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/sysdev.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
  51 #include <asm/processor.h>
  52 #include <asm/io.h>
  53 #include <asm/uaccess.h>
  54 #include <asm/pgtable.h>
  55 #include <asm-generic/bitops/le.h>
  56
  57 #include "coalesced_mmio.h"
  58 #include "async_pf.h"
  59
  60 #define CREATE_TRACE_POINTS
  61 #include <trace/events/kvm.h>
  62
  63 MODULE_AUTHOR("Qumranet");
  64 MODULE_LICENSE("GPL");
  65
  66 /*
  67  * Ordering of locks:
  68  *
  69  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  70  */
  71
  72 DEFINE_SPINLOCK(kvm_lock);
  73 LIST_HEAD(vm_list);
  74
  75 static cpumask_var_t cpus_hardware_enabled;
  76 static int kvm_usage_count = 0;
  77 static atomic_t hardware_enable_failed;
  78
  79 struct kmem_cache *kvm_vcpu_cache;
  80 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  81
  82 static __read_mostly struct preempt_ops kvm_preempt_ops;
  83
  84 struct dentry *kvm_debugfs_dir;
  85
  86 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  87                            unsigned long arg);
  88 static int hardware_enable_all(void);
  89 static void hardware_disable_all(void);
  90
  91 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  92
  93 bool kvm_rebooting;
  94 EXPORT_SYMBOL_GPL(kvm_rebooting);
  95
  96 static bool largepages_enabled = true;
  97
  98 static struct page *hwpoison_page;
  99 static pfn_t hwpoison_pfn;
 100
 101 static struct page *fault_page;
 102 static pfn_t fault_pfn;
 103
 104 inline int kvm_is_mmio_pfn(pfn_t pfn)
 105 {
 106         if (pfn_valid(pfn)) {
 107                 int reserved;
 108                 struct page *tail = pfn_to_page(pfn);
 109                 struct page *head = compound_trans_head(tail);
 110                 reserved = PageReserved(head);
 111                 if (head != tail) {
 112                         /*
 113                          * "head" is not a dangling pointer
 114                          * (compound_trans_head takes care of that)
 115                          * but the hugepage may have been splitted
 116                          * from under us (and we may not hold a
 117                          * reference count on the head page so it can
 118                          * be reused before we run PageReferenced), so
 119                          * we've to check PageTail before returning
 120                          * what we just read.
 121                          */
 122                         smp_rmb();
 123                         if (PageTail(tail))
 124                                 return reserved;
 125                 }
 126                 return PageReserved(tail);
 127         }
 128
 129         return true;
 130 }
 131
 132 /*
 133  * Switches to specified vcpu, until a matching vcpu_put()
 134  */
 135 void vcpu_load(struct kvm_vcpu *vcpu)
 136 {
 137         int cpu;
 138
 139         mutex_lock(&vcpu->mutex);
 140         cpu = get_cpu();
 141         preempt_notifier_register(&vcpu->preempt_notifier);
 142         kvm_arch_vcpu_load(vcpu, cpu);
 143         put_cpu();
 144 }
 145
 146 void vcpu_put(struct kvm_vcpu *vcpu)
 147 {
 148         preempt_disable();
 149         kvm_arch_vcpu_put(vcpu);
 150         preempt_notifier_unregister(&vcpu->preempt_notifier);
 151         preempt_enable();
 152         mutex_unlock(&vcpu->mutex);
 153 }
 154
 155 static void ack_flush(void *_completed)
 156 {
 157 }
 158
 159 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 160 {
 161         int i, cpu, me;
 162         cpumask_var_t cpus;
 163         bool called = true;
 164         struct kvm_vcpu *vcpu;
 165
 166         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 167
 168         me = get_cpu();
 169         kvm_for_each_vcpu(i, vcpu, kvm) {
 170                 kvm_make_request(req, vcpu);
 171                 cpu = vcpu->cpu;
 172
 173                 /* Set ->requests bit before we read ->mode */
 174                 smp_mb();
 175
 176                 if (cpus != NULL && cpu != -1 && cpu != me &&
 177                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 178                         cpumask_set_cpu(cpu, cpus);
 179         }
 180         if (unlikely(cpus == NULL))
 181                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 182         else if (!cpumask_empty(cpus))
 183                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 184         else
 185                 called = false;
 186         put_cpu();
 187         free_cpumask_var(cpus);
 188         return called;
 189 }
 190
 191 void kvm_flush_remote_tlbs(struct kvm *kvm)
 192 {
 193         int dirty_count = kvm->tlbs_dirty;
 194
 195         smp_mb();
 196         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 197                 ++kvm->stat.remote_tlb_flush;
 198         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 199 }
 200
 201 void kvm_reload_remote_mmus(struct kvm *kvm)
 202 {
 203         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 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         init_waitqueue_head(&vcpu->wq);
 216         kvm_async_pf_vcpu_init(vcpu);
 217
 218         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 219         if (!page) {
 220                 r = -ENOMEM;
 221                 goto fail;
 222         }
 223         vcpu->run = page_address(page);
 224
 225         r = kvm_arch_vcpu_init(vcpu);
 226         if (r < 0)
 227                 goto fail_free_run;
 228         return 0;
 229
 230 fail_free_run:
 231         free_page((unsigned long)vcpu->run);
 232 fail:
 233         return r;
 234 }
 235 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 236
 237 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 238 {
 239         kvm_arch_vcpu_uninit(vcpu);
 240         free_page((unsigned long)vcpu->run);
 241 }
 242 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 243
 244 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 245 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 246 {
 247         return container_of(mn, struct kvm, mmu_notifier);
 248 }
 249
 250 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 251                                              struct mm_struct *mm,
 252                                              unsigned long address)
 253 {
 254         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 255         int need_tlb_flush, idx;
 256
 257         /*
 258          * When ->invalidate_page runs, the linux pte has been zapped
 259          * already but the page is still allocated until
 260          * ->invalidate_page returns. So if we increase the sequence
 261          * here the kvm page fault will notice if the spte can't be
 262          * established because the page is going to be freed. If
 263          * instead the kvm page fault establishes the spte before
 264          * ->invalidate_page runs, kvm_unmap_hva will release it
 265          * before returning.
 266          *
 267          * The sequence increase only need to be seen at spin_unlock
 268          * time, and not at spin_lock time.
 269          *
 270          * Increasing the sequence after the spin_unlock would be
 271          * unsafe because the kvm page fault could then establish the
 272          * pte after kvm_unmap_hva returned, without noticing the page
 273          * is going to be freed.
 274          */
 275         idx = srcu_read_lock(&kvm->srcu);
 276         spin_lock(&kvm->mmu_lock);
 277         kvm->mmu_notifier_seq++;
 278         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 279         spin_unlock(&kvm->mmu_lock);
 280         srcu_read_unlock(&kvm->srcu, idx);
 281
 282         /* we've to flush the tlb before the pages can be freed */
 283         if (need_tlb_flush)
 284                 kvm_flush_remote_tlbs(kvm);
 285
 286 }
 287
 288 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 289                                         struct mm_struct *mm,
 290                                         unsigned long address,
 291                                         pte_t pte)
 292 {
 293         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 294         int idx;
 295
 296         idx = srcu_read_lock(&kvm->srcu);
 297         spin_lock(&kvm->mmu_lock);
 298         kvm->mmu_notifier_seq++;
 299         kvm_set_spte_hva(kvm, address, pte);
 300         spin_unlock(&kvm->mmu_lock);
 301         srcu_read_unlock(&kvm->srcu, idx);
 302 }
 303
 304 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 305                                                     struct mm_struct *mm,
 306                                                     unsigned long start,
 307                                                     unsigned long end)
 308 {
 309         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 310         int need_tlb_flush = 0, idx;
 311
 312         idx = srcu_read_lock(&kvm->srcu);
 313         spin_lock(&kvm->mmu_lock);
 314         /*
 315          * The count increase must become visible at unlock time as no
 316          * spte can be established without taking the mmu_lock and
 317          * count is also read inside the mmu_lock critical section.
 318          */
 319         kvm->mmu_notifier_count++;
 320         for (; start < end; start += PAGE_SIZE)
 321                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 322         need_tlb_flush |= kvm->tlbs_dirty;
 323         spin_unlock(&kvm->mmu_lock);
 324         srcu_read_unlock(&kvm->srcu, idx);
 325
 326         /* we've to flush the tlb before the pages can be freed */
 327         if (need_tlb_flush)
 328                 kvm_flush_remote_tlbs(kvm);
 329 }
 330
 331 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 332                                                   struct mm_struct *mm,
 333                                                   unsigned long start,
 334                                                   unsigned long end)
 335 {
 336         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 337
 338         spin_lock(&kvm->mmu_lock);
 339         /*
 340          * This sequence increase will notify the kvm page fault that
 341          * the page that is going to be mapped in the spte could have
 342          * been freed.
 343          */
 344         kvm->mmu_notifier_seq++;
 345         /*
 346          * The above sequence increase must be visible before the
 347          * below count decrease but both values are read by the kvm
 348          * page fault under mmu_lock spinlock so we don't need to add
 349          * a smb_wmb() here in between the two.
 350          */
 351         kvm->mmu_notifier_count--;
 352         spin_unlock(&kvm->mmu_lock);
 353
 354         BUG_ON(kvm->mmu_notifier_count < 0);
 355 }
 356
 357 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 358                                               struct mm_struct *mm,
 359                                               unsigned long address)
 360 {
 361         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 362         int young, idx;
 363
 364         idx = srcu_read_lock(&kvm->srcu);
 365         spin_lock(&kvm->mmu_lock);
 366         young = kvm_age_hva(kvm, address);
 367         spin_unlock(&kvm->mmu_lock);
 368         srcu_read_unlock(&kvm->srcu, idx);
 369
 370         if (young)
 371                 kvm_flush_remote_tlbs(kvm);
 372
 373         return young;
 374 }
 375
 376 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 377                                        struct mm_struct *mm,
 378                                        unsigned long address)
 379 {
 380         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 381         int young, idx;
 382
 383         idx = srcu_read_lock(&kvm->srcu);
 384         spin_lock(&kvm->mmu_lock);
 385         young = kvm_test_age_hva(kvm, address);
 386         spin_unlock(&kvm->mmu_lock);
 387         srcu_read_unlock(&kvm->srcu, idx);
 388
 389         return young;
 390 }
 391
 392 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 393                                      struct mm_struct *mm)
 394 {
 395         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 396         int idx;
 397
 398         idx = srcu_read_lock(&kvm->srcu);
 399         kvm_arch_flush_shadow(kvm);
 400         srcu_read_unlock(&kvm->srcu, idx);
 401 }
 402
 403 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 404         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 405         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 406         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 407         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 408         .test_young             = kvm_mmu_notifier_test_young,
 409         .change_pte             = kvm_mmu_notifier_change_pte,
 410         .release                = kvm_mmu_notifier_release,
 411 };
 412
 413 static int kvm_init_mmu_notifier(struct kvm *kvm)
 414 {
 415         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 416         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 417 }
 418
 419 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 420
 421 static int kvm_init_mmu_notifier(struct kvm *kvm)
 422 {
 423         return 0;
 424 }
 425
 426 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 427
 428 static struct kvm *kvm_create_vm(void)
 429 {
 430         int r, i;
 431         struct kvm *kvm = kvm_arch_alloc_vm();
 432
 433         if (!kvm)
 434                 return ERR_PTR(-ENOMEM);
 435
 436         r = kvm_arch_init_vm(kvm);
 437         if (r)
 438                 goto out_err_nodisable;
 439
 440         r = hardware_enable_all();
 441         if (r)
 442                 goto out_err_nodisable;
 443
 444 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 445         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 446         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 447 #endif
 448
 449         r = -ENOMEM;
 450         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 451         if (!kvm->memslots)
 452                 goto out_err_nosrcu;
 453         if (init_srcu_struct(&kvm->srcu))
 454                 goto out_err_nosrcu;
 455         for (i = 0; i < KVM_NR_BUSES; i++) {
 456                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 457                                         GFP_KERNEL);
 458                 if (!kvm->buses[i])
 459                         goto out_err;
 460         }
 461
 462         r = kvm_init_mmu_notifier(kvm);
 463         if (r)
 464                 goto out_err;
 465
 466         kvm->mm = current->mm;
 467         atomic_inc(&kvm->mm->mm_count);
 468         spin_lock_init(&kvm->mmu_lock);
 469         kvm_eventfd_init(kvm);
 470         mutex_init(&kvm->lock);
 471         mutex_init(&kvm->irq_lock);
 472         mutex_init(&kvm->slots_lock);
 473         atomic_set(&kvm->users_count, 1);
 474         spin_lock(&kvm_lock);
 475         list_add(&kvm->vm_list, &vm_list);
 476         spin_unlock(&kvm_lock);
 477
 478         return kvm;
 479
 480 out_err:
 481         cleanup_srcu_struct(&kvm->srcu);
 482 out_err_nosrcu:
 483         hardware_disable_all();
 484 out_err_nodisable:
 485         for (i = 0; i < KVM_NR_BUSES; i++)
 486                 kfree(kvm->buses[i]);
 487         kfree(kvm->memslots);
 488         kvm_arch_free_vm(kvm);
 489         return ERR_PTR(r);
 490 }
 491
 492 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 493 {
 494         if (!memslot->dirty_bitmap)
 495                 return;
 496
 497         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
 498                 vfree(memslot->dirty_bitmap_head);
 499         else
 500                 kfree(memslot->dirty_bitmap_head);
 501
 502         memslot->dirty_bitmap = NULL;
 503         memslot->dirty_bitmap_head = NULL;
 504 }
 505
 506 /*
 507  * Free any memory in @free but not in @dont.
 508  */
 509 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 510                                   struct kvm_memory_slot *dont)
 511 {
 512         int i;
 513
 514         if (!dont || free->rmap != dont->rmap)
 515                 vfree(free->rmap);
 516
 517         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 518                 kvm_destroy_dirty_bitmap(free);
 519
 520
 521         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 522                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 523                         vfree(free->lpage_info[i]);
 524                         free->lpage_info[i] = NULL;
 525                 }
 526         }
 527
 528         free->npages = 0;
 529         free->rmap = NULL;
 530 }
 531
 532 void kvm_free_physmem(struct kvm *kvm)
 533 {
 534         int i;
 535         struct kvm_memslots *slots = kvm->memslots;
 536
 537         for (i = 0; i < slots->nmemslots; ++i)
 538                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 539
 540         kfree(kvm->memslots);
 541 }
 542
 543 static void kvm_destroy_vm(struct kvm *kvm)
 544 {
 545         int i;
 546         struct mm_struct *mm = kvm->mm;
 547
 548         kvm_arch_sync_events(kvm);
 549         spin_lock(&kvm_lock);
 550         list_del(&kvm->vm_list);
 551         spin_unlock(&kvm_lock);
 552         kvm_free_irq_routing(kvm);
 553         for (i = 0; i < KVM_NR_BUSES; i++)
 554                 kvm_io_bus_destroy(kvm->buses[i]);
 555         kvm_coalesced_mmio_free(kvm);
 556 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 557         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 558 #else
 559         kvm_arch_flush_shadow(kvm);
 560 #endif
 561         kvm_arch_destroy_vm(kvm);
 562         kvm_free_physmem(kvm);
 563         cleanup_srcu_struct(&kvm->srcu);
 564         kvm_arch_free_vm(kvm);
 565         hardware_disable_all();
 566         mmdrop(mm);
 567 }
 568
 569 void kvm_get_kvm(struct kvm *kvm)
 570 {
 571         atomic_inc(&kvm->users_count);
 572 }
 573 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 574
 575 void kvm_put_kvm(struct kvm *kvm)
 576 {
 577         if (atomic_dec_and_test(&kvm->users_count))
 578                 kvm_destroy_vm(kvm);
 579 }
 580 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 581
 582
 583 static int kvm_vm_release(struct inode *inode, struct file *filp)
 584 {
 585         struct kvm *kvm = filp->private_data;
 586
 587         kvm_irqfd_release(kvm);
 588
 589         kvm_put_kvm(kvm);
 590         return 0;
 591 }
 592
 593 #ifndef CONFIG_S390
 594 /*
 595  * Allocation size is twice as large as the actual dirty bitmap size.
 596  * This makes it possible to do double buffering: see x86's
 597  * kvm_vm_ioctl_get_dirty_log().
 598  */
 599 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 600 {
 601         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 602
 603         if (dirty_bytes > PAGE_SIZE)
 604                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
 605         else
 606                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
 607
 608         if (!memslot->dirty_bitmap)
 609                 return -ENOMEM;
 610
 611         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
 612         return 0;
 613 }
 614 #endif /* !CONFIG_S390 */
 615
 616 /*
 617  * Allocate some memory and give it an address in the guest physical address
 618  * space.
 619  *
 620  * Discontiguous memory is allowed, mostly for framebuffers.
 621  *
 622  * Must be called holding mmap_sem for write.
 623  */
 624 int __kvm_set_memory_region(struct kvm *kvm,
 625                             struct kvm_userspace_memory_region *mem,
 626                             int user_alloc)
 627 {
 628         int r;
 629         gfn_t base_gfn;
 630         unsigned long npages;
 631         unsigned long i;
 632         struct kvm_memory_slot *memslot;
 633         struct kvm_memory_slot old, new;
 634         struct kvm_memslots *slots, *old_memslots;
 635
 636         r = -EINVAL;
 637         /* General sanity checks */
 638         if (mem->memory_size & (PAGE_SIZE - 1))
 639                 goto out;
 640         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 641                 goto out;
 642         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 643                 goto out;
 644         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 645                 goto out;
 646         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 647                 goto out;
 648
 649         memslot = &kvm->memslots->memslots[mem->slot];
 650         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 651         npages = mem->memory_size >> PAGE_SHIFT;
 652
 653         r = -EINVAL;
 654         if (npages > KVM_MEM_MAX_NR_PAGES)
 655                 goto out;
 656
 657         if (!npages)
 658                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 659
 660         new = old = *memslot;
 661
 662         new.id = mem->slot;
 663         new.base_gfn = base_gfn;
 664         new.npages = npages;
 665         new.flags = mem->flags;
 666
 667         /* Disallow changing a memory slot's size. */
 668         r = -EINVAL;
 669         if (npages && old.npages && npages != old.npages)
 670                 goto out_free;
 671
 672         /* Check for overlaps */
 673         r = -EEXIST;
 674         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 675                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 676
 677                 if (s == memslot || !s->npages)
 678                         continue;
 679                 if (!((base_gfn + npages <= s->base_gfn) ||
 680                       (base_gfn >= s->base_gfn + s->npages)))
 681                         goto out_free;
 682         }
 683
 684         /* Free page dirty bitmap if unneeded */
 685         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 686                 new.dirty_bitmap = NULL;
 687
 688         r = -ENOMEM;
 689
 690         /* Allocate if a slot is being created */
 691 #ifndef CONFIG_S390
 692         if (npages && !new.rmap) {
 693                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
 694
 695                 if (!new.rmap)
 696                         goto out_free;
 697
 698                 new.user_alloc = user_alloc;
 699                 new.userspace_addr = mem->userspace_addr;
 700         }
 701         if (!npages)
 702                 goto skip_lpage;
 703
 704         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 705                 unsigned long ugfn;
 706                 unsigned long j;
 707                 int lpages;
 708                 int level = i + 2;
 709
 710                 /* Avoid unused variable warning if no large pages */
 711                 (void)level;
 712
 713                 if (new.lpage_info[i])
 714                         continue;
 715
 716                 lpages = 1 + ((base_gfn + npages - 1)
 717                              >> KVM_HPAGE_GFN_SHIFT(level));
 718                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
 719
 720                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
 721
 722                 if (!new.lpage_info[i])
 723                         goto out_free;
 724
 725                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
 726                         new.lpage_info[i][0].write_count = 1;
 727                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
 728                         new.lpage_info[i][lpages - 1].write_count = 1;
 729                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 730                 /*
 731                  * If the gfn and userspace address are not aligned wrt each
 732                  * other, or if explicitly asked to, disable large page
 733                  * support for this slot
 734                  */
 735                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 736                     !largepages_enabled)
 737                         for (j = 0; j < lpages; ++j)
 738                                 new.lpage_info[i][j].write_count = 1;
 739         }
 740
 741 skip_lpage:
 742
 743         /* Allocate page dirty bitmap if needed */
 744         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 745                 if (kvm_create_dirty_bitmap(&new) < 0)
 746                         goto out_free;
 747                 /* destroy any largepage mappings for dirty tracking */
 748         }
 749 #else  /* not defined CONFIG_S390 */
 750         new.user_alloc = user_alloc;
 751         if (user_alloc)
 752                 new.userspace_addr = mem->userspace_addr;
 753 #endif /* not defined CONFIG_S390 */
 754
 755         if (!npages) {
 756                 r = -ENOMEM;
 757                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 758                 if (!slots)
 759                         goto out_free;
 760                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 761                 if (mem->slot >= slots->nmemslots)
 762                         slots->nmemslots = mem->slot + 1;
 763                 slots->generation++;
 764                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 765
 766                 old_memslots = kvm->memslots;
 767                 rcu_assign_pointer(kvm->memslots, slots);
 768                 synchronize_srcu_expedited(&kvm->srcu);
 769                 /* From this point no new shadow pages pointing to a deleted
 770                  * memslot will be created.
 771                  *
 772                  * validation of sp->gfn happens in:
 773                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 774                  *      - kvm_is_visible_gfn (mmu_check_roots)
 775                  */
 776                 kvm_arch_flush_shadow(kvm);
 777                 kfree(old_memslots);
 778         }
 779
 780         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 781         if (r)
 782                 goto out_free;
 783
 784         /* map the pages in iommu page table */
 785         if (npages) {
 786                 r = kvm_iommu_map_pages(kvm, &new);
 787                 if (r)
 788                         goto out_free;
 789         }
 790
 791         r = -ENOMEM;
 792         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 793         if (!slots)
 794                 goto out_free;
 795         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 796         if (mem->slot >= slots->nmemslots)
 797                 slots->nmemslots = mem->slot + 1;
 798         slots->generation++;
 799
 800         /* actual memory is freed via old in kvm_free_physmem_slot below */
 801         if (!npages) {
 802                 new.rmap = NULL;
 803                 new.dirty_bitmap = NULL;
 804                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 805                         new.lpage_info[i] = NULL;
 806         }
 807
 808         slots->memslots[mem->slot] = new;
 809         old_memslots = kvm->memslots;
 810         rcu_assign_pointer(kvm->memslots, slots);
 811         synchronize_srcu_expedited(&kvm->srcu);
 812
 813         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 814
 815         kvm_free_physmem_slot(&old, &new);
 816         kfree(old_memslots);
 817
 818         return 0;
 819
 820 out_free:
 821         kvm_free_physmem_slot(&new, &old);
 822 out:
 823         return r;
 824
 825 }
 826 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 827
 828 int kvm_set_memory_region(struct kvm *kvm,
 829                           struct kvm_userspace_memory_region *mem,
 830                           int user_alloc)
 831 {
 832         int r;
 833
 834         mutex_lock(&kvm->slots_lock);
 835         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 836         mutex_unlock(&kvm->slots_lock);
 837         return r;
 838 }
 839 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 840
 841 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 842                                    struct
 843                                    kvm_userspace_memory_region *mem,
 844                                    int user_alloc)
 845 {
 846         if (mem->slot >= KVM_MEMORY_SLOTS)
 847                 return -EINVAL;
 848         return kvm_set_memory_region(kvm, mem, user_alloc);
 849 }
 850
 851 int kvm_get_dirty_log(struct kvm *kvm,
 852                         struct kvm_dirty_log *log, int *is_dirty)
 853 {
 854         struct kvm_memory_slot *memslot;
 855         int r, i;
 856         unsigned long n;
 857         unsigned long any = 0;
 858
 859         r = -EINVAL;
 860         if (log->slot >= KVM_MEMORY_SLOTS)
 861                 goto out;
 862
 863         memslot = &kvm->memslots->memslots[log->slot];
 864         r = -ENOENT;
 865         if (!memslot->dirty_bitmap)
 866                 goto out;
 867
 868         n = kvm_dirty_bitmap_bytes(memslot);
 869
 870         for (i = 0; !any && i < n/sizeof(long); ++i)
 871                 any = memslot->dirty_bitmap[i];
 872
 873         r = -EFAULT;
 874         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 875                 goto out;
 876
 877         if (any)
 878                 *is_dirty = 1;
 879
 880         r = 0;
 881 out:
 882         return r;
 883 }
 884
 885 void kvm_disable_largepages(void)
 886 {
 887         largepages_enabled = false;
 888 }
 889 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 890
 891 int is_error_page(struct page *page)
 892 {
 893         return page == bad_page || page == hwpoison_page || page == fault_page;
 894 }
 895 EXPORT_SYMBOL_GPL(is_error_page);
 896
 897 int is_error_pfn(pfn_t pfn)
 898 {
 899         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 900 }
 901 EXPORT_SYMBOL_GPL(is_error_pfn);
 902
 903 int is_hwpoison_pfn(pfn_t pfn)
 904 {
 905         return pfn == hwpoison_pfn;
 906 }
 907 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 908
 909 int is_fault_pfn(pfn_t pfn)
 910 {
 911         return pfn == fault_pfn;
 912 }
 913 EXPORT_SYMBOL_GPL(is_fault_pfn);
 914
 915 static inline unsigned long bad_hva(void)
 916 {
 917         return PAGE_OFFSET;
 918 }
 919
 920 int kvm_is_error_hva(unsigned long addr)
 921 {
 922         return addr == bad_hva();
 923 }
 924 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 925
 926 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
 927                                                 gfn_t gfn)
 928 {
 929         int i;
 930
 931         for (i = 0; i < slots->nmemslots; ++i) {
 932                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 933
 934                 if (gfn >= memslot->base_gfn
 935                     && gfn < memslot->base_gfn + memslot->npages)
 936                         return memslot;
 937         }
 938         return NULL;
 939 }
 940
 941 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 942 {
 943         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 944 }
 945 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 946
 947 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 948 {
 949         int i;
 950         struct kvm_memslots *slots = kvm_memslots(kvm);
 951
 952         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 953                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 954
 955                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 956                         continue;
 957
 958                 if (gfn >= memslot->base_gfn
 959                     && gfn < memslot->base_gfn + memslot->npages)
 960                         return 1;
 961         }
 962         return 0;
 963 }
 964 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 965
 966 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 967 {
 968         struct vm_area_struct *vma;
 969         unsigned long addr, size;
 970
 971         size = PAGE_SIZE;
 972
 973         addr = gfn_to_hva(kvm, gfn);
 974         if (kvm_is_error_hva(addr))
 975                 return PAGE_SIZE;
 976
 977         down_read(&current->mm->mmap_sem);
 978         vma = find_vma(current->mm, addr);
 979         if (!vma)
 980                 goto out;
 981
 982         size = vma_kernel_pagesize(vma);
 983
 984 out:
 985         up_read(&current->mm->mmap_sem);
 986
 987         return size;
 988 }
 989
 990 int memslot_id(struct kvm *kvm, gfn_t gfn)
 991 {
 992         int i;
 993         struct kvm_memslots *slots = kvm_memslots(kvm);
 994         struct kvm_memory_slot *memslot = NULL;
 995
 996         for (i = 0; i < slots->nmemslots; ++i) {
 997                 memslot = &slots->memslots[i];
 998
 999                 if (gfn >= memslot->base_gfn
1000                     && gfn < memslot->base_gfn + memslot->npages)
1001                         break;
1002         }
1003
1004         return memslot - slots->memslots;
1005 }
1006
1007 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1008                                      gfn_t *nr_pages)
1009 {
1010         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1011                 return bad_hva();
1012
1013         if (nr_pages)
1014                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1015
1016         return gfn_to_hva_memslot(slot, gfn);
1017 }
1018
1019 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1020 {
1021         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1022 }
1023 EXPORT_SYMBOL_GPL(gfn_to_hva);
1024
1025 static pfn_t get_fault_pfn(void)
1026 {
1027         get_page(fault_page);
1028         return fault_pfn;
1029 }
1030
1031 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1032                         bool *async, bool write_fault, bool *writable)
1033 {
1034         struct page *page[1];
1035         int npages = 0;
1036         pfn_t pfn;
1037
1038         /* we can do it either atomically or asynchronously, not both */
1039         BUG_ON(atomic && async);
1040
1041         BUG_ON(!write_fault && !writable);
1042
1043         if (writable)
1044                 *writable = true;
1045
1046         if (atomic || async)
1047                 npages = __get_user_pages_fast(addr, 1, 1, page);
1048
1049         if (unlikely(npages != 1) && !atomic) {
1050                 might_sleep();
1051
1052                 if (writable)
1053                         *writable = write_fault;
1054
1055                 npages = get_user_pages_fast(addr, 1, write_fault, page);
1056
1057                 /* map read fault as writable if possible */
1058                 if (unlikely(!write_fault) && npages == 1) {
1059                         struct page *wpage[1];
1060
1061                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1062                         if (npages == 1) {
1063                                 *writable = true;
1064                                 put_page(page[0]);
1065                                 page[0] = wpage[0];
1066                         }
1067                         npages = 1;
1068                 }
1069         }
1070
1071         if (unlikely(npages != 1)) {
1072                 struct vm_area_struct *vma;
1073
1074                 if (atomic)
1075                         return get_fault_pfn();
1076
1077                 down_read(&current->mm->mmap_sem);
1078                 if (is_hwpoison_address(addr)) {
1079                         up_read(&current->mm->mmap_sem);
1080                         get_page(hwpoison_page);
1081                         return page_to_pfn(hwpoison_page);
1082                 }
1083
1084                 vma = find_vma_intersection(current->mm, addr, addr+1);
1085
1086                 if (vma == NULL)
1087                         pfn = get_fault_pfn();
1088                 else if ((vma->vm_flags & VM_PFNMAP)) {
1089                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1090                                 vma->vm_pgoff;
1091                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1092                 } else {
1093                         if (async && (vma->vm_flags & VM_WRITE))
1094                                 *async = true;
1095                         pfn = get_fault_pfn();
1096                 }
1097                 up_read(&current->mm->mmap_sem);
1098         } else
1099                 pfn = page_to_pfn(page[0]);
1100
1101         return pfn;
1102 }
1103
1104 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1105 {
1106         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1107 }
1108 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1109
1110 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1111                           bool write_fault, bool *writable)
1112 {
1113         unsigned long addr;
1114
1115         if (async)
1116                 *async = false;
1117
1118         addr = gfn_to_hva(kvm, gfn);
1119         if (kvm_is_error_hva(addr)) {
1120                 get_page(bad_page);
1121                 return page_to_pfn(bad_page);
1122         }
1123
1124         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1125 }
1126
1127 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1128 {
1129         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1130 }
1131 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1132
1133 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1134                        bool write_fault, bool *writable)
1135 {
1136         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1137 }
1138 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1139
1140 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1141 {
1142         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1143 }
1144 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1145
1146 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1147                       bool *writable)
1148 {
1149         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1150 }
1151 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1152
1153 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1154                          struct kvm_memory_slot *slot, gfn_t gfn)
1155 {
1156         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1157         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1158 }
1159
1160 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1161                                                                   int nr_pages)
1162 {
1163         unsigned long addr;
1164         gfn_t entry;
1165
1166         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1167         if (kvm_is_error_hva(addr))
1168                 return -1;
1169
1170         if (entry < nr_pages)
1171                 return 0;
1172
1173         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1174 }
1175 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1176
1177 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1178 {
1179         pfn_t pfn;
1180
1181         pfn = gfn_to_pfn(kvm, gfn);
1182         if (!kvm_is_mmio_pfn(pfn))
1183                 return pfn_to_page(pfn);
1184
1185         WARN_ON(kvm_is_mmio_pfn(pfn));
1186
1187         get_page(bad_page);
1188         return bad_page;
1189 }
1190
1191 EXPORT_SYMBOL_GPL(gfn_to_page);
1192
1193 void kvm_release_page_clean(struct page *page)
1194 {
1195         kvm_release_pfn_clean(page_to_pfn(page));
1196 }
1197 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1198
1199 void kvm_release_pfn_clean(pfn_t pfn)
1200 {
1201         if (!kvm_is_mmio_pfn(pfn))
1202                 put_page(pfn_to_page(pfn));
1203 }
1204 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1205
1206 void kvm_release_page_dirty(struct page *page)
1207 {
1208         kvm_release_pfn_dirty(page_to_pfn(page));
1209 }
1210 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1211
1212 void kvm_release_pfn_dirty(pfn_t pfn)
1213 {
1214         kvm_set_pfn_dirty(pfn);
1215         kvm_release_pfn_clean(pfn);
1216 }
1217 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1218
1219 void kvm_set_page_dirty(struct page *page)
1220 {
1221         kvm_set_pfn_dirty(page_to_pfn(page));
1222 }
1223 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1224
1225 void kvm_set_pfn_dirty(pfn_t pfn)
1226 {
1227         if (!kvm_is_mmio_pfn(pfn)) {
1228                 struct page *page = pfn_to_page(pfn);
1229                 if (!PageReserved(page))
1230                         SetPageDirty(page);
1231         }
1232 }
1233 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1234
1235 void kvm_set_pfn_accessed(pfn_t pfn)
1236 {
1237         if (!kvm_is_mmio_pfn(pfn))
1238                 mark_page_accessed(pfn_to_page(pfn));
1239 }
1240 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1241
1242 void kvm_get_pfn(pfn_t pfn)
1243 {
1244         if (!kvm_is_mmio_pfn(pfn))
1245                 get_page(pfn_to_page(pfn));
1246 }
1247 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1248
1249 static int next_segment(unsigned long len, int offset)
1250 {
1251         if (len > PAGE_SIZE - offset)
1252                 return PAGE_SIZE - offset;
1253         else
1254                 return len;
1255 }
1256
1257 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1258                         int len)
1259 {
1260         int r;
1261         unsigned long addr;
1262
1263         addr = gfn_to_hva(kvm, gfn);
1264         if (kvm_is_error_hva(addr))
1265                 return -EFAULT;
1266         r = copy_from_user(data, (void __user *)addr + offset, len);
1267         if (r)
1268                 return -EFAULT;
1269         return 0;
1270 }
1271 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1272
1273 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1274 {
1275         gfn_t gfn = gpa >> PAGE_SHIFT;
1276         int seg;
1277         int offset = offset_in_page(gpa);
1278         int ret;
1279
1280         while ((seg = next_segment(len, offset)) != 0) {
1281                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1282                 if (ret < 0)
1283                         return ret;
1284                 offset = 0;
1285                 len -= seg;
1286                 data += seg;
1287                 ++gfn;
1288         }
1289         return 0;
1290 }
1291 EXPORT_SYMBOL_GPL(kvm_read_guest);
1292
1293 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1294                           unsigned long len)
1295 {
1296         int r;
1297         unsigned long addr;
1298         gfn_t gfn = gpa >> PAGE_SHIFT;
1299         int offset = offset_in_page(gpa);
1300
1301         addr = gfn_to_hva(kvm, gfn);
1302         if (kvm_is_error_hva(addr))
1303                 return -EFAULT;
1304         pagefault_disable();
1305         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1306         pagefault_enable();
1307         if (r)
1308                 return -EFAULT;
1309         return 0;
1310 }
1311 EXPORT_SYMBOL(kvm_read_guest_atomic);
1312
1313 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1314                          int offset, int len)
1315 {
1316         int r;
1317         unsigned long addr;
1318
1319         addr = gfn_to_hva(kvm, gfn);
1320         if (kvm_is_error_hva(addr))
1321                 return -EFAULT;
1322         r = copy_to_user((void __user *)addr + offset, data, len);
1323         if (r)
1324                 return -EFAULT;
1325         mark_page_dirty(kvm, gfn);
1326         return 0;
1327 }
1328 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1329
1330 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1331                     unsigned long len)
1332 {
1333         gfn_t gfn = gpa >> PAGE_SHIFT;
1334         int seg;
1335         int offset = offset_in_page(gpa);
1336         int ret;
1337
1338         while ((seg = next_segment(len, offset)) != 0) {
1339                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1340                 if (ret < 0)
1341                         return ret;
1342                 offset = 0;
1343                 len -= seg;
1344                 data += seg;
1345                 ++gfn;
1346         }
1347         return 0;
1348 }
1349
1350 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1351                               gpa_t gpa)
1352 {
1353         struct kvm_memslots *slots = kvm_memslots(kvm);
1354         int offset = offset_in_page(gpa);
1355         gfn_t gfn = gpa >> PAGE_SHIFT;
1356
1357         ghc->gpa = gpa;
1358         ghc->generation = slots->generation;
1359         ghc->memslot = __gfn_to_memslot(slots, gfn);
1360         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1361         if (!kvm_is_error_hva(ghc->hva))
1362                 ghc->hva += offset;
1363         else
1364                 return -EFAULT;
1365
1366         return 0;
1367 }
1368 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1369
1370 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1371                            void *data, unsigned long len)
1372 {
1373         struct kvm_memslots *slots = kvm_memslots(kvm);
1374         int r;
1375
1376         if (slots->generation != ghc->generation)
1377                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1378
1379         if (kvm_is_error_hva(ghc->hva))
1380                 return -EFAULT;
1381
1382         r = copy_to_user((void __user *)ghc->hva, data, len);
1383         if (r)
1384                 return -EFAULT;
1385         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1386
1387         return 0;
1388 }
1389 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1390
1391 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1392 {
1393         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1394                                     offset, len);
1395 }
1396 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1397
1398 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1399 {
1400         gfn_t gfn = gpa >> PAGE_SHIFT;
1401         int seg;
1402         int offset = offset_in_page(gpa);
1403         int ret;
1404
1405         while ((seg = next_segment(len, offset)) != 0) {
1406                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1407                 if (ret < 0)
1408                         return ret;
1409                 offset = 0;
1410                 len -= seg;
1411                 ++gfn;
1412         }
1413         return 0;
1414 }
1415 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1416
1417 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1418                              gfn_t gfn)
1419 {
1420         if (memslot && memslot->dirty_bitmap) {
1421                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1422
1423                 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1424         }
1425 }
1426
1427 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1428 {
1429         struct kvm_memory_slot *memslot;
1430
1431         memslot = gfn_to_memslot(kvm, gfn);
1432         mark_page_dirty_in_slot(kvm, memslot, gfn);
1433 }
1434
1435 /*
1436  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1437  */
1438 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1439 {
1440         DEFINE_WAIT(wait);
1441
1442         for (;;) {
1443                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1444
1445                 if (kvm_arch_vcpu_runnable(vcpu)) {
1446                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1447                         break;
1448                 }
1449                 if (kvm_cpu_has_pending_timer(vcpu))
1450                         break;
1451                 if (signal_pending(current))
1452                         break;
1453
1454                 schedule();
1455         }
1456
1457         finish_wait(&vcpu->wq, &wait);
1458 }
1459
1460 void kvm_resched(struct kvm_vcpu *vcpu)
1461 {
1462         if (!need_resched())
1463                 return;
1464         cond_resched();
1465 }
1466 EXPORT_SYMBOL_GPL(kvm_resched);
1467
1468 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1469 {
1470         ktime_t expires;
1471         DEFINE_WAIT(wait);
1472
1473         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1474
1475         /* Sleep for 100 us, and hope lock-holder got scheduled */
1476         expires = ktime_add_ns(ktime_get(), 100000UL);
1477         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1478
1479         finish_wait(&vcpu->wq, &wait);
1480 }
1481 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1482
1483 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1484 {
1485         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1486         struct page *page;
1487
1488         if (vmf->pgoff == 0)
1489                 page = virt_to_page(vcpu->run);
1490 #ifdef CONFIG_X86
1491         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1492                 page = virt_to_page(vcpu->arch.pio_data);
1493 #endif
1494 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1495         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1496                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1497 #endif
1498         else
1499                 return VM_FAULT_SIGBUS;
1500         get_page(page);
1501         vmf->page = page;
1502         return 0;
1503 }
1504
1505 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1506         .fault = kvm_vcpu_fault,
1507 };
1508
1509 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1510 {
1511         vma->vm_ops = &kvm_vcpu_vm_ops;
1512         return 0;
1513 }
1514
1515 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1516 {
1517         struct kvm_vcpu *vcpu = filp->private_data;
1518
1519         kvm_put_kvm(vcpu->kvm);
1520         return 0;
1521 }
1522
1523 static struct file_operations kvm_vcpu_fops = {
1524         .release        = kvm_vcpu_release,
1525         .unlocked_ioctl = kvm_vcpu_ioctl,
1526         .compat_ioctl   = kvm_vcpu_ioctl,
1527         .mmap           = kvm_vcpu_mmap,
1528         .llseek         = noop_llseek,
1529 };
1530
1531 /*
1532  * Allocates an inode for the vcpu.
1533  */
1534 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1535 {
1536         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1537 }
1538
1539 /*
1540  * Creates some virtual cpus.  Good luck creating more than one.
1541  */
1542 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1543 {
1544         int r;
1545         struct kvm_vcpu *vcpu, *v;
1546
1547         vcpu = kvm_arch_vcpu_create(kvm, id);
1548         if (IS_ERR(vcpu))
1549                 return PTR_ERR(vcpu);
1550
1551         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1552
1553         r = kvm_arch_vcpu_setup(vcpu);
1554         if (r)
1555                 return r;
1556
1557         mutex_lock(&kvm->lock);
1558         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1559                 r = -EINVAL;
1560                 goto vcpu_destroy;
1561         }
1562
1563         kvm_for_each_vcpu(r, v, kvm)
1564                 if (v->vcpu_id == id) {
1565                         r = -EEXIST;
1566                         goto vcpu_destroy;
1567                 }
1568
1569         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1570
1571         /* Now it's all set up, let userspace reach it */
1572         kvm_get_kvm(kvm);
1573         r = create_vcpu_fd(vcpu);
1574         if (r < 0) {
1575                 kvm_put_kvm(kvm);
1576                 goto vcpu_destroy;
1577         }
1578
1579         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1580         smp_wmb();
1581         atomic_inc(&kvm->online_vcpus);
1582
1583 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1584         if (kvm->bsp_vcpu_id == id)
1585                 kvm->bsp_vcpu = vcpu;
1586 #endif
1587         mutex_unlock(&kvm->lock);
1588         return r;
1589
1590 vcpu_destroy:
1591         mutex_unlock(&kvm->lock);
1592         kvm_arch_vcpu_destroy(vcpu);
1593         return r;
1594 }
1595
1596 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1597 {
1598         if (sigset) {
1599                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1600                 vcpu->sigset_active = 1;
1601                 vcpu->sigset = *sigset;
1602         } else
1603                 vcpu->sigset_active = 0;
1604         return 0;
1605 }
1606
1607 static long kvm_vcpu_ioctl(struct file *filp,
1608                            unsigned int ioctl, unsigned long arg)
1609 {
1610         struct kvm_vcpu *vcpu = filp->private_data;
1611         void __user *argp = (void __user *)arg;
1612         int r;
1613         struct kvm_fpu *fpu = NULL;
1614         struct kvm_sregs *kvm_sregs = NULL;
1615
1616         if (vcpu->kvm->mm != current->mm)
1617                 return -EIO;
1618
1619 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1620         /*
1621          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1622          * so vcpu_load() would break it.
1623          */
1624         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1625                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1626 #endif
1627
1628
1629         vcpu_load(vcpu);
1630         switch (ioctl) {
1631         case KVM_RUN:
1632                 r = -EINVAL;
1633                 if (arg)
1634                         goto out;
1635                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1636                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1637                 break;
1638         case KVM_GET_REGS: {
1639                 struct kvm_regs *kvm_regs;
1640
1641                 r = -ENOMEM;
1642                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1643                 if (!kvm_regs)
1644                         goto out;
1645                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1646                 if (r)
1647                         goto out_free1;
1648                 r = -EFAULT;
1649                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1650                         goto out_free1;
1651                 r = 0;
1652 out_free1:
1653                 kfree(kvm_regs);
1654                 break;
1655         }
1656         case KVM_SET_REGS: {
1657                 struct kvm_regs *kvm_regs;
1658
1659                 r = -ENOMEM;
1660                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1661                 if (!kvm_regs)
1662                         goto out;
1663                 r = -EFAULT;
1664                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1665                         goto out_free2;
1666                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1667                 if (r)
1668                         goto out_free2;
1669                 r = 0;
1670 out_free2:
1671                 kfree(kvm_regs);
1672                 break;
1673         }
1674         case KVM_GET_SREGS: {
1675                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1676                 r = -ENOMEM;
1677                 if (!kvm_sregs)
1678                         goto out;
1679                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1680                 if (r)
1681                         goto out;
1682                 r = -EFAULT;
1683                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1684                         goto out;
1685                 r = 0;
1686                 break;
1687         }
1688         case KVM_SET_SREGS: {
1689                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1690                 r = -ENOMEM;
1691                 if (!kvm_sregs)
1692                         goto out;
1693                 r = -EFAULT;
1694                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1695                         goto out;
1696                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1697                 if (r)
1698                         goto out;
1699                 r = 0;
1700                 break;
1701         }
1702         case KVM_GET_MP_STATE: {
1703                 struct kvm_mp_state mp_state;
1704
1705                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1706                 if (r)
1707                         goto out;
1708                 r = -EFAULT;
1709                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1710                         goto out;
1711                 r = 0;
1712                 break;
1713         }
1714         case KVM_SET_MP_STATE: {
1715                 struct kvm_mp_state mp_state;
1716
1717                 r = -EFAULT;
1718                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1719                         goto out;
1720                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1721                 if (r)
1722                         goto out;
1723                 r = 0;
1724                 break;
1725         }
1726         case KVM_TRANSLATE: {
1727                 struct kvm_translation tr;
1728
1729                 r = -EFAULT;
1730                 if (copy_from_user(&tr, argp, sizeof tr))
1731                         goto out;
1732                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1733                 if (r)
1734                         goto out;
1735                 r = -EFAULT;
1736                 if (copy_to_user(argp, &tr, sizeof tr))
1737                         goto out;
1738                 r = 0;
1739                 break;
1740         }
1741         case KVM_SET_GUEST_DEBUG: {
1742                 struct kvm_guest_debug dbg;
1743
1744                 r = -EFAULT;
1745                 if (copy_from_user(&dbg, argp, sizeof dbg))
1746                         goto out;
1747                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1748                 if (r)
1749                         goto out;
1750                 r = 0;
1751                 break;
1752         }
1753         case KVM_SET_SIGNAL_MASK: {
1754                 struct kvm_signal_mask __user *sigmask_arg = argp;
1755                 struct kvm_signal_mask kvm_sigmask;
1756                 sigset_t sigset, *p;
1757
1758                 p = NULL;
1759                 if (argp) {
1760                         r = -EFAULT;
1761                         if (copy_from_user(&kvm_sigmask, argp,
1762                                            sizeof kvm_sigmask))
1763                                 goto out;
1764                         r = -EINVAL;
1765                         if (kvm_sigmask.len != sizeof sigset)
1766                                 goto out;
1767                         r = -EFAULT;
1768                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1769                                            sizeof sigset))
1770                                 goto out;
1771                         p = &sigset;
1772                 }
1773                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1774                 break;
1775         }
1776         case KVM_GET_FPU: {
1777                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1778                 r = -ENOMEM;
1779                 if (!fpu)
1780                         goto out;
1781                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1782                 if (r)
1783                         goto out;
1784                 r = -EFAULT;
1785                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1786                         goto out;
1787                 r = 0;
1788                 break;
1789         }
1790         case KVM_SET_FPU: {
1791                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1792                 r = -ENOMEM;
1793                 if (!fpu)
1794                         goto out;
1795                 r = -EFAULT;
1796                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1797                         goto out;
1798                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1799                 if (r)
1800                         goto out;
1801                 r = 0;
1802                 break;
1803         }
1804         default:
1805                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1806         }
1807 out:
1808         vcpu_put(vcpu);
1809         kfree(fpu);
1810         kfree(kvm_sregs);
1811         return r;
1812 }
1813
1814 static long kvm_vm_ioctl(struct file *filp,
1815                            unsigned int ioctl, unsigned long arg)
1816 {
1817         struct kvm *kvm = filp->private_data;
1818         void __user *argp = (void __user *)arg;
1819         int r;
1820
1821         if (kvm->mm != current->mm)
1822                 return -EIO;
1823         switch (ioctl) {
1824         case KVM_CREATE_VCPU:
1825                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1826                 if (r < 0)
1827                         goto out;
1828                 break;
1829         case KVM_SET_USER_MEMORY_REGION: {
1830                 struct kvm_userspace_memory_region kvm_userspace_mem;
1831
1832                 r = -EFAULT;
1833                 if (copy_from_user(&kvm_userspace_mem, argp,
1834                                                 sizeof kvm_userspace_mem))
1835                         goto out;
1836
1837                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1838                 if (r)
1839                         goto out;
1840                 break;
1841         }
1842         case KVM_GET_DIRTY_LOG: {
1843                 struct kvm_dirty_log log;
1844
1845                 r = -EFAULT;
1846                 if (copy_from_user(&log, argp, sizeof log))
1847                         goto out;
1848                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1849                 if (r)
1850                         goto out;
1851                 break;
1852         }
1853 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1854         case KVM_REGISTER_COALESCED_MMIO: {
1855                 struct kvm_coalesced_mmio_zone zone;
1856                 r = -EFAULT;
1857                 if (copy_from_user(&zone, argp, sizeof zone))
1858                         goto out;
1859                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1860                 if (r)
1861                         goto out;
1862                 r = 0;
1863                 break;
1864         }
1865         case KVM_UNREGISTER_COALESCED_MMIO: {
1866                 struct kvm_coalesced_mmio_zone zone;
1867                 r = -EFAULT;
1868                 if (copy_from_user(&zone, argp, sizeof zone))
1869                         goto out;
1870                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1871                 if (r)
1872                         goto out;
1873                 r = 0;
1874                 break;
1875         }
1876 #endif
1877         case KVM_IRQFD: {
1878                 struct kvm_irqfd data;
1879
1880                 r = -EFAULT;
1881                 if (copy_from_user(&data, argp, sizeof data))
1882                         goto out;
1883                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1884                 break;
1885         }
1886         case KVM_IOEVENTFD: {
1887                 struct kvm_ioeventfd data;
1888
1889                 r = -EFAULT;
1890                 if (copy_from_user(&data, argp, sizeof data))
1891                         goto out;
1892                 r = kvm_ioeventfd(kvm, &data);
1893                 break;
1894         }
1895 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1896         case KVM_SET_BOOT_CPU_ID:
1897                 r = 0;
1898                 mutex_lock(&kvm->lock);
1899                 if (atomic_read(&kvm->online_vcpus) != 0)
1900                         r = -EBUSY;
1901                 else
1902                         kvm->bsp_vcpu_id = arg;
1903                 mutex_unlock(&kvm->lock);
1904                 break;
1905 #endif
1906         default:
1907                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1908                 if (r == -ENOTTY)
1909                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1910         }
1911 out:
1912         return r;
1913 }
1914
1915 #ifdef CONFIG_COMPAT
1916 struct compat_kvm_dirty_log {
1917         __u32 slot;
1918         __u32 padding1;
1919         union {
1920                 compat_uptr_t dirty_bitmap; /* one bit per page */
1921                 __u64 padding2;
1922         };
1923 };
1924
1925 static long kvm_vm_compat_ioctl(struct file *filp,
1926                            unsigned int ioctl, unsigned long arg)
1927 {
1928         struct kvm *kvm = filp->private_data;
1929         int r;
1930
1931         if (kvm->mm != current->mm)
1932                 return -EIO;
1933         switch (ioctl) {
1934         case KVM_GET_DIRTY_LOG: {
1935                 struct compat_kvm_dirty_log compat_log;
1936                 struct kvm_dirty_log log;
1937
1938                 r = -EFAULT;
1939                 if (copy_from_user(&compat_log, (void __user *)arg,
1940                                    sizeof(compat_log)))
1941                         goto out;
1942                 log.slot         = compat_log.slot;
1943                 log.padding1     = compat_log.padding1;
1944                 log.padding2     = compat_log.padding2;
1945                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1946
1947                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1948                 if (r)
1949                         goto out;
1950                 break;
1951         }
1952         default:
1953                 r = kvm_vm_ioctl(filp, ioctl, arg);
1954         }
1955
1956 out:
1957         return r;
1958 }
1959 #endif
1960
1961 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1962 {
1963         struct page *page[1];
1964         unsigned long addr;
1965         int npages;
1966         gfn_t gfn = vmf->pgoff;
1967         struct kvm *kvm = vma->vm_file->private_data;
1968
1969         addr = gfn_to_hva(kvm, gfn);
1970         if (kvm_is_error_hva(addr))
1971                 return VM_FAULT_SIGBUS;
1972
1973         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1974                                 NULL);
1975         if (unlikely(npages != 1))
1976                 return VM_FAULT_SIGBUS;
1977
1978         vmf->page = page[0];
1979         return 0;
1980 }
1981
1982 static const struct vm_operations_struct kvm_vm_vm_ops = {
1983         .fault = kvm_vm_fault,
1984 };
1985
1986 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1987 {
1988         vma->vm_ops = &kvm_vm_vm_ops;
1989         return 0;
1990 }
1991
1992 static struct file_operations kvm_vm_fops = {
1993         .release        = kvm_vm_release,
1994         .unlocked_ioctl = kvm_vm_ioctl,
1995 #ifdef CONFIG_COMPAT
1996         .compat_ioctl   = kvm_vm_compat_ioctl,
1997 #endif
1998         .mmap           = kvm_vm_mmap,
1999         .llseek         = noop_llseek,
2000 };
2001
2002 static int kvm_dev_ioctl_create_vm(void)
2003 {
2004         int r;
2005         struct kvm *kvm;
2006
2007         kvm = kvm_create_vm();
2008         if (IS_ERR(kvm))
2009                 return PTR_ERR(kvm);
2010 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2011         r = kvm_coalesced_mmio_init(kvm);
2012         if (r < 0) {
2013                 kvm_put_kvm(kvm);
2014                 return r;
2015         }
2016 #endif
2017         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2018         if (r < 0)
2019                 kvm_put_kvm(kvm);
2020
2021         return r;
2022 }
2023
2024 static long kvm_dev_ioctl_check_extension_generic(long arg)
2025 {
2026         switch (arg) {
2027         case KVM_CAP_USER_MEMORY:
2028         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2029         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2030 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2031         case KVM_CAP_SET_BOOT_CPU_ID:
2032 #endif
2033         case KVM_CAP_INTERNAL_ERROR_DATA:
2034                 return 1;
2035 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2036         case KVM_CAP_IRQ_ROUTING:
2037                 return KVM_MAX_IRQ_ROUTES;
2038 #endif
2039         default:
2040                 break;
2041         }
2042         return kvm_dev_ioctl_check_extension(arg);
2043 }
2044
2045 static long kvm_dev_ioctl(struct file *filp,
2046                           unsigned int ioctl, unsigned long arg)
2047 {
2048         long r = -EINVAL;
2049
2050         switch (ioctl) {
2051         case KVM_GET_API_VERSION:
2052                 r = -EINVAL;
2053                 if (arg)
2054                         goto out;
2055                 r = KVM_API_VERSION;
2056                 break;
2057         case KVM_CREATE_VM:
2058                 r = -EINVAL;
2059                 if (arg)
2060                         goto out;
2061                 r = kvm_dev_ioctl_create_vm();
2062                 break;
2063         case KVM_CHECK_EXTENSION:
2064                 r = kvm_dev_ioctl_check_extension_generic(arg);
2065                 break;
2066         case KVM_GET_VCPU_MMAP_SIZE:
2067                 r = -EINVAL;
2068                 if (arg)
2069                         goto out;
2070                 r = PAGE_SIZE;     /* struct kvm_run */
2071 #ifdef CONFIG_X86
2072                 r += PAGE_SIZE;    /* pio data page */
2073 #endif
2074 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2075                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2076 #endif
2077                 break;
2078         case KVM_TRACE_ENABLE:
2079         case KVM_TRACE_PAUSE:
2080         case KVM_TRACE_DISABLE:
2081                 r = -EOPNOTSUPP;
2082                 break;
2083         default:
2084                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2085         }
2086 out:
2087         return r;
2088 }
2089
2090 static struct file_operations kvm_chardev_ops = {
2091         .unlocked_ioctl = kvm_dev_ioctl,
2092         .compat_ioctl   = kvm_dev_ioctl,
2093         .llseek         = noop_llseek,
2094 };
2095
2096 static struct miscdevice kvm_dev = {
2097         KVM_MINOR,
2098         "kvm",
2099         &kvm_chardev_ops,
2100 };
2101
2102 static void hardware_enable_nolock(void *junk)
2103 {
2104         int cpu = raw_smp_processor_id();
2105         int r;
2106
2107         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2108                 return;
2109
2110         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2111
2112         r = kvm_arch_hardware_enable(NULL);
2113
2114         if (r) {
2115                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2116                 atomic_inc(&hardware_enable_failed);
2117                 printk(KERN_INFO "kvm: enabling virtualization on "
2118                                  "CPU%d failed\n", cpu);
2119         }
2120 }
2121
2122 static void hardware_enable(void *junk)
2123 {
2124         spin_lock(&kvm_lock);
2125         hardware_enable_nolock(junk);
2126         spin_unlock(&kvm_lock);
2127 }
2128
2129 static void hardware_disable_nolock(void *junk)
2130 {
2131         int cpu = raw_smp_processor_id();
2132
2133         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2134                 return;
2135         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2136         kvm_arch_hardware_disable(NULL);
2137 }
2138
2139 static void hardware_disable(void *junk)
2140 {
2141         spin_lock(&kvm_lock);
2142         hardware_disable_nolock(junk);
2143         spin_unlock(&kvm_lock);
2144 }
2145
2146 static void hardware_disable_all_nolock(void)
2147 {
2148         BUG_ON(!kvm_usage_count);
2149
2150         kvm_usage_count--;
2151         if (!kvm_usage_count)
2152                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2153 }
2154
2155 static void hardware_disable_all(void)
2156 {
2157         spin_lock(&kvm_lock);
2158         hardware_disable_all_nolock();
2159         spin_unlock(&kvm_lock);
2160 }
2161
2162 static int hardware_enable_all(void)
2163 {
2164         int r = 0;
2165
2166         spin_lock(&kvm_lock);
2167
2168         kvm_usage_count++;
2169         if (kvm_usage_count == 1) {
2170                 atomic_set(&hardware_enable_failed, 0);
2171                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2172
2173                 if (atomic_read(&hardware_enable_failed)) {
2174                         hardware_disable_all_nolock();
2175                         r = -EBUSY;
2176                 }
2177         }
2178
2179         spin_unlock(&kvm_lock);
2180
2181         return r;
2182 }
2183
2184 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2185                            void *v)
2186 {
2187         int cpu = (long)v;
2188
2189         if (!kvm_usage_count)
2190                 return NOTIFY_OK;
2191
2192         val &= ~CPU_TASKS_FROZEN;
2193         switch (val) {
2194         case CPU_DYING:
2195                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2196                        cpu);
2197                 hardware_disable(NULL);
2198                 break;
2199         case CPU_STARTING:
2200                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2201                        cpu);
2202                 hardware_enable(NULL);
2203                 break;
2204         }
2205         return NOTIFY_OK;
2206 }
2207
2208
2209 asmlinkage void kvm_spurious_fault(void)
2210 {
2211         /* Fault while not rebooting.  We want the trace. */
2212         BUG();
2213 }
2214 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2215
2216 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2217                       void *v)
2218 {
2219         /*
2220          * Some (well, at least mine) BIOSes hang on reboot if
2221          * in vmx root mode.
2222          *
2223          * And Intel TXT required VMX off for all cpu when system shutdown.
2224          */
2225         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2226         kvm_rebooting = true;
2227         on_each_cpu(hardware_disable_nolock, NULL, 1);
2228         return NOTIFY_OK;
2229 }
2230
2231 static struct notifier_block kvm_reboot_notifier = {
2232         .notifier_call = kvm_reboot,
2233         .priority = 0,
2234 };
2235
2236 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2237 {
2238         int i;
2239
2240         for (i = 0; i < bus->dev_count; i++) {
2241                 struct kvm_io_device *pos = bus->devs[i];
2242
2243                 kvm_iodevice_destructor(pos);
2244         }
2245         kfree(bus);
2246 }
2247
2248 /* kvm_io_bus_write - called under kvm->slots_lock */
2249 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2250                      int len, const void *val)
2251 {
2252         int i;
2253         struct kvm_io_bus *bus;
2254
2255         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2256         for (i = 0; i < bus->dev_count; i++)
2257                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2258                         return 0;
2259         return -EOPNOTSUPP;
2260 }
2261
2262 /* kvm_io_bus_read - called under kvm->slots_lock */
2263 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2264                     int len, void *val)
2265 {
2266         int i;
2267         struct kvm_io_bus *bus;
2268
2269         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2270         for (i = 0; i < bus->dev_count; i++)
2271                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2272                         return 0;
2273         return -EOPNOTSUPP;
2274 }
2275
2276 /* Caller must hold slots_lock. */
2277 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2278                             struct kvm_io_device *dev)
2279 {
2280         struct kvm_io_bus *new_bus, *bus;
2281
2282         bus = kvm->buses[bus_idx];
2283         if (bus->dev_count > NR_IOBUS_DEVS-1)
2284                 return -ENOSPC;
2285
2286         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2287         if (!new_bus)
2288                 return -ENOMEM;
2289         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2290         new_bus->devs[new_bus->dev_count++] = dev;
2291         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2292         synchronize_srcu_expedited(&kvm->srcu);
2293         kfree(bus);
2294
2295         return 0;
2296 }
2297
2298 /* Caller must hold slots_lock. */
2299 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2300                               struct kvm_io_device *dev)
2301 {
2302         int i, r;
2303         struct kvm_io_bus *new_bus, *bus;
2304
2305         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2306         if (!new_bus)
2307                 return -ENOMEM;
2308
2309         bus = kvm->buses[bus_idx];
2310         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2311
2312         r = -ENOENT;
2313         for (i = 0; i < new_bus->dev_count; i++)
2314                 if (new_bus->devs[i] == dev) {
2315                         r = 0;
2316                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2317                         break;
2318                 }
2319
2320         if (r) {
2321                 kfree(new_bus);
2322                 return r;
2323         }
2324
2325         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2326         synchronize_srcu_expedited(&kvm->srcu);
2327         kfree(bus);
2328         return r;
2329 }
2330
2331 static struct notifier_block kvm_cpu_notifier = {
2332         .notifier_call = kvm_cpu_hotplug,
2333 };
2334
2335 static int vm_stat_get(void *_offset, u64 *val)
2336 {
2337         unsigned offset = (long)_offset;
2338         struct kvm *kvm;
2339
2340         *val = 0;
2341         spin_lock(&kvm_lock);
2342         list_for_each_entry(kvm, &vm_list, vm_list)
2343                 *val += *(u32 *)((void *)kvm + offset);
2344         spin_unlock(&kvm_lock);
2345         return 0;
2346 }
2347
2348 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2349
2350 static int vcpu_stat_get(void *_offset, u64 *val)
2351 {
2352         unsigned offset = (long)_offset;
2353         struct kvm *kvm;
2354         struct kvm_vcpu *vcpu;
2355         int i;
2356
2357         *val = 0;
2358         spin_lock(&kvm_lock);
2359         list_for_each_entry(kvm, &vm_list, vm_list)
2360                 kvm_for_each_vcpu(i, vcpu, kvm)
2361                         *val += *(u32 *)((void *)vcpu + offset);
2362
2363         spin_unlock(&kvm_lock);
2364         return 0;
2365 }
2366
2367 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2368
2369 static const struct file_operations *stat_fops[] = {
2370         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2371         [KVM_STAT_VM]   = &vm_stat_fops,
2372 };
2373
2374 static void kvm_init_debug(void)
2375 {
2376         struct kvm_stats_debugfs_item *p;
2377
2378         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2379         for (p = debugfs_entries; p->name; ++p)
2380                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2381                                                 (void *)(long)p->offset,
2382                                                 stat_fops[p->kind]);
2383 }
2384
2385 static void kvm_exit_debug(void)
2386 {
2387         struct kvm_stats_debugfs_item *p;
2388
2389         for (p = debugfs_entries; p->name; ++p)
2390                 debugfs_remove(p->dentry);
2391         debugfs_remove(kvm_debugfs_dir);
2392 }
2393
2394 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2395 {
2396         if (kvm_usage_count)
2397                 hardware_disable_nolock(NULL);
2398         return 0;
2399 }
2400
2401 static int kvm_resume(struct sys_device *dev)
2402 {
2403         if (kvm_usage_count) {
2404                 WARN_ON(spin_is_locked(&kvm_lock));
2405                 hardware_enable_nolock(NULL);
2406         }
2407         return 0;
2408 }
2409
2410 static struct sysdev_class kvm_sysdev_class = {
2411         .name = "kvm",
2412         .suspend = kvm_suspend,
2413         .resume = kvm_resume,
2414 };
2415
2416 static struct sys_device kvm_sysdev = {
2417         .id = 0,
2418         .cls = &kvm_sysdev_class,
2419 };
2420
2421 struct page *bad_page;
2422 pfn_t bad_pfn;
2423
2424 static inline
2425 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2426 {
2427         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2428 }
2429
2430 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2431 {
2432         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2433
2434         kvm_arch_vcpu_load(vcpu, cpu);
2435 }
2436
2437 static void kvm_sched_out(struct preempt_notifier *pn,
2438                           struct task_struct *next)
2439 {
2440         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2441
2442         kvm_arch_vcpu_put(vcpu);
2443 }
2444
2445 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2446                   struct module *module)
2447 {
2448         int r;
2449         int cpu;
2450
2451         r = kvm_arch_init(opaque);
2452         if (r)
2453                 goto out_fail;
2454
2455         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2456
2457         if (bad_page == NULL) {
2458                 r = -ENOMEM;
2459                 goto out;
2460         }
2461
2462         bad_pfn = page_to_pfn(bad_page);
2463
2464         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2465
2466         if (hwpoison_page == NULL) {
2467                 r = -ENOMEM;
2468                 goto out_free_0;
2469         }
2470
2471         hwpoison_pfn = page_to_pfn(hwpoison_page);
2472
2473         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2474
2475         if (fault_page == NULL) {
2476                 r = -ENOMEM;
2477                 goto out_free_0;
2478         }
2479
2480         fault_pfn = page_to_pfn(fault_page);
2481
2482         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2483                 r = -ENOMEM;
2484                 goto out_free_0;
2485         }
2486
2487         r = kvm_arch_hardware_setup();
2488         if (r < 0)
2489                 goto out_free_0a;
2490
2491         for_each_online_cpu(cpu) {
2492                 smp_call_function_single(cpu,
2493                                 kvm_arch_check_processor_compat,
2494                                 &r, 1);
2495                 if (r < 0)
2496                         goto out_free_1;
2497         }
2498
2499         r = register_cpu_notifier(&kvm_cpu_notifier);
2500         if (r)
2501                 goto out_free_2;
2502         register_reboot_notifier(&kvm_reboot_notifier);
2503
2504         r = sysdev_class_register(&kvm_sysdev_class);
2505         if (r)
2506                 goto out_free_3;
2507
2508         r = sysdev_register(&kvm_sysdev);
2509         if (r)
2510                 goto out_free_4;
2511
2512         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2513         if (!vcpu_align)
2514                 vcpu_align = __alignof__(struct kvm_vcpu);
2515         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2516                                            0, NULL);
2517         if (!kvm_vcpu_cache) {
2518                 r = -ENOMEM;
2519                 goto out_free_5;
2520         }
2521
2522         r = kvm_async_pf_init();
2523         if (r)
2524                 goto out_free;
2525
2526         kvm_chardev_ops.owner = module;
2527         kvm_vm_fops.owner = module;
2528         kvm_vcpu_fops.owner = module;
2529
2530         r = misc_register(&kvm_dev);
2531         if (r) {
2532                 printk(KERN_ERR "kvm: misc device register failed\n");
2533                 goto out_unreg;
2534         }
2535
2536         kvm_preempt_ops.sched_in = kvm_sched_in;
2537         kvm_preempt_ops.sched_out = kvm_sched_out;
2538
2539         kvm_init_debug();
2540
2541         return 0;
2542
2543 out_unreg:
2544         kvm_async_pf_deinit();
2545 out_free:
2546         kmem_cache_destroy(kvm_vcpu_cache);
2547 out_free_5:
2548         sysdev_unregister(&kvm_sysdev);
2549 out_free_4:
2550         sysdev_class_unregister(&kvm_sysdev_class);
2551 out_free_3:
2552         unregister_reboot_notifier(&kvm_reboot_notifier);
2553         unregister_cpu_notifier(&kvm_cpu_notifier);
2554 out_free_2:
2555 out_free_1:
2556         kvm_arch_hardware_unsetup();
2557 out_free_0a:
2558         free_cpumask_var(cpus_hardware_enabled);
2559 out_free_0:
2560         if (fault_page)
2561                 __free_page(fault_page);
2562         if (hwpoison_page)
2563                 __free_page(hwpoison_page);
2564         __free_page(bad_page);
2565 out:
2566         kvm_arch_exit();
2567 out_fail:
2568         return r;
2569 }
2570 EXPORT_SYMBOL_GPL(kvm_init);
2571
2572 void kvm_exit(void)
2573 {
2574         kvm_exit_debug();
2575         misc_deregister(&kvm_dev);
2576         kmem_cache_destroy(kvm_vcpu_cache);
2577         kvm_async_pf_deinit();
2578         sysdev_unregister(&kvm_sysdev);
2579         sysdev_class_unregister(&kvm_sysdev_class);
2580         unregister_reboot_notifier(&kvm_reboot_notifier);
2581         unregister_cpu_notifier(&kvm_cpu_notifier);
2582         on_each_cpu(hardware_disable_nolock, NULL, 1);
2583         kvm_arch_hardware_unsetup();
2584         kvm_arch_exit();
2585         free_cpumask_var(cpus_hardware_enabled);
2586         __free_page(hwpoison_page);
2587         __free_page(bad_page);
2588 }
2589 EXPORT_SYMBOL_GPL(kvm_exit);