drivers/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19 #include "x86_emulate.h"
  20 #include "segment_descriptor.h"
  21 #include "irq.h"
  22
  23 #include <linux/kvm.h>
  24 #include <linux/module.h>
  25 #include <linux/errno.h>
  26 #include <linux/percpu.h>
  27 #include <linux/gfp.h>
  28 #include <linux/mm.h>
  29 #include <linux/miscdevice.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/reboot.h>
  32 #include <linux/debugfs.h>
  33 #include <linux/highmem.h>
  34 #include <linux/file.h>
  35 #include <linux/sysdev.h>
  36 #include <linux/cpu.h>
  37 #include <linux/sched.h>
  38 #include <linux/cpumask.h>
  39 #include <linux/smp.h>
  40 #include <linux/anon_inodes.h>
  41
  42 #include <asm/processor.h>
  43 #include <asm/msr.h>
  44 #include <asm/io.h>
  45 #include <asm/uaccess.h>
  46 #include <asm/desc.h>
  47
  48 MODULE_AUTHOR("Qumranet");
  49 MODULE_LICENSE("GPL");
  50
  51 static DEFINE_SPINLOCK(kvm_lock);
  52 static LIST_HEAD(vm_list);
  53
  54 static cpumask_t cpus_hardware_enabled;
  55
  56 struct kvm_arch_ops *kvm_arch_ops;
  57 struct kmem_cache *kvm_vcpu_cache;
  58 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  59
  60 static __read_mostly struct preempt_ops kvm_preempt_ops;
  61
  62 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  63
  64 static struct kvm_stats_debugfs_item {
  65         const char *name;
  66         int offset;
  67         struct dentry *dentry;
  68 } debugfs_entries[] = {
  69         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  70         { "pf_guest", STAT_OFFSET(pf_guest) },
  71         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  72         { "invlpg", STAT_OFFSET(invlpg) },
  73         { "exits", STAT_OFFSET(exits) },
  74         { "io_exits", STAT_OFFSET(io_exits) },
  75         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  76         { "signal_exits", STAT_OFFSET(signal_exits) },
  77         { "irq_window", STAT_OFFSET(irq_window_exits) },
  78         { "halt_exits", STAT_OFFSET(halt_exits) },
  79         { "request_irq", STAT_OFFSET(request_irq_exits) },
  80         { "irq_exits", STAT_OFFSET(irq_exits) },
  81         { "light_exits", STAT_OFFSET(light_exits) },
  82         { "efer_reload", STAT_OFFSET(efer_reload) },
  83         { NULL }
  84 };
  85
  86 static struct dentry *debugfs_dir;
  87
  88 #define MAX_IO_MSRS 256
  89
  90 #define CR0_RESERVED_BITS                                               \
  91         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  92                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  93                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  94 #define CR4_RESERVED_BITS                                               \
  95         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  96                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  97                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  98                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  99
 100 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
 101 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
 102
 103 #ifdef CONFIG_X86_64
 104 // LDT or TSS descriptor in the GDT. 16 bytes.
 105 struct segment_descriptor_64 {
 106         struct segment_descriptor s;
 107         u32 base_higher;
 108         u32 pad_zero;
 109 };
 110
 111 #endif
 112
 113 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 114                            unsigned long arg);
 115
 116 unsigned long segment_base(u16 selector)
 117 {
 118         struct descriptor_table gdt;
 119         struct segment_descriptor *d;
 120         unsigned long table_base;
 121         typedef unsigned long ul;
 122         unsigned long v;
 123
 124         if (selector == 0)
 125                 return 0;
 126
 127         asm ("sgdt %0" : "=m"(gdt));
 128         table_base = gdt.base;
 129
 130         if (selector & 4) {           /* from ldt */
 131                 u16 ldt_selector;
 132
 133                 asm ("sldt %0" : "=g"(ldt_selector));
 134                 table_base = segment_base(ldt_selector);
 135         }
 136         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 137         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 138 #ifdef CONFIG_X86_64
 139         if (d->system == 0
 140             && (d->type == 2 || d->type == 9 || d->type == 11))
 141                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 142 #endif
 143         return v;
 144 }
 145 EXPORT_SYMBOL_GPL(segment_base);
 146
 147 static inline int valid_vcpu(int n)
 148 {
 149         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 150 }
 151
 152 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 153 {
 154         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 155                 return;
 156
 157         vcpu->guest_fpu_loaded = 1;
 158         fx_save(&vcpu->host_fx_image);
 159         fx_restore(&vcpu->guest_fx_image);
 160 }
 161 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 162
 163 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 164 {
 165         if (!vcpu->guest_fpu_loaded)
 166                 return;
 167
 168         vcpu->guest_fpu_loaded = 0;
 169         fx_save(&vcpu->guest_fx_image);
 170         fx_restore(&vcpu->host_fx_image);
 171 }
 172 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 173
 174 /*
 175  * Switches to specified vcpu, until a matching vcpu_put()
 176  */
 177 static void vcpu_load(struct kvm_vcpu *vcpu)
 178 {
 179         int cpu;
 180
 181         mutex_lock(&vcpu->mutex);
 182         cpu = get_cpu();
 183         preempt_notifier_register(&vcpu->preempt_notifier);
 184         kvm_arch_ops->vcpu_load(vcpu, cpu);
 185         put_cpu();
 186 }
 187
 188 static void vcpu_put(struct kvm_vcpu *vcpu)
 189 {
 190         preempt_disable();
 191         kvm_arch_ops->vcpu_put(vcpu);
 192         preempt_notifier_unregister(&vcpu->preempt_notifier);
 193         preempt_enable();
 194         mutex_unlock(&vcpu->mutex);
 195 }
 196
 197 static void ack_flush(void *_completed)
 198 {
 199         atomic_t *completed = _completed;
 200
 201         atomic_inc(completed);
 202 }
 203
 204 void kvm_flush_remote_tlbs(struct kvm *kvm)
 205 {
 206         int i, cpu, needed;
 207         cpumask_t cpus;
 208         struct kvm_vcpu *vcpu;
 209         atomic_t completed;
 210
 211         atomic_set(&completed, 0);
 212         cpus_clear(cpus);
 213         needed = 0;
 214         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 215                 vcpu = kvm->vcpus[i];
 216                 if (!vcpu)
 217                         continue;
 218                 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
 219                         continue;
 220                 cpu = vcpu->cpu;
 221                 if (cpu != -1 && cpu != raw_smp_processor_id())
 222                         if (!cpu_isset(cpu, cpus)) {
 223                                 cpu_set(cpu, cpus);
 224                                 ++needed;
 225                         }
 226         }
 227
 228         /*
 229          * We really want smp_call_function_mask() here.  But that's not
 230          * available, so ipi all cpus in parallel and wait for them
 231          * to complete.
 232          */
 233         for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
 234                 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
 235         while (atomic_read(&completed) != needed) {
 236                 cpu_relax();
 237                 barrier();
 238         }
 239 }
 240
 241 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 242 {
 243         struct page *page;
 244         int r;
 245
 246         mutex_init(&vcpu->mutex);
 247         vcpu->cpu = -1;
 248         vcpu->mmu.root_hpa = INVALID_PAGE;
 249         vcpu->kvm = kvm;
 250         vcpu->vcpu_id = id;
 251
 252         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 253         if (!page) {
 254                 r = -ENOMEM;
 255                 goto fail;
 256         }
 257         vcpu->run = page_address(page);
 258
 259         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 260         if (!page) {
 261                 r = -ENOMEM;
 262                 goto fail_free_run;
 263         }
 264         vcpu->pio_data = page_address(page);
 265
 266         r = kvm_mmu_create(vcpu);
 267         if (r < 0)
 268                 goto fail_free_pio_data;
 269
 270         return 0;
 271
 272 fail_free_pio_data:
 273         free_page((unsigned long)vcpu->pio_data);
 274 fail_free_run:
 275         free_page((unsigned long)vcpu->run);
 276 fail:
 277         return -ENOMEM;
 278 }
 279 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 280
 281 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 282 {
 283         kvm_mmu_destroy(vcpu);
 284         free_page((unsigned long)vcpu->pio_data);
 285         free_page((unsigned long)vcpu->run);
 286 }
 287 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 288
 289 static struct kvm *kvm_create_vm(void)
 290 {
 291         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 292
 293         if (!kvm)
 294                 return ERR_PTR(-ENOMEM);
 295
 296         kvm_io_bus_init(&kvm->pio_bus);
 297         mutex_init(&kvm->lock);
 298         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 299         kvm_io_bus_init(&kvm->mmio_bus);
 300         spin_lock(&kvm_lock);
 301         list_add(&kvm->vm_list, &vm_list);
 302         spin_unlock(&kvm_lock);
 303         return kvm;
 304 }
 305
 306 /*
 307  * Free any memory in @free but not in @dont.
 308  */
 309 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 310                                   struct kvm_memory_slot *dont)
 311 {
 312         int i;
 313
 314         if (!dont || free->phys_mem != dont->phys_mem)
 315                 if (free->phys_mem) {
 316                         for (i = 0; i < free->npages; ++i)
 317                                 if (free->phys_mem[i])
 318                                         __free_page(free->phys_mem[i]);
 319                         vfree(free->phys_mem);
 320                 }
 321
 322         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 323                 vfree(free->dirty_bitmap);
 324
 325         free->phys_mem = NULL;
 326         free->npages = 0;
 327         free->dirty_bitmap = NULL;
 328 }
 329
 330 static void kvm_free_physmem(struct kvm *kvm)
 331 {
 332         int i;
 333
 334         for (i = 0; i < kvm->nmemslots; ++i)
 335                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 336 }
 337
 338 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 339 {
 340         int i;
 341
 342         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
 343                 if (vcpu->pio.guest_pages[i]) {
 344                         __free_page(vcpu->pio.guest_pages[i]);
 345                         vcpu->pio.guest_pages[i] = NULL;
 346                 }
 347 }
 348
 349 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 350 {
 351         vcpu_load(vcpu);
 352         kvm_mmu_unload(vcpu);
 353         vcpu_put(vcpu);
 354 }
 355
 356 static void kvm_free_vcpus(struct kvm *kvm)
 357 {
 358         unsigned int i;
 359
 360         /*
 361          * Unpin any mmu pages first.
 362          */
 363         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 364                 if (kvm->vcpus[i])
 365                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
 366         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 367                 if (kvm->vcpus[i]) {
 368                         kvm_arch_ops->vcpu_free(kvm->vcpus[i]);
 369                         kvm->vcpus[i] = NULL;
 370                 }
 371         }
 372
 373 }
 374
 375 static void kvm_destroy_vm(struct kvm *kvm)
 376 {
 377         spin_lock(&kvm_lock);
 378         list_del(&kvm->vm_list);
 379         spin_unlock(&kvm_lock);
 380         kvm_io_bus_destroy(&kvm->pio_bus);
 381         kvm_io_bus_destroy(&kvm->mmio_bus);
 382         kfree(kvm->vpic);
 383         kvm_free_vcpus(kvm);
 384         kvm_free_physmem(kvm);
 385         kfree(kvm);
 386 }
 387
 388 static int kvm_vm_release(struct inode *inode, struct file *filp)
 389 {
 390         struct kvm *kvm = filp->private_data;
 391
 392         kvm_destroy_vm(kvm);
 393         return 0;
 394 }
 395
 396 static void inject_gp(struct kvm_vcpu *vcpu)
 397 {
 398         kvm_arch_ops->inject_gp(vcpu, 0);
 399 }
 400
 401 /*
 402  * Load the pae pdptrs.  Return true is they are all valid.
 403  */
 404 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 405 {
 406         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 407         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 408         int i;
 409         u64 *pdpt;
 410         int ret;
 411         struct page *page;
 412         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 413
 414         mutex_lock(&vcpu->kvm->lock);
 415         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
 416         if (!page) {
 417                 ret = 0;
 418                 goto out;
 419         }
 420
 421         pdpt = kmap_atomic(page, KM_USER0);
 422         memcpy(pdpte, pdpt+offset, sizeof(pdpte));
 423         kunmap_atomic(pdpt, KM_USER0);
 424
 425         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 426                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 427                         ret = 0;
 428                         goto out;
 429                 }
 430         }
 431         ret = 1;
 432
 433         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 434 out:
 435         mutex_unlock(&vcpu->kvm->lock);
 436
 437         return ret;
 438 }
 439
 440 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 441 {
 442         if (cr0 & CR0_RESERVED_BITS) {
 443                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 444                        cr0, vcpu->cr0);
 445                 inject_gp(vcpu);
 446                 return;
 447         }
 448
 449         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 450                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 451                 inject_gp(vcpu);
 452                 return;
 453         }
 454
 455         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 456                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 457                        "and a clear PE flag\n");
 458                 inject_gp(vcpu);
 459                 return;
 460         }
 461
 462         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 463 #ifdef CONFIG_X86_64
 464                 if ((vcpu->shadow_efer & EFER_LME)) {
 465                         int cs_db, cs_l;
 466
 467                         if (!is_pae(vcpu)) {
 468                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 469                                        "in long mode while PAE is disabled\n");
 470                                 inject_gp(vcpu);
 471                                 return;
 472                         }
 473                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 474                         if (cs_l) {
 475                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 476                                        "in long mode while CS.L == 1\n");
 477                                 inject_gp(vcpu);
 478                                 return;
 479
 480                         }
 481                 } else
 482 #endif
 483                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 484                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 485                                "reserved bits\n");
 486                         inject_gp(vcpu);
 487                         return;
 488                 }
 489
 490         }
 491
 492         kvm_arch_ops->set_cr0(vcpu, cr0);
 493         vcpu->cr0 = cr0;
 494
 495         mutex_lock(&vcpu->kvm->lock);
 496         kvm_mmu_reset_context(vcpu);
 497         mutex_unlock(&vcpu->kvm->lock);
 498         return;
 499 }
 500 EXPORT_SYMBOL_GPL(set_cr0);
 501
 502 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 503 {
 504         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 505 }
 506 EXPORT_SYMBOL_GPL(lmsw);
 507
 508 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 509 {
 510         if (cr4 & CR4_RESERVED_BITS) {
 511                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 512                 inject_gp(vcpu);
 513                 return;
 514         }
 515
 516         if (is_long_mode(vcpu)) {
 517                 if (!(cr4 & X86_CR4_PAE)) {
 518                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 519                                "in long mode\n");
 520                         inject_gp(vcpu);
 521                         return;
 522                 }
 523         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 524                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 525                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 526                 inject_gp(vcpu);
 527                 return;
 528         }
 529
 530         if (cr4 & X86_CR4_VMXE) {
 531                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 532                 inject_gp(vcpu);
 533                 return;
 534         }
 535         kvm_arch_ops->set_cr4(vcpu, cr4);
 536         mutex_lock(&vcpu->kvm->lock);
 537         kvm_mmu_reset_context(vcpu);
 538         mutex_unlock(&vcpu->kvm->lock);
 539 }
 540 EXPORT_SYMBOL_GPL(set_cr4);
 541
 542 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 543 {
 544         if (is_long_mode(vcpu)) {
 545                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 546                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 547                         inject_gp(vcpu);
 548                         return;
 549                 }
 550         } else {
 551                 if (is_pae(vcpu)) {
 552                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 553                                 printk(KERN_DEBUG
 554                                        "set_cr3: #GP, reserved bits\n");
 555                                 inject_gp(vcpu);
 556                                 return;
 557                         }
 558                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 559                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 560                                        "reserved bits\n");
 561                                 inject_gp(vcpu);
 562                                 return;
 563                         }
 564                 } else {
 565                         if (cr3 & CR3_NONPAE_RESERVED_BITS) {
 566                                 printk(KERN_DEBUG
 567                                        "set_cr3: #GP, reserved bits\n");
 568                                 inject_gp(vcpu);
 569                                 return;
 570                         }
 571                 }
 572         }
 573
 574         mutex_lock(&vcpu->kvm->lock);
 575         /*
 576          * Does the new cr3 value map to physical memory? (Note, we
 577          * catch an invalid cr3 even in real-mode, because it would
 578          * cause trouble later on when we turn on paging anyway.)
 579          *
 580          * A real CPU would silently accept an invalid cr3 and would
 581          * attempt to use it - with largely undefined (and often hard
 582          * to debug) behavior on the guest side.
 583          */
 584         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 585                 inject_gp(vcpu);
 586         else {
 587                 vcpu->cr3 = cr3;
 588                 vcpu->mmu.new_cr3(vcpu);
 589         }
 590         mutex_unlock(&vcpu->kvm->lock);
 591 }
 592 EXPORT_SYMBOL_GPL(set_cr3);
 593
 594 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 595 {
 596         if (cr8 & CR8_RESERVED_BITS) {
 597                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 598                 inject_gp(vcpu);
 599                 return;
 600         }
 601         vcpu->cr8 = cr8;
 602 }
 603 EXPORT_SYMBOL_GPL(set_cr8);
 604
 605 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 606 {
 607         return vcpu->cr8;
 608 }
 609 EXPORT_SYMBOL_GPL(get_cr8);
 610
 611 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 612 {
 613         return vcpu->apic_base;
 614 }
 615 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 616
 617 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 618 {
 619         vcpu->apic_base = data;
 620 }
 621 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 622
 623 void fx_init(struct kvm_vcpu *vcpu)
 624 {
 625         unsigned after_mxcsr_mask;
 626
 627         /* Initialize guest FPU by resetting ours and saving into guest's */
 628         preempt_disable();
 629         fx_save(&vcpu->host_fx_image);
 630         fpu_init();
 631         fx_save(&vcpu->guest_fx_image);
 632         fx_restore(&vcpu->host_fx_image);
 633         preempt_enable();
 634
 635         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
 636         vcpu->guest_fx_image.mxcsr = 0x1f80;
 637         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
 638                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
 639 }
 640 EXPORT_SYMBOL_GPL(fx_init);
 641
 642 /*
 643  * Allocate some memory and give it an address in the guest physical address
 644  * space.
 645  *
 646  * Discontiguous memory is allowed, mostly for framebuffers.
 647  */
 648 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 649                                           struct kvm_memory_region *mem)
 650 {
 651         int r;
 652         gfn_t base_gfn;
 653         unsigned long npages;
 654         unsigned long i;
 655         struct kvm_memory_slot *memslot;
 656         struct kvm_memory_slot old, new;
 657         int memory_config_version;
 658
 659         r = -EINVAL;
 660         /* General sanity checks */
 661         if (mem->memory_size & (PAGE_SIZE - 1))
 662                 goto out;
 663         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 664                 goto out;
 665         if (mem->slot >= KVM_MEMORY_SLOTS)
 666                 goto out;
 667         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 668                 goto out;
 669
 670         memslot = &kvm->memslots[mem->slot];
 671         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 672         npages = mem->memory_size >> PAGE_SHIFT;
 673
 674         if (!npages)
 675                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 676
 677 raced:
 678         mutex_lock(&kvm->lock);
 679
 680         memory_config_version = kvm->memory_config_version;
 681         new = old = *memslot;
 682
 683         new.base_gfn = base_gfn;
 684         new.npages = npages;
 685         new.flags = mem->flags;
 686
 687         /* Disallow changing a memory slot's size. */
 688         r = -EINVAL;
 689         if (npages && old.npages && npages != old.npages)
 690                 goto out_unlock;
 691
 692         /* Check for overlaps */
 693         r = -EEXIST;
 694         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 695                 struct kvm_memory_slot *s = &kvm->memslots[i];
 696
 697                 if (s == memslot)
 698                         continue;
 699                 if (!((base_gfn + npages <= s->base_gfn) ||
 700                       (base_gfn >= s->base_gfn + s->npages)))
 701                         goto out_unlock;
 702         }
 703         /*
 704          * Do memory allocations outside lock.  memory_config_version will
 705          * detect any races.
 706          */
 707         mutex_unlock(&kvm->lock);
 708
 709         /* Deallocate if slot is being removed */
 710         if (!npages)
 711                 new.phys_mem = NULL;
 712
 713         /* Free page dirty bitmap if unneeded */
 714         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 715                 new.dirty_bitmap = NULL;
 716
 717         r = -ENOMEM;
 718
 719         /* Allocate if a slot is being created */
 720         if (npages && !new.phys_mem) {
 721                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 722
 723                 if (!new.phys_mem)
 724                         goto out_free;
 725
 726                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 727                 for (i = 0; i < npages; ++i) {
 728                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 729                                                      | __GFP_ZERO);
 730                         if (!new.phys_mem[i])
 731                                 goto out_free;
 732                         set_page_private(new.phys_mem[i],0);
 733                 }
 734         }
 735
 736         /* Allocate page dirty bitmap if needed */
 737         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 738                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 739
 740                 new.dirty_bitmap = vmalloc(dirty_bytes);
 741                 if (!new.dirty_bitmap)
 742                         goto out_free;
 743                 memset(new.dirty_bitmap, 0, dirty_bytes);
 744         }
 745
 746         mutex_lock(&kvm->lock);
 747
 748         if (memory_config_version != kvm->memory_config_version) {
 749                 mutex_unlock(&kvm->lock);
 750                 kvm_free_physmem_slot(&new, &old);
 751                 goto raced;
 752         }
 753
 754         r = -EAGAIN;
 755         if (kvm->busy)
 756                 goto out_unlock;
 757
 758         if (mem->slot >= kvm->nmemslots)
 759                 kvm->nmemslots = mem->slot + 1;
 760
 761         *memslot = new;
 762         ++kvm->memory_config_version;
 763
 764         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 765         kvm_flush_remote_tlbs(kvm);
 766
 767         mutex_unlock(&kvm->lock);
 768
 769         kvm_free_physmem_slot(&old, &new);
 770         return 0;
 771
 772 out_unlock:
 773         mutex_unlock(&kvm->lock);
 774 out_free:
 775         kvm_free_physmem_slot(&new, &old);
 776 out:
 777         return r;
 778 }
 779
 780 /*
 781  * Get (and clear) the dirty memory log for a memory slot.
 782  */
 783 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 784                                       struct kvm_dirty_log *log)
 785 {
 786         struct kvm_memory_slot *memslot;
 787         int r, i;
 788         int n;
 789         unsigned long any = 0;
 790
 791         mutex_lock(&kvm->lock);
 792
 793         /*
 794          * Prevent changes to guest memory configuration even while the lock
 795          * is not taken.
 796          */
 797         ++kvm->busy;
 798         mutex_unlock(&kvm->lock);
 799         r = -EINVAL;
 800         if (log->slot >= KVM_MEMORY_SLOTS)
 801                 goto out;
 802
 803         memslot = &kvm->memslots[log->slot];
 804         r = -ENOENT;
 805         if (!memslot->dirty_bitmap)
 806                 goto out;
 807
 808         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 809
 810         for (i = 0; !any && i < n/sizeof(long); ++i)
 811                 any = memslot->dirty_bitmap[i];
 812
 813         r = -EFAULT;
 814         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 815                 goto out;
 816
 817         /* If nothing is dirty, don't bother messing with page tables. */
 818         if (any) {
 819                 mutex_lock(&kvm->lock);
 820                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 821                 kvm_flush_remote_tlbs(kvm);
 822                 memset(memslot->dirty_bitmap, 0, n);
 823                 mutex_unlock(&kvm->lock);
 824         }
 825
 826         r = 0;
 827
 828 out:
 829         mutex_lock(&kvm->lock);
 830         --kvm->busy;
 831         mutex_unlock(&kvm->lock);
 832         return r;
 833 }
 834
 835 /*
 836  * Set a new alias region.  Aliases map a portion of physical memory into
 837  * another portion.  This is useful for memory windows, for example the PC
 838  * VGA region.
 839  */
 840 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 841                                          struct kvm_memory_alias *alias)
 842 {
 843         int r, n;
 844         struct kvm_mem_alias *p;
 845
 846         r = -EINVAL;
 847         /* General sanity checks */
 848         if (alias->memory_size & (PAGE_SIZE - 1))
 849                 goto out;
 850         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 851                 goto out;
 852         if (alias->slot >= KVM_ALIAS_SLOTS)
 853                 goto out;
 854         if (alias->guest_phys_addr + alias->memory_size
 855             < alias->guest_phys_addr)
 856                 goto out;
 857         if (alias->target_phys_addr + alias->memory_size
 858             < alias->target_phys_addr)
 859                 goto out;
 860
 861         mutex_lock(&kvm->lock);
 862
 863         p = &kvm->aliases[alias->slot];
 864         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 865         p->npages = alias->memory_size >> PAGE_SHIFT;
 866         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 867
 868         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 869                 if (kvm->aliases[n - 1].npages)
 870                         break;
 871         kvm->naliases = n;
 872
 873         kvm_mmu_zap_all(kvm);
 874
 875         mutex_unlock(&kvm->lock);
 876
 877         return 0;
 878
 879 out:
 880         return r;
 881 }
 882
 883 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 884 {
 885         int i;
 886         struct kvm_mem_alias *alias;
 887
 888         for (i = 0; i < kvm->naliases; ++i) {
 889                 alias = &kvm->aliases[i];
 890                 if (gfn >= alias->base_gfn
 891                     && gfn < alias->base_gfn + alias->npages)
 892                         return alias->target_gfn + gfn - alias->base_gfn;
 893         }
 894         return gfn;
 895 }
 896
 897 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 898 {
 899         int i;
 900
 901         for (i = 0; i < kvm->nmemslots; ++i) {
 902                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 903
 904                 if (gfn >= memslot->base_gfn
 905                     && gfn < memslot->base_gfn + memslot->npages)
 906                         return memslot;
 907         }
 908         return NULL;
 909 }
 910
 911 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 912 {
 913         gfn = unalias_gfn(kvm, gfn);
 914         return __gfn_to_memslot(kvm, gfn);
 915 }
 916
 917 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 918 {
 919         struct kvm_memory_slot *slot;
 920
 921         gfn = unalias_gfn(kvm, gfn);
 922         slot = __gfn_to_memslot(kvm, gfn);
 923         if (!slot)
 924                 return NULL;
 925         return slot->phys_mem[gfn - slot->base_gfn];
 926 }
 927 EXPORT_SYMBOL_GPL(gfn_to_page);
 928
 929 /* WARNING: Does not work on aliased pages. */
 930 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 931 {
 932         struct kvm_memory_slot *memslot;
 933
 934         memslot = __gfn_to_memslot(kvm, gfn);
 935         if (memslot && memslot->dirty_bitmap) {
 936                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 937
 938                 /* avoid RMW */
 939                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 940                         set_bit(rel_gfn, memslot->dirty_bitmap);
 941         }
 942 }
 943
 944 int emulator_read_std(unsigned long addr,
 945                              void *val,
 946                              unsigned int bytes,
 947                              struct kvm_vcpu *vcpu)
 948 {
 949         void *data = val;
 950
 951         while (bytes) {
 952                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 953                 unsigned offset = addr & (PAGE_SIZE-1);
 954                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 955                 unsigned long pfn;
 956                 struct page *page;
 957                 void *page_virt;
 958
 959                 if (gpa == UNMAPPED_GVA)
 960                         return X86EMUL_PROPAGATE_FAULT;
 961                 pfn = gpa >> PAGE_SHIFT;
 962                 page = gfn_to_page(vcpu->kvm, pfn);
 963                 if (!page)
 964                         return X86EMUL_UNHANDLEABLE;
 965                 page_virt = kmap_atomic(page, KM_USER0);
 966
 967                 memcpy(data, page_virt + offset, tocopy);
 968
 969                 kunmap_atomic(page_virt, KM_USER0);
 970
 971                 bytes -= tocopy;
 972                 data += tocopy;
 973                 addr += tocopy;
 974         }
 975
 976         return X86EMUL_CONTINUE;
 977 }
 978 EXPORT_SYMBOL_GPL(emulator_read_std);
 979
 980 static int emulator_write_std(unsigned long addr,
 981                               const void *val,
 982                               unsigned int bytes,
 983                               struct kvm_vcpu *vcpu)
 984 {
 985         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
 986         return X86EMUL_UNHANDLEABLE;
 987 }
 988
 989 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
 990                                                 gpa_t addr)
 991 {
 992         /*
 993          * Note that its important to have this wrapper function because
 994          * in the very near future we will be checking for MMIOs against
 995          * the LAPIC as well as the general MMIO bus
 996          */
 997         return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
 998 }
 999
1000 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1001                                                gpa_t addr)
1002 {
1003         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1004 }
1005
1006 static int emulator_read_emulated(unsigned long addr,
1007                                   void *val,
1008                                   unsigned int bytes,
1009                                   struct kvm_vcpu *vcpu)
1010 {
1011         struct kvm_io_device *mmio_dev;
1012         gpa_t                 gpa;
1013
1014         if (vcpu->mmio_read_completed) {
1015                 memcpy(val, vcpu->mmio_data, bytes);
1016                 vcpu->mmio_read_completed = 0;
1017                 return X86EMUL_CONTINUE;
1018         } else if (emulator_read_std(addr, val, bytes, vcpu)
1019                    == X86EMUL_CONTINUE)
1020                 return X86EMUL_CONTINUE;
1021
1022         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1023         if (gpa == UNMAPPED_GVA)
1024                 return X86EMUL_PROPAGATE_FAULT;
1025
1026         /*
1027          * Is this MMIO handled locally?
1028          */
1029         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1030         if (mmio_dev) {
1031                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1032                 return X86EMUL_CONTINUE;
1033         }
1034
1035         vcpu->mmio_needed = 1;
1036         vcpu->mmio_phys_addr = gpa;
1037         vcpu->mmio_size = bytes;
1038         vcpu->mmio_is_write = 0;
1039
1040         return X86EMUL_UNHANDLEABLE;
1041 }
1042
1043 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1044                                const void *val, int bytes)
1045 {
1046         struct page *page;
1047         void *virt;
1048
1049         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1050                 return 0;
1051         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1052         if (!page)
1053                 return 0;
1054         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1055         virt = kmap_atomic(page, KM_USER0);
1056         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1057         memcpy(virt + offset_in_page(gpa), val, bytes);
1058         kunmap_atomic(virt, KM_USER0);
1059         return 1;
1060 }
1061
1062 static int emulator_write_emulated_onepage(unsigned long addr,
1063                                            const void *val,
1064                                            unsigned int bytes,
1065                                            struct kvm_vcpu *vcpu)
1066 {
1067         struct kvm_io_device *mmio_dev;
1068         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1069
1070         if (gpa == UNMAPPED_GVA) {
1071                 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1072                 return X86EMUL_PROPAGATE_FAULT;
1073         }
1074
1075         if (emulator_write_phys(vcpu, gpa, val, bytes))
1076                 return X86EMUL_CONTINUE;
1077
1078         /*
1079          * Is this MMIO handled locally?
1080          */
1081         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1082         if (mmio_dev) {
1083                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1084                 return X86EMUL_CONTINUE;
1085         }
1086
1087         vcpu->mmio_needed = 1;
1088         vcpu->mmio_phys_addr = gpa;
1089         vcpu->mmio_size = bytes;
1090         vcpu->mmio_is_write = 1;
1091         memcpy(vcpu->mmio_data, val, bytes);
1092
1093         return X86EMUL_CONTINUE;
1094 }
1095
1096 int emulator_write_emulated(unsigned long addr,
1097                                    const void *val,
1098                                    unsigned int bytes,
1099                                    struct kvm_vcpu *vcpu)
1100 {
1101         /* Crossing a page boundary? */
1102         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1103                 int rc, now;
1104
1105                 now = -addr & ~PAGE_MASK;
1106                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1107                 if (rc != X86EMUL_CONTINUE)
1108                         return rc;
1109                 addr += now;
1110                 val += now;
1111                 bytes -= now;
1112         }
1113         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1114 }
1115 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1116
1117 static int emulator_cmpxchg_emulated(unsigned long addr,
1118                                      const void *old,
1119                                      const void *new,
1120                                      unsigned int bytes,
1121                                      struct kvm_vcpu *vcpu)
1122 {
1123         static int reported;
1124
1125         if (!reported) {
1126                 reported = 1;
1127                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1128         }
1129         return emulator_write_emulated(addr, new, bytes, vcpu);
1130 }
1131
1132 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1133 {
1134         return kvm_arch_ops->get_segment_base(vcpu, seg);
1135 }
1136
1137 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1138 {
1139         return X86EMUL_CONTINUE;
1140 }
1141
1142 int emulate_clts(struct kvm_vcpu *vcpu)
1143 {
1144         unsigned long cr0;
1145
1146         cr0 = vcpu->cr0 & ~X86_CR0_TS;
1147         kvm_arch_ops->set_cr0(vcpu, cr0);
1148         return X86EMUL_CONTINUE;
1149 }
1150
1151 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1152 {
1153         struct kvm_vcpu *vcpu = ctxt->vcpu;
1154
1155         switch (dr) {
1156         case 0 ... 3:
1157                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1158                 return X86EMUL_CONTINUE;
1159         default:
1160                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1161                 return X86EMUL_UNHANDLEABLE;
1162         }
1163 }
1164
1165 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1166 {
1167         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1168         int exception;
1169
1170         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1171         if (exception) {
1172                 /* FIXME: better handling */
1173                 return X86EMUL_UNHANDLEABLE;
1174         }
1175         return X86EMUL_CONTINUE;
1176 }
1177
1178 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1179 {
1180         static int reported;
1181         u8 opcodes[4];
1182         unsigned long rip = ctxt->vcpu->rip;
1183         unsigned long rip_linear;
1184
1185         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1186
1187         if (reported)
1188                 return;
1189
1190         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt->vcpu);
1191
1192         printk(KERN_ERR "emulation failed but !mmio_needed?"
1193                " rip %lx %02x %02x %02x %02x\n",
1194                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1195         reported = 1;
1196 }
1197
1198 struct x86_emulate_ops emulate_ops = {
1199         .read_std            = emulator_read_std,
1200         .write_std           = emulator_write_std,
1201         .read_emulated       = emulator_read_emulated,
1202         .write_emulated      = emulator_write_emulated,
1203         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1204 };
1205
1206 int emulate_instruction(struct kvm_vcpu *vcpu,
1207                         struct kvm_run *run,
1208                         unsigned long cr2,
1209                         u16 error_code)
1210 {
1211         struct x86_emulate_ctxt emulate_ctxt;
1212         int r;
1213         int cs_db, cs_l;
1214
1215         vcpu->mmio_fault_cr2 = cr2;
1216         kvm_arch_ops->cache_regs(vcpu);
1217
1218         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1219
1220         emulate_ctxt.vcpu = vcpu;
1221         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1222         emulate_ctxt.cr2 = cr2;
1223         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1224                 ? X86EMUL_MODE_REAL : cs_l
1225                 ? X86EMUL_MODE_PROT64 : cs_db
1226                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1227
1228         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1229                 emulate_ctxt.cs_base = 0;
1230                 emulate_ctxt.ds_base = 0;
1231                 emulate_ctxt.es_base = 0;
1232                 emulate_ctxt.ss_base = 0;
1233         } else {
1234                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1235                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1236                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1237                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1238         }
1239
1240         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1241         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1242
1243         vcpu->mmio_is_write = 0;
1244         vcpu->pio.string = 0;
1245         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1246         if (vcpu->pio.string)
1247                 return EMULATE_DO_MMIO;
1248
1249         if ((r || vcpu->mmio_is_write) && run) {
1250                 run->exit_reason = KVM_EXIT_MMIO;
1251                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1252                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1253                 run->mmio.len = vcpu->mmio_size;
1254                 run->mmio.is_write = vcpu->mmio_is_write;
1255         }
1256
1257         if (r) {
1258                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1259                         return EMULATE_DONE;
1260                 if (!vcpu->mmio_needed) {
1261                         report_emulation_failure(&emulate_ctxt);
1262                         return EMULATE_FAIL;
1263                 }
1264                 return EMULATE_DO_MMIO;
1265         }
1266
1267         kvm_arch_ops->decache_regs(vcpu);
1268         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1269
1270         if (vcpu->mmio_is_write) {
1271                 vcpu->mmio_needed = 0;
1272                 return EMULATE_DO_MMIO;
1273         }
1274
1275         return EMULATE_DONE;
1276 }
1277 EXPORT_SYMBOL_GPL(emulate_instruction);
1278
1279 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1280 {
1281         if (vcpu->irq_summary ||
1282                 (irqchip_in_kernel(vcpu->kvm) && kvm_cpu_has_interrupt(vcpu)))
1283                 return 1;
1284
1285         vcpu->run->exit_reason = KVM_EXIT_HLT;
1286         ++vcpu->stat.halt_exits;
1287         return 0;
1288 }
1289 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1290
1291 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1292 {
1293         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1294
1295         kvm_arch_ops->cache_regs(vcpu);
1296         ret = -KVM_EINVAL;
1297 #ifdef CONFIG_X86_64
1298         if (is_long_mode(vcpu)) {
1299                 nr = vcpu->regs[VCPU_REGS_RAX];
1300                 a0 = vcpu->regs[VCPU_REGS_RDI];
1301                 a1 = vcpu->regs[VCPU_REGS_RSI];
1302                 a2 = vcpu->regs[VCPU_REGS_RDX];
1303                 a3 = vcpu->regs[VCPU_REGS_RCX];
1304                 a4 = vcpu->regs[VCPU_REGS_R8];
1305                 a5 = vcpu->regs[VCPU_REGS_R9];
1306         } else
1307 #endif
1308         {
1309                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1310                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1311                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1312                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1313                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1314                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1315                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1316         }
1317         switch (nr) {
1318         default:
1319                 run->hypercall.nr = nr;
1320                 run->hypercall.args[0] = a0;
1321                 run->hypercall.args[1] = a1;
1322                 run->hypercall.args[2] = a2;
1323                 run->hypercall.args[3] = a3;
1324                 run->hypercall.args[4] = a4;
1325                 run->hypercall.args[5] = a5;
1326                 run->hypercall.ret = ret;
1327                 run->hypercall.longmode = is_long_mode(vcpu);
1328                 kvm_arch_ops->decache_regs(vcpu);
1329                 return 0;
1330         }
1331         vcpu->regs[VCPU_REGS_RAX] = ret;
1332         kvm_arch_ops->decache_regs(vcpu);
1333         return 1;
1334 }
1335 EXPORT_SYMBOL_GPL(kvm_hypercall);
1336
1337 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1338 {
1339         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1340 }
1341
1342 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1343 {
1344         struct descriptor_table dt = { limit, base };
1345
1346         kvm_arch_ops->set_gdt(vcpu, &dt);
1347 }
1348
1349 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1350 {
1351         struct descriptor_table dt = { limit, base };
1352
1353         kvm_arch_ops->set_idt(vcpu, &dt);
1354 }
1355
1356 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1357                    unsigned long *rflags)
1358 {
1359         lmsw(vcpu, msw);
1360         *rflags = kvm_arch_ops->get_rflags(vcpu);
1361 }
1362
1363 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1364 {
1365         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1366         switch (cr) {
1367         case 0:
1368                 return vcpu->cr0;
1369         case 2:
1370                 return vcpu->cr2;
1371         case 3:
1372                 return vcpu->cr3;
1373         case 4:
1374                 return vcpu->cr4;
1375         default:
1376                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1377                 return 0;
1378         }
1379 }
1380
1381 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1382                      unsigned long *rflags)
1383 {
1384         switch (cr) {
1385         case 0:
1386                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1387                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1388                 break;
1389         case 2:
1390                 vcpu->cr2 = val;
1391                 break;
1392         case 3:
1393                 set_cr3(vcpu, val);
1394                 break;
1395         case 4:
1396                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1397                 break;
1398         default:
1399                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1400         }
1401 }
1402
1403 /*
1404  * Register the para guest with the host:
1405  */
1406 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1407 {
1408         struct kvm_vcpu_para_state *para_state;
1409         hpa_t para_state_hpa, hypercall_hpa;
1410         struct page *para_state_page;
1411         unsigned char *hypercall;
1412         gpa_t hypercall_gpa;
1413
1414         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1415         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1416
1417         /*
1418          * Needs to be page aligned:
1419          */
1420         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1421                 goto err_gp;
1422
1423         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1424         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1425         if (is_error_hpa(para_state_hpa))
1426                 goto err_gp;
1427
1428         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1429         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1430         para_state = kmap(para_state_page);
1431
1432         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1433         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1434
1435         para_state->host_version = KVM_PARA_API_VERSION;
1436         /*
1437          * We cannot support guests that try to register themselves
1438          * with a newer API version than the host supports:
1439          */
1440         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1441                 para_state->ret = -KVM_EINVAL;
1442                 goto err_kunmap_skip;
1443         }
1444
1445         hypercall_gpa = para_state->hypercall_gpa;
1446         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1447         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1448         if (is_error_hpa(hypercall_hpa)) {
1449                 para_state->ret = -KVM_EINVAL;
1450                 goto err_kunmap_skip;
1451         }
1452
1453         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1454         vcpu->para_state_page = para_state_page;
1455         vcpu->para_state_gpa = para_state_gpa;
1456         vcpu->hypercall_gpa = hypercall_gpa;
1457
1458         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1459         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1460                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1461         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1462         kunmap_atomic(hypercall, KM_USER1);
1463
1464         para_state->ret = 0;
1465 err_kunmap_skip:
1466         kunmap(para_state_page);
1467         return 0;
1468 err_gp:
1469         return 1;
1470 }
1471
1472 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1473 {
1474         u64 data;
1475
1476         switch (msr) {
1477         case 0xc0010010: /* SYSCFG */
1478         case 0xc0010015: /* HWCR */
1479         case MSR_IA32_PLATFORM_ID:
1480         case MSR_IA32_P5_MC_ADDR:
1481         case MSR_IA32_P5_MC_TYPE:
1482         case MSR_IA32_MC0_CTL:
1483         case MSR_IA32_MCG_STATUS:
1484         case MSR_IA32_MCG_CAP:
1485         case MSR_IA32_MC0_MISC:
1486         case MSR_IA32_MC0_MISC+4:
1487         case MSR_IA32_MC0_MISC+8:
1488         case MSR_IA32_MC0_MISC+12:
1489         case MSR_IA32_MC0_MISC+16:
1490         case MSR_IA32_UCODE_REV:
1491         case MSR_IA32_PERF_STATUS:
1492         case MSR_IA32_EBL_CR_POWERON:
1493                 /* MTRR registers */
1494         case 0xfe:
1495         case 0x200 ... 0x2ff:
1496                 data = 0;
1497                 break;
1498         case 0xcd: /* fsb frequency */
1499                 data = 3;
1500                 break;
1501         case MSR_IA32_APICBASE:
1502                 data = kvm_get_apic_base(vcpu);
1503                 break;
1504         case MSR_IA32_MISC_ENABLE:
1505                 data = vcpu->ia32_misc_enable_msr;
1506                 break;
1507 #ifdef CONFIG_X86_64
1508         case MSR_EFER:
1509                 data = vcpu->shadow_efer;
1510                 break;
1511 #endif
1512         default:
1513                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1514                 return 1;
1515         }
1516         *pdata = data;
1517         return 0;
1518 }
1519 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1520
1521 /*
1522  * Reads an msr value (of 'msr_index') into 'pdata'.
1523  * Returns 0 on success, non-0 otherwise.
1524  * Assumes vcpu_load() was already called.
1525  */
1526 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1527 {
1528         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1529 }
1530
1531 #ifdef CONFIG_X86_64
1532
1533 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1534 {
1535         if (efer & EFER_RESERVED_BITS) {
1536                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1537                        efer);
1538                 inject_gp(vcpu);
1539                 return;
1540         }
1541
1542         if (is_paging(vcpu)
1543             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1544                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1545                 inject_gp(vcpu);
1546                 return;
1547         }
1548
1549         kvm_arch_ops->set_efer(vcpu, efer);
1550
1551         efer &= ~EFER_LMA;
1552         efer |= vcpu->shadow_efer & EFER_LMA;
1553
1554         vcpu->shadow_efer = efer;
1555 }
1556
1557 #endif
1558
1559 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1560 {
1561         switch (msr) {
1562 #ifdef CONFIG_X86_64
1563         case MSR_EFER:
1564                 set_efer(vcpu, data);
1565                 break;
1566 #endif
1567         case MSR_IA32_MC0_STATUS:
1568                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1569                        __FUNCTION__, data);
1570                 break;
1571         case MSR_IA32_MCG_STATUS:
1572                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1573                         __FUNCTION__, data);
1574                 break;
1575         case MSR_IA32_UCODE_REV:
1576         case MSR_IA32_UCODE_WRITE:
1577         case 0x200 ... 0x2ff: /* MTRRs */
1578                 break;
1579         case MSR_IA32_APICBASE:
1580                 kvm_set_apic_base(vcpu, data);
1581                 break;
1582         case MSR_IA32_MISC_ENABLE:
1583                 vcpu->ia32_misc_enable_msr = data;
1584                 break;
1585         /*
1586          * This is the 'probe whether the host is KVM' logic:
1587          */
1588         case MSR_KVM_API_MAGIC:
1589                 return vcpu_register_para(vcpu, data);
1590
1591         default:
1592                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1593                 return 1;
1594         }
1595         return 0;
1596 }
1597 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1598
1599 /*
1600  * Writes msr value into into the appropriate "register".
1601  * Returns 0 on success, non-0 otherwise.
1602  * Assumes vcpu_load() was already called.
1603  */
1604 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1605 {
1606         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1607 }
1608
1609 void kvm_resched(struct kvm_vcpu *vcpu)
1610 {
1611         if (!need_resched())
1612                 return;
1613         cond_resched();
1614 }
1615 EXPORT_SYMBOL_GPL(kvm_resched);
1616
1617 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1618 {
1619         int i;
1620         u32 function;
1621         struct kvm_cpuid_entry *e, *best;
1622
1623         kvm_arch_ops->cache_regs(vcpu);
1624         function = vcpu->regs[VCPU_REGS_RAX];
1625         vcpu->regs[VCPU_REGS_RAX] = 0;
1626         vcpu->regs[VCPU_REGS_RBX] = 0;
1627         vcpu->regs[VCPU_REGS_RCX] = 0;
1628         vcpu->regs[VCPU_REGS_RDX] = 0;
1629         best = NULL;
1630         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1631                 e = &vcpu->cpuid_entries[i];
1632                 if (e->function == function) {
1633                         best = e;
1634                         break;
1635                 }
1636                 /*
1637                  * Both basic or both extended?
1638                  */
1639                 if (((e->function ^ function) & 0x80000000) == 0)
1640                         if (!best || e->function > best->function)
1641                                 best = e;
1642         }
1643         if (best) {
1644                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1645                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1646                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1647                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1648         }
1649         kvm_arch_ops->decache_regs(vcpu);
1650         kvm_arch_ops->skip_emulated_instruction(vcpu);
1651 }
1652 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1653
1654 static int pio_copy_data(struct kvm_vcpu *vcpu)
1655 {
1656         void *p = vcpu->pio_data;
1657         void *q;
1658         unsigned bytes;
1659         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1660
1661         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1662                  PAGE_KERNEL);
1663         if (!q) {
1664                 free_pio_guest_pages(vcpu);
1665                 return -ENOMEM;
1666         }
1667         q += vcpu->pio.guest_page_offset;
1668         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1669         if (vcpu->pio.in)
1670                 memcpy(q, p, bytes);
1671         else
1672                 memcpy(p, q, bytes);
1673         q -= vcpu->pio.guest_page_offset;
1674         vunmap(q);
1675         free_pio_guest_pages(vcpu);
1676         return 0;
1677 }
1678
1679 static int complete_pio(struct kvm_vcpu *vcpu)
1680 {
1681         struct kvm_pio_request *io = &vcpu->pio;
1682         long delta;
1683         int r;
1684
1685         kvm_arch_ops->cache_regs(vcpu);
1686
1687         if (!io->string) {
1688                 if (io->in)
1689                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1690                                io->size);
1691         } else {
1692                 if (io->in) {
1693                         r = pio_copy_data(vcpu);
1694                         if (r) {
1695                                 kvm_arch_ops->cache_regs(vcpu);
1696                                 return r;
1697                         }
1698                 }
1699
1700                 delta = 1;
1701                 if (io->rep) {
1702                         delta *= io->cur_count;
1703                         /*
1704                          * The size of the register should really depend on
1705                          * current address size.
1706                          */
1707                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1708                 }
1709                 if (io->down)
1710                         delta = -delta;
1711                 delta *= io->size;
1712                 if (io->in)
1713                         vcpu->regs[VCPU_REGS_RDI] += delta;
1714                 else
1715                         vcpu->regs[VCPU_REGS_RSI] += delta;
1716         }
1717
1718         kvm_arch_ops->decache_regs(vcpu);
1719
1720         io->count -= io->cur_count;
1721         io->cur_count = 0;
1722
1723         if (!io->count)
1724                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1725         return 0;
1726 }
1727
1728 static void kernel_pio(struct kvm_io_device *pio_dev,
1729                        struct kvm_vcpu *vcpu,
1730                        void *pd)
1731 {
1732         /* TODO: String I/O for in kernel device */
1733
1734         if (vcpu->pio.in)
1735                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1736                                   vcpu->pio.size,
1737                                   pd);
1738         else
1739                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1740                                    vcpu->pio.size,
1741                                    pd);
1742 }
1743
1744 static void pio_string_write(struct kvm_io_device *pio_dev,
1745                              struct kvm_vcpu *vcpu)
1746 {
1747         struct kvm_pio_request *io = &vcpu->pio;
1748         void *pd = vcpu->pio_data;
1749         int i;
1750
1751         for (i = 0; i < io->cur_count; i++) {
1752                 kvm_iodevice_write(pio_dev, io->port,
1753                                    io->size,
1754                                    pd);
1755                 pd += io->size;
1756         }
1757 }
1758
1759 int kvm_emulate_pio (struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1760                   int size, unsigned port)
1761 {
1762         struct kvm_io_device *pio_dev;
1763
1764         vcpu->run->exit_reason = KVM_EXIT_IO;
1765         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1766         vcpu->run->io.size = vcpu->pio.size = size;
1767         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1768         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1769         vcpu->run->io.port = vcpu->pio.port = port;
1770         vcpu->pio.in = in;
1771         vcpu->pio.string = 0;
1772         vcpu->pio.down = 0;
1773         vcpu->pio.guest_page_offset = 0;
1774         vcpu->pio.rep = 0;
1775
1776         kvm_arch_ops->cache_regs(vcpu);
1777         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1778         kvm_arch_ops->decache_regs(vcpu);
1779
1780         pio_dev = vcpu_find_pio_dev(vcpu, port);
1781         if (pio_dev) {
1782                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1783                 complete_pio(vcpu);
1784                 return 1;
1785         }
1786         return 0;
1787 }
1788 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1789
1790 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1791                   int size, unsigned long count, int down,
1792                   gva_t address, int rep, unsigned port)
1793 {
1794         unsigned now, in_page;
1795         int i, ret = 0;
1796         int nr_pages = 1;
1797         struct page *page;
1798         struct kvm_io_device *pio_dev;
1799
1800         vcpu->run->exit_reason = KVM_EXIT_IO;
1801         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1802         vcpu->run->io.size = vcpu->pio.size = size;
1803         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1804         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1805         vcpu->run->io.port = vcpu->pio.port = port;
1806         vcpu->pio.in = in;
1807         vcpu->pio.string = 1;
1808         vcpu->pio.down = down;
1809         vcpu->pio.guest_page_offset = offset_in_page(address);
1810         vcpu->pio.rep = rep;
1811
1812         if (!count) {
1813                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1814                 return 1;
1815         }
1816
1817         if (!down)
1818                 in_page = PAGE_SIZE - offset_in_page(address);
1819         else
1820                 in_page = offset_in_page(address) + size;
1821         now = min(count, (unsigned long)in_page / size);
1822         if (!now) {
1823                 /*
1824                  * String I/O straddles page boundary.  Pin two guest pages
1825                  * so that we satisfy atomicity constraints.  Do just one
1826                  * transaction to avoid complexity.
1827                  */
1828                 nr_pages = 2;
1829                 now = 1;
1830         }
1831         if (down) {
1832                 /*
1833                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1834                  */
1835                 pr_unimpl(vcpu, "guest string pio down\n");
1836                 inject_gp(vcpu);
1837                 return 1;
1838         }
1839         vcpu->run->io.count = now;
1840         vcpu->pio.cur_count = now;
1841
1842         for (i = 0; i < nr_pages; ++i) {
1843                 mutex_lock(&vcpu->kvm->lock);
1844                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1845                 if (page)
1846                         get_page(page);
1847                 vcpu->pio.guest_pages[i] = page;
1848                 mutex_unlock(&vcpu->kvm->lock);
1849                 if (!page) {
1850                         inject_gp(vcpu);
1851                         free_pio_guest_pages(vcpu);
1852                         return 1;
1853                 }
1854         }
1855
1856         pio_dev = vcpu_find_pio_dev(vcpu, port);
1857         if (!vcpu->pio.in) {
1858                 /* string PIO write */
1859                 ret = pio_copy_data(vcpu);
1860                 if (ret >= 0 && pio_dev) {
1861                         pio_string_write(pio_dev, vcpu);
1862                         complete_pio(vcpu);
1863                         if (vcpu->pio.count == 0)
1864                                 ret = 1;
1865                 }
1866         } else if (pio_dev)
1867                 pr_unimpl(vcpu, "no string pio read support yet, "
1868                        "port %x size %d count %ld\n",
1869                         port, size, count);
1870
1871         return ret;
1872 }
1873 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1874
1875 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1876 {
1877         int r;
1878         sigset_t sigsaved;
1879
1880         vcpu_load(vcpu);
1881
1882         if (vcpu->sigset_active)
1883                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1884
1885         /* re-sync apic's tpr */
1886         set_cr8(vcpu, kvm_run->cr8);
1887
1888         if (vcpu->pio.cur_count) {
1889                 r = complete_pio(vcpu);
1890                 if (r)
1891                         goto out;
1892         }
1893
1894         if (vcpu->mmio_needed) {
1895                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1896                 vcpu->mmio_read_completed = 1;
1897                 vcpu->mmio_needed = 0;
1898                 r = emulate_instruction(vcpu, kvm_run,
1899                                         vcpu->mmio_fault_cr2, 0);
1900                 if (r == EMULATE_DO_MMIO) {
1901                         /*
1902                          * Read-modify-write.  Back to userspace.
1903                          */
1904                         r = 0;
1905                         goto out;
1906                 }
1907         }
1908
1909         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1910                 kvm_arch_ops->cache_regs(vcpu);
1911                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1912                 kvm_arch_ops->decache_regs(vcpu);
1913         }
1914
1915         r = kvm_arch_ops->run(vcpu, kvm_run);
1916
1917 out:
1918         if (vcpu->sigset_active)
1919                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1920
1921         vcpu_put(vcpu);
1922         return r;
1923 }
1924
1925 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1926                                    struct kvm_regs *regs)
1927 {
1928         vcpu_load(vcpu);
1929
1930         kvm_arch_ops->cache_regs(vcpu);
1931
1932         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1933         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1934         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1935         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1936         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1937         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1938         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1939         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1940 #ifdef CONFIG_X86_64
1941         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1942         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1943         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1944         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1945         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1946         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1947         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1948         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1949 #endif
1950
1951         regs->rip = vcpu->rip;
1952         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1953
1954         /*
1955          * Don't leak debug flags in case they were set for guest debugging
1956          */
1957         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1958                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1959
1960         vcpu_put(vcpu);
1961
1962         return 0;
1963 }
1964
1965 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1966                                    struct kvm_regs *regs)
1967 {
1968         vcpu_load(vcpu);
1969
1970         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1971         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1972         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1973         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1974         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1975         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1976         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1977         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1978 #ifdef CONFIG_X86_64
1979         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1980         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1981         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1982         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1983         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1984         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1985         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1986         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1987 #endif
1988
1989         vcpu->rip = regs->rip;
1990         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
1991
1992         kvm_arch_ops->decache_regs(vcpu);
1993
1994         vcpu_put(vcpu);
1995
1996         return 0;
1997 }
1998
1999 static void get_segment(struct kvm_vcpu *vcpu,
2000                         struct kvm_segment *var, int seg)
2001 {
2002         return kvm_arch_ops->get_segment(vcpu, var, seg);
2003 }
2004
2005 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2006                                     struct kvm_sregs *sregs)
2007 {
2008         struct descriptor_table dt;
2009
2010         vcpu_load(vcpu);
2011
2012         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2013         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2014         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2015         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2016         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2017         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2018
2019         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2020         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2021
2022         kvm_arch_ops->get_idt(vcpu, &dt);
2023         sregs->idt.limit = dt.limit;
2024         sregs->idt.base = dt.base;
2025         kvm_arch_ops->get_gdt(vcpu, &dt);
2026         sregs->gdt.limit = dt.limit;
2027         sregs->gdt.base = dt.base;
2028
2029         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2030         sregs->cr0 = vcpu->cr0;
2031         sregs->cr2 = vcpu->cr2;
2032         sregs->cr3 = vcpu->cr3;
2033         sregs->cr4 = vcpu->cr4;
2034         sregs->cr8 = get_cr8(vcpu);
2035         sregs->efer = vcpu->shadow_efer;
2036         sregs->apic_base = kvm_get_apic_base(vcpu);
2037
2038         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2039                sizeof sregs->interrupt_bitmap);
2040
2041         vcpu_put(vcpu);
2042
2043         return 0;
2044 }
2045
2046 static void set_segment(struct kvm_vcpu *vcpu,
2047                         struct kvm_segment *var, int seg)
2048 {
2049         return kvm_arch_ops->set_segment(vcpu, var, seg);
2050 }
2051
2052 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2053                                     struct kvm_sregs *sregs)
2054 {
2055         int mmu_reset_needed = 0;
2056         int i;
2057         struct descriptor_table dt;
2058
2059         vcpu_load(vcpu);
2060
2061         dt.limit = sregs->idt.limit;
2062         dt.base = sregs->idt.base;
2063         kvm_arch_ops->set_idt(vcpu, &dt);
2064         dt.limit = sregs->gdt.limit;
2065         dt.base = sregs->gdt.base;
2066         kvm_arch_ops->set_gdt(vcpu, &dt);
2067
2068         vcpu->cr2 = sregs->cr2;
2069         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2070         vcpu->cr3 = sregs->cr3;
2071
2072         set_cr8(vcpu, sregs->cr8);
2073
2074         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2075 #ifdef CONFIG_X86_64
2076         kvm_arch_ops->set_efer(vcpu, sregs->efer);
2077 #endif
2078         kvm_set_apic_base(vcpu, sregs->apic_base);
2079
2080         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2081
2082         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2083         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2084
2085         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2086         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2087         if (!is_long_mode(vcpu) && is_pae(vcpu))
2088                 load_pdptrs(vcpu, vcpu->cr3);
2089
2090         if (mmu_reset_needed)
2091                 kvm_mmu_reset_context(vcpu);
2092
2093         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2094                sizeof vcpu->irq_pending);
2095         vcpu->irq_summary = 0;
2096         for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2097                 if (vcpu->irq_pending[i])
2098                         __set_bit(i, &vcpu->irq_summary);
2099
2100         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2101         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2102         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2103         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2104         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2105         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2106
2107         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2108         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2109
2110         vcpu_put(vcpu);
2111
2112         return 0;
2113 }
2114
2115 /*
2116  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2117  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2118  *
2119  * This list is modified at module load time to reflect the
2120  * capabilities of the host cpu.
2121  */
2122 static u32 msrs_to_save[] = {
2123         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2124         MSR_K6_STAR,
2125 #ifdef CONFIG_X86_64
2126         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2127 #endif
2128         MSR_IA32_TIME_STAMP_COUNTER,
2129 };
2130
2131 static unsigned num_msrs_to_save;
2132
2133 static u32 emulated_msrs[] = {
2134         MSR_IA32_MISC_ENABLE,
2135 };
2136
2137 static __init void kvm_init_msr_list(void)
2138 {
2139         u32 dummy[2];
2140         unsigned i, j;
2141
2142         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2143                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2144                         continue;
2145                 if (j < i)
2146                         msrs_to_save[j] = msrs_to_save[i];
2147                 j++;
2148         }
2149         num_msrs_to_save = j;
2150 }
2151
2152 /*
2153  * Adapt set_msr() to msr_io()'s calling convention
2154  */
2155 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2156 {
2157         return kvm_set_msr(vcpu, index, *data);
2158 }
2159
2160 /*
2161  * Read or write a bunch of msrs. All parameters are kernel addresses.
2162  *
2163  * @return number of msrs set successfully.
2164  */
2165 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2166                     struct kvm_msr_entry *entries,
2167                     int (*do_msr)(struct kvm_vcpu *vcpu,
2168                                   unsigned index, u64 *data))
2169 {
2170         int i;
2171
2172         vcpu_load(vcpu);
2173
2174         for (i = 0; i < msrs->nmsrs; ++i)
2175                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2176                         break;
2177
2178         vcpu_put(vcpu);
2179
2180         return i;
2181 }
2182
2183 /*
2184  * Read or write a bunch of msrs. Parameters are user addresses.
2185  *
2186  * @return number of msrs set successfully.
2187  */
2188 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2189                   int (*do_msr)(struct kvm_vcpu *vcpu,
2190                                 unsigned index, u64 *data),
2191                   int writeback)
2192 {
2193         struct kvm_msrs msrs;
2194         struct kvm_msr_entry *entries;
2195         int r, n;
2196         unsigned size;
2197
2198         r = -EFAULT;
2199         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2200                 goto out;
2201
2202         r = -E2BIG;
2203         if (msrs.nmsrs >= MAX_IO_MSRS)
2204                 goto out;
2205
2206         r = -ENOMEM;
2207         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2208         entries = vmalloc(size);
2209         if (!entries)
2210                 goto out;
2211
2212         r = -EFAULT;
2213         if (copy_from_user(entries, user_msrs->entries, size))
2214                 goto out_free;
2215
2216         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2217         if (r < 0)
2218                 goto out_free;
2219
2220         r = -EFAULT;
2221         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2222                 goto out_free;
2223
2224         r = n;
2225
2226 out_free:
2227         vfree(entries);
2228 out:
2229         return r;
2230 }
2231
2232 /*
2233  * Translate a guest virtual address to a guest physical address.
2234  */
2235 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2236                                     struct kvm_translation *tr)
2237 {
2238         unsigned long vaddr = tr->linear_address;
2239         gpa_t gpa;
2240
2241         vcpu_load(vcpu);
2242         mutex_lock(&vcpu->kvm->lock);
2243         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2244         tr->physical_address = gpa;
2245         tr->valid = gpa != UNMAPPED_GVA;
2246         tr->writeable = 1;
2247         tr->usermode = 0;
2248         mutex_unlock(&vcpu->kvm->lock);
2249         vcpu_put(vcpu);
2250
2251         return 0;
2252 }
2253
2254 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2255                                     struct kvm_interrupt *irq)
2256 {
2257         if (irq->irq < 0 || irq->irq >= 256)
2258                 return -EINVAL;
2259         vcpu_load(vcpu);
2260
2261         set_bit(irq->irq, vcpu->irq_pending);
2262         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2263
2264         vcpu_put(vcpu);
2265
2266         return 0;
2267 }
2268
2269 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2270                                       struct kvm_debug_guest *dbg)
2271 {
2272         int r;
2273
2274         vcpu_load(vcpu);
2275
2276         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2277
2278         vcpu_put(vcpu);
2279
2280         return r;
2281 }
2282
2283 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2284                                     unsigned long address,
2285                                     int *type)
2286 {
2287         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2288         unsigned long pgoff;
2289         struct page *page;
2290
2291         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2292         if (pgoff == 0)
2293                 page = virt_to_page(vcpu->run);
2294         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2295                 page = virt_to_page(vcpu->pio_data);
2296         else
2297                 return NOPAGE_SIGBUS;
2298         get_page(page);
2299         if (type != NULL)
2300                 *type = VM_FAULT_MINOR;
2301
2302         return page;
2303 }
2304
2305 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2306         .nopage = kvm_vcpu_nopage,
2307 };
2308
2309 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2310 {
2311         vma->vm_ops = &kvm_vcpu_vm_ops;
2312         return 0;
2313 }
2314
2315 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2316 {
2317         struct kvm_vcpu *vcpu = filp->private_data;
2318
2319         fput(vcpu->kvm->filp);
2320         return 0;
2321 }
2322
2323 static struct file_operations kvm_vcpu_fops = {
2324         .release        = kvm_vcpu_release,
2325         .unlocked_ioctl = kvm_vcpu_ioctl,
2326         .compat_ioctl   = kvm_vcpu_ioctl,
2327         .mmap           = kvm_vcpu_mmap,
2328 };
2329
2330 /*
2331  * Allocates an inode for the vcpu.
2332  */
2333 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2334 {
2335         int fd, r;
2336         struct inode *inode;
2337         struct file *file;
2338
2339         r = anon_inode_getfd(&fd, &inode, &file,
2340                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2341         if (r)
2342                 return r;
2343         atomic_inc(&vcpu->kvm->filp->f_count);
2344         return fd;
2345 }
2346
2347 /*
2348  * Creates some virtual cpus.  Good luck creating more than one.
2349  */
2350 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2351 {
2352         int r;
2353         struct kvm_vcpu *vcpu;
2354
2355         if (!valid_vcpu(n))
2356                 return -EINVAL;
2357
2358         vcpu = kvm_arch_ops->vcpu_create(kvm, n);
2359         if (IS_ERR(vcpu))
2360                 return PTR_ERR(vcpu);
2361
2362         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2363
2364         /* We do fxsave: this must be aligned. */
2365         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2366
2367         vcpu_load(vcpu);
2368         r = kvm_mmu_setup(vcpu);
2369         vcpu_put(vcpu);
2370         if (r < 0)
2371                 goto free_vcpu;
2372
2373         mutex_lock(&kvm->lock);
2374         if (kvm->vcpus[n]) {
2375                 r = -EEXIST;
2376                 mutex_unlock(&kvm->lock);
2377                 goto mmu_unload;
2378         }
2379         kvm->vcpus[n] = vcpu;
2380         mutex_unlock(&kvm->lock);
2381
2382         /* Now it's all set up, let userspace reach it */
2383         r = create_vcpu_fd(vcpu);
2384         if (r < 0)
2385                 goto unlink;
2386         return r;
2387
2388 unlink:
2389         mutex_lock(&kvm->lock);
2390         kvm->vcpus[n] = NULL;
2391         mutex_unlock(&kvm->lock);
2392
2393 mmu_unload:
2394         vcpu_load(vcpu);
2395         kvm_mmu_unload(vcpu);
2396         vcpu_put(vcpu);
2397
2398 free_vcpu:
2399         kvm_arch_ops->vcpu_free(vcpu);
2400         return r;
2401 }
2402
2403 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2404 {
2405         u64 efer;
2406         int i;
2407         struct kvm_cpuid_entry *e, *entry;
2408
2409         rdmsrl(MSR_EFER, efer);
2410         entry = NULL;
2411         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2412                 e = &vcpu->cpuid_entries[i];
2413                 if (e->function == 0x80000001) {
2414                         entry = e;
2415                         break;
2416                 }
2417         }
2418         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2419                 entry->edx &= ~(1 << 20);
2420                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2421         }
2422 }
2423
2424 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2425                                     struct kvm_cpuid *cpuid,
2426                                     struct kvm_cpuid_entry __user *entries)
2427 {
2428         int r;
2429
2430         r = -E2BIG;
2431         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2432                 goto out;
2433         r = -EFAULT;
2434         if (copy_from_user(&vcpu->cpuid_entries, entries,
2435                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2436                 goto out;
2437         vcpu->cpuid_nent = cpuid->nent;
2438         cpuid_fix_nx_cap(vcpu);
2439         return 0;
2440
2441 out:
2442         return r;
2443 }
2444
2445 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2446 {
2447         if (sigset) {
2448                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2449                 vcpu->sigset_active = 1;
2450                 vcpu->sigset = *sigset;
2451         } else
2452                 vcpu->sigset_active = 0;
2453         return 0;
2454 }
2455
2456 /*
2457  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2458  * we have asm/x86/processor.h
2459  */
2460 struct fxsave {
2461         u16     cwd;
2462         u16     swd;
2463         u16     twd;
2464         u16     fop;
2465         u64     rip;
2466         u64     rdp;
2467         u32     mxcsr;
2468         u32     mxcsr_mask;
2469         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2470 #ifdef CONFIG_X86_64
2471         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2472 #else
2473         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2474 #endif
2475 };
2476
2477 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2478 {
2479         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2480
2481         vcpu_load(vcpu);
2482
2483         memcpy(fpu->fpr, fxsave->st_space, 128);
2484         fpu->fcw = fxsave->cwd;
2485         fpu->fsw = fxsave->swd;
2486         fpu->ftwx = fxsave->twd;
2487         fpu->last_opcode = fxsave->fop;
2488         fpu->last_ip = fxsave->rip;
2489         fpu->last_dp = fxsave->rdp;
2490         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2491
2492         vcpu_put(vcpu);
2493
2494         return 0;
2495 }
2496
2497 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2498 {
2499         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2500
2501         vcpu_load(vcpu);
2502
2503         memcpy(fxsave->st_space, fpu->fpr, 128);
2504         fxsave->cwd = fpu->fcw;
2505         fxsave->swd = fpu->fsw;
2506         fxsave->twd = fpu->ftwx;
2507         fxsave->fop = fpu->last_opcode;
2508         fxsave->rip = fpu->last_ip;
2509         fxsave->rdp = fpu->last_dp;
2510         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2511
2512         vcpu_put(vcpu);
2513
2514         return 0;
2515 }
2516
2517 static long kvm_vcpu_ioctl(struct file *filp,
2518                            unsigned int ioctl, unsigned long arg)
2519 {
2520         struct kvm_vcpu *vcpu = filp->private_data;
2521         void __user *argp = (void __user *)arg;
2522         int r = -EINVAL;
2523
2524         switch (ioctl) {
2525         case KVM_RUN:
2526                 r = -EINVAL;
2527                 if (arg)
2528                         goto out;
2529                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2530                 break;
2531         case KVM_GET_REGS: {
2532                 struct kvm_regs kvm_regs;
2533
2534                 memset(&kvm_regs, 0, sizeof kvm_regs);
2535                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2536                 if (r)
2537                         goto out;
2538                 r = -EFAULT;
2539                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2540                         goto out;
2541                 r = 0;
2542                 break;
2543         }
2544         case KVM_SET_REGS: {
2545                 struct kvm_regs kvm_regs;
2546
2547                 r = -EFAULT;
2548                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2549                         goto out;
2550                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2551                 if (r)
2552                         goto out;
2553                 r = 0;
2554                 break;
2555         }
2556         case KVM_GET_SREGS: {
2557                 struct kvm_sregs kvm_sregs;
2558
2559                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2560                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2561                 if (r)
2562                         goto out;
2563                 r = -EFAULT;
2564                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2565                         goto out;
2566                 r = 0;
2567                 break;
2568         }
2569         case KVM_SET_SREGS: {
2570                 struct kvm_sregs kvm_sregs;
2571
2572                 r = -EFAULT;
2573                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2574                         goto out;
2575                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2576                 if (r)
2577                         goto out;
2578                 r = 0;
2579                 break;
2580         }
2581         case KVM_TRANSLATE: {
2582                 struct kvm_translation tr;
2583
2584                 r = -EFAULT;
2585                 if (copy_from_user(&tr, argp, sizeof tr))
2586                         goto out;
2587                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2588                 if (r)
2589                         goto out;
2590                 r = -EFAULT;
2591                 if (copy_to_user(argp, &tr, sizeof tr))
2592                         goto out;
2593                 r = 0;
2594                 break;
2595         }
2596         case KVM_INTERRUPT: {
2597                 struct kvm_interrupt irq;
2598
2599                 r = -EFAULT;
2600                 if (copy_from_user(&irq, argp, sizeof irq))
2601                         goto out;
2602                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2603                 if (r)
2604                         goto out;
2605                 r = 0;
2606                 break;
2607         }
2608         case KVM_DEBUG_GUEST: {
2609                 struct kvm_debug_guest dbg;
2610
2611                 r = -EFAULT;
2612                 if (copy_from_user(&dbg, argp, sizeof dbg))
2613                         goto out;
2614                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2615                 if (r)
2616                         goto out;
2617                 r = 0;
2618                 break;
2619         }
2620         case KVM_GET_MSRS:
2621                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2622                 break;
2623         case KVM_SET_MSRS:
2624                 r = msr_io(vcpu, argp, do_set_msr, 0);
2625                 break;
2626         case KVM_SET_CPUID: {
2627                 struct kvm_cpuid __user *cpuid_arg = argp;
2628                 struct kvm_cpuid cpuid;
2629
2630                 r = -EFAULT;
2631                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2632                         goto out;
2633                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2634                 if (r)
2635                         goto out;
2636                 break;
2637         }
2638         case KVM_SET_SIGNAL_MASK: {
2639                 struct kvm_signal_mask __user *sigmask_arg = argp;
2640                 struct kvm_signal_mask kvm_sigmask;
2641                 sigset_t sigset, *p;
2642
2643                 p = NULL;
2644                 if (argp) {
2645                         r = -EFAULT;
2646                         if (copy_from_user(&kvm_sigmask, argp,
2647                                            sizeof kvm_sigmask))
2648                                 goto out;
2649                         r = -EINVAL;
2650                         if (kvm_sigmask.len != sizeof sigset)
2651                                 goto out;
2652                         r = -EFAULT;
2653                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2654                                            sizeof sigset))
2655                                 goto out;
2656                         p = &sigset;
2657                 }
2658                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2659                 break;
2660         }
2661         case KVM_GET_FPU: {
2662                 struct kvm_fpu fpu;
2663
2664                 memset(&fpu, 0, sizeof fpu);
2665                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2666                 if (r)
2667                         goto out;
2668                 r = -EFAULT;
2669                 if (copy_to_user(argp, &fpu, sizeof fpu))
2670                         goto out;
2671                 r = 0;
2672                 break;
2673         }
2674         case KVM_SET_FPU: {
2675                 struct kvm_fpu fpu;
2676
2677                 r = -EFAULT;
2678                 if (copy_from_user(&fpu, argp, sizeof fpu))
2679                         goto out;
2680                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2681                 if (r)
2682                         goto out;
2683                 r = 0;
2684                 break;
2685         }
2686         default:
2687                 ;
2688         }
2689 out:
2690         return r;
2691 }
2692
2693 static long kvm_vm_ioctl(struct file *filp,
2694                            unsigned int ioctl, unsigned long arg)
2695 {
2696         struct kvm *kvm = filp->private_data;
2697         void __user *argp = (void __user *)arg;
2698         int r = -EINVAL;
2699
2700         switch (ioctl) {
2701         case KVM_CREATE_VCPU:
2702                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2703                 if (r < 0)
2704                         goto out;
2705                 break;
2706         case KVM_SET_MEMORY_REGION: {
2707                 struct kvm_memory_region kvm_mem;
2708
2709                 r = -EFAULT;
2710                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2711                         goto out;
2712                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2713                 if (r)
2714                         goto out;
2715                 break;
2716         }
2717         case KVM_GET_DIRTY_LOG: {
2718                 struct kvm_dirty_log log;
2719
2720                 r = -EFAULT;
2721                 if (copy_from_user(&log, argp, sizeof log))
2722                         goto out;
2723                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2724                 if (r)
2725                         goto out;
2726                 break;
2727         }
2728         case KVM_SET_MEMORY_ALIAS: {
2729                 struct kvm_memory_alias alias;
2730
2731                 r = -EFAULT;
2732                 if (copy_from_user(&alias, argp, sizeof alias))
2733                         goto out;
2734                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2735                 if (r)
2736                         goto out;
2737                 break;
2738         }
2739         case KVM_CREATE_IRQCHIP:
2740                 r = -ENOMEM;
2741                 kvm->vpic = kvm_create_pic(kvm);
2742                 if (kvm->vpic)
2743                         r = 0;
2744                 else
2745                         goto out;
2746                 break;
2747         case KVM_IRQ_LINE: {
2748                 struct kvm_irq_level irq_event;
2749
2750                 r = -EFAULT;
2751                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2752                         goto out;
2753                 if (irqchip_in_kernel(kvm)) {
2754                         if (irq_event.irq < 16)
2755                                 kvm_pic_set_irq(pic_irqchip(kvm),
2756                                         irq_event.irq,
2757                                         irq_event.level);
2758                         /* TODO: IOAPIC */
2759                         r = 0;
2760                 }
2761                 break;
2762         }
2763         default:
2764                 ;
2765         }
2766 out:
2767         return r;
2768 }
2769
2770 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2771                                   unsigned long address,
2772                                   int *type)
2773 {
2774         struct kvm *kvm = vma->vm_file->private_data;
2775         unsigned long pgoff;
2776         struct page *page;
2777
2778         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2779         page = gfn_to_page(kvm, pgoff);
2780         if (!page)
2781                 return NOPAGE_SIGBUS;
2782         get_page(page);
2783         if (type != NULL)
2784                 *type = VM_FAULT_MINOR;
2785
2786         return page;
2787 }
2788
2789 static struct vm_operations_struct kvm_vm_vm_ops = {
2790         .nopage = kvm_vm_nopage,
2791 };
2792
2793 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2794 {
2795         vma->vm_ops = &kvm_vm_vm_ops;
2796         return 0;
2797 }
2798
2799 static struct file_operations kvm_vm_fops = {
2800         .release        = kvm_vm_release,
2801         .unlocked_ioctl = kvm_vm_ioctl,
2802         .compat_ioctl   = kvm_vm_ioctl,
2803         .mmap           = kvm_vm_mmap,
2804 };
2805
2806 static int kvm_dev_ioctl_create_vm(void)
2807 {
2808         int fd, r;
2809         struct inode *inode;
2810         struct file *file;
2811         struct kvm *kvm;
2812
2813         kvm = kvm_create_vm();
2814         if (IS_ERR(kvm))
2815                 return PTR_ERR(kvm);
2816         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2817         if (r) {
2818                 kvm_destroy_vm(kvm);
2819                 return r;
2820         }
2821
2822         kvm->filp = file;
2823
2824         return fd;
2825 }
2826
2827 static long kvm_dev_ioctl(struct file *filp,
2828                           unsigned int ioctl, unsigned long arg)
2829 {
2830         void __user *argp = (void __user *)arg;
2831         long r = -EINVAL;
2832
2833         switch (ioctl) {
2834         case KVM_GET_API_VERSION:
2835                 r = -EINVAL;
2836                 if (arg)
2837                         goto out;
2838                 r = KVM_API_VERSION;
2839                 break;
2840         case KVM_CREATE_VM:
2841                 r = -EINVAL;
2842                 if (arg)
2843                         goto out;
2844                 r = kvm_dev_ioctl_create_vm();
2845                 break;
2846         case KVM_GET_MSR_INDEX_LIST: {
2847                 struct kvm_msr_list __user *user_msr_list = argp;
2848                 struct kvm_msr_list msr_list;
2849                 unsigned n;
2850
2851                 r = -EFAULT;
2852                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2853                         goto out;
2854                 n = msr_list.nmsrs;
2855                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2856                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2857                         goto out;
2858                 r = -E2BIG;
2859                 if (n < num_msrs_to_save)
2860                         goto out;
2861                 r = -EFAULT;
2862                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2863                                  num_msrs_to_save * sizeof(u32)))
2864                         goto out;
2865                 if (copy_to_user(user_msr_list->indices
2866                                  + num_msrs_to_save * sizeof(u32),
2867                                  &emulated_msrs,
2868                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2869                         goto out;
2870                 r = 0;
2871                 break;
2872         }
2873         case KVM_CHECK_EXTENSION: {
2874                 int ext = (long)argp;
2875
2876                 switch (ext) {
2877                 case KVM_CAP_IRQCHIP:
2878                         r = 1;
2879                         break;
2880                 default:
2881                         r = 0;
2882                         break;
2883                 }
2884                 break;
2885         }
2886         case KVM_GET_VCPU_MMAP_SIZE:
2887                 r = -EINVAL;
2888                 if (arg)
2889                         goto out;
2890                 r = 2 * PAGE_SIZE;
2891                 break;
2892         default:
2893                 ;
2894         }
2895 out:
2896         return r;
2897 }
2898
2899 static struct file_operations kvm_chardev_ops = {
2900         .unlocked_ioctl = kvm_dev_ioctl,
2901         .compat_ioctl   = kvm_dev_ioctl,
2902 };
2903
2904 static struct miscdevice kvm_dev = {
2905         KVM_MINOR,
2906         "kvm",
2907         &kvm_chardev_ops,
2908 };
2909
2910 /*
2911  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2912  * cached on it.
2913  */
2914 static void decache_vcpus_on_cpu(int cpu)
2915 {
2916         struct kvm *vm;
2917         struct kvm_vcpu *vcpu;
2918         int i;
2919
2920         spin_lock(&kvm_lock);
2921         list_for_each_entry(vm, &vm_list, vm_list)
2922                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2923                         vcpu = vm->vcpus[i];
2924                         if (!vcpu)
2925                                 continue;
2926                         /*
2927                          * If the vcpu is locked, then it is running on some
2928                          * other cpu and therefore it is not cached on the
2929                          * cpu in question.
2930                          *
2931                          * If it's not locked, check the last cpu it executed
2932                          * on.
2933                          */
2934                         if (mutex_trylock(&vcpu->mutex)) {
2935                                 if (vcpu->cpu == cpu) {
2936                                         kvm_arch_ops->vcpu_decache(vcpu);
2937                                         vcpu->cpu = -1;
2938                                 }
2939                                 mutex_unlock(&vcpu->mutex);
2940                         }
2941                 }
2942         spin_unlock(&kvm_lock);
2943 }
2944
2945 static void hardware_enable(void *junk)
2946 {
2947         int cpu = raw_smp_processor_id();
2948
2949         if (cpu_isset(cpu, cpus_hardware_enabled))
2950                 return;
2951         cpu_set(cpu, cpus_hardware_enabled);
2952         kvm_arch_ops->hardware_enable(NULL);
2953 }
2954
2955 static void hardware_disable(void *junk)
2956 {
2957         int cpu = raw_smp_processor_id();
2958
2959         if (!cpu_isset(cpu, cpus_hardware_enabled))
2960                 return;
2961         cpu_clear(cpu, cpus_hardware_enabled);
2962         decache_vcpus_on_cpu(cpu);
2963         kvm_arch_ops->hardware_disable(NULL);
2964 }
2965
2966 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2967                            void *v)
2968 {
2969         int cpu = (long)v;
2970
2971         switch (val) {
2972         case CPU_DYING:
2973         case CPU_DYING_FROZEN:
2974                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2975                        cpu);
2976                 hardware_disable(NULL);
2977                 break;
2978         case CPU_UP_CANCELED:
2979         case CPU_UP_CANCELED_FROZEN:
2980                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2981                        cpu);
2982                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
2983                 break;
2984         case CPU_ONLINE:
2985         case CPU_ONLINE_FROZEN:
2986                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2987                        cpu);
2988                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
2989                 break;
2990         }
2991         return NOTIFY_OK;
2992 }
2993
2994 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2995                        void *v)
2996 {
2997         if (val == SYS_RESTART) {
2998                 /*
2999                  * Some (well, at least mine) BIOSes hang on reboot if
3000                  * in vmx root mode.
3001                  */
3002                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3003                 on_each_cpu(hardware_disable, NULL, 0, 1);
3004         }
3005         return NOTIFY_OK;
3006 }
3007
3008 static struct notifier_block kvm_reboot_notifier = {
3009         .notifier_call = kvm_reboot,
3010         .priority = 0,
3011 };
3012
3013 void kvm_io_bus_init(struct kvm_io_bus *bus)
3014 {
3015         memset(bus, 0, sizeof(*bus));
3016 }
3017
3018 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3019 {
3020         int i;
3021
3022         for (i = 0; i < bus->dev_count; i++) {
3023                 struct kvm_io_device *pos = bus->devs[i];
3024
3025                 kvm_iodevice_destructor(pos);
3026         }
3027 }
3028
3029 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3030 {
3031         int i;
3032
3033         for (i = 0; i < bus->dev_count; i++) {
3034                 struct kvm_io_device *pos = bus->devs[i];
3035
3036                 if (pos->in_range(pos, addr))
3037                         return pos;
3038         }
3039
3040         return NULL;
3041 }
3042
3043 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3044 {
3045         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3046
3047         bus->devs[bus->dev_count++] = dev;
3048 }
3049
3050 static struct notifier_block kvm_cpu_notifier = {
3051         .notifier_call = kvm_cpu_hotplug,
3052         .priority = 20, /* must be > scheduler priority */
3053 };
3054
3055 static u64 stat_get(void *_offset)
3056 {
3057         unsigned offset = (long)_offset;
3058         u64 total = 0;
3059         struct kvm *kvm;
3060         struct kvm_vcpu *vcpu;
3061         int i;
3062
3063         spin_lock(&kvm_lock);
3064         list_for_each_entry(kvm, &vm_list, vm_list)
3065                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3066                         vcpu = kvm->vcpus[i];
3067                         if (vcpu)
3068                                 total += *(u32 *)((void *)vcpu + offset);
3069                 }
3070         spin_unlock(&kvm_lock);
3071         return total;
3072 }
3073
3074 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3075
3076 static __init void kvm_init_debug(void)
3077 {
3078         struct kvm_stats_debugfs_item *p;
3079
3080         debugfs_dir = debugfs_create_dir("kvm", NULL);
3081         for (p = debugfs_entries; p->name; ++p)
3082                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3083                                                 (void *)(long)p->offset,
3084                                                 &stat_fops);
3085 }
3086
3087 static void kvm_exit_debug(void)
3088 {
3089         struct kvm_stats_debugfs_item *p;
3090
3091         for (p = debugfs_entries; p->name; ++p)
3092                 debugfs_remove(p->dentry);
3093         debugfs_remove(debugfs_dir);
3094 }
3095
3096 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3097 {
3098         hardware_disable(NULL);
3099         return 0;
3100 }
3101
3102 static int kvm_resume(struct sys_device *dev)
3103 {
3104         hardware_enable(NULL);
3105         return 0;
3106 }
3107
3108 static struct sysdev_class kvm_sysdev_class = {
3109         set_kset_name("kvm"),
3110         .suspend = kvm_suspend,
3111         .resume = kvm_resume,
3112 };
3113
3114 static struct sys_device kvm_sysdev = {
3115         .id = 0,
3116         .cls = &kvm_sysdev_class,
3117 };
3118
3119 hpa_t bad_page_address;
3120
3121 static inline
3122 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3123 {
3124         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3125 }
3126
3127 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3128 {
3129         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3130
3131         kvm_arch_ops->vcpu_load(vcpu, cpu);
3132 }
3133
3134 static void kvm_sched_out(struct preempt_notifier *pn,
3135                           struct task_struct *next)
3136 {
3137         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3138
3139         kvm_arch_ops->vcpu_put(vcpu);
3140 }
3141
3142 int kvm_init_arch(struct kvm_arch_ops *ops, unsigned int vcpu_size,
3143                   struct module *module)
3144 {
3145         int r;
3146         int cpu;
3147
3148         if (kvm_arch_ops) {
3149                 printk(KERN_ERR "kvm: already loaded the other module\n");
3150                 return -EEXIST;
3151         }
3152
3153         if (!ops->cpu_has_kvm_support()) {
3154                 printk(KERN_ERR "kvm: no hardware support\n");
3155                 return -EOPNOTSUPP;
3156         }
3157         if (ops->disabled_by_bios()) {
3158                 printk(KERN_ERR "kvm: disabled by bios\n");
3159                 return -EOPNOTSUPP;
3160         }
3161
3162         kvm_arch_ops = ops;
3163
3164         r = kvm_arch_ops->hardware_setup();
3165         if (r < 0)
3166                 goto out;
3167
3168         for_each_online_cpu(cpu) {
3169                 smp_call_function_single(cpu,
3170                                 kvm_arch_ops->check_processor_compatibility,
3171                                 &r, 0, 1);
3172                 if (r < 0)
3173                         goto out_free_0;
3174         }
3175
3176         on_each_cpu(hardware_enable, NULL, 0, 1);
3177         r = register_cpu_notifier(&kvm_cpu_notifier);
3178         if (r)
3179                 goto out_free_1;
3180         register_reboot_notifier(&kvm_reboot_notifier);
3181
3182         r = sysdev_class_register(&kvm_sysdev_class);
3183         if (r)
3184                 goto out_free_2;
3185
3186         r = sysdev_register(&kvm_sysdev);
3187         if (r)
3188                 goto out_free_3;
3189
3190         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3191         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3192                                            __alignof__(struct kvm_vcpu), 0, 0);
3193         if (!kvm_vcpu_cache) {
3194                 r = -ENOMEM;
3195                 goto out_free_4;
3196         }
3197
3198         kvm_chardev_ops.owner = module;
3199
3200         r = misc_register(&kvm_dev);
3201         if (r) {
3202                 printk (KERN_ERR "kvm: misc device register failed\n");
3203                 goto out_free;
3204         }
3205
3206         kvm_preempt_ops.sched_in = kvm_sched_in;
3207         kvm_preempt_ops.sched_out = kvm_sched_out;
3208
3209         return r;
3210
3211 out_free:
3212         kmem_cache_destroy(kvm_vcpu_cache);
3213 out_free_4:
3214         sysdev_unregister(&kvm_sysdev);
3215 out_free_3:
3216         sysdev_class_unregister(&kvm_sysdev_class);
3217 out_free_2:
3218         unregister_reboot_notifier(&kvm_reboot_notifier);
3219         unregister_cpu_notifier(&kvm_cpu_notifier);
3220 out_free_1:
3221         on_each_cpu(hardware_disable, NULL, 0, 1);
3222 out_free_0:
3223         kvm_arch_ops->hardware_unsetup();
3224 out:
3225         kvm_arch_ops = NULL;
3226         return r;
3227 }
3228
3229 void kvm_exit_arch(void)
3230 {
3231         misc_deregister(&kvm_dev);
3232         kmem_cache_destroy(kvm_vcpu_cache);
3233         sysdev_unregister(&kvm_sysdev);
3234         sysdev_class_unregister(&kvm_sysdev_class);
3235         unregister_reboot_notifier(&kvm_reboot_notifier);
3236         unregister_cpu_notifier(&kvm_cpu_notifier);
3237         on_each_cpu(hardware_disable, NULL, 0, 1);
3238         kvm_arch_ops->hardware_unsetup();
3239         kvm_arch_ops = NULL;
3240 }
3241
3242 static __init int kvm_init(void)
3243 {
3244         static struct page *bad_page;
3245         int r;
3246
3247         r = kvm_mmu_module_init();
3248         if (r)
3249                 goto out4;
3250
3251         kvm_init_debug();
3252
3253         kvm_init_msr_list();
3254
3255         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3256                 r = -ENOMEM;
3257                 goto out;
3258         }
3259
3260         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3261         memset(__va(bad_page_address), 0, PAGE_SIZE);
3262
3263         return 0;
3264
3265 out:
3266         kvm_exit_debug();
3267         kvm_mmu_module_exit();
3268 out4:
3269         return r;
3270 }
3271
3272 static __exit void kvm_exit(void)
3273 {
3274         kvm_exit_debug();
3275         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3276         kvm_mmu_module_exit();
3277 }
3278
3279 module_init(kvm_init)
3280 module_exit(kvm_exit)
3281
3282 EXPORT_SYMBOL_GPL(kvm_init_arch);
3283 EXPORT_SYMBOL_GPL(kvm_exit_arch);