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_emulate.h"
  20 #include "segment_descriptor.h"
  21
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/gfp.h>
  27 #include <linux/mm.h>
  28 #include <linux/miscdevice.h>
  29 #include <linux/vmalloc.h>
  30 #include <linux/reboot.h>
  31 #include <linux/debugfs.h>
  32 #include <linux/highmem.h>
  33 #include <linux/file.h>
  34 #include <linux/sysdev.h>
  35 #include <linux/cpu.h>
  36 #include <linux/sched.h>
  37 #include <linux/cpumask.h>
  38 #include <linux/smp.h>
  39 #include <linux/anon_inodes.h>
  40
  41 #include <asm/processor.h>
  42 #include <asm/msr.h>
  43 #include <asm/io.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/desc.h>
  46
  47 MODULE_AUTHOR("Qumranet");
  48 MODULE_LICENSE("GPL");
  49
  50 static DEFINE_SPINLOCK(kvm_lock);
  51 static LIST_HEAD(vm_list);
  52
  53 static cpumask_t cpus_hardware_enabled;
  54
  55 struct kvm_arch_ops *kvm_arch_ops;
  56 struct kmem_cache *kvm_vcpu_cache;
  57 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  58
  59 static __read_mostly struct preempt_ops kvm_preempt_ops;
  60
  61 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  62
  63 static struct kvm_stats_debugfs_item {
  64         const char *name;
  65         int offset;
  66         struct dentry *dentry;
  67 } debugfs_entries[] = {
  68         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  69         { "pf_guest", STAT_OFFSET(pf_guest) },
  70         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  71         { "invlpg", STAT_OFFSET(invlpg) },
  72         { "exits", STAT_OFFSET(exits) },
  73         { "io_exits", STAT_OFFSET(io_exits) },
  74         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  75         { "signal_exits", STAT_OFFSET(signal_exits) },
  76         { "irq_window", STAT_OFFSET(irq_window_exits) },
  77         { "halt_exits", STAT_OFFSET(halt_exits) },
  78         { "request_irq", STAT_OFFSET(request_irq_exits) },
  79         { "irq_exits", STAT_OFFSET(irq_exits) },
  80         { "light_exits", STAT_OFFSET(light_exits) },
  81         { "efer_reload", STAT_OFFSET(efer_reload) },
  82         { NULL }
  83 };
  84
  85 static struct dentry *debugfs_dir;
  86
  87 #define MAX_IO_MSRS 256
  88
  89 #define CR0_RESERVED_BITS                                               \
  90         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  91                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  92                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  93 #define CR4_RESERVED_BITS                                               \
  94         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  95                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  96                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  97                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  98
  99 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
 100 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
 101
 102 #ifdef CONFIG_X86_64
 103 // LDT or TSS descriptor in the GDT. 16 bytes.
 104 struct segment_descriptor_64 {
 105         struct segment_descriptor s;
 106         u32 base_higher;
 107         u32 pad_zero;
 108 };
 109
 110 #endif
 111
 112 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 113                            unsigned long arg);
 114
 115 unsigned long segment_base(u16 selector)
 116 {
 117         struct descriptor_table gdt;
 118         struct segment_descriptor *d;
 119         unsigned long table_base;
 120         typedef unsigned long ul;
 121         unsigned long v;
 122
 123         if (selector == 0)
 124                 return 0;
 125
 126         asm ("sgdt %0" : "=m"(gdt));
 127         table_base = gdt.base;
 128
 129         if (selector & 4) {           /* from ldt */
 130                 u16 ldt_selector;
 131
 132                 asm ("sldt %0" : "=g"(ldt_selector));
 133                 table_base = segment_base(ldt_selector);
 134         }
 135         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 136         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 137 #ifdef CONFIG_X86_64
 138         if (d->system == 0
 139             && (d->type == 2 || d->type == 9 || d->type == 11))
 140                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 141 #endif
 142         return v;
 143 }
 144 EXPORT_SYMBOL_GPL(segment_base);
 145
 146 static inline int valid_vcpu(int n)
 147 {
 148         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 149 }
 150
 151 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 152 {
 153         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 154                 return;
 155
 156         vcpu->guest_fpu_loaded = 1;
 157         fx_save(&vcpu->host_fx_image);
 158         fx_restore(&vcpu->guest_fx_image);
 159 }
 160 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 161
 162 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 163 {
 164         if (!vcpu->guest_fpu_loaded)
 165                 return;
 166
 167         vcpu->guest_fpu_loaded = 0;
 168         fx_save(&vcpu->guest_fx_image);
 169         fx_restore(&vcpu->host_fx_image);
 170 }
 171 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 172
 173 /*
 174  * Switches to specified vcpu, until a matching vcpu_put()
 175  */
 176 static void vcpu_load(struct kvm_vcpu *vcpu)
 177 {
 178         int cpu;
 179
 180         mutex_lock(&vcpu->mutex);
 181         cpu = get_cpu();
 182         preempt_notifier_register(&vcpu->preempt_notifier);
 183         kvm_arch_ops->vcpu_load(vcpu, cpu);
 184         put_cpu();
 185 }
 186
 187 static void vcpu_put(struct kvm_vcpu *vcpu)
 188 {
 189         preempt_disable();
 190         kvm_arch_ops->vcpu_put(vcpu);
 191         preempt_notifier_unregister(&vcpu->preempt_notifier);
 192         preempt_enable();
 193         mutex_unlock(&vcpu->mutex);
 194 }
 195
 196 static void ack_flush(void *_completed)
 197 {
 198         atomic_t *completed = _completed;
 199
 200         atomic_inc(completed);
 201 }
 202
 203 void kvm_flush_remote_tlbs(struct kvm *kvm)
 204 {
 205         int i, cpu, needed;
 206         cpumask_t cpus;
 207         struct kvm_vcpu *vcpu;
 208         atomic_t completed;
 209
 210         atomic_set(&completed, 0);
 211         cpus_clear(cpus);
 212         needed = 0;
 213         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 214                 vcpu = kvm->vcpus[i];
 215                 if (!vcpu)
 216                         continue;
 217                 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
 218                         continue;
 219                 cpu = vcpu->cpu;
 220                 if (cpu != -1 && cpu != raw_smp_processor_id())
 221                         if (!cpu_isset(cpu, cpus)) {
 222                                 cpu_set(cpu, cpus);
 223                                 ++needed;
 224                         }
 225         }
 226
 227         /*
 228          * We really want smp_call_function_mask() here.  But that's not
 229          * available, so ipi all cpus in parallel and wait for them
 230          * to complete.
 231          */
 232         for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
 233                 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
 234         while (atomic_read(&completed) != needed) {
 235                 cpu_relax();
 236                 barrier();
 237         }
 238 }
 239
 240 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 241 {
 242         struct page *page;
 243         int r;
 244
 245         mutex_init(&vcpu->mutex);
 246         vcpu->cpu = -1;
 247         vcpu->mmu.root_hpa = INVALID_PAGE;
 248         vcpu->kvm = kvm;
 249         vcpu->vcpu_id = id;
 250
 251         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 252         if (!page) {
 253                 r = -ENOMEM;
 254                 goto fail;
 255         }
 256         vcpu->run = page_address(page);
 257
 258         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 259         if (!page) {
 260                 r = -ENOMEM;
 261                 goto fail_free_run;
 262         }
 263         vcpu->pio_data = page_address(page);
 264
 265         r = kvm_mmu_create(vcpu);
 266         if (r < 0)
 267                 goto fail_free_pio_data;
 268
 269         return 0;
 270
 271 fail_free_pio_data:
 272         free_page((unsigned long)vcpu->pio_data);
 273 fail_free_run:
 274         free_page((unsigned long)vcpu->run);
 275 fail:
 276         return -ENOMEM;
 277 }
 278 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 279
 280 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 281 {
 282         kvm_mmu_destroy(vcpu);
 283         free_page((unsigned long)vcpu->pio_data);
 284         free_page((unsigned long)vcpu->run);
 285 }
 286 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 287
 288 static struct kvm *kvm_create_vm(void)
 289 {
 290         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 291
 292         if (!kvm)
 293                 return ERR_PTR(-ENOMEM);
 294
 295         kvm_io_bus_init(&kvm->pio_bus);
 296         mutex_init(&kvm->lock);
 297         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 298         kvm_io_bus_init(&kvm->mmio_bus);
 299         spin_lock(&kvm_lock);
 300         list_add(&kvm->vm_list, &vm_list);
 301         spin_unlock(&kvm_lock);
 302         return kvm;
 303 }
 304
 305 static int kvm_dev_open(struct inode *inode, struct file *filp)
 306 {
 307         return 0;
 308 }
 309
 310 /*
 311  * Free any memory in @free but not in @dont.
 312  */
 313 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 314                                   struct kvm_memory_slot *dont)
 315 {
 316         int i;
 317
 318         if (!dont || free->phys_mem != dont->phys_mem)
 319                 if (free->phys_mem) {
 320                         for (i = 0; i < free->npages; ++i)
 321                                 if (free->phys_mem[i])
 322                                         __free_page(free->phys_mem[i]);
 323                         vfree(free->phys_mem);
 324                 }
 325
 326         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 327                 vfree(free->dirty_bitmap);
 328
 329         free->phys_mem = NULL;
 330         free->npages = 0;
 331         free->dirty_bitmap = NULL;
 332 }
 333
 334 static void kvm_free_physmem(struct kvm *kvm)
 335 {
 336         int i;
 337
 338         for (i = 0; i < kvm->nmemslots; ++i)
 339                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 340 }
 341
 342 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 343 {
 344         int i;
 345
 346         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
 347                 if (vcpu->pio.guest_pages[i]) {
 348                         __free_page(vcpu->pio.guest_pages[i]);
 349                         vcpu->pio.guest_pages[i] = NULL;
 350                 }
 351 }
 352
 353 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 354 {
 355         vcpu_load(vcpu);
 356         kvm_mmu_unload(vcpu);
 357         vcpu_put(vcpu);
 358 }
 359
 360 static void kvm_free_vcpus(struct kvm *kvm)
 361 {
 362         unsigned int i;
 363
 364         /*
 365          * Unpin any mmu pages first.
 366          */
 367         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 368                 if (kvm->vcpus[i])
 369                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
 370         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 371                 if (kvm->vcpus[i]) {
 372                         kvm_arch_ops->vcpu_free(kvm->vcpus[i]);
 373                         kvm->vcpus[i] = NULL;
 374                 }
 375         }
 376
 377 }
 378
 379 static int kvm_dev_release(struct inode *inode, struct file *filp)
 380 {
 381         return 0;
 382 }
 383
 384 static void kvm_destroy_vm(struct kvm *kvm)
 385 {
 386         spin_lock(&kvm_lock);
 387         list_del(&kvm->vm_list);
 388         spin_unlock(&kvm_lock);
 389         kvm_io_bus_destroy(&kvm->pio_bus);
 390         kvm_io_bus_destroy(&kvm->mmio_bus);
 391         kvm_free_vcpus(kvm);
 392         kvm_free_physmem(kvm);
 393         kfree(kvm);
 394 }
 395
 396 static int kvm_vm_release(struct inode *inode, struct file *filp)
 397 {
 398         struct kvm *kvm = filp->private_data;
 399
 400         kvm_destroy_vm(kvm);
 401         return 0;
 402 }
 403
 404 static void inject_gp(struct kvm_vcpu *vcpu)
 405 {
 406         kvm_arch_ops->inject_gp(vcpu, 0);
 407 }
 408
 409 /*
 410  * Load the pae pdptrs.  Return true is they are all valid.
 411  */
 412 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 413 {
 414         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 415         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 416         int i;
 417         u64 *pdpt;
 418         int ret;
 419         struct page *page;
 420         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 421
 422         mutex_lock(&vcpu->kvm->lock);
 423         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
 424         if (!page) {
 425                 ret = 0;
 426                 goto out;
 427         }
 428
 429         pdpt = kmap_atomic(page, KM_USER0);
 430         memcpy(pdpte, pdpt+offset, sizeof(pdpte));
 431         kunmap_atomic(pdpt, KM_USER0);
 432
 433         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 434                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 435                         ret = 0;
 436                         goto out;
 437                 }
 438         }
 439         ret = 1;
 440
 441         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 442 out:
 443         mutex_unlock(&vcpu->kvm->lock);
 444
 445         return ret;
 446 }
 447
 448 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 449 {
 450         if (cr0 & CR0_RESERVED_BITS) {
 451                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 452                        cr0, vcpu->cr0);
 453                 inject_gp(vcpu);
 454                 return;
 455         }
 456
 457         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 458                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 459                 inject_gp(vcpu);
 460                 return;
 461         }
 462
 463         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 464                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 465                        "and a clear PE flag\n");
 466                 inject_gp(vcpu);
 467                 return;
 468         }
 469
 470         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 471 #ifdef CONFIG_X86_64
 472                 if ((vcpu->shadow_efer & EFER_LME)) {
 473                         int cs_db, cs_l;
 474
 475                         if (!is_pae(vcpu)) {
 476                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 477                                        "in long mode while PAE is disabled\n");
 478                                 inject_gp(vcpu);
 479                                 return;
 480                         }
 481                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 482                         if (cs_l) {
 483                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 484                                        "in long mode while CS.L == 1\n");
 485                                 inject_gp(vcpu);
 486                                 return;
 487
 488                         }
 489                 } else
 490 #endif
 491                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 492                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 493                                "reserved bits\n");
 494                         inject_gp(vcpu);
 495                         return;
 496                 }
 497
 498         }
 499
 500         kvm_arch_ops->set_cr0(vcpu, cr0);
 501         vcpu->cr0 = cr0;
 502
 503         mutex_lock(&vcpu->kvm->lock);
 504         kvm_mmu_reset_context(vcpu);
 505         mutex_unlock(&vcpu->kvm->lock);
 506         return;
 507 }
 508 EXPORT_SYMBOL_GPL(set_cr0);
 509
 510 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 511 {
 512         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 513 }
 514 EXPORT_SYMBOL_GPL(lmsw);
 515
 516 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 517 {
 518         if (cr4 & CR4_RESERVED_BITS) {
 519                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 520                 inject_gp(vcpu);
 521                 return;
 522         }
 523
 524         if (is_long_mode(vcpu)) {
 525                 if (!(cr4 & X86_CR4_PAE)) {
 526                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 527                                "in long mode\n");
 528                         inject_gp(vcpu);
 529                         return;
 530                 }
 531         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 532                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 533                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 534                 inject_gp(vcpu);
 535                 return;
 536         }
 537
 538         if (cr4 & X86_CR4_VMXE) {
 539                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 540                 inject_gp(vcpu);
 541                 return;
 542         }
 543         kvm_arch_ops->set_cr4(vcpu, cr4);
 544         mutex_lock(&vcpu->kvm->lock);
 545         kvm_mmu_reset_context(vcpu);
 546         mutex_unlock(&vcpu->kvm->lock);
 547 }
 548 EXPORT_SYMBOL_GPL(set_cr4);
 549
 550 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 551 {
 552         if (is_long_mode(vcpu)) {
 553                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 554                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 555                         inject_gp(vcpu);
 556                         return;
 557                 }
 558         } else {
 559                 if (is_pae(vcpu)) {
 560                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 561                                 printk(KERN_DEBUG
 562                                        "set_cr3: #GP, reserved bits\n");
 563                                 inject_gp(vcpu);
 564                                 return;
 565                         }
 566                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 567                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 568                                        "reserved bits\n");
 569                                 inject_gp(vcpu);
 570                                 return;
 571                         }
 572                 } else {
 573                         if (cr3 & CR3_NONPAE_RESERVED_BITS) {
 574                                 printk(KERN_DEBUG
 575                                        "set_cr3: #GP, reserved bits\n");
 576                                 inject_gp(vcpu);
 577                                 return;
 578                         }
 579                 }
 580         }
 581
 582         mutex_lock(&vcpu->kvm->lock);
 583         /*
 584          * Does the new cr3 value map to physical memory? (Note, we
 585          * catch an invalid cr3 even in real-mode, because it would
 586          * cause trouble later on when we turn on paging anyway.)
 587          *
 588          * A real CPU would silently accept an invalid cr3 and would
 589          * attempt to use it - with largely undefined (and often hard
 590          * to debug) behavior on the guest side.
 591          */
 592         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 593                 inject_gp(vcpu);
 594         else {
 595                 vcpu->cr3 = cr3;
 596                 vcpu->mmu.new_cr3(vcpu);
 597         }
 598         mutex_unlock(&vcpu->kvm->lock);
 599 }
 600 EXPORT_SYMBOL_GPL(set_cr3);
 601
 602 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 603 {
 604         if (cr8 & CR8_RESERVED_BITS) {
 605                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 606                 inject_gp(vcpu);
 607                 return;
 608         }
 609         vcpu->cr8 = cr8;
 610 }
 611 EXPORT_SYMBOL_GPL(set_cr8);
 612
 613 void fx_init(struct kvm_vcpu *vcpu)
 614 {
 615         unsigned after_mxcsr_mask;
 616
 617         /* Initialize guest FPU by resetting ours and saving into guest's */
 618         preempt_disable();
 619         fx_save(&vcpu->host_fx_image);
 620         fpu_init();
 621         fx_save(&vcpu->guest_fx_image);
 622         fx_restore(&vcpu->host_fx_image);
 623         preempt_enable();
 624
 625         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
 626         vcpu->guest_fx_image.mxcsr = 0x1f80;
 627         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
 628                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
 629 }
 630 EXPORT_SYMBOL_GPL(fx_init);
 631
 632 /*
 633  * Allocate some memory and give it an address in the guest physical address
 634  * space.
 635  *
 636  * Discontiguous memory is allowed, mostly for framebuffers.
 637  */
 638 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 639                                           struct kvm_memory_region *mem)
 640 {
 641         int r;
 642         gfn_t base_gfn;
 643         unsigned long npages;
 644         unsigned long i;
 645         struct kvm_memory_slot *memslot;
 646         struct kvm_memory_slot old, new;
 647         int memory_config_version;
 648
 649         r = -EINVAL;
 650         /* General sanity checks */
 651         if (mem->memory_size & (PAGE_SIZE - 1))
 652                 goto out;
 653         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 654                 goto out;
 655         if (mem->slot >= KVM_MEMORY_SLOTS)
 656                 goto out;
 657         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 658                 goto out;
 659
 660         memslot = &kvm->memslots[mem->slot];
 661         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 662         npages = mem->memory_size >> PAGE_SHIFT;
 663
 664         if (!npages)
 665                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 666
 667 raced:
 668         mutex_lock(&kvm->lock);
 669
 670         memory_config_version = kvm->memory_config_version;
 671         new = old = *memslot;
 672
 673         new.base_gfn = base_gfn;
 674         new.npages = npages;
 675         new.flags = mem->flags;
 676
 677         /* Disallow changing a memory slot's size. */
 678         r = -EINVAL;
 679         if (npages && old.npages && npages != old.npages)
 680                 goto out_unlock;
 681
 682         /* Check for overlaps */
 683         r = -EEXIST;
 684         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 685                 struct kvm_memory_slot *s = &kvm->memslots[i];
 686
 687                 if (s == memslot)
 688                         continue;
 689                 if (!((base_gfn + npages <= s->base_gfn) ||
 690                       (base_gfn >= s->base_gfn + s->npages)))
 691                         goto out_unlock;
 692         }
 693         /*
 694          * Do memory allocations outside lock.  memory_config_version will
 695          * detect any races.
 696          */
 697         mutex_unlock(&kvm->lock);
 698
 699         /* Deallocate if slot is being removed */
 700         if (!npages)
 701                 new.phys_mem = NULL;
 702
 703         /* Free page dirty bitmap if unneeded */
 704         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 705                 new.dirty_bitmap = NULL;
 706
 707         r = -ENOMEM;
 708
 709         /* Allocate if a slot is being created */
 710         if (npages && !new.phys_mem) {
 711                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 712
 713                 if (!new.phys_mem)
 714                         goto out_free;
 715
 716                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 717                 for (i = 0; i < npages; ++i) {
 718                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 719                                                      | __GFP_ZERO);
 720                         if (!new.phys_mem[i])
 721                                 goto out_free;
 722                         set_page_private(new.phys_mem[i],0);
 723                 }
 724         }
 725
 726         /* Allocate page dirty bitmap if needed */
 727         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 728                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 729
 730                 new.dirty_bitmap = vmalloc(dirty_bytes);
 731                 if (!new.dirty_bitmap)
 732                         goto out_free;
 733                 memset(new.dirty_bitmap, 0, dirty_bytes);
 734         }
 735
 736         mutex_lock(&kvm->lock);
 737
 738         if (memory_config_version != kvm->memory_config_version) {
 739                 mutex_unlock(&kvm->lock);
 740                 kvm_free_physmem_slot(&new, &old);
 741                 goto raced;
 742         }
 743
 744         r = -EAGAIN;
 745         if (kvm->busy)
 746                 goto out_unlock;
 747
 748         if (mem->slot >= kvm->nmemslots)
 749                 kvm->nmemslots = mem->slot + 1;
 750
 751         *memslot = new;
 752         ++kvm->memory_config_version;
 753
 754         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 755         kvm_flush_remote_tlbs(kvm);
 756
 757         mutex_unlock(&kvm->lock);
 758
 759         kvm_free_physmem_slot(&old, &new);
 760         return 0;
 761
 762 out_unlock:
 763         mutex_unlock(&kvm->lock);
 764 out_free:
 765         kvm_free_physmem_slot(&new, &old);
 766 out:
 767         return r;
 768 }
 769
 770 /*
 771  * Get (and clear) the dirty memory log for a memory slot.
 772  */
 773 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 774                                       struct kvm_dirty_log *log)
 775 {
 776         struct kvm_memory_slot *memslot;
 777         int r, i;
 778         int n;
 779         unsigned long any = 0;
 780
 781         mutex_lock(&kvm->lock);
 782
 783         /*
 784          * Prevent changes to guest memory configuration even while the lock
 785          * is not taken.
 786          */
 787         ++kvm->busy;
 788         mutex_unlock(&kvm->lock);
 789         r = -EINVAL;
 790         if (log->slot >= KVM_MEMORY_SLOTS)
 791                 goto out;
 792
 793         memslot = &kvm->memslots[log->slot];
 794         r = -ENOENT;
 795         if (!memslot->dirty_bitmap)
 796                 goto out;
 797
 798         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 799
 800         for (i = 0; !any && i < n/sizeof(long); ++i)
 801                 any = memslot->dirty_bitmap[i];
 802
 803         r = -EFAULT;
 804         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 805                 goto out;
 806
 807         /* If nothing is dirty, don't bother messing with page tables. */
 808         if (any) {
 809                 mutex_lock(&kvm->lock);
 810                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 811                 kvm_flush_remote_tlbs(kvm);
 812                 memset(memslot->dirty_bitmap, 0, n);
 813                 mutex_unlock(&kvm->lock);
 814         }
 815
 816         r = 0;
 817
 818 out:
 819         mutex_lock(&kvm->lock);
 820         --kvm->busy;
 821         mutex_unlock(&kvm->lock);
 822         return r;
 823 }
 824
 825 /*
 826  * Set a new alias region.  Aliases map a portion of physical memory into
 827  * another portion.  This is useful for memory windows, for example the PC
 828  * VGA region.
 829  */
 830 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 831                                          struct kvm_memory_alias *alias)
 832 {
 833         int r, n;
 834         struct kvm_mem_alias *p;
 835
 836         r = -EINVAL;
 837         /* General sanity checks */
 838         if (alias->memory_size & (PAGE_SIZE - 1))
 839                 goto out;
 840         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 841                 goto out;
 842         if (alias->slot >= KVM_ALIAS_SLOTS)
 843                 goto out;
 844         if (alias->guest_phys_addr + alias->memory_size
 845             < alias->guest_phys_addr)
 846                 goto out;
 847         if (alias->target_phys_addr + alias->memory_size
 848             < alias->target_phys_addr)
 849                 goto out;
 850
 851         mutex_lock(&kvm->lock);
 852
 853         p = &kvm->aliases[alias->slot];
 854         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 855         p->npages = alias->memory_size >> PAGE_SHIFT;
 856         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 857
 858         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 859                 if (kvm->aliases[n - 1].npages)
 860                         break;
 861         kvm->naliases = n;
 862
 863         kvm_mmu_zap_all(kvm);
 864
 865         mutex_unlock(&kvm->lock);
 866
 867         return 0;
 868
 869 out:
 870         return r;
 871 }
 872
 873 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 874 {
 875         int i;
 876         struct kvm_mem_alias *alias;
 877
 878         for (i = 0; i < kvm->naliases; ++i) {
 879                 alias = &kvm->aliases[i];
 880                 if (gfn >= alias->base_gfn
 881                     && gfn < alias->base_gfn + alias->npages)
 882                         return alias->target_gfn + gfn - alias->base_gfn;
 883         }
 884         return gfn;
 885 }
 886
 887 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 888 {
 889         int i;
 890
 891         for (i = 0; i < kvm->nmemslots; ++i) {
 892                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 893
 894                 if (gfn >= memslot->base_gfn
 895                     && gfn < memslot->base_gfn + memslot->npages)
 896                         return memslot;
 897         }
 898         return NULL;
 899 }
 900
 901 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 902 {
 903         gfn = unalias_gfn(kvm, gfn);
 904         return __gfn_to_memslot(kvm, gfn);
 905 }
 906
 907 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 908 {
 909         struct kvm_memory_slot *slot;
 910
 911         gfn = unalias_gfn(kvm, gfn);
 912         slot = __gfn_to_memslot(kvm, gfn);
 913         if (!slot)
 914                 return NULL;
 915         return slot->phys_mem[gfn - slot->base_gfn];
 916 }
 917 EXPORT_SYMBOL_GPL(gfn_to_page);
 918
 919 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 920 {
 921         int i;
 922         struct kvm_memory_slot *memslot;
 923         unsigned long rel_gfn;
 924
 925         for (i = 0; i < kvm->nmemslots; ++i) {
 926                 memslot = &kvm->memslots[i];
 927
 928                 if (gfn >= memslot->base_gfn
 929                     && gfn < memslot->base_gfn + memslot->npages) {
 930
 931                         if (!memslot->dirty_bitmap)
 932                                 return;
 933
 934                         rel_gfn = gfn - memslot->base_gfn;
 935
 936                         /* avoid RMW */
 937                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 938                                 set_bit(rel_gfn, memslot->dirty_bitmap);
 939                         return;
 940                 }
 941         }
 942 }
 943
 944 int emulator_read_std(unsigned long addr,
 945                              void *val,
 946                              unsigned int bytes,
 947                              struct kvm_vcpu *vcpu)
 948 {
 949         void *data = val;
 950
 951         while (bytes) {
 952                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 953                 unsigned offset = addr & (PAGE_SIZE-1);
 954                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 955                 unsigned long pfn;
 956                 struct page *page;
 957                 void *page_virt;
 958
 959                 if (gpa == UNMAPPED_GVA)
 960                         return X86EMUL_PROPAGATE_FAULT;
 961                 pfn = gpa >> PAGE_SHIFT;
 962                 page = gfn_to_page(vcpu->kvm, pfn);
 963                 if (!page)
 964                         return X86EMUL_UNHANDLEABLE;
 965                 page_virt = kmap_atomic(page, KM_USER0);
 966
 967                 memcpy(data, page_virt + offset, tocopy);
 968
 969                 kunmap_atomic(page_virt, KM_USER0);
 970
 971                 bytes -= tocopy;
 972                 data += tocopy;
 973                 addr += tocopy;
 974         }
 975
 976         return X86EMUL_CONTINUE;
 977 }
 978 EXPORT_SYMBOL_GPL(emulator_read_std);
 979
 980 static int emulator_write_std(unsigned long addr,
 981                               const void *val,
 982                               unsigned int bytes,
 983                               struct kvm_vcpu *vcpu)
 984 {
 985         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
 986                addr, bytes);
 987         return X86EMUL_UNHANDLEABLE;
 988 }
 989
 990 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
 991                                                 gpa_t addr)
 992 {
 993         /*
 994          * Note that its important to have this wrapper function because
 995          * in the very near future we will be checking for MMIOs against
 996          * the LAPIC as well as the general MMIO bus
 997          */
 998         return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
 999 }
1000
1001 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1002                                                gpa_t addr)
1003 {
1004         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1005 }
1006
1007 static int emulator_read_emulated(unsigned long addr,
1008                                   void *val,
1009                                   unsigned int bytes,
1010                                   struct kvm_vcpu *vcpu)
1011 {
1012         struct kvm_io_device *mmio_dev;
1013         gpa_t                 gpa;
1014
1015         if (vcpu->mmio_read_completed) {
1016                 memcpy(val, vcpu->mmio_data, bytes);
1017                 vcpu->mmio_read_completed = 0;
1018                 return X86EMUL_CONTINUE;
1019         } else if (emulator_read_std(addr, val, bytes, vcpu)
1020                    == X86EMUL_CONTINUE)
1021                 return X86EMUL_CONTINUE;
1022
1023         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1024         if (gpa == UNMAPPED_GVA)
1025                 return X86EMUL_PROPAGATE_FAULT;
1026
1027         /*
1028          * Is this MMIO handled locally?
1029          */
1030         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1031         if (mmio_dev) {
1032                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1033                 return X86EMUL_CONTINUE;
1034         }
1035
1036         vcpu->mmio_needed = 1;
1037         vcpu->mmio_phys_addr = gpa;
1038         vcpu->mmio_size = bytes;
1039         vcpu->mmio_is_write = 0;
1040
1041         return X86EMUL_UNHANDLEABLE;
1042 }
1043
1044 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1045                                const void *val, int bytes)
1046 {
1047         struct page *page;
1048         void *virt;
1049
1050         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1051                 return 0;
1052         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1053         if (!page)
1054                 return 0;
1055         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1056         virt = kmap_atomic(page, KM_USER0);
1057         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1058         memcpy(virt + offset_in_page(gpa), val, bytes);
1059         kunmap_atomic(virt, KM_USER0);
1060         return 1;
1061 }
1062
1063 static int emulator_write_emulated_onepage(unsigned long addr,
1064                                            const void *val,
1065                                            unsigned int bytes,
1066                                            struct kvm_vcpu *vcpu)
1067 {
1068         struct kvm_io_device *mmio_dev;
1069         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1070
1071         if (gpa == UNMAPPED_GVA) {
1072                 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1073                 return X86EMUL_PROPAGATE_FAULT;
1074         }
1075
1076         if (emulator_write_phys(vcpu, gpa, val, bytes))
1077                 return X86EMUL_CONTINUE;
1078
1079         /*
1080          * Is this MMIO handled locally?
1081          */
1082         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1083         if (mmio_dev) {
1084                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1085                 return X86EMUL_CONTINUE;
1086         }
1087
1088         vcpu->mmio_needed = 1;
1089         vcpu->mmio_phys_addr = gpa;
1090         vcpu->mmio_size = bytes;
1091         vcpu->mmio_is_write = 1;
1092         memcpy(vcpu->mmio_data, val, bytes);
1093
1094         return X86EMUL_CONTINUE;
1095 }
1096
1097 int emulator_write_emulated(unsigned long addr,
1098                                    const void *val,
1099                                    unsigned int bytes,
1100                                    struct kvm_vcpu *vcpu)
1101 {
1102         /* Crossing a page boundary? */
1103         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1104                 int rc, now;
1105
1106                 now = -addr & ~PAGE_MASK;
1107                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1108                 if (rc != X86EMUL_CONTINUE)
1109                         return rc;
1110                 addr += now;
1111                 val += now;
1112                 bytes -= now;
1113         }
1114         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1115 }
1116 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1117
1118 static int emulator_cmpxchg_emulated(unsigned long addr,
1119                                      const void *old,
1120                                      const void *new,
1121                                      unsigned int bytes,
1122                                      struct kvm_vcpu *vcpu)
1123 {
1124         static int reported;
1125
1126         if (!reported) {
1127                 reported = 1;
1128                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1129         }
1130         return emulator_write_emulated(addr, new, bytes, vcpu);
1131 }
1132
1133 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1134 {
1135         return kvm_arch_ops->get_segment_base(vcpu, seg);
1136 }
1137
1138 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1139 {
1140         return X86EMUL_CONTINUE;
1141 }
1142
1143 int emulate_clts(struct kvm_vcpu *vcpu)
1144 {
1145         unsigned long cr0;
1146
1147         cr0 = vcpu->cr0 & ~X86_CR0_TS;
1148         kvm_arch_ops->set_cr0(vcpu, cr0);
1149         return X86EMUL_CONTINUE;
1150 }
1151
1152 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1153 {
1154         struct kvm_vcpu *vcpu = ctxt->vcpu;
1155
1156         switch (dr) {
1157         case 0 ... 3:
1158                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1159                 return X86EMUL_CONTINUE;
1160         default:
1161                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1162                        __FUNCTION__, dr);
1163                 return X86EMUL_UNHANDLEABLE;
1164         }
1165 }
1166
1167 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1168 {
1169         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1170         int exception;
1171
1172         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1173         if (exception) {
1174                 /* FIXME: better handling */
1175                 return X86EMUL_UNHANDLEABLE;
1176         }
1177         return X86EMUL_CONTINUE;
1178 }
1179
1180 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1181 {
1182         static int reported;
1183         u8 opcodes[4];
1184         unsigned long rip = ctxt->vcpu->rip;
1185         unsigned long rip_linear;
1186
1187         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1188
1189         if (reported)
1190                 return;
1191
1192         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt->vcpu);
1193
1194         printk(KERN_ERR "emulation failed but !mmio_needed?"
1195                " rip %lx %02x %02x %02x %02x\n",
1196                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1197         reported = 1;
1198 }
1199
1200 struct x86_emulate_ops emulate_ops = {
1201         .read_std            = emulator_read_std,
1202         .write_std           = emulator_write_std,
1203         .read_emulated       = emulator_read_emulated,
1204         .write_emulated      = emulator_write_emulated,
1205         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1206 };
1207
1208 int emulate_instruction(struct kvm_vcpu *vcpu,
1209                         struct kvm_run *run,
1210                         unsigned long cr2,
1211                         u16 error_code)
1212 {
1213         struct x86_emulate_ctxt emulate_ctxt;
1214         int r;
1215         int cs_db, cs_l;
1216
1217         vcpu->mmio_fault_cr2 = cr2;
1218         kvm_arch_ops->cache_regs(vcpu);
1219
1220         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1221
1222         emulate_ctxt.vcpu = vcpu;
1223         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1224         emulate_ctxt.cr2 = cr2;
1225         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1226                 ? X86EMUL_MODE_REAL : cs_l
1227                 ? X86EMUL_MODE_PROT64 : cs_db
1228                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1229
1230         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1231                 emulate_ctxt.cs_base = 0;
1232                 emulate_ctxt.ds_base = 0;
1233                 emulate_ctxt.es_base = 0;
1234                 emulate_ctxt.ss_base = 0;
1235         } else {
1236                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1237                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1238                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1239                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1240         }
1241
1242         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1243         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1244
1245         vcpu->mmio_is_write = 0;
1246         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1247
1248         if ((r || vcpu->mmio_is_write) && run) {
1249                 run->exit_reason = KVM_EXIT_MMIO;
1250                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1251                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1252                 run->mmio.len = vcpu->mmio_size;
1253                 run->mmio.is_write = vcpu->mmio_is_write;
1254         }
1255
1256         if (r) {
1257                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1258                         return EMULATE_DONE;
1259                 if (!vcpu->mmio_needed) {
1260                         report_emulation_failure(&emulate_ctxt);
1261                         return EMULATE_FAIL;
1262                 }
1263                 return EMULATE_DO_MMIO;
1264         }
1265
1266         kvm_arch_ops->decache_regs(vcpu);
1267         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1268
1269         if (vcpu->mmio_is_write) {
1270                 vcpu->mmio_needed = 0;
1271                 return EMULATE_DO_MMIO;
1272         }
1273
1274         return EMULATE_DONE;
1275 }
1276 EXPORT_SYMBOL_GPL(emulate_instruction);
1277
1278 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1279 {
1280         if (vcpu->irq_summary)
1281                 return 1;
1282
1283         vcpu->run->exit_reason = KVM_EXIT_HLT;
1284         ++vcpu->stat.halt_exits;
1285         return 0;
1286 }
1287 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1288
1289 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1290 {
1291         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1292
1293         kvm_arch_ops->cache_regs(vcpu);
1294         ret = -KVM_EINVAL;
1295 #ifdef CONFIG_X86_64
1296         if (is_long_mode(vcpu)) {
1297                 nr = vcpu->regs[VCPU_REGS_RAX];
1298                 a0 = vcpu->regs[VCPU_REGS_RDI];
1299                 a1 = vcpu->regs[VCPU_REGS_RSI];
1300                 a2 = vcpu->regs[VCPU_REGS_RDX];
1301                 a3 = vcpu->regs[VCPU_REGS_RCX];
1302                 a4 = vcpu->regs[VCPU_REGS_R8];
1303                 a5 = vcpu->regs[VCPU_REGS_R9];
1304         } else
1305 #endif
1306         {
1307                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1308                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1309                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1310                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1311                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1312                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1313                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1314         }
1315         switch (nr) {
1316         default:
1317                 run->hypercall.nr = nr;
1318                 run->hypercall.args[0] = a0;
1319                 run->hypercall.args[1] = a1;
1320                 run->hypercall.args[2] = a2;
1321                 run->hypercall.args[3] = a3;
1322                 run->hypercall.args[4] = a4;
1323                 run->hypercall.args[5] = a5;
1324                 run->hypercall.ret = ret;
1325                 run->hypercall.longmode = is_long_mode(vcpu);
1326                 kvm_arch_ops->decache_regs(vcpu);
1327                 return 0;
1328         }
1329         vcpu->regs[VCPU_REGS_RAX] = ret;
1330         kvm_arch_ops->decache_regs(vcpu);
1331         return 1;
1332 }
1333 EXPORT_SYMBOL_GPL(kvm_hypercall);
1334
1335 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1336 {
1337         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1338 }
1339
1340 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1341 {
1342         struct descriptor_table dt = { limit, base };
1343
1344         kvm_arch_ops->set_gdt(vcpu, &dt);
1345 }
1346
1347 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1348 {
1349         struct descriptor_table dt = { limit, base };
1350
1351         kvm_arch_ops->set_idt(vcpu, &dt);
1352 }
1353
1354 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1355                    unsigned long *rflags)
1356 {
1357         lmsw(vcpu, msw);
1358         *rflags = kvm_arch_ops->get_rflags(vcpu);
1359 }
1360
1361 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1362 {
1363         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1364         switch (cr) {
1365         case 0:
1366                 return vcpu->cr0;
1367         case 2:
1368                 return vcpu->cr2;
1369         case 3:
1370                 return vcpu->cr3;
1371         case 4:
1372                 return vcpu->cr4;
1373         default:
1374                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1375                 return 0;
1376         }
1377 }
1378
1379 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1380                      unsigned long *rflags)
1381 {
1382         switch (cr) {
1383         case 0:
1384                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1385                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1386                 break;
1387         case 2:
1388                 vcpu->cr2 = val;
1389                 break;
1390         case 3:
1391                 set_cr3(vcpu, val);
1392                 break;
1393         case 4:
1394                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1395                 break;
1396         default:
1397                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1398         }
1399 }
1400
1401 /*
1402  * Register the para guest with the host:
1403  */
1404 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1405 {
1406         struct kvm_vcpu_para_state *para_state;
1407         hpa_t para_state_hpa, hypercall_hpa;
1408         struct page *para_state_page;
1409         unsigned char *hypercall;
1410         gpa_t hypercall_gpa;
1411
1412         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1413         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1414
1415         /*
1416          * Needs to be page aligned:
1417          */
1418         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1419                 goto err_gp;
1420
1421         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1422         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1423         if (is_error_hpa(para_state_hpa))
1424                 goto err_gp;
1425
1426         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1427         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1428         para_state = kmap(para_state_page);
1429
1430         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1431         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1432
1433         para_state->host_version = KVM_PARA_API_VERSION;
1434         /*
1435          * We cannot support guests that try to register themselves
1436          * with a newer API version than the host supports:
1437          */
1438         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1439                 para_state->ret = -KVM_EINVAL;
1440                 goto err_kunmap_skip;
1441         }
1442
1443         hypercall_gpa = para_state->hypercall_gpa;
1444         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1445         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1446         if (is_error_hpa(hypercall_hpa)) {
1447                 para_state->ret = -KVM_EINVAL;
1448                 goto err_kunmap_skip;
1449         }
1450
1451         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1452         vcpu->para_state_page = para_state_page;
1453         vcpu->para_state_gpa = para_state_gpa;
1454         vcpu->hypercall_gpa = hypercall_gpa;
1455
1456         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1457         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1458                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1459         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1460         kunmap_atomic(hypercall, KM_USER1);
1461
1462         para_state->ret = 0;
1463 err_kunmap_skip:
1464         kunmap(para_state_page);
1465         return 0;
1466 err_gp:
1467         return 1;
1468 }
1469
1470 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1471 {
1472         u64 data;
1473
1474         switch (msr) {
1475         case 0xc0010010: /* SYSCFG */
1476         case 0xc0010015: /* HWCR */
1477         case MSR_IA32_PLATFORM_ID:
1478         case MSR_IA32_P5_MC_ADDR:
1479         case MSR_IA32_P5_MC_TYPE:
1480         case MSR_IA32_MC0_CTL:
1481         case MSR_IA32_MCG_STATUS:
1482         case MSR_IA32_MCG_CAP:
1483         case MSR_IA32_MC0_MISC:
1484         case MSR_IA32_MC0_MISC+4:
1485         case MSR_IA32_MC0_MISC+8:
1486         case MSR_IA32_MC0_MISC+12:
1487         case MSR_IA32_MC0_MISC+16:
1488         case MSR_IA32_UCODE_REV:
1489         case MSR_IA32_PERF_STATUS:
1490         case MSR_IA32_EBL_CR_POWERON:
1491                 /* MTRR registers */
1492         case 0xfe:
1493         case 0x200 ... 0x2ff:
1494                 data = 0;
1495                 break;
1496         case 0xcd: /* fsb frequency */
1497                 data = 3;
1498                 break;
1499         case MSR_IA32_APICBASE:
1500                 data = vcpu->apic_base;
1501                 break;
1502         case MSR_IA32_MISC_ENABLE:
1503                 data = vcpu->ia32_misc_enable_msr;
1504                 break;
1505 #ifdef CONFIG_X86_64
1506         case MSR_EFER:
1507                 data = vcpu->shadow_efer;
1508                 break;
1509 #endif
1510         default:
1511                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1512                 return 1;
1513         }
1514         *pdata = data;
1515         return 0;
1516 }
1517 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1518
1519 /*
1520  * Reads an msr value (of 'msr_index') into 'pdata'.
1521  * Returns 0 on success, non-0 otherwise.
1522  * Assumes vcpu_load() was already called.
1523  */
1524 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1525 {
1526         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1527 }
1528
1529 #ifdef CONFIG_X86_64
1530
1531 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1532 {
1533         if (efer & EFER_RESERVED_BITS) {
1534                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1535                        efer);
1536                 inject_gp(vcpu);
1537                 return;
1538         }
1539
1540         if (is_paging(vcpu)
1541             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1542                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1543                 inject_gp(vcpu);
1544                 return;
1545         }
1546
1547         kvm_arch_ops->set_efer(vcpu, efer);
1548
1549         efer &= ~EFER_LMA;
1550         efer |= vcpu->shadow_efer & EFER_LMA;
1551
1552         vcpu->shadow_efer = efer;
1553 }
1554
1555 #endif
1556
1557 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1558 {
1559         switch (msr) {
1560 #ifdef CONFIG_X86_64
1561         case MSR_EFER:
1562                 set_efer(vcpu, data);
1563                 break;
1564 #endif
1565         case MSR_IA32_MC0_STATUS:
1566                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1567                        __FUNCTION__, data);
1568                 break;
1569         case MSR_IA32_MCG_STATUS:
1570                 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1571                         __FUNCTION__, data);
1572                 break;
1573         case MSR_IA32_UCODE_REV:
1574         case MSR_IA32_UCODE_WRITE:
1575         case 0x200 ... 0x2ff: /* MTRRs */
1576                 break;
1577         case MSR_IA32_APICBASE:
1578                 vcpu->apic_base = data;
1579                 break;
1580         case MSR_IA32_MISC_ENABLE:
1581                 vcpu->ia32_misc_enable_msr = data;
1582                 break;
1583         /*
1584          * This is the 'probe whether the host is KVM' logic:
1585          */
1586         case MSR_KVM_API_MAGIC:
1587                 return vcpu_register_para(vcpu, data);
1588
1589         default:
1590                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1591                 return 1;
1592         }
1593         return 0;
1594 }
1595 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1596
1597 /*
1598  * Writes msr value into into the appropriate "register".
1599  * Returns 0 on success, non-0 otherwise.
1600  * Assumes vcpu_load() was already called.
1601  */
1602 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1603 {
1604         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1605 }
1606
1607 void kvm_resched(struct kvm_vcpu *vcpu)
1608 {
1609         if (!need_resched())
1610                 return;
1611         cond_resched();
1612 }
1613 EXPORT_SYMBOL_GPL(kvm_resched);
1614
1615 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1616 {
1617         int i;
1618         u32 function;
1619         struct kvm_cpuid_entry *e, *best;
1620
1621         kvm_arch_ops->cache_regs(vcpu);
1622         function = vcpu->regs[VCPU_REGS_RAX];
1623         vcpu->regs[VCPU_REGS_RAX] = 0;
1624         vcpu->regs[VCPU_REGS_RBX] = 0;
1625         vcpu->regs[VCPU_REGS_RCX] = 0;
1626         vcpu->regs[VCPU_REGS_RDX] = 0;
1627         best = NULL;
1628         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1629                 e = &vcpu->cpuid_entries[i];
1630                 if (e->function == function) {
1631                         best = e;
1632                         break;
1633                 }
1634                 /*
1635                  * Both basic or both extended?
1636                  */
1637                 if (((e->function ^ function) & 0x80000000) == 0)
1638                         if (!best || e->function > best->function)
1639                                 best = e;
1640         }
1641         if (best) {
1642                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1643                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1644                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1645                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1646         }
1647         kvm_arch_ops->decache_regs(vcpu);
1648         kvm_arch_ops->skip_emulated_instruction(vcpu);
1649 }
1650 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1651
1652 static int pio_copy_data(struct kvm_vcpu *vcpu)
1653 {
1654         void *p = vcpu->pio_data;
1655         void *q;
1656         unsigned bytes;
1657         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1658
1659         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1660                  PAGE_KERNEL);
1661         if (!q) {
1662                 free_pio_guest_pages(vcpu);
1663                 return -ENOMEM;
1664         }
1665         q += vcpu->pio.guest_page_offset;
1666         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1667         if (vcpu->pio.in)
1668                 memcpy(q, p, bytes);
1669         else
1670                 memcpy(p, q, bytes);
1671         q -= vcpu->pio.guest_page_offset;
1672         vunmap(q);
1673         free_pio_guest_pages(vcpu);
1674         return 0;
1675 }
1676
1677 static int complete_pio(struct kvm_vcpu *vcpu)
1678 {
1679         struct kvm_pio_request *io = &vcpu->pio;
1680         long delta;
1681         int r;
1682
1683         kvm_arch_ops->cache_regs(vcpu);
1684
1685         if (!io->string) {
1686                 if (io->in)
1687                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1688                                io->size);
1689         } else {
1690                 if (io->in) {
1691                         r = pio_copy_data(vcpu);
1692                         if (r) {
1693                                 kvm_arch_ops->cache_regs(vcpu);
1694                                 return r;
1695                         }
1696                 }
1697
1698                 delta = 1;
1699                 if (io->rep) {
1700                         delta *= io->cur_count;
1701                         /*
1702                          * The size of the register should really depend on
1703                          * current address size.
1704                          */
1705                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1706                 }
1707                 if (io->down)
1708                         delta = -delta;
1709                 delta *= io->size;
1710                 if (io->in)
1711                         vcpu->regs[VCPU_REGS_RDI] += delta;
1712                 else
1713                         vcpu->regs[VCPU_REGS_RSI] += delta;
1714         }
1715
1716         kvm_arch_ops->decache_regs(vcpu);
1717
1718         io->count -= io->cur_count;
1719         io->cur_count = 0;
1720
1721         if (!io->count)
1722                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1723         return 0;
1724 }
1725
1726 static void kernel_pio(struct kvm_io_device *pio_dev,
1727                        struct kvm_vcpu *vcpu,
1728                        void *pd)
1729 {
1730         /* TODO: String I/O for in kernel device */
1731
1732         if (vcpu->pio.in)
1733                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1734                                   vcpu->pio.size,
1735                                   pd);
1736         else
1737                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1738                                    vcpu->pio.size,
1739                                    pd);
1740 }
1741
1742 static void pio_string_write(struct kvm_io_device *pio_dev,
1743                              struct kvm_vcpu *vcpu)
1744 {
1745         struct kvm_pio_request *io = &vcpu->pio;
1746         void *pd = vcpu->pio_data;
1747         int i;
1748
1749         for (i = 0; i < io->cur_count; i++) {
1750                 kvm_iodevice_write(pio_dev, io->port,
1751                                    io->size,
1752                                    pd);
1753                 pd += io->size;
1754         }
1755 }
1756
1757 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1758                   int size, unsigned long count, int string, int down,
1759                   gva_t address, int rep, unsigned port)
1760 {
1761         unsigned now, in_page;
1762         int i, ret = 0;
1763         int nr_pages = 1;
1764         struct page *page;
1765         struct kvm_io_device *pio_dev;
1766
1767         vcpu->run->exit_reason = KVM_EXIT_IO;
1768         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1769         vcpu->run->io.size = size;
1770         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1771         vcpu->run->io.count = count;
1772         vcpu->run->io.port = port;
1773         vcpu->pio.count = count;
1774         vcpu->pio.cur_count = count;
1775         vcpu->pio.size = size;
1776         vcpu->pio.in = in;
1777         vcpu->pio.port = port;
1778         vcpu->pio.string = string;
1779         vcpu->pio.down = down;
1780         vcpu->pio.guest_page_offset = offset_in_page(address);
1781         vcpu->pio.rep = rep;
1782
1783         pio_dev = vcpu_find_pio_dev(vcpu, port);
1784         if (!string) {
1785                 kvm_arch_ops->cache_regs(vcpu);
1786                 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1787                 kvm_arch_ops->decache_regs(vcpu);
1788                 if (pio_dev) {
1789                         kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1790                         complete_pio(vcpu);
1791                         return 1;
1792                 }
1793                 return 0;
1794         }
1795
1796         if (!count) {
1797                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1798                 return 1;
1799         }
1800
1801         now = min(count, PAGE_SIZE / size);
1802
1803         if (!down)
1804                 in_page = PAGE_SIZE - offset_in_page(address);
1805         else
1806                 in_page = offset_in_page(address) + size;
1807         now = min(count, (unsigned long)in_page / size);
1808         if (!now) {
1809                 /*
1810                  * String I/O straddles page boundary.  Pin two guest pages
1811                  * so that we satisfy atomicity constraints.  Do just one
1812                  * transaction to avoid complexity.
1813                  */
1814                 nr_pages = 2;
1815                 now = 1;
1816         }
1817         if (down) {
1818                 /*
1819                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1820                  */
1821                 printk(KERN_ERR "kvm: guest string pio down\n");
1822                 inject_gp(vcpu);
1823                 return 1;
1824         }
1825         vcpu->run->io.count = now;
1826         vcpu->pio.cur_count = now;
1827
1828         for (i = 0; i < nr_pages; ++i) {
1829                 mutex_lock(&vcpu->kvm->lock);
1830                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1831                 if (page)
1832                         get_page(page);
1833                 vcpu->pio.guest_pages[i] = page;
1834                 mutex_unlock(&vcpu->kvm->lock);
1835                 if (!page) {
1836                         inject_gp(vcpu);
1837                         free_pio_guest_pages(vcpu);
1838                         return 1;
1839                 }
1840         }
1841
1842         if (!vcpu->pio.in) {
1843                 /* string PIO write */
1844                 ret = pio_copy_data(vcpu);
1845                 if (ret >= 0 && pio_dev) {
1846                         pio_string_write(pio_dev, vcpu);
1847                         complete_pio(vcpu);
1848                         if (vcpu->pio.count == 0)
1849                                 ret = 1;
1850                 }
1851         } else if (pio_dev)
1852                 printk(KERN_ERR "no string pio read support yet, "
1853                        "port %x size %d count %ld\n",
1854                         port, size, count);
1855
1856         return ret;
1857 }
1858 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1859
1860 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1861 {
1862         int r;
1863         sigset_t sigsaved;
1864
1865         vcpu_load(vcpu);
1866
1867         if (vcpu->sigset_active)
1868                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1869
1870         /* re-sync apic's tpr */
1871         vcpu->cr8 = kvm_run->cr8;
1872
1873         if (vcpu->pio.cur_count) {
1874                 r = complete_pio(vcpu);
1875                 if (r)
1876                         goto out;
1877         }
1878
1879         if (vcpu->mmio_needed) {
1880                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1881                 vcpu->mmio_read_completed = 1;
1882                 vcpu->mmio_needed = 0;
1883                 r = emulate_instruction(vcpu, kvm_run,
1884                                         vcpu->mmio_fault_cr2, 0);
1885                 if (r == EMULATE_DO_MMIO) {
1886                         /*
1887                          * Read-modify-write.  Back to userspace.
1888                          */
1889                         r = 0;
1890                         goto out;
1891                 }
1892         }
1893
1894         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1895                 kvm_arch_ops->cache_regs(vcpu);
1896                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1897                 kvm_arch_ops->decache_regs(vcpu);
1898         }
1899
1900         r = kvm_arch_ops->run(vcpu, kvm_run);
1901
1902 out:
1903         if (vcpu->sigset_active)
1904                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1905
1906         vcpu_put(vcpu);
1907         return r;
1908 }
1909
1910 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1911                                    struct kvm_regs *regs)
1912 {
1913         vcpu_load(vcpu);
1914
1915         kvm_arch_ops->cache_regs(vcpu);
1916
1917         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1918         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1919         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1920         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1921         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1922         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1923         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1924         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1925 #ifdef CONFIG_X86_64
1926         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1927         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1928         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1929         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1930         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1931         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1932         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1933         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1934 #endif
1935
1936         regs->rip = vcpu->rip;
1937         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1938
1939         /*
1940          * Don't leak debug flags in case they were set for guest debugging
1941          */
1942         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1943                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1944
1945         vcpu_put(vcpu);
1946
1947         return 0;
1948 }
1949
1950 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1951                                    struct kvm_regs *regs)
1952 {
1953         vcpu_load(vcpu);
1954
1955         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1956         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1957         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1958         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1959         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1960         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1961         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1962         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1963 #ifdef CONFIG_X86_64
1964         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1965         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1966         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1967         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1968         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1969         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1970         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1971         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1972 #endif
1973
1974         vcpu->rip = regs->rip;
1975         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1976
1977         kvm_arch_ops->decache_regs(vcpu);
1978
1979         vcpu_put(vcpu);
1980
1981         return 0;
1982 }
1983
1984 static void get_segment(struct kvm_vcpu *vcpu,
1985                         struct kvm_segment *var, int seg)
1986 {
1987         return kvm_arch_ops->get_segment(vcpu, var, seg);
1988 }
1989
1990 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1991                                     struct kvm_sregs *sregs)
1992 {
1993         struct descriptor_table dt;
1994
1995         vcpu_load(vcpu);
1996
1997         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1998         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1999         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2000         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2001         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2002         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2003
2004         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2005         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2006
2007         kvm_arch_ops->get_idt(vcpu, &dt);
2008         sregs->idt.limit = dt.limit;
2009         sregs->idt.base = dt.base;
2010         kvm_arch_ops->get_gdt(vcpu, &dt);
2011         sregs->gdt.limit = dt.limit;
2012         sregs->gdt.base = dt.base;
2013
2014         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2015         sregs->cr0 = vcpu->cr0;
2016         sregs->cr2 = vcpu->cr2;
2017         sregs->cr3 = vcpu->cr3;
2018         sregs->cr4 = vcpu->cr4;
2019         sregs->cr8 = vcpu->cr8;
2020         sregs->efer = vcpu->shadow_efer;
2021         sregs->apic_base = vcpu->apic_base;
2022
2023         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2024                sizeof sregs->interrupt_bitmap);
2025
2026         vcpu_put(vcpu);
2027
2028         return 0;
2029 }
2030
2031 static void set_segment(struct kvm_vcpu *vcpu,
2032                         struct kvm_segment *var, int seg)
2033 {
2034         return kvm_arch_ops->set_segment(vcpu, var, seg);
2035 }
2036
2037 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2038                                     struct kvm_sregs *sregs)
2039 {
2040         int mmu_reset_needed = 0;
2041         int i;
2042         struct descriptor_table dt;
2043
2044         vcpu_load(vcpu);
2045
2046         dt.limit = sregs->idt.limit;
2047         dt.base = sregs->idt.base;
2048         kvm_arch_ops->set_idt(vcpu, &dt);
2049         dt.limit = sregs->gdt.limit;
2050         dt.base = sregs->gdt.base;
2051         kvm_arch_ops->set_gdt(vcpu, &dt);
2052
2053         vcpu->cr2 = sregs->cr2;
2054         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2055         vcpu->cr3 = sregs->cr3;
2056
2057         vcpu->cr8 = sregs->cr8;
2058
2059         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2060 #ifdef CONFIG_X86_64
2061         kvm_arch_ops->set_efer(vcpu, sregs->efer);
2062 #endif
2063         vcpu->apic_base = sregs->apic_base;
2064
2065         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2066
2067         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2068         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2069
2070         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2071         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2072         if (!is_long_mode(vcpu) && is_pae(vcpu))
2073                 load_pdptrs(vcpu, vcpu->cr3);
2074
2075         if (mmu_reset_needed)
2076                 kvm_mmu_reset_context(vcpu);
2077
2078         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2079                sizeof vcpu->irq_pending);
2080         vcpu->irq_summary = 0;
2081         for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2082                 if (vcpu->irq_pending[i])
2083                         __set_bit(i, &vcpu->irq_summary);
2084
2085         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2086         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2087         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2088         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2089         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2090         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2091
2092         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2093         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2094
2095         vcpu_put(vcpu);
2096
2097         return 0;
2098 }
2099
2100 /*
2101  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2102  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2103  *
2104  * This list is modified at module load time to reflect the
2105  * capabilities of the host cpu.
2106  */
2107 static u32 msrs_to_save[] = {
2108         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2109         MSR_K6_STAR,
2110 #ifdef CONFIG_X86_64
2111         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2112 #endif
2113         MSR_IA32_TIME_STAMP_COUNTER,
2114 };
2115
2116 static unsigned num_msrs_to_save;
2117
2118 static u32 emulated_msrs[] = {
2119         MSR_IA32_MISC_ENABLE,
2120 };
2121
2122 static __init void kvm_init_msr_list(void)
2123 {
2124         u32 dummy[2];
2125         unsigned i, j;
2126
2127         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2128                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2129                         continue;
2130                 if (j < i)
2131                         msrs_to_save[j] = msrs_to_save[i];
2132                 j++;
2133         }
2134         num_msrs_to_save = j;
2135 }
2136
2137 /*
2138  * Adapt set_msr() to msr_io()'s calling convention
2139  */
2140 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2141 {
2142         return kvm_set_msr(vcpu, index, *data);
2143 }
2144
2145 /*
2146  * Read or write a bunch of msrs. All parameters are kernel addresses.
2147  *
2148  * @return number of msrs set successfully.
2149  */
2150 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2151                     struct kvm_msr_entry *entries,
2152                     int (*do_msr)(struct kvm_vcpu *vcpu,
2153                                   unsigned index, u64 *data))
2154 {
2155         int i;
2156
2157         vcpu_load(vcpu);
2158
2159         for (i = 0; i < msrs->nmsrs; ++i)
2160                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2161                         break;
2162
2163         vcpu_put(vcpu);
2164
2165         return i;
2166 }
2167
2168 /*
2169  * Read or write a bunch of msrs. Parameters are user addresses.
2170  *
2171  * @return number of msrs set successfully.
2172  */
2173 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2174                   int (*do_msr)(struct kvm_vcpu *vcpu,
2175                                 unsigned index, u64 *data),
2176                   int writeback)
2177 {
2178         struct kvm_msrs msrs;
2179         struct kvm_msr_entry *entries;
2180         int r, n;
2181         unsigned size;
2182
2183         r = -EFAULT;
2184         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2185                 goto out;
2186
2187         r = -E2BIG;
2188         if (msrs.nmsrs >= MAX_IO_MSRS)
2189                 goto out;
2190
2191         r = -ENOMEM;
2192         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2193         entries = vmalloc(size);
2194         if (!entries)
2195                 goto out;
2196
2197         r = -EFAULT;
2198         if (copy_from_user(entries, user_msrs->entries, size))
2199                 goto out_free;
2200
2201         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2202         if (r < 0)
2203                 goto out_free;
2204
2205         r = -EFAULT;
2206         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2207                 goto out_free;
2208
2209         r = n;
2210
2211 out_free:
2212         vfree(entries);
2213 out:
2214         return r;
2215 }
2216
2217 /*
2218  * Translate a guest virtual address to a guest physical address.
2219  */
2220 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2221                                     struct kvm_translation *tr)
2222 {
2223         unsigned long vaddr = tr->linear_address;
2224         gpa_t gpa;
2225
2226         vcpu_load(vcpu);
2227         mutex_lock(&vcpu->kvm->lock);
2228         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2229         tr->physical_address = gpa;
2230         tr->valid = gpa != UNMAPPED_GVA;
2231         tr->writeable = 1;
2232         tr->usermode = 0;
2233         mutex_unlock(&vcpu->kvm->lock);
2234         vcpu_put(vcpu);
2235
2236         return 0;
2237 }
2238
2239 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2240                                     struct kvm_interrupt *irq)
2241 {
2242         if (irq->irq < 0 || irq->irq >= 256)
2243                 return -EINVAL;
2244         vcpu_load(vcpu);
2245
2246         set_bit(irq->irq, vcpu->irq_pending);
2247         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2248
2249         vcpu_put(vcpu);
2250
2251         return 0;
2252 }
2253
2254 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2255                                       struct kvm_debug_guest *dbg)
2256 {
2257         int r;
2258
2259         vcpu_load(vcpu);
2260
2261         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2262
2263         vcpu_put(vcpu);
2264
2265         return r;
2266 }
2267
2268 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2269                                     unsigned long address,
2270                                     int *type)
2271 {
2272         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2273         unsigned long pgoff;
2274         struct page *page;
2275
2276         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2277         if (pgoff == 0)
2278                 page = virt_to_page(vcpu->run);
2279         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2280                 page = virt_to_page(vcpu->pio_data);
2281         else
2282                 return NOPAGE_SIGBUS;
2283         get_page(page);
2284         if (type != NULL)
2285                 *type = VM_FAULT_MINOR;
2286
2287         return page;
2288 }
2289
2290 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2291         .nopage = kvm_vcpu_nopage,
2292 };
2293
2294 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2295 {
2296         vma->vm_ops = &kvm_vcpu_vm_ops;
2297         return 0;
2298 }
2299
2300 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2301 {
2302         struct kvm_vcpu *vcpu = filp->private_data;
2303
2304         fput(vcpu->kvm->filp);
2305         return 0;
2306 }
2307
2308 static struct file_operations kvm_vcpu_fops = {
2309         .release        = kvm_vcpu_release,
2310         .unlocked_ioctl = kvm_vcpu_ioctl,
2311         .compat_ioctl   = kvm_vcpu_ioctl,
2312         .mmap           = kvm_vcpu_mmap,
2313 };
2314
2315 /*
2316  * Allocates an inode for the vcpu.
2317  */
2318 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2319 {
2320         int fd, r;
2321         struct inode *inode;
2322         struct file *file;
2323
2324         r = anon_inode_getfd(&fd, &inode, &file,
2325                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2326         if (r)
2327                 return r;
2328         atomic_inc(&vcpu->kvm->filp->f_count);
2329         return fd;
2330 }
2331
2332 /*
2333  * Creates some virtual cpus.  Good luck creating more than one.
2334  */
2335 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2336 {
2337         int r;
2338         struct kvm_vcpu *vcpu;
2339
2340         if (!valid_vcpu(n))
2341                 return -EINVAL;
2342
2343         vcpu = kvm_arch_ops->vcpu_create(kvm, n);
2344         if (IS_ERR(vcpu))
2345                 return PTR_ERR(vcpu);
2346
2347         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2348
2349         /* We do fxsave: this must be aligned. */
2350         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2351
2352         vcpu_load(vcpu);
2353         r = kvm_mmu_setup(vcpu);
2354         vcpu_put(vcpu);
2355         if (r < 0)
2356                 goto free_vcpu;
2357
2358         mutex_lock(&kvm->lock);
2359         if (kvm->vcpus[n]) {
2360                 r = -EEXIST;
2361                 mutex_unlock(&kvm->lock);
2362                 goto mmu_unload;
2363         }
2364         kvm->vcpus[n] = vcpu;
2365         mutex_unlock(&kvm->lock);
2366
2367         /* Now it's all set up, let userspace reach it */
2368         r = create_vcpu_fd(vcpu);
2369         if (r < 0)
2370                 goto unlink;
2371         return r;
2372
2373 unlink:
2374         mutex_lock(&kvm->lock);
2375         kvm->vcpus[n] = NULL;
2376         mutex_unlock(&kvm->lock);
2377
2378 mmu_unload:
2379         vcpu_load(vcpu);
2380         kvm_mmu_unload(vcpu);
2381         vcpu_put(vcpu);
2382
2383 free_vcpu:
2384         kvm_arch_ops->vcpu_free(vcpu);
2385         return r;
2386 }
2387
2388 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2389 {
2390         u64 efer;
2391         int i;
2392         struct kvm_cpuid_entry *e, *entry;
2393
2394         rdmsrl(MSR_EFER, efer);
2395         entry = NULL;
2396         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2397                 e = &vcpu->cpuid_entries[i];
2398                 if (e->function == 0x80000001) {
2399                         entry = e;
2400                         break;
2401                 }
2402         }
2403         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2404                 entry->edx &= ~(1 << 20);
2405                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2406         }
2407 }
2408
2409 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2410                                     struct kvm_cpuid *cpuid,
2411                                     struct kvm_cpuid_entry __user *entries)
2412 {
2413         int r;
2414
2415         r = -E2BIG;
2416         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2417                 goto out;
2418         r = -EFAULT;
2419         if (copy_from_user(&vcpu->cpuid_entries, entries,
2420                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2421                 goto out;
2422         vcpu->cpuid_nent = cpuid->nent;
2423         cpuid_fix_nx_cap(vcpu);
2424         return 0;
2425
2426 out:
2427         return r;
2428 }
2429
2430 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2431 {
2432         if (sigset) {
2433                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2434                 vcpu->sigset_active = 1;
2435                 vcpu->sigset = *sigset;
2436         } else
2437                 vcpu->sigset_active = 0;
2438         return 0;
2439 }
2440
2441 /*
2442  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2443  * we have asm/x86/processor.h
2444  */
2445 struct fxsave {
2446         u16     cwd;
2447         u16     swd;
2448         u16     twd;
2449         u16     fop;
2450         u64     rip;
2451         u64     rdp;
2452         u32     mxcsr;
2453         u32     mxcsr_mask;
2454         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2455 #ifdef CONFIG_X86_64
2456         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2457 #else
2458         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2459 #endif
2460 };
2461
2462 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2463 {
2464         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2465
2466         vcpu_load(vcpu);
2467
2468         memcpy(fpu->fpr, fxsave->st_space, 128);
2469         fpu->fcw = fxsave->cwd;
2470         fpu->fsw = fxsave->swd;
2471         fpu->ftwx = fxsave->twd;
2472         fpu->last_opcode = fxsave->fop;
2473         fpu->last_ip = fxsave->rip;
2474         fpu->last_dp = fxsave->rdp;
2475         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2476
2477         vcpu_put(vcpu);
2478
2479         return 0;
2480 }
2481
2482 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2483 {
2484         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2485
2486         vcpu_load(vcpu);
2487
2488         memcpy(fxsave->st_space, fpu->fpr, 128);
2489         fxsave->cwd = fpu->fcw;
2490         fxsave->swd = fpu->fsw;
2491         fxsave->twd = fpu->ftwx;
2492         fxsave->fop = fpu->last_opcode;
2493         fxsave->rip = fpu->last_ip;
2494         fxsave->rdp = fpu->last_dp;
2495         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2496
2497         vcpu_put(vcpu);
2498
2499         return 0;
2500 }
2501
2502 static long kvm_vcpu_ioctl(struct file *filp,
2503                            unsigned int ioctl, unsigned long arg)
2504 {
2505         struct kvm_vcpu *vcpu = filp->private_data;
2506         void __user *argp = (void __user *)arg;
2507         int r = -EINVAL;
2508
2509         switch (ioctl) {
2510         case KVM_RUN:
2511                 r = -EINVAL;
2512                 if (arg)
2513                         goto out;
2514                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2515                 break;
2516         case KVM_GET_REGS: {
2517                 struct kvm_regs kvm_regs;
2518
2519                 memset(&kvm_regs, 0, sizeof kvm_regs);
2520                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2521                 if (r)
2522                         goto out;
2523                 r = -EFAULT;
2524                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2525                         goto out;
2526                 r = 0;
2527                 break;
2528         }
2529         case KVM_SET_REGS: {
2530                 struct kvm_regs kvm_regs;
2531
2532                 r = -EFAULT;
2533                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2534                         goto out;
2535                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2536                 if (r)
2537                         goto out;
2538                 r = 0;
2539                 break;
2540         }
2541         case KVM_GET_SREGS: {
2542                 struct kvm_sregs kvm_sregs;
2543
2544                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2545                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2546                 if (r)
2547                         goto out;
2548                 r = -EFAULT;
2549                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2550                         goto out;
2551                 r = 0;
2552                 break;
2553         }
2554         case KVM_SET_SREGS: {
2555                 struct kvm_sregs kvm_sregs;
2556
2557                 r = -EFAULT;
2558                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2559                         goto out;
2560                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2561                 if (r)
2562                         goto out;
2563                 r = 0;
2564                 break;
2565         }
2566         case KVM_TRANSLATE: {
2567                 struct kvm_translation tr;
2568
2569                 r = -EFAULT;
2570                 if (copy_from_user(&tr, argp, sizeof tr))
2571                         goto out;
2572                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2573                 if (r)
2574                         goto out;
2575                 r = -EFAULT;
2576                 if (copy_to_user(argp, &tr, sizeof tr))
2577                         goto out;
2578                 r = 0;
2579                 break;
2580         }
2581         case KVM_INTERRUPT: {
2582                 struct kvm_interrupt irq;
2583
2584                 r = -EFAULT;
2585                 if (copy_from_user(&irq, argp, sizeof irq))
2586                         goto out;
2587                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2588                 if (r)
2589                         goto out;
2590                 r = 0;
2591                 break;
2592         }
2593         case KVM_DEBUG_GUEST: {
2594                 struct kvm_debug_guest dbg;
2595
2596                 r = -EFAULT;
2597                 if (copy_from_user(&dbg, argp, sizeof dbg))
2598                         goto out;
2599                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2600                 if (r)
2601                         goto out;
2602                 r = 0;
2603                 break;
2604         }
2605         case KVM_GET_MSRS:
2606                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2607                 break;
2608         case KVM_SET_MSRS:
2609                 r = msr_io(vcpu, argp, do_set_msr, 0);
2610                 break;
2611         case KVM_SET_CPUID: {
2612                 struct kvm_cpuid __user *cpuid_arg = argp;
2613                 struct kvm_cpuid cpuid;
2614
2615                 r = -EFAULT;
2616                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2617                         goto out;
2618                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2619                 if (r)
2620                         goto out;
2621                 break;
2622         }
2623         case KVM_SET_SIGNAL_MASK: {
2624                 struct kvm_signal_mask __user *sigmask_arg = argp;
2625                 struct kvm_signal_mask kvm_sigmask;
2626                 sigset_t sigset, *p;
2627
2628                 p = NULL;
2629                 if (argp) {
2630                         r = -EFAULT;
2631                         if (copy_from_user(&kvm_sigmask, argp,
2632                                            sizeof kvm_sigmask))
2633                                 goto out;
2634                         r = -EINVAL;
2635                         if (kvm_sigmask.len != sizeof sigset)
2636                                 goto out;
2637                         r = -EFAULT;
2638                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2639                                            sizeof sigset))
2640                                 goto out;
2641                         p = &sigset;
2642                 }
2643                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2644                 break;
2645         }
2646         case KVM_GET_FPU: {
2647                 struct kvm_fpu fpu;
2648
2649                 memset(&fpu, 0, sizeof fpu);
2650                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2651                 if (r)
2652                         goto out;
2653                 r = -EFAULT;
2654                 if (copy_to_user(argp, &fpu, sizeof fpu))
2655                         goto out;
2656                 r = 0;
2657                 break;
2658         }
2659         case KVM_SET_FPU: {
2660                 struct kvm_fpu fpu;
2661
2662                 r = -EFAULT;
2663                 if (copy_from_user(&fpu, argp, sizeof fpu))
2664                         goto out;
2665                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2666                 if (r)
2667                         goto out;
2668                 r = 0;
2669                 break;
2670         }
2671         default:
2672                 ;
2673         }
2674 out:
2675         return r;
2676 }
2677
2678 static long kvm_vm_ioctl(struct file *filp,
2679                            unsigned int ioctl, unsigned long arg)
2680 {
2681         struct kvm *kvm = filp->private_data;
2682         void __user *argp = (void __user *)arg;
2683         int r = -EINVAL;
2684
2685         switch (ioctl) {
2686         case KVM_CREATE_VCPU:
2687                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2688                 if (r < 0)
2689                         goto out;
2690                 break;
2691         case KVM_SET_MEMORY_REGION: {
2692                 struct kvm_memory_region kvm_mem;
2693
2694                 r = -EFAULT;
2695                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2696                         goto out;
2697                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2698                 if (r)
2699                         goto out;
2700                 break;
2701         }
2702         case KVM_GET_DIRTY_LOG: {
2703                 struct kvm_dirty_log log;
2704
2705                 r = -EFAULT;
2706                 if (copy_from_user(&log, argp, sizeof log))
2707                         goto out;
2708                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2709                 if (r)
2710                         goto out;
2711                 break;
2712         }
2713         case KVM_SET_MEMORY_ALIAS: {
2714                 struct kvm_memory_alias alias;
2715
2716                 r = -EFAULT;
2717                 if (copy_from_user(&alias, argp, sizeof alias))
2718                         goto out;
2719                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2720                 if (r)
2721                         goto out;
2722                 break;
2723         }
2724         default:
2725                 ;
2726         }
2727 out:
2728         return r;
2729 }
2730
2731 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2732                                   unsigned long address,
2733                                   int *type)
2734 {
2735         struct kvm *kvm = vma->vm_file->private_data;
2736         unsigned long pgoff;
2737         struct page *page;
2738
2739         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2740         page = gfn_to_page(kvm, pgoff);
2741         if (!page)
2742                 return NOPAGE_SIGBUS;
2743         get_page(page);
2744         if (type != NULL)
2745                 *type = VM_FAULT_MINOR;
2746
2747         return page;
2748 }
2749
2750 static struct vm_operations_struct kvm_vm_vm_ops = {
2751         .nopage = kvm_vm_nopage,
2752 };
2753
2754 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2755 {
2756         vma->vm_ops = &kvm_vm_vm_ops;
2757         return 0;
2758 }
2759
2760 static struct file_operations kvm_vm_fops = {
2761         .release        = kvm_vm_release,
2762         .unlocked_ioctl = kvm_vm_ioctl,
2763         .compat_ioctl   = kvm_vm_ioctl,
2764         .mmap           = kvm_vm_mmap,
2765 };
2766
2767 static int kvm_dev_ioctl_create_vm(void)
2768 {
2769         int fd, r;
2770         struct inode *inode;
2771         struct file *file;
2772         struct kvm *kvm;
2773
2774         kvm = kvm_create_vm();
2775         if (IS_ERR(kvm))
2776                 return PTR_ERR(kvm);
2777         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2778         if (r) {
2779                 kvm_destroy_vm(kvm);
2780                 return r;
2781         }
2782
2783         kvm->filp = file;
2784
2785         return fd;
2786 }
2787
2788 static long kvm_dev_ioctl(struct file *filp,
2789                           unsigned int ioctl, unsigned long arg)
2790 {
2791         void __user *argp = (void __user *)arg;
2792         long r = -EINVAL;
2793
2794         switch (ioctl) {
2795         case KVM_GET_API_VERSION:
2796                 r = -EINVAL;
2797                 if (arg)
2798                         goto out;
2799                 r = KVM_API_VERSION;
2800                 break;
2801         case KVM_CREATE_VM:
2802                 r = -EINVAL;
2803                 if (arg)
2804                         goto out;
2805                 r = kvm_dev_ioctl_create_vm();
2806                 break;
2807         case KVM_GET_MSR_INDEX_LIST: {
2808                 struct kvm_msr_list __user *user_msr_list = argp;
2809                 struct kvm_msr_list msr_list;
2810                 unsigned n;
2811
2812                 r = -EFAULT;
2813                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2814                         goto out;
2815                 n = msr_list.nmsrs;
2816                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2817                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2818                         goto out;
2819                 r = -E2BIG;
2820                 if (n < num_msrs_to_save)
2821                         goto out;
2822                 r = -EFAULT;
2823                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2824                                  num_msrs_to_save * sizeof(u32)))
2825                         goto out;
2826                 if (copy_to_user(user_msr_list->indices
2827                                  + num_msrs_to_save * sizeof(u32),
2828                                  &emulated_msrs,
2829                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2830                         goto out;
2831                 r = 0;
2832                 break;
2833         }
2834         case KVM_CHECK_EXTENSION:
2835                 /*
2836                  * No extensions defined at present.
2837                  */
2838                 r = 0;
2839                 break;
2840         case KVM_GET_VCPU_MMAP_SIZE:
2841                 r = -EINVAL;
2842                 if (arg)
2843                         goto out;
2844                 r = 2 * PAGE_SIZE;
2845                 break;
2846         default:
2847                 ;
2848         }
2849 out:
2850         return r;
2851 }
2852
2853 static struct file_operations kvm_chardev_ops = {
2854         .open           = kvm_dev_open,
2855         .release        = kvm_dev_release,
2856         .unlocked_ioctl = kvm_dev_ioctl,
2857         .compat_ioctl   = kvm_dev_ioctl,
2858 };
2859
2860 static struct miscdevice kvm_dev = {
2861         KVM_MINOR,
2862         "kvm",
2863         &kvm_chardev_ops,
2864 };
2865
2866 /*
2867  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2868  * cached on it.
2869  */
2870 static void decache_vcpus_on_cpu(int cpu)
2871 {
2872         struct kvm *vm;
2873         struct kvm_vcpu *vcpu;
2874         int i;
2875
2876         spin_lock(&kvm_lock);
2877         list_for_each_entry(vm, &vm_list, vm_list)
2878                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2879                         vcpu = vm->vcpus[i];
2880                         if (!vcpu)
2881                                 continue;
2882                         /*
2883                          * If the vcpu is locked, then it is running on some
2884                          * other cpu and therefore it is not cached on the
2885                          * cpu in question.
2886                          *
2887                          * If it's not locked, check the last cpu it executed
2888                          * on.
2889                          */
2890                         if (mutex_trylock(&vcpu->mutex)) {
2891                                 if (vcpu->cpu == cpu) {
2892                                         kvm_arch_ops->vcpu_decache(vcpu);
2893                                         vcpu->cpu = -1;
2894                                 }
2895                                 mutex_unlock(&vcpu->mutex);
2896                         }
2897                 }
2898         spin_unlock(&kvm_lock);
2899 }
2900
2901 static void hardware_enable(void *junk)
2902 {
2903         int cpu = raw_smp_processor_id();
2904
2905         if (cpu_isset(cpu, cpus_hardware_enabled))
2906                 return;
2907         cpu_set(cpu, cpus_hardware_enabled);
2908         kvm_arch_ops->hardware_enable(NULL);
2909 }
2910
2911 static void hardware_disable(void *junk)
2912 {
2913         int cpu = raw_smp_processor_id();
2914
2915         if (!cpu_isset(cpu, cpus_hardware_enabled))
2916                 return;
2917         cpu_clear(cpu, cpus_hardware_enabled);
2918         decache_vcpus_on_cpu(cpu);
2919         kvm_arch_ops->hardware_disable(NULL);
2920 }
2921
2922 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2923                            void *v)
2924 {
2925         int cpu = (long)v;
2926
2927         switch (val) {
2928         case CPU_DYING:
2929         case CPU_DYING_FROZEN:
2930                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2931                        cpu);
2932                 hardware_disable(NULL);
2933                 break;
2934         case CPU_UP_CANCELED:
2935         case CPU_UP_CANCELED_FROZEN:
2936                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2937                        cpu);
2938                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
2939                 break;
2940         case CPU_ONLINE:
2941         case CPU_ONLINE_FROZEN:
2942                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2943                        cpu);
2944                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
2945                 break;
2946         }
2947         return NOTIFY_OK;
2948 }
2949
2950 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2951                        void *v)
2952 {
2953         if (val == SYS_RESTART) {
2954                 /*
2955                  * Some (well, at least mine) BIOSes hang on reboot if
2956                  * in vmx root mode.
2957                  */
2958                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2959                 on_each_cpu(hardware_disable, NULL, 0, 1);
2960         }
2961         return NOTIFY_OK;
2962 }
2963
2964 static struct notifier_block kvm_reboot_notifier = {
2965         .notifier_call = kvm_reboot,
2966         .priority = 0,
2967 };
2968
2969 void kvm_io_bus_init(struct kvm_io_bus *bus)
2970 {
2971         memset(bus, 0, sizeof(*bus));
2972 }
2973
2974 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2975 {
2976         int i;
2977
2978         for (i = 0; i < bus->dev_count; i++) {
2979                 struct kvm_io_device *pos = bus->devs[i];
2980
2981                 kvm_iodevice_destructor(pos);
2982         }
2983 }
2984
2985 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
2986 {
2987         int i;
2988
2989         for (i = 0; i < bus->dev_count; i++) {
2990                 struct kvm_io_device *pos = bus->devs[i];
2991
2992                 if (pos->in_range(pos, addr))
2993                         return pos;
2994         }
2995
2996         return NULL;
2997 }
2998
2999 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3000 {
3001         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3002
3003         bus->devs[bus->dev_count++] = dev;
3004 }
3005
3006 static struct notifier_block kvm_cpu_notifier = {
3007         .notifier_call = kvm_cpu_hotplug,
3008         .priority = 20, /* must be > scheduler priority */
3009 };
3010
3011 static u64 stat_get(void *_offset)
3012 {
3013         unsigned offset = (long)_offset;
3014         u64 total = 0;
3015         struct kvm *kvm;
3016         struct kvm_vcpu *vcpu;
3017         int i;
3018
3019         spin_lock(&kvm_lock);
3020         list_for_each_entry(kvm, &vm_list, vm_list)
3021                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3022                         vcpu = kvm->vcpus[i];
3023                         if (vcpu)
3024                                 total += *(u32 *)((void *)vcpu + offset);
3025                 }
3026         spin_unlock(&kvm_lock);
3027         return total;
3028 }
3029
3030 static void stat_set(void *offset, u64 val)
3031 {
3032 }
3033
3034 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
3035
3036 static __init void kvm_init_debug(void)
3037 {
3038         struct kvm_stats_debugfs_item *p;
3039
3040         debugfs_dir = debugfs_create_dir("kvm", NULL);
3041         for (p = debugfs_entries; p->name; ++p)
3042                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3043                                                 (void *)(long)p->offset,
3044                                                 &stat_fops);
3045 }
3046
3047 static void kvm_exit_debug(void)
3048 {
3049         struct kvm_stats_debugfs_item *p;
3050
3051         for (p = debugfs_entries; p->name; ++p)
3052                 debugfs_remove(p->dentry);
3053         debugfs_remove(debugfs_dir);
3054 }
3055
3056 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3057 {
3058         hardware_disable(NULL);
3059         return 0;
3060 }
3061
3062 static int kvm_resume(struct sys_device *dev)
3063 {
3064         hardware_enable(NULL);
3065         return 0;
3066 }
3067
3068 static struct sysdev_class kvm_sysdev_class = {
3069         set_kset_name("kvm"),
3070         .suspend = kvm_suspend,
3071         .resume = kvm_resume,
3072 };
3073
3074 static struct sys_device kvm_sysdev = {
3075         .id = 0,
3076         .cls = &kvm_sysdev_class,
3077 };
3078
3079 hpa_t bad_page_address;
3080
3081 static inline
3082 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3083 {
3084         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3085 }
3086
3087 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3088 {
3089         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3090
3091         kvm_arch_ops->vcpu_load(vcpu, cpu);
3092 }
3093
3094 static void kvm_sched_out(struct preempt_notifier *pn,
3095                           struct task_struct *next)
3096 {
3097         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3098
3099         kvm_arch_ops->vcpu_put(vcpu);
3100 }
3101
3102 int kvm_init_arch(struct kvm_arch_ops *ops, unsigned int vcpu_size,
3103                   struct module *module)
3104 {
3105         int r;
3106         int cpu;
3107
3108         if (kvm_arch_ops) {
3109                 printk(KERN_ERR "kvm: already loaded the other module\n");
3110                 return -EEXIST;
3111         }
3112
3113         if (!ops->cpu_has_kvm_support()) {
3114                 printk(KERN_ERR "kvm: no hardware support\n");
3115                 return -EOPNOTSUPP;
3116         }
3117         if (ops->disabled_by_bios()) {
3118                 printk(KERN_ERR "kvm: disabled by bios\n");
3119                 return -EOPNOTSUPP;
3120         }
3121
3122         kvm_arch_ops = ops;
3123
3124         r = kvm_arch_ops->hardware_setup();
3125         if (r < 0)
3126                 goto out;
3127
3128         for_each_online_cpu(cpu) {
3129                 smp_call_function_single(cpu,
3130                                 kvm_arch_ops->check_processor_compatibility,
3131                                 &r, 0, 1);
3132                 if (r < 0)
3133                         goto out_free_0;
3134         }
3135
3136         on_each_cpu(hardware_enable, NULL, 0, 1);
3137         r = register_cpu_notifier(&kvm_cpu_notifier);
3138         if (r)
3139                 goto out_free_1;
3140         register_reboot_notifier(&kvm_reboot_notifier);
3141
3142         r = sysdev_class_register(&kvm_sysdev_class);
3143         if (r)
3144                 goto out_free_2;
3145
3146         r = sysdev_register(&kvm_sysdev);
3147         if (r)
3148                 goto out_free_3;
3149
3150         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3151         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3152                                            __alignof__(struct kvm_vcpu), 0, 0);
3153         if (!kvm_vcpu_cache) {
3154                 r = -ENOMEM;
3155                 goto out_free_4;
3156         }
3157
3158         kvm_chardev_ops.owner = module;
3159
3160         r = misc_register(&kvm_dev);
3161         if (r) {
3162                 printk (KERN_ERR "kvm: misc device register failed\n");
3163                 goto out_free;
3164         }
3165
3166         kvm_preempt_ops.sched_in = kvm_sched_in;
3167         kvm_preempt_ops.sched_out = kvm_sched_out;
3168
3169         return r;
3170
3171 out_free:
3172         kmem_cache_destroy(kvm_vcpu_cache);
3173 out_free_4:
3174         sysdev_unregister(&kvm_sysdev);
3175 out_free_3:
3176         sysdev_class_unregister(&kvm_sysdev_class);
3177 out_free_2:
3178         unregister_reboot_notifier(&kvm_reboot_notifier);
3179         unregister_cpu_notifier(&kvm_cpu_notifier);
3180 out_free_1:
3181         on_each_cpu(hardware_disable, NULL, 0, 1);
3182 out_free_0:
3183         kvm_arch_ops->hardware_unsetup();
3184 out:
3185         kvm_arch_ops = NULL;
3186         return r;
3187 }
3188
3189 void kvm_exit_arch(void)
3190 {
3191         misc_deregister(&kvm_dev);
3192         kmem_cache_destroy(kvm_vcpu_cache);
3193         sysdev_unregister(&kvm_sysdev);
3194         sysdev_class_unregister(&kvm_sysdev_class);
3195         unregister_reboot_notifier(&kvm_reboot_notifier);
3196         unregister_cpu_notifier(&kvm_cpu_notifier);
3197         on_each_cpu(hardware_disable, NULL, 0, 1);
3198         kvm_arch_ops->hardware_unsetup();
3199         kvm_arch_ops = NULL;
3200 }
3201
3202 static __init int kvm_init(void)
3203 {
3204         static struct page *bad_page;
3205         int r;
3206
3207         r = kvm_mmu_module_init();
3208         if (r)
3209                 goto out4;
3210
3211         kvm_init_debug();
3212
3213         kvm_init_msr_list();
3214
3215         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3216                 r = -ENOMEM;
3217                 goto out;
3218         }
3219
3220         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3221         memset(__va(bad_page_address), 0, PAGE_SIZE);
3222
3223         return 0;
3224
3225 out:
3226         kvm_exit_debug();
3227         kvm_mmu_module_exit();
3228 out4:
3229         return r;
3230 }
3231
3232 static __exit void kvm_exit(void)
3233 {
3234         kvm_exit_debug();
3235         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3236         kvm_mmu_module_exit();
3237 }
3238
3239 module_init(kvm_init)
3240 module_exit(kvm_exit)
3241
3242 EXPORT_SYMBOL_GPL(kvm_init_arch);
3243 EXPORT_SYMBOL_GPL(kvm_exit_arch);