virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/syscore_ops.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/pgtable.h>
  57
  58 #include "coalesced_mmio.h"
  59 #include "async_pf.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/kvm.h>
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 /*
  68  * Ordering of locks:
  69  *
  70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  71  */
  72
  73 DEFINE_RAW_SPINLOCK(kvm_lock);
  74 LIST_HEAD(vm_list);
  75
  76 static cpumask_var_t cpus_hardware_enabled;
  77 static int kvm_usage_count = 0;
  78 static atomic_t hardware_enable_failed;
  79
  80 struct kmem_cache *kvm_vcpu_cache;
  81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  82
  83 static __read_mostly struct preempt_ops kvm_preempt_ops;
  84
  85 struct dentry *kvm_debugfs_dir;
  86
  87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  88                            unsigned long arg);
  89 #ifdef CONFIG_COMPAT
  90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  91                                   unsigned long arg);
  92 #endif
  93 static int hardware_enable_all(void);
  94 static void hardware_disable_all(void);
  95
  96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  97
  98 bool kvm_rebooting;
  99 EXPORT_SYMBOL_GPL(kvm_rebooting);
 100
 101 static bool largepages_enabled = true;
 102
 103 bool kvm_is_mmio_pfn(pfn_t pfn)
 104 {
 105         if (pfn_valid(pfn)) {
 106                 int reserved;
 107                 struct page *tail = pfn_to_page(pfn);
 108                 struct page *head = compound_trans_head(tail);
 109                 reserved = PageReserved(head);
 110                 if (head != tail) {
 111                         /*
 112                          * "head" is not a dangling pointer
 113                          * (compound_trans_head takes care of that)
 114                          * but the hugepage may have been splitted
 115                          * from under us (and we may not hold a
 116                          * reference count on the head page so it can
 117                          * be reused before we run PageReferenced), so
 118                          * we've to check PageTail before returning
 119                          * what we just read.
 120                          */
 121                         smp_rmb();
 122                         if (PageTail(tail))
 123                                 return reserved;
 124                 }
 125                 return PageReserved(tail);
 126         }
 127
 128         return true;
 129 }
 130
 131 /*
 132  * Switches to specified vcpu, until a matching vcpu_put()
 133  */
 134 int vcpu_load(struct kvm_vcpu *vcpu)
 135 {
 136         int cpu;
 137
 138         if (mutex_lock_killable(&vcpu->mutex))
 139                 return -EINTR;
 140         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 141                 /* The thread running this VCPU changed. */
 142                 struct pid *oldpid = vcpu->pid;
 143                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 144                 rcu_assign_pointer(vcpu->pid, newpid);
 145                 synchronize_rcu();
 146                 put_pid(oldpid);
 147         }
 148         cpu = get_cpu();
 149         preempt_notifier_register(&vcpu->preempt_notifier);
 150         kvm_arch_vcpu_load(vcpu, cpu);
 151         put_cpu();
 152         return 0;
 153 }
 154
 155 void vcpu_put(struct kvm_vcpu *vcpu)
 156 {
 157         preempt_disable();
 158         kvm_arch_vcpu_put(vcpu);
 159         preempt_notifier_unregister(&vcpu->preempt_notifier);
 160         preempt_enable();
 161         mutex_unlock(&vcpu->mutex);
 162 }
 163
 164 static void ack_flush(void *_completed)
 165 {
 166 }
 167
 168 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 169 {
 170         int i, cpu, me;
 171         cpumask_var_t cpus;
 172         bool called = true;
 173         struct kvm_vcpu *vcpu;
 174
 175         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 176
 177         me = get_cpu();
 178         kvm_for_each_vcpu(i, vcpu, kvm) {
 179                 kvm_make_request(req, vcpu);
 180                 cpu = vcpu->cpu;
 181
 182                 /* Set ->requests bit before we read ->mode */
 183                 smp_mb();
 184
 185                 if (cpus != NULL && cpu != -1 && cpu != me &&
 186                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 187                         cpumask_set_cpu(cpu, cpus);
 188         }
 189         if (unlikely(cpus == NULL))
 190                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 191         else if (!cpumask_empty(cpus))
 192                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 193         else
 194                 called = false;
 195         put_cpu();
 196         free_cpumask_var(cpus);
 197         return called;
 198 }
 199
 200 void kvm_flush_remote_tlbs(struct kvm *kvm)
 201 {
 202         long dirty_count = kvm->tlbs_dirty;
 203
 204         smp_mb();
 205         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 206                 ++kvm->stat.remote_tlb_flush;
 207         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 208 }
 209
 210 void kvm_reload_remote_mmus(struct kvm *kvm)
 211 {
 212         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 213 }
 214
 215 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 216 {
 217         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 218 }
 219
 220 void kvm_make_update_eoibitmap_request(struct kvm *kvm)
 221 {
 222         make_all_cpus_request(kvm, KVM_REQ_EOIBITMAP);
 223 }
 224
 225 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 226 {
 227         struct page *page;
 228         int r;
 229
 230         mutex_init(&vcpu->mutex);
 231         vcpu->cpu = -1;
 232         vcpu->kvm = kvm;
 233         vcpu->vcpu_id = id;
 234         vcpu->pid = NULL;
 235         init_waitqueue_head(&vcpu->wq);
 236         kvm_async_pf_vcpu_init(vcpu);
 237
 238         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 239         if (!page) {
 240                 r = -ENOMEM;
 241                 goto fail;
 242         }
 243         vcpu->run = page_address(page);
 244
 245         kvm_vcpu_set_in_spin_loop(vcpu, false);
 246         kvm_vcpu_set_dy_eligible(vcpu, false);
 247
 248         r = kvm_arch_vcpu_init(vcpu);
 249         if (r < 0)
 250                 goto fail_free_run;
 251         return 0;
 252
 253 fail_free_run:
 254         free_page((unsigned long)vcpu->run);
 255 fail:
 256         return r;
 257 }
 258 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 259
 260 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 261 {
 262         put_pid(vcpu->pid);
 263         kvm_arch_vcpu_uninit(vcpu);
 264         free_page((unsigned long)vcpu->run);
 265 }
 266 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 267
 268 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 269 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 270 {
 271         return container_of(mn, struct kvm, mmu_notifier);
 272 }
 273
 274 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 275                                              struct mm_struct *mm,
 276                                              unsigned long address)
 277 {
 278         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 279         int need_tlb_flush, idx;
 280
 281         /*
 282          * When ->invalidate_page runs, the linux pte has been zapped
 283          * already but the page is still allocated until
 284          * ->invalidate_page returns. So if we increase the sequence
 285          * here the kvm page fault will notice if the spte can't be
 286          * established because the page is going to be freed. If
 287          * instead the kvm page fault establishes the spte before
 288          * ->invalidate_page runs, kvm_unmap_hva will release it
 289          * before returning.
 290          *
 291          * The sequence increase only need to be seen at spin_unlock
 292          * time, and not at spin_lock time.
 293          *
 294          * Increasing the sequence after the spin_unlock would be
 295          * unsafe because the kvm page fault could then establish the
 296          * pte after kvm_unmap_hva returned, without noticing the page
 297          * is going to be freed.
 298          */
 299         idx = srcu_read_lock(&kvm->srcu);
 300         spin_lock(&kvm->mmu_lock);
 301
 302         kvm->mmu_notifier_seq++;
 303         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 304         /* we've to flush the tlb before the pages can be freed */
 305         if (need_tlb_flush)
 306                 kvm_flush_remote_tlbs(kvm);
 307
 308         spin_unlock(&kvm->mmu_lock);
 309         srcu_read_unlock(&kvm->srcu, idx);
 310 }
 311
 312 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 313                                         struct mm_struct *mm,
 314                                         unsigned long address,
 315                                         pte_t pte)
 316 {
 317         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 318         int idx;
 319
 320         idx = srcu_read_lock(&kvm->srcu);
 321         spin_lock(&kvm->mmu_lock);
 322         kvm->mmu_notifier_seq++;
 323         kvm_set_spte_hva(kvm, address, pte);
 324         spin_unlock(&kvm->mmu_lock);
 325         srcu_read_unlock(&kvm->srcu, idx);
 326 }
 327
 328 static void kvm_mmu_notifier_invalidate_range_start(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         int need_tlb_flush = 0, idx;
 335
 336         idx = srcu_read_lock(&kvm->srcu);
 337         spin_lock(&kvm->mmu_lock);
 338         /*
 339          * The count increase must become visible at unlock time as no
 340          * spte can be established without taking the mmu_lock and
 341          * count is also read inside the mmu_lock critical section.
 342          */
 343         kvm->mmu_notifier_count++;
 344         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 345         need_tlb_flush |= kvm->tlbs_dirty;
 346         /* we've to flush the tlb before the pages can be freed */
 347         if (need_tlb_flush)
 348                 kvm_flush_remote_tlbs(kvm);
 349
 350         spin_unlock(&kvm->mmu_lock);
 351         srcu_read_unlock(&kvm->srcu, idx);
 352 }
 353
 354 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 355                                                   struct mm_struct *mm,
 356                                                   unsigned long start,
 357                                                   unsigned long end)
 358 {
 359         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 360
 361         spin_lock(&kvm->mmu_lock);
 362         /*
 363          * This sequence increase will notify the kvm page fault that
 364          * the page that is going to be mapped in the spte could have
 365          * been freed.
 366          */
 367         kvm->mmu_notifier_seq++;
 368         smp_wmb();
 369         /*
 370          * The above sequence increase must be visible before the
 371          * below count decrease, which is ensured by the smp_wmb above
 372          * in conjunction with the smp_rmb in mmu_notifier_retry().
 373          */
 374         kvm->mmu_notifier_count--;
 375         spin_unlock(&kvm->mmu_lock);
 376
 377         BUG_ON(kvm->mmu_notifier_count < 0);
 378 }
 379
 380 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 381                                               struct mm_struct *mm,
 382                                               unsigned long address)
 383 {
 384         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 385         int young, idx;
 386
 387         idx = srcu_read_lock(&kvm->srcu);
 388         spin_lock(&kvm->mmu_lock);
 389
 390         young = kvm_age_hva(kvm, address);
 391         if (young)
 392                 kvm_flush_remote_tlbs(kvm);
 393
 394         spin_unlock(&kvm->mmu_lock);
 395         srcu_read_unlock(&kvm->srcu, idx);
 396
 397         return young;
 398 }
 399
 400 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 401                                        struct mm_struct *mm,
 402                                        unsigned long address)
 403 {
 404         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 405         int young, idx;
 406
 407         idx = srcu_read_lock(&kvm->srcu);
 408         spin_lock(&kvm->mmu_lock);
 409         young = kvm_test_age_hva(kvm, address);
 410         spin_unlock(&kvm->mmu_lock);
 411         srcu_read_unlock(&kvm->srcu, idx);
 412
 413         return young;
 414 }
 415
 416 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 417                                      struct mm_struct *mm)
 418 {
 419         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 420         int idx;
 421
 422         idx = srcu_read_lock(&kvm->srcu);
 423         kvm_arch_flush_shadow_all(kvm);
 424         srcu_read_unlock(&kvm->srcu, idx);
 425 }
 426
 427 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 428         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 429         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 430         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 431         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 432         .test_young             = kvm_mmu_notifier_test_young,
 433         .change_pte             = kvm_mmu_notifier_change_pte,
 434         .release                = kvm_mmu_notifier_release,
 435 };
 436
 437 static int kvm_init_mmu_notifier(struct kvm *kvm)
 438 {
 439         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 440         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 441 }
 442
 443 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 444
 445 static int kvm_init_mmu_notifier(struct kvm *kvm)
 446 {
 447         return 0;
 448 }
 449
 450 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 451
 452 static void kvm_init_memslots_id(struct kvm *kvm)
 453 {
 454         int i;
 455         struct kvm_memslots *slots = kvm->memslots;
 456
 457         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 458                 slots->id_to_index[i] = slots->memslots[i].id = i;
 459 }
 460
 461 static struct kvm *kvm_create_vm(unsigned long type)
 462 {
 463         int r, i;
 464         struct kvm *kvm = kvm_arch_alloc_vm();
 465
 466         if (!kvm)
 467                 return ERR_PTR(-ENOMEM);
 468
 469         r = kvm_arch_init_vm(kvm, type);
 470         if (r)
 471                 goto out_err_nodisable;
 472
 473         r = hardware_enable_all();
 474         if (r)
 475                 goto out_err_nodisable;
 476
 477 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 478         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 479         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 480 #endif
 481
 482         BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 483
 484         r = -ENOMEM;
 485         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 486         if (!kvm->memslots)
 487                 goto out_err_nosrcu;
 488         kvm_init_memslots_id(kvm);
 489         if (init_srcu_struct(&kvm->srcu))
 490                 goto out_err_nosrcu;
 491         for (i = 0; i < KVM_NR_BUSES; i++) {
 492                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 493                                         GFP_KERNEL);
 494                 if (!kvm->buses[i])
 495                         goto out_err;
 496         }
 497
 498         spin_lock_init(&kvm->mmu_lock);
 499         kvm->mm = current->mm;
 500         atomic_inc(&kvm->mm->mm_count);
 501         kvm_eventfd_init(kvm);
 502         mutex_init(&kvm->lock);
 503         mutex_init(&kvm->irq_lock);
 504         mutex_init(&kvm->slots_lock);
 505         atomic_set(&kvm->users_count, 1);
 506
 507         r = kvm_init_mmu_notifier(kvm);
 508         if (r)
 509                 goto out_err;
 510
 511         raw_spin_lock(&kvm_lock);
 512         list_add(&kvm->vm_list, &vm_list);
 513         raw_spin_unlock(&kvm_lock);
 514
 515         return kvm;
 516
 517 out_err:
 518         cleanup_srcu_struct(&kvm->srcu);
 519 out_err_nosrcu:
 520         hardware_disable_all();
 521 out_err_nodisable:
 522         for (i = 0; i < KVM_NR_BUSES; i++)
 523                 kfree(kvm->buses[i]);
 524         kfree(kvm->memslots);
 525         kvm_arch_free_vm(kvm);
 526         return ERR_PTR(r);
 527 }
 528
 529 /*
 530  * Avoid using vmalloc for a small buffer.
 531  * Should not be used when the size is statically known.
 532  */
 533 void *kvm_kvzalloc(unsigned long size)
 534 {
 535         if (size > PAGE_SIZE)
 536                 return vzalloc(size);
 537         else
 538                 return kzalloc(size, GFP_KERNEL);
 539 }
 540
 541 void kvm_kvfree(const void *addr)
 542 {
 543         if (is_vmalloc_addr(addr))
 544                 vfree(addr);
 545         else
 546                 kfree(addr);
 547 }
 548
 549 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 550 {
 551         if (!memslot->dirty_bitmap)
 552                 return;
 553
 554         kvm_kvfree(memslot->dirty_bitmap);
 555         memslot->dirty_bitmap = NULL;
 556 }
 557
 558 /*
 559  * Free any memory in @free but not in @dont.
 560  */
 561 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 562                                   struct kvm_memory_slot *dont)
 563 {
 564         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 565                 kvm_destroy_dirty_bitmap(free);
 566
 567         kvm_arch_free_memslot(free, dont);
 568
 569         free->npages = 0;
 570 }
 571
 572 void kvm_free_physmem(struct kvm *kvm)
 573 {
 574         struct kvm_memslots *slots = kvm->memslots;
 575         struct kvm_memory_slot *memslot;
 576
 577         kvm_for_each_memslot(memslot, slots)
 578                 kvm_free_physmem_slot(memslot, NULL);
 579
 580         kfree(kvm->memslots);
 581 }
 582
 583 static void kvm_destroy_vm(struct kvm *kvm)
 584 {
 585         int i;
 586         struct mm_struct *mm = kvm->mm;
 587
 588         kvm_arch_sync_events(kvm);
 589         raw_spin_lock(&kvm_lock);
 590         list_del(&kvm->vm_list);
 591         raw_spin_unlock(&kvm_lock);
 592         kvm_free_irq_routing(kvm);
 593         for (i = 0; i < KVM_NR_BUSES; i++)
 594                 kvm_io_bus_destroy(kvm->buses[i]);
 595         kvm_coalesced_mmio_free(kvm);
 596 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 597         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 598 #else
 599         kvm_arch_flush_shadow_all(kvm);
 600 #endif
 601         kvm_arch_destroy_vm(kvm);
 602         kvm_free_physmem(kvm);
 603         cleanup_srcu_struct(&kvm->srcu);
 604         kvm_arch_free_vm(kvm);
 605         hardware_disable_all();
 606         mmdrop(mm);
 607 }
 608
 609 void kvm_get_kvm(struct kvm *kvm)
 610 {
 611         atomic_inc(&kvm->users_count);
 612 }
 613 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 614
 615 void kvm_put_kvm(struct kvm *kvm)
 616 {
 617         if (atomic_dec_and_test(&kvm->users_count))
 618                 kvm_destroy_vm(kvm);
 619 }
 620 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 621
 622
 623 static int kvm_vm_release(struct inode *inode, struct file *filp)
 624 {
 625         struct kvm *kvm = filp->private_data;
 626
 627         kvm_irqfd_release(kvm);
 628
 629         kvm_put_kvm(kvm);
 630         return 0;
 631 }
 632
 633 /*
 634  * Allocation size is twice as large as the actual dirty bitmap size.
 635  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 636  */
 637 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 638 {
 639 #ifndef CONFIG_S390
 640         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 641
 642         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 643         if (!memslot->dirty_bitmap)
 644                 return -ENOMEM;
 645
 646 #endif /* !CONFIG_S390 */
 647         return 0;
 648 }
 649
 650 static int cmp_memslot(const void *slot1, const void *slot2)
 651 {
 652         struct kvm_memory_slot *s1, *s2;
 653
 654         s1 = (struct kvm_memory_slot *)slot1;
 655         s2 = (struct kvm_memory_slot *)slot2;
 656
 657         if (s1->npages < s2->npages)
 658                 return 1;
 659         if (s1->npages > s2->npages)
 660                 return -1;
 661
 662         return 0;
 663 }
 664
 665 /*
 666  * Sort the memslots base on its size, so the larger slots
 667  * will get better fit.
 668  */
 669 static void sort_memslots(struct kvm_memslots *slots)
 670 {
 671         int i;
 672
 673         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 674               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 675
 676         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 677                 slots->id_to_index[slots->memslots[i].id] = i;
 678 }
 679
 680 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new,
 681                      u64 last_generation)
 682 {
 683         if (new) {
 684                 int id = new->id;
 685                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 686                 unsigned long npages = old->npages;
 687
 688                 *old = *new;
 689                 if (new->npages != npages)
 690                         sort_memslots(slots);
 691         }
 692
 693         slots->generation = last_generation + 1;
 694 }
 695
 696 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 697 {
 698         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 699
 700 #ifdef KVM_CAP_READONLY_MEM
 701         valid_flags |= KVM_MEM_READONLY;
 702 #endif
 703
 704         if (mem->flags & ~valid_flags)
 705                 return -EINVAL;
 706
 707         return 0;
 708 }
 709
 710 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 711                 struct kvm_memslots *slots, struct kvm_memory_slot *new)
 712 {
 713         struct kvm_memslots *old_memslots = kvm->memslots;
 714
 715         update_memslots(slots, new, kvm->memslots->generation);
 716         rcu_assign_pointer(kvm->memslots, slots);
 717         synchronize_srcu_expedited(&kvm->srcu);
 718         return old_memslots;
 719 }
 720
 721 /*
 722  * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations:
 723  * - create a new memory slot
 724  * - delete an existing memory slot
 725  * - modify an existing memory slot
 726  *   -- move it in the guest physical memory space
 727  *   -- just change its flags
 728  *
 729  * Since flags can be changed by some of these operations, the following
 730  * differentiation is the best we can do for __kvm_set_memory_region():
 731  */
 732 enum kvm_mr_change {
 733         KVM_MR_CREATE,
 734         KVM_MR_DELETE,
 735         KVM_MR_MOVE,
 736         KVM_MR_FLAGS_ONLY,
 737 };
 738
 739 /*
 740  * Allocate some memory and give it an address in the guest physical address
 741  * space.
 742  *
 743  * Discontiguous memory is allowed, mostly for framebuffers.
 744  *
 745  * Must be called holding mmap_sem for write.
 746  */
 747 int __kvm_set_memory_region(struct kvm *kvm,
 748                             struct kvm_userspace_memory_region *mem,
 749                             bool user_alloc)
 750 {
 751         int r;
 752         gfn_t base_gfn;
 753         unsigned long npages;
 754         struct kvm_memory_slot *slot;
 755         struct kvm_memory_slot old, new;
 756         struct kvm_memslots *slots = NULL, *old_memslots;
 757         bool old_iommu_mapped;
 758         enum kvm_mr_change change;
 759
 760         r = check_memory_region_flags(mem);
 761         if (r)
 762                 goto out;
 763
 764         r = -EINVAL;
 765         /* General sanity checks */
 766         if (mem->memory_size & (PAGE_SIZE - 1))
 767                 goto out;
 768         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 769                 goto out;
 770         /* We can read the guest memory with __xxx_user() later on. */
 771         if (user_alloc &&
 772             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 773              !access_ok(VERIFY_WRITE,
 774                         (void __user *)(unsigned long)mem->userspace_addr,
 775                         mem->memory_size)))
 776                 goto out;
 777         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 778                 goto out;
 779         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 780                 goto out;
 781
 782         slot = id_to_memslot(kvm->memslots, mem->slot);
 783         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 784         npages = mem->memory_size >> PAGE_SHIFT;
 785
 786         r = -EINVAL;
 787         if (npages > KVM_MEM_MAX_NR_PAGES)
 788                 goto out;
 789
 790         if (!npages)
 791                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 792
 793         new = old = *slot;
 794
 795         new.id = mem->slot;
 796         new.base_gfn = base_gfn;
 797         new.npages = npages;
 798         new.flags = mem->flags;
 799
 800         old_iommu_mapped = old.npages;
 801
 802         r = -EINVAL;
 803         if (npages) {
 804                 if (!old.npages)
 805                         change = KVM_MR_CREATE;
 806                 else { /* Modify an existing slot. */
 807                         if ((mem->userspace_addr != old.userspace_addr) ||
 808                             (npages != old.npages))
 809                                 goto out;
 810
 811                         if (base_gfn != old.base_gfn)
 812                                 change = KVM_MR_MOVE;
 813                         else if (new.flags != old.flags)
 814                                 change = KVM_MR_FLAGS_ONLY;
 815                         else { /* Nothing to change. */
 816                                 r = 0;
 817                                 goto out;
 818                         }
 819                 }
 820         } else if (old.npages) {
 821                 change = KVM_MR_DELETE;
 822         } else /* Modify a non-existent slot: disallowed. */
 823                 goto out;
 824
 825         if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 826                 /* Check for overlaps */
 827                 r = -EEXIST;
 828                 kvm_for_each_memslot(slot, kvm->memslots) {
 829                         if ((slot->id >= KVM_USER_MEM_SLOTS) ||
 830                             (slot->id == mem->slot))
 831                                 continue;
 832                         if (!((base_gfn + npages <= slot->base_gfn) ||
 833                               (base_gfn >= slot->base_gfn + slot->npages)))
 834                                 goto out;
 835                 }
 836         }
 837
 838         /* Free page dirty bitmap if unneeded */
 839         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 840                 new.dirty_bitmap = NULL;
 841
 842         r = -ENOMEM;
 843         if (change == KVM_MR_CREATE) {
 844                 new.user_alloc = user_alloc;
 845                 new.userspace_addr = mem->userspace_addr;
 846
 847                 if (kvm_arch_create_memslot(&new, npages))
 848                         goto out_free;
 849         }
 850
 851         /* Allocate page dirty bitmap if needed */
 852         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 853                 if (kvm_create_dirty_bitmap(&new) < 0)
 854                         goto out_free;
 855         }
 856
 857         if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
 858                 r = -ENOMEM;
 859                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 860                                 GFP_KERNEL);
 861                 if (!slots)
 862                         goto out_free;
 863                 slot = id_to_memslot(slots, mem->slot);
 864                 slot->flags |= KVM_MEMSLOT_INVALID;
 865
 866                 old_memslots = install_new_memslots(kvm, slots, NULL);
 867
 868                 /* slot was deleted or moved, clear iommu mapping */
 869                 kvm_iommu_unmap_pages(kvm, &old);
 870                 old_iommu_mapped = false;
 871                 /* From this point no new shadow pages pointing to a deleted,
 872                  * or moved, memslot will be created.
 873                  *
 874                  * validation of sp->gfn happens in:
 875                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 876                  *      - kvm_is_visible_gfn (mmu_check_roots)
 877                  */
 878                 kvm_arch_flush_shadow_memslot(kvm, slot);
 879                 slots = old_memslots;
 880         }
 881
 882         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 883         if (r)
 884                 goto out_slots;
 885
 886         r = -ENOMEM;
 887         /*
 888          * We can re-use the old_memslots from above, the only difference
 889          * from the currently installed memslots is the invalid flag.  This
 890          * will get overwritten by update_memslots anyway.
 891          */
 892         if (!slots) {
 893                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 894                                 GFP_KERNEL);
 895                 if (!slots)
 896                         goto out_free;
 897         }
 898
 899         /*
 900          * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 901          * un-mapped and re-mapped if their base changes or if flags that the
 902          * iommu cares about change (read-only).  Base change unmapping is
 903          * handled above with slot deletion, so we only unmap incompatible
 904          * flags here.  Anything else the iommu might care about for existing
 905          * slots (size changes, userspace addr changes) is disallowed above,
 906          * so any other attribute changes getting here can be skipped.
 907          */
 908         if (change != KVM_MR_DELETE) {
 909                 if (old_iommu_mapped &&
 910                     ((new.flags ^ old.flags) & KVM_MEM_READONLY)) {
 911                         kvm_iommu_unmap_pages(kvm, &old);
 912                         old_iommu_mapped = false;
 913                 }
 914
 915                 if (!old_iommu_mapped) {
 916                         r = kvm_iommu_map_pages(kvm, &new);
 917                         if (r)
 918                                 goto out_slots;
 919                 }
 920         }
 921
 922         /* actual memory is freed via old in kvm_free_physmem_slot below */
 923         if (change == KVM_MR_DELETE) {
 924                 new.dirty_bitmap = NULL;
 925                 memset(&new.arch, 0, sizeof(new.arch));
 926         }
 927
 928         old_memslots = install_new_memslots(kvm, slots, &new);
 929
 930         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 931
 932         kvm_free_physmem_slot(&old, &new);
 933         kfree(old_memslots);
 934
 935         return 0;
 936
 937 out_slots:
 938         kfree(slots);
 939 out_free:
 940         kvm_free_physmem_slot(&new, &old);
 941 out:
 942         return r;
 943 }
 944 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 945
 946 int kvm_set_memory_region(struct kvm *kvm,
 947                           struct kvm_userspace_memory_region *mem,
 948                           bool user_alloc)
 949 {
 950         int r;
 951
 952         mutex_lock(&kvm->slots_lock);
 953         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 954         mutex_unlock(&kvm->slots_lock);
 955         return r;
 956 }
 957 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 958
 959 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 960                                    struct
 961                                    kvm_userspace_memory_region *mem,
 962                                    bool user_alloc)
 963 {
 964         if (mem->slot >= KVM_USER_MEM_SLOTS)
 965                 return -EINVAL;
 966         return kvm_set_memory_region(kvm, mem, user_alloc);
 967 }
 968
 969 int kvm_get_dirty_log(struct kvm *kvm,
 970                         struct kvm_dirty_log *log, int *is_dirty)
 971 {
 972         struct kvm_memory_slot *memslot;
 973         int r, i;
 974         unsigned long n;
 975         unsigned long any = 0;
 976
 977         r = -EINVAL;
 978         if (log->slot >= KVM_USER_MEM_SLOTS)
 979                 goto out;
 980
 981         memslot = id_to_memslot(kvm->memslots, log->slot);
 982         r = -ENOENT;
 983         if (!memslot->dirty_bitmap)
 984                 goto out;
 985
 986         n = kvm_dirty_bitmap_bytes(memslot);
 987
 988         for (i = 0; !any && i < n/sizeof(long); ++i)
 989                 any = memslot->dirty_bitmap[i];
 990
 991         r = -EFAULT;
 992         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 993                 goto out;
 994
 995         if (any)
 996                 *is_dirty = 1;
 997
 998         r = 0;
 999 out:
1000         return r;
1001 }
1002
1003 bool kvm_largepages_enabled(void)
1004 {
1005         return largepages_enabled;
1006 }
1007
1008 void kvm_disable_largepages(void)
1009 {
1010         largepages_enabled = false;
1011 }
1012 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1013
1014 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1015 {
1016         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1017 }
1018 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1019
1020 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1021 {
1022         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1023
1024         if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1025               memslot->flags & KVM_MEMSLOT_INVALID)
1026                 return 0;
1027
1028         return 1;
1029 }
1030 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1031
1032 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1033 {
1034         struct vm_area_struct *vma;
1035         unsigned long addr, size;
1036
1037         size = PAGE_SIZE;
1038
1039         addr = gfn_to_hva(kvm, gfn);
1040         if (kvm_is_error_hva(addr))
1041                 return PAGE_SIZE;
1042
1043         down_read(&current->mm->mmap_sem);
1044         vma = find_vma(current->mm, addr);
1045         if (!vma)
1046                 goto out;
1047
1048         size = vma_kernel_pagesize(vma);
1049
1050 out:
1051         up_read(&current->mm->mmap_sem);
1052
1053         return size;
1054 }
1055
1056 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1057 {
1058         return slot->flags & KVM_MEM_READONLY;
1059 }
1060
1061 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1062                                        gfn_t *nr_pages, bool write)
1063 {
1064         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1065                 return KVM_HVA_ERR_BAD;
1066
1067         if (memslot_is_readonly(slot) && write)
1068                 return KVM_HVA_ERR_RO_BAD;
1069
1070         if (nr_pages)
1071                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1072
1073         return __gfn_to_hva_memslot(slot, gfn);
1074 }
1075
1076 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1077                                      gfn_t *nr_pages)
1078 {
1079         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1080 }
1081
1082 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1083                                  gfn_t gfn)
1084 {
1085         return gfn_to_hva_many(slot, gfn, NULL);
1086 }
1087 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1088
1089 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1090 {
1091         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1092 }
1093 EXPORT_SYMBOL_GPL(gfn_to_hva);
1094
1095 /*
1096  * The hva returned by this function is only allowed to be read.
1097  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1098  */
1099 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1100 {
1101         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1102 }
1103
1104 static int kvm_read_hva(void *data, void __user *hva, int len)
1105 {
1106         return __copy_from_user(data, hva, len);
1107 }
1108
1109 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1110 {
1111         return __copy_from_user_inatomic(data, hva, len);
1112 }
1113
1114 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1115         unsigned long start, int write, struct page **page)
1116 {
1117         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1118
1119         if (write)
1120                 flags |= FOLL_WRITE;
1121
1122         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1123 }
1124
1125 static inline int check_user_page_hwpoison(unsigned long addr)
1126 {
1127         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1128
1129         rc = __get_user_pages(current, current->mm, addr, 1,
1130                               flags, NULL, NULL, NULL);
1131         return rc == -EHWPOISON;
1132 }
1133
1134 /*
1135  * The atomic path to get the writable pfn which will be stored in @pfn,
1136  * true indicates success, otherwise false is returned.
1137  */
1138 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1139                             bool write_fault, bool *writable, pfn_t *pfn)
1140 {
1141         struct page *page[1];
1142         int npages;
1143
1144         if (!(async || atomic))
1145                 return false;
1146
1147         /*
1148          * Fast pin a writable pfn only if it is a write fault request
1149          * or the caller allows to map a writable pfn for a read fault
1150          * request.
1151          */
1152         if (!(write_fault || writable))
1153                 return false;
1154
1155         npages = __get_user_pages_fast(addr, 1, 1, page);
1156         if (npages == 1) {
1157                 *pfn = page_to_pfn(page[0]);
1158
1159                 if (writable)
1160                         *writable = true;
1161                 return true;
1162         }
1163
1164         return false;
1165 }
1166
1167 /*
1168  * The slow path to get the pfn of the specified host virtual address,
1169  * 1 indicates success, -errno is returned if error is detected.
1170  */
1171 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1172                            bool *writable, pfn_t *pfn)
1173 {
1174         struct page *page[1];
1175         int npages = 0;
1176
1177         might_sleep();
1178
1179         if (writable)
1180                 *writable = write_fault;
1181
1182         if (async) {
1183                 down_read(&current->mm->mmap_sem);
1184                 npages = get_user_page_nowait(current, current->mm,
1185                                               addr, write_fault, page);
1186                 up_read(&current->mm->mmap_sem);
1187         } else
1188                 npages = get_user_pages_fast(addr, 1, write_fault,
1189                                              page);
1190         if (npages != 1)
1191                 return npages;
1192
1193         /* map read fault as writable if possible */
1194         if (unlikely(!write_fault) && writable) {
1195                 struct page *wpage[1];
1196
1197                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1198                 if (npages == 1) {
1199                         *writable = true;
1200                         put_page(page[0]);
1201                         page[0] = wpage[0];
1202                 }
1203
1204                 npages = 1;
1205         }
1206         *pfn = page_to_pfn(page[0]);
1207         return npages;
1208 }
1209
1210 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1211 {
1212         if (unlikely(!(vma->vm_flags & VM_READ)))
1213                 return false;
1214
1215         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1216                 return false;
1217
1218         return true;
1219 }
1220
1221 /*
1222  * Pin guest page in memory and return its pfn.
1223  * @addr: host virtual address which maps memory to the guest
1224  * @atomic: whether this function can sleep
1225  * @async: whether this function need to wait IO complete if the
1226  *         host page is not in the memory
1227  * @write_fault: whether we should get a writable host page
1228  * @writable: whether it allows to map a writable host page for !@write_fault
1229  *
1230  * The function will map a writable host page for these two cases:
1231  * 1): @write_fault = true
1232  * 2): @write_fault = false && @writable, @writable will tell the caller
1233  *     whether the mapping is writable.
1234  */
1235 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1236                         bool write_fault, bool *writable)
1237 {
1238         struct vm_area_struct *vma;
1239         pfn_t pfn = 0;
1240         int npages;
1241
1242         /* we can do it either atomically or asynchronously, not both */
1243         BUG_ON(atomic && async);
1244
1245         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1246                 return pfn;
1247
1248         if (atomic)
1249                 return KVM_PFN_ERR_FAULT;
1250
1251         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1252         if (npages == 1)
1253                 return pfn;
1254
1255         down_read(&current->mm->mmap_sem);
1256         if (npages == -EHWPOISON ||
1257               (!async && check_user_page_hwpoison(addr))) {
1258                 pfn = KVM_PFN_ERR_HWPOISON;
1259                 goto exit;
1260         }
1261
1262         vma = find_vma_intersection(current->mm, addr, addr + 1);
1263
1264         if (vma == NULL)
1265                 pfn = KVM_PFN_ERR_FAULT;
1266         else if ((vma->vm_flags & VM_PFNMAP)) {
1267                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1268                         vma->vm_pgoff;
1269                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1270         } else {
1271                 if (async && vma_is_valid(vma, write_fault))
1272                         *async = true;
1273                 pfn = KVM_PFN_ERR_FAULT;
1274         }
1275 exit:
1276         up_read(&current->mm->mmap_sem);
1277         return pfn;
1278 }
1279
1280 static pfn_t
1281 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1282                      bool *async, bool write_fault, bool *writable)
1283 {
1284         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1285
1286         if (addr == KVM_HVA_ERR_RO_BAD)
1287                 return KVM_PFN_ERR_RO_FAULT;
1288
1289         if (kvm_is_error_hva(addr))
1290                 return KVM_PFN_NOSLOT;
1291
1292         /* Do not map writable pfn in the readonly memslot. */
1293         if (writable && memslot_is_readonly(slot)) {
1294                 *writable = false;
1295                 writable = NULL;
1296         }
1297
1298         return hva_to_pfn(addr, atomic, async, write_fault,
1299                           writable);
1300 }
1301
1302 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1303                           bool write_fault, bool *writable)
1304 {
1305         struct kvm_memory_slot *slot;
1306
1307         if (async)
1308                 *async = false;
1309
1310         slot = gfn_to_memslot(kvm, gfn);
1311
1312         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1313                                     writable);
1314 }
1315
1316 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1317 {
1318         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1319 }
1320 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1321
1322 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1323                        bool write_fault, bool *writable)
1324 {
1325         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1326 }
1327 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1328
1329 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1330 {
1331         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1332 }
1333 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1334
1335 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1336                       bool *writable)
1337 {
1338         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1339 }
1340 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1341
1342 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1343 {
1344         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1345 }
1346
1347 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1348 {
1349         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1350 }
1351 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1352
1353 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1354                                                                   int nr_pages)
1355 {
1356         unsigned long addr;
1357         gfn_t entry;
1358
1359         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1360         if (kvm_is_error_hva(addr))
1361                 return -1;
1362
1363         if (entry < nr_pages)
1364                 return 0;
1365
1366         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1367 }
1368 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1369
1370 static struct page *kvm_pfn_to_page(pfn_t pfn)
1371 {
1372         if (is_error_noslot_pfn(pfn))
1373                 return KVM_ERR_PTR_BAD_PAGE;
1374
1375         if (kvm_is_mmio_pfn(pfn)) {
1376                 WARN_ON(1);
1377                 return KVM_ERR_PTR_BAD_PAGE;
1378         }
1379
1380         return pfn_to_page(pfn);
1381 }
1382
1383 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1384 {
1385         pfn_t pfn;
1386
1387         pfn = gfn_to_pfn(kvm, gfn);
1388
1389         return kvm_pfn_to_page(pfn);
1390 }
1391
1392 EXPORT_SYMBOL_GPL(gfn_to_page);
1393
1394 void kvm_release_page_clean(struct page *page)
1395 {
1396         WARN_ON(is_error_page(page));
1397
1398         kvm_release_pfn_clean(page_to_pfn(page));
1399 }
1400 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1401
1402 void kvm_release_pfn_clean(pfn_t pfn)
1403 {
1404         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1405                 put_page(pfn_to_page(pfn));
1406 }
1407 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1408
1409 void kvm_release_page_dirty(struct page *page)
1410 {
1411         WARN_ON(is_error_page(page));
1412
1413         kvm_release_pfn_dirty(page_to_pfn(page));
1414 }
1415 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1416
1417 void kvm_release_pfn_dirty(pfn_t pfn)
1418 {
1419         kvm_set_pfn_dirty(pfn);
1420         kvm_release_pfn_clean(pfn);
1421 }
1422 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1423
1424 void kvm_set_page_dirty(struct page *page)
1425 {
1426         kvm_set_pfn_dirty(page_to_pfn(page));
1427 }
1428 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1429
1430 void kvm_set_pfn_dirty(pfn_t pfn)
1431 {
1432         if (!kvm_is_mmio_pfn(pfn)) {
1433                 struct page *page = pfn_to_page(pfn);
1434                 if (!PageReserved(page))
1435                         SetPageDirty(page);
1436         }
1437 }
1438 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1439
1440 void kvm_set_pfn_accessed(pfn_t pfn)
1441 {
1442         if (!kvm_is_mmio_pfn(pfn))
1443                 mark_page_accessed(pfn_to_page(pfn));
1444 }
1445 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1446
1447 void kvm_get_pfn(pfn_t pfn)
1448 {
1449         if (!kvm_is_mmio_pfn(pfn))
1450                 get_page(pfn_to_page(pfn));
1451 }
1452 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1453
1454 static int next_segment(unsigned long len, int offset)
1455 {
1456         if (len > PAGE_SIZE - offset)
1457                 return PAGE_SIZE - offset;
1458         else
1459                 return len;
1460 }
1461
1462 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1463                         int len)
1464 {
1465         int r;
1466         unsigned long addr;
1467
1468         addr = gfn_to_hva_read(kvm, gfn);
1469         if (kvm_is_error_hva(addr))
1470                 return -EFAULT;
1471         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1472         if (r)
1473                 return -EFAULT;
1474         return 0;
1475 }
1476 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1477
1478 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1479 {
1480         gfn_t gfn = gpa >> PAGE_SHIFT;
1481         int seg;
1482         int offset = offset_in_page(gpa);
1483         int ret;
1484
1485         while ((seg = next_segment(len, offset)) != 0) {
1486                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1487                 if (ret < 0)
1488                         return ret;
1489                 offset = 0;
1490                 len -= seg;
1491                 data += seg;
1492                 ++gfn;
1493         }
1494         return 0;
1495 }
1496 EXPORT_SYMBOL_GPL(kvm_read_guest);
1497
1498 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1499                           unsigned long len)
1500 {
1501         int r;
1502         unsigned long addr;
1503         gfn_t gfn = gpa >> PAGE_SHIFT;
1504         int offset = offset_in_page(gpa);
1505
1506         addr = gfn_to_hva_read(kvm, gfn);
1507         if (kvm_is_error_hva(addr))
1508                 return -EFAULT;
1509         pagefault_disable();
1510         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1511         pagefault_enable();
1512         if (r)
1513                 return -EFAULT;
1514         return 0;
1515 }
1516 EXPORT_SYMBOL(kvm_read_guest_atomic);
1517
1518 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1519                          int offset, int len)
1520 {
1521         int r;
1522         unsigned long addr;
1523
1524         addr = gfn_to_hva(kvm, gfn);
1525         if (kvm_is_error_hva(addr))
1526                 return -EFAULT;
1527         r = __copy_to_user((void __user *)addr + offset, data, len);
1528         if (r)
1529                 return -EFAULT;
1530         mark_page_dirty(kvm, gfn);
1531         return 0;
1532 }
1533 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1534
1535 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1536                     unsigned long len)
1537 {
1538         gfn_t gfn = gpa >> PAGE_SHIFT;
1539         int seg;
1540         int offset = offset_in_page(gpa);
1541         int ret;
1542
1543         while ((seg = next_segment(len, offset)) != 0) {
1544                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1545                 if (ret < 0)
1546                         return ret;
1547                 offset = 0;
1548                 len -= seg;
1549                 data += seg;
1550                 ++gfn;
1551         }
1552         return 0;
1553 }
1554
1555 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1556                               gpa_t gpa)
1557 {
1558         struct kvm_memslots *slots = kvm_memslots(kvm);
1559         int offset = offset_in_page(gpa);
1560         gfn_t gfn = gpa >> PAGE_SHIFT;
1561
1562         ghc->gpa = gpa;
1563         ghc->generation = slots->generation;
1564         ghc->memslot = gfn_to_memslot(kvm, gfn);
1565         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1566         if (!kvm_is_error_hva(ghc->hva))
1567                 ghc->hva += offset;
1568         else
1569                 return -EFAULT;
1570
1571         return 0;
1572 }
1573 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1574
1575 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1576                            void *data, unsigned long len)
1577 {
1578         struct kvm_memslots *slots = kvm_memslots(kvm);
1579         int r;
1580
1581         if (slots->generation != ghc->generation)
1582                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1583
1584         if (kvm_is_error_hva(ghc->hva))
1585                 return -EFAULT;
1586
1587         r = __copy_to_user((void __user *)ghc->hva, data, len);
1588         if (r)
1589                 return -EFAULT;
1590         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1591
1592         return 0;
1593 }
1594 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1595
1596 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1597                            void *data, unsigned long len)
1598 {
1599         struct kvm_memslots *slots = kvm_memslots(kvm);
1600         int r;
1601
1602         if (slots->generation != ghc->generation)
1603                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1604
1605         if (kvm_is_error_hva(ghc->hva))
1606                 return -EFAULT;
1607
1608         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1609         if (r)
1610                 return -EFAULT;
1611
1612         return 0;
1613 }
1614 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1615
1616 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1617 {
1618         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1619                                     offset, len);
1620 }
1621 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1622
1623 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1624 {
1625         gfn_t gfn = gpa >> PAGE_SHIFT;
1626         int seg;
1627         int offset = offset_in_page(gpa);
1628         int ret;
1629
1630         while ((seg = next_segment(len, offset)) != 0) {
1631                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1632                 if (ret < 0)
1633                         return ret;
1634                 offset = 0;
1635                 len -= seg;
1636                 ++gfn;
1637         }
1638         return 0;
1639 }
1640 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1641
1642 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1643                              gfn_t gfn)
1644 {
1645         if (memslot && memslot->dirty_bitmap) {
1646                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1647
1648                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1649         }
1650 }
1651
1652 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1653 {
1654         struct kvm_memory_slot *memslot;
1655
1656         memslot = gfn_to_memslot(kvm, gfn);
1657         mark_page_dirty_in_slot(kvm, memslot, gfn);
1658 }
1659
1660 /*
1661  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1662  */
1663 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1664 {
1665         DEFINE_WAIT(wait);
1666
1667         for (;;) {
1668                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1669
1670                 if (kvm_arch_vcpu_runnable(vcpu)) {
1671                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1672                         break;
1673                 }
1674                 if (kvm_cpu_has_pending_timer(vcpu))
1675                         break;
1676                 if (signal_pending(current))
1677                         break;
1678
1679                 schedule();
1680         }
1681
1682         finish_wait(&vcpu->wq, &wait);
1683 }
1684
1685 #ifndef CONFIG_S390
1686 /*
1687  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1688  */
1689 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1690 {
1691         int me;
1692         int cpu = vcpu->cpu;
1693         wait_queue_head_t *wqp;
1694
1695         wqp = kvm_arch_vcpu_wq(vcpu);
1696         if (waitqueue_active(wqp)) {
1697                 wake_up_interruptible(wqp);
1698                 ++vcpu->stat.halt_wakeup;
1699         }
1700
1701         me = get_cpu();
1702         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1703                 if (kvm_arch_vcpu_should_kick(vcpu))
1704                         smp_send_reschedule(cpu);
1705         put_cpu();
1706 }
1707 #endif /* !CONFIG_S390 */
1708
1709 void kvm_resched(struct kvm_vcpu *vcpu)
1710 {
1711         if (!need_resched())
1712                 return;
1713         cond_resched();
1714 }
1715 EXPORT_SYMBOL_GPL(kvm_resched);
1716
1717 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1718 {
1719         struct pid *pid;
1720         struct task_struct *task = NULL;
1721         bool ret = false;
1722
1723         rcu_read_lock();
1724         pid = rcu_dereference(target->pid);
1725         if (pid)
1726                 task = get_pid_task(target->pid, PIDTYPE_PID);
1727         rcu_read_unlock();
1728         if (!task)
1729                 return ret;
1730         if (task->flags & PF_VCPU) {
1731                 put_task_struct(task);
1732                 return ret;
1733         }
1734         ret = yield_to(task, 1);
1735         put_task_struct(task);
1736
1737         return ret;
1738 }
1739 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1740
1741 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1742 /*
1743  * Helper that checks whether a VCPU is eligible for directed yield.
1744  * Most eligible candidate to yield is decided by following heuristics:
1745  *
1746  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1747  *  (preempted lock holder), indicated by @in_spin_loop.
1748  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1749  *
1750  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1751  *  chance last time (mostly it has become eligible now since we have probably
1752  *  yielded to lockholder in last iteration. This is done by toggling
1753  *  @dy_eligible each time a VCPU checked for eligibility.)
1754  *
1755  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1756  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1757  *  burning. Giving priority for a potential lock-holder increases lock
1758  *  progress.
1759  *
1760  *  Since algorithm is based on heuristics, accessing another VCPU data without
1761  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1762  *  and continue with next VCPU and so on.
1763  */
1764 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1765 {
1766         bool eligible;
1767
1768         eligible = !vcpu->spin_loop.in_spin_loop ||
1769                         (vcpu->spin_loop.in_spin_loop &&
1770                          vcpu->spin_loop.dy_eligible);
1771
1772         if (vcpu->spin_loop.in_spin_loop)
1773                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1774
1775         return eligible;
1776 }
1777 #endif
1778
1779 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1780 {
1781         struct kvm *kvm = me->kvm;
1782         struct kvm_vcpu *vcpu;
1783         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1784         int yielded = 0;
1785         int try = 3;
1786         int pass;
1787         int i;
1788
1789         kvm_vcpu_set_in_spin_loop(me, true);
1790         /*
1791          * We boost the priority of a VCPU that is runnable but not
1792          * currently running, because it got preempted by something
1793          * else and called schedule in __vcpu_run.  Hopefully that
1794          * VCPU is holding the lock that we need and will release it.
1795          * We approximate round-robin by starting at the last boosted VCPU.
1796          */
1797         for (pass = 0; pass < 2 && !yielded && try; pass++) {
1798                 kvm_for_each_vcpu(i, vcpu, kvm) {
1799                         if (!pass && i <= last_boosted_vcpu) {
1800                                 i = last_boosted_vcpu;
1801                                 continue;
1802                         } else if (pass && i > last_boosted_vcpu)
1803                                 break;
1804                         if (vcpu == me)
1805                                 continue;
1806                         if (waitqueue_active(&vcpu->wq))
1807                                 continue;
1808                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1809                                 continue;
1810
1811                         yielded = kvm_vcpu_yield_to(vcpu);
1812                         if (yielded > 0) {
1813                                 kvm->last_boosted_vcpu = i;
1814                                 break;
1815                         } else if (yielded < 0) {
1816                                 try--;
1817                                 if (!try)
1818                                         break;
1819                         }
1820                 }
1821         }
1822         kvm_vcpu_set_in_spin_loop(me, false);
1823
1824         /* Ensure vcpu is not eligible during next spinloop */
1825         kvm_vcpu_set_dy_eligible(me, false);
1826 }
1827 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1828
1829 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1830 {
1831         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1832         struct page *page;
1833
1834         if (vmf->pgoff == 0)
1835                 page = virt_to_page(vcpu->run);
1836 #ifdef CONFIG_X86
1837         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1838                 page = virt_to_page(vcpu->arch.pio_data);
1839 #endif
1840 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1841         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1842                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1843 #endif
1844         else
1845                 return kvm_arch_vcpu_fault(vcpu, vmf);
1846         get_page(page);
1847         vmf->page = page;
1848         return 0;
1849 }
1850
1851 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1852         .fault = kvm_vcpu_fault,
1853 };
1854
1855 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1856 {
1857         vma->vm_ops = &kvm_vcpu_vm_ops;
1858         return 0;
1859 }
1860
1861 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1862 {
1863         struct kvm_vcpu *vcpu = filp->private_data;
1864
1865         kvm_put_kvm(vcpu->kvm);
1866         return 0;
1867 }
1868
1869 static struct file_operations kvm_vcpu_fops = {
1870         .release        = kvm_vcpu_release,
1871         .unlocked_ioctl = kvm_vcpu_ioctl,
1872 #ifdef CONFIG_COMPAT
1873         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1874 #endif
1875         .mmap           = kvm_vcpu_mmap,
1876         .llseek         = noop_llseek,
1877 };
1878
1879 /*
1880  * Allocates an inode for the vcpu.
1881  */
1882 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1883 {
1884         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1885 }
1886
1887 /*
1888  * Creates some virtual cpus.  Good luck creating more than one.
1889  */
1890 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1891 {
1892         int r;
1893         struct kvm_vcpu *vcpu, *v;
1894
1895         vcpu = kvm_arch_vcpu_create(kvm, id);
1896         if (IS_ERR(vcpu))
1897                 return PTR_ERR(vcpu);
1898
1899         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1900
1901         r = kvm_arch_vcpu_setup(vcpu);
1902         if (r)
1903                 goto vcpu_destroy;
1904
1905         mutex_lock(&kvm->lock);
1906         if (!kvm_vcpu_compatible(vcpu)) {
1907                 r = -EINVAL;
1908                 goto unlock_vcpu_destroy;
1909         }
1910         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1911                 r = -EINVAL;
1912                 goto unlock_vcpu_destroy;
1913         }
1914
1915         kvm_for_each_vcpu(r, v, kvm)
1916                 if (v->vcpu_id == id) {
1917                         r = -EEXIST;
1918                         goto unlock_vcpu_destroy;
1919                 }
1920
1921         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1922
1923         /* Now it's all set up, let userspace reach it */
1924         kvm_get_kvm(kvm);
1925         r = create_vcpu_fd(vcpu);
1926         if (r < 0) {
1927                 kvm_put_kvm(kvm);
1928                 goto unlock_vcpu_destroy;
1929         }
1930
1931         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1932         smp_wmb();
1933         atomic_inc(&kvm->online_vcpus);
1934
1935         mutex_unlock(&kvm->lock);
1936         kvm_arch_vcpu_postcreate(vcpu);
1937         return r;
1938
1939 unlock_vcpu_destroy:
1940         mutex_unlock(&kvm->lock);
1941 vcpu_destroy:
1942         kvm_arch_vcpu_destroy(vcpu);
1943         return r;
1944 }
1945
1946 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1947 {
1948         if (sigset) {
1949                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1950                 vcpu->sigset_active = 1;
1951                 vcpu->sigset = *sigset;
1952         } else
1953                 vcpu->sigset_active = 0;
1954         return 0;
1955 }
1956
1957 static long kvm_vcpu_ioctl(struct file *filp,
1958                            unsigned int ioctl, unsigned long arg)
1959 {
1960         struct kvm_vcpu *vcpu = filp->private_data;
1961         void __user *argp = (void __user *)arg;
1962         int r;
1963         struct kvm_fpu *fpu = NULL;
1964         struct kvm_sregs *kvm_sregs = NULL;
1965
1966         if (vcpu->kvm->mm != current->mm)
1967                 return -EIO;
1968
1969 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1970         /*
1971          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1972          * so vcpu_load() would break it.
1973          */
1974         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1975                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1976 #endif
1977
1978
1979         r = vcpu_load(vcpu);
1980         if (r)
1981                 return r;
1982         switch (ioctl) {
1983         case KVM_RUN:
1984                 r = -EINVAL;
1985                 if (arg)
1986                         goto out;
1987                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1988                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1989                 break;
1990         case KVM_GET_REGS: {
1991                 struct kvm_regs *kvm_regs;
1992
1993                 r = -ENOMEM;
1994                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1995                 if (!kvm_regs)
1996                         goto out;
1997                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1998                 if (r)
1999                         goto out_free1;
2000                 r = -EFAULT;
2001                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2002                         goto out_free1;
2003                 r = 0;
2004 out_free1:
2005                 kfree(kvm_regs);
2006                 break;
2007         }
2008         case KVM_SET_REGS: {
2009                 struct kvm_regs *kvm_regs;
2010
2011                 r = -ENOMEM;
2012                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2013                 if (IS_ERR(kvm_regs)) {
2014                         r = PTR_ERR(kvm_regs);
2015                         goto out;
2016                 }
2017                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2018                 kfree(kvm_regs);
2019                 break;
2020         }
2021         case KVM_GET_SREGS: {
2022                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2023                 r = -ENOMEM;
2024                 if (!kvm_sregs)
2025                         goto out;
2026                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2027                 if (r)
2028                         goto out;
2029                 r = -EFAULT;
2030                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2031                         goto out;
2032                 r = 0;
2033                 break;
2034         }
2035         case KVM_SET_SREGS: {
2036                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2037                 if (IS_ERR(kvm_sregs)) {
2038                         r = PTR_ERR(kvm_sregs);
2039                         kvm_sregs = NULL;
2040                         goto out;
2041                 }
2042                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2043                 break;
2044         }
2045         case KVM_GET_MP_STATE: {
2046                 struct kvm_mp_state mp_state;
2047
2048                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2049                 if (r)
2050                         goto out;
2051                 r = -EFAULT;
2052                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
2053                         goto out;
2054                 r = 0;
2055                 break;
2056         }
2057         case KVM_SET_MP_STATE: {
2058                 struct kvm_mp_state mp_state;
2059
2060                 r = -EFAULT;
2061                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
2062                         goto out;
2063                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2064                 break;
2065         }
2066         case KVM_TRANSLATE: {
2067                 struct kvm_translation tr;
2068
2069                 r = -EFAULT;
2070                 if (copy_from_user(&tr, argp, sizeof tr))
2071                         goto out;
2072                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2073                 if (r)
2074                         goto out;
2075                 r = -EFAULT;
2076                 if (copy_to_user(argp, &tr, sizeof tr))
2077                         goto out;
2078                 r = 0;
2079                 break;
2080         }
2081         case KVM_SET_GUEST_DEBUG: {
2082                 struct kvm_guest_debug dbg;
2083
2084                 r = -EFAULT;
2085                 if (copy_from_user(&dbg, argp, sizeof dbg))
2086                         goto out;
2087                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2088                 break;
2089         }
2090         case KVM_SET_SIGNAL_MASK: {
2091                 struct kvm_signal_mask __user *sigmask_arg = argp;
2092                 struct kvm_signal_mask kvm_sigmask;
2093                 sigset_t sigset, *p;
2094
2095                 p = NULL;
2096                 if (argp) {
2097                         r = -EFAULT;
2098                         if (copy_from_user(&kvm_sigmask, argp,
2099                                            sizeof kvm_sigmask))
2100                                 goto out;
2101                         r = -EINVAL;
2102                         if (kvm_sigmask.len != sizeof sigset)
2103                                 goto out;
2104                         r = -EFAULT;
2105                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2106                                            sizeof sigset))
2107                                 goto out;
2108                         p = &sigset;
2109                 }
2110                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2111                 break;
2112         }
2113         case KVM_GET_FPU: {
2114                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2115                 r = -ENOMEM;
2116                 if (!fpu)
2117                         goto out;
2118                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2119                 if (r)
2120                         goto out;
2121                 r = -EFAULT;
2122                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2123                         goto out;
2124                 r = 0;
2125                 break;
2126         }
2127         case KVM_SET_FPU: {
2128                 fpu = memdup_user(argp, sizeof(*fpu));
2129                 if (IS_ERR(fpu)) {
2130                         r = PTR_ERR(fpu);
2131                         fpu = NULL;
2132                         goto out;
2133                 }
2134                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2135                 break;
2136         }
2137         default:
2138                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2139         }
2140 out:
2141         vcpu_put(vcpu);
2142         kfree(fpu);
2143         kfree(kvm_sregs);
2144         return r;
2145 }
2146
2147 #ifdef CONFIG_COMPAT
2148 static long kvm_vcpu_compat_ioctl(struct file *filp,
2149                                   unsigned int ioctl, unsigned long arg)
2150 {
2151         struct kvm_vcpu *vcpu = filp->private_data;
2152         void __user *argp = compat_ptr(arg);
2153         int r;
2154
2155         if (vcpu->kvm->mm != current->mm)
2156                 return -EIO;
2157
2158         switch (ioctl) {
2159         case KVM_SET_SIGNAL_MASK: {
2160                 struct kvm_signal_mask __user *sigmask_arg = argp;
2161                 struct kvm_signal_mask kvm_sigmask;
2162                 compat_sigset_t csigset;
2163                 sigset_t sigset;
2164
2165                 if (argp) {
2166                         r = -EFAULT;
2167                         if (copy_from_user(&kvm_sigmask, argp,
2168                                            sizeof kvm_sigmask))
2169                                 goto out;
2170                         r = -EINVAL;
2171                         if (kvm_sigmask.len != sizeof csigset)
2172                                 goto out;
2173                         r = -EFAULT;
2174                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2175                                            sizeof csigset))
2176                                 goto out;
2177                         sigset_from_compat(&sigset, &csigset);
2178                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2179                 } else
2180                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2181                 break;
2182         }
2183         default:
2184                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2185         }
2186
2187 out:
2188         return r;
2189 }
2190 #endif
2191
2192 static long kvm_vm_ioctl(struct file *filp,
2193                            unsigned int ioctl, unsigned long arg)
2194 {
2195         struct kvm *kvm = filp->private_data;
2196         void __user *argp = (void __user *)arg;
2197         int r;
2198
2199         if (kvm->mm != current->mm)
2200                 return -EIO;
2201         switch (ioctl) {
2202         case KVM_CREATE_VCPU:
2203                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2204                 break;
2205         case KVM_SET_USER_MEMORY_REGION: {
2206                 struct kvm_userspace_memory_region kvm_userspace_mem;
2207
2208                 r = -EFAULT;
2209                 if (copy_from_user(&kvm_userspace_mem, argp,
2210                                                 sizeof kvm_userspace_mem))
2211                         goto out;
2212
2213                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, true);
2214                 break;
2215         }
2216         case KVM_GET_DIRTY_LOG: {
2217                 struct kvm_dirty_log log;
2218
2219                 r = -EFAULT;
2220                 if (copy_from_user(&log, argp, sizeof log))
2221                         goto out;
2222                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2223                 break;
2224         }
2225 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2226         case KVM_REGISTER_COALESCED_MMIO: {
2227                 struct kvm_coalesced_mmio_zone zone;
2228                 r = -EFAULT;
2229                 if (copy_from_user(&zone, argp, sizeof zone))
2230                         goto out;
2231                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2232                 break;
2233         }
2234         case KVM_UNREGISTER_COALESCED_MMIO: {
2235                 struct kvm_coalesced_mmio_zone zone;
2236                 r = -EFAULT;
2237                 if (copy_from_user(&zone, argp, sizeof zone))
2238                         goto out;
2239                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2240                 break;
2241         }
2242 #endif
2243         case KVM_IRQFD: {
2244                 struct kvm_irqfd data;
2245
2246                 r = -EFAULT;
2247                 if (copy_from_user(&data, argp, sizeof data))
2248                         goto out;
2249                 r = kvm_irqfd(kvm, &data);
2250                 break;
2251         }
2252         case KVM_IOEVENTFD: {
2253                 struct kvm_ioeventfd data;
2254
2255                 r = -EFAULT;
2256                 if (copy_from_user(&data, argp, sizeof data))
2257                         goto out;
2258                 r = kvm_ioeventfd(kvm, &data);
2259                 break;
2260         }
2261 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2262         case KVM_SET_BOOT_CPU_ID:
2263                 r = 0;
2264                 mutex_lock(&kvm->lock);
2265                 if (atomic_read(&kvm->online_vcpus) != 0)
2266                         r = -EBUSY;
2267                 else
2268                         kvm->bsp_vcpu_id = arg;
2269                 mutex_unlock(&kvm->lock);
2270                 break;
2271 #endif
2272 #ifdef CONFIG_HAVE_KVM_MSI
2273         case KVM_SIGNAL_MSI: {
2274                 struct kvm_msi msi;
2275
2276                 r = -EFAULT;
2277                 if (copy_from_user(&msi, argp, sizeof msi))
2278                         goto out;
2279                 r = kvm_send_userspace_msi(kvm, &msi);
2280                 break;
2281         }
2282 #endif
2283 #ifdef __KVM_HAVE_IRQ_LINE
2284         case KVM_IRQ_LINE_STATUS:
2285         case KVM_IRQ_LINE: {
2286                 struct kvm_irq_level irq_event;
2287
2288                 r = -EFAULT;
2289                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2290                         goto out;
2291
2292                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2293                 if (r)
2294                         goto out;
2295
2296                 r = -EFAULT;
2297                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2298                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2299                                 goto out;
2300                 }
2301
2302                 r = 0;
2303                 break;
2304         }
2305 #endif
2306         default:
2307                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2308                 if (r == -ENOTTY)
2309                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2310         }
2311 out:
2312         return r;
2313 }
2314
2315 #ifdef CONFIG_COMPAT
2316 struct compat_kvm_dirty_log {
2317         __u32 slot;
2318         __u32 padding1;
2319         union {
2320                 compat_uptr_t dirty_bitmap; /* one bit per page */
2321                 __u64 padding2;
2322         };
2323 };
2324
2325 static long kvm_vm_compat_ioctl(struct file *filp,
2326                            unsigned int ioctl, unsigned long arg)
2327 {
2328         struct kvm *kvm = filp->private_data;
2329         int r;
2330
2331         if (kvm->mm != current->mm)
2332                 return -EIO;
2333         switch (ioctl) {
2334         case KVM_GET_DIRTY_LOG: {
2335                 struct compat_kvm_dirty_log compat_log;
2336                 struct kvm_dirty_log log;
2337
2338                 r = -EFAULT;
2339                 if (copy_from_user(&compat_log, (void __user *)arg,
2340                                    sizeof(compat_log)))
2341                         goto out;
2342                 log.slot         = compat_log.slot;
2343                 log.padding1     = compat_log.padding1;
2344                 log.padding2     = compat_log.padding2;
2345                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2346
2347                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2348                 break;
2349         }
2350         default:
2351                 r = kvm_vm_ioctl(filp, ioctl, arg);
2352         }
2353
2354 out:
2355         return r;
2356 }
2357 #endif
2358
2359 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2360 {
2361         struct page *page[1];
2362         unsigned long addr;
2363         int npages;
2364         gfn_t gfn = vmf->pgoff;
2365         struct kvm *kvm = vma->vm_file->private_data;
2366
2367         addr = gfn_to_hva(kvm, gfn);
2368         if (kvm_is_error_hva(addr))
2369                 return VM_FAULT_SIGBUS;
2370
2371         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2372                                 NULL);
2373         if (unlikely(npages != 1))
2374                 return VM_FAULT_SIGBUS;
2375
2376         vmf->page = page[0];
2377         return 0;
2378 }
2379
2380 static const struct vm_operations_struct kvm_vm_vm_ops = {
2381         .fault = kvm_vm_fault,
2382 };
2383
2384 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2385 {
2386         vma->vm_ops = &kvm_vm_vm_ops;
2387         return 0;
2388 }
2389
2390 static struct file_operations kvm_vm_fops = {
2391         .release        = kvm_vm_release,
2392         .unlocked_ioctl = kvm_vm_ioctl,
2393 #ifdef CONFIG_COMPAT
2394         .compat_ioctl   = kvm_vm_compat_ioctl,
2395 #endif
2396         .mmap           = kvm_vm_mmap,
2397         .llseek         = noop_llseek,
2398 };
2399
2400 static int kvm_dev_ioctl_create_vm(unsigned long type)
2401 {
2402         int r;
2403         struct kvm *kvm;
2404
2405         kvm = kvm_create_vm(type);
2406         if (IS_ERR(kvm))
2407                 return PTR_ERR(kvm);
2408 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2409         r = kvm_coalesced_mmio_init(kvm);
2410         if (r < 0) {
2411                 kvm_put_kvm(kvm);
2412                 return r;
2413         }
2414 #endif
2415         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2416         if (r < 0)
2417                 kvm_put_kvm(kvm);
2418
2419         return r;
2420 }
2421
2422 static long kvm_dev_ioctl_check_extension_generic(long arg)
2423 {
2424         switch (arg) {
2425         case KVM_CAP_USER_MEMORY:
2426         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2427         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2428 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2429         case KVM_CAP_SET_BOOT_CPU_ID:
2430 #endif
2431         case KVM_CAP_INTERNAL_ERROR_DATA:
2432 #ifdef CONFIG_HAVE_KVM_MSI
2433         case KVM_CAP_SIGNAL_MSI:
2434 #endif
2435                 return 1;
2436 #ifdef KVM_CAP_IRQ_ROUTING
2437         case KVM_CAP_IRQ_ROUTING:
2438                 return KVM_MAX_IRQ_ROUTES;
2439 #endif
2440         default:
2441                 break;
2442         }
2443         return kvm_dev_ioctl_check_extension(arg);
2444 }
2445
2446 static long kvm_dev_ioctl(struct file *filp,
2447                           unsigned int ioctl, unsigned long arg)
2448 {
2449         long r = -EINVAL;
2450
2451         switch (ioctl) {
2452         case KVM_GET_API_VERSION:
2453                 r = -EINVAL;
2454                 if (arg)
2455                         goto out;
2456                 r = KVM_API_VERSION;
2457                 break;
2458         case KVM_CREATE_VM:
2459                 r = kvm_dev_ioctl_create_vm(arg);
2460                 break;
2461         case KVM_CHECK_EXTENSION:
2462                 r = kvm_dev_ioctl_check_extension_generic(arg);
2463                 break;
2464         case KVM_GET_VCPU_MMAP_SIZE:
2465                 r = -EINVAL;
2466                 if (arg)
2467                         goto out;
2468                 r = PAGE_SIZE;     /* struct kvm_run */
2469 #ifdef CONFIG_X86
2470                 r += PAGE_SIZE;    /* pio data page */
2471 #endif
2472 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2473                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2474 #endif
2475                 break;
2476         case KVM_TRACE_ENABLE:
2477         case KVM_TRACE_PAUSE:
2478         case KVM_TRACE_DISABLE:
2479                 r = -EOPNOTSUPP;
2480                 break;
2481         default:
2482                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2483         }
2484 out:
2485         return r;
2486 }
2487
2488 static struct file_operations kvm_chardev_ops = {
2489         .unlocked_ioctl = kvm_dev_ioctl,
2490         .compat_ioctl   = kvm_dev_ioctl,
2491         .llseek         = noop_llseek,
2492 };
2493
2494 static struct miscdevice kvm_dev = {
2495         KVM_MINOR,
2496         "kvm",
2497         &kvm_chardev_ops,
2498 };
2499
2500 static void hardware_enable_nolock(void *junk)
2501 {
2502         int cpu = raw_smp_processor_id();
2503         int r;
2504
2505         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2506                 return;
2507
2508         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2509
2510         r = kvm_arch_hardware_enable(NULL);
2511
2512         if (r) {
2513                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2514                 atomic_inc(&hardware_enable_failed);
2515                 printk(KERN_INFO "kvm: enabling virtualization on "
2516                                  "CPU%d failed\n", cpu);
2517         }
2518 }
2519
2520 static void hardware_enable(void *junk)
2521 {
2522         raw_spin_lock(&kvm_lock);
2523         hardware_enable_nolock(junk);
2524         raw_spin_unlock(&kvm_lock);
2525 }
2526
2527 static void hardware_disable_nolock(void *junk)
2528 {
2529         int cpu = raw_smp_processor_id();
2530
2531         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2532                 return;
2533         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2534         kvm_arch_hardware_disable(NULL);
2535 }
2536
2537 static void hardware_disable(void *junk)
2538 {
2539         raw_spin_lock(&kvm_lock);
2540         hardware_disable_nolock(junk);
2541         raw_spin_unlock(&kvm_lock);
2542 }
2543
2544 static void hardware_disable_all_nolock(void)
2545 {
2546         BUG_ON(!kvm_usage_count);
2547
2548         kvm_usage_count--;
2549         if (!kvm_usage_count)
2550                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2551 }
2552
2553 static void hardware_disable_all(void)
2554 {
2555         raw_spin_lock(&kvm_lock);
2556         hardware_disable_all_nolock();
2557         raw_spin_unlock(&kvm_lock);
2558 }
2559
2560 static int hardware_enable_all(void)
2561 {
2562         int r = 0;
2563
2564         raw_spin_lock(&kvm_lock);
2565
2566         kvm_usage_count++;
2567         if (kvm_usage_count == 1) {
2568                 atomic_set(&hardware_enable_failed, 0);
2569                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2570
2571                 if (atomic_read(&hardware_enable_failed)) {
2572                         hardware_disable_all_nolock();
2573                         r = -EBUSY;
2574                 }
2575         }
2576
2577         raw_spin_unlock(&kvm_lock);
2578
2579         return r;
2580 }
2581
2582 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2583                            void *v)
2584 {
2585         int cpu = (long)v;
2586
2587         if (!kvm_usage_count)
2588                 return NOTIFY_OK;
2589
2590         val &= ~CPU_TASKS_FROZEN;
2591         switch (val) {
2592         case CPU_DYING:
2593                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2594                        cpu);
2595                 hardware_disable(NULL);
2596                 break;
2597         case CPU_STARTING:
2598                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2599                        cpu);
2600                 hardware_enable(NULL);
2601                 break;
2602         }
2603         return NOTIFY_OK;
2604 }
2605
2606
2607 asmlinkage void kvm_spurious_fault(void)
2608 {
2609         /* Fault while not rebooting.  We want the trace. */
2610         BUG();
2611 }
2612 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2613
2614 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2615                       void *v)
2616 {
2617         /*
2618          * Some (well, at least mine) BIOSes hang on reboot if
2619          * in vmx root mode.
2620          *
2621          * And Intel TXT required VMX off for all cpu when system shutdown.
2622          */
2623         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2624         kvm_rebooting = true;
2625         on_each_cpu(hardware_disable_nolock, NULL, 1);
2626         return NOTIFY_OK;
2627 }
2628
2629 static struct notifier_block kvm_reboot_notifier = {
2630         .notifier_call = kvm_reboot,
2631         .priority = 0,
2632 };
2633
2634 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2635 {
2636         int i;
2637
2638         for (i = 0; i < bus->dev_count; i++) {
2639                 struct kvm_io_device *pos = bus->range[i].dev;
2640
2641                 kvm_iodevice_destructor(pos);
2642         }
2643         kfree(bus);
2644 }
2645
2646 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2647 {
2648         const struct kvm_io_range *r1 = p1;
2649         const struct kvm_io_range *r2 = p2;
2650
2651         if (r1->addr < r2->addr)
2652                 return -1;
2653         if (r1->addr + r1->len > r2->addr + r2->len)
2654                 return 1;
2655         return 0;
2656 }
2657
2658 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2659                           gpa_t addr, int len)
2660 {
2661         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2662                 .addr = addr,
2663                 .len = len,
2664                 .dev = dev,
2665         };
2666
2667         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2668                 kvm_io_bus_sort_cmp, NULL);
2669
2670         return 0;
2671 }
2672
2673 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2674                              gpa_t addr, int len)
2675 {
2676         struct kvm_io_range *range, key;
2677         int off;
2678
2679         key = (struct kvm_io_range) {
2680                 .addr = addr,
2681                 .len = len,
2682         };
2683
2684         range = bsearch(&key, bus->range, bus->dev_count,
2685                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2686         if (range == NULL)
2687                 return -ENOENT;
2688
2689         off = range - bus->range;
2690
2691         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2692                 off--;
2693
2694         return off;
2695 }
2696
2697 /* kvm_io_bus_write - called under kvm->slots_lock */
2698 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2699                      int len, const void *val)
2700 {
2701         int idx;
2702         struct kvm_io_bus *bus;
2703         struct kvm_io_range range;
2704
2705         range = (struct kvm_io_range) {
2706                 .addr = addr,
2707                 .len = len,
2708         };
2709
2710         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2711         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2712         if (idx < 0)
2713                 return -EOPNOTSUPP;
2714
2715         while (idx < bus->dev_count &&
2716                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2717                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2718                         return 0;
2719                 idx++;
2720         }
2721
2722         return -EOPNOTSUPP;
2723 }
2724
2725 /* kvm_io_bus_read - called under kvm->slots_lock */
2726 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2727                     int len, void *val)
2728 {
2729         int idx;
2730         struct kvm_io_bus *bus;
2731         struct kvm_io_range range;
2732
2733         range = (struct kvm_io_range) {
2734                 .addr = addr,
2735                 .len = len,
2736         };
2737
2738         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2739         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2740         if (idx < 0)
2741                 return -EOPNOTSUPP;
2742
2743         while (idx < bus->dev_count &&
2744                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2745                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2746                         return 0;
2747                 idx++;
2748         }
2749
2750         return -EOPNOTSUPP;
2751 }
2752
2753 /* Caller must hold slots_lock. */
2754 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2755                             int len, struct kvm_io_device *dev)
2756 {
2757         struct kvm_io_bus *new_bus, *bus;
2758
2759         bus = kvm->buses[bus_idx];
2760         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2761                 return -ENOSPC;
2762
2763         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2764                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2765         if (!new_bus)
2766                 return -ENOMEM;
2767         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2768                sizeof(struct kvm_io_range)));
2769         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2770         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2771         synchronize_srcu_expedited(&kvm->srcu);
2772         kfree(bus);
2773
2774         return 0;
2775 }
2776
2777 /* Caller must hold slots_lock. */
2778 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2779                               struct kvm_io_device *dev)
2780 {
2781         int i, r;
2782         struct kvm_io_bus *new_bus, *bus;
2783
2784         bus = kvm->buses[bus_idx];
2785         r = -ENOENT;
2786         for (i = 0; i < bus->dev_count; i++)
2787                 if (bus->range[i].dev == dev) {
2788                         r = 0;
2789                         break;
2790                 }
2791
2792         if (r)
2793                 return r;
2794
2795         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2796                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2797         if (!new_bus)
2798                 return -ENOMEM;
2799
2800         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2801         new_bus->dev_count--;
2802         memcpy(new_bus->range + i, bus->range + i + 1,
2803                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2804
2805         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2806         synchronize_srcu_expedited(&kvm->srcu);
2807         kfree(bus);
2808         return r;
2809 }
2810
2811 static struct notifier_block kvm_cpu_notifier = {
2812         .notifier_call = kvm_cpu_hotplug,
2813 };
2814
2815 static int vm_stat_get(void *_offset, u64 *val)
2816 {
2817         unsigned offset = (long)_offset;
2818         struct kvm *kvm;
2819
2820         *val = 0;
2821         raw_spin_lock(&kvm_lock);
2822         list_for_each_entry(kvm, &vm_list, vm_list)
2823                 *val += *(u32 *)((void *)kvm + offset);
2824         raw_spin_unlock(&kvm_lock);
2825         return 0;
2826 }
2827
2828 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2829
2830 static int vcpu_stat_get(void *_offset, u64 *val)
2831 {
2832         unsigned offset = (long)_offset;
2833         struct kvm *kvm;
2834         struct kvm_vcpu *vcpu;
2835         int i;
2836
2837         *val = 0;
2838         raw_spin_lock(&kvm_lock);
2839         list_for_each_entry(kvm, &vm_list, vm_list)
2840                 kvm_for_each_vcpu(i, vcpu, kvm)
2841                         *val += *(u32 *)((void *)vcpu + offset);
2842
2843         raw_spin_unlock(&kvm_lock);
2844         return 0;
2845 }
2846
2847 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2848
2849 static const struct file_operations *stat_fops[] = {
2850         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2851         [KVM_STAT_VM]   = &vm_stat_fops,
2852 };
2853
2854 static int kvm_init_debug(void)
2855 {
2856         int r = -EFAULT;
2857         struct kvm_stats_debugfs_item *p;
2858
2859         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2860         if (kvm_debugfs_dir == NULL)
2861                 goto out;
2862
2863         for (p = debugfs_entries; p->name; ++p) {
2864                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2865                                                 (void *)(long)p->offset,
2866                                                 stat_fops[p->kind]);
2867                 if (p->dentry == NULL)
2868                         goto out_dir;
2869         }
2870
2871         return 0;
2872
2873 out_dir:
2874         debugfs_remove_recursive(kvm_debugfs_dir);
2875 out:
2876         return r;
2877 }
2878
2879 static void kvm_exit_debug(void)
2880 {
2881         struct kvm_stats_debugfs_item *p;
2882
2883         for (p = debugfs_entries; p->name; ++p)
2884                 debugfs_remove(p->dentry);
2885         debugfs_remove(kvm_debugfs_dir);
2886 }
2887
2888 static int kvm_suspend(void)
2889 {
2890         if (kvm_usage_count)
2891                 hardware_disable_nolock(NULL);
2892         return 0;
2893 }
2894
2895 static void kvm_resume(void)
2896 {
2897         if (kvm_usage_count) {
2898                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2899                 hardware_enable_nolock(NULL);
2900         }
2901 }
2902
2903 static struct syscore_ops kvm_syscore_ops = {
2904         .suspend = kvm_suspend,
2905         .resume = kvm_resume,
2906 };
2907
2908 static inline
2909 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2910 {
2911         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2912 }
2913
2914 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2915 {
2916         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2917
2918         kvm_arch_vcpu_load(vcpu, cpu);
2919 }
2920
2921 static void kvm_sched_out(struct preempt_notifier *pn,
2922                           struct task_struct *next)
2923 {
2924         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2925
2926         kvm_arch_vcpu_put(vcpu);
2927 }
2928
2929 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2930                   struct module *module)
2931 {
2932         int r;
2933         int cpu;
2934
2935         r = kvm_arch_init(opaque);
2936         if (r)
2937                 goto out_fail;
2938
2939         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2940                 r = -ENOMEM;
2941                 goto out_free_0;
2942         }
2943
2944         r = kvm_arch_hardware_setup();
2945         if (r < 0)
2946                 goto out_free_0a;
2947
2948         for_each_online_cpu(cpu) {
2949                 smp_call_function_single(cpu,
2950                                 kvm_arch_check_processor_compat,
2951                                 &r, 1);
2952                 if (r < 0)
2953                         goto out_free_1;
2954         }
2955
2956         r = register_cpu_notifier(&kvm_cpu_notifier);
2957         if (r)
2958                 goto out_free_2;
2959         register_reboot_notifier(&kvm_reboot_notifier);
2960
2961         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2962         if (!vcpu_align)
2963                 vcpu_align = __alignof__(struct kvm_vcpu);
2964         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2965                                            0, NULL);
2966         if (!kvm_vcpu_cache) {
2967                 r = -ENOMEM;
2968                 goto out_free_3;
2969         }
2970
2971         r = kvm_async_pf_init();
2972         if (r)
2973                 goto out_free;
2974
2975         kvm_chardev_ops.owner = module;
2976         kvm_vm_fops.owner = module;
2977         kvm_vcpu_fops.owner = module;
2978
2979         r = misc_register(&kvm_dev);
2980         if (r) {
2981                 printk(KERN_ERR "kvm: misc device register failed\n");
2982                 goto out_unreg;
2983         }
2984
2985         register_syscore_ops(&kvm_syscore_ops);
2986
2987         kvm_preempt_ops.sched_in = kvm_sched_in;
2988         kvm_preempt_ops.sched_out = kvm_sched_out;
2989
2990         r = kvm_init_debug();
2991         if (r) {
2992                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2993                 goto out_undebugfs;
2994         }
2995
2996         return 0;
2997
2998 out_undebugfs:
2999         unregister_syscore_ops(&kvm_syscore_ops);
3000 out_unreg:
3001         kvm_async_pf_deinit();
3002 out_free:
3003         kmem_cache_destroy(kvm_vcpu_cache);
3004 out_free_3:
3005         unregister_reboot_notifier(&kvm_reboot_notifier);
3006         unregister_cpu_notifier(&kvm_cpu_notifier);
3007 out_free_2:
3008 out_free_1:
3009         kvm_arch_hardware_unsetup();
3010 out_free_0a:
3011         free_cpumask_var(cpus_hardware_enabled);
3012 out_free_0:
3013         kvm_arch_exit();
3014 out_fail:
3015         return r;
3016 }
3017 EXPORT_SYMBOL_GPL(kvm_init);
3018
3019 void kvm_exit(void)
3020 {
3021         kvm_exit_debug();
3022         misc_deregister(&kvm_dev);
3023         kmem_cache_destroy(kvm_vcpu_cache);
3024         kvm_async_pf_deinit();
3025         unregister_syscore_ops(&kvm_syscore_ops);
3026         unregister_reboot_notifier(&kvm_reboot_notifier);
3027         unregister_cpu_notifier(&kvm_cpu_notifier);
3028         on_each_cpu(hardware_disable_nolock, NULL, 1);
3029         kvm_arch_hardware_unsetup();
3030         kvm_arch_exit();
3031         free_cpumask_var(cpus_hardware_enabled);
3032 }
3033 EXPORT_SYMBOL_GPL(kvm_exit);