drivers/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  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19 #include "x86.h"
  20 #include "x86_emulate.h"
  21 #include "irq.h"
  22
  23 #include <linux/kvm.h>
  24 #include <linux/module.h>
  25 #include <linux/errno.h>
  26 #include <linux/percpu.h>
  27 #include <linux/gfp.h>
  28 #include <linux/mm.h>
  29 #include <linux/miscdevice.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/reboot.h>
  32 #include <linux/debugfs.h>
  33 #include <linux/highmem.h>
  34 #include <linux/file.h>
  35 #include <linux/sysdev.h>
  36 #include <linux/cpu.h>
  37 #include <linux/sched.h>
  38 #include <linux/cpumask.h>
  39 #include <linux/smp.h>
  40 #include <linux/anon_inodes.h>
  41 #include <linux/profile.h>
  42 #include <linux/kvm_para.h>
  43 #include <linux/pagemap.h>
  44 #include <linux/mman.h>
  45
  46 #include <asm/processor.h>
  47 #include <asm/msr.h>
  48 #include <asm/io.h>
  49 #include <asm/uaccess.h>
  50 #include <asm/desc.h>
  51
  52 MODULE_AUTHOR("Qumranet");
  53 MODULE_LICENSE("GPL");
  54
  55 static DEFINE_SPINLOCK(kvm_lock);
  56 static LIST_HEAD(vm_list);
  57
  58 static cpumask_t cpus_hardware_enabled;
  59
  60 struct kvm_x86_ops *kvm_x86_ops;
  61 struct kmem_cache *kvm_vcpu_cache;
  62 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  63
  64 static __read_mostly struct preempt_ops kvm_preempt_ops;
  65
  66 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  67
  68 static struct kvm_stats_debugfs_item {
  69         const char *name;
  70         int offset;
  71         struct dentry *dentry;
  72 } debugfs_entries[] = {
  73         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  74         { "pf_guest", STAT_OFFSET(pf_guest) },
  75         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  76         { "invlpg", STAT_OFFSET(invlpg) },
  77         { "exits", STAT_OFFSET(exits) },
  78         { "io_exits", STAT_OFFSET(io_exits) },
  79         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  80         { "signal_exits", STAT_OFFSET(signal_exits) },
  81         { "irq_window", STAT_OFFSET(irq_window_exits) },
  82         { "halt_exits", STAT_OFFSET(halt_exits) },
  83         { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
  84         { "request_irq", STAT_OFFSET(request_irq_exits) },
  85         { "irq_exits", STAT_OFFSET(irq_exits) },
  86         { "light_exits", STAT_OFFSET(light_exits) },
  87         { "efer_reload", STAT_OFFSET(efer_reload) },
  88         { NULL }
  89 };
  90
  91 static struct dentry *debugfs_dir;
  92
  93 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  94
  95 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  96                            unsigned long arg);
  97
  98 static inline int valid_vcpu(int n)
  99 {
 100         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 101 }
 102
 103 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 104 {
 105         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 106                 return;
 107
 108         vcpu->guest_fpu_loaded = 1;
 109         fx_save(&vcpu->host_fx_image);
 110         fx_restore(&vcpu->guest_fx_image);
 111 }
 112 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 113
 114 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 115 {
 116         if (!vcpu->guest_fpu_loaded)
 117                 return;
 118
 119         vcpu->guest_fpu_loaded = 0;
 120         fx_save(&vcpu->guest_fx_image);
 121         fx_restore(&vcpu->host_fx_image);
 122 }
 123 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 124
 125 /*
 126  * Switches to specified vcpu, until a matching vcpu_put()
 127  */
 128 void vcpu_load(struct kvm_vcpu *vcpu)
 129 {
 130         int cpu;
 131
 132         mutex_lock(&vcpu->mutex);
 133         cpu = get_cpu();
 134         preempt_notifier_register(&vcpu->preempt_notifier);
 135         kvm_arch_vcpu_load(vcpu, cpu);
 136         put_cpu();
 137 }
 138
 139 void vcpu_put(struct kvm_vcpu *vcpu)
 140 {
 141         preempt_disable();
 142         kvm_arch_vcpu_put(vcpu);
 143         preempt_notifier_unregister(&vcpu->preempt_notifier);
 144         preempt_enable();
 145         mutex_unlock(&vcpu->mutex);
 146 }
 147
 148 static void ack_flush(void *_completed)
 149 {
 150 }
 151
 152 void kvm_flush_remote_tlbs(struct kvm *kvm)
 153 {
 154         int i, cpu;
 155         cpumask_t cpus;
 156         struct kvm_vcpu *vcpu;
 157
 158         cpus_clear(cpus);
 159         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 160                 vcpu = kvm->vcpus[i];
 161                 if (!vcpu)
 162                         continue;
 163                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 164                         continue;
 165                 cpu = vcpu->cpu;
 166                 if (cpu != -1 && cpu != raw_smp_processor_id())
 167                         cpu_set(cpu, cpus);
 168         }
 169         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 170 }
 171
 172 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 173 {
 174         struct page *page;
 175         int r;
 176
 177         mutex_init(&vcpu->mutex);
 178         vcpu->cpu = -1;
 179         vcpu->mmu.root_hpa = INVALID_PAGE;
 180         vcpu->kvm = kvm;
 181         vcpu->vcpu_id = id;
 182         if (!irqchip_in_kernel(kvm) || id == 0)
 183                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
 184         else
 185                 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
 186         init_waitqueue_head(&vcpu->wq);
 187
 188         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 189         if (!page) {
 190                 r = -ENOMEM;
 191                 goto fail;
 192         }
 193         vcpu->run = page_address(page);
 194
 195         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 196         if (!page) {
 197                 r = -ENOMEM;
 198                 goto fail_free_run;
 199         }
 200         vcpu->pio_data = page_address(page);
 201
 202         r = kvm_mmu_create(vcpu);
 203         if (r < 0)
 204                 goto fail_free_pio_data;
 205
 206         if (irqchip_in_kernel(kvm)) {
 207                 r = kvm_create_lapic(vcpu);
 208                 if (r < 0)
 209                         goto fail_mmu_destroy;
 210         }
 211
 212         return 0;
 213
 214 fail_mmu_destroy:
 215         kvm_mmu_destroy(vcpu);
 216 fail_free_pio_data:
 217         free_page((unsigned long)vcpu->pio_data);
 218 fail_free_run:
 219         free_page((unsigned long)vcpu->run);
 220 fail:
 221         return r;
 222 }
 223 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 224
 225 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 226 {
 227         kvm_free_lapic(vcpu);
 228         kvm_mmu_destroy(vcpu);
 229         free_page((unsigned long)vcpu->pio_data);
 230         free_page((unsigned long)vcpu->run);
 231 }
 232 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 233
 234 static struct kvm *kvm_create_vm(void)
 235 {
 236         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 237
 238         if (!kvm)
 239                 return ERR_PTR(-ENOMEM);
 240
 241         kvm_io_bus_init(&kvm->pio_bus);
 242         mutex_init(&kvm->lock);
 243         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 244         kvm_io_bus_init(&kvm->mmio_bus);
 245         spin_lock(&kvm_lock);
 246         list_add(&kvm->vm_list, &vm_list);
 247         spin_unlock(&kvm_lock);
 248         return kvm;
 249 }
 250
 251 /*
 252  * Free any memory in @free but not in @dont.
 253  */
 254 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 255                                   struct kvm_memory_slot *dont)
 256 {
 257         if (!dont || free->rmap != dont->rmap)
 258                 vfree(free->rmap);
 259
 260         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 261                 vfree(free->dirty_bitmap);
 262
 263         free->npages = 0;
 264         free->dirty_bitmap = NULL;
 265         free->rmap = NULL;
 266 }
 267
 268 static void kvm_free_physmem(struct kvm *kvm)
 269 {
 270         int i;
 271
 272         for (i = 0; i < kvm->nmemslots; ++i)
 273                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 274 }
 275
 276 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 277 {
 278         int i;
 279
 280         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
 281                 if (vcpu->pio.guest_pages[i]) {
 282                         kvm_release_page(vcpu->pio.guest_pages[i]);
 283                         vcpu->pio.guest_pages[i] = NULL;
 284                 }
 285 }
 286
 287 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 288 {
 289         vcpu_load(vcpu);
 290         kvm_mmu_unload(vcpu);
 291         vcpu_put(vcpu);
 292 }
 293
 294 static void kvm_free_vcpus(struct kvm *kvm)
 295 {
 296         unsigned int i;
 297
 298         /*
 299          * Unpin any mmu pages first.
 300          */
 301         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 302                 if (kvm->vcpus[i])
 303                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
 304         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 305                 if (kvm->vcpus[i]) {
 306                         kvm_x86_ops->vcpu_free(kvm->vcpus[i]);
 307                         kvm->vcpus[i] = NULL;
 308                 }
 309         }
 310
 311 }
 312
 313 static void kvm_destroy_vm(struct kvm *kvm)
 314 {
 315         spin_lock(&kvm_lock);
 316         list_del(&kvm->vm_list);
 317         spin_unlock(&kvm_lock);
 318         kvm_io_bus_destroy(&kvm->pio_bus);
 319         kvm_io_bus_destroy(&kvm->mmio_bus);
 320         kfree(kvm->vpic);
 321         kfree(kvm->vioapic);
 322         kvm_free_vcpus(kvm);
 323         kvm_free_physmem(kvm);
 324         kfree(kvm);
 325 }
 326
 327 static int kvm_vm_release(struct inode *inode, struct file *filp)
 328 {
 329         struct kvm *kvm = filp->private_data;
 330
 331         kvm_destroy_vm(kvm);
 332         return 0;
 333 }
 334
 335 static void inject_gp(struct kvm_vcpu *vcpu)
 336 {
 337         kvm_x86_ops->inject_gp(vcpu, 0);
 338 }
 339
 340 void fx_init(struct kvm_vcpu *vcpu)
 341 {
 342         unsigned after_mxcsr_mask;
 343
 344         /* Initialize guest FPU by resetting ours and saving into guest's */
 345         preempt_disable();
 346         fx_save(&vcpu->host_fx_image);
 347         fpu_init();
 348         fx_save(&vcpu->guest_fx_image);
 349         fx_restore(&vcpu->host_fx_image);
 350         preempt_enable();
 351
 352         vcpu->cr0 |= X86_CR0_ET;
 353         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
 354         vcpu->guest_fx_image.mxcsr = 0x1f80;
 355         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
 356                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
 357 }
 358 EXPORT_SYMBOL_GPL(fx_init);
 359
 360 /*
 361  * Allocate some memory and give it an address in the guest physical address
 362  * space.
 363  *
 364  * Discontiguous memory is allowed, mostly for framebuffers.
 365  *
 366  * Must be called holding kvm->lock.
 367  */
 368 int __kvm_set_memory_region(struct kvm *kvm,
 369                             struct kvm_userspace_memory_region *mem,
 370                             int user_alloc)
 371 {
 372         int r;
 373         gfn_t base_gfn;
 374         unsigned long npages;
 375         unsigned long i;
 376         struct kvm_memory_slot *memslot;
 377         struct kvm_memory_slot old, new;
 378
 379         r = -EINVAL;
 380         /* General sanity checks */
 381         if (mem->memory_size & (PAGE_SIZE - 1))
 382                 goto out;
 383         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 384                 goto out;
 385         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 386                 goto out;
 387         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 388                 goto out;
 389
 390         memslot = &kvm->memslots[mem->slot];
 391         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 392         npages = mem->memory_size >> PAGE_SHIFT;
 393
 394         if (!npages)
 395                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 396
 397         new = old = *memslot;
 398
 399         new.base_gfn = base_gfn;
 400         new.npages = npages;
 401         new.flags = mem->flags;
 402
 403         /* Disallow changing a memory slot's size. */
 404         r = -EINVAL;
 405         if (npages && old.npages && npages != old.npages)
 406                 goto out_free;
 407
 408         /* Check for overlaps */
 409         r = -EEXIST;
 410         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 411                 struct kvm_memory_slot *s = &kvm->memslots[i];
 412
 413                 if (s == memslot)
 414                         continue;
 415                 if (!((base_gfn + npages <= s->base_gfn) ||
 416                       (base_gfn >= s->base_gfn + s->npages)))
 417                         goto out_free;
 418         }
 419
 420         /* Free page dirty bitmap if unneeded */
 421         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 422                 new.dirty_bitmap = NULL;
 423
 424         r = -ENOMEM;
 425
 426         /* Allocate if a slot is being created */
 427         if (npages && !new.rmap) {
 428                 new.rmap = vmalloc(npages * sizeof(struct page *));
 429
 430                 if (!new.rmap)
 431                         goto out_free;
 432
 433                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 434
 435                 new.user_alloc = user_alloc;
 436                 if (user_alloc)
 437                         new.userspace_addr = mem->userspace_addr;
 438                 else {
 439                         down_write(&current->mm->mmap_sem);
 440                         new.userspace_addr = do_mmap(NULL, 0,
 441                                                      npages * PAGE_SIZE,
 442                                                      PROT_READ | PROT_WRITE,
 443                                                      MAP_SHARED | MAP_ANONYMOUS,
 444                                                      0);
 445                         up_write(&current->mm->mmap_sem);
 446
 447                         if (IS_ERR((void *)new.userspace_addr))
 448                                 goto out_free;
 449                 }
 450         } else {
 451                 if (!old.user_alloc && old.rmap) {
 452                         int ret;
 453
 454                         down_write(&current->mm->mmap_sem);
 455                         ret = do_munmap(current->mm, old.userspace_addr,
 456                                         old.npages * PAGE_SIZE);
 457                         up_write(&current->mm->mmap_sem);
 458                         if (ret < 0)
 459                                 printk(KERN_WARNING
 460                                        "kvm_vm_ioctl_set_memory_region: "
 461                                        "failed to munmap memory\n");
 462                 }
 463         }
 464
 465         /* Allocate page dirty bitmap if needed */
 466         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 467                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 468
 469                 new.dirty_bitmap = vmalloc(dirty_bytes);
 470                 if (!new.dirty_bitmap)
 471                         goto out_free;
 472                 memset(new.dirty_bitmap, 0, dirty_bytes);
 473         }
 474
 475         if (mem->slot >= kvm->nmemslots)
 476                 kvm->nmemslots = mem->slot + 1;
 477
 478         if (!kvm->n_requested_mmu_pages) {
 479                 unsigned int n_pages;
 480
 481                 if (npages) {
 482                         n_pages = npages * KVM_PERMILLE_MMU_PAGES / 1000;
 483                         kvm_mmu_change_mmu_pages(kvm, kvm->n_alloc_mmu_pages +
 484                                                  n_pages);
 485                 } else {
 486                         unsigned int nr_mmu_pages;
 487
 488                         n_pages = old.npages * KVM_PERMILLE_MMU_PAGES / 1000;
 489                         nr_mmu_pages = kvm->n_alloc_mmu_pages - n_pages;
 490                         nr_mmu_pages = max(nr_mmu_pages,
 491                                         (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
 492                         kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
 493                 }
 494         }
 495
 496         *memslot = new;
 497
 498         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 499         kvm_flush_remote_tlbs(kvm);
 500
 501         kvm_free_physmem_slot(&old, &new);
 502         return 0;
 503
 504 out_free:
 505         kvm_free_physmem_slot(&new, &old);
 506 out:
 507         return r;
 508
 509 }
 510 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 511
 512 int kvm_set_memory_region(struct kvm *kvm,
 513                           struct kvm_userspace_memory_region *mem,
 514                           int user_alloc)
 515 {
 516         int r;
 517
 518         mutex_lock(&kvm->lock);
 519         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 520         mutex_unlock(&kvm->lock);
 521         return r;
 522 }
 523 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 524
 525 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 526                                    struct
 527                                    kvm_userspace_memory_region *mem,
 528                                    int user_alloc)
 529 {
 530         if (mem->slot >= KVM_MEMORY_SLOTS)
 531                 return -EINVAL;
 532         return kvm_set_memory_region(kvm, mem, user_alloc);
 533 }
 534
 535 /*
 536  * Get (and clear) the dirty memory log for a memory slot.
 537  */
 538 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 539                                       struct kvm_dirty_log *log)
 540 {
 541         struct kvm_memory_slot *memslot;
 542         int r, i;
 543         int n;
 544         unsigned long any = 0;
 545
 546         mutex_lock(&kvm->lock);
 547
 548         r = -EINVAL;
 549         if (log->slot >= KVM_MEMORY_SLOTS)
 550                 goto out;
 551
 552         memslot = &kvm->memslots[log->slot];
 553         r = -ENOENT;
 554         if (!memslot->dirty_bitmap)
 555                 goto out;
 556
 557         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 558
 559         for (i = 0; !any && i < n/sizeof(long); ++i)
 560                 any = memslot->dirty_bitmap[i];
 561
 562         r = -EFAULT;
 563         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 564                 goto out;
 565
 566         /* If nothing is dirty, don't bother messing with page tables. */
 567         if (any) {
 568                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 569                 kvm_flush_remote_tlbs(kvm);
 570                 memset(memslot->dirty_bitmap, 0, n);
 571         }
 572
 573         r = 0;
 574
 575 out:
 576         mutex_unlock(&kvm->lock);
 577         return r;
 578 }
 579
 580 int is_error_page(struct page *page)
 581 {
 582         return page == bad_page;
 583 }
 584 EXPORT_SYMBOL_GPL(is_error_page);
 585
 586 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 587 {
 588         int i;
 589         struct kvm_mem_alias *alias;
 590
 591         for (i = 0; i < kvm->naliases; ++i) {
 592                 alias = &kvm->aliases[i];
 593                 if (gfn >= alias->base_gfn
 594                     && gfn < alias->base_gfn + alias->npages)
 595                         return alias->target_gfn + gfn - alias->base_gfn;
 596         }
 597         return gfn;
 598 }
 599
 600 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 601 {
 602         int i;
 603
 604         for (i = 0; i < kvm->nmemslots; ++i) {
 605                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 606
 607                 if (gfn >= memslot->base_gfn
 608                     && gfn < memslot->base_gfn + memslot->npages)
 609                         return memslot;
 610         }
 611         return NULL;
 612 }
 613
 614 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 615 {
 616         gfn = unalias_gfn(kvm, gfn);
 617         return __gfn_to_memslot(kvm, gfn);
 618 }
 619
 620 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 621 {
 622         int i;
 623
 624         gfn = unalias_gfn(kvm, gfn);
 625         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 626                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 627
 628                 if (gfn >= memslot->base_gfn
 629                     && gfn < memslot->base_gfn + memslot->npages)
 630                         return 1;
 631         }
 632         return 0;
 633 }
 634 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 635
 636 /*
 637  * Requires current->mm->mmap_sem to be held
 638  */
 639 static struct page *__gfn_to_page(struct kvm *kvm, gfn_t gfn)
 640 {
 641         struct kvm_memory_slot *slot;
 642         struct page *page[1];
 643         int npages;
 644
 645         might_sleep();
 646
 647         gfn = unalias_gfn(kvm, gfn);
 648         slot = __gfn_to_memslot(kvm, gfn);
 649         if (!slot) {
 650                 get_page(bad_page);
 651                 return bad_page;
 652         }
 653
 654         npages = get_user_pages(current, current->mm,
 655                                 slot->userspace_addr
 656                                 + (gfn - slot->base_gfn) * PAGE_SIZE, 1,
 657                                 1, 1, page, NULL);
 658         if (npages != 1) {
 659                 get_page(bad_page);
 660                 return bad_page;
 661         }
 662
 663         return page[0];
 664 }
 665
 666 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 667 {
 668         struct page *page;
 669
 670         down_read(&current->mm->mmap_sem);
 671         page = __gfn_to_page(kvm, gfn);
 672         up_read(&current->mm->mmap_sem);
 673
 674         return page;
 675 }
 676
 677 EXPORT_SYMBOL_GPL(gfn_to_page);
 678
 679 void kvm_release_page(struct page *page)
 680 {
 681         if (!PageReserved(page))
 682                 SetPageDirty(page);
 683         put_page(page);
 684 }
 685 EXPORT_SYMBOL_GPL(kvm_release_page);
 686
 687 static int next_segment(unsigned long len, int offset)
 688 {
 689         if (len > PAGE_SIZE - offset)
 690                 return PAGE_SIZE - offset;
 691         else
 692                 return len;
 693 }
 694
 695 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 696                         int len)
 697 {
 698         void *page_virt;
 699         struct page *page;
 700
 701         page = gfn_to_page(kvm, gfn);
 702         if (is_error_page(page)) {
 703                 kvm_release_page(page);
 704                 return -EFAULT;
 705         }
 706         page_virt = kmap_atomic(page, KM_USER0);
 707
 708         memcpy(data, page_virt + offset, len);
 709
 710         kunmap_atomic(page_virt, KM_USER0);
 711         kvm_release_page(page);
 712         return 0;
 713 }
 714 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 715
 716 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 717 {
 718         gfn_t gfn = gpa >> PAGE_SHIFT;
 719         int seg;
 720         int offset = offset_in_page(gpa);
 721         int ret;
 722
 723         while ((seg = next_segment(len, offset)) != 0) {
 724                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 725                 if (ret < 0)
 726                         return ret;
 727                 offset = 0;
 728                 len -= seg;
 729                 data += seg;
 730                 ++gfn;
 731         }
 732         return 0;
 733 }
 734 EXPORT_SYMBOL_GPL(kvm_read_guest);
 735
 736 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
 737                          int offset, int len)
 738 {
 739         void *page_virt;
 740         struct page *page;
 741
 742         page = gfn_to_page(kvm, gfn);
 743         if (is_error_page(page)) {
 744                 kvm_release_page(page);
 745                 return -EFAULT;
 746         }
 747         page_virt = kmap_atomic(page, KM_USER0);
 748
 749         memcpy(page_virt + offset, data, len);
 750
 751         kunmap_atomic(page_virt, KM_USER0);
 752         mark_page_dirty(kvm, gfn);
 753         kvm_release_page(page);
 754         return 0;
 755 }
 756 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
 757
 758 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
 759                     unsigned long len)
 760 {
 761         gfn_t gfn = gpa >> PAGE_SHIFT;
 762         int seg;
 763         int offset = offset_in_page(gpa);
 764         int ret;
 765
 766         while ((seg = next_segment(len, offset)) != 0) {
 767                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
 768                 if (ret < 0)
 769                         return ret;
 770                 offset = 0;
 771                 len -= seg;
 772                 data += seg;
 773                 ++gfn;
 774         }
 775         return 0;
 776 }
 777
 778 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
 779 {
 780         void *page_virt;
 781         struct page *page;
 782
 783         page = gfn_to_page(kvm, gfn);
 784         if (is_error_page(page)) {
 785                 kvm_release_page(page);
 786                 return -EFAULT;
 787         }
 788         page_virt = kmap_atomic(page, KM_USER0);
 789
 790         memset(page_virt + offset, 0, len);
 791
 792         kunmap_atomic(page_virt, KM_USER0);
 793         kvm_release_page(page);
 794         return 0;
 795 }
 796 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
 797
 798 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
 799 {
 800         gfn_t gfn = gpa >> PAGE_SHIFT;
 801         int seg;
 802         int offset = offset_in_page(gpa);
 803         int ret;
 804
 805         while ((seg = next_segment(len, offset)) != 0) {
 806                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
 807                 if (ret < 0)
 808                         return ret;
 809                 offset = 0;
 810                 len -= seg;
 811                 ++gfn;
 812         }
 813         return 0;
 814 }
 815 EXPORT_SYMBOL_GPL(kvm_clear_guest);
 816
 817 /* WARNING: Does not work on aliased pages. */
 818 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 819 {
 820         struct kvm_memory_slot *memslot;
 821
 822         memslot = __gfn_to_memslot(kvm, gfn);
 823         if (memslot && memslot->dirty_bitmap) {
 824                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 825
 826                 /* avoid RMW */
 827                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 828                         set_bit(rel_gfn, memslot->dirty_bitmap);
 829         }
 830 }
 831
 832 int emulator_read_std(unsigned long addr,
 833                              void *val,
 834                              unsigned int bytes,
 835                              struct kvm_vcpu *vcpu)
 836 {
 837         void *data = val;
 838
 839         while (bytes) {
 840                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 841                 unsigned offset = addr & (PAGE_SIZE-1);
 842                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 843                 int ret;
 844
 845                 if (gpa == UNMAPPED_GVA)
 846                         return X86EMUL_PROPAGATE_FAULT;
 847                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
 848                 if (ret < 0)
 849                         return X86EMUL_UNHANDLEABLE;
 850
 851                 bytes -= tocopy;
 852                 data += tocopy;
 853                 addr += tocopy;
 854         }
 855
 856         return X86EMUL_CONTINUE;
 857 }
 858 EXPORT_SYMBOL_GPL(emulator_read_std);
 859
 860 static int emulator_write_std(unsigned long addr,
 861                               const void *val,
 862                               unsigned int bytes,
 863                               struct kvm_vcpu *vcpu)
 864 {
 865         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
 866         return X86EMUL_UNHANDLEABLE;
 867 }
 868
 869 /*
 870  * Only apic need an MMIO device hook, so shortcut now..
 871  */
 872 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
 873                                                 gpa_t addr)
 874 {
 875         struct kvm_io_device *dev;
 876
 877         if (vcpu->apic) {
 878                 dev = &vcpu->apic->dev;
 879                 if (dev->in_range(dev, addr))
 880                         return dev;
 881         }
 882         return NULL;
 883 }
 884
 885 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
 886                                                 gpa_t addr)
 887 {
 888         struct kvm_io_device *dev;
 889
 890         dev = vcpu_find_pervcpu_dev(vcpu, addr);
 891         if (dev == NULL)
 892                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
 893         return dev;
 894 }
 895
 896 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
 897                                                gpa_t addr)
 898 {
 899         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
 900 }
 901
 902 static int emulator_read_emulated(unsigned long addr,
 903                                   void *val,
 904                                   unsigned int bytes,
 905                                   struct kvm_vcpu *vcpu)
 906 {
 907         struct kvm_io_device *mmio_dev;
 908         gpa_t                 gpa;
 909
 910         if (vcpu->mmio_read_completed) {
 911                 memcpy(val, vcpu->mmio_data, bytes);
 912                 vcpu->mmio_read_completed = 0;
 913                 return X86EMUL_CONTINUE;
 914         }
 915
 916         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 917
 918         /* For APIC access vmexit */
 919         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
 920                 goto mmio;
 921
 922         if (emulator_read_std(addr, val, bytes, vcpu)
 923                         == X86EMUL_CONTINUE)
 924                 return X86EMUL_CONTINUE;
 925         if (gpa == UNMAPPED_GVA)
 926                 return X86EMUL_PROPAGATE_FAULT;
 927
 928 mmio:
 929         /*
 930          * Is this MMIO handled locally?
 931          */
 932         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
 933         if (mmio_dev) {
 934                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
 935                 return X86EMUL_CONTINUE;
 936         }
 937
 938         vcpu->mmio_needed = 1;
 939         vcpu->mmio_phys_addr = gpa;
 940         vcpu->mmio_size = bytes;
 941         vcpu->mmio_is_write = 0;
 942
 943         return X86EMUL_UNHANDLEABLE;
 944 }
 945
 946 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
 947                                const void *val, int bytes)
 948 {
 949         int ret;
 950
 951         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
 952         if (ret < 0)
 953                 return 0;
 954         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
 955         return 1;
 956 }
 957
 958 static int emulator_write_emulated_onepage(unsigned long addr,
 959                                            const void *val,
 960                                            unsigned int bytes,
 961                                            struct kvm_vcpu *vcpu)
 962 {
 963         struct kvm_io_device *mmio_dev;
 964         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 965
 966         if (gpa == UNMAPPED_GVA) {
 967                 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
 968                 return X86EMUL_PROPAGATE_FAULT;
 969         }
 970
 971         /* For APIC access vmexit */
 972         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
 973                 goto mmio;
 974
 975         if (emulator_write_phys(vcpu, gpa, val, bytes))
 976                 return X86EMUL_CONTINUE;
 977
 978 mmio:
 979         /*
 980          * Is this MMIO handled locally?
 981          */
 982         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
 983         if (mmio_dev) {
 984                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
 985                 return X86EMUL_CONTINUE;
 986         }
 987
 988         vcpu->mmio_needed = 1;
 989         vcpu->mmio_phys_addr = gpa;
 990         vcpu->mmio_size = bytes;
 991         vcpu->mmio_is_write = 1;
 992         memcpy(vcpu->mmio_data, val, bytes);
 993
 994         return X86EMUL_CONTINUE;
 995 }
 996
 997 int emulator_write_emulated(unsigned long addr,
 998                                    const void *val,
 999                                    unsigned int bytes,
1000                                    struct kvm_vcpu *vcpu)
1001 {
1002         /* Crossing a page boundary? */
1003         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1004                 int rc, now;
1005
1006                 now = -addr & ~PAGE_MASK;
1007                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1008                 if (rc != X86EMUL_CONTINUE)
1009                         return rc;
1010                 addr += now;
1011                 val += now;
1012                 bytes -= now;
1013         }
1014         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1015 }
1016 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1017
1018 static int emulator_cmpxchg_emulated(unsigned long addr,
1019                                      const void *old,
1020                                      const void *new,
1021                                      unsigned int bytes,
1022                                      struct kvm_vcpu *vcpu)
1023 {
1024         static int reported;
1025
1026         if (!reported) {
1027                 reported = 1;
1028                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1029         }
1030         return emulator_write_emulated(addr, new, bytes, vcpu);
1031 }
1032
1033 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1034 {
1035         return kvm_x86_ops->get_segment_base(vcpu, seg);
1036 }
1037
1038 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1039 {
1040         return X86EMUL_CONTINUE;
1041 }
1042
1043 int emulate_clts(struct kvm_vcpu *vcpu)
1044 {
1045         kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1046         return X86EMUL_CONTINUE;
1047 }
1048
1049 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1050 {
1051         struct kvm_vcpu *vcpu = ctxt->vcpu;
1052
1053         switch (dr) {
1054         case 0 ... 3:
1055                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1056                 return X86EMUL_CONTINUE;
1057         default:
1058                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1059                 return X86EMUL_UNHANDLEABLE;
1060         }
1061 }
1062
1063 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1064 {
1065         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1066         int exception;
1067
1068         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1069         if (exception) {
1070                 /* FIXME: better handling */
1071                 return X86EMUL_UNHANDLEABLE;
1072         }
1073         return X86EMUL_CONTINUE;
1074 }
1075
1076 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1077 {
1078         static int reported;
1079         u8 opcodes[4];
1080         unsigned long rip = vcpu->rip;
1081         unsigned long rip_linear;
1082
1083         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1084
1085         if (reported)
1086                 return;
1087
1088         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1089
1090         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1091                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1092         reported = 1;
1093 }
1094 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1095
1096 struct x86_emulate_ops emulate_ops = {
1097         .read_std            = emulator_read_std,
1098         .write_std           = emulator_write_std,
1099         .read_emulated       = emulator_read_emulated,
1100         .write_emulated      = emulator_write_emulated,
1101         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1102 };
1103
1104 int emulate_instruction(struct kvm_vcpu *vcpu,
1105                         struct kvm_run *run,
1106                         unsigned long cr2,
1107                         u16 error_code,
1108                         int no_decode)
1109 {
1110         int r;
1111
1112         vcpu->mmio_fault_cr2 = cr2;
1113         kvm_x86_ops->cache_regs(vcpu);
1114
1115         vcpu->mmio_is_write = 0;
1116         vcpu->pio.string = 0;
1117
1118         if (!no_decode) {
1119                 int cs_db, cs_l;
1120                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1121
1122                 vcpu->emulate_ctxt.vcpu = vcpu;
1123                 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1124                 vcpu->emulate_ctxt.cr2 = cr2;
1125                 vcpu->emulate_ctxt.mode =
1126                         (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1127                         ? X86EMUL_MODE_REAL : cs_l
1128                         ? X86EMUL_MODE_PROT64 : cs_db
1129                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1130
1131                 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1132                         vcpu->emulate_ctxt.cs_base = 0;
1133                         vcpu->emulate_ctxt.ds_base = 0;
1134                         vcpu->emulate_ctxt.es_base = 0;
1135                         vcpu->emulate_ctxt.ss_base = 0;
1136                 } else {
1137                         vcpu->emulate_ctxt.cs_base =
1138                                         get_segment_base(vcpu, VCPU_SREG_CS);
1139                         vcpu->emulate_ctxt.ds_base =
1140                                         get_segment_base(vcpu, VCPU_SREG_DS);
1141                         vcpu->emulate_ctxt.es_base =
1142                                         get_segment_base(vcpu, VCPU_SREG_ES);
1143                         vcpu->emulate_ctxt.ss_base =
1144                                         get_segment_base(vcpu, VCPU_SREG_SS);
1145                 }
1146
1147                 vcpu->emulate_ctxt.gs_base =
1148                                         get_segment_base(vcpu, VCPU_SREG_GS);
1149                 vcpu->emulate_ctxt.fs_base =
1150                                         get_segment_base(vcpu, VCPU_SREG_FS);
1151
1152                 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1153                 if (r)  {
1154                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1155                                 return EMULATE_DONE;
1156                         return EMULATE_FAIL;
1157                 }
1158         }
1159
1160         r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1161
1162         if (vcpu->pio.string)
1163                 return EMULATE_DO_MMIO;
1164
1165         if ((r || vcpu->mmio_is_write) && run) {
1166                 run->exit_reason = KVM_EXIT_MMIO;
1167                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1168                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1169                 run->mmio.len = vcpu->mmio_size;
1170                 run->mmio.is_write = vcpu->mmio_is_write;
1171         }
1172
1173         if (r) {
1174                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1175                         return EMULATE_DONE;
1176                 if (!vcpu->mmio_needed) {
1177                         kvm_report_emulation_failure(vcpu, "mmio");
1178                         return EMULATE_FAIL;
1179                 }
1180                 return EMULATE_DO_MMIO;
1181         }
1182
1183         kvm_x86_ops->decache_regs(vcpu);
1184         kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1185
1186         if (vcpu->mmio_is_write) {
1187                 vcpu->mmio_needed = 0;
1188                 return EMULATE_DO_MMIO;
1189         }
1190
1191         return EMULATE_DONE;
1192 }
1193 EXPORT_SYMBOL_GPL(emulate_instruction);
1194
1195 /*
1196  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1197  */
1198 static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1199 {
1200         DECLARE_WAITQUEUE(wait, current);
1201
1202         add_wait_queue(&vcpu->wq, &wait);
1203
1204         /*
1205          * We will block until either an interrupt or a signal wakes us up
1206          */
1207         while (!kvm_cpu_has_interrupt(vcpu)
1208                && !signal_pending(current)
1209                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1210                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
1211                 set_current_state(TASK_INTERRUPTIBLE);
1212                 vcpu_put(vcpu);
1213                 schedule();
1214                 vcpu_load(vcpu);
1215         }
1216
1217         __set_current_state(TASK_RUNNING);
1218         remove_wait_queue(&vcpu->wq, &wait);
1219 }
1220
1221 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1222 {
1223         ++vcpu->stat.halt_exits;
1224         if (irqchip_in_kernel(vcpu->kvm)) {
1225                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1226                 kvm_vcpu_block(vcpu);
1227                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1228                         return -EINTR;
1229                 return 1;
1230         } else {
1231                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1232                 return 0;
1233         }
1234 }
1235 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1236
1237 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1238 {
1239         unsigned long nr, a0, a1, a2, a3, ret;
1240
1241         kvm_x86_ops->cache_regs(vcpu);
1242
1243         nr = vcpu->regs[VCPU_REGS_RAX];
1244         a0 = vcpu->regs[VCPU_REGS_RBX];
1245         a1 = vcpu->regs[VCPU_REGS_RCX];
1246         a2 = vcpu->regs[VCPU_REGS_RDX];
1247         a3 = vcpu->regs[VCPU_REGS_RSI];
1248
1249         if (!is_long_mode(vcpu)) {
1250                 nr &= 0xFFFFFFFF;
1251                 a0 &= 0xFFFFFFFF;
1252                 a1 &= 0xFFFFFFFF;
1253                 a2 &= 0xFFFFFFFF;
1254                 a3 &= 0xFFFFFFFF;
1255         }
1256
1257         switch (nr) {
1258         default:
1259                 ret = -KVM_ENOSYS;
1260                 break;
1261         }
1262         vcpu->regs[VCPU_REGS_RAX] = ret;
1263         kvm_x86_ops->decache_regs(vcpu);
1264         return 0;
1265 }
1266 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1267
1268 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1269 {
1270         char instruction[3];
1271         int ret = 0;
1272
1273         mutex_lock(&vcpu->kvm->lock);
1274
1275         /*
1276          * Blow out the MMU to ensure that no other VCPU has an active mapping
1277          * to ensure that the updated hypercall appears atomically across all
1278          * VCPUs.
1279          */
1280         kvm_mmu_zap_all(vcpu->kvm);
1281
1282         kvm_x86_ops->cache_regs(vcpu);
1283         kvm_x86_ops->patch_hypercall(vcpu, instruction);
1284         if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1285             != X86EMUL_CONTINUE)
1286                 ret = -EFAULT;
1287
1288         mutex_unlock(&vcpu->kvm->lock);
1289
1290         return ret;
1291 }
1292
1293 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1294 {
1295         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1296 }
1297
1298 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1299 {
1300         struct descriptor_table dt = { limit, base };
1301
1302         kvm_x86_ops->set_gdt(vcpu, &dt);
1303 }
1304
1305 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1306 {
1307         struct descriptor_table dt = { limit, base };
1308
1309         kvm_x86_ops->set_idt(vcpu, &dt);
1310 }
1311
1312 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1313                    unsigned long *rflags)
1314 {
1315         lmsw(vcpu, msw);
1316         *rflags = kvm_x86_ops->get_rflags(vcpu);
1317 }
1318
1319 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1320 {
1321         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1322         switch (cr) {
1323         case 0:
1324                 return vcpu->cr0;
1325         case 2:
1326                 return vcpu->cr2;
1327         case 3:
1328                 return vcpu->cr3;
1329         case 4:
1330                 return vcpu->cr4;
1331         default:
1332                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1333                 return 0;
1334         }
1335 }
1336
1337 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1338                      unsigned long *rflags)
1339 {
1340         switch (cr) {
1341         case 0:
1342                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1343                 *rflags = kvm_x86_ops->get_rflags(vcpu);
1344                 break;
1345         case 2:
1346                 vcpu->cr2 = val;
1347                 break;
1348         case 3:
1349                 set_cr3(vcpu, val);
1350                 break;
1351         case 4:
1352                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1353                 break;
1354         default:
1355                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1356         }
1357 }
1358
1359 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1360 {
1361         u64 data;
1362
1363         switch (msr) {
1364         case 0xc0010010: /* SYSCFG */
1365         case 0xc0010015: /* HWCR */
1366         case MSR_IA32_PLATFORM_ID:
1367         case MSR_IA32_P5_MC_ADDR:
1368         case MSR_IA32_P5_MC_TYPE:
1369         case MSR_IA32_MC0_CTL:
1370         case MSR_IA32_MCG_STATUS:
1371         case MSR_IA32_MCG_CAP:
1372         case MSR_IA32_MC0_MISC:
1373         case MSR_IA32_MC0_MISC+4:
1374         case MSR_IA32_MC0_MISC+8:
1375         case MSR_IA32_MC0_MISC+12:
1376         case MSR_IA32_MC0_MISC+16:
1377         case MSR_IA32_UCODE_REV:
1378         case MSR_IA32_PERF_STATUS:
1379         case MSR_IA32_EBL_CR_POWERON:
1380                 /* MTRR registers */
1381         case 0xfe:
1382         case 0x200 ... 0x2ff:
1383                 data = 0;
1384                 break;
1385         case 0xcd: /* fsb frequency */
1386                 data = 3;
1387                 break;
1388         case MSR_IA32_APICBASE:
1389                 data = kvm_get_apic_base(vcpu);
1390                 break;
1391         case MSR_IA32_MISC_ENABLE:
1392                 data = vcpu->ia32_misc_enable_msr;
1393                 break;
1394 #ifdef CONFIG_X86_64
1395         case MSR_EFER:
1396                 data = vcpu->shadow_efer;
1397                 break;
1398 #endif
1399         default:
1400                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1401                 return 1;
1402         }
1403         *pdata = data;
1404         return 0;
1405 }
1406 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1407
1408 /*
1409  * Reads an msr value (of 'msr_index') into 'pdata'.
1410  * Returns 0 on success, non-0 otherwise.
1411  * Assumes vcpu_load() was already called.
1412  */
1413 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1414 {
1415         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1416 }
1417
1418 #ifdef CONFIG_X86_64
1419
1420 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1421 {
1422         if (efer & EFER_RESERVED_BITS) {
1423                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1424                        efer);
1425                 inject_gp(vcpu);
1426                 return;
1427         }
1428
1429         if (is_paging(vcpu)
1430             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1431                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1432                 inject_gp(vcpu);
1433                 return;
1434         }
1435
1436         kvm_x86_ops->set_efer(vcpu, efer);
1437
1438         efer &= ~EFER_LMA;
1439         efer |= vcpu->shadow_efer & EFER_LMA;
1440
1441         vcpu->shadow_efer = efer;
1442 }
1443
1444 #endif
1445
1446 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1447 {
1448         switch (msr) {
1449 #ifdef CONFIG_X86_64
1450         case MSR_EFER:
1451                 set_efer(vcpu, data);
1452                 break;
1453 #endif
1454         case MSR_IA32_MC0_STATUS:
1455                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1456                        __FUNCTION__, data);
1457                 break;
1458         case MSR_IA32_MCG_STATUS:
1459                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1460                         __FUNCTION__, data);
1461                 break;
1462         case MSR_IA32_UCODE_REV:
1463         case MSR_IA32_UCODE_WRITE:
1464         case 0x200 ... 0x2ff: /* MTRRs */
1465                 break;
1466         case MSR_IA32_APICBASE:
1467                 kvm_set_apic_base(vcpu, data);
1468                 break;
1469         case MSR_IA32_MISC_ENABLE:
1470                 vcpu->ia32_misc_enable_msr = data;
1471                 break;
1472         default:
1473                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1474                 return 1;
1475         }
1476         return 0;
1477 }
1478 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1479
1480 /*
1481  * Writes msr value into into the appropriate "register".
1482  * Returns 0 on success, non-0 otherwise.
1483  * Assumes vcpu_load() was already called.
1484  */
1485 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1486 {
1487         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
1488 }
1489
1490 void kvm_resched(struct kvm_vcpu *vcpu)
1491 {
1492         if (!need_resched())
1493                 return;
1494         cond_resched();
1495 }
1496 EXPORT_SYMBOL_GPL(kvm_resched);
1497
1498 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1499 {
1500         int i;
1501         u32 function;
1502         struct kvm_cpuid_entry *e, *best;
1503
1504         kvm_x86_ops->cache_regs(vcpu);
1505         function = vcpu->regs[VCPU_REGS_RAX];
1506         vcpu->regs[VCPU_REGS_RAX] = 0;
1507         vcpu->regs[VCPU_REGS_RBX] = 0;
1508         vcpu->regs[VCPU_REGS_RCX] = 0;
1509         vcpu->regs[VCPU_REGS_RDX] = 0;
1510         best = NULL;
1511         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1512                 e = &vcpu->cpuid_entries[i];
1513                 if (e->function == function) {
1514                         best = e;
1515                         break;
1516                 }
1517                 /*
1518                  * Both basic or both extended?
1519                  */
1520                 if (((e->function ^ function) & 0x80000000) == 0)
1521                         if (!best || e->function > best->function)
1522                                 best = e;
1523         }
1524         if (best) {
1525                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1526                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1527                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1528                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1529         }
1530         kvm_x86_ops->decache_regs(vcpu);
1531         kvm_x86_ops->skip_emulated_instruction(vcpu);
1532 }
1533 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1534
1535 static int pio_copy_data(struct kvm_vcpu *vcpu)
1536 {
1537         void *p = vcpu->pio_data;
1538         void *q;
1539         unsigned bytes;
1540         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1541
1542         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1543                  PAGE_KERNEL);
1544         if (!q) {
1545                 free_pio_guest_pages(vcpu);
1546                 return -ENOMEM;
1547         }
1548         q += vcpu->pio.guest_page_offset;
1549         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1550         if (vcpu->pio.in)
1551                 memcpy(q, p, bytes);
1552         else
1553                 memcpy(p, q, bytes);
1554         q -= vcpu->pio.guest_page_offset;
1555         vunmap(q);
1556         free_pio_guest_pages(vcpu);
1557         return 0;
1558 }
1559
1560 static int complete_pio(struct kvm_vcpu *vcpu)
1561 {
1562         struct kvm_pio_request *io = &vcpu->pio;
1563         long delta;
1564         int r;
1565
1566         kvm_x86_ops->cache_regs(vcpu);
1567
1568         if (!io->string) {
1569                 if (io->in)
1570                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1571                                io->size);
1572         } else {
1573                 if (io->in) {
1574                         r = pio_copy_data(vcpu);
1575                         if (r) {
1576                                 kvm_x86_ops->cache_regs(vcpu);
1577                                 return r;
1578                         }
1579                 }
1580
1581                 delta = 1;
1582                 if (io->rep) {
1583                         delta *= io->cur_count;
1584                         /*
1585                          * The size of the register should really depend on
1586                          * current address size.
1587                          */
1588                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1589                 }
1590                 if (io->down)
1591                         delta = -delta;
1592                 delta *= io->size;
1593                 if (io->in)
1594                         vcpu->regs[VCPU_REGS_RDI] += delta;
1595                 else
1596                         vcpu->regs[VCPU_REGS_RSI] += delta;
1597         }
1598
1599         kvm_x86_ops->decache_regs(vcpu);
1600
1601         io->count -= io->cur_count;
1602         io->cur_count = 0;
1603
1604         return 0;
1605 }
1606
1607 static void kernel_pio(struct kvm_io_device *pio_dev,
1608                        struct kvm_vcpu *vcpu,
1609                        void *pd)
1610 {
1611         /* TODO: String I/O for in kernel device */
1612
1613         mutex_lock(&vcpu->kvm->lock);
1614         if (vcpu->pio.in)
1615                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1616                                   vcpu->pio.size,
1617                                   pd);
1618         else
1619                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1620                                    vcpu->pio.size,
1621                                    pd);
1622         mutex_unlock(&vcpu->kvm->lock);
1623 }
1624
1625 static void pio_string_write(struct kvm_io_device *pio_dev,
1626                              struct kvm_vcpu *vcpu)
1627 {
1628         struct kvm_pio_request *io = &vcpu->pio;
1629         void *pd = vcpu->pio_data;
1630         int i;
1631
1632         mutex_lock(&vcpu->kvm->lock);
1633         for (i = 0; i < io->cur_count; i++) {
1634                 kvm_iodevice_write(pio_dev, io->port,
1635                                    io->size,
1636                                    pd);
1637                 pd += io->size;
1638         }
1639         mutex_unlock(&vcpu->kvm->lock);
1640 }
1641
1642 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1643                   int size, unsigned port)
1644 {
1645         struct kvm_io_device *pio_dev;
1646
1647         vcpu->run->exit_reason = KVM_EXIT_IO;
1648         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1649         vcpu->run->io.size = vcpu->pio.size = size;
1650         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1651         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1652         vcpu->run->io.port = vcpu->pio.port = port;
1653         vcpu->pio.in = in;
1654         vcpu->pio.string = 0;
1655         vcpu->pio.down = 0;
1656         vcpu->pio.guest_page_offset = 0;
1657         vcpu->pio.rep = 0;
1658
1659         kvm_x86_ops->cache_regs(vcpu);
1660         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1661         kvm_x86_ops->decache_regs(vcpu);
1662
1663         kvm_x86_ops->skip_emulated_instruction(vcpu);
1664
1665         pio_dev = vcpu_find_pio_dev(vcpu, port);
1666         if (pio_dev) {
1667                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1668                 complete_pio(vcpu);
1669                 return 1;
1670         }
1671         return 0;
1672 }
1673 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1674
1675 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1676                   int size, unsigned long count, int down,
1677                   gva_t address, int rep, unsigned port)
1678 {
1679         unsigned now, in_page;
1680         int i, ret = 0;
1681         int nr_pages = 1;
1682         struct page *page;
1683         struct kvm_io_device *pio_dev;
1684
1685         vcpu->run->exit_reason = KVM_EXIT_IO;
1686         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1687         vcpu->run->io.size = vcpu->pio.size = size;
1688         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1689         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1690         vcpu->run->io.port = vcpu->pio.port = port;
1691         vcpu->pio.in = in;
1692         vcpu->pio.string = 1;
1693         vcpu->pio.down = down;
1694         vcpu->pio.guest_page_offset = offset_in_page(address);
1695         vcpu->pio.rep = rep;
1696
1697         if (!count) {
1698                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1699                 return 1;
1700         }
1701
1702         if (!down)
1703                 in_page = PAGE_SIZE - offset_in_page(address);
1704         else
1705                 in_page = offset_in_page(address) + size;
1706         now = min(count, (unsigned long)in_page / size);
1707         if (!now) {
1708                 /*
1709                  * String I/O straddles page boundary.  Pin two guest pages
1710                  * so that we satisfy atomicity constraints.  Do just one
1711                  * transaction to avoid complexity.
1712                  */
1713                 nr_pages = 2;
1714                 now = 1;
1715         }
1716         if (down) {
1717                 /*
1718                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1719                  */
1720                 pr_unimpl(vcpu, "guest string pio down\n");
1721                 inject_gp(vcpu);
1722                 return 1;
1723         }
1724         vcpu->run->io.count = now;
1725         vcpu->pio.cur_count = now;
1726
1727         if (vcpu->pio.cur_count == vcpu->pio.count)
1728                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1729
1730         for (i = 0; i < nr_pages; ++i) {
1731                 mutex_lock(&vcpu->kvm->lock);
1732                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1733                 vcpu->pio.guest_pages[i] = page;
1734                 mutex_unlock(&vcpu->kvm->lock);
1735                 if (!page) {
1736                         inject_gp(vcpu);
1737                         free_pio_guest_pages(vcpu);
1738                         return 1;
1739                 }
1740         }
1741
1742         pio_dev = vcpu_find_pio_dev(vcpu, port);
1743         if (!vcpu->pio.in) {
1744                 /* string PIO write */
1745                 ret = pio_copy_data(vcpu);
1746                 if (ret >= 0 && pio_dev) {
1747                         pio_string_write(pio_dev, vcpu);
1748                         complete_pio(vcpu);
1749                         if (vcpu->pio.count == 0)
1750                                 ret = 1;
1751                 }
1752         } else if (pio_dev)
1753                 pr_unimpl(vcpu, "no string pio read support yet, "
1754                        "port %x size %d count %ld\n",
1755                         port, size, count);
1756
1757         return ret;
1758 }
1759 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1760
1761 /*
1762  * Check if userspace requested an interrupt window, and that the
1763  * interrupt window is open.
1764  *
1765  * No need to exit to userspace if we already have an interrupt queued.
1766  */
1767 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1768                                           struct kvm_run *kvm_run)
1769 {
1770         return (!vcpu->irq_summary &&
1771                 kvm_run->request_interrupt_window &&
1772                 vcpu->interrupt_window_open &&
1773                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1774 }
1775
1776 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1777                               struct kvm_run *kvm_run)
1778 {
1779         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1780         kvm_run->cr8 = get_cr8(vcpu);
1781         kvm_run->apic_base = kvm_get_apic_base(vcpu);
1782         if (irqchip_in_kernel(vcpu->kvm))
1783                 kvm_run->ready_for_interrupt_injection = 1;
1784         else
1785                 kvm_run->ready_for_interrupt_injection =
1786                                         (vcpu->interrupt_window_open &&
1787                                          vcpu->irq_summary == 0);
1788 }
1789
1790 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1791 {
1792         int r;
1793
1794         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1795                 pr_debug("vcpu %d received sipi with vector # %x\n",
1796                        vcpu->vcpu_id, vcpu->sipi_vector);
1797                 kvm_lapic_reset(vcpu);
1798                 r = kvm_x86_ops->vcpu_reset(vcpu);
1799                 if (r)
1800                         return r;
1801                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
1802         }
1803
1804 preempted:
1805         if (vcpu->guest_debug.enabled)
1806                 kvm_x86_ops->guest_debug_pre(vcpu);
1807
1808 again:
1809         r = kvm_mmu_reload(vcpu);
1810         if (unlikely(r))
1811                 goto out;
1812
1813         kvm_inject_pending_timer_irqs(vcpu);
1814
1815         preempt_disable();
1816
1817         kvm_x86_ops->prepare_guest_switch(vcpu);
1818         kvm_load_guest_fpu(vcpu);
1819
1820         local_irq_disable();
1821
1822         if (signal_pending(current)) {
1823                 local_irq_enable();
1824                 preempt_enable();
1825                 r = -EINTR;
1826                 kvm_run->exit_reason = KVM_EXIT_INTR;
1827                 ++vcpu->stat.signal_exits;
1828                 goto out;
1829         }
1830
1831         if (irqchip_in_kernel(vcpu->kvm))
1832                 kvm_x86_ops->inject_pending_irq(vcpu);
1833         else if (!vcpu->mmio_read_completed)
1834                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
1835
1836         vcpu->guest_mode = 1;
1837         kvm_guest_enter();
1838
1839         if (vcpu->requests)
1840                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
1841                         kvm_x86_ops->tlb_flush(vcpu);
1842
1843         kvm_x86_ops->run(vcpu, kvm_run);
1844
1845         vcpu->guest_mode = 0;
1846         local_irq_enable();
1847
1848         ++vcpu->stat.exits;
1849
1850         /*
1851          * We must have an instruction between local_irq_enable() and
1852          * kvm_guest_exit(), so the timer interrupt isn't delayed by
1853          * the interrupt shadow.  The stat.exits increment will do nicely.
1854          * But we need to prevent reordering, hence this barrier():
1855          */
1856         barrier();
1857
1858         kvm_guest_exit();
1859
1860         preempt_enable();
1861
1862         /*
1863          * Profile KVM exit RIPs:
1864          */
1865         if (unlikely(prof_on == KVM_PROFILING)) {
1866                 kvm_x86_ops->cache_regs(vcpu);
1867                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
1868         }
1869
1870         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
1871
1872         if (r > 0) {
1873                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1874                         r = -EINTR;
1875                         kvm_run->exit_reason = KVM_EXIT_INTR;
1876                         ++vcpu->stat.request_irq_exits;
1877                         goto out;
1878                 }
1879                 if (!need_resched()) {
1880                         ++vcpu->stat.light_exits;
1881                         goto again;
1882                 }
1883         }
1884
1885 out:
1886         if (r > 0) {
1887                 kvm_resched(vcpu);
1888                 goto preempted;
1889         }
1890
1891         post_kvm_run_save(vcpu, kvm_run);
1892
1893         return r;
1894 }
1895
1896
1897 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1898 {
1899         int r;
1900         sigset_t sigsaved;
1901
1902         vcpu_load(vcpu);
1903
1904         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
1905                 kvm_vcpu_block(vcpu);
1906                 vcpu_put(vcpu);
1907                 return -EAGAIN;
1908         }
1909
1910         if (vcpu->sigset_active)
1911                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1912
1913         /* re-sync apic's tpr */
1914         if (!irqchip_in_kernel(vcpu->kvm))
1915                 set_cr8(vcpu, kvm_run->cr8);
1916
1917         if (vcpu->pio.cur_count) {
1918                 r = complete_pio(vcpu);
1919                 if (r)
1920                         goto out;
1921         }
1922 #if CONFIG_HAS_IOMEM
1923         if (vcpu->mmio_needed) {
1924                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1925                 vcpu->mmio_read_completed = 1;
1926                 vcpu->mmio_needed = 0;
1927                 r = emulate_instruction(vcpu, kvm_run,
1928                                         vcpu->mmio_fault_cr2, 0, 1);
1929                 if (r == EMULATE_DO_MMIO) {
1930                         /*
1931                          * Read-modify-write.  Back to userspace.
1932                          */
1933                         r = 0;
1934                         goto out;
1935                 }
1936         }
1937 #endif
1938         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1939                 kvm_x86_ops->cache_regs(vcpu);
1940                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1941                 kvm_x86_ops->decache_regs(vcpu);
1942         }
1943
1944         r = __vcpu_run(vcpu, kvm_run);
1945
1946 out:
1947         if (vcpu->sigset_active)
1948                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1949
1950         vcpu_put(vcpu);
1951         return r;
1952 }
1953
1954 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1955                                    struct kvm_regs *regs)
1956 {
1957         vcpu_load(vcpu);
1958
1959         kvm_x86_ops->cache_regs(vcpu);
1960
1961         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1962         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1963         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1964         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1965         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1966         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1967         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1968         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1969 #ifdef CONFIG_X86_64
1970         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1971         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1972         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1973         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1974         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1975         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1976         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1977         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1978 #endif
1979
1980         regs->rip = vcpu->rip;
1981         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
1982
1983         /*
1984          * Don't leak debug flags in case they were set for guest debugging
1985          */
1986         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1987                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1988
1989         vcpu_put(vcpu);
1990
1991         return 0;
1992 }
1993
1994 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1995                                    struct kvm_regs *regs)
1996 {
1997         vcpu_load(vcpu);
1998
1999         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2000         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2001         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2002         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2003         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2004         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2005         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2006         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2007 #ifdef CONFIG_X86_64
2008         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2009         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2010         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2011         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2012         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2013         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2014         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2015         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2016 #endif
2017
2018         vcpu->rip = regs->rip;
2019         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2020
2021         kvm_x86_ops->decache_regs(vcpu);
2022
2023         vcpu_put(vcpu);
2024
2025         return 0;
2026 }
2027
2028 static void get_segment(struct kvm_vcpu *vcpu,
2029                         struct kvm_segment *var, int seg)
2030 {
2031         return kvm_x86_ops->get_segment(vcpu, var, seg);
2032 }
2033
2034 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2035                                     struct kvm_sregs *sregs)
2036 {
2037         struct descriptor_table dt;
2038         int pending_vec;
2039
2040         vcpu_load(vcpu);
2041
2042         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2043         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2044         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2045         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2046         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2047         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2048
2049         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2050         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2051
2052         kvm_x86_ops->get_idt(vcpu, &dt);
2053         sregs->idt.limit = dt.limit;
2054         sregs->idt.base = dt.base;
2055         kvm_x86_ops->get_gdt(vcpu, &dt);
2056         sregs->gdt.limit = dt.limit;
2057         sregs->gdt.base = dt.base;
2058
2059         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2060         sregs->cr0 = vcpu->cr0;
2061         sregs->cr2 = vcpu->cr2;
2062         sregs->cr3 = vcpu->cr3;
2063         sregs->cr4 = vcpu->cr4;
2064         sregs->cr8 = get_cr8(vcpu);
2065         sregs->efer = vcpu->shadow_efer;
2066         sregs->apic_base = kvm_get_apic_base(vcpu);
2067
2068         if (irqchip_in_kernel(vcpu->kvm)) {
2069                 memset(sregs->interrupt_bitmap, 0,
2070                        sizeof sregs->interrupt_bitmap);
2071                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2072                 if (pending_vec >= 0)
2073                         set_bit(pending_vec,
2074                                 (unsigned long *)sregs->interrupt_bitmap);
2075         } else
2076                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2077                        sizeof sregs->interrupt_bitmap);
2078
2079         vcpu_put(vcpu);
2080
2081         return 0;
2082 }
2083
2084 static void set_segment(struct kvm_vcpu *vcpu,
2085                         struct kvm_segment *var, int seg)
2086 {
2087         return kvm_x86_ops->set_segment(vcpu, var, seg);
2088 }
2089
2090 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2091                                     struct kvm_sregs *sregs)
2092 {
2093         int mmu_reset_needed = 0;
2094         int i, pending_vec, max_bits;
2095         struct descriptor_table dt;
2096
2097         vcpu_load(vcpu);
2098
2099         dt.limit = sregs->idt.limit;
2100         dt.base = sregs->idt.base;
2101         kvm_x86_ops->set_idt(vcpu, &dt);
2102         dt.limit = sregs->gdt.limit;
2103         dt.base = sregs->gdt.base;
2104         kvm_x86_ops->set_gdt(vcpu, &dt);
2105
2106         vcpu->cr2 = sregs->cr2;
2107         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2108         vcpu->cr3 = sregs->cr3;
2109
2110         set_cr8(vcpu, sregs->cr8);
2111
2112         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2113 #ifdef CONFIG_X86_64
2114         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2115 #endif
2116         kvm_set_apic_base(vcpu, sregs->apic_base);
2117
2118         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2119
2120         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2121         vcpu->cr0 = sregs->cr0;
2122         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2123
2124         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2125         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2126         if (!is_long_mode(vcpu) && is_pae(vcpu))
2127                 load_pdptrs(vcpu, vcpu->cr3);
2128
2129         if (mmu_reset_needed)
2130                 kvm_mmu_reset_context(vcpu);
2131
2132         if (!irqchip_in_kernel(vcpu->kvm)) {
2133                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2134                        sizeof vcpu->irq_pending);
2135                 vcpu->irq_summary = 0;
2136                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2137                         if (vcpu->irq_pending[i])
2138                                 __set_bit(i, &vcpu->irq_summary);
2139         } else {
2140                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2141                 pending_vec = find_first_bit(
2142                         (const unsigned long *)sregs->interrupt_bitmap,
2143                         max_bits);
2144                 /* Only pending external irq is handled here */
2145                 if (pending_vec < max_bits) {
2146                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2147                         pr_debug("Set back pending irq %d\n",
2148                                  pending_vec);
2149                 }
2150         }
2151
2152         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2153         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2154         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2155         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2156         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2157         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2158
2159         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2160         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2161
2162         vcpu_put(vcpu);
2163
2164         return 0;
2165 }
2166
2167 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2168 {
2169         struct kvm_segment cs;
2170
2171         get_segment(vcpu, &cs, VCPU_SREG_CS);
2172         *db = cs.db;
2173         *l = cs.l;
2174 }
2175 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2176
2177 /*
2178  * Translate a guest virtual address to a guest physical address.
2179  */
2180 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2181                                     struct kvm_translation *tr)
2182 {
2183         unsigned long vaddr = tr->linear_address;
2184         gpa_t gpa;
2185
2186         vcpu_load(vcpu);
2187         mutex_lock(&vcpu->kvm->lock);
2188         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2189         tr->physical_address = gpa;
2190         tr->valid = gpa != UNMAPPED_GVA;
2191         tr->writeable = 1;
2192         tr->usermode = 0;
2193         mutex_unlock(&vcpu->kvm->lock);
2194         vcpu_put(vcpu);
2195
2196         return 0;
2197 }
2198
2199 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2200                                     struct kvm_interrupt *irq)
2201 {
2202         if (irq->irq < 0 || irq->irq >= 256)
2203                 return -EINVAL;
2204         if (irqchip_in_kernel(vcpu->kvm))
2205                 return -ENXIO;
2206         vcpu_load(vcpu);
2207
2208         set_bit(irq->irq, vcpu->irq_pending);
2209         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2210
2211         vcpu_put(vcpu);
2212
2213         return 0;
2214 }
2215
2216 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2217                                       struct kvm_debug_guest *dbg)
2218 {
2219         int r;
2220
2221         vcpu_load(vcpu);
2222
2223         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2224
2225         vcpu_put(vcpu);
2226
2227         return r;
2228 }
2229
2230 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2231                                     unsigned long address,
2232                                     int *type)
2233 {
2234         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2235         unsigned long pgoff;
2236         struct page *page;
2237
2238         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2239         if (pgoff == 0)
2240                 page = virt_to_page(vcpu->run);
2241         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2242                 page = virt_to_page(vcpu->pio_data);
2243         else
2244                 return NOPAGE_SIGBUS;
2245         get_page(page);
2246         if (type != NULL)
2247                 *type = VM_FAULT_MINOR;
2248
2249         return page;
2250 }
2251
2252 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2253         .nopage = kvm_vcpu_nopage,
2254 };
2255
2256 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2257 {
2258         vma->vm_ops = &kvm_vcpu_vm_ops;
2259         return 0;
2260 }
2261
2262 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2263 {
2264         struct kvm_vcpu *vcpu = filp->private_data;
2265
2266         fput(vcpu->kvm->filp);
2267         return 0;
2268 }
2269
2270 static struct file_operations kvm_vcpu_fops = {
2271         .release        = kvm_vcpu_release,
2272         .unlocked_ioctl = kvm_vcpu_ioctl,
2273         .compat_ioctl   = kvm_vcpu_ioctl,
2274         .mmap           = kvm_vcpu_mmap,
2275 };
2276
2277 /*
2278  * Allocates an inode for the vcpu.
2279  */
2280 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2281 {
2282         int fd, r;
2283         struct inode *inode;
2284         struct file *file;
2285
2286         r = anon_inode_getfd(&fd, &inode, &file,
2287                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2288         if (r)
2289                 return r;
2290         atomic_inc(&vcpu->kvm->filp->f_count);
2291         return fd;
2292 }
2293
2294 /*
2295  * Creates some virtual cpus.  Good luck creating more than one.
2296  */
2297 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2298 {
2299         int r;
2300         struct kvm_vcpu *vcpu;
2301
2302         if (!valid_vcpu(n))
2303                 return -EINVAL;
2304
2305         vcpu = kvm_x86_ops->vcpu_create(kvm, n);
2306         if (IS_ERR(vcpu))
2307                 return PTR_ERR(vcpu);
2308
2309         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2310
2311         /* We do fxsave: this must be aligned. */
2312         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2313
2314         vcpu_load(vcpu);
2315         r = kvm_x86_ops->vcpu_reset(vcpu);
2316         if (r == 0)
2317                 r = kvm_mmu_setup(vcpu);
2318         vcpu_put(vcpu);
2319         if (r < 0)
2320                 goto free_vcpu;
2321
2322         mutex_lock(&kvm->lock);
2323         if (kvm->vcpus[n]) {
2324                 r = -EEXIST;
2325                 mutex_unlock(&kvm->lock);
2326                 goto mmu_unload;
2327         }
2328         kvm->vcpus[n] = vcpu;
2329         mutex_unlock(&kvm->lock);
2330
2331         /* Now it's all set up, let userspace reach it */
2332         r = create_vcpu_fd(vcpu);
2333         if (r < 0)
2334                 goto unlink;
2335         return r;
2336
2337 unlink:
2338         mutex_lock(&kvm->lock);
2339         kvm->vcpus[n] = NULL;
2340         mutex_unlock(&kvm->lock);
2341
2342 mmu_unload:
2343         vcpu_load(vcpu);
2344         kvm_mmu_unload(vcpu);
2345         vcpu_put(vcpu);
2346
2347 free_vcpu:
2348         kvm_x86_ops->vcpu_free(vcpu);
2349         return r;
2350 }
2351
2352 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2353 {
2354         if (sigset) {
2355                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2356                 vcpu->sigset_active = 1;
2357                 vcpu->sigset = *sigset;
2358         } else
2359                 vcpu->sigset_active = 0;
2360         return 0;
2361 }
2362
2363 /*
2364  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2365  * we have asm/x86/processor.h
2366  */
2367 struct fxsave {
2368         u16     cwd;
2369         u16     swd;
2370         u16     twd;
2371         u16     fop;
2372         u64     rip;
2373         u64     rdp;
2374         u32     mxcsr;
2375         u32     mxcsr_mask;
2376         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2377 #ifdef CONFIG_X86_64
2378         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2379 #else
2380         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2381 #endif
2382 };
2383
2384 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2385 {
2386         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2387
2388         vcpu_load(vcpu);
2389
2390         memcpy(fpu->fpr, fxsave->st_space, 128);
2391         fpu->fcw = fxsave->cwd;
2392         fpu->fsw = fxsave->swd;
2393         fpu->ftwx = fxsave->twd;
2394         fpu->last_opcode = fxsave->fop;
2395         fpu->last_ip = fxsave->rip;
2396         fpu->last_dp = fxsave->rdp;
2397         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2398
2399         vcpu_put(vcpu);
2400
2401         return 0;
2402 }
2403
2404 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2405 {
2406         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2407
2408         vcpu_load(vcpu);
2409
2410         memcpy(fxsave->st_space, fpu->fpr, 128);
2411         fxsave->cwd = fpu->fcw;
2412         fxsave->swd = fpu->fsw;
2413         fxsave->twd = fpu->ftwx;
2414         fxsave->fop = fpu->last_opcode;
2415         fxsave->rip = fpu->last_ip;
2416         fxsave->rdp = fpu->last_dp;
2417         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2418
2419         vcpu_put(vcpu);
2420
2421         return 0;
2422 }
2423
2424 static long kvm_vcpu_ioctl(struct file *filp,
2425                            unsigned int ioctl, unsigned long arg)
2426 {
2427         struct kvm_vcpu *vcpu = filp->private_data;
2428         void __user *argp = (void __user *)arg;
2429         int r;
2430
2431         switch (ioctl) {
2432         case KVM_RUN:
2433                 r = -EINVAL;
2434                 if (arg)
2435                         goto out;
2436                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2437                 break;
2438         case KVM_GET_REGS: {
2439                 struct kvm_regs kvm_regs;
2440
2441                 memset(&kvm_regs, 0, sizeof kvm_regs);
2442                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2443                 if (r)
2444                         goto out;
2445                 r = -EFAULT;
2446                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2447                         goto out;
2448                 r = 0;
2449                 break;
2450         }
2451         case KVM_SET_REGS: {
2452                 struct kvm_regs kvm_regs;
2453
2454                 r = -EFAULT;
2455                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2456                         goto out;
2457                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2458                 if (r)
2459                         goto out;
2460                 r = 0;
2461                 break;
2462         }
2463         case KVM_GET_SREGS: {
2464                 struct kvm_sregs kvm_sregs;
2465
2466                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2467                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2468                 if (r)
2469                         goto out;
2470                 r = -EFAULT;
2471                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2472                         goto out;
2473                 r = 0;
2474                 break;
2475         }
2476         case KVM_SET_SREGS: {
2477                 struct kvm_sregs kvm_sregs;
2478
2479                 r = -EFAULT;
2480                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2481                         goto out;
2482                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2483                 if (r)
2484                         goto out;
2485                 r = 0;
2486                 break;
2487         }
2488         case KVM_TRANSLATE: {
2489                 struct kvm_translation tr;
2490
2491                 r = -EFAULT;
2492                 if (copy_from_user(&tr, argp, sizeof tr))
2493                         goto out;
2494                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2495                 if (r)
2496                         goto out;
2497                 r = -EFAULT;
2498                 if (copy_to_user(argp, &tr, sizeof tr))
2499                         goto out;
2500                 r = 0;
2501                 break;
2502         }
2503         case KVM_INTERRUPT: {
2504                 struct kvm_interrupt irq;
2505
2506                 r = -EFAULT;
2507                 if (copy_from_user(&irq, argp, sizeof irq))
2508                         goto out;
2509                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2510                 if (r)
2511                         goto out;
2512                 r = 0;
2513                 break;
2514         }
2515         case KVM_DEBUG_GUEST: {
2516                 struct kvm_debug_guest dbg;
2517
2518                 r = -EFAULT;
2519                 if (copy_from_user(&dbg, argp, sizeof dbg))
2520                         goto out;
2521                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2522                 if (r)
2523                         goto out;
2524                 r = 0;
2525                 break;
2526         }
2527         case KVM_SET_SIGNAL_MASK: {
2528                 struct kvm_signal_mask __user *sigmask_arg = argp;
2529                 struct kvm_signal_mask kvm_sigmask;
2530                 sigset_t sigset, *p;
2531
2532                 p = NULL;
2533                 if (argp) {
2534                         r = -EFAULT;
2535                         if (copy_from_user(&kvm_sigmask, argp,
2536                                            sizeof kvm_sigmask))
2537                                 goto out;
2538                         r = -EINVAL;
2539                         if (kvm_sigmask.len != sizeof sigset)
2540                                 goto out;
2541                         r = -EFAULT;
2542                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2543                                            sizeof sigset))
2544                                 goto out;
2545                         p = &sigset;
2546                 }
2547                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2548                 break;
2549         }
2550         case KVM_GET_FPU: {
2551                 struct kvm_fpu fpu;
2552
2553                 memset(&fpu, 0, sizeof fpu);
2554                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2555                 if (r)
2556                         goto out;
2557                 r = -EFAULT;
2558                 if (copy_to_user(argp, &fpu, sizeof fpu))
2559                         goto out;
2560                 r = 0;
2561                 break;
2562         }
2563         case KVM_SET_FPU: {
2564                 struct kvm_fpu fpu;
2565
2566                 r = -EFAULT;
2567                 if (copy_from_user(&fpu, argp, sizeof fpu))
2568                         goto out;
2569                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2570                 if (r)
2571                         goto out;
2572                 r = 0;
2573                 break;
2574         }
2575         default:
2576                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2577         }
2578 out:
2579         return r;
2580 }
2581
2582 static long kvm_vm_ioctl(struct file *filp,
2583                            unsigned int ioctl, unsigned long arg)
2584 {
2585         struct kvm *kvm = filp->private_data;
2586         void __user *argp = (void __user *)arg;
2587         int r;
2588
2589         switch (ioctl) {
2590         case KVM_CREATE_VCPU:
2591                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2592                 if (r < 0)
2593                         goto out;
2594                 break;
2595         case KVM_SET_USER_MEMORY_REGION: {
2596                 struct kvm_userspace_memory_region kvm_userspace_mem;
2597
2598                 r = -EFAULT;
2599                 if (copy_from_user(&kvm_userspace_mem, argp,
2600                                                 sizeof kvm_userspace_mem))
2601                         goto out;
2602
2603                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2604                 if (r)
2605                         goto out;
2606                 break;
2607         }
2608         case KVM_GET_DIRTY_LOG: {
2609                 struct kvm_dirty_log log;
2610
2611                 r = -EFAULT;
2612                 if (copy_from_user(&log, argp, sizeof log))
2613                         goto out;
2614                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2615                 if (r)
2616                         goto out;
2617                 break;
2618         }
2619         default:
2620                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2621         }
2622 out:
2623         return r;
2624 }
2625
2626 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2627                                   unsigned long address,
2628                                   int *type)
2629 {
2630         struct kvm *kvm = vma->vm_file->private_data;
2631         unsigned long pgoff;
2632         struct page *page;
2633
2634         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2635         if (!kvm_is_visible_gfn(kvm, pgoff))
2636                 return NOPAGE_SIGBUS;
2637         /* current->mm->mmap_sem is already held so call lockless version */
2638         page = __gfn_to_page(kvm, pgoff);
2639         if (is_error_page(page)) {
2640                 kvm_release_page(page);
2641                 return NOPAGE_SIGBUS;
2642         }
2643         if (type != NULL)
2644                 *type = VM_FAULT_MINOR;
2645
2646         return page;
2647 }
2648
2649 static struct vm_operations_struct kvm_vm_vm_ops = {
2650         .nopage = kvm_vm_nopage,
2651 };
2652
2653 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2654 {
2655         vma->vm_ops = &kvm_vm_vm_ops;
2656         return 0;
2657 }
2658
2659 static struct file_operations kvm_vm_fops = {
2660         .release        = kvm_vm_release,
2661         .unlocked_ioctl = kvm_vm_ioctl,
2662         .compat_ioctl   = kvm_vm_ioctl,
2663         .mmap           = kvm_vm_mmap,
2664 };
2665
2666 static int kvm_dev_ioctl_create_vm(void)
2667 {
2668         int fd, r;
2669         struct inode *inode;
2670         struct file *file;
2671         struct kvm *kvm;
2672
2673         kvm = kvm_create_vm();
2674         if (IS_ERR(kvm))
2675                 return PTR_ERR(kvm);
2676         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2677         if (r) {
2678                 kvm_destroy_vm(kvm);
2679                 return r;
2680         }
2681
2682         kvm->filp = file;
2683
2684         return fd;
2685 }
2686
2687 static long kvm_dev_ioctl(struct file *filp,
2688                           unsigned int ioctl, unsigned long arg)
2689 {
2690         void __user *argp = (void __user *)arg;
2691         long r = -EINVAL;
2692
2693         switch (ioctl) {
2694         case KVM_GET_API_VERSION:
2695                 r = -EINVAL;
2696                 if (arg)
2697                         goto out;
2698                 r = KVM_API_VERSION;
2699                 break;
2700         case KVM_CREATE_VM:
2701                 r = -EINVAL;
2702                 if (arg)
2703                         goto out;
2704                 r = kvm_dev_ioctl_create_vm();
2705                 break;
2706         case KVM_CHECK_EXTENSION: {
2707                 int ext = (long)argp;
2708
2709                 switch (ext) {
2710                 case KVM_CAP_IRQCHIP:
2711                 case KVM_CAP_HLT:
2712                 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2713                 case KVM_CAP_USER_MEMORY:
2714                 case KVM_CAP_SET_TSS_ADDR:
2715                         r = 1;
2716                         break;
2717                 default:
2718                         r = 0;
2719                         break;
2720                 }
2721                 break;
2722         }
2723         case KVM_GET_VCPU_MMAP_SIZE:
2724                 r = -EINVAL;
2725                 if (arg)
2726                         goto out;
2727                 r = 2 * PAGE_SIZE;
2728                 break;
2729         default:
2730                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2731         }
2732 out:
2733         return r;
2734 }
2735
2736 static struct file_operations kvm_chardev_ops = {
2737         .unlocked_ioctl = kvm_dev_ioctl,
2738         .compat_ioctl   = kvm_dev_ioctl,
2739 };
2740
2741 static struct miscdevice kvm_dev = {
2742         KVM_MINOR,
2743         "kvm",
2744         &kvm_chardev_ops,
2745 };
2746
2747 /*
2748  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2749  * cached on it.
2750  */
2751 static void decache_vcpus_on_cpu(int cpu)
2752 {
2753         struct kvm *vm;
2754         struct kvm_vcpu *vcpu;
2755         int i;
2756
2757         spin_lock(&kvm_lock);
2758         list_for_each_entry(vm, &vm_list, vm_list)
2759                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2760                         vcpu = vm->vcpus[i];
2761                         if (!vcpu)
2762                                 continue;
2763                         /*
2764                          * If the vcpu is locked, then it is running on some
2765                          * other cpu and therefore it is not cached on the
2766                          * cpu in question.
2767                          *
2768                          * If it's not locked, check the last cpu it executed
2769                          * on.
2770                          */
2771                         if (mutex_trylock(&vcpu->mutex)) {
2772                                 if (vcpu->cpu == cpu) {
2773                                         kvm_x86_ops->vcpu_decache(vcpu);
2774                                         vcpu->cpu = -1;
2775                                 }
2776                                 mutex_unlock(&vcpu->mutex);
2777                         }
2778                 }
2779         spin_unlock(&kvm_lock);
2780 }
2781
2782 static void hardware_enable(void *junk)
2783 {
2784         int cpu = raw_smp_processor_id();
2785
2786         if (cpu_isset(cpu, cpus_hardware_enabled))
2787                 return;
2788         cpu_set(cpu, cpus_hardware_enabled);
2789         kvm_x86_ops->hardware_enable(NULL);
2790 }
2791
2792 static void hardware_disable(void *junk)
2793 {
2794         int cpu = raw_smp_processor_id();
2795
2796         if (!cpu_isset(cpu, cpus_hardware_enabled))
2797                 return;
2798         cpu_clear(cpu, cpus_hardware_enabled);
2799         decache_vcpus_on_cpu(cpu);
2800         kvm_x86_ops->hardware_disable(NULL);
2801 }
2802
2803 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2804                            void *v)
2805 {
2806         int cpu = (long)v;
2807
2808         switch (val) {
2809         case CPU_DYING:
2810         case CPU_DYING_FROZEN:
2811                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2812                        cpu);
2813                 hardware_disable(NULL);
2814                 break;
2815         case CPU_UP_CANCELED:
2816         case CPU_UP_CANCELED_FROZEN:
2817                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2818                        cpu);
2819                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
2820                 break;
2821         case CPU_ONLINE:
2822         case CPU_ONLINE_FROZEN:
2823                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2824                        cpu);
2825                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
2826                 break;
2827         }
2828         return NOTIFY_OK;
2829 }
2830
2831 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2832                       void *v)
2833 {
2834         if (val == SYS_RESTART) {
2835                 /*
2836                  * Some (well, at least mine) BIOSes hang on reboot if
2837                  * in vmx root mode.
2838                  */
2839                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2840                 on_each_cpu(hardware_disable, NULL, 0, 1);
2841         }
2842         return NOTIFY_OK;
2843 }
2844
2845 static struct notifier_block kvm_reboot_notifier = {
2846         .notifier_call = kvm_reboot,
2847         .priority = 0,
2848 };
2849
2850 void kvm_io_bus_init(struct kvm_io_bus *bus)
2851 {
2852         memset(bus, 0, sizeof(*bus));
2853 }
2854
2855 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2856 {
2857         int i;
2858
2859         for (i = 0; i < bus->dev_count; i++) {
2860                 struct kvm_io_device *pos = bus->devs[i];
2861
2862                 kvm_iodevice_destructor(pos);
2863         }
2864 }
2865
2866 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
2867 {
2868         int i;
2869
2870         for (i = 0; i < bus->dev_count; i++) {
2871                 struct kvm_io_device *pos = bus->devs[i];
2872
2873                 if (pos->in_range(pos, addr))
2874                         return pos;
2875         }
2876
2877         return NULL;
2878 }
2879
2880 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2881 {
2882         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2883
2884         bus->devs[bus->dev_count++] = dev;
2885 }
2886
2887 static struct notifier_block kvm_cpu_notifier = {
2888         .notifier_call = kvm_cpu_hotplug,
2889         .priority = 20, /* must be > scheduler priority */
2890 };
2891
2892 static u64 stat_get(void *_offset)
2893 {
2894         unsigned offset = (long)_offset;
2895         u64 total = 0;
2896         struct kvm *kvm;
2897         struct kvm_vcpu *vcpu;
2898         int i;
2899
2900         spin_lock(&kvm_lock);
2901         list_for_each_entry(kvm, &vm_list, vm_list)
2902                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2903                         vcpu = kvm->vcpus[i];
2904                         if (vcpu)
2905                                 total += *(u32 *)((void *)vcpu + offset);
2906                 }
2907         spin_unlock(&kvm_lock);
2908         return total;
2909 }
2910
2911 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
2912
2913 static __init void kvm_init_debug(void)
2914 {
2915         struct kvm_stats_debugfs_item *p;
2916
2917         debugfs_dir = debugfs_create_dir("kvm", NULL);
2918         for (p = debugfs_entries; p->name; ++p)
2919                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
2920                                                 (void *)(long)p->offset,
2921                                                 &stat_fops);
2922 }
2923
2924 static void kvm_exit_debug(void)
2925 {
2926         struct kvm_stats_debugfs_item *p;
2927
2928         for (p = debugfs_entries; p->name; ++p)
2929                 debugfs_remove(p->dentry);
2930         debugfs_remove(debugfs_dir);
2931 }
2932
2933 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2934 {
2935         hardware_disable(NULL);
2936         return 0;
2937 }
2938
2939 static int kvm_resume(struct sys_device *dev)
2940 {
2941         hardware_enable(NULL);
2942         return 0;
2943 }
2944
2945 static struct sysdev_class kvm_sysdev_class = {
2946         .name = "kvm",
2947         .suspend = kvm_suspend,
2948         .resume = kvm_resume,
2949 };
2950
2951 static struct sys_device kvm_sysdev = {
2952         .id = 0,
2953         .cls = &kvm_sysdev_class,
2954 };
2955
2956 struct page *bad_page;
2957
2958 static inline
2959 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2960 {
2961         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2962 }
2963
2964 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2965 {
2966         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2967
2968         kvm_x86_ops->vcpu_load(vcpu, cpu);
2969 }
2970
2971 static void kvm_sched_out(struct preempt_notifier *pn,
2972                           struct task_struct *next)
2973 {
2974         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2975
2976         kvm_x86_ops->vcpu_put(vcpu);
2977 }
2978
2979 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
2980                   struct module *module)
2981 {
2982         int r;
2983         int cpu;
2984
2985         if (kvm_x86_ops) {
2986                 printk(KERN_ERR "kvm: already loaded the other module\n");
2987                 return -EEXIST;
2988         }
2989
2990         if (!ops->cpu_has_kvm_support()) {
2991                 printk(KERN_ERR "kvm: no hardware support\n");
2992                 return -EOPNOTSUPP;
2993         }
2994         if (ops->disabled_by_bios()) {
2995                 printk(KERN_ERR "kvm: disabled by bios\n");
2996                 return -EOPNOTSUPP;
2997         }
2998
2999         kvm_x86_ops = ops;
3000
3001         r = kvm_x86_ops->hardware_setup();
3002         if (r < 0)
3003                 goto out;
3004
3005         for_each_online_cpu(cpu) {
3006                 smp_call_function_single(cpu,
3007                                 kvm_x86_ops->check_processor_compatibility,
3008                                 &r, 0, 1);
3009                 if (r < 0)
3010                         goto out_free_0;
3011         }
3012
3013         on_each_cpu(hardware_enable, NULL, 0, 1);
3014         r = register_cpu_notifier(&kvm_cpu_notifier);
3015         if (r)
3016                 goto out_free_1;
3017         register_reboot_notifier(&kvm_reboot_notifier);
3018
3019         r = sysdev_class_register(&kvm_sysdev_class);
3020         if (r)
3021                 goto out_free_2;
3022
3023         r = sysdev_register(&kvm_sysdev);
3024         if (r)
3025                 goto out_free_3;
3026
3027         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3028         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3029                                            __alignof__(struct kvm_vcpu), 0, 0);
3030         if (!kvm_vcpu_cache) {
3031                 r = -ENOMEM;
3032                 goto out_free_4;
3033         }
3034
3035         kvm_chardev_ops.owner = module;
3036
3037         r = misc_register(&kvm_dev);
3038         if (r) {
3039                 printk(KERN_ERR "kvm: misc device register failed\n");
3040                 goto out_free;
3041         }
3042
3043         kvm_preempt_ops.sched_in = kvm_sched_in;
3044         kvm_preempt_ops.sched_out = kvm_sched_out;
3045
3046         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3047
3048         return 0;
3049
3050 out_free:
3051         kmem_cache_destroy(kvm_vcpu_cache);
3052 out_free_4:
3053         sysdev_unregister(&kvm_sysdev);
3054 out_free_3:
3055         sysdev_class_unregister(&kvm_sysdev_class);
3056 out_free_2:
3057         unregister_reboot_notifier(&kvm_reboot_notifier);
3058         unregister_cpu_notifier(&kvm_cpu_notifier);
3059 out_free_1:
3060         on_each_cpu(hardware_disable, NULL, 0, 1);
3061 out_free_0:
3062         kvm_x86_ops->hardware_unsetup();
3063 out:
3064         kvm_x86_ops = NULL;
3065         return r;
3066 }
3067 EXPORT_SYMBOL_GPL(kvm_init_x86);
3068
3069 void kvm_exit_x86(void)
3070 {
3071         misc_deregister(&kvm_dev);
3072         kmem_cache_destroy(kvm_vcpu_cache);
3073         sysdev_unregister(&kvm_sysdev);
3074         sysdev_class_unregister(&kvm_sysdev_class);
3075         unregister_reboot_notifier(&kvm_reboot_notifier);
3076         unregister_cpu_notifier(&kvm_cpu_notifier);
3077         on_each_cpu(hardware_disable, NULL, 0, 1);
3078         kvm_x86_ops->hardware_unsetup();
3079         kvm_x86_ops = NULL;
3080 }
3081 EXPORT_SYMBOL_GPL(kvm_exit_x86);
3082
3083 static __init int kvm_init(void)
3084 {
3085         int r;
3086
3087         r = kvm_mmu_module_init();
3088         if (r)
3089                 goto out4;
3090
3091         kvm_init_debug();
3092
3093         kvm_arch_init();
3094
3095         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3096
3097         if (bad_page == NULL) {
3098                 r = -ENOMEM;
3099                 goto out;
3100         }
3101
3102         return 0;
3103
3104 out:
3105         kvm_exit_debug();
3106         kvm_mmu_module_exit();
3107 out4:
3108         return r;
3109 }
3110
3111 static __exit void kvm_exit(void)
3112 {
3113         kvm_exit_debug();
3114         __free_page(bad_page);
3115         kvm_mmu_module_exit();
3116 }
3117
3118 module_init(kvm_init)
3119 module_exit(kvm_exit)