SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 637 {
 638         struct kvm_memory_slot *slot;
 639         struct page *page[1];
 640         int npages;
 641
 642         might_sleep();
 643
 644         gfn = unalias_gfn(kvm, gfn);
 645         slot = __gfn_to_memslot(kvm, gfn);
 646         if (!slot) {
 647                 get_page(bad_page);
 648                 return bad_page;
 649         }
 650
 651         down_read(&current->mm->mmap_sem);
 652         npages = get_user_pages(current, current->mm,
 653                                 slot->userspace_addr
 654                                 + (gfn - slot->base_gfn) * PAGE_SIZE, 1,
 655                                 1, 1, page, NULL);
 656         up_read(&current->mm->mmap_sem);
 657         if (npages != 1) {
 658                 get_page(bad_page);
 659                 return bad_page;
 660         }
 661
 662         return page[0];
 663 }
 664 EXPORT_SYMBOL_GPL(gfn_to_page);
 665
 666 void kvm_release_page(struct page *page)
 667 {
 668         if (!PageReserved(page))
 669                 SetPageDirty(page);
 670         put_page(page);
 671 }
 672 EXPORT_SYMBOL_GPL(kvm_release_page);
 673
 674 static int next_segment(unsigned long len, int offset)
 675 {
 676         if (len > PAGE_SIZE - offset)
 677                 return PAGE_SIZE - offset;
 678         else
 679                 return len;
 680 }
 681
 682 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 683                         int len)
 684 {
 685         void *page_virt;
 686         struct page *page;
 687
 688         page = gfn_to_page(kvm, gfn);
 689         if (is_error_page(page)) {
 690                 kvm_release_page(page);
 691                 return -EFAULT;
 692         }
 693         page_virt = kmap_atomic(page, KM_USER0);
 694
 695         memcpy(data, page_virt + offset, len);
 696
 697         kunmap_atomic(page_virt, KM_USER0);
 698         kvm_release_page(page);
 699         return 0;
 700 }
 701 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 702
 703 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 704 {
 705         gfn_t gfn = gpa >> PAGE_SHIFT;
 706         int seg;
 707         int offset = offset_in_page(gpa);
 708         int ret;
 709
 710         while ((seg = next_segment(len, offset)) != 0) {
 711                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 712                 if (ret < 0)
 713                         return ret;
 714                 offset = 0;
 715                 len -= seg;
 716                 data += seg;
 717                 ++gfn;
 718         }
 719         return 0;
 720 }
 721 EXPORT_SYMBOL_GPL(kvm_read_guest);
 722
 723 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
 724                          int offset, int len)
 725 {
 726         void *page_virt;
 727         struct page *page;
 728
 729         page = gfn_to_page(kvm, gfn);
 730         if (is_error_page(page)) {
 731                 kvm_release_page(page);
 732                 return -EFAULT;
 733         }
 734         page_virt = kmap_atomic(page, KM_USER0);
 735
 736         memcpy(page_virt + offset, data, len);
 737
 738         kunmap_atomic(page_virt, KM_USER0);
 739         mark_page_dirty(kvm, gfn);
 740         kvm_release_page(page);
 741         return 0;
 742 }
 743 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
 744
 745 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
 746                     unsigned long len)
 747 {
 748         gfn_t gfn = gpa >> PAGE_SHIFT;
 749         int seg;
 750         int offset = offset_in_page(gpa);
 751         int ret;
 752
 753         while ((seg = next_segment(len, offset)) != 0) {
 754                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
 755                 if (ret < 0)
 756                         return ret;
 757                 offset = 0;
 758                 len -= seg;
 759                 data += seg;
 760                 ++gfn;
 761         }
 762         return 0;
 763 }
 764
 765 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
 766 {
 767         void *page_virt;
 768         struct page *page;
 769
 770         page = gfn_to_page(kvm, gfn);
 771         if (is_error_page(page)) {
 772                 kvm_release_page(page);
 773                 return -EFAULT;
 774         }
 775         page_virt = kmap_atomic(page, KM_USER0);
 776
 777         memset(page_virt + offset, 0, len);
 778
 779         kunmap_atomic(page_virt, KM_USER0);
 780         kvm_release_page(page);
 781         return 0;
 782 }
 783 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
 784
 785 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
 786 {
 787         gfn_t gfn = gpa >> PAGE_SHIFT;
 788         int seg;
 789         int offset = offset_in_page(gpa);
 790         int ret;
 791
 792         while ((seg = next_segment(len, offset)) != 0) {
 793                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
 794                 if (ret < 0)
 795                         return ret;
 796                 offset = 0;
 797                 len -= seg;
 798                 ++gfn;
 799         }
 800         return 0;
 801 }
 802 EXPORT_SYMBOL_GPL(kvm_clear_guest);
 803
 804 /* WARNING: Does not work on aliased pages. */
 805 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 806 {
 807         struct kvm_memory_slot *memslot;
 808
 809         memslot = __gfn_to_memslot(kvm, gfn);
 810         if (memslot && memslot->dirty_bitmap) {
 811                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 812
 813                 /* avoid RMW */
 814                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 815                         set_bit(rel_gfn, memslot->dirty_bitmap);
 816         }
 817 }
 818
 819 int emulator_read_std(unsigned long addr,
 820                              void *val,
 821                              unsigned int bytes,
 822                              struct kvm_vcpu *vcpu)
 823 {
 824         void *data = val;
 825
 826         while (bytes) {
 827                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 828                 unsigned offset = addr & (PAGE_SIZE-1);
 829                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 830                 int ret;
 831
 832                 if (gpa == UNMAPPED_GVA)
 833                         return X86EMUL_PROPAGATE_FAULT;
 834                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
 835                 if (ret < 0)
 836                         return X86EMUL_UNHANDLEABLE;
 837
 838                 bytes -= tocopy;
 839                 data += tocopy;
 840                 addr += tocopy;
 841         }
 842
 843         return X86EMUL_CONTINUE;
 844 }
 845 EXPORT_SYMBOL_GPL(emulator_read_std);
 846
 847 static int emulator_write_std(unsigned long addr,
 848                               const void *val,
 849                               unsigned int bytes,
 850                               struct kvm_vcpu *vcpu)
 851 {
 852         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
 853         return X86EMUL_UNHANDLEABLE;
 854 }
 855
 856 /*
 857  * Only apic need an MMIO device hook, so shortcut now..
 858  */
 859 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
 860                                                 gpa_t addr)
 861 {
 862         struct kvm_io_device *dev;
 863
 864         if (vcpu->apic) {
 865                 dev = &vcpu->apic->dev;
 866                 if (dev->in_range(dev, addr))
 867                         return dev;
 868         }
 869         return NULL;
 870 }
 871
 872 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
 873                                                 gpa_t addr)
 874 {
 875         struct kvm_io_device *dev;
 876
 877         dev = vcpu_find_pervcpu_dev(vcpu, addr);
 878         if (dev == NULL)
 879                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
 880         return dev;
 881 }
 882
 883 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
 884                                                gpa_t addr)
 885 {
 886         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
 887 }
 888
 889 static int emulator_read_emulated(unsigned long addr,
 890                                   void *val,
 891                                   unsigned int bytes,
 892                                   struct kvm_vcpu *vcpu)
 893 {
 894         struct kvm_io_device *mmio_dev;
 895         gpa_t                 gpa;
 896
 897         if (vcpu->mmio_read_completed) {
 898                 memcpy(val, vcpu->mmio_data, bytes);
 899                 vcpu->mmio_read_completed = 0;
 900                 return X86EMUL_CONTINUE;
 901         }
 902
 903         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 904
 905         /* For APIC access vmexit */
 906         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
 907                 goto mmio;
 908
 909         if (emulator_read_std(addr, val, bytes, vcpu)
 910                         == X86EMUL_CONTINUE)
 911                 return X86EMUL_CONTINUE;
 912         if (gpa == UNMAPPED_GVA)
 913                 return X86EMUL_PROPAGATE_FAULT;
 914
 915 mmio:
 916         /*
 917          * Is this MMIO handled locally?
 918          */
 919         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
 920         if (mmio_dev) {
 921                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
 922                 return X86EMUL_CONTINUE;
 923         }
 924
 925         vcpu->mmio_needed = 1;
 926         vcpu->mmio_phys_addr = gpa;
 927         vcpu->mmio_size = bytes;
 928         vcpu->mmio_is_write = 0;
 929
 930         return X86EMUL_UNHANDLEABLE;
 931 }
 932
 933 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
 934                                const void *val, int bytes)
 935 {
 936         int ret;
 937
 938         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
 939         if (ret < 0)
 940                 return 0;
 941         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
 942         return 1;
 943 }
 944
 945 static int emulator_write_emulated_onepage(unsigned long addr,
 946                                            const void *val,
 947                                            unsigned int bytes,
 948                                            struct kvm_vcpu *vcpu)
 949 {
 950         struct kvm_io_device *mmio_dev;
 951         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 952
 953         if (gpa == UNMAPPED_GVA) {
 954                 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
 955                 return X86EMUL_PROPAGATE_FAULT;
 956         }
 957
 958         /* For APIC access vmexit */
 959         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
 960                 goto mmio;
 961
 962         if (emulator_write_phys(vcpu, gpa, val, bytes))
 963                 return X86EMUL_CONTINUE;
 964
 965 mmio:
 966         /*
 967          * Is this MMIO handled locally?
 968          */
 969         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
 970         if (mmio_dev) {
 971                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
 972                 return X86EMUL_CONTINUE;
 973         }
 974
 975         vcpu->mmio_needed = 1;
 976         vcpu->mmio_phys_addr = gpa;
 977         vcpu->mmio_size = bytes;
 978         vcpu->mmio_is_write = 1;
 979         memcpy(vcpu->mmio_data, val, bytes);
 980
 981         return X86EMUL_CONTINUE;
 982 }
 983
 984 int emulator_write_emulated(unsigned long addr,
 985                                    const void *val,
 986                                    unsigned int bytes,
 987                                    struct kvm_vcpu *vcpu)
 988 {
 989         /* Crossing a page boundary? */
 990         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
 991                 int rc, now;
 992
 993                 now = -addr & ~PAGE_MASK;
 994                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
 995                 if (rc != X86EMUL_CONTINUE)
 996                         return rc;
 997                 addr += now;
 998                 val += now;
 999                 bytes -= now;
1000         }
1001         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1002 }
1003 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1004
1005 static int emulator_cmpxchg_emulated(unsigned long addr,
1006                                      const void *old,
1007                                      const void *new,
1008                                      unsigned int bytes,
1009                                      struct kvm_vcpu *vcpu)
1010 {
1011         static int reported;
1012
1013         if (!reported) {
1014                 reported = 1;
1015                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1016         }
1017         return emulator_write_emulated(addr, new, bytes, vcpu);
1018 }
1019
1020 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1021 {
1022         return kvm_x86_ops->get_segment_base(vcpu, seg);
1023 }
1024
1025 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1026 {
1027         return X86EMUL_CONTINUE;
1028 }
1029
1030 int emulate_clts(struct kvm_vcpu *vcpu)
1031 {
1032         kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1033         return X86EMUL_CONTINUE;
1034 }
1035
1036 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1037 {
1038         struct kvm_vcpu *vcpu = ctxt->vcpu;
1039
1040         switch (dr) {
1041         case 0 ... 3:
1042                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1043                 return X86EMUL_CONTINUE;
1044         default:
1045                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1046                 return X86EMUL_UNHANDLEABLE;
1047         }
1048 }
1049
1050 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1051 {
1052         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1053         int exception;
1054
1055         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1056         if (exception) {
1057                 /* FIXME: better handling */
1058                 return X86EMUL_UNHANDLEABLE;
1059         }
1060         return X86EMUL_CONTINUE;
1061 }
1062
1063 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1064 {
1065         static int reported;
1066         u8 opcodes[4];
1067         unsigned long rip = vcpu->rip;
1068         unsigned long rip_linear;
1069
1070         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1071
1072         if (reported)
1073                 return;
1074
1075         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1076
1077         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1078                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1079         reported = 1;
1080 }
1081 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1082
1083 struct x86_emulate_ops emulate_ops = {
1084         .read_std            = emulator_read_std,
1085         .write_std           = emulator_write_std,
1086         .read_emulated       = emulator_read_emulated,
1087         .write_emulated      = emulator_write_emulated,
1088         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1089 };
1090
1091 int emulate_instruction(struct kvm_vcpu *vcpu,
1092                         struct kvm_run *run,
1093                         unsigned long cr2,
1094                         u16 error_code,
1095                         int no_decode)
1096 {
1097         int r;
1098
1099         vcpu->mmio_fault_cr2 = cr2;
1100         kvm_x86_ops->cache_regs(vcpu);
1101
1102         vcpu->mmio_is_write = 0;
1103         vcpu->pio.string = 0;
1104
1105         if (!no_decode) {
1106                 int cs_db, cs_l;
1107                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1108
1109                 vcpu->emulate_ctxt.vcpu = vcpu;
1110                 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1111                 vcpu->emulate_ctxt.cr2 = cr2;
1112                 vcpu->emulate_ctxt.mode =
1113                         (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1114                         ? X86EMUL_MODE_REAL : cs_l
1115                         ? X86EMUL_MODE_PROT64 : cs_db
1116                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1117
1118                 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1119                         vcpu->emulate_ctxt.cs_base = 0;
1120                         vcpu->emulate_ctxt.ds_base = 0;
1121                         vcpu->emulate_ctxt.es_base = 0;
1122                         vcpu->emulate_ctxt.ss_base = 0;
1123                 } else {
1124                         vcpu->emulate_ctxt.cs_base =
1125                                         get_segment_base(vcpu, VCPU_SREG_CS);
1126                         vcpu->emulate_ctxt.ds_base =
1127                                         get_segment_base(vcpu, VCPU_SREG_DS);
1128                         vcpu->emulate_ctxt.es_base =
1129                                         get_segment_base(vcpu, VCPU_SREG_ES);
1130                         vcpu->emulate_ctxt.ss_base =
1131                                         get_segment_base(vcpu, VCPU_SREG_SS);
1132                 }
1133
1134                 vcpu->emulate_ctxt.gs_base =
1135                                         get_segment_base(vcpu, VCPU_SREG_GS);
1136                 vcpu->emulate_ctxt.fs_base =
1137                                         get_segment_base(vcpu, VCPU_SREG_FS);
1138
1139                 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1140                 if (r)  {
1141                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1142                                 return EMULATE_DONE;
1143                         return EMULATE_FAIL;
1144                 }
1145         }
1146
1147         r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1148
1149         if (vcpu->pio.string)
1150                 return EMULATE_DO_MMIO;
1151
1152         if ((r || vcpu->mmio_is_write) && run) {
1153                 run->exit_reason = KVM_EXIT_MMIO;
1154                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1155                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1156                 run->mmio.len = vcpu->mmio_size;
1157                 run->mmio.is_write = vcpu->mmio_is_write;
1158         }
1159
1160         if (r) {
1161                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1162                         return EMULATE_DONE;
1163                 if (!vcpu->mmio_needed) {
1164                         kvm_report_emulation_failure(vcpu, "mmio");
1165                         return EMULATE_FAIL;
1166                 }
1167                 return EMULATE_DO_MMIO;
1168         }
1169
1170         kvm_x86_ops->decache_regs(vcpu);
1171         kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1172
1173         if (vcpu->mmio_is_write) {
1174                 vcpu->mmio_needed = 0;
1175                 return EMULATE_DO_MMIO;
1176         }
1177
1178         return EMULATE_DONE;
1179 }
1180 EXPORT_SYMBOL_GPL(emulate_instruction);
1181
1182 /*
1183  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1184  */
1185 static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1186 {
1187         DECLARE_WAITQUEUE(wait, current);
1188
1189         add_wait_queue(&vcpu->wq, &wait);
1190
1191         /*
1192          * We will block until either an interrupt or a signal wakes us up
1193          */
1194         while (!kvm_cpu_has_interrupt(vcpu)
1195                && !signal_pending(current)
1196                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1197                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
1198                 set_current_state(TASK_INTERRUPTIBLE);
1199                 vcpu_put(vcpu);
1200                 schedule();
1201                 vcpu_load(vcpu);
1202         }
1203
1204         __set_current_state(TASK_RUNNING);
1205         remove_wait_queue(&vcpu->wq, &wait);
1206 }
1207
1208 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1209 {
1210         ++vcpu->stat.halt_exits;
1211         if (irqchip_in_kernel(vcpu->kvm)) {
1212                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1213                 kvm_vcpu_block(vcpu);
1214                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1215                         return -EINTR;
1216                 return 1;
1217         } else {
1218                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1219                 return 0;
1220         }
1221 }
1222 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1223
1224 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1225 {
1226         unsigned long nr, a0, a1, a2, a3, ret;
1227
1228         kvm_x86_ops->cache_regs(vcpu);
1229
1230         nr = vcpu->regs[VCPU_REGS_RAX];
1231         a0 = vcpu->regs[VCPU_REGS_RBX];
1232         a1 = vcpu->regs[VCPU_REGS_RCX];
1233         a2 = vcpu->regs[VCPU_REGS_RDX];
1234         a3 = vcpu->regs[VCPU_REGS_RSI];
1235
1236         if (!is_long_mode(vcpu)) {
1237                 nr &= 0xFFFFFFFF;
1238                 a0 &= 0xFFFFFFFF;
1239                 a1 &= 0xFFFFFFFF;
1240                 a2 &= 0xFFFFFFFF;
1241                 a3 &= 0xFFFFFFFF;
1242         }
1243
1244         switch (nr) {
1245         default:
1246                 ret = -KVM_ENOSYS;
1247                 break;
1248         }
1249         vcpu->regs[VCPU_REGS_RAX] = ret;
1250         kvm_x86_ops->decache_regs(vcpu);
1251         return 0;
1252 }
1253 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1254
1255 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1256 {
1257         char instruction[3];
1258         int ret = 0;
1259
1260         mutex_lock(&vcpu->kvm->lock);
1261
1262         /*
1263          * Blow out the MMU to ensure that no other VCPU has an active mapping
1264          * to ensure that the updated hypercall appears atomically across all
1265          * VCPUs.
1266          */
1267         kvm_mmu_zap_all(vcpu->kvm);
1268
1269         kvm_x86_ops->cache_regs(vcpu);
1270         kvm_x86_ops->patch_hypercall(vcpu, instruction);
1271         if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1272             != X86EMUL_CONTINUE)
1273                 ret = -EFAULT;
1274
1275         mutex_unlock(&vcpu->kvm->lock);
1276
1277         return ret;
1278 }
1279
1280 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1281 {
1282         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1283 }
1284
1285 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1286 {
1287         struct descriptor_table dt = { limit, base };
1288
1289         kvm_x86_ops->set_gdt(vcpu, &dt);
1290 }
1291
1292 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1293 {
1294         struct descriptor_table dt = { limit, base };
1295
1296         kvm_x86_ops->set_idt(vcpu, &dt);
1297 }
1298
1299 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1300                    unsigned long *rflags)
1301 {
1302         lmsw(vcpu, msw);
1303         *rflags = kvm_x86_ops->get_rflags(vcpu);
1304 }
1305
1306 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1307 {
1308         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1309         switch (cr) {
1310         case 0:
1311                 return vcpu->cr0;
1312         case 2:
1313                 return vcpu->cr2;
1314         case 3:
1315                 return vcpu->cr3;
1316         case 4:
1317                 return vcpu->cr4;
1318         default:
1319                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1320                 return 0;
1321         }
1322 }
1323
1324 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1325                      unsigned long *rflags)
1326 {
1327         switch (cr) {
1328         case 0:
1329                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1330                 *rflags = kvm_x86_ops->get_rflags(vcpu);
1331                 break;
1332         case 2:
1333                 vcpu->cr2 = val;
1334                 break;
1335         case 3:
1336                 set_cr3(vcpu, val);
1337                 break;
1338         case 4:
1339                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1340                 break;
1341         default:
1342                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1343         }
1344 }
1345
1346 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1347 {
1348         u64 data;
1349
1350         switch (msr) {
1351         case 0xc0010010: /* SYSCFG */
1352         case 0xc0010015: /* HWCR */
1353         case MSR_IA32_PLATFORM_ID:
1354         case MSR_IA32_P5_MC_ADDR:
1355         case MSR_IA32_P5_MC_TYPE:
1356         case MSR_IA32_MC0_CTL:
1357         case MSR_IA32_MCG_STATUS:
1358         case MSR_IA32_MCG_CAP:
1359         case MSR_IA32_MC0_MISC:
1360         case MSR_IA32_MC0_MISC+4:
1361         case MSR_IA32_MC0_MISC+8:
1362         case MSR_IA32_MC0_MISC+12:
1363         case MSR_IA32_MC0_MISC+16:
1364         case MSR_IA32_UCODE_REV:
1365         case MSR_IA32_PERF_STATUS:
1366         case MSR_IA32_EBL_CR_POWERON:
1367                 /* MTRR registers */
1368         case 0xfe:
1369         case 0x200 ... 0x2ff:
1370                 data = 0;
1371                 break;
1372         case 0xcd: /* fsb frequency */
1373                 data = 3;
1374                 break;
1375         case MSR_IA32_APICBASE:
1376                 data = kvm_get_apic_base(vcpu);
1377                 break;
1378         case MSR_IA32_MISC_ENABLE:
1379                 data = vcpu->ia32_misc_enable_msr;
1380                 break;
1381 #ifdef CONFIG_X86_64
1382         case MSR_EFER:
1383                 data = vcpu->shadow_efer;
1384                 break;
1385 #endif
1386         default:
1387                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1388                 return 1;
1389         }
1390         *pdata = data;
1391         return 0;
1392 }
1393 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1394
1395 /*
1396  * Reads an msr value (of 'msr_index') into 'pdata'.
1397  * Returns 0 on success, non-0 otherwise.
1398  * Assumes vcpu_load() was already called.
1399  */
1400 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1401 {
1402         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1403 }
1404
1405 #ifdef CONFIG_X86_64
1406
1407 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1408 {
1409         if (efer & EFER_RESERVED_BITS) {
1410                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1411                        efer);
1412                 inject_gp(vcpu);
1413                 return;
1414         }
1415
1416         if (is_paging(vcpu)
1417             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1418                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1419                 inject_gp(vcpu);
1420                 return;
1421         }
1422
1423         kvm_x86_ops->set_efer(vcpu, efer);
1424
1425         efer &= ~EFER_LMA;
1426         efer |= vcpu->shadow_efer & EFER_LMA;
1427
1428         vcpu->shadow_efer = efer;
1429 }
1430
1431 #endif
1432
1433 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1434 {
1435         switch (msr) {
1436 #ifdef CONFIG_X86_64
1437         case MSR_EFER:
1438                 set_efer(vcpu, data);
1439                 break;
1440 #endif
1441         case MSR_IA32_MC0_STATUS:
1442                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1443                        __FUNCTION__, data);
1444                 break;
1445         case MSR_IA32_MCG_STATUS:
1446                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1447                         __FUNCTION__, data);
1448                 break;
1449         case MSR_IA32_UCODE_REV:
1450         case MSR_IA32_UCODE_WRITE:
1451         case 0x200 ... 0x2ff: /* MTRRs */
1452                 break;
1453         case MSR_IA32_APICBASE:
1454                 kvm_set_apic_base(vcpu, data);
1455                 break;
1456         case MSR_IA32_MISC_ENABLE:
1457                 vcpu->ia32_misc_enable_msr = data;
1458                 break;
1459         default:
1460                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1461                 return 1;
1462         }
1463         return 0;
1464 }
1465 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1466
1467 /*
1468  * Writes msr value into into the appropriate "register".
1469  * Returns 0 on success, non-0 otherwise.
1470  * Assumes vcpu_load() was already called.
1471  */
1472 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1473 {
1474         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
1475 }
1476
1477 void kvm_resched(struct kvm_vcpu *vcpu)
1478 {
1479         if (!need_resched())
1480                 return;
1481         cond_resched();
1482 }
1483 EXPORT_SYMBOL_GPL(kvm_resched);
1484
1485 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1486 {
1487         int i;
1488         u32 function;
1489         struct kvm_cpuid_entry *e, *best;
1490
1491         kvm_x86_ops->cache_regs(vcpu);
1492         function = vcpu->regs[VCPU_REGS_RAX];
1493         vcpu->regs[VCPU_REGS_RAX] = 0;
1494         vcpu->regs[VCPU_REGS_RBX] = 0;
1495         vcpu->regs[VCPU_REGS_RCX] = 0;
1496         vcpu->regs[VCPU_REGS_RDX] = 0;
1497         best = NULL;
1498         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1499                 e = &vcpu->cpuid_entries[i];
1500                 if (e->function == function) {
1501                         best = e;
1502                         break;
1503                 }
1504                 /*
1505                  * Both basic or both extended?
1506                  */
1507                 if (((e->function ^ function) & 0x80000000) == 0)
1508                         if (!best || e->function > best->function)
1509                                 best = e;
1510         }
1511         if (best) {
1512                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1513                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1514                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1515                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1516         }
1517         kvm_x86_ops->decache_regs(vcpu);
1518         kvm_x86_ops->skip_emulated_instruction(vcpu);
1519 }
1520 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1521
1522 static int pio_copy_data(struct kvm_vcpu *vcpu)
1523 {
1524         void *p = vcpu->pio_data;
1525         void *q;
1526         unsigned bytes;
1527         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1528
1529         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1530                  PAGE_KERNEL);
1531         if (!q) {
1532                 free_pio_guest_pages(vcpu);
1533                 return -ENOMEM;
1534         }
1535         q += vcpu->pio.guest_page_offset;
1536         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1537         if (vcpu->pio.in)
1538                 memcpy(q, p, bytes);
1539         else
1540                 memcpy(p, q, bytes);
1541         q -= vcpu->pio.guest_page_offset;
1542         vunmap(q);
1543         free_pio_guest_pages(vcpu);
1544         return 0;
1545 }
1546
1547 static int complete_pio(struct kvm_vcpu *vcpu)
1548 {
1549         struct kvm_pio_request *io = &vcpu->pio;
1550         long delta;
1551         int r;
1552
1553         kvm_x86_ops->cache_regs(vcpu);
1554
1555         if (!io->string) {
1556                 if (io->in)
1557                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1558                                io->size);
1559         } else {
1560                 if (io->in) {
1561                         r = pio_copy_data(vcpu);
1562                         if (r) {
1563                                 kvm_x86_ops->cache_regs(vcpu);
1564                                 return r;
1565                         }
1566                 }
1567
1568                 delta = 1;
1569                 if (io->rep) {
1570                         delta *= io->cur_count;
1571                         /*
1572                          * The size of the register should really depend on
1573                          * current address size.
1574                          */
1575                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1576                 }
1577                 if (io->down)
1578                         delta = -delta;
1579                 delta *= io->size;
1580                 if (io->in)
1581                         vcpu->regs[VCPU_REGS_RDI] += delta;
1582                 else
1583                         vcpu->regs[VCPU_REGS_RSI] += delta;
1584         }
1585
1586         kvm_x86_ops->decache_regs(vcpu);
1587
1588         io->count -= io->cur_count;
1589         io->cur_count = 0;
1590
1591         return 0;
1592 }
1593
1594 static void kernel_pio(struct kvm_io_device *pio_dev,
1595                        struct kvm_vcpu *vcpu,
1596                        void *pd)
1597 {
1598         /* TODO: String I/O for in kernel device */
1599
1600         mutex_lock(&vcpu->kvm->lock);
1601         if (vcpu->pio.in)
1602                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1603                                   vcpu->pio.size,
1604                                   pd);
1605         else
1606                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1607                                    vcpu->pio.size,
1608                                    pd);
1609         mutex_unlock(&vcpu->kvm->lock);
1610 }
1611
1612 static void pio_string_write(struct kvm_io_device *pio_dev,
1613                              struct kvm_vcpu *vcpu)
1614 {
1615         struct kvm_pio_request *io = &vcpu->pio;
1616         void *pd = vcpu->pio_data;
1617         int i;
1618
1619         mutex_lock(&vcpu->kvm->lock);
1620         for (i = 0; i < io->cur_count; i++) {
1621                 kvm_iodevice_write(pio_dev, io->port,
1622                                    io->size,
1623                                    pd);
1624                 pd += io->size;
1625         }
1626         mutex_unlock(&vcpu->kvm->lock);
1627 }
1628
1629 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1630                   int size, unsigned port)
1631 {
1632         struct kvm_io_device *pio_dev;
1633
1634         vcpu->run->exit_reason = KVM_EXIT_IO;
1635         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1636         vcpu->run->io.size = vcpu->pio.size = size;
1637         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1638         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1639         vcpu->run->io.port = vcpu->pio.port = port;
1640         vcpu->pio.in = in;
1641         vcpu->pio.string = 0;
1642         vcpu->pio.down = 0;
1643         vcpu->pio.guest_page_offset = 0;
1644         vcpu->pio.rep = 0;
1645
1646         kvm_x86_ops->cache_regs(vcpu);
1647         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1648         kvm_x86_ops->decache_regs(vcpu);
1649
1650         kvm_x86_ops->skip_emulated_instruction(vcpu);
1651
1652         pio_dev = vcpu_find_pio_dev(vcpu, port);
1653         if (pio_dev) {
1654                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1655                 complete_pio(vcpu);
1656                 return 1;
1657         }
1658         return 0;
1659 }
1660 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1661
1662 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1663                   int size, unsigned long count, int down,
1664                   gva_t address, int rep, unsigned port)
1665 {
1666         unsigned now, in_page;
1667         int i, ret = 0;
1668         int nr_pages = 1;
1669         struct page *page;
1670         struct kvm_io_device *pio_dev;
1671
1672         vcpu->run->exit_reason = KVM_EXIT_IO;
1673         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1674         vcpu->run->io.size = vcpu->pio.size = size;
1675         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1676         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1677         vcpu->run->io.port = vcpu->pio.port = port;
1678         vcpu->pio.in = in;
1679         vcpu->pio.string = 1;
1680         vcpu->pio.down = down;
1681         vcpu->pio.guest_page_offset = offset_in_page(address);
1682         vcpu->pio.rep = rep;
1683
1684         if (!count) {
1685                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1686                 return 1;
1687         }
1688
1689         if (!down)
1690                 in_page = PAGE_SIZE - offset_in_page(address);
1691         else
1692                 in_page = offset_in_page(address) + size;
1693         now = min(count, (unsigned long)in_page / size);
1694         if (!now) {
1695                 /*
1696                  * String I/O straddles page boundary.  Pin two guest pages
1697                  * so that we satisfy atomicity constraints.  Do just one
1698                  * transaction to avoid complexity.
1699                  */
1700                 nr_pages = 2;
1701                 now = 1;
1702         }
1703         if (down) {
1704                 /*
1705                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1706                  */
1707                 pr_unimpl(vcpu, "guest string pio down\n");
1708                 inject_gp(vcpu);
1709                 return 1;
1710         }
1711         vcpu->run->io.count = now;
1712         vcpu->pio.cur_count = now;
1713
1714         if (vcpu->pio.cur_count == vcpu->pio.count)
1715                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1716
1717         for (i = 0; i < nr_pages; ++i) {
1718                 mutex_lock(&vcpu->kvm->lock);
1719                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1720                 vcpu->pio.guest_pages[i] = page;
1721                 mutex_unlock(&vcpu->kvm->lock);
1722                 if (!page) {
1723                         inject_gp(vcpu);
1724                         free_pio_guest_pages(vcpu);
1725                         return 1;
1726                 }
1727         }
1728
1729         pio_dev = vcpu_find_pio_dev(vcpu, port);
1730         if (!vcpu->pio.in) {
1731                 /* string PIO write */
1732                 ret = pio_copy_data(vcpu);
1733                 if (ret >= 0 && pio_dev) {
1734                         pio_string_write(pio_dev, vcpu);
1735                         complete_pio(vcpu);
1736                         if (vcpu->pio.count == 0)
1737                                 ret = 1;
1738                 }
1739         } else if (pio_dev)
1740                 pr_unimpl(vcpu, "no string pio read support yet, "
1741                        "port %x size %d count %ld\n",
1742                         port, size, count);
1743
1744         return ret;
1745 }
1746 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1747
1748 /*
1749  * Check if userspace requested an interrupt window, and that the
1750  * interrupt window is open.
1751  *
1752  * No need to exit to userspace if we already have an interrupt queued.
1753  */
1754 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1755                                           struct kvm_run *kvm_run)
1756 {
1757         return (!vcpu->irq_summary &&
1758                 kvm_run->request_interrupt_window &&
1759                 vcpu->interrupt_window_open &&
1760                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1761 }
1762
1763 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1764                               struct kvm_run *kvm_run)
1765 {
1766         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1767         kvm_run->cr8 = get_cr8(vcpu);
1768         kvm_run->apic_base = kvm_get_apic_base(vcpu);
1769         if (irqchip_in_kernel(vcpu->kvm))
1770                 kvm_run->ready_for_interrupt_injection = 1;
1771         else
1772                 kvm_run->ready_for_interrupt_injection =
1773                                         (vcpu->interrupt_window_open &&
1774                                          vcpu->irq_summary == 0);
1775 }
1776
1777 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1778 {
1779         int r;
1780
1781         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1782                 pr_debug("vcpu %d received sipi with vector # %x\n",
1783                        vcpu->vcpu_id, vcpu->sipi_vector);
1784                 kvm_lapic_reset(vcpu);
1785                 r = kvm_x86_ops->vcpu_reset(vcpu);
1786                 if (r)
1787                         return r;
1788                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
1789         }
1790
1791 preempted:
1792         if (vcpu->guest_debug.enabled)
1793                 kvm_x86_ops->guest_debug_pre(vcpu);
1794
1795 again:
1796         r = kvm_mmu_reload(vcpu);
1797         if (unlikely(r))
1798                 goto out;
1799
1800         kvm_inject_pending_timer_irqs(vcpu);
1801
1802         preempt_disable();
1803
1804         kvm_x86_ops->prepare_guest_switch(vcpu);
1805         kvm_load_guest_fpu(vcpu);
1806
1807         local_irq_disable();
1808
1809         if (signal_pending(current)) {
1810                 local_irq_enable();
1811                 preempt_enable();
1812                 r = -EINTR;
1813                 kvm_run->exit_reason = KVM_EXIT_INTR;
1814                 ++vcpu->stat.signal_exits;
1815                 goto out;
1816         }
1817
1818         if (irqchip_in_kernel(vcpu->kvm))
1819                 kvm_x86_ops->inject_pending_irq(vcpu);
1820         else if (!vcpu->mmio_read_completed)
1821                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
1822
1823         vcpu->guest_mode = 1;
1824         kvm_guest_enter();
1825
1826         if (vcpu->requests)
1827                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
1828                         kvm_x86_ops->tlb_flush(vcpu);
1829
1830         kvm_x86_ops->run(vcpu, kvm_run);
1831
1832         vcpu->guest_mode = 0;
1833         local_irq_enable();
1834
1835         ++vcpu->stat.exits;
1836
1837         /*
1838          * We must have an instruction between local_irq_enable() and
1839          * kvm_guest_exit(), so the timer interrupt isn't delayed by
1840          * the interrupt shadow.  The stat.exits increment will do nicely.
1841          * But we need to prevent reordering, hence this barrier():
1842          */
1843         barrier();
1844
1845         kvm_guest_exit();
1846
1847         preempt_enable();
1848
1849         /*
1850          * Profile KVM exit RIPs:
1851          */
1852         if (unlikely(prof_on == KVM_PROFILING)) {
1853                 kvm_x86_ops->cache_regs(vcpu);
1854                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
1855         }
1856
1857         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
1858
1859         if (r > 0) {
1860                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1861                         r = -EINTR;
1862                         kvm_run->exit_reason = KVM_EXIT_INTR;
1863                         ++vcpu->stat.request_irq_exits;
1864                         goto out;
1865                 }
1866                 if (!need_resched()) {
1867                         ++vcpu->stat.light_exits;
1868                         goto again;
1869                 }
1870         }
1871
1872 out:
1873         if (r > 0) {
1874                 kvm_resched(vcpu);
1875                 goto preempted;
1876         }
1877
1878         post_kvm_run_save(vcpu, kvm_run);
1879
1880         return r;
1881 }
1882
1883
1884 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1885 {
1886         int r;
1887         sigset_t sigsaved;
1888
1889         vcpu_load(vcpu);
1890
1891         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
1892                 kvm_vcpu_block(vcpu);
1893                 vcpu_put(vcpu);
1894                 return -EAGAIN;
1895         }
1896
1897         if (vcpu->sigset_active)
1898                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1899
1900         /* re-sync apic's tpr */
1901         if (!irqchip_in_kernel(vcpu->kvm))
1902                 set_cr8(vcpu, kvm_run->cr8);
1903
1904         if (vcpu->pio.cur_count) {
1905                 r = complete_pio(vcpu);
1906                 if (r)
1907                         goto out;
1908         }
1909 #if CONFIG_HAS_IOMEM
1910         if (vcpu->mmio_needed) {
1911                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1912                 vcpu->mmio_read_completed = 1;
1913                 vcpu->mmio_needed = 0;
1914                 r = emulate_instruction(vcpu, kvm_run,
1915                                         vcpu->mmio_fault_cr2, 0, 1);
1916                 if (r == EMULATE_DO_MMIO) {
1917                         /*
1918                          * Read-modify-write.  Back to userspace.
1919                          */
1920                         r = 0;
1921                         goto out;
1922                 }
1923         }
1924 #endif
1925         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1926                 kvm_x86_ops->cache_regs(vcpu);
1927                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1928                 kvm_x86_ops->decache_regs(vcpu);
1929         }
1930
1931         r = __vcpu_run(vcpu, kvm_run);
1932
1933 out:
1934         if (vcpu->sigset_active)
1935                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1936
1937         vcpu_put(vcpu);
1938         return r;
1939 }
1940
1941 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1942                                    struct kvm_regs *regs)
1943 {
1944         vcpu_load(vcpu);
1945
1946         kvm_x86_ops->cache_regs(vcpu);
1947
1948         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1949         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1950         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1951         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1952         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1953         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1954         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1955         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1956 #ifdef CONFIG_X86_64
1957         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1958         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1959         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1960         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1961         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1962         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1963         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1964         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1965 #endif
1966
1967         regs->rip = vcpu->rip;
1968         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
1969
1970         /*
1971          * Don't leak debug flags in case they were set for guest debugging
1972          */
1973         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1974                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1975
1976         vcpu_put(vcpu);
1977
1978         return 0;
1979 }
1980
1981 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1982                                    struct kvm_regs *regs)
1983 {
1984         vcpu_load(vcpu);
1985
1986         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1987         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1988         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1989         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1990         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1991         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1992         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1993         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1994 #ifdef CONFIG_X86_64
1995         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1996         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1997         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1998         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1999         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2000         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2001         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2002         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2003 #endif
2004
2005         vcpu->rip = regs->rip;
2006         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2007
2008         kvm_x86_ops->decache_regs(vcpu);
2009
2010         vcpu_put(vcpu);
2011
2012         return 0;
2013 }
2014
2015 static void get_segment(struct kvm_vcpu *vcpu,
2016                         struct kvm_segment *var, int seg)
2017 {
2018         return kvm_x86_ops->get_segment(vcpu, var, seg);
2019 }
2020
2021 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2022                                     struct kvm_sregs *sregs)
2023 {
2024         struct descriptor_table dt;
2025         int pending_vec;
2026
2027         vcpu_load(vcpu);
2028
2029         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2030         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2031         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2032         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2033         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2034         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2035
2036         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2037         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2038
2039         kvm_x86_ops->get_idt(vcpu, &dt);
2040         sregs->idt.limit = dt.limit;
2041         sregs->idt.base = dt.base;
2042         kvm_x86_ops->get_gdt(vcpu, &dt);
2043         sregs->gdt.limit = dt.limit;
2044         sregs->gdt.base = dt.base;
2045
2046         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2047         sregs->cr0 = vcpu->cr0;
2048         sregs->cr2 = vcpu->cr2;
2049         sregs->cr3 = vcpu->cr3;
2050         sregs->cr4 = vcpu->cr4;
2051         sregs->cr8 = get_cr8(vcpu);
2052         sregs->efer = vcpu->shadow_efer;
2053         sregs->apic_base = kvm_get_apic_base(vcpu);
2054
2055         if (irqchip_in_kernel(vcpu->kvm)) {
2056                 memset(sregs->interrupt_bitmap, 0,
2057                        sizeof sregs->interrupt_bitmap);
2058                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2059                 if (pending_vec >= 0)
2060                         set_bit(pending_vec,
2061                                 (unsigned long *)sregs->interrupt_bitmap);
2062         } else
2063                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2064                        sizeof sregs->interrupt_bitmap);
2065
2066         vcpu_put(vcpu);
2067
2068         return 0;
2069 }
2070
2071 static void set_segment(struct kvm_vcpu *vcpu,
2072                         struct kvm_segment *var, int seg)
2073 {
2074         return kvm_x86_ops->set_segment(vcpu, var, seg);
2075 }
2076
2077 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2078                                     struct kvm_sregs *sregs)
2079 {
2080         int mmu_reset_needed = 0;
2081         int i, pending_vec, max_bits;
2082         struct descriptor_table dt;
2083
2084         vcpu_load(vcpu);
2085
2086         dt.limit = sregs->idt.limit;
2087         dt.base = sregs->idt.base;
2088         kvm_x86_ops->set_idt(vcpu, &dt);
2089         dt.limit = sregs->gdt.limit;
2090         dt.base = sregs->gdt.base;
2091         kvm_x86_ops->set_gdt(vcpu, &dt);
2092
2093         vcpu->cr2 = sregs->cr2;
2094         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2095         vcpu->cr3 = sregs->cr3;
2096
2097         set_cr8(vcpu, sregs->cr8);
2098
2099         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2100 #ifdef CONFIG_X86_64
2101         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2102 #endif
2103         kvm_set_apic_base(vcpu, sregs->apic_base);
2104
2105         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2106
2107         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2108         vcpu->cr0 = sregs->cr0;
2109         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2110
2111         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2112         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2113         if (!is_long_mode(vcpu) && is_pae(vcpu))
2114                 load_pdptrs(vcpu, vcpu->cr3);
2115
2116         if (mmu_reset_needed)
2117                 kvm_mmu_reset_context(vcpu);
2118
2119         if (!irqchip_in_kernel(vcpu->kvm)) {
2120                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2121                        sizeof vcpu->irq_pending);
2122                 vcpu->irq_summary = 0;
2123                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2124                         if (vcpu->irq_pending[i])
2125                                 __set_bit(i, &vcpu->irq_summary);
2126         } else {
2127                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2128                 pending_vec = find_first_bit(
2129                         (const unsigned long *)sregs->interrupt_bitmap,
2130                         max_bits);
2131                 /* Only pending external irq is handled here */
2132                 if (pending_vec < max_bits) {
2133                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2134                         pr_debug("Set back pending irq %d\n",
2135                                  pending_vec);
2136                 }
2137         }
2138
2139         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2140         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2141         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2142         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2143         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2144         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2145
2146         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2147         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2148
2149         vcpu_put(vcpu);
2150
2151         return 0;
2152 }
2153
2154 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2155 {
2156         struct kvm_segment cs;
2157
2158         get_segment(vcpu, &cs, VCPU_SREG_CS);
2159         *db = cs.db;
2160         *l = cs.l;
2161 }
2162 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2163
2164 /*
2165  * Translate a guest virtual address to a guest physical address.
2166  */
2167 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2168                                     struct kvm_translation *tr)
2169 {
2170         unsigned long vaddr = tr->linear_address;
2171         gpa_t gpa;
2172
2173         vcpu_load(vcpu);
2174         mutex_lock(&vcpu->kvm->lock);
2175         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2176         tr->physical_address = gpa;
2177         tr->valid = gpa != UNMAPPED_GVA;
2178         tr->writeable = 1;
2179         tr->usermode = 0;
2180         mutex_unlock(&vcpu->kvm->lock);
2181         vcpu_put(vcpu);
2182
2183         return 0;
2184 }
2185
2186 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2187                                     struct kvm_interrupt *irq)
2188 {
2189         if (irq->irq < 0 || irq->irq >= 256)
2190                 return -EINVAL;
2191         if (irqchip_in_kernel(vcpu->kvm))
2192                 return -ENXIO;
2193         vcpu_load(vcpu);
2194
2195         set_bit(irq->irq, vcpu->irq_pending);
2196         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2197
2198         vcpu_put(vcpu);
2199
2200         return 0;
2201 }
2202
2203 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2204                                       struct kvm_debug_guest *dbg)
2205 {
2206         int r;
2207
2208         vcpu_load(vcpu);
2209
2210         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2211
2212         vcpu_put(vcpu);
2213
2214         return r;
2215 }
2216
2217 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2218                                     unsigned long address,
2219                                     int *type)
2220 {
2221         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2222         unsigned long pgoff;
2223         struct page *page;
2224
2225         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2226         if (pgoff == 0)
2227                 page = virt_to_page(vcpu->run);
2228         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2229                 page = virt_to_page(vcpu->pio_data);
2230         else
2231                 return NOPAGE_SIGBUS;
2232         get_page(page);
2233         if (type != NULL)
2234                 *type = VM_FAULT_MINOR;
2235
2236         return page;
2237 }
2238
2239 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2240         .nopage = kvm_vcpu_nopage,
2241 };
2242
2243 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2244 {
2245         vma->vm_ops = &kvm_vcpu_vm_ops;
2246         return 0;
2247 }
2248
2249 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2250 {
2251         struct kvm_vcpu *vcpu = filp->private_data;
2252
2253         fput(vcpu->kvm->filp);
2254         return 0;
2255 }
2256
2257 static struct file_operations kvm_vcpu_fops = {
2258         .release        = kvm_vcpu_release,
2259         .unlocked_ioctl = kvm_vcpu_ioctl,
2260         .compat_ioctl   = kvm_vcpu_ioctl,
2261         .mmap           = kvm_vcpu_mmap,
2262 };
2263
2264 /*
2265  * Allocates an inode for the vcpu.
2266  */
2267 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2268 {
2269         int fd, r;
2270         struct inode *inode;
2271         struct file *file;
2272
2273         r = anon_inode_getfd(&fd, &inode, &file,
2274                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2275         if (r)
2276                 return r;
2277         atomic_inc(&vcpu->kvm->filp->f_count);
2278         return fd;
2279 }
2280
2281 /*
2282  * Creates some virtual cpus.  Good luck creating more than one.
2283  */
2284 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2285 {
2286         int r;
2287         struct kvm_vcpu *vcpu;
2288
2289         if (!valid_vcpu(n))
2290                 return -EINVAL;
2291
2292         vcpu = kvm_x86_ops->vcpu_create(kvm, n);
2293         if (IS_ERR(vcpu))
2294                 return PTR_ERR(vcpu);
2295
2296         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2297
2298         /* We do fxsave: this must be aligned. */
2299         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2300
2301         vcpu_load(vcpu);
2302         r = kvm_x86_ops->vcpu_reset(vcpu);
2303         if (r == 0)
2304                 r = kvm_mmu_setup(vcpu);
2305         vcpu_put(vcpu);
2306         if (r < 0)
2307                 goto free_vcpu;
2308
2309         mutex_lock(&kvm->lock);
2310         if (kvm->vcpus[n]) {
2311                 r = -EEXIST;
2312                 mutex_unlock(&kvm->lock);
2313                 goto mmu_unload;
2314         }
2315         kvm->vcpus[n] = vcpu;
2316         mutex_unlock(&kvm->lock);
2317
2318         /* Now it's all set up, let userspace reach it */
2319         r = create_vcpu_fd(vcpu);
2320         if (r < 0)
2321                 goto unlink;
2322         return r;
2323
2324 unlink:
2325         mutex_lock(&kvm->lock);
2326         kvm->vcpus[n] = NULL;
2327         mutex_unlock(&kvm->lock);
2328
2329 mmu_unload:
2330         vcpu_load(vcpu);
2331         kvm_mmu_unload(vcpu);
2332         vcpu_put(vcpu);
2333
2334 free_vcpu:
2335         kvm_x86_ops->vcpu_free(vcpu);
2336         return r;
2337 }
2338
2339 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2340 {
2341         if (sigset) {
2342                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2343                 vcpu->sigset_active = 1;
2344                 vcpu->sigset = *sigset;
2345         } else
2346                 vcpu->sigset_active = 0;
2347         return 0;
2348 }
2349
2350 /*
2351  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2352  * we have asm/x86/processor.h
2353  */
2354 struct fxsave {
2355         u16     cwd;
2356         u16     swd;
2357         u16     twd;
2358         u16     fop;
2359         u64     rip;
2360         u64     rdp;
2361         u32     mxcsr;
2362         u32     mxcsr_mask;
2363         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2364 #ifdef CONFIG_X86_64
2365         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2366 #else
2367         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2368 #endif
2369 };
2370
2371 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2372 {
2373         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2374
2375         vcpu_load(vcpu);
2376
2377         memcpy(fpu->fpr, fxsave->st_space, 128);
2378         fpu->fcw = fxsave->cwd;
2379         fpu->fsw = fxsave->swd;
2380         fpu->ftwx = fxsave->twd;
2381         fpu->last_opcode = fxsave->fop;
2382         fpu->last_ip = fxsave->rip;
2383         fpu->last_dp = fxsave->rdp;
2384         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2385
2386         vcpu_put(vcpu);
2387
2388         return 0;
2389 }
2390
2391 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2392 {
2393         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2394
2395         vcpu_load(vcpu);
2396
2397         memcpy(fxsave->st_space, fpu->fpr, 128);
2398         fxsave->cwd = fpu->fcw;
2399         fxsave->swd = fpu->fsw;
2400         fxsave->twd = fpu->ftwx;
2401         fxsave->fop = fpu->last_opcode;
2402         fxsave->rip = fpu->last_ip;
2403         fxsave->rdp = fpu->last_dp;
2404         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2405
2406         vcpu_put(vcpu);
2407
2408         return 0;
2409 }
2410
2411 static long kvm_vcpu_ioctl(struct file *filp,
2412                            unsigned int ioctl, unsigned long arg)
2413 {
2414         struct kvm_vcpu *vcpu = filp->private_data;
2415         void __user *argp = (void __user *)arg;
2416         int r;
2417
2418         switch (ioctl) {
2419         case KVM_RUN:
2420                 r = -EINVAL;
2421                 if (arg)
2422                         goto out;
2423                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2424                 break;
2425         case KVM_GET_REGS: {
2426                 struct kvm_regs kvm_regs;
2427
2428                 memset(&kvm_regs, 0, sizeof kvm_regs);
2429                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2430                 if (r)
2431                         goto out;
2432                 r = -EFAULT;
2433                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2434                         goto out;
2435                 r = 0;
2436                 break;
2437         }
2438         case KVM_SET_REGS: {
2439                 struct kvm_regs kvm_regs;
2440
2441                 r = -EFAULT;
2442                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2443                         goto out;
2444                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2445                 if (r)
2446                         goto out;
2447                 r = 0;
2448                 break;
2449         }
2450         case KVM_GET_SREGS: {
2451                 struct kvm_sregs kvm_sregs;
2452
2453                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2454                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2455                 if (r)
2456                         goto out;
2457                 r = -EFAULT;
2458                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2459                         goto out;
2460                 r = 0;
2461                 break;
2462         }
2463         case KVM_SET_SREGS: {
2464                 struct kvm_sregs kvm_sregs;
2465
2466                 r = -EFAULT;
2467                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2468                         goto out;
2469                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2470                 if (r)
2471                         goto out;
2472                 r = 0;
2473                 break;
2474         }
2475         case KVM_TRANSLATE: {
2476                 struct kvm_translation tr;
2477
2478                 r = -EFAULT;
2479                 if (copy_from_user(&tr, argp, sizeof tr))
2480                         goto out;
2481                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2482                 if (r)
2483                         goto out;
2484                 r = -EFAULT;
2485                 if (copy_to_user(argp, &tr, sizeof tr))
2486                         goto out;
2487                 r = 0;
2488                 break;
2489         }
2490         case KVM_INTERRUPT: {
2491                 struct kvm_interrupt irq;
2492
2493                 r = -EFAULT;
2494                 if (copy_from_user(&irq, argp, sizeof irq))
2495                         goto out;
2496                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2497                 if (r)
2498                         goto out;
2499                 r = 0;
2500                 break;
2501         }
2502         case KVM_DEBUG_GUEST: {
2503                 struct kvm_debug_guest dbg;
2504
2505                 r = -EFAULT;
2506                 if (copy_from_user(&dbg, argp, sizeof dbg))
2507                         goto out;
2508                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2509                 if (r)
2510                         goto out;
2511                 r = 0;
2512                 break;
2513         }
2514         case KVM_SET_SIGNAL_MASK: {
2515                 struct kvm_signal_mask __user *sigmask_arg = argp;
2516                 struct kvm_signal_mask kvm_sigmask;
2517                 sigset_t sigset, *p;
2518
2519                 p = NULL;
2520                 if (argp) {
2521                         r = -EFAULT;
2522                         if (copy_from_user(&kvm_sigmask, argp,
2523                                            sizeof kvm_sigmask))
2524                                 goto out;
2525                         r = -EINVAL;
2526                         if (kvm_sigmask.len != sizeof sigset)
2527                                 goto out;
2528                         r = -EFAULT;
2529                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2530                                            sizeof sigset))
2531                                 goto out;
2532                         p = &sigset;
2533                 }
2534                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2535                 break;
2536         }
2537         case KVM_GET_FPU: {
2538                 struct kvm_fpu fpu;
2539
2540                 memset(&fpu, 0, sizeof fpu);
2541                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2542                 if (r)
2543                         goto out;
2544                 r = -EFAULT;
2545                 if (copy_to_user(argp, &fpu, sizeof fpu))
2546                         goto out;
2547                 r = 0;
2548                 break;
2549         }
2550         case KVM_SET_FPU: {
2551                 struct kvm_fpu fpu;
2552
2553                 r = -EFAULT;
2554                 if (copy_from_user(&fpu, argp, sizeof fpu))
2555                         goto out;
2556                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2557                 if (r)
2558                         goto out;
2559                 r = 0;
2560                 break;
2561         }
2562         default:
2563                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2564         }
2565 out:
2566         return r;
2567 }
2568
2569 static long kvm_vm_ioctl(struct file *filp,
2570                            unsigned int ioctl, unsigned long arg)
2571 {
2572         struct kvm *kvm = filp->private_data;
2573         void __user *argp = (void __user *)arg;
2574         int r;
2575
2576         switch (ioctl) {
2577         case KVM_CREATE_VCPU:
2578                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2579                 if (r < 0)
2580                         goto out;
2581                 break;
2582         case KVM_SET_USER_MEMORY_REGION: {
2583                 struct kvm_userspace_memory_region kvm_userspace_mem;
2584
2585                 r = -EFAULT;
2586                 if (copy_from_user(&kvm_userspace_mem, argp,
2587                                                 sizeof kvm_userspace_mem))
2588                         goto out;
2589
2590                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2591                 if (r)
2592                         goto out;
2593                 break;
2594         }
2595         case KVM_GET_DIRTY_LOG: {
2596                 struct kvm_dirty_log log;
2597
2598                 r = -EFAULT;
2599                 if (copy_from_user(&log, argp, sizeof log))
2600                         goto out;
2601                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2602                 if (r)
2603                         goto out;
2604                 break;
2605         }
2606         default:
2607                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2608         }
2609 out:
2610         return r;
2611 }
2612
2613 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2614                                   unsigned long address,
2615                                   int *type)
2616 {
2617         struct kvm *kvm = vma->vm_file->private_data;
2618         unsigned long pgoff;
2619         struct page *page;
2620
2621         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2622         if (!kvm_is_visible_gfn(kvm, pgoff))
2623                 return NOPAGE_SIGBUS;
2624         page = gfn_to_page(kvm, pgoff);
2625         if (is_error_page(page)) {
2626                 kvm_release_page(page);
2627                 return NOPAGE_SIGBUS;
2628         }
2629         if (type != NULL)
2630                 *type = VM_FAULT_MINOR;
2631
2632         return page;
2633 }
2634
2635 static struct vm_operations_struct kvm_vm_vm_ops = {
2636         .nopage = kvm_vm_nopage,
2637 };
2638
2639 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2640 {
2641         vma->vm_ops = &kvm_vm_vm_ops;
2642         return 0;
2643 }
2644
2645 static struct file_operations kvm_vm_fops = {
2646         .release        = kvm_vm_release,
2647         .unlocked_ioctl = kvm_vm_ioctl,
2648         .compat_ioctl   = kvm_vm_ioctl,
2649         .mmap           = kvm_vm_mmap,
2650 };
2651
2652 static int kvm_dev_ioctl_create_vm(void)
2653 {
2654         int fd, r;
2655         struct inode *inode;
2656         struct file *file;
2657         struct kvm *kvm;
2658
2659         kvm = kvm_create_vm();
2660         if (IS_ERR(kvm))
2661                 return PTR_ERR(kvm);
2662         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2663         if (r) {
2664                 kvm_destroy_vm(kvm);
2665                 return r;
2666         }
2667
2668         kvm->filp = file;
2669
2670         return fd;
2671 }
2672
2673 static long kvm_dev_ioctl(struct file *filp,
2674                           unsigned int ioctl, unsigned long arg)
2675 {
2676         void __user *argp = (void __user *)arg;
2677         long r = -EINVAL;
2678
2679         switch (ioctl) {
2680         case KVM_GET_API_VERSION:
2681                 r = -EINVAL;
2682                 if (arg)
2683                         goto out;
2684                 r = KVM_API_VERSION;
2685                 break;
2686         case KVM_CREATE_VM:
2687                 r = -EINVAL;
2688                 if (arg)
2689                         goto out;
2690                 r = kvm_dev_ioctl_create_vm();
2691                 break;
2692         case KVM_CHECK_EXTENSION: {
2693                 int ext = (long)argp;
2694
2695                 switch (ext) {
2696                 case KVM_CAP_IRQCHIP:
2697                 case KVM_CAP_HLT:
2698                 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2699                 case KVM_CAP_USER_MEMORY:
2700                 case KVM_CAP_SET_TSS_ADDR:
2701                         r = 1;
2702                         break;
2703                 default:
2704                         r = 0;
2705                         break;
2706                 }
2707                 break;
2708         }
2709         case KVM_GET_VCPU_MMAP_SIZE:
2710                 r = -EINVAL;
2711                 if (arg)
2712                         goto out;
2713                 r = 2 * PAGE_SIZE;
2714                 break;
2715         default:
2716                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2717         }
2718 out:
2719         return r;
2720 }
2721
2722 static struct file_operations kvm_chardev_ops = {
2723         .unlocked_ioctl = kvm_dev_ioctl,
2724         .compat_ioctl   = kvm_dev_ioctl,
2725 };
2726
2727 static struct miscdevice kvm_dev = {
2728         KVM_MINOR,
2729         "kvm",
2730         &kvm_chardev_ops,
2731 };
2732
2733 /*
2734  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2735  * cached on it.
2736  */
2737 static void decache_vcpus_on_cpu(int cpu)
2738 {
2739         struct kvm *vm;
2740         struct kvm_vcpu *vcpu;
2741         int i;
2742
2743         spin_lock(&kvm_lock);
2744         list_for_each_entry(vm, &vm_list, vm_list)
2745                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2746                         vcpu = vm->vcpus[i];
2747                         if (!vcpu)
2748                                 continue;
2749                         /*
2750                          * If the vcpu is locked, then it is running on some
2751                          * other cpu and therefore it is not cached on the
2752                          * cpu in question.
2753                          *
2754                          * If it's not locked, check the last cpu it executed
2755                          * on.
2756                          */
2757                         if (mutex_trylock(&vcpu->mutex)) {
2758                                 if (vcpu->cpu == cpu) {
2759                                         kvm_x86_ops->vcpu_decache(vcpu);
2760                                         vcpu->cpu = -1;
2761                                 }
2762                                 mutex_unlock(&vcpu->mutex);
2763                         }
2764                 }
2765         spin_unlock(&kvm_lock);
2766 }
2767
2768 static void hardware_enable(void *junk)
2769 {
2770         int cpu = raw_smp_processor_id();
2771
2772         if (cpu_isset(cpu, cpus_hardware_enabled))
2773                 return;
2774         cpu_set(cpu, cpus_hardware_enabled);
2775         kvm_x86_ops->hardware_enable(NULL);
2776 }
2777
2778 static void hardware_disable(void *junk)
2779 {
2780         int cpu = raw_smp_processor_id();
2781
2782         if (!cpu_isset(cpu, cpus_hardware_enabled))
2783                 return;
2784         cpu_clear(cpu, cpus_hardware_enabled);
2785         decache_vcpus_on_cpu(cpu);
2786         kvm_x86_ops->hardware_disable(NULL);
2787 }
2788
2789 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2790                            void *v)
2791 {
2792         int cpu = (long)v;
2793
2794         switch (val) {
2795         case CPU_DYING:
2796         case CPU_DYING_FROZEN:
2797                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2798                        cpu);
2799                 hardware_disable(NULL);
2800                 break;
2801         case CPU_UP_CANCELED:
2802         case CPU_UP_CANCELED_FROZEN:
2803                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2804                        cpu);
2805                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
2806                 break;
2807         case CPU_ONLINE:
2808         case CPU_ONLINE_FROZEN:
2809                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2810                        cpu);
2811                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
2812                 break;
2813         }
2814         return NOTIFY_OK;
2815 }
2816
2817 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2818                       void *v)
2819 {
2820         if (val == SYS_RESTART) {
2821                 /*
2822                  * Some (well, at least mine) BIOSes hang on reboot if
2823                  * in vmx root mode.
2824                  */
2825                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2826                 on_each_cpu(hardware_disable, NULL, 0, 1);
2827         }
2828         return NOTIFY_OK;
2829 }
2830
2831 static struct notifier_block kvm_reboot_notifier = {
2832         .notifier_call = kvm_reboot,
2833         .priority = 0,
2834 };
2835
2836 void kvm_io_bus_init(struct kvm_io_bus *bus)
2837 {
2838         memset(bus, 0, sizeof(*bus));
2839 }
2840
2841 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2842 {
2843         int i;
2844
2845         for (i = 0; i < bus->dev_count; i++) {
2846                 struct kvm_io_device *pos = bus->devs[i];
2847
2848                 kvm_iodevice_destructor(pos);
2849         }
2850 }
2851
2852 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
2853 {
2854         int i;
2855
2856         for (i = 0; i < bus->dev_count; i++) {
2857                 struct kvm_io_device *pos = bus->devs[i];
2858
2859                 if (pos->in_range(pos, addr))
2860                         return pos;
2861         }
2862
2863         return NULL;
2864 }
2865
2866 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
2867 {
2868         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
2869
2870         bus->devs[bus->dev_count++] = dev;
2871 }
2872
2873 static struct notifier_block kvm_cpu_notifier = {
2874         .notifier_call = kvm_cpu_hotplug,
2875         .priority = 20, /* must be > scheduler priority */
2876 };
2877
2878 static u64 stat_get(void *_offset)
2879 {
2880         unsigned offset = (long)_offset;
2881         u64 total = 0;
2882         struct kvm *kvm;
2883         struct kvm_vcpu *vcpu;
2884         int i;
2885
2886         spin_lock(&kvm_lock);
2887         list_for_each_entry(kvm, &vm_list, vm_list)
2888                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2889                         vcpu = kvm->vcpus[i];
2890                         if (vcpu)
2891                                 total += *(u32 *)((void *)vcpu + offset);
2892                 }
2893         spin_unlock(&kvm_lock);
2894         return total;
2895 }
2896
2897 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
2898
2899 static __init void kvm_init_debug(void)
2900 {
2901         struct kvm_stats_debugfs_item *p;
2902
2903         debugfs_dir = debugfs_create_dir("kvm", NULL);
2904         for (p = debugfs_entries; p->name; ++p)
2905                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
2906                                                 (void *)(long)p->offset,
2907                                                 &stat_fops);
2908 }
2909
2910 static void kvm_exit_debug(void)
2911 {
2912         struct kvm_stats_debugfs_item *p;
2913
2914         for (p = debugfs_entries; p->name; ++p)
2915                 debugfs_remove(p->dentry);
2916         debugfs_remove(debugfs_dir);
2917 }
2918
2919 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2920 {
2921         hardware_disable(NULL);
2922         return 0;
2923 }
2924
2925 static int kvm_resume(struct sys_device *dev)
2926 {
2927         hardware_enable(NULL);
2928         return 0;
2929 }
2930
2931 static struct sysdev_class kvm_sysdev_class = {
2932         .name = "kvm",
2933         .suspend = kvm_suspend,
2934         .resume = kvm_resume,
2935 };
2936
2937 static struct sys_device kvm_sysdev = {
2938         .id = 0,
2939         .cls = &kvm_sysdev_class,
2940 };
2941
2942 struct page *bad_page;
2943
2944 static inline
2945 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2946 {
2947         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2948 }
2949
2950 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2951 {
2952         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2953
2954         kvm_x86_ops->vcpu_load(vcpu, cpu);
2955 }
2956
2957 static void kvm_sched_out(struct preempt_notifier *pn,
2958                           struct task_struct *next)
2959 {
2960         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2961
2962         kvm_x86_ops->vcpu_put(vcpu);
2963 }
2964
2965 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
2966                   struct module *module)
2967 {
2968         int r;
2969         int cpu;
2970
2971         if (kvm_x86_ops) {
2972                 printk(KERN_ERR "kvm: already loaded the other module\n");
2973                 return -EEXIST;
2974         }
2975
2976         if (!ops->cpu_has_kvm_support()) {
2977                 printk(KERN_ERR "kvm: no hardware support\n");
2978                 return -EOPNOTSUPP;
2979         }
2980         if (ops->disabled_by_bios()) {
2981                 printk(KERN_ERR "kvm: disabled by bios\n");
2982                 return -EOPNOTSUPP;
2983         }
2984
2985         kvm_x86_ops = ops;
2986
2987         r = kvm_x86_ops->hardware_setup();
2988         if (r < 0)
2989                 goto out;
2990
2991         for_each_online_cpu(cpu) {
2992                 smp_call_function_single(cpu,
2993                                 kvm_x86_ops->check_processor_compatibility,
2994                                 &r, 0, 1);
2995                 if (r < 0)
2996                         goto out_free_0;
2997         }
2998
2999         on_each_cpu(hardware_enable, NULL, 0, 1);
3000         r = register_cpu_notifier(&kvm_cpu_notifier);
3001         if (r)
3002                 goto out_free_1;
3003         register_reboot_notifier(&kvm_reboot_notifier);
3004
3005         r = sysdev_class_register(&kvm_sysdev_class);
3006         if (r)
3007                 goto out_free_2;
3008
3009         r = sysdev_register(&kvm_sysdev);
3010         if (r)
3011                 goto out_free_3;
3012
3013         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3014         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3015                                            __alignof__(struct kvm_vcpu), 0, 0);
3016         if (!kvm_vcpu_cache) {
3017                 r = -ENOMEM;
3018                 goto out_free_4;
3019         }
3020
3021         kvm_chardev_ops.owner = module;
3022
3023         r = misc_register(&kvm_dev);
3024         if (r) {
3025                 printk(KERN_ERR "kvm: misc device register failed\n");
3026                 goto out_free;
3027         }
3028
3029         kvm_preempt_ops.sched_in = kvm_sched_in;
3030         kvm_preempt_ops.sched_out = kvm_sched_out;
3031
3032         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3033
3034         return 0;
3035
3036 out_free:
3037         kmem_cache_destroy(kvm_vcpu_cache);
3038 out_free_4:
3039         sysdev_unregister(&kvm_sysdev);
3040 out_free_3:
3041         sysdev_class_unregister(&kvm_sysdev_class);
3042 out_free_2:
3043         unregister_reboot_notifier(&kvm_reboot_notifier);
3044         unregister_cpu_notifier(&kvm_cpu_notifier);
3045 out_free_1:
3046         on_each_cpu(hardware_disable, NULL, 0, 1);
3047 out_free_0:
3048         kvm_x86_ops->hardware_unsetup();
3049 out:
3050         kvm_x86_ops = NULL;
3051         return r;
3052 }
3053 EXPORT_SYMBOL_GPL(kvm_init_x86);
3054
3055 void kvm_exit_x86(void)
3056 {
3057         misc_deregister(&kvm_dev);
3058         kmem_cache_destroy(kvm_vcpu_cache);
3059         sysdev_unregister(&kvm_sysdev);
3060         sysdev_class_unregister(&kvm_sysdev_class);
3061         unregister_reboot_notifier(&kvm_reboot_notifier);
3062         unregister_cpu_notifier(&kvm_cpu_notifier);
3063         on_each_cpu(hardware_disable, NULL, 0, 1);
3064         kvm_x86_ops->hardware_unsetup();
3065         kvm_x86_ops = NULL;
3066 }
3067 EXPORT_SYMBOL_GPL(kvm_exit_x86);
3068
3069 static __init int kvm_init(void)
3070 {
3071         int r;
3072
3073         r = kvm_mmu_module_init();
3074         if (r)
3075                 goto out4;
3076
3077         kvm_init_debug();
3078
3079         kvm_arch_init();
3080
3081         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3082
3083         if (bad_page == NULL) {
3084                 r = -ENOMEM;
3085                 goto out;
3086         }
3087
3088         return 0;
3089
3090 out:
3091         kvm_exit_debug();
3092         kvm_mmu_module_exit();
3093 out4:
3094         return r;
3095 }
3096
3097 static __exit void kvm_exit(void)
3098 {
3099         kvm_exit_debug();
3100         __free_page(bad_page);
3101         kvm_mmu_module_exit();
3102 }
3103
3104 module_init(kvm_init)
3105 module_exit(kvm_exit)