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