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         raw_spin_lock(&kvm->requests_lock);
 169         me = smp_processor_id();
 170         kvm_for_each_vcpu(i, vcpu, kvm) {
 171                 if (kvm_make_check_request(req, vcpu))
 172                         continue;
 173                 cpu = vcpu->cpu;
 174                 if (cpus != NULL && cpu != -1 && cpu != me)
 175                         cpumask_set_cpu(cpu, cpus);
 176         }
 177         if (unlikely(cpus == NULL))
 178                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 179         else if (!cpumask_empty(cpus))
 180                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 181         else
 182                 called = false;
 183         raw_spin_unlock(&kvm->requests_lock);
 184         free_cpumask_var(cpus);
 185         return called;
 186 }
 187
 188 void kvm_flush_remote_tlbs(struct kvm *kvm)
 189 {
 190         int dirty_count = kvm->tlbs_dirty;
 191
 192         smp_mb();
 193         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 194                 ++kvm->stat.remote_tlb_flush;
 195         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 196 }
 197
 198 void kvm_reload_remote_mmus(struct kvm *kvm)
 199 {
 200         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 201 }
 202
 203 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 204 {
 205         struct page *page;
 206         int r;
 207
 208         mutex_init(&vcpu->mutex);
 209         vcpu->cpu = -1;
 210         vcpu->kvm = kvm;
 211         vcpu->vcpu_id = id;
 212         init_waitqueue_head(&vcpu->wq);
 213         kvm_async_pf_vcpu_init(vcpu);
 214
 215         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 216         if (!page) {
 217                 r = -ENOMEM;
 218                 goto fail;
 219         }
 220         vcpu->run = page_address(page);
 221
 222         r = kvm_arch_vcpu_init(vcpu);
 223         if (r < 0)
 224                 goto fail_free_run;
 225         return 0;
 226
 227 fail_free_run:
 228         free_page((unsigned long)vcpu->run);
 229 fail:
 230         return r;
 231 }
 232 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 233
 234 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 235 {
 236         kvm_arch_vcpu_uninit(vcpu);
 237         free_page((unsigned long)vcpu->run);
 238 }
 239 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 240
 241 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 242 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 243 {
 244         return container_of(mn, struct kvm, mmu_notifier);
 245 }
 246
 247 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 248                                              struct mm_struct *mm,
 249                                              unsigned long address)
 250 {
 251         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 252         int need_tlb_flush, idx;
 253
 254         /*
 255          * When ->invalidate_page runs, the linux pte has been zapped
 256          * already but the page is still allocated until
 257          * ->invalidate_page returns. So if we increase the sequence
 258          * here the kvm page fault will notice if the spte can't be
 259          * established because the page is going to be freed. If
 260          * instead the kvm page fault establishes the spte before
 261          * ->invalidate_page runs, kvm_unmap_hva will release it
 262          * before returning.
 263          *
 264          * The sequence increase only need to be seen at spin_unlock
 265          * time, and not at spin_lock time.
 266          *
 267          * Increasing the sequence after the spin_unlock would be
 268          * unsafe because the kvm page fault could then establish the
 269          * pte after kvm_unmap_hva returned, without noticing the page
 270          * is going to be freed.
 271          */
 272         idx = srcu_read_lock(&kvm->srcu);
 273         spin_lock(&kvm->mmu_lock);
 274         kvm->mmu_notifier_seq++;
 275         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 276         spin_unlock(&kvm->mmu_lock);
 277         srcu_read_unlock(&kvm->srcu, idx);
 278
 279         /* we've to flush the tlb before the pages can be freed */
 280         if (need_tlb_flush)
 281                 kvm_flush_remote_tlbs(kvm);
 282
 283 }
 284
 285 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 286                                         struct mm_struct *mm,
 287                                         unsigned long address,
 288                                         pte_t pte)
 289 {
 290         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 291         int idx;
 292
 293         idx = srcu_read_lock(&kvm->srcu);
 294         spin_lock(&kvm->mmu_lock);
 295         kvm->mmu_notifier_seq++;
 296         kvm_set_spte_hva(kvm, address, pte);
 297         spin_unlock(&kvm->mmu_lock);
 298         srcu_read_unlock(&kvm->srcu, idx);
 299 }
 300
 301 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 302                                                     struct mm_struct *mm,
 303                                                     unsigned long start,
 304                                                     unsigned long end)
 305 {
 306         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 307         int need_tlb_flush = 0, idx;
 308
 309         idx = srcu_read_lock(&kvm->srcu);
 310         spin_lock(&kvm->mmu_lock);
 311         /*
 312          * The count increase must become visible at unlock time as no
 313          * spte can be established without taking the mmu_lock and
 314          * count is also read inside the mmu_lock critical section.
 315          */
 316         kvm->mmu_notifier_count++;
 317         for (; start < end; start += PAGE_SIZE)
 318                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 319         need_tlb_flush |= kvm->tlbs_dirty;
 320         spin_unlock(&kvm->mmu_lock);
 321         srcu_read_unlock(&kvm->srcu, idx);
 322
 323         /* we've to flush the tlb before the pages can be freed */
 324         if (need_tlb_flush)
 325                 kvm_flush_remote_tlbs(kvm);
 326 }
 327
 328 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 329                                                   struct mm_struct *mm,
 330                                                   unsigned long start,
 331                                                   unsigned long end)
 332 {
 333         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 334
 335         spin_lock(&kvm->mmu_lock);
 336         /*
 337          * This sequence increase will notify the kvm page fault that
 338          * the page that is going to be mapped in the spte could have
 339          * been freed.
 340          */
 341         kvm->mmu_notifier_seq++;
 342         /*
 343          * The above sequence increase must be visible before the
 344          * below count decrease but both values are read by the kvm
 345          * page fault under mmu_lock spinlock so we don't need to add
 346          * a smb_wmb() here in between the two.
 347          */
 348         kvm->mmu_notifier_count--;
 349         spin_unlock(&kvm->mmu_lock);
 350
 351         BUG_ON(kvm->mmu_notifier_count < 0);
 352 }
 353
 354 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 355                                               struct mm_struct *mm,
 356                                               unsigned long address)
 357 {
 358         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 359         int young, idx;
 360
 361         idx = srcu_read_lock(&kvm->srcu);
 362         spin_lock(&kvm->mmu_lock);
 363         young = kvm_age_hva(kvm, address);
 364         spin_unlock(&kvm->mmu_lock);
 365         srcu_read_unlock(&kvm->srcu, idx);
 366
 367         if (young)
 368                 kvm_flush_remote_tlbs(kvm);
 369
 370         return young;
 371 }
 372
 373 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 374                                        struct mm_struct *mm,
 375                                        unsigned long address)
 376 {
 377         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 378         int young, idx;
 379
 380         idx = srcu_read_lock(&kvm->srcu);
 381         spin_lock(&kvm->mmu_lock);
 382         young = kvm_test_age_hva(kvm, address);
 383         spin_unlock(&kvm->mmu_lock);
 384         srcu_read_unlock(&kvm->srcu, idx);
 385
 386         return young;
 387 }
 388
 389 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 390                                      struct mm_struct *mm)
 391 {
 392         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 393         int idx;
 394
 395         idx = srcu_read_lock(&kvm->srcu);
 396         kvm_arch_flush_shadow(kvm);
 397         srcu_read_unlock(&kvm->srcu, idx);
 398 }
 399
 400 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 401         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 402         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 403         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 404         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 405         .test_young             = kvm_mmu_notifier_test_young,
 406         .change_pte             = kvm_mmu_notifier_change_pte,
 407         .release                = kvm_mmu_notifier_release,
 408 };
 409
 410 static int kvm_init_mmu_notifier(struct kvm *kvm)
 411 {
 412         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 413         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 414 }
 415
 416 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 417
 418 static int kvm_init_mmu_notifier(struct kvm *kvm)
 419 {
 420         return 0;
 421 }
 422
 423 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 424
 425 static struct kvm *kvm_create_vm(void)
 426 {
 427         int r, i;
 428         struct kvm *kvm = kvm_arch_alloc_vm();
 429
 430         if (!kvm)
 431                 return ERR_PTR(-ENOMEM);
 432
 433         r = kvm_arch_init_vm(kvm);
 434         if (r)
 435                 goto out_err_nodisable;
 436
 437         r = hardware_enable_all();
 438         if (r)
 439                 goto out_err_nodisable;
 440
 441 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 442         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 443         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 444 #endif
 445
 446         r = -ENOMEM;
 447         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 448         if (!kvm->memslots)
 449                 goto out_err_nosrcu;
 450         if (init_srcu_struct(&kvm->srcu))
 451                 goto out_err_nosrcu;
 452         for (i = 0; i < KVM_NR_BUSES; i++) {
 453                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 454                                         GFP_KERNEL);
 455                 if (!kvm->buses[i])
 456                         goto out_err;
 457         }
 458
 459         r = kvm_init_mmu_notifier(kvm);
 460         if (r)
 461                 goto out_err;
 462
 463         kvm->mm = current->mm;
 464         atomic_inc(&kvm->mm->mm_count);
 465         spin_lock_init(&kvm->mmu_lock);
 466         raw_spin_lock_init(&kvm->requests_lock);
 467         kvm_eventfd_init(kvm);
 468         mutex_init(&kvm->lock);
 469         mutex_init(&kvm->irq_lock);
 470         mutex_init(&kvm->slots_lock);
 471         atomic_set(&kvm->users_count, 1);
 472         spin_lock(&kvm_lock);
 473         list_add(&kvm->vm_list, &vm_list);
 474         spin_unlock(&kvm_lock);
 475
 476         return kvm;
 477
 478 out_err:
 479         cleanup_srcu_struct(&kvm->srcu);
 480 out_err_nosrcu:
 481         hardware_disable_all();
 482 out_err_nodisable:
 483         for (i = 0; i < KVM_NR_BUSES; i++)
 484                 kfree(kvm->buses[i]);
 485         kfree(kvm->memslots);
 486         kvm_arch_free_vm(kvm);
 487         return ERR_PTR(r);
 488 }
 489
 490 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 491 {
 492         if (!memslot->dirty_bitmap)
 493                 return;
 494
 495         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
 496                 vfree(memslot->dirty_bitmap_head);
 497         else
 498                 kfree(memslot->dirty_bitmap_head);
 499
 500         memslot->dirty_bitmap = NULL;
 501         memslot->dirty_bitmap_head = NULL;
 502 }
 503
 504 /*
 505  * Free any memory in @free but not in @dont.
 506  */
 507 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 508                                   struct kvm_memory_slot *dont)
 509 {
 510         int i;
 511
 512         if (!dont || free->rmap != dont->rmap)
 513                 vfree(free->rmap);
 514
 515         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 516                 kvm_destroy_dirty_bitmap(free);
 517
 518
 519         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 520                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 521                         vfree(free->lpage_info[i]);
 522                         free->lpage_info[i] = NULL;
 523                 }
 524         }
 525
 526         free->npages = 0;
 527         free->rmap = NULL;
 528 }
 529
 530 void kvm_free_physmem(struct kvm *kvm)
 531 {
 532         int i;
 533         struct kvm_memslots *slots = kvm->memslots;
 534
 535         for (i = 0; i < slots->nmemslots; ++i)
 536                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 537
 538         kfree(kvm->memslots);
 539 }
 540
 541 static void kvm_destroy_vm(struct kvm *kvm)
 542 {
 543         int i;
 544         struct mm_struct *mm = kvm->mm;
 545
 546         kvm_arch_sync_events(kvm);
 547         spin_lock(&kvm_lock);
 548         list_del(&kvm->vm_list);
 549         spin_unlock(&kvm_lock);
 550         kvm_free_irq_routing(kvm);
 551         for (i = 0; i < KVM_NR_BUSES; i++)
 552                 kvm_io_bus_destroy(kvm->buses[i]);
 553         kvm_coalesced_mmio_free(kvm);
 554 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 555         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 556 #else
 557         kvm_arch_flush_shadow(kvm);
 558 #endif
 559         kvm_arch_destroy_vm(kvm);
 560         kvm_free_physmem(kvm);
 561         cleanup_srcu_struct(&kvm->srcu);
 562         kvm_arch_free_vm(kvm);
 563         hardware_disable_all();
 564         mmdrop(mm);
 565 }
 566
 567 void kvm_get_kvm(struct kvm *kvm)
 568 {
 569         atomic_inc(&kvm->users_count);
 570 }
 571 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 572
 573 void kvm_put_kvm(struct kvm *kvm)
 574 {
 575         if (atomic_dec_and_test(&kvm->users_count))
 576                 kvm_destroy_vm(kvm);
 577 }
 578 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 579
 580
 581 static int kvm_vm_release(struct inode *inode, struct file *filp)
 582 {
 583         struct kvm *kvm = filp->private_data;
 584
 585         kvm_irqfd_release(kvm);
 586
 587         kvm_put_kvm(kvm);
 588         return 0;
 589 }
 590
 591 /*
 592  * Allocation size is twice as large as the actual dirty bitmap size.
 593  * This makes it possible to do double buffering: see x86's
 594  * kvm_vm_ioctl_get_dirty_log().
 595  */
 596 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 597 {
 598         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 599
 600         if (dirty_bytes > PAGE_SIZE)
 601                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
 602         else
 603                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
 604
 605         if (!memslot->dirty_bitmap)
 606                 return -ENOMEM;
 607
 608         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
 609         return 0;
 610 }
 611
 612 /*
 613  * Allocate some memory and give it an address in the guest physical address
 614  * space.
 615  *
 616  * Discontiguous memory is allowed, mostly for framebuffers.
 617  *
 618  * Must be called holding mmap_sem for write.
 619  */
 620 int __kvm_set_memory_region(struct kvm *kvm,
 621                             struct kvm_userspace_memory_region *mem,
 622                             int user_alloc)
 623 {
 624         int r;
 625         gfn_t base_gfn;
 626         unsigned long npages;
 627         unsigned long i;
 628         struct kvm_memory_slot *memslot;
 629         struct kvm_memory_slot old, new;
 630         struct kvm_memslots *slots, *old_memslots;
 631
 632         r = -EINVAL;
 633         /* General sanity checks */
 634         if (mem->memory_size & (PAGE_SIZE - 1))
 635                 goto out;
 636         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 637                 goto out;
 638         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 639                 goto out;
 640         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 641                 goto out;
 642         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 643                 goto out;
 644
 645         memslot = &kvm->memslots->memslots[mem->slot];
 646         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 647         npages = mem->memory_size >> PAGE_SHIFT;
 648
 649         r = -EINVAL;
 650         if (npages > KVM_MEM_MAX_NR_PAGES)
 651                 goto out;
 652
 653         if (!npages)
 654                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 655
 656         new = old = *memslot;
 657
 658         new.id = mem->slot;
 659         new.base_gfn = base_gfn;
 660         new.npages = npages;
 661         new.flags = mem->flags;
 662
 663         /* Disallow changing a memory slot's size. */
 664         r = -EINVAL;
 665         if (npages && old.npages && npages != old.npages)
 666                 goto out_free;
 667
 668         /* Check for overlaps */
 669         r = -EEXIST;
 670         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 671                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 672
 673                 if (s == memslot || !s->npages)
 674                         continue;
 675                 if (!((base_gfn + npages <= s->base_gfn) ||
 676                       (base_gfn >= s->base_gfn + s->npages)))
 677                         goto out_free;
 678         }
 679
 680         /* Free page dirty bitmap if unneeded */
 681         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 682                 new.dirty_bitmap = NULL;
 683
 684         r = -ENOMEM;
 685
 686         /* Allocate if a slot is being created */
 687 #ifndef CONFIG_S390
 688         if (npages && !new.rmap) {
 689                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
 690
 691                 if (!new.rmap)
 692                         goto out_free;
 693
 694                 new.user_alloc = user_alloc;
 695                 new.userspace_addr = mem->userspace_addr;
 696         }
 697         if (!npages)
 698                 goto skip_lpage;
 699
 700         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 701                 unsigned long ugfn;
 702                 unsigned long j;
 703                 int lpages;
 704                 int level = i + 2;
 705
 706                 /* Avoid unused variable warning if no large pages */
 707                 (void)level;
 708
 709                 if (new.lpage_info[i])
 710                         continue;
 711
 712                 lpages = 1 + ((base_gfn + npages - 1)
 713                              >> KVM_HPAGE_GFN_SHIFT(level));
 714                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
 715
 716                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
 717
 718                 if (!new.lpage_info[i])
 719                         goto out_free;
 720
 721                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
 722                         new.lpage_info[i][0].write_count = 1;
 723                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
 724                         new.lpage_info[i][lpages - 1].write_count = 1;
 725                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 726                 /*
 727                  * If the gfn and userspace address are not aligned wrt each
 728                  * other, or if explicitly asked to, disable large page
 729                  * support for this slot
 730                  */
 731                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 732                     !largepages_enabled)
 733                         for (j = 0; j < lpages; ++j)
 734                                 new.lpage_info[i][j].write_count = 1;
 735         }
 736
 737 skip_lpage:
 738
 739         /* Allocate page dirty bitmap if needed */
 740         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 741                 if (kvm_create_dirty_bitmap(&new) < 0)
 742                         goto out_free;
 743                 /* destroy any largepage mappings for dirty tracking */
 744         }
 745 #else  /* not defined CONFIG_S390 */
 746         new.user_alloc = user_alloc;
 747         if (user_alloc)
 748                 new.userspace_addr = mem->userspace_addr;
 749 #endif /* not defined CONFIG_S390 */
 750
 751         if (!npages) {
 752                 r = -ENOMEM;
 753                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 754                 if (!slots)
 755                         goto out_free;
 756                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 757                 if (mem->slot >= slots->nmemslots)
 758                         slots->nmemslots = mem->slot + 1;
 759                 slots->generation++;
 760                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 761
 762                 old_memslots = kvm->memslots;
 763                 rcu_assign_pointer(kvm->memslots, slots);
 764                 synchronize_srcu_expedited(&kvm->srcu);
 765                 /* From this point no new shadow pages pointing to a deleted
 766                  * memslot will be created.
 767                  *
 768                  * validation of sp->gfn happens in:
 769                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 770                  *      - kvm_is_visible_gfn (mmu_check_roots)
 771                  */
 772                 kvm_arch_flush_shadow(kvm);
 773                 kfree(old_memslots);
 774         }
 775
 776         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 777         if (r)
 778                 goto out_free;
 779
 780         /* map the pages in iommu page table */
 781         if (npages) {
 782                 r = kvm_iommu_map_pages(kvm, &new);
 783                 if (r)
 784                         goto out_free;
 785         }
 786
 787         r = -ENOMEM;
 788         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 789         if (!slots)
 790                 goto out_free;
 791         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 792         if (mem->slot >= slots->nmemslots)
 793                 slots->nmemslots = mem->slot + 1;
 794         slots->generation++;
 795
 796         /* actual memory is freed via old in kvm_free_physmem_slot below */
 797         if (!npages) {
 798                 new.rmap = NULL;
 799                 new.dirty_bitmap = NULL;
 800                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 801                         new.lpage_info[i] = NULL;
 802         }
 803
 804         slots->memslots[mem->slot] = new;
 805         old_memslots = kvm->memslots;
 806         rcu_assign_pointer(kvm->memslots, slots);
 807         synchronize_srcu_expedited(&kvm->srcu);
 808
 809         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 810
 811         kvm_free_physmem_slot(&old, &new);
 812         kfree(old_memslots);
 813
 814         return 0;
 815
 816 out_free:
 817         kvm_free_physmem_slot(&new, &old);
 818 out:
 819         return r;
 820
 821 }
 822 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 823
 824 int kvm_set_memory_region(struct kvm *kvm,
 825                           struct kvm_userspace_memory_region *mem,
 826                           int user_alloc)
 827 {
 828         int r;
 829
 830         mutex_lock(&kvm->slots_lock);
 831         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 832         mutex_unlock(&kvm->slots_lock);
 833         return r;
 834 }
 835 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 836
 837 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 838                                    struct
 839                                    kvm_userspace_memory_region *mem,
 840                                    int user_alloc)
 841 {
 842         if (mem->slot >= KVM_MEMORY_SLOTS)
 843                 return -EINVAL;
 844         return kvm_set_memory_region(kvm, mem, user_alloc);
 845 }
 846
 847 int kvm_get_dirty_log(struct kvm *kvm,
 848                         struct kvm_dirty_log *log, int *is_dirty)
 849 {
 850         struct kvm_memory_slot *memslot;
 851         int r, i;
 852         unsigned long n;
 853         unsigned long any = 0;
 854
 855         r = -EINVAL;
 856         if (log->slot >= KVM_MEMORY_SLOTS)
 857                 goto out;
 858
 859         memslot = &kvm->memslots->memslots[log->slot];
 860         r = -ENOENT;
 861         if (!memslot->dirty_bitmap)
 862                 goto out;
 863
 864         n = kvm_dirty_bitmap_bytes(memslot);
 865
 866         for (i = 0; !any && i < n/sizeof(long); ++i)
 867                 any = memslot->dirty_bitmap[i];
 868
 869         r = -EFAULT;
 870         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 871                 goto out;
 872
 873         if (any)
 874                 *is_dirty = 1;
 875
 876         r = 0;
 877 out:
 878         return r;
 879 }
 880
 881 void kvm_disable_largepages(void)
 882 {
 883         largepages_enabled = false;
 884 }
 885 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 886
 887 int is_error_page(struct page *page)
 888 {
 889         return page == bad_page || page == hwpoison_page || page == fault_page;
 890 }
 891 EXPORT_SYMBOL_GPL(is_error_page);
 892
 893 int is_error_pfn(pfn_t pfn)
 894 {
 895         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 896 }
 897 EXPORT_SYMBOL_GPL(is_error_pfn);
 898
 899 int is_hwpoison_pfn(pfn_t pfn)
 900 {
 901         return pfn == hwpoison_pfn;
 902 }
 903 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 904
 905 int is_fault_pfn(pfn_t pfn)
 906 {
 907         return pfn == fault_pfn;
 908 }
 909 EXPORT_SYMBOL_GPL(is_fault_pfn);
 910
 911 static inline unsigned long bad_hva(void)
 912 {
 913         return PAGE_OFFSET;
 914 }
 915
 916 int kvm_is_error_hva(unsigned long addr)
 917 {
 918         return addr == bad_hva();
 919 }
 920 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 921
 922 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
 923                                                 gfn_t gfn)
 924 {
 925         int i;
 926
 927         for (i = 0; i < slots->nmemslots; ++i) {
 928                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 929
 930                 if (gfn >= memslot->base_gfn
 931                     && gfn < memslot->base_gfn + memslot->npages)
 932                         return memslot;
 933         }
 934         return NULL;
 935 }
 936
 937 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 938 {
 939         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 940 }
 941 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 942
 943 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 944 {
 945         int i;
 946         struct kvm_memslots *slots = kvm_memslots(kvm);
 947
 948         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 949                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 950
 951                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 952                         continue;
 953
 954                 if (gfn >= memslot->base_gfn
 955                     && gfn < memslot->base_gfn + memslot->npages)
 956                         return 1;
 957         }
 958         return 0;
 959 }
 960 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 961
 962 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 963 {
 964         struct vm_area_struct *vma;
 965         unsigned long addr, size;
 966
 967         size = PAGE_SIZE;
 968
 969         addr = gfn_to_hva(kvm, gfn);
 970         if (kvm_is_error_hva(addr))
 971                 return PAGE_SIZE;
 972
 973         down_read(&current->mm->mmap_sem);
 974         vma = find_vma(current->mm, addr);
 975         if (!vma)
 976                 goto out;
 977
 978         size = vma_kernel_pagesize(vma);
 979
 980 out:
 981         up_read(&current->mm->mmap_sem);
 982
 983         return size;
 984 }
 985
 986 int memslot_id(struct kvm *kvm, gfn_t gfn)
 987 {
 988         int i;
 989         struct kvm_memslots *slots = kvm_memslots(kvm);
 990         struct kvm_memory_slot *memslot = NULL;
 991
 992         for (i = 0; i < slots->nmemslots; ++i) {
 993                 memslot = &slots->memslots[i];
 994
 995                 if (gfn >= memslot->base_gfn
 996                     && gfn < memslot->base_gfn + memslot->npages)
 997                         break;
 998         }
 999
1000         return memslot - slots->memslots;
1001 }
1002
1003 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1004                                      gfn_t *nr_pages)
1005 {
1006         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1007                 return bad_hva();
1008
1009         if (nr_pages)
1010                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1011
1012         return gfn_to_hva_memslot(slot, gfn);
1013 }
1014
1015 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1016 {
1017         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1018 }
1019 EXPORT_SYMBOL_GPL(gfn_to_hva);
1020
1021 static pfn_t get_fault_pfn(void)
1022 {
1023         get_page(fault_page);
1024         return fault_pfn;
1025 }
1026
1027 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1028                         bool *async, bool write_fault, bool *writable)
1029 {
1030         struct page *page[1];
1031         int npages = 0;
1032         pfn_t pfn;
1033
1034         /* we can do it either atomically or asynchronously, not both */
1035         BUG_ON(atomic && async);
1036
1037         BUG_ON(!write_fault && !writable);
1038
1039         if (writable)
1040                 *writable = true;
1041
1042         if (atomic || async)
1043                 npages = __get_user_pages_fast(addr, 1, 1, page);
1044
1045         if (unlikely(npages != 1) && !atomic) {
1046                 might_sleep();
1047
1048                 if (writable)
1049                         *writable = write_fault;
1050
1051                 npages = get_user_pages_fast(addr, 1, write_fault, page);
1052
1053                 /* map read fault as writable if possible */
1054                 if (unlikely(!write_fault) && npages == 1) {
1055                         struct page *wpage[1];
1056
1057                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1058                         if (npages == 1) {
1059                                 *writable = true;
1060                                 put_page(page[0]);
1061                                 page[0] = wpage[0];
1062                         }
1063                         npages = 1;
1064                 }
1065         }
1066
1067         if (unlikely(npages != 1)) {
1068                 struct vm_area_struct *vma;
1069
1070                 if (atomic)
1071                         return get_fault_pfn();
1072
1073                 down_read(&current->mm->mmap_sem);
1074                 if (is_hwpoison_address(addr)) {
1075                         up_read(&current->mm->mmap_sem);
1076                         get_page(hwpoison_page);
1077                         return page_to_pfn(hwpoison_page);
1078                 }
1079
1080                 vma = find_vma_intersection(current->mm, addr, addr+1);
1081
1082                 if (vma == NULL)
1083                         pfn = get_fault_pfn();
1084                 else if ((vma->vm_flags & VM_PFNMAP)) {
1085                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1086                                 vma->vm_pgoff;
1087                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1088                 } else {
1089                         if (async && (vma->vm_flags & VM_WRITE))
1090                                 *async = true;
1091                         pfn = get_fault_pfn();
1092                 }
1093                 up_read(&current->mm->mmap_sem);
1094         } else
1095                 pfn = page_to_pfn(page[0]);
1096
1097         return pfn;
1098 }
1099
1100 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1101 {
1102         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1103 }
1104 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1105
1106 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1107                           bool write_fault, bool *writable)
1108 {
1109         unsigned long addr;
1110
1111         if (async)
1112                 *async = false;
1113
1114         addr = gfn_to_hva(kvm, gfn);
1115         if (kvm_is_error_hva(addr)) {
1116                 get_page(bad_page);
1117                 return page_to_pfn(bad_page);
1118         }
1119
1120         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1121 }
1122
1123 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1124 {
1125         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1126 }
1127 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1128
1129 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1130                        bool write_fault, bool *writable)
1131 {
1132         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1133 }
1134 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1135
1136 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1137 {
1138         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1139 }
1140 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1141
1142 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1143                       bool *writable)
1144 {
1145         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1146 }
1147 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1148
1149 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1150                          struct kvm_memory_slot *slot, gfn_t gfn)
1151 {
1152         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1153         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1154 }
1155
1156 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1157                                                                   int nr_pages)
1158 {
1159         unsigned long addr;
1160         gfn_t entry;
1161
1162         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1163         if (kvm_is_error_hva(addr))
1164                 return -1;
1165
1166         if (entry < nr_pages)
1167                 return 0;
1168
1169         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1170 }
1171 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1172
1173 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1174 {
1175         pfn_t pfn;
1176
1177         pfn = gfn_to_pfn(kvm, gfn);
1178         if (!kvm_is_mmio_pfn(pfn))
1179                 return pfn_to_page(pfn);
1180
1181         WARN_ON(kvm_is_mmio_pfn(pfn));
1182
1183         get_page(bad_page);
1184         return bad_page;
1185 }
1186
1187 EXPORT_SYMBOL_GPL(gfn_to_page);
1188
1189 void kvm_release_page_clean(struct page *page)
1190 {
1191         kvm_release_pfn_clean(page_to_pfn(page));
1192 }
1193 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1194
1195 void kvm_release_pfn_clean(pfn_t pfn)
1196 {
1197         if (!kvm_is_mmio_pfn(pfn))
1198                 put_page(pfn_to_page(pfn));
1199 }
1200 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1201
1202 void kvm_release_page_dirty(struct page *page)
1203 {
1204         kvm_release_pfn_dirty(page_to_pfn(page));
1205 }
1206 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1207
1208 void kvm_release_pfn_dirty(pfn_t pfn)
1209 {
1210         kvm_set_pfn_dirty(pfn);
1211         kvm_release_pfn_clean(pfn);
1212 }
1213 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1214
1215 void kvm_set_page_dirty(struct page *page)
1216 {
1217         kvm_set_pfn_dirty(page_to_pfn(page));
1218 }
1219 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1220
1221 void kvm_set_pfn_dirty(pfn_t pfn)
1222 {
1223         if (!kvm_is_mmio_pfn(pfn)) {
1224                 struct page *page = pfn_to_page(pfn);
1225                 if (!PageReserved(page))
1226                         SetPageDirty(page);
1227         }
1228 }
1229 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1230
1231 void kvm_set_pfn_accessed(pfn_t pfn)
1232 {
1233         if (!kvm_is_mmio_pfn(pfn))
1234                 mark_page_accessed(pfn_to_page(pfn));
1235 }
1236 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1237
1238 void kvm_get_pfn(pfn_t pfn)
1239 {
1240         if (!kvm_is_mmio_pfn(pfn))
1241                 get_page(pfn_to_page(pfn));
1242 }
1243 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1244
1245 static int next_segment(unsigned long len, int offset)
1246 {
1247         if (len > PAGE_SIZE - offset)
1248                 return PAGE_SIZE - offset;
1249         else
1250                 return len;
1251 }
1252
1253 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1254                         int len)
1255 {
1256         int r;
1257         unsigned long addr;
1258
1259         addr = gfn_to_hva(kvm, gfn);
1260         if (kvm_is_error_hva(addr))
1261                 return -EFAULT;
1262         r = copy_from_user(data, (void __user *)addr + offset, len);
1263         if (r)
1264                 return -EFAULT;
1265         return 0;
1266 }
1267 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1268
1269 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1270 {
1271         gfn_t gfn = gpa >> PAGE_SHIFT;
1272         int seg;
1273         int offset = offset_in_page(gpa);
1274         int ret;
1275
1276         while ((seg = next_segment(len, offset)) != 0) {
1277                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1278                 if (ret < 0)
1279                         return ret;
1280                 offset = 0;
1281                 len -= seg;
1282                 data += seg;
1283                 ++gfn;
1284         }
1285         return 0;
1286 }
1287 EXPORT_SYMBOL_GPL(kvm_read_guest);
1288
1289 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1290                           unsigned long len)
1291 {
1292         int r;
1293         unsigned long addr;
1294         gfn_t gfn = gpa >> PAGE_SHIFT;
1295         int offset = offset_in_page(gpa);
1296
1297         addr = gfn_to_hva(kvm, gfn);
1298         if (kvm_is_error_hva(addr))
1299                 return -EFAULT;
1300         pagefault_disable();
1301         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1302         pagefault_enable();
1303         if (r)
1304                 return -EFAULT;
1305         return 0;
1306 }
1307 EXPORT_SYMBOL(kvm_read_guest_atomic);
1308
1309 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1310                          int offset, int len)
1311 {
1312         int r;
1313         unsigned long addr;
1314
1315         addr = gfn_to_hva(kvm, gfn);
1316         if (kvm_is_error_hva(addr))
1317                 return -EFAULT;
1318         r = copy_to_user((void __user *)addr + offset, data, len);
1319         if (r)
1320                 return -EFAULT;
1321         mark_page_dirty(kvm, gfn);
1322         return 0;
1323 }
1324 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1325
1326 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1327                     unsigned long len)
1328 {
1329         gfn_t gfn = gpa >> PAGE_SHIFT;
1330         int seg;
1331         int offset = offset_in_page(gpa);
1332         int ret;
1333
1334         while ((seg = next_segment(len, offset)) != 0) {
1335                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1336                 if (ret < 0)
1337                         return ret;
1338                 offset = 0;
1339                 len -= seg;
1340                 data += seg;
1341                 ++gfn;
1342         }
1343         return 0;
1344 }
1345
1346 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1347                               gpa_t gpa)
1348 {
1349         struct kvm_memslots *slots = kvm_memslots(kvm);
1350         int offset = offset_in_page(gpa);
1351         gfn_t gfn = gpa >> PAGE_SHIFT;
1352
1353         ghc->gpa = gpa;
1354         ghc->generation = slots->generation;
1355         ghc->memslot = __gfn_to_memslot(slots, gfn);
1356         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1357         if (!kvm_is_error_hva(ghc->hva))
1358                 ghc->hva += offset;
1359         else
1360                 return -EFAULT;
1361
1362         return 0;
1363 }
1364 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1365
1366 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1367                            void *data, unsigned long len)
1368 {
1369         struct kvm_memslots *slots = kvm_memslots(kvm);
1370         int r;
1371
1372         if (slots->generation != ghc->generation)
1373                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1374
1375         if (kvm_is_error_hva(ghc->hva))
1376                 return -EFAULT;
1377
1378         r = copy_to_user((void __user *)ghc->hva, data, len);
1379         if (r)
1380                 return -EFAULT;
1381         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1382
1383         return 0;
1384 }
1385 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1386
1387 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1388 {
1389         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1390                                     offset, len);
1391 }
1392 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1393
1394 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1395 {
1396         gfn_t gfn = gpa >> PAGE_SHIFT;
1397         int seg;
1398         int offset = offset_in_page(gpa);
1399         int ret;
1400
1401         while ((seg = next_segment(len, offset)) != 0) {
1402                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1403                 if (ret < 0)
1404                         return ret;
1405                 offset = 0;
1406                 len -= seg;
1407                 ++gfn;
1408         }
1409         return 0;
1410 }
1411 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1412
1413 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1414                              gfn_t gfn)
1415 {
1416         if (memslot && memslot->dirty_bitmap) {
1417                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1418
1419                 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1420         }
1421 }
1422
1423 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1424 {
1425         struct kvm_memory_slot *memslot;
1426
1427         memslot = gfn_to_memslot(kvm, gfn);
1428         mark_page_dirty_in_slot(kvm, memslot, gfn);
1429 }
1430
1431 /*
1432  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1433  */
1434 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1435 {
1436         DEFINE_WAIT(wait);
1437
1438         for (;;) {
1439                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1440
1441                 if (kvm_arch_vcpu_runnable(vcpu)) {
1442                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1443                         break;
1444                 }
1445                 if (kvm_cpu_has_pending_timer(vcpu))
1446                         break;
1447                 if (signal_pending(current))
1448                         break;
1449
1450                 schedule();
1451         }
1452
1453         finish_wait(&vcpu->wq, &wait);
1454 }
1455
1456 void kvm_resched(struct kvm_vcpu *vcpu)
1457 {
1458         if (!need_resched())
1459                 return;
1460         cond_resched();
1461 }
1462 EXPORT_SYMBOL_GPL(kvm_resched);
1463
1464 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1465 {
1466         ktime_t expires;
1467         DEFINE_WAIT(wait);
1468
1469         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1470
1471         /* Sleep for 100 us, and hope lock-holder got scheduled */
1472         expires = ktime_add_ns(ktime_get(), 100000UL);
1473         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1474
1475         finish_wait(&vcpu->wq, &wait);
1476 }
1477 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1478
1479 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1480 {
1481         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1482         struct page *page;
1483
1484         if (vmf->pgoff == 0)
1485                 page = virt_to_page(vcpu->run);
1486 #ifdef CONFIG_X86
1487         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1488                 page = virt_to_page(vcpu->arch.pio_data);
1489 #endif
1490 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1491         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1492                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1493 #endif
1494         else
1495                 return VM_FAULT_SIGBUS;
1496         get_page(page);
1497         vmf->page = page;
1498         return 0;
1499 }
1500
1501 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1502         .fault = kvm_vcpu_fault,
1503 };
1504
1505 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1506 {
1507         vma->vm_ops = &kvm_vcpu_vm_ops;
1508         return 0;
1509 }
1510
1511 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1512 {
1513         struct kvm_vcpu *vcpu = filp->private_data;
1514
1515         kvm_put_kvm(vcpu->kvm);
1516         return 0;
1517 }
1518
1519 static struct file_operations kvm_vcpu_fops = {
1520         .release        = kvm_vcpu_release,
1521         .unlocked_ioctl = kvm_vcpu_ioctl,
1522         .compat_ioctl   = kvm_vcpu_ioctl,
1523         .mmap           = kvm_vcpu_mmap,
1524         .llseek         = noop_llseek,
1525 };
1526
1527 /*
1528  * Allocates an inode for the vcpu.
1529  */
1530 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1531 {
1532         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1533 }
1534
1535 /*
1536  * Creates some virtual cpus.  Good luck creating more than one.
1537  */
1538 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1539 {
1540         int r;
1541         struct kvm_vcpu *vcpu, *v;
1542
1543         vcpu = kvm_arch_vcpu_create(kvm, id);
1544         if (IS_ERR(vcpu))
1545                 return PTR_ERR(vcpu);
1546
1547         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1548
1549         r = kvm_arch_vcpu_setup(vcpu);
1550         if (r)
1551                 return r;
1552
1553         mutex_lock(&kvm->lock);
1554         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1555                 r = -EINVAL;
1556                 goto vcpu_destroy;
1557         }
1558
1559         kvm_for_each_vcpu(r, v, kvm)
1560                 if (v->vcpu_id == id) {
1561                         r = -EEXIST;
1562                         goto vcpu_destroy;
1563                 }
1564
1565         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1566
1567         /* Now it's all set up, let userspace reach it */
1568         kvm_get_kvm(kvm);
1569         r = create_vcpu_fd(vcpu);
1570         if (r < 0) {
1571                 kvm_put_kvm(kvm);
1572                 goto vcpu_destroy;
1573         }
1574
1575         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1576         smp_wmb();
1577         atomic_inc(&kvm->online_vcpus);
1578
1579 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1580         if (kvm->bsp_vcpu_id == id)
1581                 kvm->bsp_vcpu = vcpu;
1582 #endif
1583         mutex_unlock(&kvm->lock);
1584         return r;
1585
1586 vcpu_destroy:
1587         mutex_unlock(&kvm->lock);
1588         kvm_arch_vcpu_destroy(vcpu);
1589         return r;
1590 }
1591
1592 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1593 {
1594         if (sigset) {
1595                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1596                 vcpu->sigset_active = 1;
1597                 vcpu->sigset = *sigset;
1598         } else
1599                 vcpu->sigset_active = 0;
1600         return 0;
1601 }
1602
1603 static long kvm_vcpu_ioctl(struct file *filp,
1604                            unsigned int ioctl, unsigned long arg)
1605 {
1606         struct kvm_vcpu *vcpu = filp->private_data;
1607         void __user *argp = (void __user *)arg;
1608         int r;
1609         struct kvm_fpu *fpu = NULL;
1610         struct kvm_sregs *kvm_sregs = NULL;
1611
1612         if (vcpu->kvm->mm != current->mm)
1613                 return -EIO;
1614
1615 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1616         /*
1617          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1618          * so vcpu_load() would break it.
1619          */
1620         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1621                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1622 #endif
1623
1624
1625         vcpu_load(vcpu);
1626         switch (ioctl) {
1627         case KVM_RUN:
1628                 r = -EINVAL;
1629                 if (arg)
1630                         goto out;
1631                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1632                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1633                 break;
1634         case KVM_GET_REGS: {
1635                 struct kvm_regs *kvm_regs;
1636
1637                 r = -ENOMEM;
1638                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1639                 if (!kvm_regs)
1640                         goto out;
1641                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1642                 if (r)
1643                         goto out_free1;
1644                 r = -EFAULT;
1645                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1646                         goto out_free1;
1647                 r = 0;
1648 out_free1:
1649                 kfree(kvm_regs);
1650                 break;
1651         }
1652         case KVM_SET_REGS: {
1653                 struct kvm_regs *kvm_regs;
1654
1655                 r = -ENOMEM;
1656                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1657                 if (!kvm_regs)
1658                         goto out;
1659                 r = -EFAULT;
1660                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1661                         goto out_free2;
1662                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1663                 if (r)
1664                         goto out_free2;
1665                 r = 0;
1666 out_free2:
1667                 kfree(kvm_regs);
1668                 break;
1669         }
1670         case KVM_GET_SREGS: {
1671                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1672                 r = -ENOMEM;
1673                 if (!kvm_sregs)
1674                         goto out;
1675                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1676                 if (r)
1677                         goto out;
1678                 r = -EFAULT;
1679                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1680                         goto out;
1681                 r = 0;
1682                 break;
1683         }
1684         case KVM_SET_SREGS: {
1685                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1686                 r = -ENOMEM;
1687                 if (!kvm_sregs)
1688                         goto out;
1689                 r = -EFAULT;
1690                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1691                         goto out;
1692                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1693                 if (r)
1694                         goto out;
1695                 r = 0;
1696                 break;
1697         }
1698         case KVM_GET_MP_STATE: {
1699                 struct kvm_mp_state mp_state;
1700
1701                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1702                 if (r)
1703                         goto out;
1704                 r = -EFAULT;
1705                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1706                         goto out;
1707                 r = 0;
1708                 break;
1709         }
1710         case KVM_SET_MP_STATE: {
1711                 struct kvm_mp_state mp_state;
1712
1713                 r = -EFAULT;
1714                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1715                         goto out;
1716                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1717                 if (r)
1718                         goto out;
1719                 r = 0;
1720                 break;
1721         }
1722         case KVM_TRANSLATE: {
1723                 struct kvm_translation tr;
1724
1725                 r = -EFAULT;
1726                 if (copy_from_user(&tr, argp, sizeof tr))
1727                         goto out;
1728                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1729                 if (r)
1730                         goto out;
1731                 r = -EFAULT;
1732                 if (copy_to_user(argp, &tr, sizeof tr))
1733                         goto out;
1734                 r = 0;
1735                 break;
1736         }
1737         case KVM_SET_GUEST_DEBUG: {
1738                 struct kvm_guest_debug dbg;
1739
1740                 r = -EFAULT;
1741                 if (copy_from_user(&dbg, argp, sizeof dbg))
1742                         goto out;
1743                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1744                 if (r)
1745                         goto out;
1746                 r = 0;
1747                 break;
1748         }
1749         case KVM_SET_SIGNAL_MASK: {
1750                 struct kvm_signal_mask __user *sigmask_arg = argp;
1751                 struct kvm_signal_mask kvm_sigmask;
1752                 sigset_t sigset, *p;
1753
1754                 p = NULL;
1755                 if (argp) {
1756                         r = -EFAULT;
1757                         if (copy_from_user(&kvm_sigmask, argp,
1758                                            sizeof kvm_sigmask))
1759                                 goto out;
1760                         r = -EINVAL;
1761                         if (kvm_sigmask.len != sizeof sigset)
1762                                 goto out;
1763                         r = -EFAULT;
1764                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1765                                            sizeof sigset))
1766                                 goto out;
1767                         p = &sigset;
1768                 }
1769                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1770                 break;
1771         }
1772         case KVM_GET_FPU: {
1773                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1774                 r = -ENOMEM;
1775                 if (!fpu)
1776                         goto out;
1777                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1778                 if (r)
1779                         goto out;
1780                 r = -EFAULT;
1781                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1782                         goto out;
1783                 r = 0;
1784                 break;
1785         }
1786         case KVM_SET_FPU: {
1787                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1788                 r = -ENOMEM;
1789                 if (!fpu)
1790                         goto out;
1791                 r = -EFAULT;
1792                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1793                         goto out;
1794                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1795                 if (r)
1796                         goto out;
1797                 r = 0;
1798                 break;
1799         }
1800         default:
1801                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1802         }
1803 out:
1804         vcpu_put(vcpu);
1805         kfree(fpu);
1806         kfree(kvm_sregs);
1807         return r;
1808 }
1809
1810 static long kvm_vm_ioctl(struct file *filp,
1811                            unsigned int ioctl, unsigned long arg)
1812 {
1813         struct kvm *kvm = filp->private_data;
1814         void __user *argp = (void __user *)arg;
1815         int r;
1816
1817         if (kvm->mm != current->mm)
1818                 return -EIO;
1819         switch (ioctl) {
1820         case KVM_CREATE_VCPU:
1821                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1822                 if (r < 0)
1823                         goto out;
1824                 break;
1825         case KVM_SET_USER_MEMORY_REGION: {
1826                 struct kvm_userspace_memory_region kvm_userspace_mem;
1827
1828                 r = -EFAULT;
1829                 if (copy_from_user(&kvm_userspace_mem, argp,
1830                                                 sizeof kvm_userspace_mem))
1831                         goto out;
1832
1833                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1834                 if (r)
1835                         goto out;
1836                 break;
1837         }
1838         case KVM_GET_DIRTY_LOG: {
1839                 struct kvm_dirty_log log;
1840
1841                 r = -EFAULT;
1842                 if (copy_from_user(&log, argp, sizeof log))
1843                         goto out;
1844                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1845                 if (r)
1846                         goto out;
1847                 break;
1848         }
1849 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1850         case KVM_REGISTER_COALESCED_MMIO: {
1851                 struct kvm_coalesced_mmio_zone zone;
1852                 r = -EFAULT;
1853                 if (copy_from_user(&zone, argp, sizeof zone))
1854                         goto out;
1855                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1856                 if (r)
1857                         goto out;
1858                 r = 0;
1859                 break;
1860         }
1861         case KVM_UNREGISTER_COALESCED_MMIO: {
1862                 struct kvm_coalesced_mmio_zone zone;
1863                 r = -EFAULT;
1864                 if (copy_from_user(&zone, argp, sizeof zone))
1865                         goto out;
1866                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1867                 if (r)
1868                         goto out;
1869                 r = 0;
1870                 break;
1871         }
1872 #endif
1873         case KVM_IRQFD: {
1874                 struct kvm_irqfd data;
1875
1876                 r = -EFAULT;
1877                 if (copy_from_user(&data, argp, sizeof data))
1878                         goto out;
1879                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1880                 break;
1881         }
1882         case KVM_IOEVENTFD: {
1883                 struct kvm_ioeventfd data;
1884
1885                 r = -EFAULT;
1886                 if (copy_from_user(&data, argp, sizeof data))
1887                         goto out;
1888                 r = kvm_ioeventfd(kvm, &data);
1889                 break;
1890         }
1891 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1892         case KVM_SET_BOOT_CPU_ID:
1893                 r = 0;
1894                 mutex_lock(&kvm->lock);
1895                 if (atomic_read(&kvm->online_vcpus) != 0)
1896                         r = -EBUSY;
1897                 else
1898                         kvm->bsp_vcpu_id = arg;
1899                 mutex_unlock(&kvm->lock);
1900                 break;
1901 #endif
1902         default:
1903                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1904                 if (r == -ENOTTY)
1905                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1906         }
1907 out:
1908         return r;
1909 }
1910
1911 #ifdef CONFIG_COMPAT
1912 struct compat_kvm_dirty_log {
1913         __u32 slot;
1914         __u32 padding1;
1915         union {
1916                 compat_uptr_t dirty_bitmap; /* one bit per page */
1917                 __u64 padding2;
1918         };
1919 };
1920
1921 static long kvm_vm_compat_ioctl(struct file *filp,
1922                            unsigned int ioctl, unsigned long arg)
1923 {
1924         struct kvm *kvm = filp->private_data;
1925         int r;
1926
1927         if (kvm->mm != current->mm)
1928                 return -EIO;
1929         switch (ioctl) {
1930         case KVM_GET_DIRTY_LOG: {
1931                 struct compat_kvm_dirty_log compat_log;
1932                 struct kvm_dirty_log log;
1933
1934                 r = -EFAULT;
1935                 if (copy_from_user(&compat_log, (void __user *)arg,
1936                                    sizeof(compat_log)))
1937                         goto out;
1938                 log.slot         = compat_log.slot;
1939                 log.padding1     = compat_log.padding1;
1940                 log.padding2     = compat_log.padding2;
1941                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1942
1943                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1944                 if (r)
1945                         goto out;
1946                 break;
1947         }
1948         default:
1949                 r = kvm_vm_ioctl(filp, ioctl, arg);
1950         }
1951
1952 out:
1953         return r;
1954 }
1955 #endif
1956
1957 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1958 {
1959         struct page *page[1];
1960         unsigned long addr;
1961         int npages;
1962         gfn_t gfn = vmf->pgoff;
1963         struct kvm *kvm = vma->vm_file->private_data;
1964
1965         addr = gfn_to_hva(kvm, gfn);
1966         if (kvm_is_error_hva(addr))
1967                 return VM_FAULT_SIGBUS;
1968
1969         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1970                                 NULL);
1971         if (unlikely(npages != 1))
1972                 return VM_FAULT_SIGBUS;
1973
1974         vmf->page = page[0];
1975         return 0;
1976 }
1977
1978 static const struct vm_operations_struct kvm_vm_vm_ops = {
1979         .fault = kvm_vm_fault,
1980 };
1981
1982 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1983 {
1984         vma->vm_ops = &kvm_vm_vm_ops;
1985         return 0;
1986 }
1987
1988 static struct file_operations kvm_vm_fops = {
1989         .release        = kvm_vm_release,
1990         .unlocked_ioctl = kvm_vm_ioctl,
1991 #ifdef CONFIG_COMPAT
1992         .compat_ioctl   = kvm_vm_compat_ioctl,
1993 #endif
1994         .mmap           = kvm_vm_mmap,
1995         .llseek         = noop_llseek,
1996 };
1997
1998 static int kvm_dev_ioctl_create_vm(void)
1999 {
2000         int r;
2001         struct kvm *kvm;
2002
2003         kvm = kvm_create_vm();
2004         if (IS_ERR(kvm))
2005                 return PTR_ERR(kvm);
2006 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2007         r = kvm_coalesced_mmio_init(kvm);
2008         if (r < 0) {
2009                 kvm_put_kvm(kvm);
2010                 return r;
2011         }
2012 #endif
2013         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2014         if (r < 0)
2015                 kvm_put_kvm(kvm);
2016
2017         return r;
2018 }
2019
2020 static long kvm_dev_ioctl_check_extension_generic(long arg)
2021 {
2022         switch (arg) {
2023         case KVM_CAP_USER_MEMORY:
2024         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2025         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2026 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2027         case KVM_CAP_SET_BOOT_CPU_ID:
2028 #endif
2029         case KVM_CAP_INTERNAL_ERROR_DATA:
2030                 return 1;
2031 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2032         case KVM_CAP_IRQ_ROUTING:
2033                 return KVM_MAX_IRQ_ROUTES;
2034 #endif
2035         default:
2036                 break;
2037         }
2038         return kvm_dev_ioctl_check_extension(arg);
2039 }
2040
2041 static long kvm_dev_ioctl(struct file *filp,
2042                           unsigned int ioctl, unsigned long arg)
2043 {
2044         long r = -EINVAL;
2045
2046         switch (ioctl) {
2047         case KVM_GET_API_VERSION:
2048                 r = -EINVAL;
2049                 if (arg)
2050                         goto out;
2051                 r = KVM_API_VERSION;
2052                 break;
2053         case KVM_CREATE_VM:
2054                 r = -EINVAL;
2055                 if (arg)
2056                         goto out;
2057                 r = kvm_dev_ioctl_create_vm();
2058                 break;
2059         case KVM_CHECK_EXTENSION:
2060                 r = kvm_dev_ioctl_check_extension_generic(arg);
2061                 break;
2062         case KVM_GET_VCPU_MMAP_SIZE:
2063                 r = -EINVAL;
2064                 if (arg)
2065                         goto out;
2066                 r = PAGE_SIZE;     /* struct kvm_run */
2067 #ifdef CONFIG_X86
2068                 r += PAGE_SIZE;    /* pio data page */
2069 #endif
2070 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2071                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2072 #endif
2073                 break;
2074         case KVM_TRACE_ENABLE:
2075         case KVM_TRACE_PAUSE:
2076         case KVM_TRACE_DISABLE:
2077                 r = -EOPNOTSUPP;
2078                 break;
2079         default:
2080                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2081         }
2082 out:
2083         return r;
2084 }
2085
2086 static struct file_operations kvm_chardev_ops = {
2087         .unlocked_ioctl = kvm_dev_ioctl,
2088         .compat_ioctl   = kvm_dev_ioctl,
2089         .llseek         = noop_llseek,
2090 };
2091
2092 static struct miscdevice kvm_dev = {
2093         KVM_MINOR,
2094         "kvm",
2095         &kvm_chardev_ops,
2096 };
2097
2098 static void hardware_enable_nolock(void *junk)
2099 {
2100         int cpu = raw_smp_processor_id();
2101         int r;
2102
2103         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2104                 return;
2105
2106         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2107
2108         r = kvm_arch_hardware_enable(NULL);
2109
2110         if (r) {
2111                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2112                 atomic_inc(&hardware_enable_failed);
2113                 printk(KERN_INFO "kvm: enabling virtualization on "
2114                                  "CPU%d failed\n", cpu);
2115         }
2116 }
2117
2118 static void hardware_enable(void *junk)
2119 {
2120         spin_lock(&kvm_lock);
2121         hardware_enable_nolock(junk);
2122         spin_unlock(&kvm_lock);
2123 }
2124
2125 static void hardware_disable_nolock(void *junk)
2126 {
2127         int cpu = raw_smp_processor_id();
2128
2129         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2130                 return;
2131         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2132         kvm_arch_hardware_disable(NULL);
2133 }
2134
2135 static void hardware_disable(void *junk)
2136 {
2137         spin_lock(&kvm_lock);
2138         hardware_disable_nolock(junk);
2139         spin_unlock(&kvm_lock);
2140 }
2141
2142 static void hardware_disable_all_nolock(void)
2143 {
2144         BUG_ON(!kvm_usage_count);
2145
2146         kvm_usage_count--;
2147         if (!kvm_usage_count)
2148                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2149 }
2150
2151 static void hardware_disable_all(void)
2152 {
2153         spin_lock(&kvm_lock);
2154         hardware_disable_all_nolock();
2155         spin_unlock(&kvm_lock);
2156 }
2157
2158 static int hardware_enable_all(void)
2159 {
2160         int r = 0;
2161
2162         spin_lock(&kvm_lock);
2163
2164         kvm_usage_count++;
2165         if (kvm_usage_count == 1) {
2166                 atomic_set(&hardware_enable_failed, 0);
2167                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2168
2169                 if (atomic_read(&hardware_enable_failed)) {
2170                         hardware_disable_all_nolock();
2171                         r = -EBUSY;
2172                 }
2173         }
2174
2175         spin_unlock(&kvm_lock);
2176
2177         return r;
2178 }
2179
2180 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2181                            void *v)
2182 {
2183         int cpu = (long)v;
2184
2185         if (!kvm_usage_count)
2186                 return NOTIFY_OK;
2187
2188         val &= ~CPU_TASKS_FROZEN;
2189         switch (val) {
2190         case CPU_DYING:
2191                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2192                        cpu);
2193                 hardware_disable(NULL);
2194                 break;
2195         case CPU_STARTING:
2196                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2197                        cpu);
2198                 hardware_enable(NULL);
2199                 break;
2200         }
2201         return NOTIFY_OK;
2202 }
2203
2204
2205 asmlinkage void kvm_spurious_fault(void)
2206 {
2207         /* Fault while not rebooting.  We want the trace. */
2208         BUG();
2209 }
2210 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2211
2212 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2213                       void *v)
2214 {
2215         /*
2216          * Some (well, at least mine) BIOSes hang on reboot if
2217          * in vmx root mode.
2218          *
2219          * And Intel TXT required VMX off for all cpu when system shutdown.
2220          */
2221         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2222         kvm_rebooting = true;
2223         on_each_cpu(hardware_disable_nolock, NULL, 1);
2224         return NOTIFY_OK;
2225 }
2226
2227 static struct notifier_block kvm_reboot_notifier = {
2228         .notifier_call = kvm_reboot,
2229         .priority = 0,
2230 };
2231
2232 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2233 {
2234         int i;
2235
2236         for (i = 0; i < bus->dev_count; i++) {
2237                 struct kvm_io_device *pos = bus->devs[i];
2238
2239                 kvm_iodevice_destructor(pos);
2240         }
2241         kfree(bus);
2242 }
2243
2244 /* kvm_io_bus_write - called under kvm->slots_lock */
2245 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2246                      int len, const void *val)
2247 {
2248         int i;
2249         struct kvm_io_bus *bus;
2250
2251         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2252         for (i = 0; i < bus->dev_count; i++)
2253                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2254                         return 0;
2255         return -EOPNOTSUPP;
2256 }
2257
2258 /* kvm_io_bus_read - called under kvm->slots_lock */
2259 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2260                     int len, void *val)
2261 {
2262         int i;
2263         struct kvm_io_bus *bus;
2264
2265         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2266         for (i = 0; i < bus->dev_count; i++)
2267                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2268                         return 0;
2269         return -EOPNOTSUPP;
2270 }
2271
2272 /* Caller must hold slots_lock. */
2273 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2274                             struct kvm_io_device *dev)
2275 {
2276         struct kvm_io_bus *new_bus, *bus;
2277
2278         bus = kvm->buses[bus_idx];
2279         if (bus->dev_count > NR_IOBUS_DEVS-1)
2280                 return -ENOSPC;
2281
2282         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2283         if (!new_bus)
2284                 return -ENOMEM;
2285         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2286         new_bus->devs[new_bus->dev_count++] = dev;
2287         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2288         synchronize_srcu_expedited(&kvm->srcu);
2289         kfree(bus);
2290
2291         return 0;
2292 }
2293
2294 /* Caller must hold slots_lock. */
2295 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2296                               struct kvm_io_device *dev)
2297 {
2298         int i, r;
2299         struct kvm_io_bus *new_bus, *bus;
2300
2301         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2302         if (!new_bus)
2303                 return -ENOMEM;
2304
2305         bus = kvm->buses[bus_idx];
2306         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2307
2308         r = -ENOENT;
2309         for (i = 0; i < new_bus->dev_count; i++)
2310                 if (new_bus->devs[i] == dev) {
2311                         r = 0;
2312                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2313                         break;
2314                 }
2315
2316         if (r) {
2317                 kfree(new_bus);
2318                 return r;
2319         }
2320
2321         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2322         synchronize_srcu_expedited(&kvm->srcu);
2323         kfree(bus);
2324         return r;
2325 }
2326
2327 static struct notifier_block kvm_cpu_notifier = {
2328         .notifier_call = kvm_cpu_hotplug,
2329 };
2330
2331 static int vm_stat_get(void *_offset, u64 *val)
2332 {
2333         unsigned offset = (long)_offset;
2334         struct kvm *kvm;
2335
2336         *val = 0;
2337         spin_lock(&kvm_lock);
2338         list_for_each_entry(kvm, &vm_list, vm_list)
2339                 *val += *(u32 *)((void *)kvm + offset);
2340         spin_unlock(&kvm_lock);
2341         return 0;
2342 }
2343
2344 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2345
2346 static int vcpu_stat_get(void *_offset, u64 *val)
2347 {
2348         unsigned offset = (long)_offset;
2349         struct kvm *kvm;
2350         struct kvm_vcpu *vcpu;
2351         int i;
2352
2353         *val = 0;
2354         spin_lock(&kvm_lock);
2355         list_for_each_entry(kvm, &vm_list, vm_list)
2356                 kvm_for_each_vcpu(i, vcpu, kvm)
2357                         *val += *(u32 *)((void *)vcpu + offset);
2358
2359         spin_unlock(&kvm_lock);
2360         return 0;
2361 }
2362
2363 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2364
2365 static const struct file_operations *stat_fops[] = {
2366         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2367         [KVM_STAT_VM]   = &vm_stat_fops,
2368 };
2369
2370 static void kvm_init_debug(void)
2371 {
2372         struct kvm_stats_debugfs_item *p;
2373
2374         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2375         for (p = debugfs_entries; p->name; ++p)
2376                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2377                                                 (void *)(long)p->offset,
2378                                                 stat_fops[p->kind]);
2379 }
2380
2381 static void kvm_exit_debug(void)
2382 {
2383         struct kvm_stats_debugfs_item *p;
2384
2385         for (p = debugfs_entries; p->name; ++p)
2386                 debugfs_remove(p->dentry);
2387         debugfs_remove(kvm_debugfs_dir);
2388 }
2389
2390 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2391 {
2392         if (kvm_usage_count)
2393                 hardware_disable_nolock(NULL);
2394         return 0;
2395 }
2396
2397 static int kvm_resume(struct sys_device *dev)
2398 {
2399         if (kvm_usage_count) {
2400                 WARN_ON(spin_is_locked(&kvm_lock));
2401                 hardware_enable_nolock(NULL);
2402         }
2403         return 0;
2404 }
2405
2406 static struct sysdev_class kvm_sysdev_class = {
2407         .name = "kvm",
2408         .suspend = kvm_suspend,
2409         .resume = kvm_resume,
2410 };
2411
2412 static struct sys_device kvm_sysdev = {
2413         .id = 0,
2414         .cls = &kvm_sysdev_class,
2415 };
2416
2417 struct page *bad_page;
2418 pfn_t bad_pfn;
2419
2420 static inline
2421 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2422 {
2423         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2424 }
2425
2426 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2427 {
2428         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2429
2430         kvm_arch_vcpu_load(vcpu, cpu);
2431 }
2432
2433 static void kvm_sched_out(struct preempt_notifier *pn,
2434                           struct task_struct *next)
2435 {
2436         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2437
2438         kvm_arch_vcpu_put(vcpu);
2439 }
2440
2441 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2442                   struct module *module)
2443 {
2444         int r;
2445         int cpu;
2446
2447         r = kvm_arch_init(opaque);
2448         if (r)
2449                 goto out_fail;
2450
2451         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2452
2453         if (bad_page == NULL) {
2454                 r = -ENOMEM;
2455                 goto out;
2456         }
2457
2458         bad_pfn = page_to_pfn(bad_page);
2459
2460         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2461
2462         if (hwpoison_page == NULL) {
2463                 r = -ENOMEM;
2464                 goto out_free_0;
2465         }
2466
2467         hwpoison_pfn = page_to_pfn(hwpoison_page);
2468
2469         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2470
2471         if (fault_page == NULL) {
2472                 r = -ENOMEM;
2473                 goto out_free_0;
2474         }
2475
2476         fault_pfn = page_to_pfn(fault_page);
2477
2478         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2479                 r = -ENOMEM;
2480                 goto out_free_0;
2481         }
2482
2483         r = kvm_arch_hardware_setup();
2484         if (r < 0)
2485                 goto out_free_0a;
2486
2487         for_each_online_cpu(cpu) {
2488                 smp_call_function_single(cpu,
2489                                 kvm_arch_check_processor_compat,
2490                                 &r, 1);
2491                 if (r < 0)
2492                         goto out_free_1;
2493         }
2494
2495         r = register_cpu_notifier(&kvm_cpu_notifier);
2496         if (r)
2497                 goto out_free_2;
2498         register_reboot_notifier(&kvm_reboot_notifier);
2499
2500         r = sysdev_class_register(&kvm_sysdev_class);
2501         if (r)
2502                 goto out_free_3;
2503
2504         r = sysdev_register(&kvm_sysdev);
2505         if (r)
2506                 goto out_free_4;
2507
2508         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2509         if (!vcpu_align)
2510                 vcpu_align = __alignof__(struct kvm_vcpu);
2511         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2512                                            0, NULL);
2513         if (!kvm_vcpu_cache) {
2514                 r = -ENOMEM;
2515                 goto out_free_5;
2516         }
2517
2518         r = kvm_async_pf_init();
2519         if (r)
2520                 goto out_free;
2521
2522         kvm_chardev_ops.owner = module;
2523         kvm_vm_fops.owner = module;
2524         kvm_vcpu_fops.owner = module;
2525
2526         r = misc_register(&kvm_dev);
2527         if (r) {
2528                 printk(KERN_ERR "kvm: misc device register failed\n");
2529                 goto out_unreg;
2530         }
2531
2532         kvm_preempt_ops.sched_in = kvm_sched_in;
2533         kvm_preempt_ops.sched_out = kvm_sched_out;
2534
2535         kvm_init_debug();
2536
2537         return 0;
2538
2539 out_unreg:
2540         kvm_async_pf_deinit();
2541 out_free:
2542         kmem_cache_destroy(kvm_vcpu_cache);
2543 out_free_5:
2544         sysdev_unregister(&kvm_sysdev);
2545 out_free_4:
2546         sysdev_class_unregister(&kvm_sysdev_class);
2547 out_free_3:
2548         unregister_reboot_notifier(&kvm_reboot_notifier);
2549         unregister_cpu_notifier(&kvm_cpu_notifier);
2550 out_free_2:
2551 out_free_1:
2552         kvm_arch_hardware_unsetup();
2553 out_free_0a:
2554         free_cpumask_var(cpus_hardware_enabled);
2555 out_free_0:
2556         if (fault_page)
2557                 __free_page(fault_page);
2558         if (hwpoison_page)
2559                 __free_page(hwpoison_page);
2560         __free_page(bad_page);
2561 out:
2562         kvm_arch_exit();
2563 out_fail:
2564         return r;
2565 }
2566 EXPORT_SYMBOL_GPL(kvm_init);
2567
2568 void kvm_exit(void)
2569 {
2570         kvm_exit_debug();
2571         misc_deregister(&kvm_dev);
2572         kmem_cache_destroy(kvm_vcpu_cache);
2573         kvm_async_pf_deinit();
2574         sysdev_unregister(&kvm_sysdev);
2575         sysdev_class_unregister(&kvm_sysdev_class);
2576         unregister_reboot_notifier(&kvm_reboot_notifier);
2577         unregister_cpu_notifier(&kvm_cpu_notifier);
2578         on_each_cpu(hardware_disable_nolock, NULL, 1);
2579         kvm_arch_hardware_unsetup();
2580         kvm_arch_exit();
2581         free_cpumask_var(cpus_hardware_enabled);
2582         __free_page(hwpoison_page);
2583         __free_page(bad_page);
2584 }
2585 EXPORT_SYMBOL_GPL(kvm_exit);