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