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 void vcpu_load(struct kvm_vcpu *vcpu)
 135 {
 136         int cpu;
 137
 138         mutex_lock(&vcpu->mutex);
 139         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 140                 /* The thread running this VCPU changed. */
 141                 struct pid *oldpid = vcpu->pid;
 142                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 143                 rcu_assign_pointer(vcpu->pid, newpid);
 144                 synchronize_rcu();
 145                 put_pid(oldpid);
 146         }
 147         cpu = get_cpu();
 148         preempt_notifier_register(&vcpu->preempt_notifier);
 149         kvm_arch_vcpu_load(vcpu, cpu);
 150         put_cpu();
 151 }
 152
 153 void vcpu_put(struct kvm_vcpu *vcpu)
 154 {
 155         preempt_disable();
 156         kvm_arch_vcpu_put(vcpu);
 157         preempt_notifier_unregister(&vcpu->preempt_notifier);
 158         preempt_enable();
 159         mutex_unlock(&vcpu->mutex);
 160 }
 161
 162 static void ack_flush(void *_completed)
 163 {
 164 }
 165
 166 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 167 {
 168         int i, cpu, me;
 169         cpumask_var_t cpus;
 170         bool called = true;
 171         struct kvm_vcpu *vcpu;
 172
 173         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 174
 175         me = get_cpu();
 176         kvm_for_each_vcpu(i, vcpu, kvm) {
 177                 kvm_make_request(req, vcpu);
 178                 cpu = vcpu->cpu;
 179
 180                 /* Set ->requests bit before we read ->mode */
 181                 smp_mb();
 182
 183                 if (cpus != NULL && cpu != -1 && cpu != me &&
 184                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 185                         cpumask_set_cpu(cpu, cpus);
 186         }
 187         if (unlikely(cpus == NULL))
 188                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 189         else if (!cpumask_empty(cpus))
 190                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 191         else
 192                 called = false;
 193         put_cpu();
 194         free_cpumask_var(cpus);
 195         return called;
 196 }
 197
 198 void kvm_flush_remote_tlbs(struct kvm *kvm)
 199 {
 200         long dirty_count = kvm->tlbs_dirty;
 201
 202         smp_mb();
 203         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 204                 ++kvm->stat.remote_tlb_flush;
 205         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 206 }
 207
 208 void kvm_reload_remote_mmus(struct kvm *kvm)
 209 {
 210         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 211 }
 212
 213 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 214 {
 215         struct page *page;
 216         int r;
 217
 218         mutex_init(&vcpu->mutex);
 219         vcpu->cpu = -1;
 220         vcpu->kvm = kvm;
 221         vcpu->vcpu_id = id;
 222         vcpu->pid = NULL;
 223         init_waitqueue_head(&vcpu->wq);
 224         kvm_async_pf_vcpu_init(vcpu);
 225
 226         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 227         if (!page) {
 228                 r = -ENOMEM;
 229                 goto fail;
 230         }
 231         vcpu->run = page_address(page);
 232
 233         kvm_vcpu_set_in_spin_loop(vcpu, false);
 234         kvm_vcpu_set_dy_eligible(vcpu, false);
 235
 236         r = kvm_arch_vcpu_init(vcpu);
 237         if (r < 0)
 238                 goto fail_free_run;
 239         return 0;
 240
 241 fail_free_run:
 242         free_page((unsigned long)vcpu->run);
 243 fail:
 244         return r;
 245 }
 246 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 247
 248 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 249 {
 250         put_pid(vcpu->pid);
 251         kvm_arch_vcpu_uninit(vcpu);
 252         free_page((unsigned long)vcpu->run);
 253 }
 254 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 255
 256 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 257 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 258 {
 259         return container_of(mn, struct kvm, mmu_notifier);
 260 }
 261
 262 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 263                                              struct mm_struct *mm,
 264                                              unsigned long address)
 265 {
 266         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 267         int need_tlb_flush, idx;
 268
 269         /*
 270          * When ->invalidate_page runs, the linux pte has been zapped
 271          * already but the page is still allocated until
 272          * ->invalidate_page returns. So if we increase the sequence
 273          * here the kvm page fault will notice if the spte can't be
 274          * established because the page is going to be freed. If
 275          * instead the kvm page fault establishes the spte before
 276          * ->invalidate_page runs, kvm_unmap_hva will release it
 277          * before returning.
 278          *
 279          * The sequence increase only need to be seen at spin_unlock
 280          * time, and not at spin_lock time.
 281          *
 282          * Increasing the sequence after the spin_unlock would be
 283          * unsafe because the kvm page fault could then establish the
 284          * pte after kvm_unmap_hva returned, without noticing the page
 285          * is going to be freed.
 286          */
 287         idx = srcu_read_lock(&kvm->srcu);
 288         spin_lock(&kvm->mmu_lock);
 289
 290         kvm->mmu_notifier_seq++;
 291         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 292         /* we've to flush the tlb before the pages can be freed */
 293         if (need_tlb_flush)
 294                 kvm_flush_remote_tlbs(kvm);
 295
 296         spin_unlock(&kvm->mmu_lock);
 297         srcu_read_unlock(&kvm->srcu, idx);
 298 }
 299
 300 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 301                                         struct mm_struct *mm,
 302                                         unsigned long address,
 303                                         pte_t pte)
 304 {
 305         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 306         int idx;
 307
 308         idx = srcu_read_lock(&kvm->srcu);
 309         spin_lock(&kvm->mmu_lock);
 310         kvm->mmu_notifier_seq++;
 311         kvm_set_spte_hva(kvm, address, pte);
 312         spin_unlock(&kvm->mmu_lock);
 313         srcu_read_unlock(&kvm->srcu, idx);
 314 }
 315
 316 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 317                                                     struct mm_struct *mm,
 318                                                     unsigned long start,
 319                                                     unsigned long end)
 320 {
 321         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 322         int need_tlb_flush = 0, idx;
 323
 324         idx = srcu_read_lock(&kvm->srcu);
 325         spin_lock(&kvm->mmu_lock);
 326         /*
 327          * The count increase must become visible at unlock time as no
 328          * spte can be established without taking the mmu_lock and
 329          * count is also read inside the mmu_lock critical section.
 330          */
 331         kvm->mmu_notifier_count++;
 332         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 333         need_tlb_flush |= kvm->tlbs_dirty;
 334         /* we've to flush the tlb before the pages can be freed */
 335         if (need_tlb_flush)
 336                 kvm_flush_remote_tlbs(kvm);
 337
 338         spin_unlock(&kvm->mmu_lock);
 339         srcu_read_unlock(&kvm->srcu, idx);
 340 }
 341
 342 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 343                                                   struct mm_struct *mm,
 344                                                   unsigned long start,
 345                                                   unsigned long end)
 346 {
 347         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 348
 349         spin_lock(&kvm->mmu_lock);
 350         /*
 351          * This sequence increase will notify the kvm page fault that
 352          * the page that is going to be mapped in the spte could have
 353          * been freed.
 354          */
 355         kvm->mmu_notifier_seq++;
 356         smp_wmb();
 357         /*
 358          * The above sequence increase must be visible before the
 359          * below count decrease, which is ensured by the smp_wmb above
 360          * in conjunction with the smp_rmb in mmu_notifier_retry().
 361          */
 362         kvm->mmu_notifier_count--;
 363         spin_unlock(&kvm->mmu_lock);
 364
 365         BUG_ON(kvm->mmu_notifier_count < 0);
 366 }
 367
 368 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 369                                               struct mm_struct *mm,
 370                                               unsigned long address)
 371 {
 372         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 373         int young, idx;
 374
 375         idx = srcu_read_lock(&kvm->srcu);
 376         spin_lock(&kvm->mmu_lock);
 377
 378         young = kvm_age_hva(kvm, address);
 379         if (young)
 380                 kvm_flush_remote_tlbs(kvm);
 381
 382         spin_unlock(&kvm->mmu_lock);
 383         srcu_read_unlock(&kvm->srcu, idx);
 384
 385         return young;
 386 }
 387
 388 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 389                                        struct mm_struct *mm,
 390                                        unsigned long address)
 391 {
 392         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 393         int young, idx;
 394
 395         idx = srcu_read_lock(&kvm->srcu);
 396         spin_lock(&kvm->mmu_lock);
 397         young = kvm_test_age_hva(kvm, address);
 398         spin_unlock(&kvm->mmu_lock);
 399         srcu_read_unlock(&kvm->srcu, idx);
 400
 401         return young;
 402 }
 403
 404 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 405                                      struct mm_struct *mm)
 406 {
 407         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 408         int idx;
 409
 410         idx = srcu_read_lock(&kvm->srcu);
 411         kvm_arch_flush_shadow(kvm);
 412         srcu_read_unlock(&kvm->srcu, idx);
 413 }
 414
 415 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 416         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 417         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 418         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 419         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 420         .test_young             = kvm_mmu_notifier_test_young,
 421         .change_pte             = kvm_mmu_notifier_change_pte,
 422         .release                = kvm_mmu_notifier_release,
 423 };
 424
 425 static int kvm_init_mmu_notifier(struct kvm *kvm)
 426 {
 427         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 428         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 429 }
 430
 431 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 432
 433 static int kvm_init_mmu_notifier(struct kvm *kvm)
 434 {
 435         return 0;
 436 }
 437
 438 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 439
 440 static void kvm_init_memslots_id(struct kvm *kvm)
 441 {
 442         int i;
 443         struct kvm_memslots *slots = kvm->memslots;
 444
 445         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 446                 slots->id_to_index[i] = slots->memslots[i].id = i;
 447 }
 448
 449 static struct kvm *kvm_create_vm(unsigned long type)
 450 {
 451         int r, i;
 452         struct kvm *kvm = kvm_arch_alloc_vm();
 453
 454         if (!kvm)
 455                 return ERR_PTR(-ENOMEM);
 456
 457         r = kvm_arch_init_vm(kvm, type);
 458         if (r)
 459                 goto out_err_nodisable;
 460
 461         r = hardware_enable_all();
 462         if (r)
 463                 goto out_err_nodisable;
 464
 465 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 466         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 467         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 468 #endif
 469
 470         r = -ENOMEM;
 471         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 472         if (!kvm->memslots)
 473                 goto out_err_nosrcu;
 474         kvm_init_memslots_id(kvm);
 475         if (init_srcu_struct(&kvm->srcu))
 476                 goto out_err_nosrcu;
 477         for (i = 0; i < KVM_NR_BUSES; i++) {
 478                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 479                                         GFP_KERNEL);
 480                 if (!kvm->buses[i])
 481                         goto out_err;
 482         }
 483
 484         spin_lock_init(&kvm->mmu_lock);
 485         kvm->mm = current->mm;
 486         atomic_inc(&kvm->mm->mm_count);
 487         kvm_eventfd_init(kvm);
 488         mutex_init(&kvm->lock);
 489         mutex_init(&kvm->irq_lock);
 490         mutex_init(&kvm->slots_lock);
 491         atomic_set(&kvm->users_count, 1);
 492
 493         r = kvm_init_mmu_notifier(kvm);
 494         if (r)
 495                 goto out_err;
 496
 497         raw_spin_lock(&kvm_lock);
 498         list_add(&kvm->vm_list, &vm_list);
 499         raw_spin_unlock(&kvm_lock);
 500
 501         return kvm;
 502
 503 out_err:
 504         cleanup_srcu_struct(&kvm->srcu);
 505 out_err_nosrcu:
 506         hardware_disable_all();
 507 out_err_nodisable:
 508         for (i = 0; i < KVM_NR_BUSES; i++)
 509                 kfree(kvm->buses[i]);
 510         kfree(kvm->memslots);
 511         kvm_arch_free_vm(kvm);
 512         return ERR_PTR(r);
 513 }
 514
 515 /*
 516  * Avoid using vmalloc for a small buffer.
 517  * Should not be used when the size is statically known.
 518  */
 519 void *kvm_kvzalloc(unsigned long size)
 520 {
 521         if (size > PAGE_SIZE)
 522                 return vzalloc(size);
 523         else
 524                 return kzalloc(size, GFP_KERNEL);
 525 }
 526
 527 void kvm_kvfree(const void *addr)
 528 {
 529         if (is_vmalloc_addr(addr))
 530                 vfree(addr);
 531         else
 532                 kfree(addr);
 533 }
 534
 535 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 536 {
 537         if (!memslot->dirty_bitmap)
 538                 return;
 539
 540         kvm_kvfree(memslot->dirty_bitmap);
 541         memslot->dirty_bitmap = NULL;
 542 }
 543
 544 /*
 545  * Free any memory in @free but not in @dont.
 546  */
 547 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 548                                   struct kvm_memory_slot *dont)
 549 {
 550         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 551                 kvm_destroy_dirty_bitmap(free);
 552
 553         kvm_arch_free_memslot(free, dont);
 554
 555         free->npages = 0;
 556 }
 557
 558 void kvm_free_physmem(struct kvm *kvm)
 559 {
 560         struct kvm_memslots *slots = kvm->memslots;
 561         struct kvm_memory_slot *memslot;
 562
 563         kvm_for_each_memslot(memslot, slots)
 564                 kvm_free_physmem_slot(memslot, NULL);
 565
 566         kfree(kvm->memslots);
 567 }
 568
 569 static void kvm_destroy_vm(struct kvm *kvm)
 570 {
 571         int i;
 572         struct mm_struct *mm = kvm->mm;
 573
 574         kvm_arch_sync_events(kvm);
 575         raw_spin_lock(&kvm_lock);
 576         list_del(&kvm->vm_list);
 577         raw_spin_unlock(&kvm_lock);
 578         kvm_free_irq_routing(kvm);
 579         for (i = 0; i < KVM_NR_BUSES; i++)
 580                 kvm_io_bus_destroy(kvm->buses[i]);
 581         kvm_coalesced_mmio_free(kvm);
 582 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 583         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 584 #else
 585         kvm_arch_flush_shadow(kvm);
 586 #endif
 587         kvm_arch_destroy_vm(kvm);
 588         kvm_free_physmem(kvm);
 589         cleanup_srcu_struct(&kvm->srcu);
 590         kvm_arch_free_vm(kvm);
 591         hardware_disable_all();
 592         mmdrop(mm);
 593 }
 594
 595 void kvm_get_kvm(struct kvm *kvm)
 596 {
 597         atomic_inc(&kvm->users_count);
 598 }
 599 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 600
 601 void kvm_put_kvm(struct kvm *kvm)
 602 {
 603         if (atomic_dec_and_test(&kvm->users_count))
 604                 kvm_destroy_vm(kvm);
 605 }
 606 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 607
 608
 609 static int kvm_vm_release(struct inode *inode, struct file *filp)
 610 {
 611         struct kvm *kvm = filp->private_data;
 612
 613         kvm_irqfd_release(kvm);
 614
 615         kvm_put_kvm(kvm);
 616         return 0;
 617 }
 618
 619 /*
 620  * Allocation size is twice as large as the actual dirty bitmap size.
 621  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 622  */
 623 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 624 {
 625 #ifndef CONFIG_S390
 626         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 627
 628         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 629         if (!memslot->dirty_bitmap)
 630                 return -ENOMEM;
 631
 632 #endif /* !CONFIG_S390 */
 633         return 0;
 634 }
 635
 636 static int cmp_memslot(const void *slot1, const void *slot2)
 637 {
 638         struct kvm_memory_slot *s1, *s2;
 639
 640         s1 = (struct kvm_memory_slot *)slot1;
 641         s2 = (struct kvm_memory_slot *)slot2;
 642
 643         if (s1->npages < s2->npages)
 644                 return 1;
 645         if (s1->npages > s2->npages)
 646                 return -1;
 647
 648         return 0;
 649 }
 650
 651 /*
 652  * Sort the memslots base on its size, so the larger slots
 653  * will get better fit.
 654  */
 655 static void sort_memslots(struct kvm_memslots *slots)
 656 {
 657         int i;
 658
 659         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 660               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 661
 662         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 663                 slots->id_to_index[slots->memslots[i].id] = i;
 664 }
 665
 666 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
 667 {
 668         if (new) {
 669                 int id = new->id;
 670                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 671                 unsigned long npages = old->npages;
 672
 673                 *old = *new;
 674                 if (new->npages != npages)
 675                         sort_memslots(slots);
 676         }
 677
 678         slots->generation++;
 679 }
 680
 681 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 682 {
 683         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 684
 685 #ifdef KVM_CAP_READONLY_MEM
 686         valid_flags |= KVM_MEM_READONLY;
 687 #endif
 688
 689         if (mem->flags & ~valid_flags)
 690                 return -EINVAL;
 691
 692         return 0;
 693 }
 694
 695 /*
 696  * Allocate some memory and give it an address in the guest physical address
 697  * space.
 698  *
 699  * Discontiguous memory is allowed, mostly for framebuffers.
 700  *
 701  * Must be called holding mmap_sem for write.
 702  */
 703 int __kvm_set_memory_region(struct kvm *kvm,
 704                             struct kvm_userspace_memory_region *mem,
 705                             int user_alloc)
 706 {
 707         int r;
 708         gfn_t base_gfn;
 709         unsigned long npages;
 710         unsigned long i;
 711         struct kvm_memory_slot *memslot;
 712         struct kvm_memory_slot old, new;
 713         struct kvm_memslots *slots, *old_memslots;
 714
 715         r = check_memory_region_flags(mem);
 716         if (r)
 717                 goto out;
 718
 719         r = -EINVAL;
 720         /* General sanity checks */
 721         if (mem->memory_size & (PAGE_SIZE - 1))
 722                 goto out;
 723         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 724                 goto out;
 725         /* We can read the guest memory with __xxx_user() later on. */
 726         if (user_alloc &&
 727             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 728              !access_ok(VERIFY_WRITE,
 729                         (void __user *)(unsigned long)mem->userspace_addr,
 730                         mem->memory_size)))
 731                 goto out;
 732         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 733                 goto out;
 734         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 735                 goto out;
 736
 737         memslot = id_to_memslot(kvm->memslots, mem->slot);
 738         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 739         npages = mem->memory_size >> PAGE_SHIFT;
 740
 741         r = -EINVAL;
 742         if (npages > KVM_MEM_MAX_NR_PAGES)
 743                 goto out;
 744
 745         if (!npages)
 746                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 747
 748         new = old = *memslot;
 749
 750         new.id = mem->slot;
 751         new.base_gfn = base_gfn;
 752         new.npages = npages;
 753         new.flags = mem->flags;
 754
 755         /* Disallow changing a memory slot's size. */
 756         r = -EINVAL;
 757         if (npages && old.npages && npages != old.npages)
 758                 goto out_free;
 759
 760         /* Check for overlaps */
 761         r = -EEXIST;
 762         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 763                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 764
 765                 if (s == memslot || !s->npages)
 766                         continue;
 767                 if (!((base_gfn + npages <= s->base_gfn) ||
 768                       (base_gfn >= s->base_gfn + s->npages)))
 769                         goto out_free;
 770         }
 771
 772         /* Free page dirty bitmap if unneeded */
 773         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 774                 new.dirty_bitmap = NULL;
 775
 776         r = -ENOMEM;
 777
 778         /* Allocate if a slot is being created */
 779         if (npages && !old.npages) {
 780                 new.user_alloc = user_alloc;
 781                 new.userspace_addr = mem->userspace_addr;
 782
 783                 if (kvm_arch_create_memslot(&new, npages))
 784                         goto out_free;
 785         }
 786
 787         /* Allocate page dirty bitmap if needed */
 788         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 789                 if (kvm_create_dirty_bitmap(&new) < 0)
 790                         goto out_free;
 791                 /* destroy any largepage mappings for dirty tracking */
 792         }
 793
 794         if (!npages) {
 795                 struct kvm_memory_slot *slot;
 796
 797                 r = -ENOMEM;
 798                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 799                                 GFP_KERNEL);
 800                 if (!slots)
 801                         goto out_free;
 802                 slot = id_to_memslot(slots, mem->slot);
 803                 slot->flags |= KVM_MEMSLOT_INVALID;
 804
 805                 update_memslots(slots, NULL);
 806
 807                 old_memslots = kvm->memslots;
 808                 rcu_assign_pointer(kvm->memslots, slots);
 809                 synchronize_srcu_expedited(&kvm->srcu);
 810                 /* From this point no new shadow pages pointing to a deleted
 811                  * memslot will be created.
 812                  *
 813                  * validation of sp->gfn happens in:
 814                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 815                  *      - kvm_is_visible_gfn (mmu_check_roots)
 816                  */
 817                 kvm_arch_flush_shadow(kvm);
 818                 kfree(old_memslots);
 819         }
 820
 821         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 822         if (r)
 823                 goto out_free;
 824
 825         /* map/unmap the pages in iommu page table */
 826         if (npages) {
 827                 r = kvm_iommu_map_pages(kvm, &new);
 828                 if (r)
 829                         goto out_free;
 830         } else
 831                 kvm_iommu_unmap_pages(kvm, &old);
 832
 833         r = -ENOMEM;
 834         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 835                         GFP_KERNEL);
 836         if (!slots)
 837                 goto out_free;
 838
 839         /* actual memory is freed via old in kvm_free_physmem_slot below */
 840         if (!npages) {
 841                 new.dirty_bitmap = NULL;
 842                 memset(&new.arch, 0, sizeof(new.arch));
 843         }
 844
 845         update_memslots(slots, &new);
 846         old_memslots = kvm->memslots;
 847         rcu_assign_pointer(kvm->memslots, slots);
 848         synchronize_srcu_expedited(&kvm->srcu);
 849
 850         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 851
 852         /*
 853          * If the new memory slot is created, we need to clear all
 854          * mmio sptes.
 855          */
 856         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 857                 kvm_arch_flush_shadow(kvm);
 858
 859         kvm_free_physmem_slot(&old, &new);
 860         kfree(old_memslots);
 861
 862         return 0;
 863
 864 out_free:
 865         kvm_free_physmem_slot(&new, &old);
 866 out:
 867         return r;
 868
 869 }
 870 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 871
 872 int kvm_set_memory_region(struct kvm *kvm,
 873                           struct kvm_userspace_memory_region *mem,
 874                           int user_alloc)
 875 {
 876         int r;
 877
 878         mutex_lock(&kvm->slots_lock);
 879         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 880         mutex_unlock(&kvm->slots_lock);
 881         return r;
 882 }
 883 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 884
 885 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 886                                    struct
 887                                    kvm_userspace_memory_region *mem,
 888                                    int user_alloc)
 889 {
 890         if (mem->slot >= KVM_MEMORY_SLOTS)
 891                 return -EINVAL;
 892         return kvm_set_memory_region(kvm, mem, user_alloc);
 893 }
 894
 895 int kvm_get_dirty_log(struct kvm *kvm,
 896                         struct kvm_dirty_log *log, int *is_dirty)
 897 {
 898         struct kvm_memory_slot *memslot;
 899         int r, i;
 900         unsigned long n;
 901         unsigned long any = 0;
 902
 903         r = -EINVAL;
 904         if (log->slot >= KVM_MEMORY_SLOTS)
 905                 goto out;
 906
 907         memslot = id_to_memslot(kvm->memslots, log->slot);
 908         r = -ENOENT;
 909         if (!memslot->dirty_bitmap)
 910                 goto out;
 911
 912         n = kvm_dirty_bitmap_bytes(memslot);
 913
 914         for (i = 0; !any && i < n/sizeof(long); ++i)
 915                 any = memslot->dirty_bitmap[i];
 916
 917         r = -EFAULT;
 918         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 919                 goto out;
 920
 921         if (any)
 922                 *is_dirty = 1;
 923
 924         r = 0;
 925 out:
 926         return r;
 927 }
 928
 929 bool kvm_largepages_enabled(void)
 930 {
 931         return largepages_enabled;
 932 }
 933
 934 void kvm_disable_largepages(void)
 935 {
 936         largepages_enabled = false;
 937 }
 938 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 939
 940 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 941 {
 942         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 943 }
 944 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 945
 946 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 947 {
 948         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 949
 950         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
 951               memslot->flags & KVM_MEMSLOT_INVALID)
 952                 return 0;
 953
 954         return 1;
 955 }
 956 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 957
 958 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 959 {
 960         struct vm_area_struct *vma;
 961         unsigned long addr, size;
 962
 963         size = PAGE_SIZE;
 964
 965         addr = gfn_to_hva(kvm, gfn);
 966         if (kvm_is_error_hva(addr))
 967                 return PAGE_SIZE;
 968
 969         down_read(&current->mm->mmap_sem);
 970         vma = find_vma(current->mm, addr);
 971         if (!vma)
 972                 goto out;
 973
 974         size = vma_kernel_pagesize(vma);
 975
 976 out:
 977         up_read(&current->mm->mmap_sem);
 978
 979         return size;
 980 }
 981
 982 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
 983 {
 984         return slot->flags & KVM_MEM_READONLY;
 985 }
 986
 987 static unsigned long __gfn_to_hva_memslot(struct kvm_memory_slot *slot,
 988                                           gfn_t gfn)
 989 {
 990         return slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE;
 991 }
 992
 993 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
 994                                        gfn_t *nr_pages, bool write)
 995 {
 996         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 997                 return KVM_HVA_ERR_BAD;
 998
 999         if (memslot_is_readonly(slot) && write)
1000                 return KVM_HVA_ERR_RO_BAD;
1001
1002         if (nr_pages)
1003                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1004
1005         return __gfn_to_hva_memslot(slot, gfn);
1006 }
1007
1008 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1009                                      gfn_t *nr_pages)
1010 {
1011         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1012 }
1013
1014 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1015                                  gfn_t gfn)
1016 {
1017         return gfn_to_hva_many(slot, gfn, NULL);
1018 }
1019 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1020
1021 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1022 {
1023         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1024 }
1025 EXPORT_SYMBOL_GPL(gfn_to_hva);
1026
1027 /*
1028  * The hva returned by this function is only allowed to be read.
1029  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1030  */
1031 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1032 {
1033         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1034 }
1035
1036 static int kvm_read_hva(void *data, void __user *hva, int len)
1037 {
1038         return __copy_from_user(data, hva, len);
1039 }
1040
1041 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1042 {
1043         return __copy_from_user_inatomic(data, hva, len);
1044 }
1045
1046 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1047         unsigned long start, int write, struct page **page)
1048 {
1049         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1050
1051         if (write)
1052                 flags |= FOLL_WRITE;
1053
1054         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1055 }
1056
1057 static inline int check_user_page_hwpoison(unsigned long addr)
1058 {
1059         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1060
1061         rc = __get_user_pages(current, current->mm, addr, 1,
1062                               flags, NULL, NULL, NULL);
1063         return rc == -EHWPOISON;
1064 }
1065
1066 /*
1067  * The atomic path to get the writable pfn which will be stored in @pfn,
1068  * true indicates success, otherwise false is returned.
1069  */
1070 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1071                             bool write_fault, bool *writable, pfn_t *pfn)
1072 {
1073         struct page *page[1];
1074         int npages;
1075
1076         if (!(async || atomic))
1077                 return false;
1078
1079         /*
1080          * Fast pin a writable pfn only if it is a write fault request
1081          * or the caller allows to map a writable pfn for a read fault
1082          * request.
1083          */
1084         if (!(write_fault || writable))
1085                 return false;
1086
1087         npages = __get_user_pages_fast(addr, 1, 1, page);
1088         if (npages == 1) {
1089                 *pfn = page_to_pfn(page[0]);
1090
1091                 if (writable)
1092                         *writable = true;
1093                 return true;
1094         }
1095
1096         return false;
1097 }
1098
1099 /*
1100  * The slow path to get the pfn of the specified host virtual address,
1101  * 1 indicates success, -errno is returned if error is detected.
1102  */
1103 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1104                            bool *writable, pfn_t *pfn)
1105 {
1106         struct page *page[1];
1107         int npages = 0;
1108
1109         might_sleep();
1110
1111         if (writable)
1112                 *writable = write_fault;
1113
1114         if (async) {
1115                 down_read(&current->mm->mmap_sem);
1116                 npages = get_user_page_nowait(current, current->mm,
1117                                               addr, write_fault, page);
1118                 up_read(&current->mm->mmap_sem);
1119         } else
1120                 npages = get_user_pages_fast(addr, 1, write_fault,
1121                                              page);
1122         if (npages != 1)
1123                 return npages;
1124
1125         /* map read fault as writable if possible */
1126         if (unlikely(!write_fault) && writable) {
1127                 struct page *wpage[1];
1128
1129                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1130                 if (npages == 1) {
1131                         *writable = true;
1132                         put_page(page[0]);
1133                         page[0] = wpage[0];
1134                 }
1135
1136                 npages = 1;
1137         }
1138         *pfn = page_to_pfn(page[0]);
1139         return npages;
1140 }
1141
1142 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1143 {
1144         if (unlikely(!(vma->vm_flags & VM_READ)))
1145                 return false;
1146
1147         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1148                 return false;
1149
1150         return true;
1151 }
1152
1153 /*
1154  * Pin guest page in memory and return its pfn.
1155  * @addr: host virtual address which maps memory to the guest
1156  * @atomic: whether this function can sleep
1157  * @async: whether this function need to wait IO complete if the
1158  *         host page is not in the memory
1159  * @write_fault: whether we should get a writable host page
1160  * @writable: whether it allows to map a writable host page for !@write_fault
1161  *
1162  * The function will map a writable host page for these two cases:
1163  * 1): @write_fault = true
1164  * 2): @write_fault = false && @writable, @writable will tell the caller
1165  *     whether the mapping is writable.
1166  */
1167 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1168                         bool write_fault, bool *writable)
1169 {
1170         struct vm_area_struct *vma;
1171         pfn_t pfn = 0;
1172         int npages;
1173
1174         /* we can do it either atomically or asynchronously, not both */
1175         BUG_ON(atomic && async);
1176
1177         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1178                 return pfn;
1179
1180         if (atomic)
1181                 return KVM_PFN_ERR_FAULT;
1182
1183         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1184         if (npages == 1)
1185                 return pfn;
1186
1187         down_read(&current->mm->mmap_sem);
1188         if (npages == -EHWPOISON ||
1189               (!async && check_user_page_hwpoison(addr))) {
1190                 pfn = KVM_PFN_ERR_HWPOISON;
1191                 goto exit;
1192         }
1193
1194         vma = find_vma_intersection(current->mm, addr, addr + 1);
1195
1196         if (vma == NULL)
1197                 pfn = KVM_PFN_ERR_FAULT;
1198         else if ((vma->vm_flags & VM_PFNMAP)) {
1199                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1200                         vma->vm_pgoff;
1201                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1202         } else {
1203                 if (async && vma_is_valid(vma, write_fault))
1204                         *async = true;
1205                 pfn = KVM_PFN_ERR_FAULT;
1206         }
1207 exit:
1208         up_read(&current->mm->mmap_sem);
1209         return pfn;
1210 }
1211
1212 static pfn_t
1213 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1214                      bool *async, bool write_fault, bool *writable)
1215 {
1216         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1217
1218         if (addr == KVM_HVA_ERR_RO_BAD)
1219                 return KVM_PFN_ERR_RO_FAULT;
1220
1221         if (kvm_is_error_hva(addr))
1222                 return KVM_PFN_ERR_BAD;
1223
1224         /* Do not map writable pfn in the readonly memslot. */
1225         if (writable && memslot_is_readonly(slot)) {
1226                 *writable = false;
1227                 writable = NULL;
1228         }
1229
1230         return hva_to_pfn(addr, atomic, async, write_fault,
1231                           writable);
1232 }
1233
1234 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1235                           bool write_fault, bool *writable)
1236 {
1237         struct kvm_memory_slot *slot;
1238
1239         if (async)
1240                 *async = false;
1241
1242         slot = gfn_to_memslot(kvm, gfn);
1243
1244         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1245                                     writable);
1246 }
1247
1248 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1249 {
1250         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1251 }
1252 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1253
1254 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1255                        bool write_fault, bool *writable)
1256 {
1257         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1258 }
1259 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1260
1261 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1262 {
1263         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1264 }
1265 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1266
1267 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1268                       bool *writable)
1269 {
1270         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1271 }
1272 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1273
1274 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1275 {
1276         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1277 }
1278
1279 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1280 {
1281         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1282 }
1283 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1284
1285 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1286                                                                   int nr_pages)
1287 {
1288         unsigned long addr;
1289         gfn_t entry;
1290
1291         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1292         if (kvm_is_error_hva(addr))
1293                 return -1;
1294
1295         if (entry < nr_pages)
1296                 return 0;
1297
1298         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1299 }
1300 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1301
1302 static struct page *kvm_pfn_to_page(pfn_t pfn)
1303 {
1304         if (is_error_pfn(pfn))
1305                 return KVM_ERR_PTR_BAD_PAGE;
1306
1307         if (kvm_is_mmio_pfn(pfn)) {
1308                 WARN_ON(1);
1309                 return KVM_ERR_PTR_BAD_PAGE;
1310         }
1311
1312         return pfn_to_page(pfn);
1313 }
1314
1315 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1316 {
1317         pfn_t pfn;
1318
1319         pfn = gfn_to_pfn(kvm, gfn);
1320
1321         return kvm_pfn_to_page(pfn);
1322 }
1323
1324 EXPORT_SYMBOL_GPL(gfn_to_page);
1325
1326 void kvm_release_page_clean(struct page *page)
1327 {
1328         WARN_ON(is_error_page(page));
1329
1330         kvm_release_pfn_clean(page_to_pfn(page));
1331 }
1332 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1333
1334 void kvm_release_pfn_clean(pfn_t pfn)
1335 {
1336         WARN_ON(is_error_pfn(pfn));
1337
1338         if (!kvm_is_mmio_pfn(pfn))
1339                 put_page(pfn_to_page(pfn));
1340 }
1341 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1342
1343 void kvm_release_page_dirty(struct page *page)
1344 {
1345         WARN_ON(is_error_page(page));
1346
1347         kvm_release_pfn_dirty(page_to_pfn(page));
1348 }
1349 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1350
1351 void kvm_release_pfn_dirty(pfn_t pfn)
1352 {
1353         kvm_set_pfn_dirty(pfn);
1354         kvm_release_pfn_clean(pfn);
1355 }
1356 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1357
1358 void kvm_set_page_dirty(struct page *page)
1359 {
1360         kvm_set_pfn_dirty(page_to_pfn(page));
1361 }
1362 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1363
1364 void kvm_set_pfn_dirty(pfn_t pfn)
1365 {
1366         if (!kvm_is_mmio_pfn(pfn)) {
1367                 struct page *page = pfn_to_page(pfn);
1368                 if (!PageReserved(page))
1369                         SetPageDirty(page);
1370         }
1371 }
1372 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1373
1374 void kvm_set_pfn_accessed(pfn_t pfn)
1375 {
1376         if (!kvm_is_mmio_pfn(pfn))
1377                 mark_page_accessed(pfn_to_page(pfn));
1378 }
1379 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1380
1381 void kvm_get_pfn(pfn_t pfn)
1382 {
1383         if (!kvm_is_mmio_pfn(pfn))
1384                 get_page(pfn_to_page(pfn));
1385 }
1386 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1387
1388 static int next_segment(unsigned long len, int offset)
1389 {
1390         if (len > PAGE_SIZE - offset)
1391                 return PAGE_SIZE - offset;
1392         else
1393                 return len;
1394 }
1395
1396 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1397                         int len)
1398 {
1399         int r;
1400         unsigned long addr;
1401
1402         addr = gfn_to_hva_read(kvm, gfn);
1403         if (kvm_is_error_hva(addr))
1404                 return -EFAULT;
1405         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1406         if (r)
1407                 return -EFAULT;
1408         return 0;
1409 }
1410 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1411
1412 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1413 {
1414         gfn_t gfn = gpa >> PAGE_SHIFT;
1415         int seg;
1416         int offset = offset_in_page(gpa);
1417         int ret;
1418
1419         while ((seg = next_segment(len, offset)) != 0) {
1420                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1421                 if (ret < 0)
1422                         return ret;
1423                 offset = 0;
1424                 len -= seg;
1425                 data += seg;
1426                 ++gfn;
1427         }
1428         return 0;
1429 }
1430 EXPORT_SYMBOL_GPL(kvm_read_guest);
1431
1432 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1433                           unsigned long len)
1434 {
1435         int r;
1436         unsigned long addr;
1437         gfn_t gfn = gpa >> PAGE_SHIFT;
1438         int offset = offset_in_page(gpa);
1439
1440         addr = gfn_to_hva_read(kvm, gfn);
1441         if (kvm_is_error_hva(addr))
1442                 return -EFAULT;
1443         pagefault_disable();
1444         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1445         pagefault_enable();
1446         if (r)
1447                 return -EFAULT;
1448         return 0;
1449 }
1450 EXPORT_SYMBOL(kvm_read_guest_atomic);
1451
1452 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1453                          int offset, int len)
1454 {
1455         int r;
1456         unsigned long addr;
1457
1458         addr = gfn_to_hva(kvm, gfn);
1459         if (kvm_is_error_hva(addr))
1460                 return -EFAULT;
1461         r = __copy_to_user((void __user *)addr + offset, data, len);
1462         if (r)
1463                 return -EFAULT;
1464         mark_page_dirty(kvm, gfn);
1465         return 0;
1466 }
1467 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1468
1469 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1470                     unsigned long len)
1471 {
1472         gfn_t gfn = gpa >> PAGE_SHIFT;
1473         int seg;
1474         int offset = offset_in_page(gpa);
1475         int ret;
1476
1477         while ((seg = next_segment(len, offset)) != 0) {
1478                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1479                 if (ret < 0)
1480                         return ret;
1481                 offset = 0;
1482                 len -= seg;
1483                 data += seg;
1484                 ++gfn;
1485         }
1486         return 0;
1487 }
1488
1489 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1490                               gpa_t gpa)
1491 {
1492         struct kvm_memslots *slots = kvm_memslots(kvm);
1493         int offset = offset_in_page(gpa);
1494         gfn_t gfn = gpa >> PAGE_SHIFT;
1495
1496         ghc->gpa = gpa;
1497         ghc->generation = slots->generation;
1498         ghc->memslot = gfn_to_memslot(kvm, gfn);
1499         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1500         if (!kvm_is_error_hva(ghc->hva))
1501                 ghc->hva += offset;
1502         else
1503                 return -EFAULT;
1504
1505         return 0;
1506 }
1507 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1508
1509 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1510                            void *data, unsigned long len)
1511 {
1512         struct kvm_memslots *slots = kvm_memslots(kvm);
1513         int r;
1514
1515         if (slots->generation != ghc->generation)
1516                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1517
1518         if (kvm_is_error_hva(ghc->hva))
1519                 return -EFAULT;
1520
1521         r = __copy_to_user((void __user *)ghc->hva, data, len);
1522         if (r)
1523                 return -EFAULT;
1524         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1525
1526         return 0;
1527 }
1528 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1529
1530 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1531                            void *data, unsigned long len)
1532 {
1533         struct kvm_memslots *slots = kvm_memslots(kvm);
1534         int r;
1535
1536         if (slots->generation != ghc->generation)
1537                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1538
1539         if (kvm_is_error_hva(ghc->hva))
1540                 return -EFAULT;
1541
1542         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1543         if (r)
1544                 return -EFAULT;
1545
1546         return 0;
1547 }
1548 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1549
1550 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1551 {
1552         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1553                                     offset, len);
1554 }
1555 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1556
1557 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1558 {
1559         gfn_t gfn = gpa >> PAGE_SHIFT;
1560         int seg;
1561         int offset = offset_in_page(gpa);
1562         int ret;
1563
1564         while ((seg = next_segment(len, offset)) != 0) {
1565                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1566                 if (ret < 0)
1567                         return ret;
1568                 offset = 0;
1569                 len -= seg;
1570                 ++gfn;
1571         }
1572         return 0;
1573 }
1574 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1575
1576 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1577                              gfn_t gfn)
1578 {
1579         if (memslot && memslot->dirty_bitmap) {
1580                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1581
1582                 /* TODO: introduce set_bit_le() and use it */
1583                 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1584         }
1585 }
1586
1587 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1588 {
1589         struct kvm_memory_slot *memslot;
1590
1591         memslot = gfn_to_memslot(kvm, gfn);
1592         mark_page_dirty_in_slot(kvm, memslot, gfn);
1593 }
1594
1595 /*
1596  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1597  */
1598 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1599 {
1600         DEFINE_WAIT(wait);
1601
1602         for (;;) {
1603                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1604
1605                 if (kvm_arch_vcpu_runnable(vcpu)) {
1606                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1607                         break;
1608                 }
1609                 if (kvm_cpu_has_pending_timer(vcpu))
1610                         break;
1611                 if (signal_pending(current))
1612                         break;
1613
1614                 schedule();
1615         }
1616
1617         finish_wait(&vcpu->wq, &wait);
1618 }
1619
1620 #ifndef CONFIG_S390
1621 /*
1622  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1623  */
1624 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1625 {
1626         int me;
1627         int cpu = vcpu->cpu;
1628         wait_queue_head_t *wqp;
1629
1630         wqp = kvm_arch_vcpu_wq(vcpu);
1631         if (waitqueue_active(wqp)) {
1632                 wake_up_interruptible(wqp);
1633                 ++vcpu->stat.halt_wakeup;
1634         }
1635
1636         me = get_cpu();
1637         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1638                 if (kvm_arch_vcpu_should_kick(vcpu))
1639                         smp_send_reschedule(cpu);
1640         put_cpu();
1641 }
1642 #endif /* !CONFIG_S390 */
1643
1644 void kvm_resched(struct kvm_vcpu *vcpu)
1645 {
1646         if (!need_resched())
1647                 return;
1648         cond_resched();
1649 }
1650 EXPORT_SYMBOL_GPL(kvm_resched);
1651
1652 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1653 {
1654         struct pid *pid;
1655         struct task_struct *task = NULL;
1656
1657         rcu_read_lock();
1658         pid = rcu_dereference(target->pid);
1659         if (pid)
1660                 task = get_pid_task(target->pid, PIDTYPE_PID);
1661         rcu_read_unlock();
1662         if (!task)
1663                 return false;
1664         if (task->flags & PF_VCPU) {
1665                 put_task_struct(task);
1666                 return false;
1667         }
1668         if (yield_to(task, 1)) {
1669                 put_task_struct(task);
1670                 return true;
1671         }
1672         put_task_struct(task);
1673         return false;
1674 }
1675 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1676
1677 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1678 /*
1679  * Helper that checks whether a VCPU is eligible for directed yield.
1680  * Most eligible candidate to yield is decided by following heuristics:
1681  *
1682  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1683  *  (preempted lock holder), indicated by @in_spin_loop.
1684  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1685  *
1686  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1687  *  chance last time (mostly it has become eligible now since we have probably
1688  *  yielded to lockholder in last iteration. This is done by toggling
1689  *  @dy_eligible each time a VCPU checked for eligibility.)
1690  *
1691  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1692  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1693  *  burning. Giving priority for a potential lock-holder increases lock
1694  *  progress.
1695  *
1696  *  Since algorithm is based on heuristics, accessing another VCPU data without
1697  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1698  *  and continue with next VCPU and so on.
1699  */
1700 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1701 {
1702         bool eligible;
1703
1704         eligible = !vcpu->spin_loop.in_spin_loop ||
1705                         (vcpu->spin_loop.in_spin_loop &&
1706                          vcpu->spin_loop.dy_eligible);
1707
1708         if (vcpu->spin_loop.in_spin_loop)
1709                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1710
1711         return eligible;
1712 }
1713 #endif
1714 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1715 {
1716         struct kvm *kvm = me->kvm;
1717         struct kvm_vcpu *vcpu;
1718         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1719         int yielded = 0;
1720         int pass;
1721         int i;
1722
1723         kvm_vcpu_set_in_spin_loop(me, true);
1724         /*
1725          * We boost the priority of a VCPU that is runnable but not
1726          * currently running, because it got preempted by something
1727          * else and called schedule in __vcpu_run.  Hopefully that
1728          * VCPU is holding the lock that we need and will release it.
1729          * We approximate round-robin by starting at the last boosted VCPU.
1730          */
1731         for (pass = 0; pass < 2 && !yielded; pass++) {
1732                 kvm_for_each_vcpu(i, vcpu, kvm) {
1733                         if (!pass && i <= last_boosted_vcpu) {
1734                                 i = last_boosted_vcpu;
1735                                 continue;
1736                         } else if (pass && i > last_boosted_vcpu)
1737                                 break;
1738                         if (vcpu == me)
1739                                 continue;
1740                         if (waitqueue_active(&vcpu->wq))
1741                                 continue;
1742                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1743                                 continue;
1744                         if (kvm_vcpu_yield_to(vcpu)) {
1745                                 kvm->last_boosted_vcpu = i;
1746                                 yielded = 1;
1747                                 break;
1748                         }
1749                 }
1750         }
1751         kvm_vcpu_set_in_spin_loop(me, false);
1752
1753         /* Ensure vcpu is not eligible during next spinloop */
1754         kvm_vcpu_set_dy_eligible(me, false);
1755 }
1756 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1757
1758 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1759 {
1760         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1761         struct page *page;
1762
1763         if (vmf->pgoff == 0)
1764                 page = virt_to_page(vcpu->run);
1765 #ifdef CONFIG_X86
1766         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1767                 page = virt_to_page(vcpu->arch.pio_data);
1768 #endif
1769 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1770         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1771                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1772 #endif
1773         else
1774                 return kvm_arch_vcpu_fault(vcpu, vmf);
1775         get_page(page);
1776         vmf->page = page;
1777         return 0;
1778 }
1779
1780 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1781         .fault = kvm_vcpu_fault,
1782 };
1783
1784 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1785 {
1786         vma->vm_ops = &kvm_vcpu_vm_ops;
1787         return 0;
1788 }
1789
1790 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1791 {
1792         struct kvm_vcpu *vcpu = filp->private_data;
1793
1794         kvm_put_kvm(vcpu->kvm);
1795         return 0;
1796 }
1797
1798 static struct file_operations kvm_vcpu_fops = {
1799         .release        = kvm_vcpu_release,
1800         .unlocked_ioctl = kvm_vcpu_ioctl,
1801 #ifdef CONFIG_COMPAT
1802         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1803 #endif
1804         .mmap           = kvm_vcpu_mmap,
1805         .llseek         = noop_llseek,
1806 };
1807
1808 /*
1809  * Allocates an inode for the vcpu.
1810  */
1811 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1812 {
1813         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1814 }
1815
1816 /*
1817  * Creates some virtual cpus.  Good luck creating more than one.
1818  */
1819 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1820 {
1821         int r;
1822         struct kvm_vcpu *vcpu, *v;
1823
1824         vcpu = kvm_arch_vcpu_create(kvm, id);
1825         if (IS_ERR(vcpu))
1826                 return PTR_ERR(vcpu);
1827
1828         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1829
1830         r = kvm_arch_vcpu_setup(vcpu);
1831         if (r)
1832                 goto vcpu_destroy;
1833
1834         mutex_lock(&kvm->lock);
1835         if (!kvm_vcpu_compatible(vcpu)) {
1836                 r = -EINVAL;
1837                 goto unlock_vcpu_destroy;
1838         }
1839         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1840                 r = -EINVAL;
1841                 goto unlock_vcpu_destroy;
1842         }
1843
1844         kvm_for_each_vcpu(r, v, kvm)
1845                 if (v->vcpu_id == id) {
1846                         r = -EEXIST;
1847                         goto unlock_vcpu_destroy;
1848                 }
1849
1850         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1851
1852         /* Now it's all set up, let userspace reach it */
1853         kvm_get_kvm(kvm);
1854         r = create_vcpu_fd(vcpu);
1855         if (r < 0) {
1856                 kvm_put_kvm(kvm);
1857                 goto unlock_vcpu_destroy;
1858         }
1859
1860         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1861         smp_wmb();
1862         atomic_inc(&kvm->online_vcpus);
1863
1864         mutex_unlock(&kvm->lock);
1865         return r;
1866
1867 unlock_vcpu_destroy:
1868         mutex_unlock(&kvm->lock);
1869 vcpu_destroy:
1870         kvm_arch_vcpu_destroy(vcpu);
1871         return r;
1872 }
1873
1874 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1875 {
1876         if (sigset) {
1877                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1878                 vcpu->sigset_active = 1;
1879                 vcpu->sigset = *sigset;
1880         } else
1881                 vcpu->sigset_active = 0;
1882         return 0;
1883 }
1884
1885 static long kvm_vcpu_ioctl(struct file *filp,
1886                            unsigned int ioctl, unsigned long arg)
1887 {
1888         struct kvm_vcpu *vcpu = filp->private_data;
1889         void __user *argp = (void __user *)arg;
1890         int r;
1891         struct kvm_fpu *fpu = NULL;
1892         struct kvm_sregs *kvm_sregs = NULL;
1893
1894         if (vcpu->kvm->mm != current->mm)
1895                 return -EIO;
1896
1897 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1898         /*
1899          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1900          * so vcpu_load() would break it.
1901          */
1902         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1903                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1904 #endif
1905
1906
1907         vcpu_load(vcpu);
1908         switch (ioctl) {
1909         case KVM_RUN:
1910                 r = -EINVAL;
1911                 if (arg)
1912                         goto out;
1913                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1914                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1915                 break;
1916         case KVM_GET_REGS: {
1917                 struct kvm_regs *kvm_regs;
1918
1919                 r = -ENOMEM;
1920                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1921                 if (!kvm_regs)
1922                         goto out;
1923                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1924                 if (r)
1925                         goto out_free1;
1926                 r = -EFAULT;
1927                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1928                         goto out_free1;
1929                 r = 0;
1930 out_free1:
1931                 kfree(kvm_regs);
1932                 break;
1933         }
1934         case KVM_SET_REGS: {
1935                 struct kvm_regs *kvm_regs;
1936
1937                 r = -ENOMEM;
1938                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1939                 if (IS_ERR(kvm_regs)) {
1940                         r = PTR_ERR(kvm_regs);
1941                         goto out;
1942                 }
1943                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1944                 if (r)
1945                         goto out_free2;
1946                 r = 0;
1947 out_free2:
1948                 kfree(kvm_regs);
1949                 break;
1950         }
1951         case KVM_GET_SREGS: {
1952                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1953                 r = -ENOMEM;
1954                 if (!kvm_sregs)
1955                         goto out;
1956                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1957                 if (r)
1958                         goto out;
1959                 r = -EFAULT;
1960                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1961                         goto out;
1962                 r = 0;
1963                 break;
1964         }
1965         case KVM_SET_SREGS: {
1966                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1967                 if (IS_ERR(kvm_sregs)) {
1968                         r = PTR_ERR(kvm_sregs);
1969                         goto out;
1970                 }
1971                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1972                 if (r)
1973                         goto out;
1974                 r = 0;
1975                 break;
1976         }
1977         case KVM_GET_MP_STATE: {
1978                 struct kvm_mp_state mp_state;
1979
1980                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1981                 if (r)
1982                         goto out;
1983                 r = -EFAULT;
1984                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1985                         goto out;
1986                 r = 0;
1987                 break;
1988         }
1989         case KVM_SET_MP_STATE: {
1990                 struct kvm_mp_state mp_state;
1991
1992                 r = -EFAULT;
1993                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1994                         goto out;
1995                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1996                 if (r)
1997                         goto out;
1998                 r = 0;
1999                 break;
2000         }
2001         case KVM_TRANSLATE: {
2002                 struct kvm_translation tr;
2003
2004                 r = -EFAULT;
2005                 if (copy_from_user(&tr, argp, sizeof tr))
2006                         goto out;
2007                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2008                 if (r)
2009                         goto out;
2010                 r = -EFAULT;
2011                 if (copy_to_user(argp, &tr, sizeof tr))
2012                         goto out;
2013                 r = 0;
2014                 break;
2015         }
2016         case KVM_SET_GUEST_DEBUG: {
2017                 struct kvm_guest_debug dbg;
2018
2019                 r = -EFAULT;
2020                 if (copy_from_user(&dbg, argp, sizeof dbg))
2021                         goto out;
2022                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2023                 if (r)
2024                         goto out;
2025                 r = 0;
2026                 break;
2027         }
2028         case KVM_SET_SIGNAL_MASK: {
2029                 struct kvm_signal_mask __user *sigmask_arg = argp;
2030                 struct kvm_signal_mask kvm_sigmask;
2031                 sigset_t sigset, *p;
2032
2033                 p = NULL;
2034                 if (argp) {
2035                         r = -EFAULT;
2036                         if (copy_from_user(&kvm_sigmask, argp,
2037                                            sizeof kvm_sigmask))
2038                                 goto out;
2039                         r = -EINVAL;
2040                         if (kvm_sigmask.len != sizeof sigset)
2041                                 goto out;
2042                         r = -EFAULT;
2043                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2044                                            sizeof sigset))
2045                                 goto out;
2046                         p = &sigset;
2047                 }
2048                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2049                 break;
2050         }
2051         case KVM_GET_FPU: {
2052                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2053                 r = -ENOMEM;
2054                 if (!fpu)
2055                         goto out;
2056                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2057                 if (r)
2058                         goto out;
2059                 r = -EFAULT;
2060                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2061                         goto out;
2062                 r = 0;
2063                 break;
2064         }
2065         case KVM_SET_FPU: {
2066                 fpu = memdup_user(argp, sizeof(*fpu));
2067                 if (IS_ERR(fpu)) {
2068                         r = PTR_ERR(fpu);
2069                         goto out;
2070                 }
2071                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2072                 if (r)
2073                         goto out;
2074                 r = 0;
2075                 break;
2076         }
2077         default:
2078                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2079         }
2080 out:
2081         vcpu_put(vcpu);
2082         kfree(fpu);
2083         kfree(kvm_sregs);
2084         return r;
2085 }
2086
2087 #ifdef CONFIG_COMPAT
2088 static long kvm_vcpu_compat_ioctl(struct file *filp,
2089                                   unsigned int ioctl, unsigned long arg)
2090 {
2091         struct kvm_vcpu *vcpu = filp->private_data;
2092         void __user *argp = compat_ptr(arg);
2093         int r;
2094
2095         if (vcpu->kvm->mm != current->mm)
2096                 return -EIO;
2097
2098         switch (ioctl) {
2099         case KVM_SET_SIGNAL_MASK: {
2100                 struct kvm_signal_mask __user *sigmask_arg = argp;
2101                 struct kvm_signal_mask kvm_sigmask;
2102                 compat_sigset_t csigset;
2103                 sigset_t sigset;
2104
2105                 if (argp) {
2106                         r = -EFAULT;
2107                         if (copy_from_user(&kvm_sigmask, argp,
2108                                            sizeof kvm_sigmask))
2109                                 goto out;
2110                         r = -EINVAL;
2111                         if (kvm_sigmask.len != sizeof csigset)
2112                                 goto out;
2113                         r = -EFAULT;
2114                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2115                                            sizeof csigset))
2116                                 goto out;
2117                 }
2118                 sigset_from_compat(&sigset, &csigset);
2119                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2120                 break;
2121         }
2122         default:
2123                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2124         }
2125
2126 out:
2127         return r;
2128 }
2129 #endif
2130
2131 static long kvm_vm_ioctl(struct file *filp,
2132                            unsigned int ioctl, unsigned long arg)
2133 {
2134         struct kvm *kvm = filp->private_data;
2135         void __user *argp = (void __user *)arg;
2136         int r;
2137
2138         if (kvm->mm != current->mm)
2139                 return -EIO;
2140         switch (ioctl) {
2141         case KVM_CREATE_VCPU:
2142                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2143                 if (r < 0)
2144                         goto out;
2145                 break;
2146         case KVM_SET_USER_MEMORY_REGION: {
2147                 struct kvm_userspace_memory_region kvm_userspace_mem;
2148
2149                 r = -EFAULT;
2150                 if (copy_from_user(&kvm_userspace_mem, argp,
2151                                                 sizeof kvm_userspace_mem))
2152                         goto out;
2153
2154                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2155                 if (r)
2156                         goto out;
2157                 break;
2158         }
2159         case KVM_GET_DIRTY_LOG: {
2160                 struct kvm_dirty_log log;
2161
2162                 r = -EFAULT;
2163                 if (copy_from_user(&log, argp, sizeof log))
2164                         goto out;
2165                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2166                 if (r)
2167                         goto out;
2168                 break;
2169         }
2170 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2171         case KVM_REGISTER_COALESCED_MMIO: {
2172                 struct kvm_coalesced_mmio_zone zone;
2173                 r = -EFAULT;
2174                 if (copy_from_user(&zone, argp, sizeof zone))
2175                         goto out;
2176                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2177                 if (r)
2178                         goto out;
2179                 r = 0;
2180                 break;
2181         }
2182         case KVM_UNREGISTER_COALESCED_MMIO: {
2183                 struct kvm_coalesced_mmio_zone zone;
2184                 r = -EFAULT;
2185                 if (copy_from_user(&zone, argp, sizeof zone))
2186                         goto out;
2187                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2188                 if (r)
2189                         goto out;
2190                 r = 0;
2191                 break;
2192         }
2193 #endif
2194         case KVM_IRQFD: {
2195                 struct kvm_irqfd data;
2196
2197                 r = -EFAULT;
2198                 if (copy_from_user(&data, argp, sizeof data))
2199                         goto out;
2200                 r = kvm_irqfd(kvm, &data);
2201                 break;
2202         }
2203         case KVM_IOEVENTFD: {
2204                 struct kvm_ioeventfd data;
2205
2206                 r = -EFAULT;
2207                 if (copy_from_user(&data, argp, sizeof data))
2208                         goto out;
2209                 r = kvm_ioeventfd(kvm, &data);
2210                 break;
2211         }
2212 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2213         case KVM_SET_BOOT_CPU_ID:
2214                 r = 0;
2215                 mutex_lock(&kvm->lock);
2216                 if (atomic_read(&kvm->online_vcpus) != 0)
2217                         r = -EBUSY;
2218                 else
2219                         kvm->bsp_vcpu_id = arg;
2220                 mutex_unlock(&kvm->lock);
2221                 break;
2222 #endif
2223 #ifdef CONFIG_HAVE_KVM_MSI
2224         case KVM_SIGNAL_MSI: {
2225                 struct kvm_msi msi;
2226
2227                 r = -EFAULT;
2228                 if (copy_from_user(&msi, argp, sizeof msi))
2229                         goto out;
2230                 r = kvm_send_userspace_msi(kvm, &msi);
2231                 break;
2232         }
2233 #endif
2234 #ifdef __KVM_HAVE_IRQ_LINE
2235         case KVM_IRQ_LINE_STATUS:
2236         case KVM_IRQ_LINE: {
2237                 struct kvm_irq_level irq_event;
2238
2239                 r = -EFAULT;
2240                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2241                         goto out;
2242
2243                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2244                 if (r)
2245                         goto out;
2246
2247                 r = -EFAULT;
2248                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2249                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2250                                 goto out;
2251                 }
2252
2253                 r = 0;
2254                 break;
2255         }
2256 #endif
2257         default:
2258                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2259                 if (r == -ENOTTY)
2260                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2261         }
2262 out:
2263         return r;
2264 }
2265
2266 #ifdef CONFIG_COMPAT
2267 struct compat_kvm_dirty_log {
2268         __u32 slot;
2269         __u32 padding1;
2270         union {
2271                 compat_uptr_t dirty_bitmap; /* one bit per page */
2272                 __u64 padding2;
2273         };
2274 };
2275
2276 static long kvm_vm_compat_ioctl(struct file *filp,
2277                            unsigned int ioctl, unsigned long arg)
2278 {
2279         struct kvm *kvm = filp->private_data;
2280         int r;
2281
2282         if (kvm->mm != current->mm)
2283                 return -EIO;
2284         switch (ioctl) {
2285         case KVM_GET_DIRTY_LOG: {
2286                 struct compat_kvm_dirty_log compat_log;
2287                 struct kvm_dirty_log log;
2288
2289                 r = -EFAULT;
2290                 if (copy_from_user(&compat_log, (void __user *)arg,
2291                                    sizeof(compat_log)))
2292                         goto out;
2293                 log.slot         = compat_log.slot;
2294                 log.padding1     = compat_log.padding1;
2295                 log.padding2     = compat_log.padding2;
2296                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2297
2298                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2299                 if (r)
2300                         goto out;
2301                 break;
2302         }
2303         default:
2304                 r = kvm_vm_ioctl(filp, ioctl, arg);
2305         }
2306
2307 out:
2308         return r;
2309 }
2310 #endif
2311
2312 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2313 {
2314         struct page *page[1];
2315         unsigned long addr;
2316         int npages;
2317         gfn_t gfn = vmf->pgoff;
2318         struct kvm *kvm = vma->vm_file->private_data;
2319
2320         addr = gfn_to_hva(kvm, gfn);
2321         if (kvm_is_error_hva(addr))
2322                 return VM_FAULT_SIGBUS;
2323
2324         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2325                                 NULL);
2326         if (unlikely(npages != 1))
2327                 return VM_FAULT_SIGBUS;
2328
2329         vmf->page = page[0];
2330         return 0;
2331 }
2332
2333 static const struct vm_operations_struct kvm_vm_vm_ops = {
2334         .fault = kvm_vm_fault,
2335 };
2336
2337 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2338 {
2339         vma->vm_ops = &kvm_vm_vm_ops;
2340         return 0;
2341 }
2342
2343 static struct file_operations kvm_vm_fops = {
2344         .release        = kvm_vm_release,
2345         .unlocked_ioctl = kvm_vm_ioctl,
2346 #ifdef CONFIG_COMPAT
2347         .compat_ioctl   = kvm_vm_compat_ioctl,
2348 #endif
2349         .mmap           = kvm_vm_mmap,
2350         .llseek         = noop_llseek,
2351 };
2352
2353 static int kvm_dev_ioctl_create_vm(unsigned long type)
2354 {
2355         int r;
2356         struct kvm *kvm;
2357
2358         kvm = kvm_create_vm(type);
2359         if (IS_ERR(kvm))
2360                 return PTR_ERR(kvm);
2361 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2362         r = kvm_coalesced_mmio_init(kvm);
2363         if (r < 0) {
2364                 kvm_put_kvm(kvm);
2365                 return r;
2366         }
2367 #endif
2368         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2369         if (r < 0)
2370                 kvm_put_kvm(kvm);
2371
2372         return r;
2373 }
2374
2375 static long kvm_dev_ioctl_check_extension_generic(long arg)
2376 {
2377         switch (arg) {
2378         case KVM_CAP_USER_MEMORY:
2379         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2380         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2381 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2382         case KVM_CAP_SET_BOOT_CPU_ID:
2383 #endif
2384         case KVM_CAP_INTERNAL_ERROR_DATA:
2385 #ifdef CONFIG_HAVE_KVM_MSI
2386         case KVM_CAP_SIGNAL_MSI:
2387 #endif
2388                 return 1;
2389 #ifdef KVM_CAP_IRQ_ROUTING
2390         case KVM_CAP_IRQ_ROUTING:
2391                 return KVM_MAX_IRQ_ROUTES;
2392 #endif
2393         default:
2394                 break;
2395         }
2396         return kvm_dev_ioctl_check_extension(arg);
2397 }
2398
2399 static long kvm_dev_ioctl(struct file *filp,
2400                           unsigned int ioctl, unsigned long arg)
2401 {
2402         long r = -EINVAL;
2403
2404         switch (ioctl) {
2405         case KVM_GET_API_VERSION:
2406                 r = -EINVAL;
2407                 if (arg)
2408                         goto out;
2409                 r = KVM_API_VERSION;
2410                 break;
2411         case KVM_CREATE_VM:
2412                 r = kvm_dev_ioctl_create_vm(arg);
2413                 break;
2414         case KVM_CHECK_EXTENSION:
2415                 r = kvm_dev_ioctl_check_extension_generic(arg);
2416                 break;
2417         case KVM_GET_VCPU_MMAP_SIZE:
2418                 r = -EINVAL;
2419                 if (arg)
2420                         goto out;
2421                 r = PAGE_SIZE;     /* struct kvm_run */
2422 #ifdef CONFIG_X86
2423                 r += PAGE_SIZE;    /* pio data page */
2424 #endif
2425 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2426                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2427 #endif
2428                 break;
2429         case KVM_TRACE_ENABLE:
2430         case KVM_TRACE_PAUSE:
2431         case KVM_TRACE_DISABLE:
2432                 r = -EOPNOTSUPP;
2433                 break;
2434         default:
2435                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2436         }
2437 out:
2438         return r;
2439 }
2440
2441 static struct file_operations kvm_chardev_ops = {
2442         .unlocked_ioctl = kvm_dev_ioctl,
2443         .compat_ioctl   = kvm_dev_ioctl,
2444         .llseek         = noop_llseek,
2445 };
2446
2447 static struct miscdevice kvm_dev = {
2448         KVM_MINOR,
2449         "kvm",
2450         &kvm_chardev_ops,
2451 };
2452
2453 static void hardware_enable_nolock(void *junk)
2454 {
2455         int cpu = raw_smp_processor_id();
2456         int r;
2457
2458         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2459                 return;
2460
2461         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2462
2463         r = kvm_arch_hardware_enable(NULL);
2464
2465         if (r) {
2466                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2467                 atomic_inc(&hardware_enable_failed);
2468                 printk(KERN_INFO "kvm: enabling virtualization on "
2469                                  "CPU%d failed\n", cpu);
2470         }
2471 }
2472
2473 static void hardware_enable(void *junk)
2474 {
2475         raw_spin_lock(&kvm_lock);
2476         hardware_enable_nolock(junk);
2477         raw_spin_unlock(&kvm_lock);
2478 }
2479
2480 static void hardware_disable_nolock(void *junk)
2481 {
2482         int cpu = raw_smp_processor_id();
2483
2484         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2485                 return;
2486         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2487         kvm_arch_hardware_disable(NULL);
2488 }
2489
2490 static void hardware_disable(void *junk)
2491 {
2492         raw_spin_lock(&kvm_lock);
2493         hardware_disable_nolock(junk);
2494         raw_spin_unlock(&kvm_lock);
2495 }
2496
2497 static void hardware_disable_all_nolock(void)
2498 {
2499         BUG_ON(!kvm_usage_count);
2500
2501         kvm_usage_count--;
2502         if (!kvm_usage_count)
2503                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2504 }
2505
2506 static void hardware_disable_all(void)
2507 {
2508         raw_spin_lock(&kvm_lock);
2509         hardware_disable_all_nolock();
2510         raw_spin_unlock(&kvm_lock);
2511 }
2512
2513 static int hardware_enable_all(void)
2514 {
2515         int r = 0;
2516
2517         raw_spin_lock(&kvm_lock);
2518
2519         kvm_usage_count++;
2520         if (kvm_usage_count == 1) {
2521                 atomic_set(&hardware_enable_failed, 0);
2522                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2523
2524                 if (atomic_read(&hardware_enable_failed)) {
2525                         hardware_disable_all_nolock();
2526                         r = -EBUSY;
2527                 }
2528         }
2529
2530         raw_spin_unlock(&kvm_lock);
2531
2532         return r;
2533 }
2534
2535 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2536                            void *v)
2537 {
2538         int cpu = (long)v;
2539
2540         if (!kvm_usage_count)
2541                 return NOTIFY_OK;
2542
2543         val &= ~CPU_TASKS_FROZEN;
2544         switch (val) {
2545         case CPU_DYING:
2546                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2547                        cpu);
2548                 hardware_disable(NULL);
2549                 break;
2550         case CPU_STARTING:
2551                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2552                        cpu);
2553                 hardware_enable(NULL);
2554                 break;
2555         }
2556         return NOTIFY_OK;
2557 }
2558
2559
2560 asmlinkage void kvm_spurious_fault(void)
2561 {
2562         /* Fault while not rebooting.  We want the trace. */
2563         BUG();
2564 }
2565 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2566
2567 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2568                       void *v)
2569 {
2570         /*
2571          * Some (well, at least mine) BIOSes hang on reboot if
2572          * in vmx root mode.
2573          *
2574          * And Intel TXT required VMX off for all cpu when system shutdown.
2575          */
2576         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2577         kvm_rebooting = true;
2578         on_each_cpu(hardware_disable_nolock, NULL, 1);
2579         return NOTIFY_OK;
2580 }
2581
2582 static struct notifier_block kvm_reboot_notifier = {
2583         .notifier_call = kvm_reboot,
2584         .priority = 0,
2585 };
2586
2587 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2588 {
2589         int i;
2590
2591         for (i = 0; i < bus->dev_count; i++) {
2592                 struct kvm_io_device *pos = bus->range[i].dev;
2593
2594                 kvm_iodevice_destructor(pos);
2595         }
2596         kfree(bus);
2597 }
2598
2599 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2600 {
2601         const struct kvm_io_range *r1 = p1;
2602         const struct kvm_io_range *r2 = p2;
2603
2604         if (r1->addr < r2->addr)
2605                 return -1;
2606         if (r1->addr + r1->len > r2->addr + r2->len)
2607                 return 1;
2608         return 0;
2609 }
2610
2611 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2612                           gpa_t addr, int len)
2613 {
2614         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2615                 .addr = addr,
2616                 .len = len,
2617                 .dev = dev,
2618         };
2619
2620         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2621                 kvm_io_bus_sort_cmp, NULL);
2622
2623         return 0;
2624 }
2625
2626 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2627                              gpa_t addr, int len)
2628 {
2629         struct kvm_io_range *range, key;
2630         int off;
2631
2632         key = (struct kvm_io_range) {
2633                 .addr = addr,
2634                 .len = len,
2635         };
2636
2637         range = bsearch(&key, bus->range, bus->dev_count,
2638                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2639         if (range == NULL)
2640                 return -ENOENT;
2641
2642         off = range - bus->range;
2643
2644         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2645                 off--;
2646
2647         return off;
2648 }
2649
2650 /* kvm_io_bus_write - called under kvm->slots_lock */
2651 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2652                      int len, const void *val)
2653 {
2654         int idx;
2655         struct kvm_io_bus *bus;
2656         struct kvm_io_range range;
2657
2658         range = (struct kvm_io_range) {
2659                 .addr = addr,
2660                 .len = len,
2661         };
2662
2663         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2664         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2665         if (idx < 0)
2666                 return -EOPNOTSUPP;
2667
2668         while (idx < bus->dev_count &&
2669                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2670                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2671                         return 0;
2672                 idx++;
2673         }
2674
2675         return -EOPNOTSUPP;
2676 }
2677
2678 /* kvm_io_bus_read - called under kvm->slots_lock */
2679 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2680                     int len, void *val)
2681 {
2682         int idx;
2683         struct kvm_io_bus *bus;
2684         struct kvm_io_range range;
2685
2686         range = (struct kvm_io_range) {
2687                 .addr = addr,
2688                 .len = len,
2689         };
2690
2691         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2692         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2693         if (idx < 0)
2694                 return -EOPNOTSUPP;
2695
2696         while (idx < bus->dev_count &&
2697                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2698                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2699                         return 0;
2700                 idx++;
2701         }
2702
2703         return -EOPNOTSUPP;
2704 }
2705
2706 /* Caller must hold slots_lock. */
2707 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2708                             int len, struct kvm_io_device *dev)
2709 {
2710         struct kvm_io_bus *new_bus, *bus;
2711
2712         bus = kvm->buses[bus_idx];
2713         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2714                 return -ENOSPC;
2715
2716         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2717                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2718         if (!new_bus)
2719                 return -ENOMEM;
2720         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2721                sizeof(struct kvm_io_range)));
2722         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2723         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2724         synchronize_srcu_expedited(&kvm->srcu);
2725         kfree(bus);
2726
2727         return 0;
2728 }
2729
2730 /* Caller must hold slots_lock. */
2731 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2732                               struct kvm_io_device *dev)
2733 {
2734         int i, r;
2735         struct kvm_io_bus *new_bus, *bus;
2736
2737         bus = kvm->buses[bus_idx];
2738         r = -ENOENT;
2739         for (i = 0; i < bus->dev_count; i++)
2740                 if (bus->range[i].dev == dev) {
2741                         r = 0;
2742                         break;
2743                 }
2744
2745         if (r)
2746                 return r;
2747
2748         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2749                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2750         if (!new_bus)
2751                 return -ENOMEM;
2752
2753         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2754         new_bus->dev_count--;
2755         memcpy(new_bus->range + i, bus->range + i + 1,
2756                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2757
2758         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2759         synchronize_srcu_expedited(&kvm->srcu);
2760         kfree(bus);
2761         return r;
2762 }
2763
2764 static struct notifier_block kvm_cpu_notifier = {
2765         .notifier_call = kvm_cpu_hotplug,
2766 };
2767
2768 static int vm_stat_get(void *_offset, u64 *val)
2769 {
2770         unsigned offset = (long)_offset;
2771         struct kvm *kvm;
2772
2773         *val = 0;
2774         raw_spin_lock(&kvm_lock);
2775         list_for_each_entry(kvm, &vm_list, vm_list)
2776                 *val += *(u32 *)((void *)kvm + offset);
2777         raw_spin_unlock(&kvm_lock);
2778         return 0;
2779 }
2780
2781 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2782
2783 static int vcpu_stat_get(void *_offset, u64 *val)
2784 {
2785         unsigned offset = (long)_offset;
2786         struct kvm *kvm;
2787         struct kvm_vcpu *vcpu;
2788         int i;
2789
2790         *val = 0;
2791         raw_spin_lock(&kvm_lock);
2792         list_for_each_entry(kvm, &vm_list, vm_list)
2793                 kvm_for_each_vcpu(i, vcpu, kvm)
2794                         *val += *(u32 *)((void *)vcpu + offset);
2795
2796         raw_spin_unlock(&kvm_lock);
2797         return 0;
2798 }
2799
2800 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2801
2802 static const struct file_operations *stat_fops[] = {
2803         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2804         [KVM_STAT_VM]   = &vm_stat_fops,
2805 };
2806
2807 static int kvm_init_debug(void)
2808 {
2809         int r = -EFAULT;
2810         struct kvm_stats_debugfs_item *p;
2811
2812         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2813         if (kvm_debugfs_dir == NULL)
2814                 goto out;
2815
2816         for (p = debugfs_entries; p->name; ++p) {
2817                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2818                                                 (void *)(long)p->offset,
2819                                                 stat_fops[p->kind]);
2820                 if (p->dentry == NULL)
2821                         goto out_dir;
2822         }
2823
2824         return 0;
2825
2826 out_dir:
2827         debugfs_remove_recursive(kvm_debugfs_dir);
2828 out:
2829         return r;
2830 }
2831
2832 static void kvm_exit_debug(void)
2833 {
2834         struct kvm_stats_debugfs_item *p;
2835
2836         for (p = debugfs_entries; p->name; ++p)
2837                 debugfs_remove(p->dentry);
2838         debugfs_remove(kvm_debugfs_dir);
2839 }
2840
2841 static int kvm_suspend(void)
2842 {
2843         if (kvm_usage_count)
2844                 hardware_disable_nolock(NULL);
2845         return 0;
2846 }
2847
2848 static void kvm_resume(void)
2849 {
2850         if (kvm_usage_count) {
2851                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2852                 hardware_enable_nolock(NULL);
2853         }
2854 }
2855
2856 static struct syscore_ops kvm_syscore_ops = {
2857         .suspend = kvm_suspend,
2858         .resume = kvm_resume,
2859 };
2860
2861 static inline
2862 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2863 {
2864         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2865 }
2866
2867 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2868 {
2869         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2870
2871         kvm_arch_vcpu_load(vcpu, cpu);
2872 }
2873
2874 static void kvm_sched_out(struct preempt_notifier *pn,
2875                           struct task_struct *next)
2876 {
2877         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2878
2879         kvm_arch_vcpu_put(vcpu);
2880 }
2881
2882 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2883                   struct module *module)
2884 {
2885         int r;
2886         int cpu;
2887
2888         r = kvm_arch_init(opaque);
2889         if (r)
2890                 goto out_fail;
2891
2892         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2893                 r = -ENOMEM;
2894                 goto out_free_0;
2895         }
2896
2897         r = kvm_arch_hardware_setup();
2898         if (r < 0)
2899                 goto out_free_0a;
2900
2901         for_each_online_cpu(cpu) {
2902                 smp_call_function_single(cpu,
2903                                 kvm_arch_check_processor_compat,
2904                                 &r, 1);
2905                 if (r < 0)
2906                         goto out_free_1;
2907         }
2908
2909         r = register_cpu_notifier(&kvm_cpu_notifier);
2910         if (r)
2911                 goto out_free_2;
2912         register_reboot_notifier(&kvm_reboot_notifier);
2913
2914         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2915         if (!vcpu_align)
2916                 vcpu_align = __alignof__(struct kvm_vcpu);
2917         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2918                                            0, NULL);
2919         if (!kvm_vcpu_cache) {
2920                 r = -ENOMEM;
2921                 goto out_free_3;
2922         }
2923
2924         r = kvm_async_pf_init();
2925         if (r)
2926                 goto out_free;
2927
2928         kvm_chardev_ops.owner = module;
2929         kvm_vm_fops.owner = module;
2930         kvm_vcpu_fops.owner = module;
2931
2932         r = misc_register(&kvm_dev);
2933         if (r) {
2934                 printk(KERN_ERR "kvm: misc device register failed\n");
2935                 goto out_unreg;
2936         }
2937
2938         register_syscore_ops(&kvm_syscore_ops);
2939
2940         kvm_preempt_ops.sched_in = kvm_sched_in;
2941         kvm_preempt_ops.sched_out = kvm_sched_out;
2942
2943         r = kvm_init_debug();
2944         if (r) {
2945                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2946                 goto out_undebugfs;
2947         }
2948
2949         return 0;
2950
2951 out_undebugfs:
2952         unregister_syscore_ops(&kvm_syscore_ops);
2953 out_unreg:
2954         kvm_async_pf_deinit();
2955 out_free:
2956         kmem_cache_destroy(kvm_vcpu_cache);
2957 out_free_3:
2958         unregister_reboot_notifier(&kvm_reboot_notifier);
2959         unregister_cpu_notifier(&kvm_cpu_notifier);
2960 out_free_2:
2961 out_free_1:
2962         kvm_arch_hardware_unsetup();
2963 out_free_0a:
2964         free_cpumask_var(cpus_hardware_enabled);
2965 out_free_0:
2966         kvm_arch_exit();
2967 out_fail:
2968         return r;
2969 }
2970 EXPORT_SYMBOL_GPL(kvm_init);
2971
2972 void kvm_exit(void)
2973 {
2974         kvm_exit_debug();
2975         misc_deregister(&kvm_dev);
2976         kmem_cache_destroy(kvm_vcpu_cache);
2977         kvm_async_pf_deinit();
2978         unregister_syscore_ops(&kvm_syscore_ops);
2979         unregister_reboot_notifier(&kvm_reboot_notifier);
2980         unregister_cpu_notifier(&kvm_cpu_notifier);
2981         on_each_cpu(hardware_disable_nolock, NULL, 1);
2982         kvm_arch_hardware_unsetup();
2983         kvm_arch_exit();
2984         free_cpumask_var(cpus_hardware_enabled);
2985 }
2986 EXPORT_SYMBOL_GPL(kvm_exit);