFIG_X86
1831         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1832                 page = virt_to_page(vcpu->arch.pio_data);
1833 #endif
1834 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1835         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1836                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1837 #endif
1838         else
1839                 return kvm_arch_vcpu_fault(vcpu, vmf);
1840         get_page(page);
1841         vmf->page = page;
1842         return 0;
1843 }
1844
1845 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1846         .fault = kvm_vcpu_fault,
1847 };
1848
1849 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1850 {
1851         vma->vm_ops = &kvm_vcpu_vm_ops;
1852         return 0;
1853 }
1854
1855 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1856 {
1857         struct kvm_vcpu *vcpu = filp->private_data;
1858
1859         kvm_put_kvm(vcpu->kvm);
1860         return 0;
1861 }
1862
1863 static struct file_operations kvm_vcpu_fops = {
1864         .release        = kvm_vcpu_release,
1865         .unlocked_ioctl = kvm_vcpu_ioctl,
1866 #ifdef CONFIG_COMPAT
1867         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1868 #endif
1869         .mmap           = kvm_vcpu_mmap,
1870         .llseek         = noop_llseek,
1871 };
1872
1873 /*
1874  * Allocates an inode for the vcpu.
1875  */
1876 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1877 {
1878         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
1879 }
1880
1881 /*
1882  * Creates some virtual cpus.  Good luck creating more than one.
1883  */
1884 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1885 {
1886         int r;
1887         struct kvm_vcpu *vcpu, *v;
1888
1889         if (id >= KVM_MAX_VCPUS)
1890                 return -EINVAL;
1891
1892         vcpu = kvm_arch_vcpu_create(kvm, id);
1893         if (IS_ERR(vcpu))
1894                 return PTR_ERR(vcpu);
1895
1896         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1897
1898         r = kvm_arch_vcpu_setup(vcpu);
1899         if (r)
1900                 goto vcpu_destroy;
1901
1902         mutex_lock(&kvm->lock);
1903         if (!kvm_vcpu_compatible(vcpu)) {
1904                 r = -EINVAL;
1905                 goto unlock_vcpu_destroy;
1906         }
1907         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1908                 r = -EINVAL;
1909                 goto unlock_vcpu_destroy;
1910         }
1911
1912         kvm_for_each_vcpu(r, v, kvm)
1913                 if (v->vcpu_id == id) {
1914                         r = -EEXIST;
1915                         goto unlock_vcpu_destroy;
1916                 }
1917
1918         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1919
1920         /* Now it's all set up, let userspace reach it */
1921         kvm_get_kvm(kvm);
1922         r = create_vcpu_fd(vcpu);
1923         if (r < 0) {
1924                 kvm_put_kvm(kvm);
1925                 goto unlock_vcpu_destroy;
1926         }
1927
1928         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1929         smp_wmb();
1930         atomic_inc(&kvm->online_vcpus);
1931
1932         mutex_unlock(&kvm->lock);
1933         kvm_arch_vcpu_postcreate(vcpu);
1934         return r;
1935
1936 unlock_vcpu_destroy:
1937         mutex_unlock(&kvm->lock);
1938 vcpu_destroy:
1939         kvm_arch_vcpu_destroy(vcpu);
1940         return r;
1941 }
1942
1943 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1944 {
1945         if (sigset) {
1946                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1947                 vcpu->sigset_active = 1;
1948                 vcpu->sigset = *sigset;
1949         } else
1950                 vcpu->sigset_active = 0;
1951         return 0;
1952 }
1953
1954 static long kvm_vcpu_ioctl(struct file *filp,
1955                            unsigned int ioctl, unsigned long arg)
1956 {
1957         struct kvm_vcpu *vcpu = filp->private_data;
1958         void __user *argp = (void __user *)arg;
1959         int r;
1960         struct kvm_fpu *fpu = NULL;
1961         struct kvm_sregs *kvm_sregs = NULL;
1962
1963         if (vcpu->kvm->mm != current->mm)
1964                 return -EIO;
1965
1966 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
1967         /*
1968          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1969          * so vcpu_load() would break it.
1970          */
1971         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1972                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1973 #endif
1974
1975
1976         r = vcpu_load(vcpu);
1977         if (r)
1978                 return r;
1979         switch (ioctl) {
1980         case KVM_RUN:
1981                 r = -EINVAL;
1982                 if (arg)
1983                         goto out;
1984                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1985                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1986                 break;
1987         case KVM_GET_REGS: {
1988                 struct kvm_regs *kvm_regs;
1989
1990                 r = -ENOMEM;
1991                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1992                 if (!kvm_regs)
1993                         goto out;
1994                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1995                 if (r)
1996                         goto out_free1;
1997                 r = -EFAULT;
1998                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1999                         goto out_free1;
2000                 r = 0;
2001 out_free1:
2002                 kfree(kvm_regs);
2003                 break;
2004         }
2005         case KVM_SET_REGS: {
2006                 struct kvm_regs *kvm_regs;
2007
2008                 r = -ENOMEM;
2009                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2010                 if (IS_ERR(kvm_regs)) {
2011                         r = PTR_ERR(kvm_regs);
2012                         goto out;
2013                 }
2014                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2015                 kfree(kvm_regs);
2016                 break;
2017         }
2018         case KVM_GET_SREGS: {
2019                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2020                 r = -ENOMEM;
2021                 if (!kvm_sregs)
2022                         goto out;
2023                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2024                 if (r)
2025                         goto out;
2026                 r = -EFAULT;
2027                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2028                         goto out;
2029                 r = 0;
2030                 break;
2031         }
2032         case KVM_SET_SREGS: {
2033                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2034                 if (IS_ERR(kvm_sregs)) {
2035                         r = PTR_ERR(kvm_sregs);
2036                         kvm_sregs = NULL;
2037                         goto out;
2038                 }
2039                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2040                 break;
2041         }
2042         case KVM_GET_MP_STATE: {
2043                 struct kvm_mp_state mp_state;
2044
2045                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2046                 if (r)
2047                         goto out;
2048                 r = -EFAULT;
2049                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2050                         goto out;
2051                 r = 0;
2052                 break;
2053         }
2054         case KVM_SET_MP_STATE: {
2055                 struct kvm_mp_state mp_state;
2056
2057                 r = -EFAULT;
2058                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2059                         goto out;
2060                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2061                 break;
2062         }
2063         case KVM_TRANSLATE: {
2064                 struct kvm_translation tr;
2065
2066                 r = -EFAULT;
2067                 if (copy_from_user(&tr, argp, sizeof tr))
2068                         goto out;
2069                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2070                 if (r)
2071                         goto out;
2072                 r = -EFAULT;
2073                 if (copy_to_user(argp, &tr, sizeof tr))
2074                         goto out;
2075                 r = 0;
2076                 break;
2077         }
2078         case KVM_SET_GUEST_DEBUG: {
2079                 struct kvm_guest_debug dbg;
2080
2081                 r = -EFAULT;
2082                 if (copy_from_user(&dbg, argp, sizeof dbg))
2083                         goto out;
2084                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2085                 break;
2086         }
2087         case KVM_SET_SIGNAL_MASK: {
2088                 struct kvm_signal_mask __user *sigmask_arg = argp;
2089                 struct kvm_signal_mask kvm_sigmask;
2090                 sigset_t sigset, *p;
2091
2092                 p = NULL;
2093                 if (argp) {
2094                         r = -EFAULT;
2095                         if (copy_from_user(&kvm_sigmask, argp,
2096                                            sizeof kvm_sigmask))
2097                                 goto out;
2098                         r = -EINVAL;
2099                         if (kvm_sigmask.len != sizeof sigset)
2100                                 goto out;
2101                         r = -EFAULT;
2102                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2103                                            sizeof sigset))
2104                                 goto out;
2105                         p = &sigset;
2106                 }
2107                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2108                 break;
2109         }
2110         case KVM_GET_FPU: {
2111                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2112                 r = -ENOMEM;
2113                 if (!fpu)
2114                         goto out;
2115                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2116                 if (r)
2117                         goto out;
2118                 r = -EFAULT;
2119                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2120                         goto out;
2121                 r = 0;
2122                 break;
2123         }
2124         case KVM_SET_FPU: {
2125                 fpu = memdup_user(argp, sizeof(*fpu));
2126                 if (IS_ERR(fpu)) {
2127                         r = PTR_ERR(fpu);
2128                         fpu = NULL;
2129                         goto out;
2130                 }
2131                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2132                 break;
2133         }
2134         default:
2135                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2136         }
2137 out:
2138         vcpu_put(vcpu);
2139         kfree(fpu);
2140         kfree(kvm_sregs);
2141         return r;
2142 }
2143
2144 #ifdef CONFIG_COMPAT
2145 static long kvm_vcpu_compat_ioctl(struct file *filp,
2146                                   unsigned int ioctl, unsigned long arg)
2147 {
2148         struct kvm_vcpu *vcpu = filp->private_data;
2149         void __user *argp = compat_ptr(arg);
2150         int r;
2151
2152         if (vcpu->kvm->mm != current->mm)
2153                 return -EIO;
2154
2155         switch (ioctl) {
2156         case KVM_SET_SIGNAL_MASK: {
2157                 struct kvm_signal_mask __user *sigmask_arg = argp;
2158                 struct kvm_signal_mask kvm_sigmask;
2159                 compat_sigset_t csigset;
2160                 sigset_t sigset;
2161
2162                 if (argp) {
2163                         r = -EFAULT;
2164                         if (copy_from_user(&kvm_sigmask, argp,
2165                                            sizeof kvm_sigmask))
2166                                 goto out;
2167                         r = -EINVAL;
2168                         if (kvm_sigmask.len != sizeof csigset)
2169                                 goto out;
2170                         r = -EFAULT;
2171                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2172                                            sizeof csigset))
2173                                 goto out;
2174                         sigset_from_compat(&sigset, &csigset);
2175                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2176                 } else
2177                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2178                 break;
2179         }
2180         default:
2181                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2182         }
2183
2184 out:
2185         return r;
2186 }
2187 #endif
2188
2189 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2190                                  int (*accessor)(struct kvm_device *dev,
2191                                                  struct kvm_device_attr *attr),
2192                                  unsigned long arg)
2193 {
2194         struct kvm_device_attr attr;
2195
2196         if (!accessor)
2197                 return -EPERM;
2198
2199         if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2200                 return -EFAULT;
2201
2202         return accessor(dev, &attr);
2203 }
2204
2205 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2206                              unsigned long arg)
2207 {
2208         struct kvm_device *dev = filp->private_data;
2209
2210         switch (ioctl) {
2211         case KVM_SET_DEVICE_ATTR:
2212                 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2213         case KVM_GET_DEVICE_ATTR:
2214                 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2215         case KVM_HAS_DEVICE_ATTR:
2216                 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2217         default:
2218                 if (dev->ops->ioctl)
2219                         return dev->ops->ioctl(dev, ioctl, arg);
2220
2221                 return -ENOTTY;
2222         }
2223 }
2224
2225 static int kvm_device_release(struct inode *inode, struct file *filp)
2226 {
2227         struct kvm_device *dev = filp->private_data;
2228         struct kvm *kvm = dev->kvm;
2229
2230         kvm_put_kvm(kvm);
2231         return 0;
2232 }
2233
2234 static const struct file_operations kvm_device_fops = {
2235         .unlocked_ioctl = kvm_device_ioctl,
2236 #ifdef CONFIG_COMPAT
2237         .compat_ioctl = kvm_device_ioctl,
2238 #endif
2239         .release = kvm_device_release,
2240 };
2241
2242 struct kvm_device *kvm_device_from_filp(struct file *filp)
2243 {
2244         if (filp->f_op != &kvm_device_fops)
2245                 return NULL;
2246
2247         return filp->private_data;
2248 }
2249
2250 static int kvm_ioctl_create_device(struct kvm *kvm,
2251                                    struct kvm_create_device *cd)
2252 {
2253         struct kvm_device_ops *ops = NULL;
2254         struct kvm_device *dev;
2255         bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2256         int ret;
2257
2258         switch (cd->type) {
2259 #ifdef CONFIG_KVM_MPIC
2260         case KVM_DEV_TYPE_FSL_MPIC_20:
2261         case KVM_DEV_TYPE_FSL_MPIC_42:
2262                 ops = &kvm_mpic_ops;
2263                 break;
2264 #endif
2265 #ifdef CONFIG_KVM_XICS
2266         case KVM_DEV_TYPE_XICS:
2267                 ops = &kvm_xics_ops;
2268                 break;
2269 #endif
2270 #ifdef CONFIG_KVM_VFIO
2271         case KVM_DEV_TYPE_VFIO:
2272                 ops = &kvm_vfio_ops;
2273                 break;
2274 #endif
2275 #ifdef CONFIG_KVM_ARM_VGIC
2276         case KVM_DEV_TYPE_ARM_VGIC_V2:
2277                 ops = &kvm_arm_vgic_v2_ops;
2278                 break;
2279 #endif
2280         default:
2281                 return -ENODEV;
2282         }
2283
2284         if (test)
2285                 return 0;
2286
2287         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2288         if (!dev)
2289                 return -ENOMEM;
2290
2291         dev->ops = ops;
2292         dev->kvm = kvm;
2293
2294         ret = ops->create(dev, cd->type);
2295         if (ret < 0) {
2296                 kfree(dev);
2297                 return ret;
2298         }
2299
2300         ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2301         if (ret < 0) {
2302                 ops->destroy(dev);
2303                 return ret;
2304         }
2305
2306         list_add(&dev->vm_node, &kvm->devices);
2307         kvm_get_kvm(kvm);
2308         cd->fd = ret;
2309         return 0;
2310 }
2311
2312 static long kvm_vm_ioctl(struct file *filp,
2313                            unsigned int ioctl, unsigned long arg)
2314 {
2315         struct kvm *kvm = filp->private_data;
2316         void __user *argp = (void __user *)arg;
2317         int r;
2318
2319         if (kvm->mm != current->mm)
2320                 return -EIO;
2321         switch (ioctl) {
2322         case KVM_CREATE_VCPU:
2323                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2324                 break;
2325         case KVM_SET_USER_MEMORY_REGION: {
2326                 struct kvm_userspace_memory_region kvm_userspace_mem;
2327
2328                 r = -EFAULT;
2329                 if (copy_from_user(&kvm_userspace_mem, argp,
2330                                                 sizeof kvm_userspace_mem))
2331                         goto out;
2332
2333                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2334                 break;
2335         }
2336         case KVM_GET_DIRTY_LOG: {
2337                 struct kvm_dirty_log log;
2338
2339                 r = -EFAULT;
2340                 if (copy_from_user(&log, argp, sizeof log))
2341                         goto out;
2342                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2343                 break;
2344         }
2345 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2346         case KVM_REGISTER_COALESCED_MMIO: {
2347                 struct kvm_coalesced_mmio_zone zone;
2348                 r = -EFAULT;
2349                 if (copy_from_user(&zone, argp, sizeof zone))
2350                         goto out;
2351                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2352                 break;
2353         }
2354         case KVM_UNREGISTER_COALESCED_MMIO: {
2355                 struct kvm_coalesced_mmio_zone zone;
2356                 r = -EFAULT;
2357                 if (copy_from_user(&zone, argp, sizeof zone))
2358                         goto out;
2359                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2360                 break;
2361         }
2362 #endif
2363         case KVM_IRQFD: {
2364                 struct kvm_irqfd data;
2365
2366                 r = -EFAULT;
2367                 if (copy_from_user(&data, argp, sizeof data))
2368                         goto out;
2369                 r = kvm_irqfd(kvm, &data);
2370                 break;
2371         }
2372         case KVM_IOEVENTFD: {
2373                 struct kvm_ioeventfd data;
2374
2375                 r = -EFAULT;
2376                 if (copy_from_user(&data, argp, sizeof data))
2377                         goto out;
2378                 r = kvm_ioeventfd(kvm, &data);
2379                 break;
2380         }
2381 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2382         case KVM_SET_BOOT_CPU_ID:
2383                 r = 0;
2384                 mutex_lock(&kvm->lock);
2385                 if (atomic_read(&kvm->online_vcpus) != 0)
2386                         r = -EBUSY;
2387                 else
2388                         kvm->bsp_vcpu_id = arg;
2389                 mutex_unlock(&kvm->lock);
2390                 break;
2391 #endif
2392 #ifdef CONFIG_HAVE_KVM_MSI
2393         case KVM_SIGNAL_MSI: {
2394                 struct kvm_msi msi;
2395
2396                 r = -EFAULT;
2397                 if (copy_from_user(&msi, argp, sizeof msi))
2398                         goto out;
2399                 r = kvm_send_userspace_msi(kvm, &msi);
2400                 break;
2401         }
2402 #endif
2403 #ifdef __KVM_HAVE_IRQ_LINE
2404         case KVM_IRQ_LINE_STATUS:
2405         case KVM_IRQ_LINE: {
2406                 struct kvm_irq_level irq_event;
2407
2408                 r = -EFAULT;
2409                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2410                         goto out;
2411
2412                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2413                                         ioctl == KVM_IRQ_LINE_STATUS);
2414                 if (r)
2415                         goto out;
2416
2417                 r = -EFAULT;
2418                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2419                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2420                                 goto out;
2421                 }
2422
2423                 r = 0;
2424                 break;
2425         }
2426 #endif
2427 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2428         case KVM_SET_GSI_ROUTING: {
2429                 struct kvm_irq_routing routing;
2430                 struct kvm_irq_routing __user *urouting;
2431                 struct kvm_irq_routing_entry *entries;
2432
2433                 r = -EFAULT;
2434                 if (copy_from_user(&routing, argp, sizeof(routing)))
2435                         goto out;
2436                 r = -EINVAL;
2437                 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2438                         goto out;
2439                 if (routing.flags)
2440                         goto out;
2441                 r = -ENOMEM;
2442                 entries = vmalloc(routing.nr * sizeof(*entries));
2443                 if (!entries)
2444                         goto out;
2445                 r = -EFAULT;
2446                 urouting = argp;
2447                 if (copy_from_user(entries, urouting->entries,
2448                                    routing.nr * sizeof(*entries)))
2449                         goto out_free_irq_routing;
2450                 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2451                                         routing.flags);
2452         out_free_irq_routing:
2453                 vfree(entries);
2454                 break;
2455         }
2456 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2457         case KVM_CREATE_DEVICE: {
2458                 struct kvm_create_device cd;
2459
2460                 r = -EFAULT;
2461                 if (copy_from_user(&cd, argp, sizeof(cd)))
2462                         goto out;
2463
2464                 r = kvm_ioctl_create_device(kvm, &cd);
2465                 if (r)
2466                         goto out;
2467
2468                 r = -EFAULT;
2469                 if (copy_to_user(argp, &cd, sizeof(cd)))
2470                         goto out;
2471
2472                 r = 0;
2473                 break;
2474         }
2475         default:
2476                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2477                 if (r == -ENOTTY)
2478                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2479         }
2480 out:
2481         return r;
2482 }
2483
2484 #ifdef CONFIG_COMPAT
2485 struct compat_kvm_dirty_log {
2486         __u32 slot;
2487         __u32 padding1;
2488         union {
2489                 compat_uptr_t dirty_bitmap; /* one bit per page */
2490                 __u64 padding2;
2491         };
2492 };
2493
2494 static long kvm_vm_compat_ioctl(struct file *filp,
2495                            unsigned int ioctl, unsigned long arg)
2496 {
2497         struct kvm *kvm = filp->private_data;
2498         int r;
2499
2500         if (kvm->mm != current->mm)
2501                 return -EIO;
2502         switch (ioctl) {
2503         case KVM_GET_DIRTY_LOG: {
2504                 struct compat_kvm_dirty_log compat_log;
2505                 struct kvm_dirty_log log;
2506
2507                 r = -EFAULT;
2508                 if (copy_from_user(&compat_log, (void __user *)arg,
2509                                    sizeof(compat_log)))
2510                         goto out;
2511                 log.slot         = compat_log.slot;
2512                 log.padding1     = compat_log.padding1;
2513                 log.padding2     = compat_log.padding2;
2514                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2515
2516                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2517                 break;
2518         }
2519         default:
2520                 r = kvm_vm_ioctl(filp, ioctl, arg);
2521         }
2522
2523 out:
2524         return r;
2525 }
2526 #endif
2527
2528 static struct file_operations kvm_vm_fops = {
2529         .release        = kvm_vm_release,
2530         .unlocked_ioctl = kvm_vm_ioctl,
2531 #ifdef CONFIG_COMPAT
2532         .compat_ioctl   = kvm_vm_compat_ioctl,
2533 #endif
2534         .llseek         = noop_llseek,
2535 };
2536
2537 static int kvm_dev_ioctl_create_vm(unsigned long type)
2538 {
2539         int r;
2540         struct kvm *kvm;
2541
2542         kvm = kvm_create_vm(type);
2543         if (IS_ERR(kvm))
2544                 return PTR_ERR(kvm);
2545 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2546         r = kvm_coalesced_mmio_init(kvm);
2547         if (r < 0) {
2548                 kvm_put_kvm(kvm);
2549                 return r;
2550         }
2551 #endif
2552         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2553         if (r < 0)
2554                 kvm_put_kvm(kvm);
2555
2556         return r;
2557 }
2558
2559 static long kvm_dev_ioctl_check_extension_generic(long arg)
2560 {
2561         switch (arg) {
2562         case KVM_CAP_USER_MEMORY:
2563         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2564         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2565 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2566         case KVM_CAP_SET_BOOT_CPU_ID:
2567 #endif
2568         case KVM_CAP_INTERNAL_ERROR_DATA:
2569 #ifdef CONFIG_HAVE_KVM_MSI
2570         case KVM_CAP_SIGNAL_MSI:
2571 #endif
2572 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2573         case KVM_CAP_IRQFD_RESAMPLE:
2574 #endif
2575                 return 1;
2576 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2577         case KVM_CAP_IRQ_ROUTING:
2578                 return KVM_MAX_IRQ_ROUTES;
2579 #endif
2580         default:
2581                 break;
2582         }
2583         return kvm_dev_ioctl_check_extension(arg);
2584 }
2585
2586 static long kvm_dev_ioctl(struct file *filp,
2587                           unsigned int ioctl, unsigned long arg)
2588 {
2589         long r = -EINVAL;
2590
2591         switch (ioctl) {
2592         case KVM_GET_API_VERSION:
2593                 r = -EINVAL;
2594                 if (arg)
2595                         goto out;
2596                 r = KVM_API_VERSION;
2597                 break;
2598         case KVM_CREATE_VM:
2599                 r = kvm_dev_ioctl_create_vm(arg);
2600                 break;
2601         case KVM_CHECK_EXTENSION:
2602                 r = kvm_dev_ioctl_check_extension_generic(arg);
2603                 break;
2604         case KVM_GET_VCPU_MMAP_SIZE:
2605                 r = -EINVAL;
2606                 if (arg)
2607                         goto out;
2608                 r = PAGE_SIZE;     /* struct kvm_run */
2609 #ifdef CONFIG_X86
2610                 r += PAGE_SIZE;    /* pio data page */
2611 #endif
2612 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2613                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2614 #endif
2615                 break;
2616         case KVM_TRACE_ENABLE:
2617         case KVM_TRACE_PAUSE:
2618         case KVM_TRACE_DISABLE:
2619                 r = -EOPNOTSUPP;
2620                 break;
2621         default:
2622                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2623         }
2624 out:
2625         return r;
2626 }
2627
2628 static struct file_operations kvm_chardev_ops = {
2629         .unlocked_ioctl = kvm_dev_ioctl,
2630         .compat_ioctl   = kvm_dev_ioctl,
2631         .llseek         = noop_llseek,
2632 };
2633
2634 static struct miscdevice kvm_dev = {
2635         KVM_MINOR,
2636         "kvm",
2637         &kvm_chardev_ops,
2638 };
2639
2640 static void hardware_enable_nolock(void *junk)
2641 {
2642         int cpu = raw_smp_processor_id();
2643         int r;
2644
2645         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2646                 return;
2647
2648         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2649
2650         r = kvm_arch_hardware_enable(NULL);
2651
2652         if (r) {
2653                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2654                 atomic_inc(&hardware_enable_failed);
2655                 printk(KERN_INFO "kvm: enabling virtualization on "
2656                                  "CPU%d failed\n", cpu);
2657         }
2658 }
2659
2660 static void hardware_enable(void)
2661 {
2662         raw_spin_lock(&kvm_count_lock);
2663         if (kvm_usage_count)
2664                 hardware_enable_nolock(NULL);
2665         raw_spin_unlock(&kvm_count_lock);
2666 }
2667
2668 static void hardware_disable_nolock(void *junk)
2669 {
2670         int cpu = raw_smp_processor_id();
2671
2672         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2673                 return;
2674         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2675         kvm_arch_hardware_disable(NULL);
2676 }
2677
2678 static void hardware_disable(void)
2679 {
2680         raw_spin_lock(&kvm_count_lock);
2681         if (kvm_usage_count)
2682                 hardware_disable_nolock(NULL);
2683         raw_spin_unlock(&kvm_count_lock);
2684 }
2685
2686 static void hardware_disable_all_nolock(void)
2687 {
2688         BUG_ON(!kvm_usage_count);
2689
2690         kvm_usage_count--;
2691         if (!kvm_usage_count)
2692                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2693 }
2694
2695 static void hardware_disable_all(void)
2696 {
2697         raw_spin_lock(&kvm_count_lock);
2698         hardware_disable_all_nolock();
2699         raw_spin_unlock(&kvm_count_lock);
2700 }
2701
2702 static int hardware_enable_all(void)
2703 {
2704         int r = 0;
2705
2706         raw_spin_lock(&kvm_count_lock);
2707
2708         kvm_usage_count++;
2709         if (kvm_usage_count == 1) {
2710                 atomic_set(&hardware_enable_failed, 0);
2711                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2712
2713                 if (atomic_read(&hardware_enable_failed)) {
2714                         hardware_disable_all_nolock();
2715                         r = -EBUSY;
2716                 }
2717         }
2718
2719         raw_spin_unlock(&kvm_count_lock);
2720
2721         return r;
2722 }
2723
2724 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2725                            void *v)
2726 {
2727         int cpu = (long)v;
2728
2729         val &= ~CPU_TASKS_FROZEN;
2730         switch (val) {
2731         case CPU_DYING:
2732                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2733                        cpu);
2734                 hardware_disable();
2735                 break;
2736         case CPU_STARTING:
2737                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2738                        cpu);
2739                 hardware_enable();
2740                 break;
2741         }
2742         return NOTIFY_OK;
2743 }
2744
2745 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2746                       void *v)
2747 {
2748         /*
2749          * Some (well, at least mine) BIOSes hang on reboot if
2750          * in vmx root mode.
2751          *
2752          * And Intel TXT required VMX off for all cpu when system shutdown.
2753          */
2754         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2755         kvm_rebooting = true;
2756         on_each_cpu(hardware_disable_nolock, NULL, 1);
2757         return NOTIFY_OK;
2758 }
2759
2760 static struct notifier_block kvm_reboot_notifier = {
2761         .notifier_call = kvm_reboot,
2762         .priority = 0,
2763 };
2764
2765 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2766 {
2767         int i;
2768
2769         for (i = 0; i < bus->dev_count; i++) {
2770                 struct kvm_io_device *pos = bus->range[i].dev;
2771
2772                 kvm_iodevice_destructor(pos);
2773         }
2774         kfree(bus);
2775 }
2776
2777 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2778 {
2779         const struct kvm_io_range *r1 = p1;
2780         const struct kvm_io_range *r2 = p2;
2781
2782         if (r1->addr < r2->addr)
2783                 return -1;
2784         if (r1->addr + r1->len > r2->addr + r2->len)
2785                 return 1;
2786         return 0;
2787 }
2788
2789 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2790                           gpa_t addr, int len)
2791 {
2792         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2793                 .addr = addr,
2794                 .len = len,
2795                 .dev = dev,
2796         };
2797
2798         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2799                 kvm_io_bus_sort_cmp, NULL);
2800
2801         return 0;
2802 }
2803
2804 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2805                              gpa_t addr, int len)
2806 {
2807         struct kvm_io_range *range, key;
2808         int off;
2809
2810         key = (struct kvm_io_range) {
2811                 .addr = addr,
2812                 .len = len,
2813         };
2814
2815         range = bsearch(&key, bus->range, bus->dev_count,
2816                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2817         if (range == NULL)
2818                 return -ENOENT;
2819
2820         off = range - bus->range;
2821
2822         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2823                 off--;
2824
2825         return off;
2826 }
2827
2828 /* kvm_io_bus_write - called under kvm->slots_lock */
2829 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2830                      int len, const void *val)
2831 {
2832         int idx;
2833         struct kvm_io_bus *bus;
2834         struct kvm_io_range range;
2835
2836         range = (struct kvm_io_range) {
2837                 .addr = addr,
2838                 .len = len,
2839         };
2840
2841         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2842         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2843         if (idx < 0)
2844                 return -EOPNOTSUPP;
2845
2846         while (idx < bus->dev_count &&
2847                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2848                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2849                         return 0;
2850                 idx++;
2851         }
2852
2853         return -EOPNOTSUPP;
2854 }
2855
2856 /* kvm_io_bus_read - called under kvm->slots_lock */
2857 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2858                     int len, void *val)
2859 {
2860         int idx;
2861         struct kvm_io_bus *bus;
2862         struct kvm_io_range range;
2863
2864         range = (struct kvm_io_range) {
2865                 .addr = addr,
2866                 .len = len,
2867         };
2868
2869         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2870         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2871         if (idx < 0)
2872                 return -EOPNOTSUPP;
2873
2874         while (idx < bus->dev_count &&
2875                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2876                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2877                         return 0;
2878                 idx++;
2879         }
2880
2881         return -EOPNOTSUPP;
2882 }
2883
2884 /* Caller must hold slots_lock. */
2885 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2886                             int len, struct kvm_io_device *dev)
2887 {
2888         struct kvm_io_bus *new_bus, *bus;
2889
2890         bus = kvm->buses[bus_idx];
2891         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2892                 return -ENOSPC;
2893
2894         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2895                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2896         if (!new_bus)
2897                 return -ENOMEM;
2898         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2899                sizeof(struct kvm_io_range)));
2900         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2901         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2902         synchronize_srcu_expedited(&kvm->srcu);
2903         kfree(bus);
2904
2905         return 0;
2906 }
2907
2908 /* Caller must hold slots_lock. */
2909 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2910                               struct kvm_io_device *dev)
2911 {
2912         int i, r;
2913         struct kvm_io_bus *new_bus, *bus;
2914
2915         bus = kvm->buses[bus_idx];
2916         r = -ENOENT;
2917         for (i = 0; i < bus->dev_count; i++)
2918                 if (bus->range[i].dev == dev) {
2919                         r = 0;
2920                         break;
2921                 }
2922
2923         if (r)
2924                 return r;
2925
2926         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2927                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2928         if (!new_bus)
2929                 return -ENOMEM;
2930
2931         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2932         new_bus->dev_count--;
2933         memcpy(new_bus->range + i, bus->range + i + 1,
2934                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2935
2936         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2937         synchronize_srcu_expedited(&kvm->srcu);
2938         kfree(bus);
2939         return r;
2940 }
2941
2942 static struct notifier_block kvm_cpu_notifier = {
2943         .notifier_call = kvm_cpu_hotplug,
2944 };
2945
2946 static int vm_stat_get(void *_offset, u64 *val)
2947 {
2948         unsigned offset = (long)_offset;
2949         struct kvm *kvm;
2950
2951         *val = 0;
2952         spin_lock(&kvm_lock);
2953         list_for_each_entry(kvm, &vm_list, vm_list)
2954                 *val += *(u32 *)((void *)kvm + offset);
2955         spin_unlock(&kvm_lock);
2956         return 0;
2957 }
2958
2959 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2960
2961 static int vcpu_stat_get(void *_offset, u64 *val)
2962 {
2963         unsigned offset = (long)_offset;
2964         struct kvm *kvm;
2965         struct kvm_vcpu *vcpu;
2966         int i;
2967
2968         *val = 0;
2969         spin_lock(&kvm_lock);
2970         list_for_each_entry(kvm, &vm_list, vm_list)
2971                 kvm_for_each_vcpu(i, vcpu, kvm)
2972                         *val += *(u32 *)((void *)vcpu + offset);
2973
2974         spin_unlock(&kvm_lock);
2975         return 0;
2976 }
2977
2978 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2979
2980 static const struct file_operations *stat_fops[] = {
2981         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2982         [KVM_STAT_VM]   = &vm_stat_fops,
2983 };
2984
2985 static int kvm_init_debug(void)
2986 {
2987         int r = -EEXIST;
2988         struct kvm_stats_debugfs_item *p;
2989
2990         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2991         if (kvm_debugfs_dir == NULL)
2992                 goto out;
2993
2994         for (p = debugfs_entries; p->name; ++p) {
2995                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2996                                                 (void *)(long)p->offset,
2997                                                 stat_fops[p->kind]);
2998                 if (p->dentry == NULL)
2999                         goto out_dir;
3000         }
3001
3002         return 0;
3003
3004 out_dir:
3005         debugfs_remove_recursive(kvm_debugfs_dir);
3006 out:
3007         return r;
3008 }
3009
3010 static void kvm_exit_debug(void)
3011 {
3012         struct kvm_stats_debugfs_item *p;
3013
3014         for (p = debugfs_entries; p->name; ++p)
3015                 debugfs_remove(p->dentry);
3016         debugfs_remove(kvm_debugfs_dir);
3017 }
3018
3019 static int kvm_suspend(void)
3020 {
3021         if (kvm_usage_count)
3022                 hardware_disable_nolock(NULL);
3023         return 0;
3024 }
3025
3026 static void kvm_resume(void)
3027 {
3028         if (kvm_usage_count) {
3029                 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3030                 hardware_enable_nolock(NULL);
3031         }
3032 }
3033
3034 static struct syscore_ops kvm_syscore_ops = {
3035         .suspend = kvm_suspend,
3036         .resume = kvm_resume,
3037 };
3038
3039 static inline
3040 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3041 {
3042         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3043 }
3044
3045 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3046 {
3047         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3048         if (vcpu->preempted)
3049                 vcpu->preempted = false;
3050
3051         kvm_arch_vcpu_load(vcpu, cpu);
3052 }
3053
3054 static void kvm_sched_out(struct preempt_notifier *pn,
3055                           struct task_struct *next)
3056 {
3057         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3058
3059         if (current->state == TASK_RUNNING)
3060                 vcpu->preempted = true;
3061         kvm_arch_vcpu_put(vcpu);
3062 }
3063
3064 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3065                   struct module *module)
3066 {
3067         int r;
3068         int cpu;
3069
3070         r = kvm_arch_init(opaque);
3071         if (r)
3072                 goto out_fail;
3073
3074         /*
3075          * kvm_arch_init makes sure there's at most one caller
3076          * for architectures that support multiple implementations,
3077          * like intel and amd on x86.
3078          * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3079          * conflicts in case kvm is already setup for another implementation.
3080          */
3081         r = kvm_irqfd_init();
3082         if (r)
3083                 goto out_irqfd;
3084
3085         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3086                 r = -ENOMEM;
3087                 goto out_free_0;
3088         }
3089
3090         r = kvm_arch_hardware_setup();
3091         if (r < 0)
3092                 goto out_free_0a;
3093
3094         for_each_online_cpu(cpu) {
3095                 smp_call_function_single(cpu,
3096                                 kvm_arch_check_processor_compat,
3097                                 &r, 1);
3098                 if (r < 0)
3099                         goto out_free_1;
3100         }
3101
3102         r = register_cpu_notifier(&kvm_cpu_notifier);
3103         if (r)
3104                 goto out_free_2;
3105         register_reboot_notifier(&kvm_reboot_notifier);
3106
3107         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3108         if (!vcpu_align)
3109                 vcpu_align = __alignof__(struct kvm_vcpu);
3110         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3111                                            0, NULL);
3112         if (!kvm_vcpu_cache) {
3113                 r = -ENOMEM;
3114                 goto out_free_3;
3115         }
3116
3117         r = kvm_async_pf_init();
3118         if (r)
3119                 goto out_free;
3120
3121         kvm_chardev_ops.owner = module;
3122         kvm_vm_fops.owner = module;
3123         kvm_vcpu_fops.owner = module;
3124
3125         r = misc_register(&kvm_dev);
3126         if (r) {
3127                 printk(KERN_ERR "kvm: misc device register failed\n");
3128                 goto out_unreg;
3129         }
3130
3131         register_syscore_ops(&kvm_syscore_ops);
3132
3133         kvm_preempt_ops.sched_in = kvm_sched_in;
3134         kvm_preempt_ops.sched_out = kvm_sched_out;
3135
3136         r = kvm_init_debug();
3137         if (r) {
3138                 printk(KERN_ERR "kvm: create debugfs files failed\n");
3139                 goto out_undebugfs;
3140         }
3141
3142         return 0;
3143
3144 out_undebugfs:
3145         unregister_syscore_ops(&kvm_syscore_ops);
3146         misc_deregister(&kvm_dev);
3147 out_unreg:
3148         kvm_async_pf_deinit();
3149 out_free:
3150         kmem_cache_destroy(kvm_vcpu_cache);
3151 out_free_3:
3152         unregister_reboot_notifier(&kvm_reboot_notifier);
3153         unregister_cpu_notifier(&kvm_cpu_notifier);
3154 out_free_2:
3155 out_free_1:
3156         kvm_arch_hardware_unsetup();
3157 out_free_0a:
3158         free_cpumask_var(cpus_hardware_enabled);
3159 out_free_0:
3160         kvm_irqfd_exit();
3161 out_irqfd:
3162         kvm_arch_exit();
3163 out_fail:
3164         return r;
3165 }
3166 EXPORT_SYMBOL_GPL(kvm_init);
3167
3168 void kvm_exit(void)
3169 {
3170         kvm_exit_debug();
3171         misc_deregister(&kvm_dev);
3172         kmem_cache_destroy(kvm_vcpu_cache);
3173         kvm_async_pf_deinit();
3174         unregister_syscore_ops(&kvm_syscore_ops);
3175         unregister_reboot_notifier(&kvm_reboot_notifier);
3176         unregister_cpu_notifier(&kvm_cpu_notifier);
3177         on_each_cpu(hardware_disable_nolock, NULL, 1);
3178         kvm_arch_hardware_unsetup();
3179         kvm_arch_exit();
3180         kvm_irqfd_exit();
3181         free_cpumask_var(cpus_hardware_enabled);
3182 }
3183 EXPORT_SYMBOL_GPL(kvm_exit);