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/sysdev.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
  51 #include <asm/processor.h>
  52 #include <asm/io.h>
  53 #include <asm/uaccess.h>
  54 #include <asm/pgtable.h>
  55 #include <asm-generic/bitops/le.h>
  56
  57 #include "coalesced_mmio.h"
  58 #include "async_pf.h"
  59
  60 #define CREATE_TRACE_POINTS
  61 #include <trace/events/kvm.h>
  62
  63 MODULE_AUTHOR("Qumranet");
  64 MODULE_LICENSE("GPL");
  65
  66 /*
  67  * Ordering of locks:
  68  *
  69  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  70  */
  71
  72 DEFINE_SPINLOCK(kvm_lock);
  73 LIST_HEAD(vm_list);
  74
  75 static cpumask_var_t cpus_hardware_enabled;
  76 static int kvm_usage_count = 0;
  77 static atomic_t hardware_enable_failed;
  78
  79 struct kmem_cache *kvm_vcpu_cache;
  80 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  81
  82 static __read_mostly struct preempt_ops kvm_preempt_ops;
  83
  84 struct dentry *kvm_debugfs_dir;
  85
  86 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  87                            unsigned long arg);
  88 static int hardware_enable_all(void);
  89 static void hardware_disable_all(void);
  90
  91 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  92
  93 static bool kvm_rebooting;
  94
  95 static bool largepages_enabled = true;
  96
  97 static struct page *hwpoison_page;
  98 static pfn_t hwpoison_pfn;
  99
 100 static struct page *fault_page;
 101 static pfn_t fault_pfn;
 102
 103 inline int kvm_is_mmio_pfn(pfn_t pfn)
 104 {
 105         if (pfn_valid(pfn)) {
 106                 struct page *page = compound_head(pfn_to_page(pfn));
 107                 return PageReserved(page);
 108         }
 109
 110         return true;
 111 }
 112
 113 /*
 114  * Switches to specified vcpu, until a matching vcpu_put()
 115  */
 116 void vcpu_load(struct kvm_vcpu *vcpu)
 117 {
 118         int cpu;
 119
 120         mutex_lock(&vcpu->mutex);
 121         cpu = get_cpu();
 122         preempt_notifier_register(&vcpu->preempt_notifier);
 123         kvm_arch_vcpu_load(vcpu, cpu);
 124         put_cpu();
 125 }
 126
 127 void vcpu_put(struct kvm_vcpu *vcpu)
 128 {
 129         preempt_disable();
 130         kvm_arch_vcpu_put(vcpu);
 131         preempt_notifier_unregister(&vcpu->preempt_notifier);
 132         preempt_enable();
 133         mutex_unlock(&vcpu->mutex);
 134 }
 135
 136 static void ack_flush(void *_completed)
 137 {
 138 }
 139
 140 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 141 {
 142         int i, cpu, me;
 143         cpumask_var_t cpus;
 144         bool called = true;
 145         struct kvm_vcpu *vcpu;
 146
 147         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 148
 149         raw_spin_lock(&kvm->requests_lock);
 150         me = smp_processor_id();
 151         kvm_for_each_vcpu(i, vcpu, kvm) {
 152                 if (kvm_make_check_request(req, vcpu))
 153                         continue;
 154                 cpu = vcpu->cpu;
 155                 if (cpus != NULL && cpu != -1 && cpu != me)
 156                         cpumask_set_cpu(cpu, cpus);
 157         }
 158         if (unlikely(cpus == NULL))
 159                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 160         else if (!cpumask_empty(cpus))
 161                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 162         else
 163                 called = false;
 164         raw_spin_unlock(&kvm->requests_lock);
 165         free_cpumask_var(cpus);
 166         return called;
 167 }
 168
 169 void kvm_flush_remote_tlbs(struct kvm *kvm)
 170 {
 171         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 172                 ++kvm->stat.remote_tlb_flush;
 173 }
 174
 175 void kvm_reload_remote_mmus(struct kvm *kvm)
 176 {
 177         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 178 }
 179
 180 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 181 {
 182         struct page *page;
 183         int r;
 184
 185         mutex_init(&vcpu->mutex);
 186         vcpu->cpu = -1;
 187         vcpu->kvm = kvm;
 188         vcpu->vcpu_id = id;
 189         init_waitqueue_head(&vcpu->wq);
 190         kvm_async_pf_vcpu_init(vcpu);
 191
 192         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 193         if (!page) {
 194                 r = -ENOMEM;
 195                 goto fail;
 196         }
 197         vcpu->run = page_address(page);
 198
 199         r = kvm_arch_vcpu_init(vcpu);
 200         if (r < 0)
 201                 goto fail_free_run;
 202         return 0;
 203
 204 fail_free_run:
 205         free_page((unsigned long)vcpu->run);
 206 fail:
 207         return r;
 208 }
 209 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 210
 211 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 212 {
 213         kvm_arch_vcpu_uninit(vcpu);
 214         free_page((unsigned long)vcpu->run);
 215 }
 216 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 217
 218 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 219 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 220 {
 221         return container_of(mn, struct kvm, mmu_notifier);
 222 }
 223
 224 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 225                                              struct mm_struct *mm,
 226                                              unsigned long address)
 227 {
 228         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 229         int need_tlb_flush, idx;
 230
 231         /*
 232          * When ->invalidate_page runs, the linux pte has been zapped
 233          * already but the page is still allocated until
 234          * ->invalidate_page returns. So if we increase the sequence
 235          * here the kvm page fault will notice if the spte can't be
 236          * established because the page is going to be freed. If
 237          * instead the kvm page fault establishes the spte before
 238          * ->invalidate_page runs, kvm_unmap_hva will release it
 239          * before returning.
 240          *
 241          * The sequence increase only need to be seen at spin_unlock
 242          * time, and not at spin_lock time.
 243          *
 244          * Increasing the sequence after the spin_unlock would be
 245          * unsafe because the kvm page fault could then establish the
 246          * pte after kvm_unmap_hva returned, without noticing the page
 247          * is going to be freed.
 248          */
 249         idx = srcu_read_lock(&kvm->srcu);
 250         spin_lock(&kvm->mmu_lock);
 251         kvm->mmu_notifier_seq++;
 252         need_tlb_flush = kvm_unmap_hva(kvm, address);
 253         spin_unlock(&kvm->mmu_lock);
 254         srcu_read_unlock(&kvm->srcu, idx);
 255
 256         /* we've to flush the tlb before the pages can be freed */
 257         if (need_tlb_flush)
 258                 kvm_flush_remote_tlbs(kvm);
 259
 260 }
 261
 262 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 263                                         struct mm_struct *mm,
 264                                         unsigned long address,
 265                                         pte_t pte)
 266 {
 267         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 268         int idx;
 269
 270         idx = srcu_read_lock(&kvm->srcu);
 271         spin_lock(&kvm->mmu_lock);
 272         kvm->mmu_notifier_seq++;
 273         kvm_set_spte_hva(kvm, address, pte);
 274         spin_unlock(&kvm->mmu_lock);
 275         srcu_read_unlock(&kvm->srcu, idx);
 276 }
 277
 278 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 279                                                     struct mm_struct *mm,
 280                                                     unsigned long start,
 281                                                     unsigned long end)
 282 {
 283         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 284         int need_tlb_flush = 0, idx;
 285
 286         idx = srcu_read_lock(&kvm->srcu);
 287         spin_lock(&kvm->mmu_lock);
 288         /*
 289          * The count increase must become visible at unlock time as no
 290          * spte can be established without taking the mmu_lock and
 291          * count is also read inside the mmu_lock critical section.
 292          */
 293         kvm->mmu_notifier_count++;
 294         for (; start < end; start += PAGE_SIZE)
 295                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 296         spin_unlock(&kvm->mmu_lock);
 297         srcu_read_unlock(&kvm->srcu, idx);
 298
 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
 304 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 305                                                   struct mm_struct *mm,
 306                                                   unsigned long start,
 307                                                   unsigned long end)
 308 {
 309         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 310
 311         spin_lock(&kvm->mmu_lock);
 312         /*
 313          * This sequence increase will notify the kvm page fault that
 314          * the page that is going to be mapped in the spte could have
 315          * been freed.
 316          */
 317         kvm->mmu_notifier_seq++;
 318         /*
 319          * The above sequence increase must be visible before the
 320          * below count decrease but both values are read by the kvm
 321          * page fault under mmu_lock spinlock so we don't need to add
 322          * a smb_wmb() here in between the two.
 323          */
 324         kvm->mmu_notifier_count--;
 325         spin_unlock(&kvm->mmu_lock);
 326
 327         BUG_ON(kvm->mmu_notifier_count < 0);
 328 }
 329
 330 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 331                                               struct mm_struct *mm,
 332                                               unsigned long address)
 333 {
 334         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 335         int young, idx;
 336
 337         idx = srcu_read_lock(&kvm->srcu);
 338         spin_lock(&kvm->mmu_lock);
 339         young = kvm_age_hva(kvm, address);
 340         spin_unlock(&kvm->mmu_lock);
 341         srcu_read_unlock(&kvm->srcu, idx);
 342
 343         if (young)
 344                 kvm_flush_remote_tlbs(kvm);
 345
 346         return young;
 347 }
 348
 349 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 350                                      struct mm_struct *mm)
 351 {
 352         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 353         int idx;
 354
 355         idx = srcu_read_lock(&kvm->srcu);
 356         kvm_arch_flush_shadow(kvm);
 357         srcu_read_unlock(&kvm->srcu, idx);
 358 }
 359
 360 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 361         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 362         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 363         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 364         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 365         .change_pte             = kvm_mmu_notifier_change_pte,
 366         .release                = kvm_mmu_notifier_release,
 367 };
 368
 369 static int kvm_init_mmu_notifier(struct kvm *kvm)
 370 {
 371         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 372         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 373 }
 374
 375 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 376
 377 static int kvm_init_mmu_notifier(struct kvm *kvm)
 378 {
 379         return 0;
 380 }
 381
 382 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 383
 384 static struct kvm *kvm_create_vm(void)
 385 {
 386         int r = 0, i;
 387         struct kvm *kvm = kvm_arch_create_vm();
 388
 389         if (IS_ERR(kvm))
 390                 goto out;
 391
 392         r = hardware_enable_all();
 393         if (r)
 394                 goto out_err_nodisable;
 395
 396 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 397         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 398         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 399 #endif
 400
 401         r = -ENOMEM;
 402         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 403         if (!kvm->memslots)
 404                 goto out_err;
 405         if (init_srcu_struct(&kvm->srcu))
 406                 goto out_err;
 407         for (i = 0; i < KVM_NR_BUSES; i++) {
 408                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 409                                         GFP_KERNEL);
 410                 if (!kvm->buses[i]) {
 411                         cleanup_srcu_struct(&kvm->srcu);
 412                         goto out_err;
 413                 }
 414         }
 415
 416         r = kvm_init_mmu_notifier(kvm);
 417         if (r) {
 418                 cleanup_srcu_struct(&kvm->srcu);
 419                 goto out_err;
 420         }
 421
 422         kvm->mm = current->mm;
 423         atomic_inc(&kvm->mm->mm_count);
 424         spin_lock_init(&kvm->mmu_lock);
 425         raw_spin_lock_init(&kvm->requests_lock);
 426         kvm_eventfd_init(kvm);
 427         mutex_init(&kvm->lock);
 428         mutex_init(&kvm->irq_lock);
 429         mutex_init(&kvm->slots_lock);
 430         atomic_set(&kvm->users_count, 1);
 431         spin_lock(&kvm_lock);
 432         list_add(&kvm->vm_list, &vm_list);
 433         spin_unlock(&kvm_lock);
 434 out:
 435         return kvm;
 436
 437 out_err:
 438         hardware_disable_all();
 439 out_err_nodisable:
 440         for (i = 0; i < KVM_NR_BUSES; i++)
 441                 kfree(kvm->buses[i]);
 442         kfree(kvm->memslots);
 443         kfree(kvm);
 444         return ERR_PTR(r);
 445 }
 446
 447 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 448 {
 449         if (!memslot->dirty_bitmap)
 450                 return;
 451
 452         vfree(memslot->dirty_bitmap_head);
 453         memslot->dirty_bitmap = NULL;
 454         memslot->dirty_bitmap_head = NULL;
 455 }
 456
 457 /*
 458  * Free any memory in @free but not in @dont.
 459  */
 460 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 461                                   struct kvm_memory_slot *dont)
 462 {
 463         int i;
 464
 465         if (!dont || free->rmap != dont->rmap)
 466                 vfree(free->rmap);
 467
 468         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 469                 kvm_destroy_dirty_bitmap(free);
 470
 471
 472         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 473                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 474                         vfree(free->lpage_info[i]);
 475                         free->lpage_info[i] = NULL;
 476                 }
 477         }
 478
 479         free->npages = 0;
 480         free->rmap = NULL;
 481 }
 482
 483 void kvm_free_physmem(struct kvm *kvm)
 484 {
 485         int i;
 486         struct kvm_memslots *slots = kvm->memslots;
 487
 488         for (i = 0; i < slots->nmemslots; ++i)
 489                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 490
 491         kfree(kvm->memslots);
 492 }
 493
 494 static void kvm_destroy_vm(struct kvm *kvm)
 495 {
 496         int i;
 497         struct mm_struct *mm = kvm->mm;
 498
 499         kvm_arch_sync_events(kvm);
 500         spin_lock(&kvm_lock);
 501         list_del(&kvm->vm_list);
 502         spin_unlock(&kvm_lock);
 503         kvm_free_irq_routing(kvm);
 504         for (i = 0; i < KVM_NR_BUSES; i++)
 505                 kvm_io_bus_destroy(kvm->buses[i]);
 506         kvm_coalesced_mmio_free(kvm);
 507 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 508         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 509 #else
 510         kvm_arch_flush_shadow(kvm);
 511 #endif
 512         kvm_arch_destroy_vm(kvm);
 513         hardware_disable_all();
 514         mmdrop(mm);
 515 }
 516
 517 void kvm_get_kvm(struct kvm *kvm)
 518 {
 519         atomic_inc(&kvm->users_count);
 520 }
 521 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 522
 523 void kvm_put_kvm(struct kvm *kvm)
 524 {
 525         if (atomic_dec_and_test(&kvm->users_count))
 526                 kvm_destroy_vm(kvm);
 527 }
 528 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 529
 530
 531 static int kvm_vm_release(struct inode *inode, struct file *filp)
 532 {
 533         struct kvm *kvm = filp->private_data;
 534
 535         kvm_irqfd_release(kvm);
 536
 537         kvm_put_kvm(kvm);
 538         return 0;
 539 }
 540
 541 /*
 542  * Allocation size is twice as large as the actual dirty bitmap size.
 543  * This makes it possible to do double buffering: see x86's
 544  * kvm_vm_ioctl_get_dirty_log().
 545  */
 546 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 547 {
 548         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 549
 550         memslot->dirty_bitmap = vmalloc(dirty_bytes);
 551         if (!memslot->dirty_bitmap)
 552                 return -ENOMEM;
 553
 554         memset(memslot->dirty_bitmap, 0, dirty_bytes);
 555         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
 556         return 0;
 557 }
 558
 559 /*
 560  * Allocate some memory and give it an address in the guest physical address
 561  * space.
 562  *
 563  * Discontiguous memory is allowed, mostly for framebuffers.
 564  *
 565  * Must be called holding mmap_sem for write.
 566  */
 567 int __kvm_set_memory_region(struct kvm *kvm,
 568                             struct kvm_userspace_memory_region *mem,
 569                             int user_alloc)
 570 {
 571         int r, flush_shadow = 0;
 572         gfn_t base_gfn;
 573         unsigned long npages;
 574         unsigned long i;
 575         struct kvm_memory_slot *memslot;
 576         struct kvm_memory_slot old, new;
 577         struct kvm_memslots *slots, *old_memslots;
 578
 579         r = -EINVAL;
 580         /* General sanity checks */
 581         if (mem->memory_size & (PAGE_SIZE - 1))
 582                 goto out;
 583         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 584                 goto out;
 585         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 586                 goto out;
 587         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 588                 goto out;
 589         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 590                 goto out;
 591
 592         memslot = &kvm->memslots->memslots[mem->slot];
 593         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 594         npages = mem->memory_size >> PAGE_SHIFT;
 595
 596         r = -EINVAL;
 597         if (npages > KVM_MEM_MAX_NR_PAGES)
 598                 goto out;
 599
 600         if (!npages)
 601                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 602
 603         new = old = *memslot;
 604
 605         new.id = mem->slot;
 606         new.base_gfn = base_gfn;
 607         new.npages = npages;
 608         new.flags = mem->flags;
 609
 610         /* Disallow changing a memory slot's size. */
 611         r = -EINVAL;
 612         if (npages && old.npages && npages != old.npages)
 613                 goto out_free;
 614
 615         /* Check for overlaps */
 616         r = -EEXIST;
 617         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 618                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 619
 620                 if (s == memslot || !s->npages)
 621                         continue;
 622                 if (!((base_gfn + npages <= s->base_gfn) ||
 623                       (base_gfn >= s->base_gfn + s->npages)))
 624                         goto out_free;
 625         }
 626
 627         /* Free page dirty bitmap if unneeded */
 628         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 629                 new.dirty_bitmap = NULL;
 630
 631         r = -ENOMEM;
 632
 633         /* Allocate if a slot is being created */
 634 #ifndef CONFIG_S390
 635         if (npages && !new.rmap) {
 636                 new.rmap = vmalloc(npages * sizeof(*new.rmap));
 637
 638                 if (!new.rmap)
 639                         goto out_free;
 640
 641                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 642
 643                 new.user_alloc = user_alloc;
 644                 new.userspace_addr = mem->userspace_addr;
 645         }
 646         if (!npages)
 647                 goto skip_lpage;
 648
 649         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 650                 unsigned long ugfn;
 651                 unsigned long j;
 652                 int lpages;
 653                 int level = i + 2;
 654
 655                 /* Avoid unused variable warning if no large pages */
 656                 (void)level;
 657
 658                 if (new.lpage_info[i])
 659                         continue;
 660
 661                 lpages = 1 + ((base_gfn + npages - 1)
 662                              >> KVM_HPAGE_GFN_SHIFT(level));
 663                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
 664
 665                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
 666
 667                 if (!new.lpage_info[i])
 668                         goto out_free;
 669
 670                 memset(new.lpage_info[i], 0,
 671                        lpages * sizeof(*new.lpage_info[i]));
 672
 673                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
 674                         new.lpage_info[i][0].write_count = 1;
 675                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
 676                         new.lpage_info[i][lpages - 1].write_count = 1;
 677                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 678                 /*
 679                  * If the gfn and userspace address are not aligned wrt each
 680                  * other, or if explicitly asked to, disable large page
 681                  * support for this slot
 682                  */
 683                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 684                     !largepages_enabled)
 685                         for (j = 0; j < lpages; ++j)
 686                                 new.lpage_info[i][j].write_count = 1;
 687         }
 688
 689 skip_lpage:
 690
 691         /* Allocate page dirty bitmap if needed */
 692         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 693                 if (kvm_create_dirty_bitmap(&new) < 0)
 694                         goto out_free;
 695                 /* destroy any largepage mappings for dirty tracking */
 696                 if (old.npages)
 697                         flush_shadow = 1;
 698         }
 699 #else  /* not defined CONFIG_S390 */
 700         new.user_alloc = user_alloc;
 701         if (user_alloc)
 702                 new.userspace_addr = mem->userspace_addr;
 703 #endif /* not defined CONFIG_S390 */
 704
 705         if (!npages) {
 706                 r = -ENOMEM;
 707                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 708                 if (!slots)
 709                         goto out_free;
 710                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 711                 if (mem->slot >= slots->nmemslots)
 712                         slots->nmemslots = mem->slot + 1;
 713                 slots->generation++;
 714                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 715
 716                 old_memslots = kvm->memslots;
 717                 rcu_assign_pointer(kvm->memslots, slots);
 718                 synchronize_srcu_expedited(&kvm->srcu);
 719                 /* From this point no new shadow pages pointing to a deleted
 720                  * memslot will be created.
 721                  *
 722                  * validation of sp->gfn happens in:
 723                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 724                  *      - kvm_is_visible_gfn (mmu_check_roots)
 725                  */
 726                 kvm_arch_flush_shadow(kvm);
 727                 kfree(old_memslots);
 728         }
 729
 730         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 731         if (r)
 732                 goto out_free;
 733
 734         /* map the pages in iommu page table */
 735         if (npages) {
 736                 r = kvm_iommu_map_pages(kvm, &new);
 737                 if (r)
 738                         goto out_free;
 739         }
 740
 741         r = -ENOMEM;
 742         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 743         if (!slots)
 744                 goto out_free;
 745         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 746         if (mem->slot >= slots->nmemslots)
 747                 slots->nmemslots = mem->slot + 1;
 748         slots->generation++;
 749
 750         /* actual memory is freed via old in kvm_free_physmem_slot below */
 751         if (!npages) {
 752                 new.rmap = NULL;
 753                 new.dirty_bitmap = NULL;
 754                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 755                         new.lpage_info[i] = NULL;
 756         }
 757
 758         slots->memslots[mem->slot] = new;
 759         old_memslots = kvm->memslots;
 760         rcu_assign_pointer(kvm->memslots, slots);
 761         synchronize_srcu_expedited(&kvm->srcu);
 762
 763         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 764
 765         kvm_free_physmem_slot(&old, &new);
 766         kfree(old_memslots);
 767
 768         if (flush_shadow)
 769                 kvm_arch_flush_shadow(kvm);
 770
 771         return 0;
 772
 773 out_free:
 774         kvm_free_physmem_slot(&new, &old);
 775 out:
 776         return r;
 777
 778 }
 779 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 780
 781 int kvm_set_memory_region(struct kvm *kvm,
 782                           struct kvm_userspace_memory_region *mem,
 783                           int user_alloc)
 784 {
 785         int r;
 786
 787         mutex_lock(&kvm->slots_lock);
 788         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 789         mutex_unlock(&kvm->slots_lock);
 790         return r;
 791 }
 792 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 793
 794 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 795                                    struct
 796                                    kvm_userspace_memory_region *mem,
 797                                    int user_alloc)
 798 {
 799         if (mem->slot >= KVM_MEMORY_SLOTS)
 800                 return -EINVAL;
 801         return kvm_set_memory_region(kvm, mem, user_alloc);
 802 }
 803
 804 int kvm_get_dirty_log(struct kvm *kvm,
 805                         struct kvm_dirty_log *log, int *is_dirty)
 806 {
 807         struct kvm_memory_slot *memslot;
 808         int r, i;
 809         unsigned long n;
 810         unsigned long any = 0;
 811
 812         r = -EINVAL;
 813         if (log->slot >= KVM_MEMORY_SLOTS)
 814                 goto out;
 815
 816         memslot = &kvm->memslots->memslots[log->slot];
 817         r = -ENOENT;
 818         if (!memslot->dirty_bitmap)
 819                 goto out;
 820
 821         n = kvm_dirty_bitmap_bytes(memslot);
 822
 823         for (i = 0; !any && i < n/sizeof(long); ++i)
 824                 any = memslot->dirty_bitmap[i];
 825
 826         r = -EFAULT;
 827         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 828                 goto out;
 829
 830         if (any)
 831                 *is_dirty = 1;
 832
 833         r = 0;
 834 out:
 835         return r;
 836 }
 837
 838 void kvm_disable_largepages(void)
 839 {
 840         largepages_enabled = false;
 841 }
 842 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 843
 844 int is_error_page(struct page *page)
 845 {
 846         return page == bad_page || page == hwpoison_page || page == fault_page;
 847 }
 848 EXPORT_SYMBOL_GPL(is_error_page);
 849
 850 int is_error_pfn(pfn_t pfn)
 851 {
 852         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 853 }
 854 EXPORT_SYMBOL_GPL(is_error_pfn);
 855
 856 int is_hwpoison_pfn(pfn_t pfn)
 857 {
 858         return pfn == hwpoison_pfn;
 859 }
 860 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 861
 862 int is_fault_pfn(pfn_t pfn)
 863 {
 864         return pfn == fault_pfn;
 865 }
 866 EXPORT_SYMBOL_GPL(is_fault_pfn);
 867
 868 static inline unsigned long bad_hva(void)
 869 {
 870         return PAGE_OFFSET;
 871 }
 872
 873 int kvm_is_error_hva(unsigned long addr)
 874 {
 875         return addr == bad_hva();
 876 }
 877 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 878
 879 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
 880                                                 gfn_t gfn)
 881 {
 882         int i;
 883
 884         for (i = 0; i < slots->nmemslots; ++i) {
 885                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 886
 887                 if (gfn >= memslot->base_gfn
 888                     && gfn < memslot->base_gfn + memslot->npages)
 889                         return memslot;
 890         }
 891         return NULL;
 892 }
 893
 894 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 895 {
 896         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 897 }
 898 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 899
 900 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 901 {
 902         int i;
 903         struct kvm_memslots *slots = kvm_memslots(kvm);
 904
 905         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 906                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 907
 908                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 909                         continue;
 910
 911                 if (gfn >= memslot->base_gfn
 912                     && gfn < memslot->base_gfn + memslot->npages)
 913                         return 1;
 914         }
 915         return 0;
 916 }
 917 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 918
 919 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 920 {
 921         struct vm_area_struct *vma;
 922         unsigned long addr, size;
 923
 924         size = PAGE_SIZE;
 925
 926         addr = gfn_to_hva(kvm, gfn);
 927         if (kvm_is_error_hva(addr))
 928                 return PAGE_SIZE;
 929
 930         down_read(&current->mm->mmap_sem);
 931         vma = find_vma(current->mm, addr);
 932         if (!vma)
 933                 goto out;
 934
 935         size = vma_kernel_pagesize(vma);
 936
 937 out:
 938         up_read(&current->mm->mmap_sem);
 939
 940         return size;
 941 }
 942
 943 int memslot_id(struct kvm *kvm, gfn_t gfn)
 944 {
 945         int i;
 946         struct kvm_memslots *slots = kvm_memslots(kvm);
 947         struct kvm_memory_slot *memslot = NULL;
 948
 949         for (i = 0; i < slots->nmemslots; ++i) {
 950                 memslot = &slots->memslots[i];
 951
 952                 if (gfn >= memslot->base_gfn
 953                     && gfn < memslot->base_gfn + memslot->npages)
 954                         break;
 955         }
 956
 957         return memslot - slots->memslots;
 958 }
 959
 960 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
 961                                      gfn_t *nr_pages)
 962 {
 963         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 964                 return bad_hva();
 965
 966         if (nr_pages)
 967                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
 968
 969         return gfn_to_hva_memslot(slot, gfn);
 970 }
 971
 972 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 973 {
 974         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
 975 }
 976 EXPORT_SYMBOL_GPL(gfn_to_hva);
 977
 978 static pfn_t get_fault_pfn(void)
 979 {
 980         get_page(fault_page);
 981         return fault_pfn;
 982 }
 983
 984 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
 985                         bool *async, bool write_fault, bool *writable)
 986 {
 987         struct page *page[1];
 988         int npages = 0;
 989         pfn_t pfn;
 990
 991         /* we can do it either atomically or asynchronously, not both */
 992         BUG_ON(atomic && async);
 993
 994         BUG_ON(!write_fault && !writable);
 995
 996         if (writable)
 997                 *writable = true;
 998
 999         if (atomic || async)
1000                 npages = __get_user_pages_fast(addr, 1, 1, page);
1001
1002         if (unlikely(npages != 1) && !atomic) {
1003                 might_sleep();
1004
1005                 if (writable)
1006                         *writable = write_fault;
1007
1008                 npages = get_user_pages_fast(addr, 1, write_fault, page);
1009
1010                 /* map read fault as writable if possible */
1011                 if (unlikely(!write_fault) && npages == 1) {
1012                         struct page *wpage[1];
1013
1014                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1015                         if (npages == 1) {
1016                                 *writable = true;
1017                                 put_page(page[0]);
1018                                 page[0] = wpage[0];
1019                         }
1020                         npages = 1;
1021                 }
1022         }
1023
1024         if (unlikely(npages != 1)) {
1025                 struct vm_area_struct *vma;
1026
1027                 if (atomic)
1028                         return get_fault_pfn();
1029
1030                 down_read(&current->mm->mmap_sem);
1031                 if (is_hwpoison_address(addr)) {
1032                         up_read(&current->mm->mmap_sem);
1033                         get_page(hwpoison_page);
1034                         return page_to_pfn(hwpoison_page);
1035                 }
1036
1037                 vma = find_vma_intersection(current->mm, addr, addr+1);
1038
1039                 if (vma == NULL)
1040                         pfn = get_fault_pfn();
1041                 else if ((vma->vm_flags & VM_PFNMAP)) {
1042                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1043                                 vma->vm_pgoff;
1044                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1045                 } else {
1046                         if (async && (vma->vm_flags & VM_WRITE))
1047                                 *async = true;
1048                         pfn = get_fault_pfn();
1049                 }
1050                 up_read(&current->mm->mmap_sem);
1051         } else
1052                 pfn = page_to_pfn(page[0]);
1053
1054         return pfn;
1055 }
1056
1057 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1058 {
1059         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1060 }
1061 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1062
1063 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1064                           bool write_fault, bool *writable)
1065 {
1066         unsigned long addr;
1067
1068         if (async)
1069                 *async = false;
1070
1071         addr = gfn_to_hva(kvm, gfn);
1072         if (kvm_is_error_hva(addr)) {
1073                 get_page(bad_page);
1074                 return page_to_pfn(bad_page);
1075         }
1076
1077         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1078 }
1079
1080 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1081 {
1082         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1083 }
1084 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1085
1086 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1087                        bool write_fault, bool *writable)
1088 {
1089         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1090 }
1091 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1092
1093 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1094 {
1095         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1096 }
1097 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1098
1099 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1100                       bool *writable)
1101 {
1102         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1103 }
1104 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1105
1106 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1107                          struct kvm_memory_slot *slot, gfn_t gfn)
1108 {
1109         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1110         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1111 }
1112
1113 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1114                                                                   int nr_pages)
1115 {
1116         unsigned long addr;
1117         gfn_t entry;
1118
1119         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1120         if (kvm_is_error_hva(addr))
1121                 return -1;
1122
1123         if (entry < nr_pages)
1124                 return 0;
1125
1126         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1127 }
1128 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1129
1130 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1131 {
1132         pfn_t pfn;
1133
1134         pfn = gfn_to_pfn(kvm, gfn);
1135         if (!kvm_is_mmio_pfn(pfn))
1136                 return pfn_to_page(pfn);
1137
1138         WARN_ON(kvm_is_mmio_pfn(pfn));
1139
1140         get_page(bad_page);
1141         return bad_page;
1142 }
1143
1144 EXPORT_SYMBOL_GPL(gfn_to_page);
1145
1146 void kvm_release_page_clean(struct page *page)
1147 {
1148         kvm_release_pfn_clean(page_to_pfn(page));
1149 }
1150 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1151
1152 void kvm_release_pfn_clean(pfn_t pfn)
1153 {
1154         if (!kvm_is_mmio_pfn(pfn))
1155                 put_page(pfn_to_page(pfn));
1156 }
1157 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1158
1159 void kvm_release_page_dirty(struct page *page)
1160 {
1161         kvm_release_pfn_dirty(page_to_pfn(page));
1162 }
1163 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1164
1165 void kvm_release_pfn_dirty(pfn_t pfn)
1166 {
1167         kvm_set_pfn_dirty(pfn);
1168         kvm_release_pfn_clean(pfn);
1169 }
1170 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1171
1172 void kvm_set_page_dirty(struct page *page)
1173 {
1174         kvm_set_pfn_dirty(page_to_pfn(page));
1175 }
1176 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1177
1178 void kvm_set_pfn_dirty(pfn_t pfn)
1179 {
1180         if (!kvm_is_mmio_pfn(pfn)) {
1181                 struct page *page = pfn_to_page(pfn);
1182                 if (!PageReserved(page))
1183                         SetPageDirty(page);
1184         }
1185 }
1186 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1187
1188 void kvm_set_pfn_accessed(pfn_t pfn)
1189 {
1190         if (!kvm_is_mmio_pfn(pfn))
1191                 mark_page_accessed(pfn_to_page(pfn));
1192 }
1193 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1194
1195 void kvm_get_pfn(pfn_t pfn)
1196 {
1197         if (!kvm_is_mmio_pfn(pfn))
1198                 get_page(pfn_to_page(pfn));
1199 }
1200 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1201
1202 static int next_segment(unsigned long len, int offset)
1203 {
1204         if (len > PAGE_SIZE - offset)
1205                 return PAGE_SIZE - offset;
1206         else
1207                 return len;
1208 }
1209
1210 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1211                         int len)
1212 {
1213         int r;
1214         unsigned long addr;
1215
1216         addr = gfn_to_hva(kvm, gfn);
1217         if (kvm_is_error_hva(addr))
1218                 return -EFAULT;
1219         r = copy_from_user(data, (void __user *)addr + offset, len);
1220         if (r)
1221                 return -EFAULT;
1222         return 0;
1223 }
1224 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1225
1226 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1227 {
1228         gfn_t gfn = gpa >> PAGE_SHIFT;
1229         int seg;
1230         int offset = offset_in_page(gpa);
1231         int ret;
1232
1233         while ((seg = next_segment(len, offset)) != 0) {
1234                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1235                 if (ret < 0)
1236                         return ret;
1237                 offset = 0;
1238                 len -= seg;
1239                 data += seg;
1240                 ++gfn;
1241         }
1242         return 0;
1243 }
1244 EXPORT_SYMBOL_GPL(kvm_read_guest);
1245
1246 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1247                           unsigned long len)
1248 {
1249         int r;
1250         unsigned long addr;
1251         gfn_t gfn = gpa >> PAGE_SHIFT;
1252         int offset = offset_in_page(gpa);
1253
1254         addr = gfn_to_hva(kvm, gfn);
1255         if (kvm_is_error_hva(addr))
1256                 return -EFAULT;
1257         pagefault_disable();
1258         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1259         pagefault_enable();
1260         if (r)
1261                 return -EFAULT;
1262         return 0;
1263 }
1264 EXPORT_SYMBOL(kvm_read_guest_atomic);
1265
1266 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1267                          int offset, int len)
1268 {
1269         int r;
1270         unsigned long addr;
1271
1272         addr = gfn_to_hva(kvm, gfn);
1273         if (kvm_is_error_hva(addr))
1274                 return -EFAULT;
1275         r = copy_to_user((void __user *)addr + offset, data, len);
1276         if (r)
1277                 return -EFAULT;
1278         mark_page_dirty(kvm, gfn);
1279         return 0;
1280 }
1281 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1282
1283 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1284                     unsigned long len)
1285 {
1286         gfn_t gfn = gpa >> PAGE_SHIFT;
1287         int seg;
1288         int offset = offset_in_page(gpa);
1289         int ret;
1290
1291         while ((seg = next_segment(len, offset)) != 0) {
1292                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1293                 if (ret < 0)
1294                         return ret;
1295                 offset = 0;
1296                 len -= seg;
1297                 data += seg;
1298                 ++gfn;
1299         }
1300         return 0;
1301 }
1302
1303 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1304                               gpa_t gpa)
1305 {
1306         struct kvm_memslots *slots = kvm_memslots(kvm);
1307         int offset = offset_in_page(gpa);
1308         gfn_t gfn = gpa >> PAGE_SHIFT;
1309
1310         ghc->gpa = gpa;
1311         ghc->generation = slots->generation;
1312         ghc->memslot = __gfn_to_memslot(slots, gfn);
1313         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1314         if (!kvm_is_error_hva(ghc->hva))
1315                 ghc->hva += offset;
1316         else
1317                 return -EFAULT;
1318
1319         return 0;
1320 }
1321 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1322
1323 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1324                            void *data, unsigned long len)
1325 {
1326         struct kvm_memslots *slots = kvm_memslots(kvm);
1327         int r;
1328
1329         if (slots->generation != ghc->generation)
1330                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1331
1332         if (kvm_is_error_hva(ghc->hva))
1333                 return -EFAULT;
1334
1335         r = copy_to_user((void __user *)ghc->hva, data, len);
1336         if (r)
1337                 return -EFAULT;
1338         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1339
1340         return 0;
1341 }
1342 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1343
1344 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1345 {
1346         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1347 }
1348 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1349
1350 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1351 {
1352         gfn_t gfn = gpa >> PAGE_SHIFT;
1353         int seg;
1354         int offset = offset_in_page(gpa);
1355         int ret;
1356
1357         while ((seg = next_segment(len, offset)) != 0) {
1358                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1359                 if (ret < 0)
1360                         return ret;
1361                 offset = 0;
1362                 len -= seg;
1363                 ++gfn;
1364         }
1365         return 0;
1366 }
1367 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1368
1369 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1370                              gfn_t gfn)
1371 {
1372         if (memslot && memslot->dirty_bitmap) {
1373                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1374
1375                 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1376         }
1377 }
1378
1379 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1380 {
1381         struct kvm_memory_slot *memslot;
1382
1383         memslot = gfn_to_memslot(kvm, gfn);
1384         mark_page_dirty_in_slot(kvm, memslot, gfn);
1385 }
1386
1387 /*
1388  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1389  */
1390 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1391 {
1392         DEFINE_WAIT(wait);
1393
1394         for (;;) {
1395                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1396
1397                 if (kvm_arch_vcpu_runnable(vcpu)) {
1398                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1399                         break;
1400                 }
1401                 if (kvm_cpu_has_pending_timer(vcpu))
1402                         break;
1403                 if (signal_pending(current))
1404                         break;
1405
1406                 schedule();
1407         }
1408
1409         finish_wait(&vcpu->wq, &wait);
1410 }
1411
1412 void kvm_resched(struct kvm_vcpu *vcpu)
1413 {
1414         if (!need_resched())
1415                 return;
1416         cond_resched();
1417 }
1418 EXPORT_SYMBOL_GPL(kvm_resched);
1419
1420 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1421 {
1422         ktime_t expires;
1423         DEFINE_WAIT(wait);
1424
1425         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1426
1427         /* Sleep for 100 us, and hope lock-holder got scheduled */
1428         expires = ktime_add_ns(ktime_get(), 100000UL);
1429         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1430
1431         finish_wait(&vcpu->wq, &wait);
1432 }
1433 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1434
1435 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1436 {
1437         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1438         struct page *page;
1439
1440         if (vmf->pgoff == 0)
1441                 page = virt_to_page(vcpu->run);
1442 #ifdef CONFIG_X86
1443         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1444                 page = virt_to_page(vcpu->arch.pio_data);
1445 #endif
1446 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1447         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1448                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1449 #endif
1450         else
1451                 return VM_FAULT_SIGBUS;
1452         get_page(page);
1453         vmf->page = page;
1454         return 0;
1455 }
1456
1457 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1458         .fault = kvm_vcpu_fault,
1459 };
1460
1461 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1462 {
1463         vma->vm_ops = &kvm_vcpu_vm_ops;
1464         return 0;
1465 }
1466
1467 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1468 {
1469         struct kvm_vcpu *vcpu = filp->private_data;
1470
1471         kvm_put_kvm(vcpu->kvm);
1472         return 0;
1473 }
1474
1475 static struct file_operations kvm_vcpu_fops = {
1476         .release        = kvm_vcpu_release,
1477         .unlocked_ioctl = kvm_vcpu_ioctl,
1478         .compat_ioctl   = kvm_vcpu_ioctl,
1479         .mmap           = kvm_vcpu_mmap,
1480         .llseek         = noop_llseek,
1481 };
1482
1483 /*
1484  * Allocates an inode for the vcpu.
1485  */
1486 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1487 {
1488         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1489 }
1490
1491 /*
1492  * Creates some virtual cpus.  Good luck creating more than one.
1493  */
1494 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1495 {
1496         int r;
1497         struct kvm_vcpu *vcpu, *v;
1498
1499         vcpu = kvm_arch_vcpu_create(kvm, id);
1500         if (IS_ERR(vcpu))
1501                 return PTR_ERR(vcpu);
1502
1503         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1504
1505         r = kvm_arch_vcpu_setup(vcpu);
1506         if (r)
1507                 return r;
1508
1509         mutex_lock(&kvm->lock);
1510         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1511                 r = -EINVAL;
1512                 goto vcpu_destroy;
1513         }
1514
1515         kvm_for_each_vcpu(r, v, kvm)
1516                 if (v->vcpu_id == id) {
1517                         r = -EEXIST;
1518                         goto vcpu_destroy;
1519                 }
1520
1521         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1522
1523         /* Now it's all set up, let userspace reach it */
1524         kvm_get_kvm(kvm);
1525         r = create_vcpu_fd(vcpu);
1526         if (r < 0) {
1527                 kvm_put_kvm(kvm);
1528                 goto vcpu_destroy;
1529         }
1530
1531         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1532         smp_wmb();
1533         atomic_inc(&kvm->online_vcpus);
1534
1535 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1536         if (kvm->bsp_vcpu_id == id)
1537                 kvm->bsp_vcpu = vcpu;
1538 #endif
1539         mutex_unlock(&kvm->lock);
1540         return r;
1541
1542 vcpu_destroy:
1543         mutex_unlock(&kvm->lock);
1544         kvm_arch_vcpu_destroy(vcpu);
1545         return r;
1546 }
1547
1548 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1549 {
1550         if (sigset) {
1551                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1552                 vcpu->sigset_active = 1;
1553                 vcpu->sigset = *sigset;
1554         } else
1555                 vcpu->sigset_active = 0;
1556         return 0;
1557 }
1558
1559 static long kvm_vcpu_ioctl(struct file *filp,
1560                            unsigned int ioctl, unsigned long arg)
1561 {
1562         struct kvm_vcpu *vcpu = filp->private_data;
1563         void __user *argp = (void __user *)arg;
1564         int r;
1565         struct kvm_fpu *fpu = NULL;
1566         struct kvm_sregs *kvm_sregs = NULL;
1567
1568         if (vcpu->kvm->mm != current->mm)
1569                 return -EIO;
1570
1571 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1572         /*
1573          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1574          * so vcpu_load() would break it.
1575          */
1576         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1577                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1578 #endif
1579
1580
1581         vcpu_load(vcpu);
1582         switch (ioctl) {
1583         case KVM_RUN:
1584                 r = -EINVAL;
1585                 if (arg)
1586                         goto out;
1587                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1588                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1589                 break;
1590         case KVM_GET_REGS: {
1591                 struct kvm_regs *kvm_regs;
1592
1593                 r = -ENOMEM;
1594                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1595                 if (!kvm_regs)
1596                         goto out;
1597                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1598                 if (r)
1599                         goto out_free1;
1600                 r = -EFAULT;
1601                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1602                         goto out_free1;
1603                 r = 0;
1604 out_free1:
1605                 kfree(kvm_regs);
1606                 break;
1607         }
1608         case KVM_SET_REGS: {
1609                 struct kvm_regs *kvm_regs;
1610
1611                 r = -ENOMEM;
1612                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1613                 if (!kvm_regs)
1614                         goto out;
1615                 r = -EFAULT;
1616                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1617                         goto out_free2;
1618                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1619                 if (r)
1620                         goto out_free2;
1621                 r = 0;
1622 out_free2:
1623                 kfree(kvm_regs);
1624                 break;
1625         }
1626         case KVM_GET_SREGS: {
1627                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1628                 r = -ENOMEM;
1629                 if (!kvm_sregs)
1630                         goto out;
1631                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1632                 if (r)
1633                         goto out;
1634                 r = -EFAULT;
1635                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1636                         goto out;
1637                 r = 0;
1638                 break;
1639         }
1640         case KVM_SET_SREGS: {
1641                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1642                 r = -ENOMEM;
1643                 if (!kvm_sregs)
1644                         goto out;
1645                 r = -EFAULT;
1646                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1647                         goto out;
1648                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1649                 if (r)
1650                         goto out;
1651                 r = 0;
1652                 break;
1653         }
1654         case KVM_GET_MP_STATE: {
1655                 struct kvm_mp_state mp_state;
1656
1657                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1658                 if (r)
1659                         goto out;
1660                 r = -EFAULT;
1661                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1662                         goto out;
1663                 r = 0;
1664                 break;
1665         }
1666         case KVM_SET_MP_STATE: {
1667                 struct kvm_mp_state mp_state;
1668
1669                 r = -EFAULT;
1670                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1671                         goto out;
1672                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1673                 if (r)
1674                         goto out;
1675                 r = 0;
1676                 break;
1677         }
1678         case KVM_TRANSLATE: {
1679                 struct kvm_translation tr;
1680
1681                 r = -EFAULT;
1682                 if (copy_from_user(&tr, argp, sizeof tr))
1683                         goto out;
1684                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1685                 if (r)
1686                         goto out;
1687                 r = -EFAULT;
1688                 if (copy_to_user(argp, &tr, sizeof tr))
1689                         goto out;
1690                 r = 0;
1691                 break;
1692         }
1693         case KVM_SET_GUEST_DEBUG: {
1694                 struct kvm_guest_debug dbg;
1695
1696                 r = -EFAULT;
1697                 if (copy_from_user(&dbg, argp, sizeof dbg))
1698                         goto out;
1699                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1700                 if (r)
1701                         goto out;
1702                 r = 0;
1703                 break;
1704         }
1705         case KVM_SET_SIGNAL_MASK: {
1706                 struct kvm_signal_mask __user *sigmask_arg = argp;
1707                 struct kvm_signal_mask kvm_sigmask;
1708                 sigset_t sigset, *p;
1709
1710                 p = NULL;
1711                 if (argp) {
1712                         r = -EFAULT;
1713                         if (copy_from_user(&kvm_sigmask, argp,
1714                                            sizeof kvm_sigmask))
1715                                 goto out;
1716                         r = -EINVAL;
1717                         if (kvm_sigmask.len != sizeof sigset)
1718                                 goto out;
1719                         r = -EFAULT;
1720                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1721                                            sizeof sigset))
1722                                 goto out;
1723                         p = &sigset;
1724                 }
1725                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1726                 break;
1727         }
1728         case KVM_GET_FPU: {
1729                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1730                 r = -ENOMEM;
1731                 if (!fpu)
1732                         goto out;
1733                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1734                 if (r)
1735                         goto out;
1736                 r = -EFAULT;
1737                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1738                         goto out;
1739                 r = 0;
1740                 break;
1741         }
1742         case KVM_SET_FPU: {
1743                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1744                 r = -ENOMEM;
1745                 if (!fpu)
1746                         goto out;
1747                 r = -EFAULT;
1748                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1749                         goto out;
1750                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1751                 if (r)
1752                         goto out;
1753                 r = 0;
1754                 break;
1755         }
1756         default:
1757                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1758         }
1759 out:
1760         vcpu_put(vcpu);
1761         kfree(fpu);
1762         kfree(kvm_sregs);
1763         return r;
1764 }
1765
1766 static long kvm_vm_ioctl(struct file *filp,
1767                            unsigned int ioctl, unsigned long arg)
1768 {
1769         struct kvm *kvm = filp->private_data;
1770         void __user *argp = (void __user *)arg;
1771         int r;
1772
1773         if (kvm->mm != current->mm)
1774                 return -EIO;
1775         switch (ioctl) {
1776         case KVM_CREATE_VCPU:
1777                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1778                 if (r < 0)
1779                         goto out;
1780                 break;
1781         case KVM_SET_USER_MEMORY_REGION: {
1782                 struct kvm_userspace_memory_region kvm_userspace_mem;
1783
1784                 r = -EFAULT;
1785                 if (copy_from_user(&kvm_userspace_mem, argp,
1786                                                 sizeof kvm_userspace_mem))
1787                         goto out;
1788
1789                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1790                 if (r)
1791                         goto out;
1792                 break;
1793         }
1794         case KVM_GET_DIRTY_LOG: {
1795                 struct kvm_dirty_log log;
1796
1797                 r = -EFAULT;
1798                 if (copy_from_user(&log, argp, sizeof log))
1799                         goto out;
1800                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1801                 if (r)
1802                         goto out;
1803                 break;
1804         }
1805 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1806         case KVM_REGISTER_COALESCED_MMIO: {
1807                 struct kvm_coalesced_mmio_zone zone;
1808                 r = -EFAULT;
1809                 if (copy_from_user(&zone, argp, sizeof zone))
1810                         goto out;
1811                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1812                 if (r)
1813                         goto out;
1814                 r = 0;
1815                 break;
1816         }
1817         case KVM_UNREGISTER_COALESCED_MMIO: {
1818                 struct kvm_coalesced_mmio_zone zone;
1819                 r = -EFAULT;
1820                 if (copy_from_user(&zone, argp, sizeof zone))
1821                         goto out;
1822                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1823                 if (r)
1824                         goto out;
1825                 r = 0;
1826                 break;
1827         }
1828 #endif
1829         case KVM_IRQFD: {
1830                 struct kvm_irqfd data;
1831
1832                 r = -EFAULT;
1833                 if (copy_from_user(&data, argp, sizeof data))
1834                         goto out;
1835                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1836                 break;
1837         }
1838         case KVM_IOEVENTFD: {
1839                 struct kvm_ioeventfd data;
1840
1841                 r = -EFAULT;
1842                 if (copy_from_user(&data, argp, sizeof data))
1843                         goto out;
1844                 r = kvm_ioeventfd(kvm, &data);
1845                 break;
1846         }
1847 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1848         case KVM_SET_BOOT_CPU_ID:
1849                 r = 0;
1850                 mutex_lock(&kvm->lock);
1851                 if (atomic_read(&kvm->online_vcpus) != 0)
1852                         r = -EBUSY;
1853                 else
1854                         kvm->bsp_vcpu_id = arg;
1855                 mutex_unlock(&kvm->lock);
1856                 break;
1857 #endif
1858         default:
1859                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1860                 if (r == -ENOTTY)
1861                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1862         }
1863 out:
1864         return r;
1865 }
1866
1867 #ifdef CONFIG_COMPAT
1868 struct compat_kvm_dirty_log {
1869         __u32 slot;
1870         __u32 padding1;
1871         union {
1872                 compat_uptr_t dirty_bitmap; /* one bit per page */
1873                 __u64 padding2;
1874         };
1875 };
1876
1877 static long kvm_vm_compat_ioctl(struct file *filp,
1878                            unsigned int ioctl, unsigned long arg)
1879 {
1880         struct kvm *kvm = filp->private_data;
1881         int r;
1882
1883         if (kvm->mm != current->mm)
1884                 return -EIO;
1885         switch (ioctl) {
1886         case KVM_GET_DIRTY_LOG: {
1887                 struct compat_kvm_dirty_log compat_log;
1888                 struct kvm_dirty_log log;
1889
1890                 r = -EFAULT;
1891                 if (copy_from_user(&compat_log, (void __user *)arg,
1892                                    sizeof(compat_log)))
1893                         goto out;
1894                 log.slot         = compat_log.slot;
1895                 log.padding1     = compat_log.padding1;
1896                 log.padding2     = compat_log.padding2;
1897                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1898
1899                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1900                 if (r)
1901                         goto out;
1902                 break;
1903         }
1904         default:
1905                 r = kvm_vm_ioctl(filp, ioctl, arg);
1906         }
1907
1908 out:
1909         return r;
1910 }
1911 #endif
1912
1913 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1914 {
1915         struct page *page[1];
1916         unsigned long addr;
1917         int npages;
1918         gfn_t gfn = vmf->pgoff;
1919         struct kvm *kvm = vma->vm_file->private_data;
1920
1921         addr = gfn_to_hva(kvm, gfn);
1922         if (kvm_is_error_hva(addr))
1923                 return VM_FAULT_SIGBUS;
1924
1925         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1926                                 NULL);
1927         if (unlikely(npages != 1))
1928                 return VM_FAULT_SIGBUS;
1929
1930         vmf->page = page[0];
1931         return 0;
1932 }
1933
1934 static const struct vm_operations_struct kvm_vm_vm_ops = {
1935         .fault = kvm_vm_fault,
1936 };
1937
1938 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1939 {
1940         vma->vm_ops = &kvm_vm_vm_ops;
1941         return 0;
1942 }
1943
1944 static struct file_operations kvm_vm_fops = {
1945         .release        = kvm_vm_release,
1946         .unlocked_ioctl = kvm_vm_ioctl,
1947 #ifdef CONFIG_COMPAT
1948         .compat_ioctl   = kvm_vm_compat_ioctl,
1949 #endif
1950         .mmap           = kvm_vm_mmap,
1951         .llseek         = noop_llseek,
1952 };
1953
1954 static int kvm_dev_ioctl_create_vm(void)
1955 {
1956         int fd, r;
1957         struct kvm *kvm;
1958
1959         kvm = kvm_create_vm();
1960         if (IS_ERR(kvm))
1961                 return PTR_ERR(kvm);
1962 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1963         r = kvm_coalesced_mmio_init(kvm);
1964         if (r < 0) {
1965                 kvm_put_kvm(kvm);
1966                 return r;
1967         }
1968 #endif
1969         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1970         if (fd < 0)
1971                 kvm_put_kvm(kvm);
1972
1973         return fd;
1974 }
1975
1976 static long kvm_dev_ioctl_check_extension_generic(long arg)
1977 {
1978         switch (arg) {
1979         case KVM_CAP_USER_MEMORY:
1980         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1981         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1982 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1983         case KVM_CAP_SET_BOOT_CPU_ID:
1984 #endif
1985         case KVM_CAP_INTERNAL_ERROR_DATA:
1986                 return 1;
1987 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1988         case KVM_CAP_IRQ_ROUTING:
1989                 return KVM_MAX_IRQ_ROUTES;
1990 #endif
1991         default:
1992                 break;
1993         }
1994         return kvm_dev_ioctl_check_extension(arg);
1995 }
1996
1997 static long kvm_dev_ioctl(struct file *filp,
1998                           unsigned int ioctl, unsigned long arg)
1999 {
2000         long r = -EINVAL;
2001
2002         switch (ioctl) {
2003         case KVM_GET_API_VERSION:
2004                 r = -EINVAL;
2005                 if (arg)
2006                         goto out;
2007                 r = KVM_API_VERSION;
2008                 break;
2009         case KVM_CREATE_VM:
2010                 r = -EINVAL;
2011                 if (arg)
2012                         goto out;
2013                 r = kvm_dev_ioctl_create_vm();
2014                 break;
2015         case KVM_CHECK_EXTENSION:
2016                 r = kvm_dev_ioctl_check_extension_generic(arg);
2017                 break;
2018         case KVM_GET_VCPU_MMAP_SIZE:
2019                 r = -EINVAL;
2020                 if (arg)
2021                         goto out;
2022                 r = PAGE_SIZE;     /* struct kvm_run */
2023 #ifdef CONFIG_X86
2024                 r += PAGE_SIZE;    /* pio data page */
2025 #endif
2026 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2027                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2028 #endif
2029                 break;
2030         case KVM_TRACE_ENABLE:
2031         case KVM_TRACE_PAUSE:
2032         case KVM_TRACE_DISABLE:
2033                 r = -EOPNOTSUPP;
2034                 break;
2035         default:
2036                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2037         }
2038 out:
2039         return r;
2040 }
2041
2042 static struct file_operations kvm_chardev_ops = {
2043         .unlocked_ioctl = kvm_dev_ioctl,
2044         .compat_ioctl   = kvm_dev_ioctl,
2045         .llseek         = noop_llseek,
2046 };
2047
2048 static struct miscdevice kvm_dev = {
2049         KVM_MINOR,
2050         "kvm",
2051         &kvm_chardev_ops,
2052 };
2053
2054 static void hardware_enable(void *junk)
2055 {
2056         int cpu = raw_smp_processor_id();
2057         int r;
2058
2059         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2060                 return;
2061
2062         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2063
2064         r = kvm_arch_hardware_enable(NULL);
2065
2066         if (r) {
2067                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2068                 atomic_inc(&hardware_enable_failed);
2069                 printk(KERN_INFO "kvm: enabling virtualization on "
2070                                  "CPU%d failed\n", cpu);
2071         }
2072 }
2073
2074 static void hardware_disable(void *junk)
2075 {
2076         int cpu = raw_smp_processor_id();
2077
2078         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2079                 return;
2080         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2081         kvm_arch_hardware_disable(NULL);
2082 }
2083
2084 static void hardware_disable_all_nolock(void)
2085 {
2086         BUG_ON(!kvm_usage_count);
2087
2088         kvm_usage_count--;
2089         if (!kvm_usage_count)
2090                 on_each_cpu(hardware_disable, NULL, 1);
2091 }
2092
2093 static void hardware_disable_all(void)
2094 {
2095         spin_lock(&kvm_lock);
2096         hardware_disable_all_nolock();
2097         spin_unlock(&kvm_lock);
2098 }
2099
2100 static int hardware_enable_all(void)
2101 {
2102         int r = 0;
2103
2104         spin_lock(&kvm_lock);
2105
2106         kvm_usage_count++;
2107         if (kvm_usage_count == 1) {
2108                 atomic_set(&hardware_enable_failed, 0);
2109                 on_each_cpu(hardware_enable, NULL, 1);
2110
2111                 if (atomic_read(&hardware_enable_failed)) {
2112                         hardware_disable_all_nolock();
2113                         r = -EBUSY;
2114                 }
2115         }
2116
2117         spin_unlock(&kvm_lock);
2118
2119         return r;
2120 }
2121
2122 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2123                            void *v)
2124 {
2125         int cpu = (long)v;
2126
2127         if (!kvm_usage_count)
2128                 return NOTIFY_OK;
2129
2130         val &= ~CPU_TASKS_FROZEN;
2131         switch (val) {
2132         case CPU_DYING:
2133                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2134                        cpu);
2135                 hardware_disable(NULL);
2136                 break;
2137         case CPU_STARTING:
2138                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2139                        cpu);
2140                 spin_lock(&kvm_lock);
2141                 hardware_enable(NULL);
2142                 spin_unlock(&kvm_lock);
2143                 break;
2144         }
2145         return NOTIFY_OK;
2146 }
2147
2148
2149 asmlinkage void kvm_handle_fault_on_reboot(void)
2150 {
2151         if (kvm_rebooting) {
2152                 /* spin while reset goes on */
2153                 local_irq_enable();
2154                 while (true)
2155                         cpu_relax();
2156         }
2157         /* Fault while not rebooting.  We want the trace. */
2158         BUG();
2159 }
2160 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2161
2162 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2163                       void *v)
2164 {
2165         /*
2166          * Some (well, at least mine) BIOSes hang on reboot if
2167          * in vmx root mode.
2168          *
2169          * And Intel TXT required VMX off for all cpu when system shutdown.
2170          */
2171         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2172         kvm_rebooting = true;
2173         on_each_cpu(hardware_disable, NULL, 1);
2174         return NOTIFY_OK;
2175 }
2176
2177 static struct notifier_block kvm_reboot_notifier = {
2178         .notifier_call = kvm_reboot,
2179         .priority = 0,
2180 };
2181
2182 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2183 {
2184         int i;
2185
2186         for (i = 0; i < bus->dev_count; i++) {
2187                 struct kvm_io_device *pos = bus->devs[i];
2188
2189                 kvm_iodevice_destructor(pos);
2190         }
2191         kfree(bus);
2192 }
2193
2194 /* kvm_io_bus_write - called under kvm->slots_lock */
2195 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2196                      int len, const void *val)
2197 {
2198         int i;
2199         struct kvm_io_bus *bus;
2200
2201         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2202         for (i = 0; i < bus->dev_count; i++)
2203                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2204                         return 0;
2205         return -EOPNOTSUPP;
2206 }
2207
2208 /* kvm_io_bus_read - called under kvm->slots_lock */
2209 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2210                     int len, void *val)
2211 {
2212         int i;
2213         struct kvm_io_bus *bus;
2214
2215         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2216         for (i = 0; i < bus->dev_count; i++)
2217                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2218                         return 0;
2219         return -EOPNOTSUPP;
2220 }
2221
2222 /* Caller must hold slots_lock. */
2223 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2224                             struct kvm_io_device *dev)
2225 {
2226         struct kvm_io_bus *new_bus, *bus;
2227
2228         bus = kvm->buses[bus_idx];
2229         if (bus->dev_count > NR_IOBUS_DEVS-1)
2230                 return -ENOSPC;
2231
2232         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2233         if (!new_bus)
2234                 return -ENOMEM;
2235         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2236         new_bus->devs[new_bus->dev_count++] = dev;
2237         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2238         synchronize_srcu_expedited(&kvm->srcu);
2239         kfree(bus);
2240
2241         return 0;
2242 }
2243
2244 /* Caller must hold slots_lock. */
2245 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2246                               struct kvm_io_device *dev)
2247 {
2248         int i, r;
2249         struct kvm_io_bus *new_bus, *bus;
2250
2251         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2252         if (!new_bus)
2253                 return -ENOMEM;
2254
2255         bus = kvm->buses[bus_idx];
2256         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2257
2258         r = -ENOENT;
2259         for (i = 0; i < new_bus->dev_count; i++)
2260                 if (new_bus->devs[i] == dev) {
2261                         r = 0;
2262                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2263                         break;
2264                 }
2265
2266         if (r) {
2267                 kfree(new_bus);
2268                 return r;
2269         }
2270
2271         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2272         synchronize_srcu_expedited(&kvm->srcu);
2273         kfree(bus);
2274         return r;
2275 }
2276
2277 static struct notifier_block kvm_cpu_notifier = {
2278         .notifier_call = kvm_cpu_hotplug,
2279 };
2280
2281 static int vm_stat_get(void *_offset, u64 *val)
2282 {
2283         unsigned offset = (long)_offset;
2284         struct kvm *kvm;
2285
2286         *val = 0;
2287         spin_lock(&kvm_lock);
2288         list_for_each_entry(kvm, &vm_list, vm_list)
2289                 *val += *(u32 *)((void *)kvm + offset);
2290         spin_unlock(&kvm_lock);
2291         return 0;
2292 }
2293
2294 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2295
2296 static int vcpu_stat_get(void *_offset, u64 *val)
2297 {
2298         unsigned offset = (long)_offset;
2299         struct kvm *kvm;
2300         struct kvm_vcpu *vcpu;
2301         int i;
2302
2303         *val = 0;
2304         spin_lock(&kvm_lock);
2305         list_for_each_entry(kvm, &vm_list, vm_list)
2306                 kvm_for_each_vcpu(i, vcpu, kvm)
2307                         *val += *(u32 *)((void *)vcpu + offset);
2308
2309         spin_unlock(&kvm_lock);
2310         return 0;
2311 }
2312
2313 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2314
2315 static const struct file_operations *stat_fops[] = {
2316         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2317         [KVM_STAT_VM]   = &vm_stat_fops,
2318 };
2319
2320 static void kvm_init_debug(void)
2321 {
2322         struct kvm_stats_debugfs_item *p;
2323
2324         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2325         for (p = debugfs_entries; p->name; ++p)
2326                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2327                                                 (void *)(long)p->offset,
2328                                                 stat_fops[p->kind]);
2329 }
2330
2331 static void kvm_exit_debug(void)
2332 {
2333         struct kvm_stats_debugfs_item *p;
2334
2335         for (p = debugfs_entries; p->name; ++p)
2336                 debugfs_remove(p->dentry);
2337         debugfs_remove(kvm_debugfs_dir);
2338 }
2339
2340 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2341 {
2342         if (kvm_usage_count)
2343                 hardware_disable(NULL);
2344         return 0;
2345 }
2346
2347 static int kvm_resume(struct sys_device *dev)
2348 {
2349         if (kvm_usage_count) {
2350                 WARN_ON(spin_is_locked(&kvm_lock));
2351                 hardware_enable(NULL);
2352         }
2353         return 0;
2354 }
2355
2356 static struct sysdev_class kvm_sysdev_class = {
2357         .name = "kvm",
2358         .suspend = kvm_suspend,
2359         .resume = kvm_resume,
2360 };
2361
2362 static struct sys_device kvm_sysdev = {
2363         .id = 0,
2364         .cls = &kvm_sysdev_class,
2365 };
2366
2367 struct page *bad_page;
2368 pfn_t bad_pfn;
2369
2370 static inline
2371 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2372 {
2373         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2374 }
2375
2376 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2377 {
2378         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2379
2380         kvm_arch_vcpu_load(vcpu, cpu);
2381 }
2382
2383 static void kvm_sched_out(struct preempt_notifier *pn,
2384                           struct task_struct *next)
2385 {
2386         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2387
2388         kvm_arch_vcpu_put(vcpu);
2389 }
2390
2391 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2392                   struct module *module)
2393 {
2394         int r;
2395         int cpu;
2396
2397         r = kvm_arch_init(opaque);
2398         if (r)
2399                 goto out_fail;
2400
2401         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2402
2403         if (bad_page == NULL) {
2404                 r = -ENOMEM;
2405                 goto out;
2406         }
2407
2408         bad_pfn = page_to_pfn(bad_page);
2409
2410         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2411
2412         if (hwpoison_page == NULL) {
2413                 r = -ENOMEM;
2414                 goto out_free_0;
2415         }
2416
2417         hwpoison_pfn = page_to_pfn(hwpoison_page);
2418
2419         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2420
2421         if (fault_page == NULL) {
2422                 r = -ENOMEM;
2423                 goto out_free_0;
2424         }
2425
2426         fault_pfn = page_to_pfn(fault_page);
2427
2428         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2429                 r = -ENOMEM;
2430                 goto out_free_0;
2431         }
2432
2433         r = kvm_arch_hardware_setup();
2434         if (r < 0)
2435                 goto out_free_0a;
2436
2437         for_each_online_cpu(cpu) {
2438                 smp_call_function_single(cpu,
2439                                 kvm_arch_check_processor_compat,
2440                                 &r, 1);
2441                 if (r < 0)
2442                         goto out_free_1;
2443         }
2444
2445         r = register_cpu_notifier(&kvm_cpu_notifier);
2446         if (r)
2447                 goto out_free_2;
2448         register_reboot_notifier(&kvm_reboot_notifier);
2449
2450         r = sysdev_class_register(&kvm_sysdev_class);
2451         if (r)
2452                 goto out_free_3;
2453
2454         r = sysdev_register(&kvm_sysdev);
2455         if (r)
2456                 goto out_free_4;
2457
2458         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2459         if (!vcpu_align)
2460                 vcpu_align = __alignof__(struct kvm_vcpu);
2461         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2462                                            0, NULL);
2463         if (!kvm_vcpu_cache) {
2464                 r = -ENOMEM;
2465                 goto out_free_5;
2466         }
2467
2468         r = kvm_async_pf_init();
2469         if (r)
2470                 goto out_free;
2471
2472         kvm_chardev_ops.owner = module;
2473         kvm_vm_fops.owner = module;
2474         kvm_vcpu_fops.owner = module;
2475
2476         r = misc_register(&kvm_dev);
2477         if (r) {
2478                 printk(KERN_ERR "kvm: misc device register failed\n");
2479                 goto out_unreg;
2480         }
2481
2482         kvm_preempt_ops.sched_in = kvm_sched_in;
2483         kvm_preempt_ops.sched_out = kvm_sched_out;
2484
2485         kvm_init_debug();
2486
2487         return 0;
2488
2489 out_unreg:
2490         kvm_async_pf_deinit();
2491 out_free:
2492         kmem_cache_destroy(kvm_vcpu_cache);
2493 out_free_5:
2494         sysdev_unregister(&kvm_sysdev);
2495 out_free_4:
2496         sysdev_class_unregister(&kvm_sysdev_class);
2497 out_free_3:
2498         unregister_reboot_notifier(&kvm_reboot_notifier);
2499         unregister_cpu_notifier(&kvm_cpu_notifier);
2500 out_free_2:
2501 out_free_1:
2502         kvm_arch_hardware_unsetup();
2503 out_free_0a:
2504         free_cpumask_var(cpus_hardware_enabled);
2505 out_free_0:
2506         if (fault_page)
2507                 __free_page(fault_page);
2508         if (hwpoison_page)
2509                 __free_page(hwpoison_page);
2510         __free_page(bad_page);
2511 out:
2512         kvm_arch_exit();
2513 out_fail:
2514         return r;
2515 }
2516 EXPORT_SYMBOL_GPL(kvm_init);
2517
2518 void kvm_exit(void)
2519 {
2520         kvm_exit_debug();
2521         misc_deregister(&kvm_dev);
2522         kmem_cache_destroy(kvm_vcpu_cache);
2523         kvm_async_pf_deinit();
2524         sysdev_unregister(&kvm_sysdev);
2525         sysdev_class_unregister(&kvm_sysdev_class);
2526         unregister_reboot_notifier(&kvm_reboot_notifier);
2527         unregister_cpu_notifier(&kvm_cpu_notifier);
2528         on_each_cpu(hardware_disable, NULL, 1);
2529         kvm_arch_hardware_unsetup();
2530         kvm_arch_exit();
2531         free_cpumask_var(cpus_hardware_enabled);
2532         __free_page(hwpoison_page);
2533         __free_page(bad_page);
2534 }
2535 EXPORT_SYMBOL_GPL(kvm_exit);