virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/syscore_ops.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/pgtable.h>
  57
  58 #include "coalesced_mmio.h"
  59 #include "async_pf.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/kvm.h>
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 /*
  68  * Ordering of locks:
  69  *
  70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  71  */
  72
  73 DEFINE_RAW_SPINLOCK(kvm_lock);
  74 LIST_HEAD(vm_list);
  75
  76 static cpumask_var_t cpus_hardware_enabled;
  77 static int kvm_usage_count = 0;
  78 static atomic_t hardware_enable_failed;
  79
  80 struct kmem_cache *kvm_vcpu_cache;
  81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  82
  83 static __read_mostly struct preempt_ops kvm_preempt_ops;
  84
  85 struct dentry *kvm_debugfs_dir;
  86
  87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  88                            unsigned long arg);
  89 #ifdef CONFIG_COMPAT
  90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  91                                   unsigned long arg);
  92 #endif
  93 static int hardware_enable_all(void);
  94 static void hardware_disable_all(void);
  95
  96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  97
  98 bool kvm_rebooting;
  99 EXPORT_SYMBOL_GPL(kvm_rebooting);
 100
 101 static bool largepages_enabled = true;
 102
 103 bool kvm_is_mmio_pfn(pfn_t pfn)
 104 {
 105         if (pfn_valid(pfn)) {
 106                 int reserved;
 107                 struct page *tail = pfn_to_page(pfn);
 108                 struct page *head = compound_trans_head(tail);
 109                 reserved = PageReserved(head);
 110                 if (head != tail) {
 111                         /*
 112                          * "head" is not a dangling pointer
 113                          * (compound_trans_head takes care of that)
 114                          * but the hugepage may have been splitted
 115                          * from under us (and we may not hold a
 116                          * reference count on the head page so it can
 117                          * be reused before we run PageReferenced), so
 118                          * we've to check PageTail before returning
 119                          * what we just read.
 120                          */
 121                         smp_rmb();
 122                         if (PageTail(tail))
 123                                 return reserved;
 124                 }
 125                 return PageReserved(tail);
 126         }
 127
 128         return true;
 129 }
 130
 131 /*
 132  * Switches to specified vcpu, until a matching vcpu_put()
 133  */
 134 int vcpu_load(struct kvm_vcpu *vcpu)
 135 {
 136         int cpu;
 137
 138         if (mutex_lock_killable(&vcpu->mutex))
 139                 return -EINTR;
 140         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 141                 /* The thread running this VCPU changed. */
 142                 struct pid *oldpid = vcpu->pid;
 143                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 144                 rcu_assign_pointer(vcpu->pid, newpid);
 145                 synchronize_rcu();
 146                 put_pid(oldpid);
 147         }
 148         cpu = get_cpu();
 149         preempt_notifier_register(&vcpu->preempt_notifier);
 150         kvm_arch_vcpu_load(vcpu, cpu);
 151         put_cpu();
 152         return 0;
 153 }
 154
 155 void vcpu_put(struct kvm_vcpu *vcpu)
 156 {
 157         preempt_disable();
 158         kvm_arch_vcpu_put(vcpu);
 159         preempt_notifier_unregister(&vcpu->preempt_notifier);
 160         preempt_enable();
 161         mutex_unlock(&vcpu->mutex);
 162 }
 163
 164 static void ack_flush(void *_completed)
 165 {
 166 }
 167
 168 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 169 {
 170         int i, cpu, me;
 171         cpumask_var_t cpus;
 172         bool called = true;
 173         struct kvm_vcpu *vcpu;
 174
 175         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 176
 177         me = get_cpu();
 178         kvm_for_each_vcpu(i, vcpu, kvm) {
 179                 kvm_make_request(req, vcpu);
 180                 cpu = vcpu->cpu;
 181
 182                 /* Set ->requests bit before we read ->mode */
 183                 smp_mb();
 184
 185                 if (cpus != NULL && cpu != -1 && cpu != me &&
 186                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 187                         cpumask_set_cpu(cpu, cpus);
 188         }
 189         if (unlikely(cpus == NULL))
 190                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 191         else if (!cpumask_empty(cpus))
 192                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 193         else
 194                 called = false;
 195         put_cpu();
 196         free_cpumask_var(cpus);
 197         return called;
 198 }
 199
 200 void kvm_flush_remote_tlbs(struct kvm *kvm)
 201 {
 202         long dirty_count = kvm->tlbs_dirty;
 203
 204         smp_mb();
 205         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 206                 ++kvm->stat.remote_tlb_flush;
 207         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 208 }
 209
 210 void kvm_reload_remote_mmus(struct kvm *kvm)
 211 {
 212         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 213 }
 214
 215 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
 216 {
 217         make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
 218 }
 219
 220 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 221 {
 222         struct page *page;
 223         int r;
 224
 225         mutex_init(&vcpu->mutex);
 226         vcpu->cpu = -1;
 227         vcpu->kvm = kvm;
 228         vcpu->vcpu_id = id;
 229         vcpu->pid = NULL;
 230         init_waitqueue_head(&vcpu->wq);
 231         kvm_async_pf_vcpu_init(vcpu);
 232
 233         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 234         if (!page) {
 235                 r = -ENOMEM;
 236                 goto fail;
 237         }
 238         vcpu->run = page_address(page);
 239
 240         kvm_vcpu_set_in_spin_loop(vcpu, false);
 241         kvm_vcpu_set_dy_eligible(vcpu, false);
 242
 243         r = kvm_arch_vcpu_init(vcpu);
 244         if (r < 0)
 245                 goto fail_free_run;
 246         return 0;
 247
 248 fail_free_run:
 249         free_page((unsigned long)vcpu->run);
 250 fail:
 251         return r;
 252 }
 253 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 254
 255 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 256 {
 257         put_pid(vcpu->pid);
 258         kvm_arch_vcpu_uninit(vcpu);
 259         free_page((unsigned long)vcpu->run);
 260 }
 261 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 262
 263 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 264 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 265 {
 266         return container_of(mn, struct kvm, mmu_notifier);
 267 }
 268
 269 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 270                                              struct mm_struct *mm,
 271                                              unsigned long address)
 272 {
 273         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 274         int need_tlb_flush, idx;
 275
 276         /*
 277          * When ->invalidate_page runs, the linux pte has been zapped
 278          * already but the page is still allocated until
 279          * ->invalidate_page returns. So if we increase the sequence
 280          * here the kvm page fault will notice if the spte can't be
 281          * established because the page is going to be freed. If
 282          * instead the kvm page fault establishes the spte before
 283          * ->invalidate_page runs, kvm_unmap_hva will release it
 284          * before returning.
 285          *
 286          * The sequence increase only need to be seen at spin_unlock
 287          * time, and not at spin_lock time.
 288          *
 289          * Increasing the sequence after the spin_unlock would be
 290          * unsafe because the kvm page fault could then establish the
 291          * pte after kvm_unmap_hva returned, without noticing the page
 292          * is going to be freed.
 293          */
 294         idx = srcu_read_lock(&kvm->srcu);
 295         spin_lock(&kvm->mmu_lock);
 296
 297         kvm->mmu_notifier_seq++;
 298         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 299         /* we've to flush the tlb before the pages can be freed */
 300         if (need_tlb_flush)
 301                 kvm_flush_remote_tlbs(kvm);
 302
 303         spin_unlock(&kvm->mmu_lock);
 304         srcu_read_unlock(&kvm->srcu, idx);
 305 }
 306
 307 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 308                                         struct mm_struct *mm,
 309                                         unsigned long address,
 310                                         pte_t pte)
 311 {
 312         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 313         int idx;
 314
 315         idx = srcu_read_lock(&kvm->srcu);
 316         spin_lock(&kvm->mmu_lock);
 317         kvm->mmu_notifier_seq++;
 318         kvm_set_spte_hva(kvm, address, pte);
 319         spin_unlock(&kvm->mmu_lock);
 320         srcu_read_unlock(&kvm->srcu, idx);
 321 }
 322
 323 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 324                                                     struct mm_struct *mm,
 325                                                     unsigned long start,
 326                                                     unsigned long end)
 327 {
 328         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 329         int need_tlb_flush = 0, idx;
 330
 331         idx = srcu_read_lock(&kvm->srcu);
 332         spin_lock(&kvm->mmu_lock);
 333         /*
 334          * The count increase must become visible at unlock time as no
 335          * spte can be established without taking the mmu_lock and
 336          * count is also read inside the mmu_lock critical section.
 337          */
 338         kvm->mmu_notifier_count++;
 339         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 340         need_tlb_flush |= kvm->tlbs_dirty;
 341         /* we've to flush the tlb before the pages can be freed */
 342         if (need_tlb_flush)
 343                 kvm_flush_remote_tlbs(kvm);
 344
 345         spin_unlock(&kvm->mmu_lock);
 346         srcu_read_unlock(&kvm->srcu, idx);
 347 }
 348
 349 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 350                                                   struct mm_struct *mm,
 351                                                   unsigned long start,
 352                                                   unsigned long end)
 353 {
 354         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 355
 356         spin_lock(&kvm->mmu_lock);
 357         /*
 358          * This sequence increase will notify the kvm page fault that
 359          * the page that is going to be mapped in the spte could have
 360          * been freed.
 361          */
 362         kvm->mmu_notifier_seq++;
 363         smp_wmb();
 364         /*
 365          * The above sequence increase must be visible before the
 366          * below count decrease, which is ensured by the smp_wmb above
 367          * in conjunction with the smp_rmb in mmu_notifier_retry().
 368          */
 369         kvm->mmu_notifier_count--;
 370         spin_unlock(&kvm->mmu_lock);
 371
 372         BUG_ON(kvm->mmu_notifier_count < 0);
 373 }
 374
 375 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 376                                               struct mm_struct *mm,
 377                                               unsigned long address)
 378 {
 379         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 380         int young, idx;
 381
 382         idx = srcu_read_lock(&kvm->srcu);
 383         spin_lock(&kvm->mmu_lock);
 384
 385         young = kvm_age_hva(kvm, address);
 386         if (young)
 387                 kvm_flush_remote_tlbs(kvm);
 388
 389         spin_unlock(&kvm->mmu_lock);
 390         srcu_read_unlock(&kvm->srcu, idx);
 391
 392         return young;
 393 }
 394
 395 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 396                                        struct mm_struct *mm,
 397                                        unsigned long address)
 398 {
 399         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 400         int young, idx;
 401
 402         idx = srcu_read_lock(&kvm->srcu);
 403         spin_lock(&kvm->mmu_lock);
 404         young = kvm_test_age_hva(kvm, address);
 405         spin_unlock(&kvm->mmu_lock);
 406         srcu_read_unlock(&kvm->srcu, idx);
 407
 408         return young;
 409 }
 410
 411 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 412                                      struct mm_struct *mm)
 413 {
 414         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 415         int idx;
 416
 417         idx = srcu_read_lock(&kvm->srcu);
 418         kvm_arch_flush_shadow_all(kvm);
 419         srcu_read_unlock(&kvm->srcu, idx);
 420 }
 421
 422 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 423         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 424         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 425         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 426         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 427         .test_young             = kvm_mmu_notifier_test_young,
 428         .change_pte             = kvm_mmu_notifier_change_pte,
 429         .release                = kvm_mmu_notifier_release,
 430 };
 431
 432 static int kvm_init_mmu_notifier(struct kvm *kvm)
 433 {
 434         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 435         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 436 }
 437
 438 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 439
 440 static int kvm_init_mmu_notifier(struct kvm *kvm)
 441 {
 442         return 0;
 443 }
 444
 445 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 446
 447 static void kvm_init_memslots_id(struct kvm *kvm)
 448 {
 449         int i;
 450         struct kvm_memslots *slots = kvm->memslots;
 451
 452         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 453                 slots->id_to_index[i] = slots->memslots[i].id = i;
 454 }
 455
 456 static struct kvm *kvm_create_vm(unsigned long type)
 457 {
 458         int r, i;
 459         struct kvm *kvm = kvm_arch_alloc_vm();
 460
 461         if (!kvm)
 462                 return ERR_PTR(-ENOMEM);
 463
 464         r = kvm_arch_init_vm(kvm, type);
 465         if (r)
 466                 goto out_err_nodisable;
 467
 468         r = hardware_enable_all();
 469         if (r)
 470                 goto out_err_nodisable;
 471
 472 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 473         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 474         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 475 #endif
 476
 477         r = -ENOMEM;
 478         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 479         if (!kvm->memslots)
 480                 goto out_err_nosrcu;
 481         kvm_init_memslots_id(kvm);
 482         if (init_srcu_struct(&kvm->srcu))
 483                 goto out_err_nosrcu;
 484         for (i = 0; i < KVM_NR_BUSES; i++) {
 485                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 486                                         GFP_KERNEL);
 487                 if (!kvm->buses[i])
 488                         goto out_err;
 489         }
 490
 491         spin_lock_init(&kvm->mmu_lock);
 492         kvm->mm = current->mm;
 493         atomic_inc(&kvm->mm->mm_count);
 494         kvm_eventfd_init(kvm);
 495         mutex_init(&kvm->lock);
 496         mutex_init(&kvm->irq_lock);
 497         mutex_init(&kvm->slots_lock);
 498         atomic_set(&kvm->users_count, 1);
 499
 500         r = kvm_init_mmu_notifier(kvm);
 501         if (r)
 502                 goto out_err;
 503
 504         raw_spin_lock(&kvm_lock);
 505         list_add(&kvm->vm_list, &vm_list);
 506         raw_spin_unlock(&kvm_lock);
 507
 508         return kvm;
 509
 510 out_err:
 511         cleanup_srcu_struct(&kvm->srcu);
 512 out_err_nosrcu:
 513         hardware_disable_all();
 514 out_err_nodisable:
 515         for (i = 0; i < KVM_NR_BUSES; i++)
 516                 kfree(kvm->buses[i]);
 517         kfree(kvm->memslots);
 518         kvm_arch_free_vm(kvm);
 519         return ERR_PTR(r);
 520 }
 521
 522 /*
 523  * Avoid using vmalloc for a small buffer.
 524  * Should not be used when the size is statically known.
 525  */
 526 void *kvm_kvzalloc(unsigned long size)
 527 {
 528         if (size > PAGE_SIZE)
 529                 return vzalloc(size);
 530         else
 531                 return kzalloc(size, GFP_KERNEL);
 532 }
 533
 534 void kvm_kvfree(const void *addr)
 535 {
 536         if (is_vmalloc_addr(addr))
 537                 vfree(addr);
 538         else
 539                 kfree(addr);
 540 }
 541
 542 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 543 {
 544         if (!memslot->dirty_bitmap)
 545                 return;
 546
 547         kvm_kvfree(memslot->dirty_bitmap);
 548         memslot->dirty_bitmap = NULL;
 549 }
 550
 551 /*
 552  * Free any memory in @free but not in @dont.
 553  */
 554 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 555                                   struct kvm_memory_slot *dont)
 556 {
 557         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 558                 kvm_destroy_dirty_bitmap(free);
 559
 560         kvm_arch_free_memslot(free, dont);
 561
 562         free->npages = 0;
 563 }
 564
 565 void kvm_free_physmem(struct kvm *kvm)
 566 {
 567         struct kvm_memslots *slots = kvm->memslots;
 568         struct kvm_memory_slot *memslot;
 569
 570         kvm_for_each_memslot(memslot, slots)
 571                 kvm_free_physmem_slot(memslot, NULL);
 572
 573         kfree(kvm->memslots);
 574 }
 575
 576 static void kvm_destroy_vm(struct kvm *kvm)
 577 {
 578         int i;
 579         struct mm_struct *mm = kvm->mm;
 580
 581         kvm_arch_sync_events(kvm);
 582         raw_spin_lock(&kvm_lock);
 583         list_del(&kvm->vm_list);
 584         raw_spin_unlock(&kvm_lock);
 585         kvm_free_irq_routing(kvm);
 586         for (i = 0; i < KVM_NR_BUSES; i++)
 587                 kvm_io_bus_destroy(kvm->buses[i]);
 588         kvm_coalesced_mmio_free(kvm);
 589 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 590         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 591 #else
 592         kvm_arch_flush_shadow_all(kvm);
 593 #endif
 594         kvm_arch_destroy_vm(kvm);
 595         kvm_free_physmem(kvm);
 596         cleanup_srcu_struct(&kvm->srcu);
 597         kvm_arch_free_vm(kvm);
 598         hardware_disable_all();
 599         mmdrop(mm);
 600 }
 601
 602 void kvm_get_kvm(struct kvm *kvm)
 603 {
 604         atomic_inc(&kvm->users_count);
 605 }
 606 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 607
 608 void kvm_put_kvm(struct kvm *kvm)
 609 {
 610         if (atomic_dec_and_test(&kvm->users_count))
 611                 kvm_destroy_vm(kvm);
 612 }
 613 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 614
 615
 616 static int kvm_vm_release(struct inode *inode, struct file *filp)
 617 {
 618         struct kvm *kvm = filp->private_data;
 619
 620         kvm_irqfd_release(kvm);
 621
 622         kvm_put_kvm(kvm);
 623         return 0;
 624 }
 625
 626 /*
 627  * Allocation size is twice as large as the actual dirty bitmap size.
 628  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 629  */
 630 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 631 {
 632 #ifndef CONFIG_S390
 633         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 634
 635         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 636         if (!memslot->dirty_bitmap)
 637                 return -ENOMEM;
 638
 639 #endif /* !CONFIG_S390 */
 640         return 0;
 641 }
 642
 643 static int cmp_memslot(const void *slot1, const void *slot2)
 644 {
 645         struct kvm_memory_slot *s1, *s2;
 646
 647         s1 = (struct kvm_memory_slot *)slot1;
 648         s2 = (struct kvm_memory_slot *)slot2;
 649
 650         if (s1->npages < s2->npages)
 651                 return 1;
 652         if (s1->npages > s2->npages)
 653                 return -1;
 654
 655         return 0;
 656 }
 657
 658 /*
 659  * Sort the memslots base on its size, so the larger slots
 660  * will get better fit.
 661  */
 662 static void sort_memslots(struct kvm_memslots *slots)
 663 {
 664         int i;
 665
 666         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 667               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 668
 669         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 670                 slots->id_to_index[slots->memslots[i].id] = i;
 671 }
 672
 673 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
 674 {
 675         if (new) {
 676                 int id = new->id;
 677                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 678                 unsigned long npages = old->npages;
 679
 680                 *old = *new;
 681                 if (new->npages != npages)
 682                         sort_memslots(slots);
 683         }
 684
 685         slots->generation++;
 686 }
 687
 688 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 689 {
 690         u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 691
 692 #ifdef KVM_CAP_READONLY_MEM
 693         valid_flags |= KVM_MEM_READONLY;
 694 #endif
 695
 696         if (mem->flags & ~valid_flags)
 697                 return -EINVAL;
 698
 699         return 0;
 700 }
 701
 702 /*
 703  * Allocate some memory and give it an address in the guest physical address
 704  * space.
 705  *
 706  * Discontiguous memory is allowed, mostly for framebuffers.
 707  *
 708  * Must be called holding mmap_sem for write.
 709  */
 710 int __kvm_set_memory_region(struct kvm *kvm,
 711                             struct kvm_userspace_memory_region *mem,
 712                             int user_alloc)
 713 {
 714         int r;
 715         gfn_t base_gfn;
 716         unsigned long npages;
 717         struct kvm_memory_slot *memslot, *slot;
 718         struct kvm_memory_slot old, new;
 719         struct kvm_memslots *slots, *old_memslots;
 720
 721         r = check_memory_region_flags(mem);
 722         if (r)
 723                 goto out;
 724
 725         r = -EINVAL;
 726         /* General sanity checks */
 727         if (mem->memory_size & (PAGE_SIZE - 1))
 728                 goto out;
 729         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 730                 goto out;
 731         /* We can read the guest memory with __xxx_user() later on. */
 732         if (user_alloc &&
 733             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 734              !access_ok(VERIFY_WRITE,
 735                         (void __user *)(unsigned long)mem->userspace_addr,
 736                         mem->memory_size)))
 737                 goto out;
 738         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 739                 goto out;
 740         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 741                 goto out;
 742
 743         memslot = id_to_memslot(kvm->memslots, mem->slot);
 744         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 745         npages = mem->memory_size >> PAGE_SHIFT;
 746
 747         r = -EINVAL;
 748         if (npages > KVM_MEM_MAX_NR_PAGES)
 749                 goto out;
 750
 751         if (!npages)
 752                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 753
 754         new = old = *memslot;
 755
 756         new.id = mem->slot;
 757         new.base_gfn = base_gfn;
 758         new.npages = npages;
 759         new.flags = mem->flags;
 760
 761         /*
 762          * Disallow changing a memory slot's size or changing anything about
 763          * zero sized slots that doesn't involve making them non-zero.
 764          */
 765         r = -EINVAL;
 766         if (npages && old.npages && npages != old.npages)
 767                 goto out_free;
 768         if (!npages && !old.npages)
 769                 goto out_free;
 770
 771         /* Check for overlaps */
 772         r = -EEXIST;
 773         kvm_for_each_memslot(slot, kvm->memslots) {
 774                 if (slot->id >= KVM_MEMORY_SLOTS || slot == memslot)
 775                         continue;
 776                 if (!((base_gfn + npages <= slot->base_gfn) ||
 777                       (base_gfn >= slot->base_gfn + slot->npages)))
 778                         goto out_free;
 779         }
 780
 781         /* Free page dirty bitmap if unneeded */
 782         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 783                 new.dirty_bitmap = NULL;
 784
 785         r = -ENOMEM;
 786
 787         /*
 788          * Allocate if a slot is being created.  If modifying a slot,
 789          * the userspace_addr cannot change.
 790          */
 791         if (!old.npages) {
 792                 new.user_alloc = user_alloc;
 793                 new.userspace_addr = mem->userspace_addr;
 794
 795                 if (kvm_arch_create_memslot(&new, npages))
 796                         goto out_free;
 797         } else if (npages && mem->userspace_addr != old.userspace_addr) {
 798                 r = -EINVAL;
 799                 goto out_free;
 800         }
 801
 802         /* Allocate page dirty bitmap if needed */
 803         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 804                 if (kvm_create_dirty_bitmap(&new) < 0)
 805                         goto out_free;
 806                 /* destroy any largepage mappings for dirty tracking */
 807         }
 808
 809         if (!npages || base_gfn != old.base_gfn) {
 810                 struct kvm_memory_slot *slot;
 811
 812                 r = -ENOMEM;
 813                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 814                                 GFP_KERNEL);
 815                 if (!slots)
 816                         goto out_free;
 817                 slot = id_to_memslot(slots, mem->slot);
 818                 slot->flags |= KVM_MEMSLOT_INVALID;
 819
 820                 update_memslots(slots, NULL);
 821
 822                 old_memslots = kvm->memslots;
 823                 rcu_assign_pointer(kvm->memslots, slots);
 824                 synchronize_srcu_expedited(&kvm->srcu);
 825                 /* slot was deleted or moved, clear iommu mapping */
 826                 kvm_iommu_unmap_pages(kvm, &old);
 827                 /* From this point no new shadow pages pointing to a deleted,
 828                  * or moved, memslot will be created.
 829                  *
 830                  * validation of sp->gfn happens in:
 831                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 832                  *      - kvm_is_visible_gfn (mmu_check_roots)
 833                  */
 834                 kvm_arch_flush_shadow_memslot(kvm, slot);
 835                 kfree(old_memslots);
 836         }
 837
 838         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 839         if (r)
 840                 goto out_free;
 841
 842         r = -ENOMEM;
 843         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 844                         GFP_KERNEL);
 845         if (!slots)
 846                 goto out_free;
 847
 848         /* map new memory slot into the iommu */
 849         if (npages) {
 850                 r = kvm_iommu_map_pages(kvm, &new);
 851                 if (r)
 852                         goto out_slots;
 853         }
 854
 855         /* actual memory is freed via old in kvm_free_physmem_slot below */
 856         if (!npages) {
 857                 new.dirty_bitmap = NULL;
 858                 memset(&new.arch, 0, sizeof(new.arch));
 859         }
 860
 861         update_memslots(slots, &new);
 862         old_memslots = kvm->memslots;
 863         rcu_assign_pointer(kvm->memslots, slots);
 864         synchronize_srcu_expedited(&kvm->srcu);
 865
 866         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 867
 868         kvm_free_physmem_slot(&old, &new);
 869         kfree(old_memslots);
 870
 871         return 0;
 872
 873 out_slots:
 874         kfree(slots);
 875 out_free:
 876         kvm_free_physmem_slot(&new, &old);
 877 out:
 878         return r;
 879
 880 }
 881 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 882
 883 int kvm_set_memory_region(struct kvm *kvm,
 884                           struct kvm_userspace_memory_region *mem,
 885                           int user_alloc)
 886 {
 887         int r;
 888
 889         mutex_lock(&kvm->slots_lock);
 890         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 891         mutex_unlock(&kvm->slots_lock);
 892         return r;
 893 }
 894 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 895
 896 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 897                                    struct
 898                                    kvm_userspace_memory_region *mem,
 899                                    int user_alloc)
 900 {
 901         if (mem->slot >= KVM_MEMORY_SLOTS)
 902                 return -EINVAL;
 903         return kvm_set_memory_region(kvm, mem, user_alloc);
 904 }
 905
 906 int kvm_get_dirty_log(struct kvm *kvm,
 907                         struct kvm_dirty_log *log, int *is_dirty)
 908 {
 909         struct kvm_memory_slot *memslot;
 910         int r, i;
 911         unsigned long n;
 912         unsigned long any = 0;
 913
 914         r = -EINVAL;
 915         if (log->slot >= KVM_MEMORY_SLOTS)
 916                 goto out;
 917
 918         memslot = id_to_memslot(kvm->memslots, log->slot);
 919         r = -ENOENT;
 920         if (!memslot->dirty_bitmap)
 921                 goto out;
 922
 923         n = kvm_dirty_bitmap_bytes(memslot);
 924
 925         for (i = 0; !any && i < n/sizeof(long); ++i)
 926                 any = memslot->dirty_bitmap[i];
 927
 928         r = -EFAULT;
 929         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 930                 goto out;
 931
 932         if (any)
 933                 *is_dirty = 1;
 934
 935         r = 0;
 936 out:
 937         return r;
 938 }
 939
 940 bool kvm_largepages_enabled(void)
 941 {
 942         return largepages_enabled;
 943 }
 944
 945 void kvm_disable_largepages(void)
 946 {
 947         largepages_enabled = false;
 948 }
 949 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 950
 951 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 952 {
 953         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 954 }
 955 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 956
 957 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 958 {
 959         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 960
 961         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
 962               memslot->flags & KVM_MEMSLOT_INVALID)
 963                 return 0;
 964
 965         return 1;
 966 }
 967 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 968
 969 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 970 {
 971         struct vm_area_struct *vma;
 972         unsigned long addr, size;
 973
 974         size = PAGE_SIZE;
 975
 976         addr = gfn_to_hva(kvm, gfn);
 977         if (kvm_is_error_hva(addr))
 978                 return PAGE_SIZE;
 979
 980         down_read(&current->mm->mmap_sem);
 981         vma = find_vma(current->mm, addr);
 982         if (!vma)
 983                 goto out;
 984
 985         size = vma_kernel_pagesize(vma);
 986
 987 out:
 988         up_read(&current->mm->mmap_sem);
 989
 990         return size;
 991 }
 992
 993 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
 994 {
 995         return slot->flags & KVM_MEM_READONLY;
 996 }
 997
 998 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
 999                                        gfn_t *nr_pages, bool write)
1000 {
1001         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1002                 return KVM_HVA_ERR_BAD;
1003
1004         if (memslot_is_readonly(slot) && write)
1005                 return KVM_HVA_ERR_RO_BAD;
1006
1007         if (nr_pages)
1008                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1009
1010         return __gfn_to_hva_memslot(slot, gfn);
1011 }
1012
1013 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1014                                      gfn_t *nr_pages)
1015 {
1016         return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1017 }
1018
1019 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1020                                  gfn_t gfn)
1021 {
1022         return gfn_to_hva_many(slot, gfn, NULL);
1023 }
1024 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1025
1026 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1027 {
1028         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1029 }
1030 EXPORT_SYMBOL_GPL(gfn_to_hva);
1031
1032 /*
1033  * The hva returned by this function is only allowed to be read.
1034  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1035  */
1036 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1037 {
1038         return __gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL, false);
1039 }
1040
1041 static int kvm_read_hva(void *data, void __user *hva, int len)
1042 {
1043         return __copy_from_user(data, hva, len);
1044 }
1045
1046 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1047 {
1048         return __copy_from_user_inatomic(data, hva, len);
1049 }
1050
1051 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1052         unsigned long start, int write, struct page **page)
1053 {
1054         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1055
1056         if (write)
1057                 flags |= FOLL_WRITE;
1058
1059         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1060 }
1061
1062 static inline int check_user_page_hwpoison(unsigned long addr)
1063 {
1064         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1065
1066         rc = __get_user_pages(current, current->mm, addr, 1,
1067                               flags, NULL, NULL, NULL);
1068         return rc == -EHWPOISON;
1069 }
1070
1071 /*
1072  * The atomic path to get the writable pfn which will be stored in @pfn,
1073  * true indicates success, otherwise false is returned.
1074  */
1075 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1076                             bool write_fault, bool *writable, pfn_t *pfn)
1077 {
1078         struct page *page[1];
1079         int npages;
1080
1081         if (!(async || atomic))
1082                 return false;
1083
1084         /*
1085          * Fast pin a writable pfn only if it is a write fault request
1086          * or the caller allows to map a writable pfn for a read fault
1087          * request.
1088          */
1089         if (!(write_fault || writable))
1090                 return false;
1091
1092         npages = __get_user_pages_fast(addr, 1, 1, page);
1093         if (npages == 1) {
1094                 *pfn = page_to_pfn(page[0]);
1095
1096                 if (writable)
1097                         *writable = true;
1098                 return true;
1099         }
1100
1101         return false;
1102 }
1103
1104 /*
1105  * The slow path to get the pfn of the specified host virtual address,
1106  * 1 indicates success, -errno is returned if error is detected.
1107  */
1108 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1109                            bool *writable, pfn_t *pfn)
1110 {
1111         struct page *page[1];
1112         int npages = 0;
1113
1114         might_sleep();
1115
1116         if (writable)
1117                 *writable = write_fault;
1118
1119         if (async) {
1120                 down_read(&current->mm->mmap_sem);
1121                 npages = get_user_page_nowait(current, current->mm,
1122                                               addr, write_fault, page);
1123                 up_read(&current->mm->mmap_sem);
1124         } else
1125                 npages = get_user_pages_fast(addr, 1, write_fault,
1126                                              page);
1127         if (npages != 1)
1128                 return npages;
1129
1130         /* map read fault as writable if possible */
1131         if (unlikely(!write_fault) && writable) {
1132                 struct page *wpage[1];
1133
1134                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1135                 if (npages == 1) {
1136                         *writable = true;
1137                         put_page(page[0]);
1138                         page[0] = wpage[0];
1139                 }
1140
1141                 npages = 1;
1142         }
1143         *pfn = page_to_pfn(page[0]);
1144         return npages;
1145 }
1146
1147 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1148 {
1149         if (unlikely(!(vma->vm_flags & VM_READ)))
1150                 return false;
1151
1152         if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1153                 return false;
1154
1155         return true;
1156 }
1157
1158 /*
1159  * Pin guest page in memory and return its pfn.
1160  * @addr: host virtual address which maps memory to the guest
1161  * @atomic: whether this function can sleep
1162  * @async: whether this function need to wait IO complete if the
1163  *         host page is not in the memory
1164  * @write_fault: whether we should get a writable host page
1165  * @writable: whether it allows to map a writable host page for !@write_fault
1166  *
1167  * The function will map a writable host page for these two cases:
1168  * 1): @write_fault = true
1169  * 2): @write_fault = false && @writable, @writable will tell the caller
1170  *     whether the mapping is writable.
1171  */
1172 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1173                         bool write_fault, bool *writable)
1174 {
1175         struct vm_area_struct *vma;
1176         pfn_t pfn = 0;
1177         int npages;
1178
1179         /* we can do it either atomically or asynchronously, not both */
1180         BUG_ON(atomic && async);
1181
1182         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1183                 return pfn;
1184
1185         if (atomic)
1186                 return KVM_PFN_ERR_FAULT;
1187
1188         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1189         if (npages == 1)
1190                 return pfn;
1191
1192         down_read(&current->mm->mmap_sem);
1193         if (npages == -EHWPOISON ||
1194               (!async && check_user_page_hwpoison(addr))) {
1195                 pfn = KVM_PFN_ERR_HWPOISON;
1196                 goto exit;
1197         }
1198
1199         vma = find_vma_intersection(current->mm, addr, addr + 1);
1200
1201         if (vma == NULL)
1202                 pfn = KVM_PFN_ERR_FAULT;
1203         else if ((vma->vm_flags & VM_PFNMAP)) {
1204                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1205                         vma->vm_pgoff;
1206                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1207         } else {
1208                 if (async && vma_is_valid(vma, write_fault))
1209                         *async = true;
1210                 pfn = KVM_PFN_ERR_FAULT;
1211         }
1212 exit:
1213         up_read(&current->mm->mmap_sem);
1214         return pfn;
1215 }
1216
1217 static pfn_t
1218 __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1219                      bool *async, bool write_fault, bool *writable)
1220 {
1221         unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1222
1223         if (addr == KVM_HVA_ERR_RO_BAD)
1224                 return KVM_PFN_ERR_RO_FAULT;
1225
1226         if (kvm_is_error_hva(addr))
1227                 return KVM_PFN_NOSLOT;
1228
1229         /* Do not map writable pfn in the readonly memslot. */
1230         if (writable && memslot_is_readonly(slot)) {
1231                 *writable = false;
1232                 writable = NULL;
1233         }
1234
1235         return hva_to_pfn(addr, atomic, async, write_fault,
1236                           writable);
1237 }
1238
1239 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1240                           bool write_fault, bool *writable)
1241 {
1242         struct kvm_memory_slot *slot;
1243
1244         if (async)
1245                 *async = false;
1246
1247         slot = gfn_to_memslot(kvm, gfn);
1248
1249         return __gfn_to_pfn_memslot(slot, gfn, atomic, async, write_fault,
1250                                     writable);
1251 }
1252
1253 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1254 {
1255         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1256 }
1257 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1258
1259 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1260                        bool write_fault, bool *writable)
1261 {
1262         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1263 }
1264 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1265
1266 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1267 {
1268         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1269 }
1270 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1271
1272 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1273                       bool *writable)
1274 {
1275         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1276 }
1277 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1278
1279 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1280 {
1281         return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1282 }
1283
1284 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1285 {
1286         return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1287 }
1288 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1289
1290 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1291                                                                   int nr_pages)
1292 {
1293         unsigned long addr;
1294         gfn_t entry;
1295
1296         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1297         if (kvm_is_error_hva(addr))
1298                 return -1;
1299
1300         if (entry < nr_pages)
1301                 return 0;
1302
1303         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1304 }
1305 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1306
1307 static struct page *kvm_pfn_to_page(pfn_t pfn)
1308 {
1309         if (is_error_noslot_pfn(pfn))
1310                 return KVM_ERR_PTR_BAD_PAGE;
1311
1312         if (kvm_is_mmio_pfn(pfn)) {
1313                 WARN_ON(1);
1314                 return KVM_ERR_PTR_BAD_PAGE;
1315         }
1316
1317         return pfn_to_page(pfn);
1318 }
1319
1320 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1321 {
1322         pfn_t pfn;
1323
1324         pfn = gfn_to_pfn(kvm, gfn);
1325
1326         return kvm_pfn_to_page(pfn);
1327 }
1328
1329 EXPORT_SYMBOL_GPL(gfn_to_page);
1330
1331 void kvm_release_page_clean(struct page *page)
1332 {
1333         WARN_ON(is_error_page(page));
1334
1335         kvm_release_pfn_clean(page_to_pfn(page));
1336 }
1337 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1338
1339 void kvm_release_pfn_clean(pfn_t pfn)
1340 {
1341         if (!is_error_noslot_pfn(pfn) && !kvm_is_mmio_pfn(pfn))
1342                 put_page(pfn_to_page(pfn));
1343 }
1344 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1345
1346 void kvm_release_page_dirty(struct page *page)
1347 {
1348         WARN_ON(is_error_page(page));
1349
1350         kvm_release_pfn_dirty(page_to_pfn(page));
1351 }
1352 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1353
1354 void kvm_release_pfn_dirty(pfn_t pfn)
1355 {
1356         kvm_set_pfn_dirty(pfn);
1357         kvm_release_pfn_clean(pfn);
1358 }
1359 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1360
1361 void kvm_set_page_dirty(struct page *page)
1362 {
1363         kvm_set_pfn_dirty(page_to_pfn(page));
1364 }
1365 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1366
1367 void kvm_set_pfn_dirty(pfn_t pfn)
1368 {
1369         if (!kvm_is_mmio_pfn(pfn)) {
1370                 struct page *page = pfn_to_page(pfn);
1371                 if (!PageReserved(page))
1372                         SetPageDirty(page);
1373         }
1374 }
1375 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1376
1377 void kvm_set_pfn_accessed(pfn_t pfn)
1378 {
1379         if (!kvm_is_mmio_pfn(pfn))
1380                 mark_page_accessed(pfn_to_page(pfn));
1381 }
1382 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1383
1384 void kvm_get_pfn(pfn_t pfn)
1385 {
1386         if (!kvm_is_mmio_pfn(pfn))
1387                 get_page(pfn_to_page(pfn));
1388 }
1389 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1390
1391 static int next_segment(unsigned long len, int offset)
1392 {
1393         if (len > PAGE_SIZE - offset)
1394                 return PAGE_SIZE - offset;
1395         else
1396                 return len;
1397 }
1398
1399 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1400                         int len)
1401 {
1402         int r;
1403         unsigned long addr;
1404
1405         addr = gfn_to_hva_read(kvm, gfn);
1406         if (kvm_is_error_hva(addr))
1407                 return -EFAULT;
1408         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1409         if (r)
1410                 return -EFAULT;
1411         return 0;
1412 }
1413 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1414
1415 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1416 {
1417         gfn_t gfn = gpa >> PAGE_SHIFT;
1418         int seg;
1419         int offset = offset_in_page(gpa);
1420         int ret;
1421
1422         while ((seg = next_segment(len, offset)) != 0) {
1423                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1424                 if (ret < 0)
1425                         return ret;
1426                 offset = 0;
1427                 len -= seg;
1428                 data += seg;
1429                 ++gfn;
1430         }
1431         return 0;
1432 }
1433 EXPORT_SYMBOL_GPL(kvm_read_guest);
1434
1435 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1436                           unsigned long len)
1437 {
1438         int r;
1439         unsigned long addr;
1440         gfn_t gfn = gpa >> PAGE_SHIFT;
1441         int offset = offset_in_page(gpa);
1442
1443         addr = gfn_to_hva_read(kvm, gfn);
1444         if (kvm_is_error_hva(addr))
1445                 return -EFAULT;
1446         pagefault_disable();
1447         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1448         pagefault_enable();
1449         if (r)
1450                 return -EFAULT;
1451         return 0;
1452 }
1453 EXPORT_SYMBOL(kvm_read_guest_atomic);
1454
1455 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1456                          int offset, int len)
1457 {
1458         int r;
1459         unsigned long addr;
1460
1461         addr = gfn_to_hva(kvm, gfn);
1462         if (kvm_is_error_hva(addr))
1463                 return -EFAULT;
1464         r = __copy_to_user((void __user *)addr + offset, data, len);
1465         if (r)
1466                 return -EFAULT;
1467         mark_page_dirty(kvm, gfn);
1468         return 0;
1469 }
1470 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1471
1472 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1473                     unsigned long len)
1474 {
1475         gfn_t gfn = gpa >> PAGE_SHIFT;
1476         int seg;
1477         int offset = offset_in_page(gpa);
1478         int ret;
1479
1480         while ((seg = next_segment(len, offset)) != 0) {
1481                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1482                 if (ret < 0)
1483                         return ret;
1484                 offset = 0;
1485                 len -= seg;
1486                 data += seg;
1487                 ++gfn;
1488         }
1489         return 0;
1490 }
1491
1492 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1493                               gpa_t gpa)
1494 {
1495         struct kvm_memslots *slots = kvm_memslots(kvm);
1496         int offset = offset_in_page(gpa);
1497         gfn_t gfn = gpa >> PAGE_SHIFT;
1498
1499         ghc->gpa = gpa;
1500         ghc->generation = slots->generation;
1501         ghc->memslot = gfn_to_memslot(kvm, gfn);
1502         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1503         if (!kvm_is_error_hva(ghc->hva))
1504                 ghc->hva += offset;
1505         else
1506                 return -EFAULT;
1507
1508         return 0;
1509 }
1510 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1511
1512 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1513                            void *data, unsigned long len)
1514 {
1515         struct kvm_memslots *slots = kvm_memslots(kvm);
1516         int r;
1517
1518         if (slots->generation != ghc->generation)
1519                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1520
1521         if (kvm_is_error_hva(ghc->hva))
1522                 return -EFAULT;
1523
1524         r = __copy_to_user((void __user *)ghc->hva, data, len);
1525         if (r)
1526                 return -EFAULT;
1527         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1528
1529         return 0;
1530 }
1531 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1532
1533 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1534                            void *data, unsigned long len)
1535 {
1536         struct kvm_memslots *slots = kvm_memslots(kvm);
1537         int r;
1538
1539         if (slots->generation != ghc->generation)
1540                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1541
1542         if (kvm_is_error_hva(ghc->hva))
1543                 return -EFAULT;
1544
1545         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1546         if (r)
1547                 return -EFAULT;
1548
1549         return 0;
1550 }
1551 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1552
1553 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1554 {
1555         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1556                                     offset, len);
1557 }
1558 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1559
1560 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1561 {
1562         gfn_t gfn = gpa >> PAGE_SHIFT;
1563         int seg;
1564         int offset = offset_in_page(gpa);
1565         int ret;
1566
1567         while ((seg = next_segment(len, offset)) != 0) {
1568                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1569                 if (ret < 0)
1570                         return ret;
1571                 offset = 0;
1572                 len -= seg;
1573                 ++gfn;
1574         }
1575         return 0;
1576 }
1577 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1578
1579 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1580                              gfn_t gfn)
1581 {
1582         if (memslot && memslot->dirty_bitmap) {
1583                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1584
1585                 set_bit_le(rel_gfn, memslot->dirty_bitmap);
1586         }
1587 }
1588
1589 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1590 {
1591         struct kvm_memory_slot *memslot;
1592
1593         memslot = gfn_to_memslot(kvm, gfn);
1594         mark_page_dirty_in_slot(kvm, memslot, gfn);
1595 }
1596
1597 /*
1598  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1599  */
1600 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1601 {
1602         DEFINE_WAIT(wait);
1603
1604         for (;;) {
1605                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1606
1607                 if (kvm_arch_vcpu_runnable(vcpu)) {
1608                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1609                         break;
1610                 }
1611                 if (kvm_cpu_has_pending_timer(vcpu))
1612                         break;
1613                 if (signal_pending(current))
1614                         break;
1615
1616                 schedule();
1617         }
1618
1619         finish_wait(&vcpu->wq, &wait);
1620 }
1621
1622 #ifndef CONFIG_S390
1623 /*
1624  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1625  */
1626 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1627 {
1628         int me;
1629         int cpu = vcpu->cpu;
1630         wait_queue_head_t *wqp;
1631
1632         wqp = kvm_arch_vcpu_wq(vcpu);
1633         if (waitqueue_active(wqp)) {
1634                 wake_up_interruptible(wqp);
1635                 ++vcpu->stat.halt_wakeup;
1636         }
1637
1638         me = get_cpu();
1639         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1640                 if (kvm_arch_vcpu_should_kick(vcpu))
1641                         smp_send_reschedule(cpu);
1642         put_cpu();
1643 }
1644 #endif /* !CONFIG_S390 */
1645
1646 void kvm_resched(struct kvm_vcpu *vcpu)
1647 {
1648         if (!need_resched())
1649                 return;
1650         cond_resched();
1651 }
1652 EXPORT_SYMBOL_GPL(kvm_resched);
1653
1654 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1655 {
1656         struct pid *pid;
1657         struct task_struct *task = NULL;
1658
1659         rcu_read_lock();
1660         pid = rcu_dereference(target->pid);
1661         if (pid)
1662                 task = get_pid_task(target->pid, PIDTYPE_PID);
1663         rcu_read_unlock();
1664         if (!task)
1665                 return false;
1666         if (task->flags & PF_VCPU) {
1667                 put_task_struct(task);
1668                 return false;
1669         }
1670         if (yield_to(task, 1)) {
1671                 put_task_struct(task);
1672                 return true;
1673         }
1674         put_task_struct(task);
1675         return false;
1676 }
1677 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1678
1679 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1680 /*
1681  * Helper that checks whether a VCPU is eligible for directed yield.
1682  * Most eligible candidate to yield is decided by following heuristics:
1683  *
1684  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1685  *  (preempted lock holder), indicated by @in_spin_loop.
1686  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1687  *
1688  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1689  *  chance last time (mostly it has become eligible now since we have probably
1690  *  yielded to lockholder in last iteration. This is done by toggling
1691  *  @dy_eligible each time a VCPU checked for eligibility.)
1692  *
1693  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1694  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1695  *  burning. Giving priority for a potential lock-holder increases lock
1696  *  progress.
1697  *
1698  *  Since algorithm is based on heuristics, accessing another VCPU data without
1699  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1700  *  and continue with next VCPU and so on.
1701  */
1702 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1703 {
1704         bool eligible;
1705
1706         eligible = !vcpu->spin_loop.in_spin_loop ||
1707                         (vcpu->spin_loop.in_spin_loop &&
1708                          vcpu->spin_loop.dy_eligible);
1709
1710         if (vcpu->spin_loop.in_spin_loop)
1711                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1712
1713         return eligible;
1714 }
1715 #endif
1716 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1717 {
1718         struct kvm *kvm = me->kvm;
1719         struct kvm_vcpu *vcpu;
1720         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1721         int yielded = 0;
1722         int pass;
1723         int i;
1724
1725         kvm_vcpu_set_in_spin_loop(me, true);
1726         /*
1727          * We boost the priority of a VCPU that is runnable but not
1728          * currently running, because it got preempted by something
1729          * else and called schedule in __vcpu_run.  Hopefully that
1730          * VCPU is holding the lock that we need and will release it.
1731          * We approximate round-robin by starting at the last boosted VCPU.
1732          */
1733         for (pass = 0; pass < 2 && !yielded; pass++) {
1734                 kvm_for_each_vcpu(i, vcpu, kvm) {
1735                         if (!pass && i <= last_boosted_vcpu) {
1736                                 i = last_boosted_vcpu;
1737                                 continue;
1738                         } else if (pass && i > last_boosted_vcpu)
1739                                 break;
1740                         if (vcpu == me)
1741                                 continue;
1742                         if (waitqueue_active(&vcpu->wq))
1743                                 continue;
1744                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1745                                 continue;
1746                         if (kvm_vcpu_yield_to(vcpu)) {
1747                                 kvm->last_boosted_vcpu = i;
1748                                 yielded = 1;
1749                                 break;
1750                         }
1751                 }
1752         }
1753         kvm_vcpu_set_in_spin_loop(me, false);
1754
1755         /* Ensure vcpu is not eligible during next spinloop */
1756         kvm_vcpu_set_dy_eligible(me, false);
1757 }
1758 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1759
1760 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1761 {
1762         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1763         struct page *page;
1764
1765         if (vmf->pgoff == 0)
1766                 page = virt_to_page(vcpu->run);
1767 #ifdef CONFIG_X86
1768         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1769                 page = virt_to_page(vcpu->arch.pio_data);
1770 #endif
1771 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1772         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1773                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1774 #endif
1775         else
1776                 return kvm_arch_vcpu_fault(vcpu, vmf);
1777         get_page(page);
1778         vmf->page = page;
1779         return 0;
1780 }
1781
1782 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1783         .fault = kvm_vcpu_fault,
1784 };
1785
1786 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1787 {
1788         vma->vm_ops = &kvm_vcpu_vm_ops;
1789         return 0;
1790 }
1791
1792 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1793 {
1794         struct kvm_vcpu *vcpu = filp->private_data;
1795
1796         kvm_put_kvm(vcpu->kvm);
1797         return 0;
1798 }
1799
1800 static struct file_operations kvm_vcpu_fops = {
1801         .release        = kvm_vcpu_release,
1802         .unlocked_ioctl = kvm_vcpu_ioctl,
1803 #ifdef CONFIG_COMPAT
1804         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1805 #endif
1806         .mmap           = kvm_vcpu_mmap,
1807         .llseek         = noop_llseek,
1808 };
1809
1810 /*
1811  * Allocates an inode for the vcpu.
1812  */
1813 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1814 {
1815         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1816 }
1817
1818 /*
1819  * Creates some virtual cpus.  Good luck creating more than one.
1820  */
1821 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1822 {
1823         int r;
1824         struct kvm_vcpu *vcpu, *v;
1825
1826         vcpu = kvm_arch_vcpu_create(kvm, id);
1827         if (IS_ERR(vcpu))
1828                 return PTR_ERR(vcpu);
1829
1830         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1831
1832         r = kvm_arch_vcpu_setup(vcpu);
1833         if (r)
1834                 goto vcpu_destroy;
1835
1836         mutex_lock(&kvm->lock);
1837         if (!kvm_vcpu_compatible(vcpu)) {
1838                 r = -EINVAL;
1839                 goto unlock_vcpu_destroy;
1840         }
1841         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1842                 r = -EINVAL;
1843                 goto unlock_vcpu_destroy;
1844         }
1845
1846         kvm_for_each_vcpu(r, v, kvm)
1847                 if (v->vcpu_id == id) {
1848                         r = -EEXIST;
1849                         goto unlock_vcpu_destroy;
1850                 }
1851
1852         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1853
1854         /* Now it's all set up, let userspace reach it */
1855         kvm_get_kvm(kvm);
1856         r = create_vcpu_fd(vcpu);
1857         if (r < 0) {
1858                 kvm_put_kvm(kvm);
1859                 goto unlock_vcpu_destroy;
1860         }
1861
1862         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1863         smp_wmb();
1864         atomic_inc(&kvm->online_vcpus);
1865
1866         mutex_unlock(&kvm->lock);
1867         kvm_arch_vcpu_postcreate(vcpu);
1868         return r;
1869
1870 unlock_vcpu_destroy:
1871         mutex_unlock(&kvm->lock);
1872 vcpu_destroy:
1873         kvm_arch_vcpu_destroy(vcpu);
1874         return r;
1875 }
1876
1877 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1878 {
1879         if (sigset) {
1880                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1881                 vcpu->sigset_active = 1;
1882                 vcpu->sigset = *sigset;
1883         } else
1884                 vcpu->sigset_active = 0;
1885         return 0;
1886 }
1887
1888 static long kvm_vcpu_ioctl(struct file *filp,
1889                            unsigned int ioctl, unsigned long arg)
1890 {
1891         struct kvm_vcpu *vcpu = filp->private_data;
1892         void __user *argp = (void __user *)arg;
1893         int r;
1894         struct kvm_fpu *fpu = NULL;
1895         struct kvm_sregs *kvm_sregs = NULL;
1896
1897         if (vcpu->kvm->mm != current->mm)
1898                 return -EIO;
1899
1900 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1901         /*
1902          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1903          * so vcpu_load() would break it.
1904          */
1905         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1906                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1907 #endif
1908
1909
1910         r = vcpu_load(vcpu);
1911         if (r)
1912                 return r;
1913         switch (ioctl) {
1914         case KVM_RUN:
1915                 r = -EINVAL;
1916                 if (arg)
1917                         goto out;
1918                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1919                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1920                 break;
1921         case KVM_GET_REGS: {
1922                 struct kvm_regs *kvm_regs;
1923
1924                 r = -ENOMEM;
1925                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1926                 if (!kvm_regs)
1927                         goto out;
1928                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1929                 if (r)
1930                         goto out_free1;
1931                 r = -EFAULT;
1932                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1933                         goto out_free1;
1934                 r = 0;
1935 out_free1:
1936                 kfree(kvm_regs);
1937                 break;
1938         }
1939         case KVM_SET_REGS: {
1940                 struct kvm_regs *kvm_regs;
1941
1942                 r = -ENOMEM;
1943                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1944                 if (IS_ERR(kvm_regs)) {
1945                         r = PTR_ERR(kvm_regs);
1946                         goto out;
1947                 }
1948                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1949                 kfree(kvm_regs);
1950                 break;
1951         }
1952         case KVM_GET_SREGS: {
1953                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1954                 r = -ENOMEM;
1955                 if (!kvm_sregs)
1956                         goto out;
1957                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1958                 if (r)
1959                         goto out;
1960                 r = -EFAULT;
1961                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1962                         goto out;
1963                 r = 0;
1964                 break;
1965         }
1966         case KVM_SET_SREGS: {
1967                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1968                 if (IS_ERR(kvm_sregs)) {
1969                         r = PTR_ERR(kvm_sregs);
1970                         kvm_sregs = NULL;
1971                         goto out;
1972                 }
1973                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1974                 break;
1975         }
1976         case KVM_GET_MP_STATE: {
1977                 struct kvm_mp_state mp_state;
1978
1979                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1980                 if (r)
1981                         goto out;
1982                 r = -EFAULT;
1983                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1984                         goto out;
1985                 r = 0;
1986                 break;
1987         }
1988         case KVM_SET_MP_STATE: {
1989                 struct kvm_mp_state mp_state;
1990
1991                 r = -EFAULT;
1992                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1993                         goto out;
1994                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1995                 break;
1996         }
1997         case KVM_TRANSLATE: {
1998                 struct kvm_translation tr;
1999
2000                 r = -EFAULT;
2001                 if (copy_from_user(&tr, argp, sizeof tr))
2002                         goto out;
2003                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2004                 if (r)
2005                         goto out;
2006                 r = -EFAULT;
2007                 if (copy_to_user(argp, &tr, sizeof tr))
2008                         goto out;
2009                 r = 0;
2010                 break;
2011         }
2012         case KVM_SET_GUEST_DEBUG: {
2013                 struct kvm_guest_debug dbg;
2014
2015                 r = -EFAULT;
2016                 if (copy_from_user(&dbg, argp, sizeof dbg))
2017                         goto out;
2018                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2019                 break;
2020         }
2021         case KVM_SET_SIGNAL_MASK: {
2022                 struct kvm_signal_mask __user *sigmask_arg = argp;
2023                 struct kvm_signal_mask kvm_sigmask;
2024                 sigset_t sigset, *p;
2025
2026                 p = NULL;
2027                 if (argp) {
2028                         r = -EFAULT;
2029                         if (copy_from_user(&kvm_sigmask, argp,
2030                                            sizeof kvm_sigmask))
2031                                 goto out;
2032                         r = -EINVAL;
2033                         if (kvm_sigmask.len != sizeof sigset)
2034                                 goto out;
2035                         r = -EFAULT;
2036                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2037                                            sizeof sigset))
2038                                 goto out;
2039                         p = &sigset;
2040                 }
2041                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2042                 break;
2043         }
2044         case KVM_GET_FPU: {
2045                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2046                 r = -ENOMEM;
2047                 if (!fpu)
2048                         goto out;
2049                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2050                 if (r)
2051                         goto out;
2052                 r = -EFAULT;
2053                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2054                         goto out;
2055                 r = 0;
2056                 break;
2057         }
2058         case KVM_SET_FPU: {
2059                 fpu = memdup_user(argp, sizeof(*fpu));
2060                 if (IS_ERR(fpu)) {
2061                         r = PTR_ERR(fpu);
2062                         fpu = NULL;
2063                         goto out;
2064                 }
2065                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2066                 break;
2067         }
2068         default:
2069                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2070         }
2071 out:
2072         vcpu_put(vcpu);
2073         kfree(fpu);
2074         kfree(kvm_sregs);
2075         return r;
2076 }
2077
2078 #ifdef CONFIG_COMPAT
2079 static long kvm_vcpu_compat_ioctl(struct file *filp,
2080                                   unsigned int ioctl, unsigned long arg)
2081 {
2082         struct kvm_vcpu *vcpu = filp->private_data;
2083         void __user *argp = compat_ptr(arg);
2084         int r;
2085
2086         if (vcpu->kvm->mm != current->mm)
2087                 return -EIO;
2088
2089         switch (ioctl) {
2090         case KVM_SET_SIGNAL_MASK: {
2091                 struct kvm_signal_mask __user *sigmask_arg = argp;
2092                 struct kvm_signal_mask kvm_sigmask;
2093                 compat_sigset_t csigset;
2094                 sigset_t sigset;
2095
2096                 if (argp) {
2097                         r = -EFAULT;
2098                         if (copy_from_user(&kvm_sigmask, argp,
2099                                            sizeof kvm_sigmask))
2100                                 goto out;
2101                         r = -EINVAL;
2102                         if (kvm_sigmask.len != sizeof csigset)
2103                                 goto out;
2104                         r = -EFAULT;
2105                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2106                                            sizeof csigset))
2107                                 goto out;
2108                         sigset_from_compat(&sigset, &csigset);
2109                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2110                 } else
2111                         r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2112                 break;
2113         }
2114         default:
2115                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2116         }
2117
2118 out:
2119         return r;
2120 }
2121 #endif
2122
2123 static long kvm_vm_ioctl(struct file *filp,
2124                            unsigned int ioctl, unsigned long arg)
2125 {
2126         struct kvm *kvm = filp->private_data;
2127         void __user *argp = (void __user *)arg;
2128         int r;
2129
2130         if (kvm->mm != current->mm)
2131                 return -EIO;
2132         switch (ioctl) {
2133         case KVM_CREATE_VCPU:
2134                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2135                 break;
2136         case KVM_SET_USER_MEMORY_REGION: {
2137                 struct kvm_userspace_memory_region kvm_userspace_mem;
2138
2139                 r = -EFAULT;
2140                 if (copy_from_user(&kvm_userspace_mem, argp,
2141                                                 sizeof kvm_userspace_mem))
2142                         goto out;
2143
2144                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2145                 break;
2146         }
2147         case KVM_GET_DIRTY_LOG: {
2148                 struct kvm_dirty_log log;
2149
2150                 r = -EFAULT;
2151                 if (copy_from_user(&log, argp, sizeof log))
2152                         goto out;
2153                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2154                 break;
2155         }
2156 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2157         case KVM_REGISTER_COALESCED_MMIO: {
2158                 struct kvm_coalesced_mmio_zone zone;
2159                 r = -EFAULT;
2160                 if (copy_from_user(&zone, argp, sizeof zone))
2161                         goto out;
2162                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2163                 break;
2164         }
2165         case KVM_UNREGISTER_COALESCED_MMIO: {
2166                 struct kvm_coalesced_mmio_zone zone;
2167                 r = -EFAULT;
2168                 if (copy_from_user(&zone, argp, sizeof zone))
2169                         goto out;
2170                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2171                 break;
2172         }
2173 #endif
2174         case KVM_IRQFD: {
2175                 struct kvm_irqfd data;
2176
2177                 r = -EFAULT;
2178                 if (copy_from_user(&data, argp, sizeof data))
2179                         goto out;
2180                 r = kvm_irqfd(kvm, &data);
2181                 break;
2182         }
2183         case KVM_IOEVENTFD: {
2184                 struct kvm_ioeventfd data;
2185
2186                 r = -EFAULT;
2187                 if (copy_from_user(&data, argp, sizeof data))
2188                         goto out;
2189                 r = kvm_ioeventfd(kvm, &data);
2190                 break;
2191         }
2192 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2193         case KVM_SET_BOOT_CPU_ID:
2194                 r = 0;
2195                 mutex_lock(&kvm->lock);
2196                 if (atomic_read(&kvm->online_vcpus) != 0)
2197                         r = -EBUSY;
2198                 else
2199                         kvm->bsp_vcpu_id = arg;
2200                 mutex_unlock(&kvm->lock);
2201                 break;
2202 #endif
2203 #ifdef CONFIG_HAVE_KVM_MSI
2204         case KVM_SIGNAL_MSI: {
2205                 struct kvm_msi msi;
2206
2207                 r = -EFAULT;
2208                 if (copy_from_user(&msi, argp, sizeof msi))
2209                         goto out;
2210                 r = kvm_send_userspace_msi(kvm, &msi);
2211                 break;
2212         }
2213 #endif
2214 #ifdef __KVM_HAVE_IRQ_LINE
2215         case KVM_IRQ_LINE_STATUS:
2216         case KVM_IRQ_LINE: {
2217                 struct kvm_irq_level irq_event;
2218
2219                 r = -EFAULT;
2220                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2221                         goto out;
2222
2223                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2224                 if (r)
2225                         goto out;
2226
2227                 r = -EFAULT;
2228                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2229                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2230                                 goto out;
2231                 }
2232
2233                 r = 0;
2234                 break;
2235         }
2236 #endif
2237         default:
2238                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2239                 if (r == -ENOTTY)
2240                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2241         }
2242 out:
2243         return r;
2244 }
2245
2246 #ifdef CONFIG_COMPAT
2247 struct compat_kvm_dirty_log {
2248         __u32 slot;
2249         __u32 padding1;
2250         union {
2251                 compat_uptr_t dirty_bitmap; /* one bit per page */
2252                 __u64 padding2;
2253         };
2254 };
2255
2256 static long kvm_vm_compat_ioctl(struct file *filp,
2257                            unsigned int ioctl, unsigned long arg)
2258 {
2259         struct kvm *kvm = filp->private_data;
2260         int r;
2261
2262         if (kvm->mm != current->mm)
2263                 return -EIO;
2264         switch (ioctl) {
2265         case KVM_GET_DIRTY_LOG: {
2266                 struct compat_kvm_dirty_log compat_log;
2267                 struct kvm_dirty_log log;
2268
2269                 r = -EFAULT;
2270                 if (copy_from_user(&compat_log, (void __user *)arg,
2271                                    sizeof(compat_log)))
2272                         goto out;
2273                 log.slot         = compat_log.slot;
2274                 log.padding1     = compat_log.padding1;
2275                 log.padding2     = compat_log.padding2;
2276                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2277
2278                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2279                 break;
2280         }
2281         default:
2282                 r = kvm_vm_ioctl(filp, ioctl, arg);
2283         }
2284
2285 out:
2286         return r;
2287 }
2288 #endif
2289
2290 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2291 {
2292         struct page *page[1];
2293         unsigned long addr;
2294         int npages;
2295         gfn_t gfn = vmf->pgoff;
2296         struct kvm *kvm = vma->vm_file->private_data;
2297
2298         addr = gfn_to_hva(kvm, gfn);
2299         if (kvm_is_error_hva(addr))
2300                 return VM_FAULT_SIGBUS;
2301
2302         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2303                                 NULL);
2304         if (unlikely(npages != 1))
2305                 return VM_FAULT_SIGBUS;
2306
2307         vmf->page = page[0];
2308         return 0;
2309 }
2310
2311 static const struct vm_operations_struct kvm_vm_vm_ops = {
2312         .fault = kvm_vm_fault,
2313 };
2314
2315 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2316 {
2317         vma->vm_ops = &kvm_vm_vm_ops;
2318         return 0;
2319 }
2320
2321 static struct file_operations kvm_vm_fops = {
2322         .release        = kvm_vm_release,
2323         .unlocked_ioctl = kvm_vm_ioctl,
2324 #ifdef CONFIG_COMPAT
2325         .compat_ioctl   = kvm_vm_compat_ioctl,
2326 #endif
2327         .mmap           = kvm_vm_mmap,
2328         .llseek         = noop_llseek,
2329 };
2330
2331 static int kvm_dev_ioctl_create_vm(unsigned long type)
2332 {
2333         int r;
2334         struct kvm *kvm;
2335
2336         kvm = kvm_create_vm(type);
2337         if (IS_ERR(kvm))
2338                 return PTR_ERR(kvm);
2339 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2340         r = kvm_coalesced_mmio_init(kvm);
2341         if (r < 0) {
2342                 kvm_put_kvm(kvm);
2343                 return r;
2344         }
2345 #endif
2346         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2347         if (r < 0)
2348                 kvm_put_kvm(kvm);
2349
2350         return r;
2351 }
2352
2353 static long kvm_dev_ioctl_check_extension_generic(long arg)
2354 {
2355         switch (arg) {
2356         case KVM_CAP_USER_MEMORY:
2357         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2358         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2359 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2360         case KVM_CAP_SET_BOOT_CPU_ID:
2361 #endif
2362         case KVM_CAP_INTERNAL_ERROR_DATA:
2363 #ifdef CONFIG_HAVE_KVM_MSI
2364         case KVM_CAP_SIGNAL_MSI:
2365 #endif
2366                 return 1;
2367 #ifdef KVM_CAP_IRQ_ROUTING
2368         case KVM_CAP_IRQ_ROUTING:
2369                 return KVM_MAX_IRQ_ROUTES;
2370 #endif
2371         default:
2372                 break;
2373         }
2374         return kvm_dev_ioctl_check_extension(arg);
2375 }
2376
2377 static long kvm_dev_ioctl(struct file *filp,
2378                           unsigned int ioctl, unsigned long arg)
2379 {
2380         long r = -EINVAL;
2381
2382         switch (ioctl) {
2383         case KVM_GET_API_VERSION:
2384                 r = -EINVAL;
2385                 if (arg)
2386                         goto out;
2387                 r = KVM_API_VERSION;
2388                 break;
2389         case KVM_CREATE_VM:
2390                 r = kvm_dev_ioctl_create_vm(arg);
2391                 break;
2392         case KVM_CHECK_EXTENSION:
2393                 r = kvm_dev_ioctl_check_extension_generic(arg);
2394                 break;
2395         case KVM_GET_VCPU_MMAP_SIZE:
2396                 r = -EINVAL;
2397                 if (arg)
2398                         goto out;
2399                 r = PAGE_SIZE;     /* struct kvm_run */
2400 #ifdef CONFIG_X86
2401                 r += PAGE_SIZE;    /* pio data page */
2402 #endif
2403 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2404                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2405 #endif
2406                 break;
2407         case KVM_TRACE_ENABLE:
2408         case KVM_TRACE_PAUSE:
2409         case KVM_TRACE_DISABLE:
2410                 r = -EOPNOTSUPP;
2411                 break;
2412         default:
2413                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2414         }
2415 out:
2416         return r;
2417 }
2418
2419 static struct file_operations kvm_chardev_ops = {
2420         .unlocked_ioctl = kvm_dev_ioctl,
2421         .compat_ioctl   = kvm_dev_ioctl,
2422         .llseek         = noop_llseek,
2423 };
2424
2425 static struct miscdevice kvm_dev = {
2426         KVM_MINOR,
2427         "kvm",
2428         &kvm_chardev_ops,
2429 };
2430
2431 static void hardware_enable_nolock(void *junk)
2432 {
2433         int cpu = raw_smp_processor_id();
2434         int r;
2435
2436         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2437                 return;
2438
2439         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2440
2441         r = kvm_arch_hardware_enable(NULL);
2442
2443         if (r) {
2444                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2445                 atomic_inc(&hardware_enable_failed);
2446                 printk(KERN_INFO "kvm: enabling virtualization on "
2447                                  "CPU%d failed\n", cpu);
2448         }
2449 }
2450
2451 static void hardware_enable(void *junk)
2452 {
2453         raw_spin_lock(&kvm_lock);
2454         hardware_enable_nolock(junk);
2455         raw_spin_unlock(&kvm_lock);
2456 }
2457
2458 static void hardware_disable_nolock(void *junk)
2459 {
2460         int cpu = raw_smp_processor_id();
2461
2462         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2463                 return;
2464         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2465         kvm_arch_hardware_disable(NULL);
2466 }
2467
2468 static void hardware_disable(void *junk)
2469 {
2470         raw_spin_lock(&kvm_lock);
2471         hardware_disable_nolock(junk);
2472         raw_spin_unlock(&kvm_lock);
2473 }
2474
2475 static void hardware_disable_all_nolock(void)
2476 {
2477         BUG_ON(!kvm_usage_count);
2478
2479         kvm_usage_count--;
2480         if (!kvm_usage_count)
2481                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2482 }
2483
2484 static void hardware_disable_all(void)
2485 {
2486         raw_spin_lock(&kvm_lock);
2487         hardware_disable_all_nolock();
2488         raw_spin_unlock(&kvm_lock);
2489 }
2490
2491 static int hardware_enable_all(void)
2492 {
2493         int r = 0;
2494
2495         raw_spin_lock(&kvm_lock);
2496
2497         kvm_usage_count++;
2498         if (kvm_usage_count == 1) {
2499                 atomic_set(&hardware_enable_failed, 0);
2500                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2501
2502                 if (atomic_read(&hardware_enable_failed)) {
2503                         hardware_disable_all_nolock();
2504                         r = -EBUSY;
2505                 }
2506         }
2507
2508         raw_spin_unlock(&kvm_lock);
2509
2510         return r;
2511 }
2512
2513 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2514                            void *v)
2515 {
2516         int cpu = (long)v;
2517
2518         if (!kvm_usage_count)
2519                 return NOTIFY_OK;
2520
2521         val &= ~CPU_TASKS_FROZEN;
2522         switch (val) {
2523         case CPU_DYING:
2524                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2525                        cpu);
2526                 hardware_disable(NULL);
2527                 break;
2528         case CPU_STARTING:
2529                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2530                        cpu);
2531                 hardware_enable(NULL);
2532                 break;
2533         }
2534         return NOTIFY_OK;
2535 }
2536
2537
2538 asmlinkage void kvm_spurious_fault(void)
2539 {
2540         /* Fault while not rebooting.  We want the trace. */
2541         BUG();
2542 }
2543 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2544
2545 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2546                       void *v)
2547 {
2548         /*
2549          * Some (well, at least mine) BIOSes hang on reboot if
2550          * in vmx root mode.
2551          *
2552          * And Intel TXT required VMX off for all cpu when system shutdown.
2553          */
2554         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2555         kvm_rebooting = true;
2556         on_each_cpu(hardware_disable_nolock, NULL, 1);
2557         return NOTIFY_OK;
2558 }
2559
2560 static struct notifier_block kvm_reboot_notifier = {
2561         .notifier_call = kvm_reboot,
2562         .priority = 0,
2563 };
2564
2565 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2566 {
2567         int i;
2568
2569         for (i = 0; i < bus->dev_count; i++) {
2570                 struct kvm_io_device *pos = bus->range[i].dev;
2571
2572                 kvm_iodevice_destructor(pos);
2573         }
2574         kfree(bus);
2575 }
2576
2577 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2578 {
2579         const struct kvm_io_range *r1 = p1;
2580         const struct kvm_io_range *r2 = p2;
2581
2582         if (r1->addr < r2->addr)
2583                 return -1;
2584         if (r1->addr + r1->len > r2->addr + r2->len)
2585                 return 1;
2586         return 0;
2587 }
2588
2589 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2590                           gpa_t addr, int len)
2591 {
2592         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2593                 .addr = addr,
2594                 .len = len,
2595                 .dev = dev,
2596         };
2597
2598         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2599                 kvm_io_bus_sort_cmp, NULL);
2600
2601         return 0;
2602 }
2603
2604 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2605                              gpa_t addr, int len)
2606 {
2607         struct kvm_io_range *range, key;
2608         int off;
2609
2610         key = (struct kvm_io_range) {
2611                 .addr = addr,
2612                 .len = len,
2613         };
2614
2615         range = bsearch(&key, bus->range, bus->dev_count,
2616                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2617         if (range == NULL)
2618                 return -ENOENT;
2619
2620         off = range - bus->range;
2621
2622         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2623                 off--;
2624
2625         return off;
2626 }
2627
2628 /* kvm_io_bus_write - called under kvm->slots_lock */
2629 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2630                      int len, const void *val)
2631 {
2632         int idx;
2633         struct kvm_io_bus *bus;
2634         struct kvm_io_range range;
2635
2636         range = (struct kvm_io_range) {
2637                 .addr = addr,
2638                 .len = len,
2639         };
2640
2641         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2642         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2643         if (idx < 0)
2644                 return -EOPNOTSUPP;
2645
2646         while (idx < bus->dev_count &&
2647                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2648                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2649                         return 0;
2650                 idx++;
2651         }
2652
2653         return -EOPNOTSUPP;
2654 }
2655
2656 /* kvm_io_bus_read - called under kvm->slots_lock */
2657 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2658                     int len, void *val)
2659 {
2660         int idx;
2661         struct kvm_io_bus *bus;
2662         struct kvm_io_range range;
2663
2664         range = (struct kvm_io_range) {
2665                 .addr = addr,
2666                 .len = len,
2667         };
2668
2669         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2670         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2671         if (idx < 0)
2672                 return -EOPNOTSUPP;
2673
2674         while (idx < bus->dev_count &&
2675                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2676                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2677                         return 0;
2678                 idx++;
2679         }
2680
2681         return -EOPNOTSUPP;
2682 }
2683
2684 /* Caller must hold slots_lock. */
2685 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2686                             int len, struct kvm_io_device *dev)
2687 {
2688         struct kvm_io_bus *new_bus, *bus;
2689
2690         bus = kvm->buses[bus_idx];
2691         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2692                 return -ENOSPC;
2693
2694         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2695                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2696         if (!new_bus)
2697                 return -ENOMEM;
2698         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2699                sizeof(struct kvm_io_range)));
2700         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2701         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2702         synchronize_srcu_expedited(&kvm->srcu);
2703         kfree(bus);
2704
2705         return 0;
2706 }
2707
2708 /* Caller must hold slots_lock. */
2709 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2710                               struct kvm_io_device *dev)
2711 {
2712         int i, r;
2713         struct kvm_io_bus *new_bus, *bus;
2714
2715         bus = kvm->buses[bus_idx];
2716         r = -ENOENT;
2717         for (i = 0; i < bus->dev_count; i++)
2718                 if (bus->range[i].dev == dev) {
2719                         r = 0;
2720                         break;
2721                 }
2722
2723         if (r)
2724                 return r;
2725
2726         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2727                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2728         if (!new_bus)
2729                 return -ENOMEM;
2730
2731         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2732         new_bus->dev_count--;
2733         memcpy(new_bus->range + i, bus->range + i + 1,
2734                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2735
2736         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2737         synchronize_srcu_expedited(&kvm->srcu);
2738         kfree(bus);
2739         return r;
2740 }
2741
2742 static struct notifier_block kvm_cpu_notifier = {
2743         .notifier_call = kvm_cpu_hotplug,
2744 };
2745
2746 static int vm_stat_get(void *_offset, u64 *val)
2747 {
2748         unsigned offset = (long)_offset;
2749         struct kvm *kvm;
2750
2751         *val = 0;
2752         raw_spin_lock(&kvm_lock);
2753         list_for_each_entry(kvm, &vm_list, vm_list)
2754                 *val += *(u32 *)((void *)kvm + offset);
2755         raw_spin_unlock(&kvm_lock);
2756         return 0;
2757 }
2758
2759 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2760
2761 static int vcpu_stat_get(void *_offset, u64 *val)
2762 {
2763         unsigned offset = (long)_offset;
2764         struct kvm *kvm;
2765         struct kvm_vcpu *vcpu;
2766         int i;
2767
2768         *val = 0;
2769         raw_spin_lock(&kvm_lock);
2770         list_for_each_entry(kvm, &vm_list, vm_list)
2771                 kvm_for_each_vcpu(i, vcpu, kvm)
2772                         *val += *(u32 *)((void *)vcpu + offset);
2773
2774         raw_spin_unlock(&kvm_lock);
2775         return 0;
2776 }
2777
2778 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2779
2780 static const struct file_operations *stat_fops[] = {
2781         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2782         [KVM_STAT_VM]   = &vm_stat_fops,
2783 };
2784
2785 static int kvm_init_debug(void)
2786 {
2787         int r = -EFAULT;
2788         struct kvm_stats_debugfs_item *p;
2789
2790         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2791         if (kvm_debugfs_dir == NULL)
2792                 goto out;
2793
2794         for (p = debugfs_entries; p->name; ++p) {
2795                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2796                                                 (void *)(long)p->offset,
2797                                                 stat_fops[p->kind]);
2798                 if (p->dentry == NULL)
2799                         goto out_dir;
2800         }
2801
2802         return 0;
2803
2804 out_dir:
2805         debugfs_remove_recursive(kvm_debugfs_dir);
2806 out:
2807         return r;
2808 }
2809
2810 static void kvm_exit_debug(void)
2811 {
2812         struct kvm_stats_debugfs_item *p;
2813
2814         for (p = debugfs_entries; p->name; ++p)
2815                 debugfs_remove(p->dentry);
2816         debugfs_remove(kvm_debugfs_dir);
2817 }
2818
2819 static int kvm_suspend(void)
2820 {
2821         if (kvm_usage_count)
2822                 hardware_disable_nolock(NULL);
2823         return 0;
2824 }
2825
2826 static void kvm_resume(void)
2827 {
2828         if (kvm_usage_count) {
2829                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2830                 hardware_enable_nolock(NULL);
2831         }
2832 }
2833
2834 static struct syscore_ops kvm_syscore_ops = {
2835         .suspend = kvm_suspend,
2836         .resume = kvm_resume,
2837 };
2838
2839 static inline
2840 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2841 {
2842         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2843 }
2844
2845 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2846 {
2847         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2848
2849         kvm_arch_vcpu_load(vcpu, cpu);
2850 }
2851
2852 static void kvm_sched_out(struct preempt_notifier *pn,
2853                           struct task_struct *next)
2854 {
2855         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2856
2857         kvm_arch_vcpu_put(vcpu);
2858 }
2859
2860 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2861                   struct module *module)
2862 {
2863         int r;
2864         int cpu;
2865
2866         r = kvm_arch_init(opaque);
2867         if (r)
2868                 goto out_fail;
2869
2870         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2871                 r = -ENOMEM;
2872                 goto out_free_0;
2873         }
2874
2875         r = kvm_arch_hardware_setup();
2876         if (r < 0)
2877                 goto out_free_0a;
2878
2879         for_each_online_cpu(cpu) {
2880                 smp_call_function_single(cpu,
2881                                 kvm_arch_check_processor_compat,
2882                                 &r, 1);
2883                 if (r < 0)
2884                         goto out_free_1;
2885         }
2886
2887         r = register_cpu_notifier(&kvm_cpu_notifier);
2888         if (r)
2889                 goto out_free_2;
2890         register_reboot_notifier(&kvm_reboot_notifier);
2891
2892         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2893         if (!vcpu_align)
2894                 vcpu_align = __alignof__(struct kvm_vcpu);
2895         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2896                                            0, NULL);
2897         if (!kvm_vcpu_cache) {
2898                 r = -ENOMEM;
2899                 goto out_free_3;
2900         }
2901
2902         r = kvm_async_pf_init();
2903         if (r)
2904                 goto out_free;
2905
2906         kvm_chardev_ops.owner = module;
2907         kvm_vm_fops.owner = module;
2908         kvm_vcpu_fops.owner = module;
2909
2910         r = misc_register(&kvm_dev);
2911         if (r) {
2912                 printk(KERN_ERR "kvm: misc device register failed\n");
2913                 goto out_unreg;
2914         }
2915
2916         register_syscore_ops(&kvm_syscore_ops);
2917
2918         kvm_preempt_ops.sched_in = kvm_sched_in;
2919         kvm_preempt_ops.sched_out = kvm_sched_out;
2920
2921         r = kvm_init_debug();
2922         if (r) {
2923                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2924                 goto out_undebugfs;
2925         }
2926
2927         return 0;
2928
2929 out_undebugfs:
2930         unregister_syscore_ops(&kvm_syscore_ops);
2931 out_unreg:
2932         kvm_async_pf_deinit();
2933 out_free:
2934         kmem_cache_destroy(kvm_vcpu_cache);
2935 out_free_3:
2936         unregister_reboot_notifier(&kvm_reboot_notifier);
2937         unregister_cpu_notifier(&kvm_cpu_notifier);
2938 out_free_2:
2939 out_free_1:
2940         kvm_arch_hardware_unsetup();
2941 out_free_0a:
2942         free_cpumask_var(cpus_hardware_enabled);
2943 out_free_0:
2944         kvm_arch_exit();
2945 out_fail:
2946         return r;
2947 }
2948 EXPORT_SYMBOL_GPL(kvm_init);
2949
2950 void kvm_exit(void)
2951 {
2952         kvm_exit_debug();
2953         misc_deregister(&kvm_dev);
2954         kmem_cache_destroy(kvm_vcpu_cache);
2955         kvm_async_pf_deinit();
2956         unregister_syscore_ops(&kvm_syscore_ops);
2957         unregister_reboot_notifier(&kvm_reboot_notifier);
2958         unregister_cpu_notifier(&kvm_cpu_notifier);
2959         on_each_cpu(hardware_disable_nolock, NULL, 1);
2960         kvm_arch_hardware_unsetup();
2961         kvm_arch_exit();
2962         free_cpumask_var(cpus_hardware_enabled);
2963 }
2964 EXPORT_SYMBOL_GPL(kvm_exit);