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 (is_error_pfn(pfn))
 106                 return false;
 107
 108         if (pfn_valid(pfn)) {
 109                 int reserved;
 110                 struct page *tail = pfn_to_page(pfn);
 111                 struct page *head = compound_trans_head(tail);
 112                 reserved = PageReserved(head);
 113                 if (head != tail) {
 114                         /*
 115                          * "head" is not a dangling pointer
 116                          * (compound_trans_head takes care of that)
 117                          * but the hugepage may have been splitted
 118                          * from under us (and we may not hold a
 119                          * reference count on the head page so it can
 120                          * be reused before we run PageReferenced), so
 121                          * we've to check PageTail before returning
 122                          * what we just read.
 123                          */
 124                         smp_rmb();
 125                         if (PageTail(tail))
 126                                 return reserved;
 127                 }
 128                 return PageReserved(tail);
 129         }
 130
 131         return true;
 132 }
 133
 134 /*
 135  * Switches to specified vcpu, until a matching vcpu_put()
 136  */
 137 void vcpu_load(struct kvm_vcpu *vcpu)
 138 {
 139         int cpu;
 140
 141         mutex_lock(&vcpu->mutex);
 142         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 143                 /* The thread running this VCPU changed. */
 144                 struct pid *oldpid = vcpu->pid;
 145                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 146                 rcu_assign_pointer(vcpu->pid, newpid);
 147                 synchronize_rcu();
 148                 put_pid(oldpid);
 149         }
 150         cpu = get_cpu();
 151         preempt_notifier_register(&vcpu->preempt_notifier);
 152         kvm_arch_vcpu_load(vcpu, cpu);
 153         put_cpu();
 154 }
 155
 156 void vcpu_put(struct kvm_vcpu *vcpu)
 157 {
 158         preempt_disable();
 159         kvm_arch_vcpu_put(vcpu);
 160         preempt_notifier_unregister(&vcpu->preempt_notifier);
 161         preempt_enable();
 162         mutex_unlock(&vcpu->mutex);
 163 }
 164
 165 static void ack_flush(void *_completed)
 166 {
 167 }
 168
 169 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 170 {
 171         int i, cpu, me;
 172         cpumask_var_t cpus;
 173         bool called = true;
 174         struct kvm_vcpu *vcpu;
 175
 176         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 177
 178         me = get_cpu();
 179         kvm_for_each_vcpu(i, vcpu, kvm) {
 180                 kvm_make_request(req, vcpu);
 181                 cpu = vcpu->cpu;
 182
 183                 /* Set ->requests bit before we read ->mode */
 184                 smp_mb();
 185
 186                 if (cpus != NULL && cpu != -1 && cpu != me &&
 187                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 188                         cpumask_set_cpu(cpu, cpus);
 189         }
 190         if (unlikely(cpus == NULL))
 191                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 192         else if (!cpumask_empty(cpus))
 193                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 194         else
 195                 called = false;
 196         put_cpu();
 197         free_cpumask_var(cpus);
 198         return called;
 199 }
 200
 201 void kvm_flush_remote_tlbs(struct kvm *kvm)
 202 {
 203         long dirty_count = kvm->tlbs_dirty;
 204
 205         smp_mb();
 206         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 207                 ++kvm->stat.remote_tlb_flush;
 208         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 209 }
 210
 211 void kvm_reload_remote_mmus(struct kvm *kvm)
 212 {
 213         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 214 }
 215
 216 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 217 {
 218         struct page *page;
 219         int r;
 220
 221         mutex_init(&vcpu->mutex);
 222         vcpu->cpu = -1;
 223         vcpu->kvm = kvm;
 224         vcpu->vcpu_id = id;
 225         vcpu->pid = NULL;
 226         init_waitqueue_head(&vcpu->wq);
 227         kvm_async_pf_vcpu_init(vcpu);
 228
 229         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 230         if (!page) {
 231                 r = -ENOMEM;
 232                 goto fail;
 233         }
 234         vcpu->run = page_address(page);
 235
 236         kvm_vcpu_set_in_spin_loop(vcpu, false);
 237         kvm_vcpu_set_dy_eligible(vcpu, false);
 238
 239         r = kvm_arch_vcpu_init(vcpu);
 240         if (r < 0)
 241                 goto fail_free_run;
 242         return 0;
 243
 244 fail_free_run:
 245         free_page((unsigned long)vcpu->run);
 246 fail:
 247         return r;
 248 }
 249 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 250
 251 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 252 {
 253         put_pid(vcpu->pid);
 254         kvm_arch_vcpu_uninit(vcpu);
 255         free_page((unsigned long)vcpu->run);
 256 }
 257 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 258
 259 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 260 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 261 {
 262         return container_of(mn, struct kvm, mmu_notifier);
 263 }
 264
 265 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 266                                              struct mm_struct *mm,
 267                                              unsigned long address)
 268 {
 269         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 270         int need_tlb_flush, idx;
 271
 272         /*
 273          * When ->invalidate_page runs, the linux pte has been zapped
 274          * already but the page is still allocated until
 275          * ->invalidate_page returns. So if we increase the sequence
 276          * here the kvm page fault will notice if the spte can't be
 277          * established because the page is going to be freed. If
 278          * instead the kvm page fault establishes the spte before
 279          * ->invalidate_page runs, kvm_unmap_hva will release it
 280          * before returning.
 281          *
 282          * The sequence increase only need to be seen at spin_unlock
 283          * time, and not at spin_lock time.
 284          *
 285          * Increasing the sequence after the spin_unlock would be
 286          * unsafe because the kvm page fault could then establish the
 287          * pte after kvm_unmap_hva returned, without noticing the page
 288          * is going to be freed.
 289          */
 290         idx = srcu_read_lock(&kvm->srcu);
 291         spin_lock(&kvm->mmu_lock);
 292
 293         kvm->mmu_notifier_seq++;
 294         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 295         /* we've to flush the tlb before the pages can be freed */
 296         if (need_tlb_flush)
 297                 kvm_flush_remote_tlbs(kvm);
 298
 299         spin_unlock(&kvm->mmu_lock);
 300         srcu_read_unlock(&kvm->srcu, idx);
 301 }
 302
 303 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 304                                         struct mm_struct *mm,
 305                                         unsigned long address,
 306                                         pte_t pte)
 307 {
 308         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 309         int idx;
 310
 311         idx = srcu_read_lock(&kvm->srcu);
 312         spin_lock(&kvm->mmu_lock);
 313         kvm->mmu_notifier_seq++;
 314         kvm_set_spte_hva(kvm, address, pte);
 315         spin_unlock(&kvm->mmu_lock);
 316         srcu_read_unlock(&kvm->srcu, idx);
 317 }
 318
 319 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 320                                                     struct mm_struct *mm,
 321                                                     unsigned long start,
 322                                                     unsigned long end)
 323 {
 324         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 325         int need_tlb_flush = 0, idx;
 326
 327         idx = srcu_read_lock(&kvm->srcu);
 328         spin_lock(&kvm->mmu_lock);
 329         /*
 330          * The count increase must become visible at unlock time as no
 331          * spte can be established without taking the mmu_lock and
 332          * count is also read inside the mmu_lock critical section.
 333          */
 334         kvm->mmu_notifier_count++;
 335         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 336         need_tlb_flush |= kvm->tlbs_dirty;
 337         /* we've to flush the tlb before the pages can be freed */
 338         if (need_tlb_flush)
 339                 kvm_flush_remote_tlbs(kvm);
 340
 341         spin_unlock(&kvm->mmu_lock);
 342         srcu_read_unlock(&kvm->srcu, idx);
 343 }
 344
 345 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 346                                                   struct mm_struct *mm,
 347                                                   unsigned long start,
 348                                                   unsigned long end)
 349 {
 350         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 351
 352         spin_lock(&kvm->mmu_lock);
 353         /*
 354          * This sequence increase will notify the kvm page fault that
 355          * the page that is going to be mapped in the spte could have
 356          * been freed.
 357          */
 358         kvm->mmu_notifier_seq++;
 359         smp_wmb();
 360         /*
 361          * The above sequence increase must be visible before the
 362          * below count decrease, which is ensured by the smp_wmb above
 363          * in conjunction with the smp_rmb in mmu_notifier_retry().
 364          */
 365         kvm->mmu_notifier_count--;
 366         spin_unlock(&kvm->mmu_lock);
 367
 368         BUG_ON(kvm->mmu_notifier_count < 0);
 369 }
 370
 371 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 372                                               struct mm_struct *mm,
 373                                               unsigned long address)
 374 {
 375         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 376         int young, idx;
 377
 378         idx = srcu_read_lock(&kvm->srcu);
 379         spin_lock(&kvm->mmu_lock);
 380
 381         young = kvm_age_hva(kvm, address);
 382         if (young)
 383                 kvm_flush_remote_tlbs(kvm);
 384
 385         spin_unlock(&kvm->mmu_lock);
 386         srcu_read_unlock(&kvm->srcu, idx);
 387
 388         return young;
 389 }
 390
 391 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 392                                        struct mm_struct *mm,
 393                                        unsigned long address)
 394 {
 395         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 396         int young, idx;
 397
 398         idx = srcu_read_lock(&kvm->srcu);
 399         spin_lock(&kvm->mmu_lock);
 400         young = kvm_test_age_hva(kvm, address);
 401         spin_unlock(&kvm->mmu_lock);
 402         srcu_read_unlock(&kvm->srcu, idx);
 403
 404         return young;
 405 }
 406
 407 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 408                                      struct mm_struct *mm)
 409 {
 410         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 411         int idx;
 412
 413         idx = srcu_read_lock(&kvm->srcu);
 414         kvm_arch_flush_shadow(kvm);
 415         srcu_read_unlock(&kvm->srcu, idx);
 416 }
 417
 418 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 419         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 420         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 421         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 422         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 423         .test_young             = kvm_mmu_notifier_test_young,
 424         .change_pte             = kvm_mmu_notifier_change_pte,
 425         .release                = kvm_mmu_notifier_release,
 426 };
 427
 428 static int kvm_init_mmu_notifier(struct kvm *kvm)
 429 {
 430         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 431         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 432 }
 433
 434 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 435
 436 static int kvm_init_mmu_notifier(struct kvm *kvm)
 437 {
 438         return 0;
 439 }
 440
 441 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 442
 443 static void kvm_init_memslots_id(struct kvm *kvm)
 444 {
 445         int i;
 446         struct kvm_memslots *slots = kvm->memslots;
 447
 448         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 449                 slots->id_to_index[i] = slots->memslots[i].id = i;
 450 }
 451
 452 static struct kvm *kvm_create_vm(unsigned long type)
 453 {
 454         int r, i;
 455         struct kvm *kvm = kvm_arch_alloc_vm();
 456
 457         if (!kvm)
 458                 return ERR_PTR(-ENOMEM);
 459
 460         r = kvm_arch_init_vm(kvm, type);
 461         if (r)
 462                 goto out_err_nodisable;
 463
 464         r = hardware_enable_all();
 465         if (r)
 466                 goto out_err_nodisable;
 467
 468 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 469         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 470         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 471 #endif
 472
 473         r = -ENOMEM;
 474         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 475         if (!kvm->memslots)
 476                 goto out_err_nosrcu;
 477         kvm_init_memslots_id(kvm);
 478         if (init_srcu_struct(&kvm->srcu))
 479                 goto out_err_nosrcu;
 480         for (i = 0; i < KVM_NR_BUSES; i++) {
 481                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 482                                         GFP_KERNEL);
 483                 if (!kvm->buses[i])
 484                         goto out_err;
 485         }
 486
 487         spin_lock_init(&kvm->mmu_lock);
 488         kvm->mm = current->mm;
 489         atomic_inc(&kvm->mm->mm_count);
 490         kvm_eventfd_init(kvm);
 491         mutex_init(&kvm->lock);
 492         mutex_init(&kvm->irq_lock);
 493         mutex_init(&kvm->slots_lock);
 494         atomic_set(&kvm->users_count, 1);
 495
 496         r = kvm_init_mmu_notifier(kvm);
 497         if (r)
 498                 goto out_err;
 499
 500         raw_spin_lock(&kvm_lock);
 501         list_add(&kvm->vm_list, &vm_list);
 502         raw_spin_unlock(&kvm_lock);
 503
 504         return kvm;
 505
 506 out_err:
 507         cleanup_srcu_struct(&kvm->srcu);
 508 out_err_nosrcu:
 509         hardware_disable_all();
 510 out_err_nodisable:
 511         for (i = 0; i < KVM_NR_BUSES; i++)
 512                 kfree(kvm->buses[i]);
 513         kfree(kvm->memslots);
 514         kvm_arch_free_vm(kvm);
 515         return ERR_PTR(r);
 516 }
 517
 518 /*
 519  * Avoid using vmalloc for a small buffer.
 520  * Should not be used when the size is statically known.
 521  */
 522 void *kvm_kvzalloc(unsigned long size)
 523 {
 524         if (size > PAGE_SIZE)
 525                 return vzalloc(size);
 526         else
 527                 return kzalloc(size, GFP_KERNEL);
 528 }
 529
 530 void kvm_kvfree(const void *addr)
 531 {
 532         if (is_vmalloc_addr(addr))
 533                 vfree(addr);
 534         else
 535                 kfree(addr);
 536 }
 537
 538 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 539 {
 540         if (!memslot->dirty_bitmap)
 541                 return;
 542
 543         kvm_kvfree(memslot->dirty_bitmap);
 544         memslot->dirty_bitmap = NULL;
 545 }
 546
 547 /*
 548  * Free any memory in @free but not in @dont.
 549  */
 550 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 551                                   struct kvm_memory_slot *dont)
 552 {
 553         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 554                 kvm_destroy_dirty_bitmap(free);
 555
 556         kvm_arch_free_memslot(free, dont);
 557
 558         free->npages = 0;
 559 }
 560
 561 void kvm_free_physmem(struct kvm *kvm)
 562 {
 563         struct kvm_memslots *slots = kvm->memslots;
 564         struct kvm_memory_slot *memslot;
 565
 566         kvm_for_each_memslot(memslot, slots)
 567                 kvm_free_physmem_slot(memslot, NULL);
 568
 569         kfree(kvm->memslots);
 570 }
 571
 572 static void kvm_destroy_vm(struct kvm *kvm)
 573 {
 574         int i;
 575         struct mm_struct *mm = kvm->mm;
 576
 577         kvm_arch_sync_events(kvm);
 578         raw_spin_lock(&kvm_lock);
 579         list_del(&kvm->vm_list);
 580         raw_spin_unlock(&kvm_lock);
 581         kvm_free_irq_routing(kvm);
 582         for (i = 0; i < KVM_NR_BUSES; i++)
 583                 kvm_io_bus_destroy(kvm->buses[i]);
 584         kvm_coalesced_mmio_free(kvm);
 585 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 586         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 587 #else
 588         kvm_arch_flush_shadow(kvm);
 589 #endif
 590         kvm_arch_destroy_vm(kvm);
 591         kvm_free_physmem(kvm);
 592         cleanup_srcu_struct(&kvm->srcu);
 593         kvm_arch_free_vm(kvm);
 594         hardware_disable_all();
 595         mmdrop(mm);
 596 }
 597
 598 void kvm_get_kvm(struct kvm *kvm)
 599 {
 600         atomic_inc(&kvm->users_count);
 601 }
 602 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 603
 604 void kvm_put_kvm(struct kvm *kvm)
 605 {
 606         if (atomic_dec_and_test(&kvm->users_count))
 607                 kvm_destroy_vm(kvm);
 608 }
 609 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 610
 611
 612 static int kvm_vm_release(struct inode *inode, struct file *filp)
 613 {
 614         struct kvm *kvm = filp->private_data;
 615
 616         kvm_irqfd_release(kvm);
 617
 618         kvm_put_kvm(kvm);
 619         return 0;
 620 }
 621
 622 /*
 623  * Allocation size is twice as large as the actual dirty bitmap size.
 624  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 625  */
 626 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 627 {
 628 #ifndef CONFIG_S390
 629         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 630
 631         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 632         if (!memslot->dirty_bitmap)
 633                 return -ENOMEM;
 634
 635 #endif /* !CONFIG_S390 */
 636         return 0;
 637 }
 638
 639 static int cmp_memslot(const void *slot1, const void *slot2)
 640 {
 641         struct kvm_memory_slot *s1, *s2;
 642
 643         s1 = (struct kvm_memory_slot *)slot1;
 644         s2 = (struct kvm_memory_slot *)slot2;
 645
 646         if (s1->npages < s2->npages)
 647                 return 1;
 648         if (s1->npages > s2->npages)
 649                 return -1;
 650
 651         return 0;
 652 }
 653
 654 /*
 655  * Sort the memslots base on its size, so the larger slots
 656  * will get better fit.
 657  */
 658 static void sort_memslots(struct kvm_memslots *slots)
 659 {
 660         int i;
 661
 662         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 663               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 664
 665         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 666                 slots->id_to_index[slots->memslots[i].id] = i;
 667 }
 668
 669 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
 670 {
 671         if (new) {
 672                 int id = new->id;
 673                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 674                 unsigned long npages = old->npages;
 675
 676                 *old = *new;
 677                 if (new->npages != npages)
 678                         sort_memslots(slots);
 679         }
 680
 681         slots->generation++;
 682 }
 683
 684 /*
 685  * Allocate some memory and give it an address in the guest physical address
 686  * space.
 687  *
 688  * Discontiguous memory is allowed, mostly for framebuffers.
 689  *
 690  * Must be called holding mmap_sem for write.
 691  */
 692 int __kvm_set_memory_region(struct kvm *kvm,
 693                             struct kvm_userspace_memory_region *mem,
 694                             int user_alloc)
 695 {
 696         int r;
 697         gfn_t base_gfn;
 698         unsigned long npages;
 699         unsigned long i;
 700         struct kvm_memory_slot *memslot;
 701         struct kvm_memory_slot old, new;
 702         struct kvm_memslots *slots, *old_memslots;
 703
 704         r = -EINVAL;
 705         /* General sanity checks */
 706         if (mem->memory_size & (PAGE_SIZE - 1))
 707                 goto out;
 708         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 709                 goto out;
 710         /* We can read the guest memory with __xxx_user() later on. */
 711         if (user_alloc &&
 712             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 713              !access_ok(VERIFY_WRITE,
 714                         (void __user *)(unsigned long)mem->userspace_addr,
 715                         mem->memory_size)))
 716                 goto out;
 717         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 718                 goto out;
 719         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 720                 goto out;
 721
 722         memslot = id_to_memslot(kvm->memslots, mem->slot);
 723         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 724         npages = mem->memory_size >> PAGE_SHIFT;
 725
 726         r = -EINVAL;
 727         if (npages > KVM_MEM_MAX_NR_PAGES)
 728                 goto out;
 729
 730         if (!npages)
 731                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 732
 733         new = old = *memslot;
 734
 735         new.id = mem->slot;
 736         new.base_gfn = base_gfn;
 737         new.npages = npages;
 738         new.flags = mem->flags;
 739
 740         /* Disallow changing a memory slot's size. */
 741         r = -EINVAL;
 742         if (npages && old.npages && npages != old.npages)
 743                 goto out_free;
 744
 745         /* Check for overlaps */
 746         r = -EEXIST;
 747         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 748                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 749
 750                 if (s == memslot || !s->npages)
 751                         continue;
 752                 if (!((base_gfn + npages <= s->base_gfn) ||
 753                       (base_gfn >= s->base_gfn + s->npages)))
 754                         goto out_free;
 755         }
 756
 757         /* Free page dirty bitmap if unneeded */
 758         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 759                 new.dirty_bitmap = NULL;
 760
 761         r = -ENOMEM;
 762
 763         /* Allocate if a slot is being created */
 764         if (npages && !old.npages) {
 765                 new.user_alloc = user_alloc;
 766                 new.userspace_addr = mem->userspace_addr;
 767
 768                 if (kvm_arch_create_memslot(&new, npages))
 769                         goto out_free;
 770         }
 771
 772         /* Allocate page dirty bitmap if needed */
 773         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 774                 if (kvm_create_dirty_bitmap(&new) < 0)
 775                         goto out_free;
 776                 /* destroy any largepage mappings for dirty tracking */
 777         }
 778
 779         if (!npages) {
 780                 struct kvm_memory_slot *slot;
 781
 782                 r = -ENOMEM;
 783                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 784                                 GFP_KERNEL);
 785                 if (!slots)
 786                         goto out_free;
 787                 slot = id_to_memslot(slots, mem->slot);
 788                 slot->flags |= KVM_MEMSLOT_INVALID;
 789
 790                 update_memslots(slots, NULL);
 791
 792                 old_memslots = kvm->memslots;
 793                 rcu_assign_pointer(kvm->memslots, slots);
 794                 synchronize_srcu_expedited(&kvm->srcu);
 795                 /* From this point no new shadow pages pointing to a deleted
 796                  * memslot will be created.
 797                  *
 798                  * validation of sp->gfn happens in:
 799                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 800                  *      - kvm_is_visible_gfn (mmu_check_roots)
 801                  */
 802                 kvm_arch_flush_shadow(kvm);
 803                 kfree(old_memslots);
 804         }
 805
 806         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 807         if (r)
 808                 goto out_free;
 809
 810         /* map/unmap the pages in iommu page table */
 811         if (npages) {
 812                 r = kvm_iommu_map_pages(kvm, &new);
 813                 if (r)
 814                         goto out_free;
 815         } else
 816                 kvm_iommu_unmap_pages(kvm, &old);
 817
 818         r = -ENOMEM;
 819         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 820                         GFP_KERNEL);
 821         if (!slots)
 822                 goto out_free;
 823
 824         /* actual memory is freed via old in kvm_free_physmem_slot below */
 825         if (!npages) {
 826                 new.dirty_bitmap = NULL;
 827                 memset(&new.arch, 0, sizeof(new.arch));
 828         }
 829
 830         update_memslots(slots, &new);
 831         old_memslots = kvm->memslots;
 832         rcu_assign_pointer(kvm->memslots, slots);
 833         synchronize_srcu_expedited(&kvm->srcu);
 834
 835         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 836
 837         /*
 838          * If the new memory slot is created, we need to clear all
 839          * mmio sptes.
 840          */
 841         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 842                 kvm_arch_flush_shadow(kvm);
 843
 844         kvm_free_physmem_slot(&old, &new);
 845         kfree(old_memslots);
 846
 847         return 0;
 848
 849 out_free:
 850         kvm_free_physmem_slot(&new, &old);
 851 out:
 852         return r;
 853
 854 }
 855 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 856
 857 int kvm_set_memory_region(struct kvm *kvm,
 858                           struct kvm_userspace_memory_region *mem,
 859                           int user_alloc)
 860 {
 861         int r;
 862
 863         mutex_lock(&kvm->slots_lock);
 864         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 865         mutex_unlock(&kvm->slots_lock);
 866         return r;
 867 }
 868 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 869
 870 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 871                                    struct
 872                                    kvm_userspace_memory_region *mem,
 873                                    int user_alloc)
 874 {
 875         if (mem->slot >= KVM_MEMORY_SLOTS)
 876                 return -EINVAL;
 877         return kvm_set_memory_region(kvm, mem, user_alloc);
 878 }
 879
 880 int kvm_get_dirty_log(struct kvm *kvm,
 881                         struct kvm_dirty_log *log, int *is_dirty)
 882 {
 883         struct kvm_memory_slot *memslot;
 884         int r, i;
 885         unsigned long n;
 886         unsigned long any = 0;
 887
 888         r = -EINVAL;
 889         if (log->slot >= KVM_MEMORY_SLOTS)
 890                 goto out;
 891
 892         memslot = id_to_memslot(kvm->memslots, log->slot);
 893         r = -ENOENT;
 894         if (!memslot->dirty_bitmap)
 895                 goto out;
 896
 897         n = kvm_dirty_bitmap_bytes(memslot);
 898
 899         for (i = 0; !any && i < n/sizeof(long); ++i)
 900                 any = memslot->dirty_bitmap[i];
 901
 902         r = -EFAULT;
 903         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 904                 goto out;
 905
 906         if (any)
 907                 *is_dirty = 1;
 908
 909         r = 0;
 910 out:
 911         return r;
 912 }
 913
 914 bool kvm_largepages_enabled(void)
 915 {
 916         return largepages_enabled;
 917 }
 918
 919 void kvm_disable_largepages(void)
 920 {
 921         largepages_enabled = false;
 922 }
 923 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 924
 925 int is_error_page(struct page *page)
 926 {
 927         return IS_ERR(page);
 928 }
 929 EXPORT_SYMBOL_GPL(is_error_page);
 930
 931 struct page *get_bad_page(void)
 932 {
 933         return ERR_PTR(-ENOENT);
 934 }
 935
 936 static inline unsigned long bad_hva(void)
 937 {
 938         return PAGE_OFFSET;
 939 }
 940
 941 int kvm_is_error_hva(unsigned long addr)
 942 {
 943         return addr == bad_hva();
 944 }
 945 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 946
 947 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 948 {
 949         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 950 }
 951 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 952
 953 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 954 {
 955         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 956
 957         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
 958               memslot->flags & KVM_MEMSLOT_INVALID)
 959                 return 0;
 960
 961         return 1;
 962 }
 963 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 964
 965 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 966 {
 967         struct vm_area_struct *vma;
 968         unsigned long addr, size;
 969
 970         size = PAGE_SIZE;
 971
 972         addr = gfn_to_hva(kvm, gfn);
 973         if (kvm_is_error_hva(addr))
 974                 return PAGE_SIZE;
 975
 976         down_read(&current->mm->mmap_sem);
 977         vma = find_vma(current->mm, addr);
 978         if (!vma)
 979                 goto out;
 980
 981         size = vma_kernel_pagesize(vma);
 982
 983 out:
 984         up_read(&current->mm->mmap_sem);
 985
 986         return size;
 987 }
 988
 989 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
 990                                      gfn_t *nr_pages)
 991 {
 992         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 993                 return bad_hva();
 994
 995         if (nr_pages)
 996                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
 997
 998         return gfn_to_hva_memslot(slot, gfn);
 999 }
1000
1001 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1002 {
1003         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1004 }
1005 EXPORT_SYMBOL_GPL(gfn_to_hva);
1006
1007 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1008         unsigned long start, int write, struct page **page)
1009 {
1010         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1011
1012         if (write)
1013                 flags |= FOLL_WRITE;
1014
1015         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1016 }
1017
1018 static inline int check_user_page_hwpoison(unsigned long addr)
1019 {
1020         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1021
1022         rc = __get_user_pages(current, current->mm, addr, 1,
1023                               flags, NULL, NULL, NULL);
1024         return rc == -EHWPOISON;
1025 }
1026
1027 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1028                         bool write_fault, bool *writable)
1029 {
1030         struct page *page[1];
1031         int npages = 0;
1032         pfn_t pfn;
1033
1034         /* we can do it either atomically or asynchronously, not both */
1035         BUG_ON(atomic && async);
1036
1037         BUG_ON(!write_fault && !writable);
1038
1039         if (writable)
1040                 *writable = true;
1041
1042         if (atomic || async)
1043                 npages = __get_user_pages_fast(addr, 1, 1, page);
1044
1045         if (unlikely(npages != 1) && !atomic) {
1046                 might_sleep();
1047
1048                 if (writable)
1049                         *writable = write_fault;
1050
1051                 if (async) {
1052                         down_read(&current->mm->mmap_sem);
1053                         npages = get_user_page_nowait(current, current->mm,
1054                                                      addr, write_fault, page);
1055                         up_read(&current->mm->mmap_sem);
1056                 } else
1057                         npages = get_user_pages_fast(addr, 1, write_fault,
1058                                                      page);
1059
1060                 /* map read fault as writable if possible */
1061                 if (unlikely(!write_fault) && npages == 1) {
1062                         struct page *wpage[1];
1063
1064                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1065                         if (npages == 1) {
1066                                 *writable = true;
1067                                 put_page(page[0]);
1068                                 page[0] = wpage[0];
1069                         }
1070                         npages = 1;
1071                 }
1072         }
1073
1074         if (unlikely(npages != 1)) {
1075                 struct vm_area_struct *vma;
1076
1077                 if (atomic)
1078                         return KVM_PFN_ERR_FAULT;
1079
1080                 down_read(&current->mm->mmap_sem);
1081                 if (npages == -EHWPOISON ||
1082                         (!async && check_user_page_hwpoison(addr))) {
1083                         up_read(&current->mm->mmap_sem);
1084                         return KVM_PFN_ERR_HWPOISON;
1085                 }
1086
1087                 vma = find_vma_intersection(current->mm, addr, addr+1);
1088
1089                 if (vma == NULL)
1090                         pfn = KVM_PFN_ERR_FAULT;
1091                 else if ((vma->vm_flags & VM_PFNMAP)) {
1092                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1093                                 vma->vm_pgoff;
1094                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1095                 } else {
1096                         if (async && (vma->vm_flags & VM_WRITE))
1097                                 *async = true;
1098                         pfn = KVM_PFN_ERR_FAULT;
1099                 }
1100                 up_read(&current->mm->mmap_sem);
1101         } else
1102                 pfn = page_to_pfn(page[0]);
1103
1104         return pfn;
1105 }
1106
1107 pfn_t hva_to_pfn_atomic(unsigned long addr)
1108 {
1109         return hva_to_pfn(addr, true, NULL, true, NULL);
1110 }
1111 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1112
1113 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1114                           bool write_fault, bool *writable)
1115 {
1116         unsigned long addr;
1117
1118         if (async)
1119                 *async = false;
1120
1121         addr = gfn_to_hva(kvm, gfn);
1122         if (kvm_is_error_hva(addr))
1123                 return KVM_PFN_ERR_BAD;
1124
1125         return hva_to_pfn(addr, atomic, async, write_fault, writable);
1126 }
1127
1128 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1129 {
1130         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1131 }
1132 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1133
1134 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1135                        bool write_fault, bool *writable)
1136 {
1137         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1138 }
1139 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1140
1141 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1142 {
1143         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1144 }
1145 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1146
1147 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1148                       bool *writable)
1149 {
1150         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1151 }
1152 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1153
1154 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1155 {
1156         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1157         return hva_to_pfn(addr, false, NULL, true, NULL);
1158 }
1159
1160 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1161                                                                   int nr_pages)
1162 {
1163         unsigned long addr;
1164         gfn_t entry;
1165
1166         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1167         if (kvm_is_error_hva(addr))
1168                 return -1;
1169
1170         if (entry < nr_pages)
1171                 return 0;
1172
1173         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1174 }
1175 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1176
1177 static struct page *kvm_pfn_to_page(pfn_t pfn)
1178 {
1179         WARN_ON(kvm_is_mmio_pfn(pfn));
1180
1181         if (is_error_pfn(pfn) || kvm_is_mmio_pfn(pfn))
1182                 return get_bad_page();
1183
1184         return pfn_to_page(pfn);
1185 }
1186
1187 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1188 {
1189         pfn_t pfn;
1190
1191         pfn = gfn_to_pfn(kvm, gfn);
1192
1193         return kvm_pfn_to_page(pfn);
1194 }
1195
1196 EXPORT_SYMBOL_GPL(gfn_to_page);
1197
1198 void kvm_release_page_clean(struct page *page)
1199 {
1200         if (!is_error_page(page))
1201                 kvm_release_pfn_clean(page_to_pfn(page));
1202 }
1203 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1204
1205 void kvm_release_pfn_clean(pfn_t pfn)
1206 {
1207         if (!is_error_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1208                 put_page(pfn_to_page(pfn));
1209 }
1210 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1211
1212 void kvm_release_page_dirty(struct page *page)
1213 {
1214         WARN_ON(is_error_page(page));
1215
1216         kvm_release_pfn_dirty(page_to_pfn(page));
1217 }
1218 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1219
1220 void kvm_release_pfn_dirty(pfn_t pfn)
1221 {
1222         kvm_set_pfn_dirty(pfn);
1223         kvm_release_pfn_clean(pfn);
1224 }
1225 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1226
1227 void kvm_set_page_dirty(struct page *page)
1228 {
1229         kvm_set_pfn_dirty(page_to_pfn(page));
1230 }
1231 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1232
1233 void kvm_set_pfn_dirty(pfn_t pfn)
1234 {
1235         if (!kvm_is_mmio_pfn(pfn)) {
1236                 struct page *page = pfn_to_page(pfn);
1237                 if (!PageReserved(page))
1238                         SetPageDirty(page);
1239         }
1240 }
1241 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1242
1243 void kvm_set_pfn_accessed(pfn_t pfn)
1244 {
1245         if (!kvm_is_mmio_pfn(pfn))
1246                 mark_page_accessed(pfn_to_page(pfn));
1247 }
1248 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1249
1250 void kvm_get_pfn(pfn_t pfn)
1251 {
1252         if (!kvm_is_mmio_pfn(pfn))
1253                 get_page(pfn_to_page(pfn));
1254 }
1255 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1256
1257 static int next_segment(unsigned long len, int offset)
1258 {
1259         if (len > PAGE_SIZE - offset)
1260                 return PAGE_SIZE - offset;
1261         else
1262                 return len;
1263 }
1264
1265 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1266                         int len)
1267 {
1268         int r;
1269         unsigned long addr;
1270
1271         addr = gfn_to_hva(kvm, gfn);
1272         if (kvm_is_error_hva(addr))
1273                 return -EFAULT;
1274         r = __copy_from_user(data, (void __user *)addr + offset, len);
1275         if (r)
1276                 return -EFAULT;
1277         return 0;
1278 }
1279 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1280
1281 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1282 {
1283         gfn_t gfn = gpa >> PAGE_SHIFT;
1284         int seg;
1285         int offset = offset_in_page(gpa);
1286         int ret;
1287
1288         while ((seg = next_segment(len, offset)) != 0) {
1289                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1290                 if (ret < 0)
1291                         return ret;
1292                 offset = 0;
1293                 len -= seg;
1294                 data += seg;
1295                 ++gfn;
1296         }
1297         return 0;
1298 }
1299 EXPORT_SYMBOL_GPL(kvm_read_guest);
1300
1301 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1302                           unsigned long len)
1303 {
1304         int r;
1305         unsigned long addr;
1306         gfn_t gfn = gpa >> PAGE_SHIFT;
1307         int offset = offset_in_page(gpa);
1308
1309         addr = gfn_to_hva(kvm, gfn);
1310         if (kvm_is_error_hva(addr))
1311                 return -EFAULT;
1312         pagefault_disable();
1313         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1314         pagefault_enable();
1315         if (r)
1316                 return -EFAULT;
1317         return 0;
1318 }
1319 EXPORT_SYMBOL(kvm_read_guest_atomic);
1320
1321 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1322                          int offset, int len)
1323 {
1324         int r;
1325         unsigned long addr;
1326
1327         addr = gfn_to_hva(kvm, gfn);
1328         if (kvm_is_error_hva(addr))
1329                 return -EFAULT;
1330         r = __copy_to_user((void __user *)addr + offset, data, len);
1331         if (r)
1332                 return -EFAULT;
1333         mark_page_dirty(kvm, gfn);
1334         return 0;
1335 }
1336 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1337
1338 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1339                     unsigned long len)
1340 {
1341         gfn_t gfn = gpa >> PAGE_SHIFT;
1342         int seg;
1343         int offset = offset_in_page(gpa);
1344         int ret;
1345
1346         while ((seg = next_segment(len, offset)) != 0) {
1347                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1348                 if (ret < 0)
1349                         return ret;
1350                 offset = 0;
1351                 len -= seg;
1352                 data += seg;
1353                 ++gfn;
1354         }
1355         return 0;
1356 }
1357
1358 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1359                               gpa_t gpa)
1360 {
1361         struct kvm_memslots *slots = kvm_memslots(kvm);
1362         int offset = offset_in_page(gpa);
1363         gfn_t gfn = gpa >> PAGE_SHIFT;
1364
1365         ghc->gpa = gpa;
1366         ghc->generation = slots->generation;
1367         ghc->memslot = gfn_to_memslot(kvm, gfn);
1368         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1369         if (!kvm_is_error_hva(ghc->hva))
1370                 ghc->hva += offset;
1371         else
1372                 return -EFAULT;
1373
1374         return 0;
1375 }
1376 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1377
1378 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1379                            void *data, unsigned long len)
1380 {
1381         struct kvm_memslots *slots = kvm_memslots(kvm);
1382         int r;
1383
1384         if (slots->generation != ghc->generation)
1385                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1386
1387         if (kvm_is_error_hva(ghc->hva))
1388                 return -EFAULT;
1389
1390         r = __copy_to_user((void __user *)ghc->hva, data, len);
1391         if (r)
1392                 return -EFAULT;
1393         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1394
1395         return 0;
1396 }
1397 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1398
1399 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1400                            void *data, unsigned long len)
1401 {
1402         struct kvm_memslots *slots = kvm_memslots(kvm);
1403         int r;
1404
1405         if (slots->generation != ghc->generation)
1406                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1407
1408         if (kvm_is_error_hva(ghc->hva))
1409                 return -EFAULT;
1410
1411         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1412         if (r)
1413                 return -EFAULT;
1414
1415         return 0;
1416 }
1417 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1418
1419 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1420 {
1421         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1422                                     offset, len);
1423 }
1424 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1425
1426 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1427 {
1428         gfn_t gfn = gpa >> PAGE_SHIFT;
1429         int seg;
1430         int offset = offset_in_page(gpa);
1431         int ret;
1432
1433         while ((seg = next_segment(len, offset)) != 0) {
1434                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1435                 if (ret < 0)
1436                         return ret;
1437                 offset = 0;
1438                 len -= seg;
1439                 ++gfn;
1440         }
1441         return 0;
1442 }
1443 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1444
1445 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1446                              gfn_t gfn)
1447 {
1448         if (memslot && memslot->dirty_bitmap) {
1449                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1450
1451                 /* TODO: introduce set_bit_le() and use it */
1452                 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1453         }
1454 }
1455
1456 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1457 {
1458         struct kvm_memory_slot *memslot;
1459
1460         memslot = gfn_to_memslot(kvm, gfn);
1461         mark_page_dirty_in_slot(kvm, memslot, gfn);
1462 }
1463
1464 /*
1465  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1466  */
1467 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1468 {
1469         DEFINE_WAIT(wait);
1470
1471         for (;;) {
1472                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1473
1474                 if (kvm_arch_vcpu_runnable(vcpu)) {
1475                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1476                         break;
1477                 }
1478                 if (kvm_cpu_has_pending_timer(vcpu))
1479                         break;
1480                 if (signal_pending(current))
1481                         break;
1482
1483                 schedule();
1484         }
1485
1486         finish_wait(&vcpu->wq, &wait);
1487 }
1488
1489 #ifndef CONFIG_S390
1490 /*
1491  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1492  */
1493 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1494 {
1495         int me;
1496         int cpu = vcpu->cpu;
1497         wait_queue_head_t *wqp;
1498
1499         wqp = kvm_arch_vcpu_wq(vcpu);
1500         if (waitqueue_active(wqp)) {
1501                 wake_up_interruptible(wqp);
1502                 ++vcpu->stat.halt_wakeup;
1503         }
1504
1505         me = get_cpu();
1506         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1507                 if (kvm_arch_vcpu_should_kick(vcpu))
1508                         smp_send_reschedule(cpu);
1509         put_cpu();
1510 }
1511 #endif /* !CONFIG_S390 */
1512
1513 void kvm_resched(struct kvm_vcpu *vcpu)
1514 {
1515         if (!need_resched())
1516                 return;
1517         cond_resched();
1518 }
1519 EXPORT_SYMBOL_GPL(kvm_resched);
1520
1521 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1522 {
1523         struct pid *pid;
1524         struct task_struct *task = NULL;
1525
1526         rcu_read_lock();
1527         pid = rcu_dereference(target->pid);
1528         if (pid)
1529                 task = get_pid_task(target->pid, PIDTYPE_PID);
1530         rcu_read_unlock();
1531         if (!task)
1532                 return false;
1533         if (task->flags & PF_VCPU) {
1534                 put_task_struct(task);
1535                 return false;
1536         }
1537         if (yield_to(task, 1)) {
1538                 put_task_struct(task);
1539                 return true;
1540         }
1541         put_task_struct(task);
1542         return false;
1543 }
1544 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1545
1546 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1547 /*
1548  * Helper that checks whether a VCPU is eligible for directed yield.
1549  * Most eligible candidate to yield is decided by following heuristics:
1550  *
1551  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1552  *  (preempted lock holder), indicated by @in_spin_loop.
1553  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1554  *
1555  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1556  *  chance last time (mostly it has become eligible now since we have probably
1557  *  yielded to lockholder in last iteration. This is done by toggling
1558  *  @dy_eligible each time a VCPU checked for eligibility.)
1559  *
1560  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1561  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1562  *  burning. Giving priority for a potential lock-holder increases lock
1563  *  progress.
1564  *
1565  *  Since algorithm is based on heuristics, accessing another VCPU data without
1566  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1567  *  and continue with next VCPU and so on.
1568  */
1569 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1570 {
1571         bool eligible;
1572
1573         eligible = !vcpu->spin_loop.in_spin_loop ||
1574                         (vcpu->spin_loop.in_spin_loop &&
1575                          vcpu->spin_loop.dy_eligible);
1576
1577         if (vcpu->spin_loop.in_spin_loop)
1578                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1579
1580         return eligible;
1581 }
1582 #endif
1583 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1584 {
1585         struct kvm *kvm = me->kvm;
1586         struct kvm_vcpu *vcpu;
1587         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1588         int yielded = 0;
1589         int pass;
1590         int i;
1591
1592         kvm_vcpu_set_in_spin_loop(me, true);
1593         /*
1594          * We boost the priority of a VCPU that is runnable but not
1595          * currently running, because it got preempted by something
1596          * else and called schedule in __vcpu_run.  Hopefully that
1597          * VCPU is holding the lock that we need and will release it.
1598          * We approximate round-robin by starting at the last boosted VCPU.
1599          */
1600         for (pass = 0; pass < 2 && !yielded; pass++) {
1601                 kvm_for_each_vcpu(i, vcpu, kvm) {
1602                         if (!pass && i <= last_boosted_vcpu) {
1603                                 i = last_boosted_vcpu;
1604                                 continue;
1605                         } else if (pass && i > last_boosted_vcpu)
1606                                 break;
1607                         if (vcpu == me)
1608                                 continue;
1609                         if (waitqueue_active(&vcpu->wq))
1610                                 continue;
1611                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1612                                 continue;
1613                         if (kvm_vcpu_yield_to(vcpu)) {
1614                                 kvm->last_boosted_vcpu = i;
1615                                 yielded = 1;
1616                                 break;
1617                         }
1618                 }
1619         }
1620         kvm_vcpu_set_in_spin_loop(me, false);
1621
1622         /* Ensure vcpu is not eligible during next spinloop */
1623         kvm_vcpu_set_dy_eligible(me, false);
1624 }
1625 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1626
1627 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1628 {
1629         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1630         struct page *page;
1631
1632         if (vmf->pgoff == 0)
1633                 page = virt_to_page(vcpu->run);
1634 #ifdef CONFIG_X86
1635         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1636                 page = virt_to_page(vcpu->arch.pio_data);
1637 #endif
1638 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1639         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1640                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1641 #endif
1642         else
1643                 return kvm_arch_vcpu_fault(vcpu, vmf);
1644         get_page(page);
1645         vmf->page = page;
1646         return 0;
1647 }
1648
1649 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1650         .fault = kvm_vcpu_fault,
1651 };
1652
1653 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1654 {
1655         vma->vm_ops = &kvm_vcpu_vm_ops;
1656         return 0;
1657 }
1658
1659 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1660 {
1661         struct kvm_vcpu *vcpu = filp->private_data;
1662
1663         kvm_put_kvm(vcpu->kvm);
1664         return 0;
1665 }
1666
1667 static struct file_operations kvm_vcpu_fops = {
1668         .release        = kvm_vcpu_release,
1669         .unlocked_ioctl = kvm_vcpu_ioctl,
1670 #ifdef CONFIG_COMPAT
1671         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1672 #endif
1673         .mmap           = kvm_vcpu_mmap,
1674         .llseek         = noop_llseek,
1675 };
1676
1677 /*
1678  * Allocates an inode for the vcpu.
1679  */
1680 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1681 {
1682         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1683 }
1684
1685 /*
1686  * Creates some virtual cpus.  Good luck creating more than one.
1687  */
1688 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1689 {
1690         int r;
1691         struct kvm_vcpu *vcpu, *v;
1692
1693         vcpu = kvm_arch_vcpu_create(kvm, id);
1694         if (IS_ERR(vcpu))
1695                 return PTR_ERR(vcpu);
1696
1697         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1698
1699         r = kvm_arch_vcpu_setup(vcpu);
1700         if (r)
1701                 goto vcpu_destroy;
1702
1703         mutex_lock(&kvm->lock);
1704         if (!kvm_vcpu_compatible(vcpu)) {
1705                 r = -EINVAL;
1706                 goto unlock_vcpu_destroy;
1707         }
1708         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1709                 r = -EINVAL;
1710                 goto unlock_vcpu_destroy;
1711         }
1712
1713         kvm_for_each_vcpu(r, v, kvm)
1714                 if (v->vcpu_id == id) {
1715                         r = -EEXIST;
1716                         goto unlock_vcpu_destroy;
1717                 }
1718
1719         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1720
1721         /* Now it's all set up, let userspace reach it */
1722         kvm_get_kvm(kvm);
1723         r = create_vcpu_fd(vcpu);
1724         if (r < 0) {
1725                 kvm_put_kvm(kvm);
1726                 goto unlock_vcpu_destroy;
1727         }
1728
1729         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1730         smp_wmb();
1731         atomic_inc(&kvm->online_vcpus);
1732
1733         mutex_unlock(&kvm->lock);
1734         return r;
1735
1736 unlock_vcpu_destroy:
1737         mutex_unlock(&kvm->lock);
1738 vcpu_destroy:
1739         kvm_arch_vcpu_destroy(vcpu);
1740         return r;
1741 }
1742
1743 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1744 {
1745         if (sigset) {
1746                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1747                 vcpu->sigset_active = 1;
1748                 vcpu->sigset = *sigset;
1749         } else
1750                 vcpu->sigset_active = 0;
1751         return 0;
1752 }
1753
1754 static long kvm_vcpu_ioctl(struct file *filp,
1755                            unsigned int ioctl, unsigned long arg)
1756 {
1757         struct kvm_vcpu *vcpu = filp->private_data;
1758         void __user *argp = (void __user *)arg;
1759         int r;
1760         struct kvm_fpu *fpu = NULL;
1761         struct kvm_sregs *kvm_sregs = NULL;
1762
1763         if (vcpu->kvm->mm != current->mm)
1764                 return -EIO;
1765
1766 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1767         /*
1768          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1769          * so vcpu_load() would break it.
1770          */
1771         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1772                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1773 #endif
1774
1775
1776         vcpu_load(vcpu);
1777         switch (ioctl) {
1778         case KVM_RUN:
1779                 r = -EINVAL;
1780                 if (arg)
1781                         goto out;
1782                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1783                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1784                 break;
1785         case KVM_GET_REGS: {
1786                 struct kvm_regs *kvm_regs;
1787
1788                 r = -ENOMEM;
1789                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1790                 if (!kvm_regs)
1791                         goto out;
1792                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1793                 if (r)
1794                         goto out_free1;
1795                 r = -EFAULT;
1796                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1797                         goto out_free1;
1798                 r = 0;
1799 out_free1:
1800                 kfree(kvm_regs);
1801                 break;
1802         }
1803         case KVM_SET_REGS: {
1804                 struct kvm_regs *kvm_regs;
1805
1806                 r = -ENOMEM;
1807                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1808                 if (IS_ERR(kvm_regs)) {
1809                         r = PTR_ERR(kvm_regs);
1810                         goto out;
1811                 }
1812                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1813                 if (r)
1814                         goto out_free2;
1815                 r = 0;
1816 out_free2:
1817                 kfree(kvm_regs);
1818                 break;
1819         }
1820         case KVM_GET_SREGS: {
1821                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1822                 r = -ENOMEM;
1823                 if (!kvm_sregs)
1824                         goto out;
1825                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1826                 if (r)
1827                         goto out;
1828                 r = -EFAULT;
1829                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1830                         goto out;
1831                 r = 0;
1832                 break;
1833         }
1834         case KVM_SET_SREGS: {
1835                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1836                 if (IS_ERR(kvm_sregs)) {
1837                         r = PTR_ERR(kvm_sregs);
1838                         goto out;
1839                 }
1840                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1841                 if (r)
1842                         goto out;
1843                 r = 0;
1844                 break;
1845         }
1846         case KVM_GET_MP_STATE: {
1847                 struct kvm_mp_state mp_state;
1848
1849                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1850                 if (r)
1851                         goto out;
1852                 r = -EFAULT;
1853                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1854                         goto out;
1855                 r = 0;
1856                 break;
1857         }
1858         case KVM_SET_MP_STATE: {
1859                 struct kvm_mp_state mp_state;
1860
1861                 r = -EFAULT;
1862                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1863                         goto out;
1864                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1865                 if (r)
1866                         goto out;
1867                 r = 0;
1868                 break;
1869         }
1870         case KVM_TRANSLATE: {
1871                 struct kvm_translation tr;
1872
1873                 r = -EFAULT;
1874                 if (copy_from_user(&tr, argp, sizeof tr))
1875                         goto out;
1876                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1877                 if (r)
1878                         goto out;
1879                 r = -EFAULT;
1880                 if (copy_to_user(argp, &tr, sizeof tr))
1881                         goto out;
1882                 r = 0;
1883                 break;
1884         }
1885         case KVM_SET_GUEST_DEBUG: {
1886                 struct kvm_guest_debug dbg;
1887
1888                 r = -EFAULT;
1889                 if (copy_from_user(&dbg, argp, sizeof dbg))
1890                         goto out;
1891                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1892                 if (r)
1893                         goto out;
1894                 r = 0;
1895                 break;
1896         }
1897         case KVM_SET_SIGNAL_MASK: {
1898                 struct kvm_signal_mask __user *sigmask_arg = argp;
1899                 struct kvm_signal_mask kvm_sigmask;
1900                 sigset_t sigset, *p;
1901
1902                 p = NULL;
1903                 if (argp) {
1904                         r = -EFAULT;
1905                         if (copy_from_user(&kvm_sigmask, argp,
1906                                            sizeof kvm_sigmask))
1907                                 goto out;
1908                         r = -EINVAL;
1909                         if (kvm_sigmask.len != sizeof sigset)
1910                                 goto out;
1911                         r = -EFAULT;
1912                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1913                                            sizeof sigset))
1914                                 goto out;
1915                         p = &sigset;
1916                 }
1917                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1918                 break;
1919         }
1920         case KVM_GET_FPU: {
1921                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1922                 r = -ENOMEM;
1923                 if (!fpu)
1924                         goto out;
1925                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1926                 if (r)
1927                         goto out;
1928                 r = -EFAULT;
1929                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1930                         goto out;
1931                 r = 0;
1932                 break;
1933         }
1934         case KVM_SET_FPU: {
1935                 fpu = memdup_user(argp, sizeof(*fpu));
1936                 if (IS_ERR(fpu)) {
1937                         r = PTR_ERR(fpu);
1938                         goto out;
1939                 }
1940                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1941                 if (r)
1942                         goto out;
1943                 r = 0;
1944                 break;
1945         }
1946         default:
1947                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1948         }
1949 out:
1950         vcpu_put(vcpu);
1951         kfree(fpu);
1952         kfree(kvm_sregs);
1953         return r;
1954 }
1955
1956 #ifdef CONFIG_COMPAT
1957 static long kvm_vcpu_compat_ioctl(struct file *filp,
1958                                   unsigned int ioctl, unsigned long arg)
1959 {
1960         struct kvm_vcpu *vcpu = filp->private_data;
1961         void __user *argp = compat_ptr(arg);
1962         int r;
1963
1964         if (vcpu->kvm->mm != current->mm)
1965                 return -EIO;
1966
1967         switch (ioctl) {
1968         case KVM_SET_SIGNAL_MASK: {
1969                 struct kvm_signal_mask __user *sigmask_arg = argp;
1970                 struct kvm_signal_mask kvm_sigmask;
1971                 compat_sigset_t csigset;
1972                 sigset_t sigset;
1973
1974                 if (argp) {
1975                         r = -EFAULT;
1976                         if (copy_from_user(&kvm_sigmask, argp,
1977                                            sizeof kvm_sigmask))
1978                                 goto out;
1979                         r = -EINVAL;
1980                         if (kvm_sigmask.len != sizeof csigset)
1981                                 goto out;
1982                         r = -EFAULT;
1983                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1984                                            sizeof csigset))
1985                                 goto out;
1986                 }
1987                 sigset_from_compat(&sigset, &csigset);
1988                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1989                 break;
1990         }
1991         default:
1992                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1993         }
1994
1995 out:
1996         return r;
1997 }
1998 #endif
1999
2000 static long kvm_vm_ioctl(struct file *filp,
2001                            unsigned int ioctl, unsigned long arg)
2002 {
2003         struct kvm *kvm = filp->private_data;
2004         void __user *argp = (void __user *)arg;
2005         int r;
2006
2007         if (kvm->mm != current->mm)
2008                 return -EIO;
2009         switch (ioctl) {
2010         case KVM_CREATE_VCPU:
2011                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2012                 if (r < 0)
2013                         goto out;
2014                 break;
2015         case KVM_SET_USER_MEMORY_REGION: {
2016                 struct kvm_userspace_memory_region kvm_userspace_mem;
2017
2018                 r = -EFAULT;
2019                 if (copy_from_user(&kvm_userspace_mem, argp,
2020                                                 sizeof kvm_userspace_mem))
2021                         goto out;
2022
2023                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2024                 if (r)
2025                         goto out;
2026                 break;
2027         }
2028         case KVM_GET_DIRTY_LOG: {
2029                 struct kvm_dirty_log log;
2030
2031                 r = -EFAULT;
2032                 if (copy_from_user(&log, argp, sizeof log))
2033                         goto out;
2034                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2035                 if (r)
2036                         goto out;
2037                 break;
2038         }
2039 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2040         case KVM_REGISTER_COALESCED_MMIO: {
2041                 struct kvm_coalesced_mmio_zone zone;
2042                 r = -EFAULT;
2043                 if (copy_from_user(&zone, argp, sizeof zone))
2044                         goto out;
2045                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2046                 if (r)
2047                         goto out;
2048                 r = 0;
2049                 break;
2050         }
2051         case KVM_UNREGISTER_COALESCED_MMIO: {
2052                 struct kvm_coalesced_mmio_zone zone;
2053                 r = -EFAULT;
2054                 if (copy_from_user(&zone, argp, sizeof zone))
2055                         goto out;
2056                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2057                 if (r)
2058                         goto out;
2059                 r = 0;
2060                 break;
2061         }
2062 #endif
2063         case KVM_IRQFD: {
2064                 struct kvm_irqfd data;
2065
2066                 r = -EFAULT;
2067                 if (copy_from_user(&data, argp, sizeof data))
2068                         goto out;
2069                 r = kvm_irqfd(kvm, &data);
2070                 break;
2071         }
2072         case KVM_IOEVENTFD: {
2073                 struct kvm_ioeventfd data;
2074
2075                 r = -EFAULT;
2076                 if (copy_from_user(&data, argp, sizeof data))
2077                         goto out;
2078                 r = kvm_ioeventfd(kvm, &data);
2079                 break;
2080         }
2081 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2082         case KVM_SET_BOOT_CPU_ID:
2083                 r = 0;
2084                 mutex_lock(&kvm->lock);
2085                 if (atomic_read(&kvm->online_vcpus) != 0)
2086                         r = -EBUSY;
2087                 else
2088                         kvm->bsp_vcpu_id = arg;
2089                 mutex_unlock(&kvm->lock);
2090                 break;
2091 #endif
2092 #ifdef CONFIG_HAVE_KVM_MSI
2093         case KVM_SIGNAL_MSI: {
2094                 struct kvm_msi msi;
2095
2096                 r = -EFAULT;
2097                 if (copy_from_user(&msi, argp, sizeof msi))
2098                         goto out;
2099                 r = kvm_send_userspace_msi(kvm, &msi);
2100                 break;
2101         }
2102 #endif
2103 #ifdef __KVM_HAVE_IRQ_LINE
2104         case KVM_IRQ_LINE_STATUS:
2105         case KVM_IRQ_LINE: {
2106                 struct kvm_irq_level irq_event;
2107
2108                 r = -EFAULT;
2109                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2110                         goto out;
2111
2112                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2113                 if (r)
2114                         goto out;
2115
2116                 r = -EFAULT;
2117                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2118                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2119                                 goto out;
2120                 }
2121
2122                 r = 0;
2123                 break;
2124         }
2125 #endif
2126         default:
2127                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2128                 if (r == -ENOTTY)
2129                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2130         }
2131 out:
2132         return r;
2133 }
2134
2135 #ifdef CONFIG_COMPAT
2136 struct compat_kvm_dirty_log {
2137         __u32 slot;
2138         __u32 padding1;
2139         union {
2140                 compat_uptr_t dirty_bitmap; /* one bit per page */
2141                 __u64 padding2;
2142         };
2143 };
2144
2145 static long kvm_vm_compat_ioctl(struct file *filp,
2146                            unsigned int ioctl, unsigned long arg)
2147 {
2148         struct kvm *kvm = filp->private_data;
2149         int r;
2150
2151         if (kvm->mm != current->mm)
2152                 return -EIO;
2153         switch (ioctl) {
2154         case KVM_GET_DIRTY_LOG: {
2155                 struct compat_kvm_dirty_log compat_log;
2156                 struct kvm_dirty_log log;
2157
2158                 r = -EFAULT;
2159                 if (copy_from_user(&compat_log, (void __user *)arg,
2160                                    sizeof(compat_log)))
2161                         goto out;
2162                 log.slot         = compat_log.slot;
2163                 log.padding1     = compat_log.padding1;
2164                 log.padding2     = compat_log.padding2;
2165                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2166
2167                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2168                 if (r)
2169                         goto out;
2170                 break;
2171         }
2172         default:
2173                 r = kvm_vm_ioctl(filp, ioctl, arg);
2174         }
2175
2176 out:
2177         return r;
2178 }
2179 #endif
2180
2181 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2182 {
2183         struct page *page[1];
2184         unsigned long addr;
2185         int npages;
2186         gfn_t gfn = vmf->pgoff;
2187         struct kvm *kvm = vma->vm_file->private_data;
2188
2189         addr = gfn_to_hva(kvm, gfn);
2190         if (kvm_is_error_hva(addr))
2191                 return VM_FAULT_SIGBUS;
2192
2193         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2194                                 NULL);
2195         if (unlikely(npages != 1))
2196                 return VM_FAULT_SIGBUS;
2197
2198         vmf->page = page[0];
2199         return 0;
2200 }
2201
2202 static const struct vm_operations_struct kvm_vm_vm_ops = {
2203         .fault = kvm_vm_fault,
2204 };
2205
2206 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2207 {
2208         vma->vm_ops = &kvm_vm_vm_ops;
2209         return 0;
2210 }
2211
2212 static struct file_operations kvm_vm_fops = {
2213         .release        = kvm_vm_release,
2214         .unlocked_ioctl = kvm_vm_ioctl,
2215 #ifdef CONFIG_COMPAT
2216         .compat_ioctl   = kvm_vm_compat_ioctl,
2217 #endif
2218         .mmap           = kvm_vm_mmap,
2219         .llseek         = noop_llseek,
2220 };
2221
2222 static int kvm_dev_ioctl_create_vm(unsigned long type)
2223 {
2224         int r;
2225         struct kvm *kvm;
2226
2227         kvm = kvm_create_vm(type);
2228         if (IS_ERR(kvm))
2229                 return PTR_ERR(kvm);
2230 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2231         r = kvm_coalesced_mmio_init(kvm);
2232         if (r < 0) {
2233                 kvm_put_kvm(kvm);
2234                 return r;
2235         }
2236 #endif
2237         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2238         if (r < 0)
2239                 kvm_put_kvm(kvm);
2240
2241         return r;
2242 }
2243
2244 static long kvm_dev_ioctl_check_extension_generic(long arg)
2245 {
2246         switch (arg) {
2247         case KVM_CAP_USER_MEMORY:
2248         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2249         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2250 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2251         case KVM_CAP_SET_BOOT_CPU_ID:
2252 #endif
2253         case KVM_CAP_INTERNAL_ERROR_DATA:
2254 #ifdef CONFIG_HAVE_KVM_MSI
2255         case KVM_CAP_SIGNAL_MSI:
2256 #endif
2257                 return 1;
2258 #ifdef KVM_CAP_IRQ_ROUTING
2259         case KVM_CAP_IRQ_ROUTING:
2260                 return KVM_MAX_IRQ_ROUTES;
2261 #endif
2262         default:
2263                 break;
2264         }
2265         return kvm_dev_ioctl_check_extension(arg);
2266 }
2267
2268 static long kvm_dev_ioctl(struct file *filp,
2269                           unsigned int ioctl, unsigned long arg)
2270 {
2271         long r = -EINVAL;
2272
2273         switch (ioctl) {
2274         case KVM_GET_API_VERSION:
2275                 r = -EINVAL;
2276                 if (arg)
2277                         goto out;
2278                 r = KVM_API_VERSION;
2279                 break;
2280         case KVM_CREATE_VM:
2281                 r = kvm_dev_ioctl_create_vm(arg);
2282                 break;
2283         case KVM_CHECK_EXTENSION:
2284                 r = kvm_dev_ioctl_check_extension_generic(arg);
2285                 break;
2286         case KVM_GET_VCPU_MMAP_SIZE:
2287                 r = -EINVAL;
2288                 if (arg)
2289                         goto out;
2290                 r = PAGE_SIZE;     /* struct kvm_run */
2291 #ifdef CONFIG_X86
2292                 r += PAGE_SIZE;    /* pio data page */
2293 #endif
2294 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2295                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2296 #endif
2297                 break;
2298         case KVM_TRACE_ENABLE:
2299         case KVM_TRACE_PAUSE:
2300         case KVM_TRACE_DISABLE:
2301                 r = -EOPNOTSUPP;
2302                 break;
2303         default:
2304                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2305         }
2306 out:
2307         return r;
2308 }
2309
2310 static struct file_operations kvm_chardev_ops = {
2311         .unlocked_ioctl = kvm_dev_ioctl,
2312         .compat_ioctl   = kvm_dev_ioctl,
2313         .llseek         = noop_llseek,
2314 };
2315
2316 static struct miscdevice kvm_dev = {
2317         KVM_MINOR,
2318         "kvm",
2319         &kvm_chardev_ops,
2320 };
2321
2322 static void hardware_enable_nolock(void *junk)
2323 {
2324         int cpu = raw_smp_processor_id();
2325         int r;
2326
2327         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2328                 return;
2329
2330         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2331
2332         r = kvm_arch_hardware_enable(NULL);
2333
2334         if (r) {
2335                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2336                 atomic_inc(&hardware_enable_failed);
2337                 printk(KERN_INFO "kvm: enabling virtualization on "
2338                                  "CPU%d failed\n", cpu);
2339         }
2340 }
2341
2342 static void hardware_enable(void *junk)
2343 {
2344         raw_spin_lock(&kvm_lock);
2345         hardware_enable_nolock(junk);
2346         raw_spin_unlock(&kvm_lock);
2347 }
2348
2349 static void hardware_disable_nolock(void *junk)
2350 {
2351         int cpu = raw_smp_processor_id();
2352
2353         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2354                 return;
2355         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2356         kvm_arch_hardware_disable(NULL);
2357 }
2358
2359 static void hardware_disable(void *junk)
2360 {
2361         raw_spin_lock(&kvm_lock);
2362         hardware_disable_nolock(junk);
2363         raw_spin_unlock(&kvm_lock);
2364 }
2365
2366 static void hardware_disable_all_nolock(void)
2367 {
2368         BUG_ON(!kvm_usage_count);
2369
2370         kvm_usage_count--;
2371         if (!kvm_usage_count)
2372                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2373 }
2374
2375 static void hardware_disable_all(void)
2376 {
2377         raw_spin_lock(&kvm_lock);
2378         hardware_disable_all_nolock();
2379         raw_spin_unlock(&kvm_lock);
2380 }
2381
2382 static int hardware_enable_all(void)
2383 {
2384         int r = 0;
2385
2386         raw_spin_lock(&kvm_lock);
2387
2388         kvm_usage_count++;
2389         if (kvm_usage_count == 1) {
2390                 atomic_set(&hardware_enable_failed, 0);
2391                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2392
2393                 if (atomic_read(&hardware_enable_failed)) {
2394                         hardware_disable_all_nolock();
2395                         r = -EBUSY;
2396                 }
2397         }
2398
2399         raw_spin_unlock(&kvm_lock);
2400
2401         return r;
2402 }
2403
2404 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2405                            void *v)
2406 {
2407         int cpu = (long)v;
2408
2409         if (!kvm_usage_count)
2410                 return NOTIFY_OK;
2411
2412         val &= ~CPU_TASKS_FROZEN;
2413         switch (val) {
2414         case CPU_DYING:
2415                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2416                        cpu);
2417                 hardware_disable(NULL);
2418                 break;
2419         case CPU_STARTING:
2420                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2421                        cpu);
2422                 hardware_enable(NULL);
2423                 break;
2424         }
2425         return NOTIFY_OK;
2426 }
2427
2428
2429 asmlinkage void kvm_spurious_fault(void)
2430 {
2431         /* Fault while not rebooting.  We want the trace. */
2432         BUG();
2433 }
2434 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2435
2436 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2437                       void *v)
2438 {
2439         /*
2440          * Some (well, at least mine) BIOSes hang on reboot if
2441          * in vmx root mode.
2442          *
2443          * And Intel TXT required VMX off for all cpu when system shutdown.
2444          */
2445         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2446         kvm_rebooting = true;
2447         on_each_cpu(hardware_disable_nolock, NULL, 1);
2448         return NOTIFY_OK;
2449 }
2450
2451 static struct notifier_block kvm_reboot_notifier = {
2452         .notifier_call = kvm_reboot,
2453         .priority = 0,
2454 };
2455
2456 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2457 {
2458         int i;
2459
2460         for (i = 0; i < bus->dev_count; i++) {
2461                 struct kvm_io_device *pos = bus->range[i].dev;
2462
2463                 kvm_iodevice_destructor(pos);
2464         }
2465         kfree(bus);
2466 }
2467
2468 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2469 {
2470         const struct kvm_io_range *r1 = p1;
2471         const struct kvm_io_range *r2 = p2;
2472
2473         if (r1->addr < r2->addr)
2474                 return -1;
2475         if (r1->addr + r1->len > r2->addr + r2->len)
2476                 return 1;
2477         return 0;
2478 }
2479
2480 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2481                           gpa_t addr, int len)
2482 {
2483         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2484                 .addr = addr,
2485                 .len = len,
2486                 .dev = dev,
2487         };
2488
2489         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2490                 kvm_io_bus_sort_cmp, NULL);
2491
2492         return 0;
2493 }
2494
2495 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2496                              gpa_t addr, int len)
2497 {
2498         struct kvm_io_range *range, key;
2499         int off;
2500
2501         key = (struct kvm_io_range) {
2502                 .addr = addr,
2503                 .len = len,
2504         };
2505
2506         range = bsearch(&key, bus->range, bus->dev_count,
2507                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2508         if (range == NULL)
2509                 return -ENOENT;
2510
2511         off = range - bus->range;
2512
2513         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2514                 off--;
2515
2516         return off;
2517 }
2518
2519 /* kvm_io_bus_write - called under kvm->slots_lock */
2520 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2521                      int len, const void *val)
2522 {
2523         int idx;
2524         struct kvm_io_bus *bus;
2525         struct kvm_io_range range;
2526
2527         range = (struct kvm_io_range) {
2528                 .addr = addr,
2529                 .len = len,
2530         };
2531
2532         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2533         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2534         if (idx < 0)
2535                 return -EOPNOTSUPP;
2536
2537         while (idx < bus->dev_count &&
2538                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2539                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2540                         return 0;
2541                 idx++;
2542         }
2543
2544         return -EOPNOTSUPP;
2545 }
2546
2547 /* kvm_io_bus_read - called under kvm->slots_lock */
2548 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2549                     int len, void *val)
2550 {
2551         int idx;
2552         struct kvm_io_bus *bus;
2553         struct kvm_io_range range;
2554
2555         range = (struct kvm_io_range) {
2556                 .addr = addr,
2557                 .len = len,
2558         };
2559
2560         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2561         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2562         if (idx < 0)
2563                 return -EOPNOTSUPP;
2564
2565         while (idx < bus->dev_count &&
2566                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2567                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2568                         return 0;
2569                 idx++;
2570         }
2571
2572         return -EOPNOTSUPP;
2573 }
2574
2575 /* Caller must hold slots_lock. */
2576 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2577                             int len, struct kvm_io_device *dev)
2578 {
2579         struct kvm_io_bus *new_bus, *bus;
2580
2581         bus = kvm->buses[bus_idx];
2582         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2583                 return -ENOSPC;
2584
2585         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2586                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2587         if (!new_bus)
2588                 return -ENOMEM;
2589         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2590                sizeof(struct kvm_io_range)));
2591         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2592         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2593         synchronize_srcu_expedited(&kvm->srcu);
2594         kfree(bus);
2595
2596         return 0;
2597 }
2598
2599 /* Caller must hold slots_lock. */
2600 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2601                               struct kvm_io_device *dev)
2602 {
2603         int i, r;
2604         struct kvm_io_bus *new_bus, *bus;
2605
2606         bus = kvm->buses[bus_idx];
2607         r = -ENOENT;
2608         for (i = 0; i < bus->dev_count; i++)
2609                 if (bus->range[i].dev == dev) {
2610                         r = 0;
2611                         break;
2612                 }
2613
2614         if (r)
2615                 return r;
2616
2617         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2618                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2619         if (!new_bus)
2620                 return -ENOMEM;
2621
2622         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2623         new_bus->dev_count--;
2624         memcpy(new_bus->range + i, bus->range + i + 1,
2625                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2626
2627         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2628         synchronize_srcu_expedited(&kvm->srcu);
2629         kfree(bus);
2630         return r;
2631 }
2632
2633 static struct notifier_block kvm_cpu_notifier = {
2634         .notifier_call = kvm_cpu_hotplug,
2635 };
2636
2637 static int vm_stat_get(void *_offset, u64 *val)
2638 {
2639         unsigned offset = (long)_offset;
2640         struct kvm *kvm;
2641
2642         *val = 0;
2643         raw_spin_lock(&kvm_lock);
2644         list_for_each_entry(kvm, &vm_list, vm_list)
2645                 *val += *(u32 *)((void *)kvm + offset);
2646         raw_spin_unlock(&kvm_lock);
2647         return 0;
2648 }
2649
2650 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2651
2652 static int vcpu_stat_get(void *_offset, u64 *val)
2653 {
2654         unsigned offset = (long)_offset;
2655         struct kvm *kvm;
2656         struct kvm_vcpu *vcpu;
2657         int i;
2658
2659         *val = 0;
2660         raw_spin_lock(&kvm_lock);
2661         list_for_each_entry(kvm, &vm_list, vm_list)
2662                 kvm_for_each_vcpu(i, vcpu, kvm)
2663                         *val += *(u32 *)((void *)vcpu + offset);
2664
2665         raw_spin_unlock(&kvm_lock);
2666         return 0;
2667 }
2668
2669 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2670
2671 static const struct file_operations *stat_fops[] = {
2672         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2673         [KVM_STAT_VM]   = &vm_stat_fops,
2674 };
2675
2676 static int kvm_init_debug(void)
2677 {
2678         int r = -EFAULT;
2679         struct kvm_stats_debugfs_item *p;
2680
2681         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2682         if (kvm_debugfs_dir == NULL)
2683                 goto out;
2684
2685         for (p = debugfs_entries; p->name; ++p) {
2686                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2687                                                 (void *)(long)p->offset,
2688                                                 stat_fops[p->kind]);
2689                 if (p->dentry == NULL)
2690                         goto out_dir;
2691         }
2692
2693         return 0;
2694
2695 out_dir:
2696         debugfs_remove_recursive(kvm_debugfs_dir);
2697 out:
2698         return r;
2699 }
2700
2701 static void kvm_exit_debug(void)
2702 {
2703         struct kvm_stats_debugfs_item *p;
2704
2705         for (p = debugfs_entries; p->name; ++p)
2706                 debugfs_remove(p->dentry);
2707         debugfs_remove(kvm_debugfs_dir);
2708 }
2709
2710 static int kvm_suspend(void)
2711 {
2712         if (kvm_usage_count)
2713                 hardware_disable_nolock(NULL);
2714         return 0;
2715 }
2716
2717 static void kvm_resume(void)
2718 {
2719         if (kvm_usage_count) {
2720                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2721                 hardware_enable_nolock(NULL);
2722         }
2723 }
2724
2725 static struct syscore_ops kvm_syscore_ops = {
2726         .suspend = kvm_suspend,
2727         .resume = kvm_resume,
2728 };
2729
2730 static inline
2731 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2732 {
2733         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2734 }
2735
2736 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2737 {
2738         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2739
2740         kvm_arch_vcpu_load(vcpu, cpu);
2741 }
2742
2743 static void kvm_sched_out(struct preempt_notifier *pn,
2744                           struct task_struct *next)
2745 {
2746         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2747
2748         kvm_arch_vcpu_put(vcpu);
2749 }
2750
2751 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2752                   struct module *module)
2753 {
2754         int r;
2755         int cpu;
2756
2757         r = kvm_arch_init(opaque);
2758         if (r)
2759                 goto out_fail;
2760
2761         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2762                 r = -ENOMEM;
2763                 goto out_free_0;
2764         }
2765
2766         r = kvm_arch_hardware_setup();
2767         if (r < 0)
2768                 goto out_free_0a;
2769
2770         for_each_online_cpu(cpu) {
2771                 smp_call_function_single(cpu,
2772                                 kvm_arch_check_processor_compat,
2773                                 &r, 1);
2774                 if (r < 0)
2775                         goto out_free_1;
2776         }
2777
2778         r = register_cpu_notifier(&kvm_cpu_notifier);
2779         if (r)
2780                 goto out_free_2;
2781         register_reboot_notifier(&kvm_reboot_notifier);
2782
2783         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2784         if (!vcpu_align)
2785                 vcpu_align = __alignof__(struct kvm_vcpu);
2786         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2787                                            0, NULL);
2788         if (!kvm_vcpu_cache) {
2789                 r = -ENOMEM;
2790                 goto out_free_3;
2791         }
2792
2793         r = kvm_async_pf_init();
2794         if (r)
2795                 goto out_free;
2796
2797         kvm_chardev_ops.owner = module;
2798         kvm_vm_fops.owner = module;
2799         kvm_vcpu_fops.owner = module;
2800
2801         r = misc_register(&kvm_dev);
2802         if (r) {
2803                 printk(KERN_ERR "kvm: misc device register failed\n");
2804                 goto out_unreg;
2805         }
2806
2807         register_syscore_ops(&kvm_syscore_ops);
2808
2809         kvm_preempt_ops.sched_in = kvm_sched_in;
2810         kvm_preempt_ops.sched_out = kvm_sched_out;
2811
2812         r = kvm_init_debug();
2813         if (r) {
2814                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2815                 goto out_undebugfs;
2816         }
2817
2818         return 0;
2819
2820 out_undebugfs:
2821         unregister_syscore_ops(&kvm_syscore_ops);
2822 out_unreg:
2823         kvm_async_pf_deinit();
2824 out_free:
2825         kmem_cache_destroy(kvm_vcpu_cache);
2826 out_free_3:
2827         unregister_reboot_notifier(&kvm_reboot_notifier);
2828         unregister_cpu_notifier(&kvm_cpu_notifier);
2829 out_free_2:
2830 out_free_1:
2831         kvm_arch_hardware_unsetup();
2832 out_free_0a:
2833         free_cpumask_var(cpus_hardware_enabled);
2834 out_free_0:
2835         kvm_arch_exit();
2836 out_fail:
2837         return r;
2838 }
2839 EXPORT_SYMBOL_GPL(kvm_init);
2840
2841 void kvm_exit(void)
2842 {
2843         kvm_exit_debug();
2844         misc_deregister(&kvm_dev);
2845         kmem_cache_destroy(kvm_vcpu_cache);
2846         kvm_async_pf_deinit();
2847         unregister_syscore_ops(&kvm_syscore_ops);
2848         unregister_reboot_notifier(&kvm_reboot_notifier);
2849         unregister_cpu_notifier(&kvm_cpu_notifier);
2850         on_each_cpu(hardware_disable_nolock, NULL, 1);
2851         kvm_arch_hardware_unsetup();
2852         kvm_arch_exit();
2853         free_cpumask_var(cpus_hardware_enabled);
2854 }
2855 EXPORT_SYMBOL_GPL(kvm_exit);