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         { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
  80         { "request_irq", STAT_OFFSET(request_irq_exits) },
  81         { "irq_exits", STAT_OFFSET(irq_exits) },
  82         { "light_exits", STAT_OFFSET(light_exits) },
  83         { "efer_reload", STAT_OFFSET(efer_reload) },
  84         { NULL }
  85 };
  86
  87 static struct dentry *debugfs_dir;
  88
  89 #define MAX_IO_MSRS 256
  90
  91 #define CR0_RESERVED_BITS                                               \
  92         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  93                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  94                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  95 #define CR4_RESERVED_BITS                                               \
  96         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  97                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  98                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  99                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
 100
 101 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
 102 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
 103
 104 #ifdef CONFIG_X86_64
 105 // LDT or TSS descriptor in the GDT. 16 bytes.
 106 struct segment_descriptor_64 {
 107         struct segment_descriptor s;
 108         u32 base_higher;
 109         u32 pad_zero;
 110 };
 111
 112 #endif
 113
 114 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 115                            unsigned long arg);
 116
 117 unsigned long segment_base(u16 selector)
 118 {
 119         struct descriptor_table gdt;
 120         struct segment_descriptor *d;
 121         unsigned long table_base;
 122         typedef unsigned long ul;
 123         unsigned long v;
 124
 125         if (selector == 0)
 126                 return 0;
 127
 128         asm ("sgdt %0" : "=m"(gdt));
 129         table_base = gdt.base;
 130
 131         if (selector & 4) {           /* from ldt */
 132                 u16 ldt_selector;
 133
 134                 asm ("sldt %0" : "=g"(ldt_selector));
 135                 table_base = segment_base(ldt_selector);
 136         }
 137         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 138         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 139 #ifdef CONFIG_X86_64
 140         if (d->system == 0
 141             && (d->type == 2 || d->type == 9 || d->type == 11))
 142                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 143 #endif
 144         return v;
 145 }
 146 EXPORT_SYMBOL_GPL(segment_base);
 147
 148 static inline int valid_vcpu(int n)
 149 {
 150         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 151 }
 152
 153 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 154 {
 155         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 156                 return;
 157
 158         vcpu->guest_fpu_loaded = 1;
 159         fx_save(&vcpu->host_fx_image);
 160         fx_restore(&vcpu->guest_fx_image);
 161 }
 162 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 163
 164 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 165 {
 166         if (!vcpu->guest_fpu_loaded)
 167                 return;
 168
 169         vcpu->guest_fpu_loaded = 0;
 170         fx_save(&vcpu->guest_fx_image);
 171         fx_restore(&vcpu->host_fx_image);
 172 }
 173 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 174
 175 /*
 176  * Switches to specified vcpu, until a matching vcpu_put()
 177  */
 178 static void vcpu_load(struct kvm_vcpu *vcpu)
 179 {
 180         int cpu;
 181
 182         mutex_lock(&vcpu->mutex);
 183         cpu = get_cpu();
 184         preempt_notifier_register(&vcpu->preempt_notifier);
 185         kvm_arch_ops->vcpu_load(vcpu, cpu);
 186         put_cpu();
 187 }
 188
 189 static void vcpu_put(struct kvm_vcpu *vcpu)
 190 {
 191         preempt_disable();
 192         kvm_arch_ops->vcpu_put(vcpu);
 193         preempt_notifier_unregister(&vcpu->preempt_notifier);
 194         preempt_enable();
 195         mutex_unlock(&vcpu->mutex);
 196 }
 197
 198 static void ack_flush(void *_completed)
 199 {
 200         atomic_t *completed = _completed;
 201
 202         atomic_inc(completed);
 203 }
 204
 205 void kvm_flush_remote_tlbs(struct kvm *kvm)
 206 {
 207         int i, cpu, needed;
 208         cpumask_t cpus;
 209         struct kvm_vcpu *vcpu;
 210         atomic_t completed;
 211
 212         atomic_set(&completed, 0);
 213         cpus_clear(cpus);
 214         needed = 0;
 215         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 216                 vcpu = kvm->vcpus[i];
 217                 if (!vcpu)
 218                         continue;
 219                 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
 220                         continue;
 221                 cpu = vcpu->cpu;
 222                 if (cpu != -1 && cpu != raw_smp_processor_id())
 223                         if (!cpu_isset(cpu, cpus)) {
 224                                 cpu_set(cpu, cpus);
 225                                 ++needed;
 226                         }
 227         }
 228
 229         /*
 230          * We really want smp_call_function_mask() here.  But that's not
 231          * available, so ipi all cpus in parallel and wait for them
 232          * to complete.
 233          */
 234         for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
 235                 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
 236         while (atomic_read(&completed) != needed) {
 237                 cpu_relax();
 238                 barrier();
 239         }
 240 }
 241
 242 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 243 {
 244         struct page *page;
 245         int r;
 246
 247         mutex_init(&vcpu->mutex);
 248         vcpu->cpu = -1;
 249         vcpu->mmu.root_hpa = INVALID_PAGE;
 250         vcpu->kvm = kvm;
 251         vcpu->vcpu_id = id;
 252         init_waitqueue_head(&vcpu->wq);
 253
 254         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 255         if (!page) {
 256                 r = -ENOMEM;
 257                 goto fail;
 258         }
 259         vcpu->run = page_address(page);
 260
 261         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 262         if (!page) {
 263                 r = -ENOMEM;
 264                 goto fail_free_run;
 265         }
 266         vcpu->pio_data = page_address(page);
 267
 268         r = kvm_mmu_create(vcpu);
 269         if (r < 0)
 270                 goto fail_free_pio_data;
 271
 272         return 0;
 273
 274 fail_free_pio_data:
 275         free_page((unsigned long)vcpu->pio_data);
 276 fail_free_run:
 277         free_page((unsigned long)vcpu->run);
 278 fail:
 279         return -ENOMEM;
 280 }
 281 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 282
 283 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 284 {
 285         kvm_mmu_destroy(vcpu);
 286         kvm_free_apic(vcpu->apic);
 287         free_page((unsigned long)vcpu->pio_data);
 288         free_page((unsigned long)vcpu->run);
 289 }
 290 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 291
 292 static struct kvm *kvm_create_vm(void)
 293 {
 294         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 295
 296         if (!kvm)
 297                 return ERR_PTR(-ENOMEM);
 298
 299         kvm_io_bus_init(&kvm->pio_bus);
 300         mutex_init(&kvm->lock);
 301         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 302         kvm_io_bus_init(&kvm->mmio_bus);
 303         spin_lock(&kvm_lock);
 304         list_add(&kvm->vm_list, &vm_list);
 305         spin_unlock(&kvm_lock);
 306         return kvm;
 307 }
 308
 309 /*
 310  * Free any memory in @free but not in @dont.
 311  */
 312 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 313                                   struct kvm_memory_slot *dont)
 314 {
 315         int i;
 316
 317         if (!dont || free->phys_mem != dont->phys_mem)
 318                 if (free->phys_mem) {
 319                         for (i = 0; i < free->npages; ++i)
 320                                 if (free->phys_mem[i])
 321                                         __free_page(free->phys_mem[i]);
 322                         vfree(free->phys_mem);
 323                 }
 324
 325         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 326                 vfree(free->dirty_bitmap);
 327
 328         free->phys_mem = NULL;
 329         free->npages = 0;
 330         free->dirty_bitmap = NULL;
 331 }
 332
 333 static void kvm_free_physmem(struct kvm *kvm)
 334 {
 335         int i;
 336
 337         for (i = 0; i < kvm->nmemslots; ++i)
 338                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 339 }
 340
 341 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 342 {
 343         int i;
 344
 345         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
 346                 if (vcpu->pio.guest_pages[i]) {
 347                         __free_page(vcpu->pio.guest_pages[i]);
 348                         vcpu->pio.guest_pages[i] = NULL;
 349                 }
 350 }
 351
 352 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 353 {
 354         vcpu_load(vcpu);
 355         kvm_mmu_unload(vcpu);
 356         vcpu_put(vcpu);
 357 }
 358
 359 static void kvm_free_vcpus(struct kvm *kvm)
 360 {
 361         unsigned int i;
 362
 363         /*
 364          * Unpin any mmu pages first.
 365          */
 366         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 367                 if (kvm->vcpus[i])
 368                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
 369         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 370                 if (kvm->vcpus[i]) {
 371                         kvm_arch_ops->vcpu_free(kvm->vcpus[i]);
 372                         kvm->vcpus[i] = NULL;
 373                 }
 374         }
 375
 376 }
 377
 378 static void kvm_destroy_vm(struct kvm *kvm)
 379 {
 380         spin_lock(&kvm_lock);
 381         list_del(&kvm->vm_list);
 382         spin_unlock(&kvm_lock);
 383         kvm_io_bus_destroy(&kvm->pio_bus);
 384         kvm_io_bus_destroy(&kvm->mmio_bus);
 385         kfree(kvm->vpic);
 386         kfree(kvm->vioapic);
 387         kvm_free_vcpus(kvm);
 388         kvm_free_physmem(kvm);
 389         kfree(kvm);
 390 }
 391
 392 static int kvm_vm_release(struct inode *inode, struct file *filp)
 393 {
 394         struct kvm *kvm = filp->private_data;
 395
 396         kvm_destroy_vm(kvm);
 397         return 0;
 398 }
 399
 400 static void inject_gp(struct kvm_vcpu *vcpu)
 401 {
 402         kvm_arch_ops->inject_gp(vcpu, 0);
 403 }
 404
 405 /*
 406  * Load the pae pdptrs.  Return true is they are all valid.
 407  */
 408 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 409 {
 410         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 411         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 412         int i;
 413         u64 *pdpt;
 414         int ret;
 415         struct page *page;
 416         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 417
 418         mutex_lock(&vcpu->kvm->lock);
 419         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
 420         if (!page) {
 421                 ret = 0;
 422                 goto out;
 423         }
 424
 425         pdpt = kmap_atomic(page, KM_USER0);
 426         memcpy(pdpte, pdpt+offset, sizeof(pdpte));
 427         kunmap_atomic(pdpt, KM_USER0);
 428
 429         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 430                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 431                         ret = 0;
 432                         goto out;
 433                 }
 434         }
 435         ret = 1;
 436
 437         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 438 out:
 439         mutex_unlock(&vcpu->kvm->lock);
 440
 441         return ret;
 442 }
 443
 444 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 445 {
 446         if (cr0 & CR0_RESERVED_BITS) {
 447                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 448                        cr0, vcpu->cr0);
 449                 inject_gp(vcpu);
 450                 return;
 451         }
 452
 453         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 454                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 455                 inject_gp(vcpu);
 456                 return;
 457         }
 458
 459         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 460                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 461                        "and a clear PE flag\n");
 462                 inject_gp(vcpu);
 463                 return;
 464         }
 465
 466         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 467 #ifdef CONFIG_X86_64
 468                 if ((vcpu->shadow_efer & EFER_LME)) {
 469                         int cs_db, cs_l;
 470
 471                         if (!is_pae(vcpu)) {
 472                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 473                                        "in long mode while PAE is disabled\n");
 474                                 inject_gp(vcpu);
 475                                 return;
 476                         }
 477                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 478                         if (cs_l) {
 479                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 480                                        "in long mode while CS.L == 1\n");
 481                                 inject_gp(vcpu);
 482                                 return;
 483
 484                         }
 485                 } else
 486 #endif
 487                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 488                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 489                                "reserved bits\n");
 490                         inject_gp(vcpu);
 491                         return;
 492                 }
 493
 494         }
 495
 496         kvm_arch_ops->set_cr0(vcpu, cr0);
 497         vcpu->cr0 = cr0;
 498
 499         mutex_lock(&vcpu->kvm->lock);
 500         kvm_mmu_reset_context(vcpu);
 501         mutex_unlock(&vcpu->kvm->lock);
 502         return;
 503 }
 504 EXPORT_SYMBOL_GPL(set_cr0);
 505
 506 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 507 {
 508         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 509 }
 510 EXPORT_SYMBOL_GPL(lmsw);
 511
 512 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 513 {
 514         if (cr4 & CR4_RESERVED_BITS) {
 515                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 516                 inject_gp(vcpu);
 517                 return;
 518         }
 519
 520         if (is_long_mode(vcpu)) {
 521                 if (!(cr4 & X86_CR4_PAE)) {
 522                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 523                                "in long mode\n");
 524                         inject_gp(vcpu);
 525                         return;
 526                 }
 527         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 528                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 529                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 530                 inject_gp(vcpu);
 531                 return;
 532         }
 533
 534         if (cr4 & X86_CR4_VMXE) {
 535                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 536                 inject_gp(vcpu);
 537                 return;
 538         }
 539         kvm_arch_ops->set_cr4(vcpu, cr4);
 540         mutex_lock(&vcpu->kvm->lock);
 541         kvm_mmu_reset_context(vcpu);
 542         mutex_unlock(&vcpu->kvm->lock);
 543 }
 544 EXPORT_SYMBOL_GPL(set_cr4);
 545
 546 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 547 {
 548         if (is_long_mode(vcpu)) {
 549                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 550                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 551                         inject_gp(vcpu);
 552                         return;
 553                 }
 554         } else {
 555                 if (is_pae(vcpu)) {
 556                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 557                                 printk(KERN_DEBUG
 558                                        "set_cr3: #GP, reserved bits\n");
 559                                 inject_gp(vcpu);
 560                                 return;
 561                         }
 562                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 563                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 564                                        "reserved bits\n");
 565                                 inject_gp(vcpu);
 566                                 return;
 567                         }
 568                 } else {
 569                         if (cr3 & CR3_NONPAE_RESERVED_BITS) {
 570                                 printk(KERN_DEBUG
 571                                        "set_cr3: #GP, reserved bits\n");
 572                                 inject_gp(vcpu);
 573                                 return;
 574                         }
 575                 }
 576         }
 577
 578         mutex_lock(&vcpu->kvm->lock);
 579         /*
 580          * Does the new cr3 value map to physical memory? (Note, we
 581          * catch an invalid cr3 even in real-mode, because it would
 582          * cause trouble later on when we turn on paging anyway.)
 583          *
 584          * A real CPU would silently accept an invalid cr3 and would
 585          * attempt to use it - with largely undefined (and often hard
 586          * to debug) behavior on the guest side.
 587          */
 588         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 589                 inject_gp(vcpu);
 590         else {
 591                 vcpu->cr3 = cr3;
 592                 vcpu->mmu.new_cr3(vcpu);
 593         }
 594         mutex_unlock(&vcpu->kvm->lock);
 595 }
 596 EXPORT_SYMBOL_GPL(set_cr3);
 597
 598 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 599 {
 600         if (cr8 & CR8_RESERVED_BITS) {
 601                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 602                 inject_gp(vcpu);
 603                 return;
 604         }
 605         if (irqchip_in_kernel(vcpu->kvm))
 606                 kvm_lapic_set_tpr(vcpu, cr8);
 607         else
 608                 vcpu->cr8 = cr8;
 609 }
 610 EXPORT_SYMBOL_GPL(set_cr8);
 611
 612 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 613 {
 614         if (irqchip_in_kernel(vcpu->kvm))
 615                 return kvm_lapic_get_cr8(vcpu);
 616         else
 617                 return vcpu->cr8;
 618 }
 619 EXPORT_SYMBOL_GPL(get_cr8);
 620
 621 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 622 {
 623         if (irqchip_in_kernel(vcpu->kvm))
 624                 return vcpu->apic_base;
 625         else
 626                 return vcpu->apic_base;
 627 }
 628 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 629
 630 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 631 {
 632         /* TODO: reserve bits check */
 633         if (irqchip_in_kernel(vcpu->kvm))
 634                 kvm_lapic_set_base(vcpu, data);
 635         else
 636                 vcpu->apic_base = data;
 637 }
 638 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 639
 640 void fx_init(struct kvm_vcpu *vcpu)
 641 {
 642         unsigned after_mxcsr_mask;
 643
 644         /* Initialize guest FPU by resetting ours and saving into guest's */
 645         preempt_disable();
 646         fx_save(&vcpu->host_fx_image);
 647         fpu_init();
 648         fx_save(&vcpu->guest_fx_image);
 649         fx_restore(&vcpu->host_fx_image);
 650         preempt_enable();
 651
 652         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
 653         vcpu->guest_fx_image.mxcsr = 0x1f80;
 654         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
 655                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
 656 }
 657 EXPORT_SYMBOL_GPL(fx_init);
 658
 659 /*
 660  * Allocate some memory and give it an address in the guest physical address
 661  * space.
 662  *
 663  * Discontiguous memory is allowed, mostly for framebuffers.
 664  */
 665 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 666                                           struct kvm_memory_region *mem)
 667 {
 668         int r;
 669         gfn_t base_gfn;
 670         unsigned long npages;
 671         unsigned long i;
 672         struct kvm_memory_slot *memslot;
 673         struct kvm_memory_slot old, new;
 674         int memory_config_version;
 675
 676         r = -EINVAL;
 677         /* General sanity checks */
 678         if (mem->memory_size & (PAGE_SIZE - 1))
 679                 goto out;
 680         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 681                 goto out;
 682         if (mem->slot >= KVM_MEMORY_SLOTS)
 683                 goto out;
 684         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 685                 goto out;
 686
 687         memslot = &kvm->memslots[mem->slot];
 688         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 689         npages = mem->memory_size >> PAGE_SHIFT;
 690
 691         if (!npages)
 692                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 693
 694 raced:
 695         mutex_lock(&kvm->lock);
 696
 697         memory_config_version = kvm->memory_config_version;
 698         new = old = *memslot;
 699
 700         new.base_gfn = base_gfn;
 701         new.npages = npages;
 702         new.flags = mem->flags;
 703
 704         /* Disallow changing a memory slot's size. */
 705         r = -EINVAL;
 706         if (npages && old.npages && npages != old.npages)
 707                 goto out_unlock;
 708
 709         /* Check for overlaps */
 710         r = -EEXIST;
 711         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 712                 struct kvm_memory_slot *s = &kvm->memslots[i];
 713
 714                 if (s == memslot)
 715                         continue;
 716                 if (!((base_gfn + npages <= s->base_gfn) ||
 717                       (base_gfn >= s->base_gfn + s->npages)))
 718                         goto out_unlock;
 719         }
 720         /*
 721          * Do memory allocations outside lock.  memory_config_version will
 722          * detect any races.
 723          */
 724         mutex_unlock(&kvm->lock);
 725
 726         /* Deallocate if slot is being removed */
 727         if (!npages)
 728                 new.phys_mem = NULL;
 729
 730         /* Free page dirty bitmap if unneeded */
 731         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 732                 new.dirty_bitmap = NULL;
 733
 734         r = -ENOMEM;
 735
 736         /* Allocate if a slot is being created */
 737         if (npages && !new.phys_mem) {
 738                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 739
 740                 if (!new.phys_mem)
 741                         goto out_free;
 742
 743                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 744                 for (i = 0; i < npages; ++i) {
 745                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 746                                                      | __GFP_ZERO);
 747                         if (!new.phys_mem[i])
 748                                 goto out_free;
 749                         set_page_private(new.phys_mem[i],0);
 750                 }
 751         }
 752
 753         /* Allocate page dirty bitmap if needed */
 754         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 755                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 756
 757                 new.dirty_bitmap = vmalloc(dirty_bytes);
 758                 if (!new.dirty_bitmap)
 759                         goto out_free;
 760                 memset(new.dirty_bitmap, 0, dirty_bytes);
 761         }
 762
 763         mutex_lock(&kvm->lock);
 764
 765         if (memory_config_version != kvm->memory_config_version) {
 766                 mutex_unlock(&kvm->lock);
 767                 kvm_free_physmem_slot(&new, &old);
 768                 goto raced;
 769         }
 770
 771         r = -EAGAIN;
 772         if (kvm->busy)
 773                 goto out_unlock;
 774
 775         if (mem->slot >= kvm->nmemslots)
 776                 kvm->nmemslots = mem->slot + 1;
 777
 778         *memslot = new;
 779         ++kvm->memory_config_version;
 780
 781         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 782         kvm_flush_remote_tlbs(kvm);
 783
 784         mutex_unlock(&kvm->lock);
 785
 786         kvm_free_physmem_slot(&old, &new);
 787         return 0;
 788
 789 out_unlock:
 790         mutex_unlock(&kvm->lock);
 791 out_free:
 792         kvm_free_physmem_slot(&new, &old);
 793 out:
 794         return r;
 795 }
 796
 797 /*
 798  * Get (and clear) the dirty memory log for a memory slot.
 799  */
 800 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 801                                       struct kvm_dirty_log *log)
 802 {
 803         struct kvm_memory_slot *memslot;
 804         int r, i;
 805         int n;
 806         unsigned long any = 0;
 807
 808         mutex_lock(&kvm->lock);
 809
 810         /*
 811          * Prevent changes to guest memory configuration even while the lock
 812          * is not taken.
 813          */
 814         ++kvm->busy;
 815         mutex_unlock(&kvm->lock);
 816         r = -EINVAL;
 817         if (log->slot >= KVM_MEMORY_SLOTS)
 818                 goto out;
 819
 820         memslot = &kvm->memslots[log->slot];
 821         r = -ENOENT;
 822         if (!memslot->dirty_bitmap)
 823                 goto out;
 824
 825         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 826
 827         for (i = 0; !any && i < n/sizeof(long); ++i)
 828                 any = memslot->dirty_bitmap[i];
 829
 830         r = -EFAULT;
 831         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 832                 goto out;
 833
 834         /* If nothing is dirty, don't bother messing with page tables. */
 835         if (any) {
 836                 mutex_lock(&kvm->lock);
 837                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 838                 kvm_flush_remote_tlbs(kvm);
 839                 memset(memslot->dirty_bitmap, 0, n);
 840                 mutex_unlock(&kvm->lock);
 841         }
 842
 843         r = 0;
 844
 845 out:
 846         mutex_lock(&kvm->lock);
 847         --kvm->busy;
 848         mutex_unlock(&kvm->lock);
 849         return r;
 850 }
 851
 852 /*
 853  * Set a new alias region.  Aliases map a portion of physical memory into
 854  * another portion.  This is useful for memory windows, for example the PC
 855  * VGA region.
 856  */
 857 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 858                                          struct kvm_memory_alias *alias)
 859 {
 860         int r, n;
 861         struct kvm_mem_alias *p;
 862
 863         r = -EINVAL;
 864         /* General sanity checks */
 865         if (alias->memory_size & (PAGE_SIZE - 1))
 866                 goto out;
 867         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 868                 goto out;
 869         if (alias->slot >= KVM_ALIAS_SLOTS)
 870                 goto out;
 871         if (alias->guest_phys_addr + alias->memory_size
 872             < alias->guest_phys_addr)
 873                 goto out;
 874         if (alias->target_phys_addr + alias->memory_size
 875             < alias->target_phys_addr)
 876                 goto out;
 877
 878         mutex_lock(&kvm->lock);
 879
 880         p = &kvm->aliases[alias->slot];
 881         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 882         p->npages = alias->memory_size >> PAGE_SHIFT;
 883         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 884
 885         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 886                 if (kvm->aliases[n - 1].npages)
 887                         break;
 888         kvm->naliases = n;
 889
 890         kvm_mmu_zap_all(kvm);
 891
 892         mutex_unlock(&kvm->lock);
 893
 894         return 0;
 895
 896 out:
 897         return r;
 898 }
 899
 900 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 901 {
 902         int i;
 903         struct kvm_mem_alias *alias;
 904
 905         for (i = 0; i < kvm->naliases; ++i) {
 906                 alias = &kvm->aliases[i];
 907                 if (gfn >= alias->base_gfn
 908                     && gfn < alias->base_gfn + alias->npages)
 909                         return alias->target_gfn + gfn - alias->base_gfn;
 910         }
 911         return gfn;
 912 }
 913
 914 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 915 {
 916         int i;
 917
 918         for (i = 0; i < kvm->nmemslots; ++i) {
 919                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 920
 921                 if (gfn >= memslot->base_gfn
 922                     && gfn < memslot->base_gfn + memslot->npages)
 923                         return memslot;
 924         }
 925         return NULL;
 926 }
 927
 928 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 929 {
 930         gfn = unalias_gfn(kvm, gfn);
 931         return __gfn_to_memslot(kvm, gfn);
 932 }
 933
 934 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 935 {
 936         struct kvm_memory_slot *slot;
 937
 938         gfn = unalias_gfn(kvm, gfn);
 939         slot = __gfn_to_memslot(kvm, gfn);
 940         if (!slot)
 941                 return NULL;
 942         return slot->phys_mem[gfn - slot->base_gfn];
 943 }
 944 EXPORT_SYMBOL_GPL(gfn_to_page);
 945
 946 /* WARNING: Does not work on aliased pages. */
 947 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 948 {
 949         struct kvm_memory_slot *memslot;
 950
 951         memslot = __gfn_to_memslot(kvm, gfn);
 952         if (memslot && memslot->dirty_bitmap) {
 953                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 954
 955                 /* avoid RMW */
 956                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 957                         set_bit(rel_gfn, memslot->dirty_bitmap);
 958         }
 959 }
 960
 961 int emulator_read_std(unsigned long addr,
 962                              void *val,
 963                              unsigned int bytes,
 964                              struct kvm_vcpu *vcpu)
 965 {
 966         void *data = val;
 967
 968         while (bytes) {
 969                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 970                 unsigned offset = addr & (PAGE_SIZE-1);
 971                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 972                 unsigned long pfn;
 973                 struct page *page;
 974                 void *page_virt;
 975
 976                 if (gpa == UNMAPPED_GVA)
 977                         return X86EMUL_PROPAGATE_FAULT;
 978                 pfn = gpa >> PAGE_SHIFT;
 979                 page = gfn_to_page(vcpu->kvm, pfn);
 980                 if (!page)
 981                         return X86EMUL_UNHANDLEABLE;
 982                 page_virt = kmap_atomic(page, KM_USER0);
 983
 984                 memcpy(data, page_virt + offset, tocopy);
 985
 986                 kunmap_atomic(page_virt, KM_USER0);
 987
 988                 bytes -= tocopy;
 989                 data += tocopy;
 990                 addr += tocopy;
 991         }
 992
 993         return X86EMUL_CONTINUE;
 994 }
 995 EXPORT_SYMBOL_GPL(emulator_read_std);
 996
 997 static int emulator_write_std(unsigned long addr,
 998                               const void *val,
 999                               unsigned int bytes,
1000                               struct kvm_vcpu *vcpu)
1001 {
1002         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1003         return X86EMUL_UNHANDLEABLE;
1004 }
1005
1006 /*
1007  * Only apic need an MMIO device hook, so shortcut now..
1008  */
1009 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1010                                                 gpa_t addr)
1011 {
1012         struct kvm_io_device *dev;
1013
1014         if (vcpu->apic) {
1015                 dev = &vcpu->apic->dev;
1016                 if (dev->in_range(dev, addr))
1017                         return dev;
1018         }
1019         return NULL;
1020 }
1021
1022 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1023                                                 gpa_t addr)
1024 {
1025         struct kvm_io_device *dev;
1026
1027         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1028         if (dev == NULL)
1029                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1030         return dev;
1031 }
1032
1033 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1034                                                gpa_t addr)
1035 {
1036         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1037 }
1038
1039 static int emulator_read_emulated(unsigned long addr,
1040                                   void *val,
1041                                   unsigned int bytes,
1042                                   struct kvm_vcpu *vcpu)
1043 {
1044         struct kvm_io_device *mmio_dev;
1045         gpa_t                 gpa;
1046
1047         if (vcpu->mmio_read_completed) {
1048                 memcpy(val, vcpu->mmio_data, bytes);
1049                 vcpu->mmio_read_completed = 0;
1050                 return X86EMUL_CONTINUE;
1051         } else if (emulator_read_std(addr, val, bytes, vcpu)
1052                    == X86EMUL_CONTINUE)
1053                 return X86EMUL_CONTINUE;
1054
1055         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1056         if (gpa == UNMAPPED_GVA)
1057                 return X86EMUL_PROPAGATE_FAULT;
1058
1059         /*
1060          * Is this MMIO handled locally?
1061          */
1062         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1063         if (mmio_dev) {
1064                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1065                 return X86EMUL_CONTINUE;
1066         }
1067
1068         vcpu->mmio_needed = 1;
1069         vcpu->mmio_phys_addr = gpa;
1070         vcpu->mmio_size = bytes;
1071         vcpu->mmio_is_write = 0;
1072
1073         return X86EMUL_UNHANDLEABLE;
1074 }
1075
1076 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1077                                const void *val, int bytes)
1078 {
1079         struct page *page;
1080         void *virt;
1081
1082         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1083                 return 0;
1084         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1085         if (!page)
1086                 return 0;
1087         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1088         virt = kmap_atomic(page, KM_USER0);
1089         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1090         memcpy(virt + offset_in_page(gpa), val, bytes);
1091         kunmap_atomic(virt, KM_USER0);
1092         return 1;
1093 }
1094
1095 static int emulator_write_emulated_onepage(unsigned long addr,
1096                                            const void *val,
1097                                            unsigned int bytes,
1098                                            struct kvm_vcpu *vcpu)
1099 {
1100         struct kvm_io_device *mmio_dev;
1101         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1102
1103         if (gpa == UNMAPPED_GVA) {
1104                 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1105                 return X86EMUL_PROPAGATE_FAULT;
1106         }
1107
1108         if (emulator_write_phys(vcpu, gpa, val, bytes))
1109                 return X86EMUL_CONTINUE;
1110
1111         /*
1112          * Is this MMIO handled locally?
1113          */
1114         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1115         if (mmio_dev) {
1116                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1117                 return X86EMUL_CONTINUE;
1118         }
1119
1120         vcpu->mmio_needed = 1;
1121         vcpu->mmio_phys_addr = gpa;
1122         vcpu->mmio_size = bytes;
1123         vcpu->mmio_is_write = 1;
1124         memcpy(vcpu->mmio_data, val, bytes);
1125
1126         return X86EMUL_CONTINUE;
1127 }
1128
1129 int emulator_write_emulated(unsigned long addr,
1130                                    const void *val,
1131                                    unsigned int bytes,
1132                                    struct kvm_vcpu *vcpu)
1133 {
1134         /* Crossing a page boundary? */
1135         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1136                 int rc, now;
1137
1138                 now = -addr & ~PAGE_MASK;
1139                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1140                 if (rc != X86EMUL_CONTINUE)
1141                         return rc;
1142                 addr += now;
1143                 val += now;
1144                 bytes -= now;
1145         }
1146         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1147 }
1148 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1149
1150 static int emulator_cmpxchg_emulated(unsigned long addr,
1151                                      const void *old,
1152                                      const void *new,
1153                                      unsigned int bytes,
1154                                      struct kvm_vcpu *vcpu)
1155 {
1156         static int reported;
1157
1158         if (!reported) {
1159                 reported = 1;
1160                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1161         }
1162         return emulator_write_emulated(addr, new, bytes, vcpu);
1163 }
1164
1165 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1166 {
1167         return kvm_arch_ops->get_segment_base(vcpu, seg);
1168 }
1169
1170 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1171 {
1172         return X86EMUL_CONTINUE;
1173 }
1174
1175 int emulate_clts(struct kvm_vcpu *vcpu)
1176 {
1177         unsigned long cr0;
1178
1179         cr0 = vcpu->cr0 & ~X86_CR0_TS;
1180         kvm_arch_ops->set_cr0(vcpu, cr0);
1181         return X86EMUL_CONTINUE;
1182 }
1183
1184 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1185 {
1186         struct kvm_vcpu *vcpu = ctxt->vcpu;
1187
1188         switch (dr) {
1189         case 0 ... 3:
1190                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1191                 return X86EMUL_CONTINUE;
1192         default:
1193                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1194                 return X86EMUL_UNHANDLEABLE;
1195         }
1196 }
1197
1198 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1199 {
1200         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1201         int exception;
1202
1203         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1204         if (exception) {
1205                 /* FIXME: better handling */
1206                 return X86EMUL_UNHANDLEABLE;
1207         }
1208         return X86EMUL_CONTINUE;
1209 }
1210
1211 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1212 {
1213         static int reported;
1214         u8 opcodes[4];
1215         unsigned long rip = ctxt->vcpu->rip;
1216         unsigned long rip_linear;
1217
1218         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1219
1220         if (reported)
1221                 return;
1222
1223         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt->vcpu);
1224
1225         printk(KERN_ERR "emulation failed but !mmio_needed?"
1226                " rip %lx %02x %02x %02x %02x\n",
1227                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1228         reported = 1;
1229 }
1230
1231 struct x86_emulate_ops emulate_ops = {
1232         .read_std            = emulator_read_std,
1233         .write_std           = emulator_write_std,
1234         .read_emulated       = emulator_read_emulated,
1235         .write_emulated      = emulator_write_emulated,
1236         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1237 };
1238
1239 int emulate_instruction(struct kvm_vcpu *vcpu,
1240                         struct kvm_run *run,
1241                         unsigned long cr2,
1242                         u16 error_code)
1243 {
1244         struct x86_emulate_ctxt emulate_ctxt;
1245         int r;
1246         int cs_db, cs_l;
1247
1248         vcpu->mmio_fault_cr2 = cr2;
1249         kvm_arch_ops->cache_regs(vcpu);
1250
1251         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1252
1253         emulate_ctxt.vcpu = vcpu;
1254         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1255         emulate_ctxt.cr2 = cr2;
1256         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1257                 ? X86EMUL_MODE_REAL : cs_l
1258                 ? X86EMUL_MODE_PROT64 : cs_db
1259                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1260
1261         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1262                 emulate_ctxt.cs_base = 0;
1263                 emulate_ctxt.ds_base = 0;
1264                 emulate_ctxt.es_base = 0;
1265                 emulate_ctxt.ss_base = 0;
1266         } else {
1267                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1268                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1269                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1270                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1271         }
1272
1273         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1274         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1275
1276         vcpu->mmio_is_write = 0;
1277         vcpu->pio.string = 0;
1278         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1279         if (vcpu->pio.string)
1280                 return EMULATE_DO_MMIO;
1281
1282         if ((r || vcpu->mmio_is_write) && run) {
1283                 run->exit_reason = KVM_EXIT_MMIO;
1284                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1285                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1286                 run->mmio.len = vcpu->mmio_size;
1287                 run->mmio.is_write = vcpu->mmio_is_write;
1288         }
1289
1290         if (r) {
1291                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1292                         return EMULATE_DONE;
1293                 if (!vcpu->mmio_needed) {
1294                         report_emulation_failure(&emulate_ctxt);
1295                         return EMULATE_FAIL;
1296                 }
1297                 return EMULATE_DO_MMIO;
1298         }
1299
1300         kvm_arch_ops->decache_regs(vcpu);
1301         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1302
1303         if (vcpu->mmio_is_write) {
1304                 vcpu->mmio_needed = 0;
1305                 return EMULATE_DO_MMIO;
1306         }
1307
1308         return EMULATE_DONE;
1309 }
1310 EXPORT_SYMBOL_GPL(emulate_instruction);
1311
1312 /*
1313  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1314  */
1315 static void kvm_vcpu_kernel_halt(struct kvm_vcpu *vcpu)
1316 {
1317         DECLARE_WAITQUEUE(wait, current);
1318
1319         add_wait_queue(&vcpu->wq, &wait);
1320
1321         /*
1322          * We will block until either an interrupt or a signal wakes us up
1323          */
1324         while(!(irqchip_in_kernel(vcpu->kvm) && kvm_cpu_has_interrupt(vcpu))
1325               && !vcpu->irq_summary
1326               && !signal_pending(current)) {
1327                 set_current_state(TASK_INTERRUPTIBLE);
1328                 vcpu_put(vcpu);
1329                 schedule();
1330                 vcpu_load(vcpu);
1331         }
1332
1333         remove_wait_queue(&vcpu->wq, &wait);
1334         set_current_state(TASK_RUNNING);
1335 }
1336
1337 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1338 {
1339         ++vcpu->stat.halt_exits;
1340         if (irqchip_in_kernel(vcpu->kvm)) {
1341                 kvm_vcpu_kernel_halt(vcpu);
1342                 return 1;
1343         } else {
1344                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1345                 return 0;
1346         }
1347 }
1348 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1349
1350 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1351 {
1352         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1353
1354         kvm_arch_ops->cache_regs(vcpu);
1355         ret = -KVM_EINVAL;
1356 #ifdef CONFIG_X86_64
1357         if (is_long_mode(vcpu)) {
1358                 nr = vcpu->regs[VCPU_REGS_RAX];
1359                 a0 = vcpu->regs[VCPU_REGS_RDI];
1360                 a1 = vcpu->regs[VCPU_REGS_RSI];
1361                 a2 = vcpu->regs[VCPU_REGS_RDX];
1362                 a3 = vcpu->regs[VCPU_REGS_RCX];
1363                 a4 = vcpu->regs[VCPU_REGS_R8];
1364                 a5 = vcpu->regs[VCPU_REGS_R9];
1365         } else
1366 #endif
1367         {
1368                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1369                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1370                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1371                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1372                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1373                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1374                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1375         }
1376         switch (nr) {
1377         default:
1378                 run->hypercall.nr = nr;
1379                 run->hypercall.args[0] = a0;
1380                 run->hypercall.args[1] = a1;
1381                 run->hypercall.args[2] = a2;
1382                 run->hypercall.args[3] = a3;
1383                 run->hypercall.args[4] = a4;
1384                 run->hypercall.args[5] = a5;
1385                 run->hypercall.ret = ret;
1386                 run->hypercall.longmode = is_long_mode(vcpu);
1387                 kvm_arch_ops->decache_regs(vcpu);
1388                 return 0;
1389         }
1390         vcpu->regs[VCPU_REGS_RAX] = ret;
1391         kvm_arch_ops->decache_regs(vcpu);
1392         return 1;
1393 }
1394 EXPORT_SYMBOL_GPL(kvm_hypercall);
1395
1396 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1397 {
1398         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1399 }
1400
1401 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1402 {
1403         struct descriptor_table dt = { limit, base };
1404
1405         kvm_arch_ops->set_gdt(vcpu, &dt);
1406 }
1407
1408 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1409 {
1410         struct descriptor_table dt = { limit, base };
1411
1412         kvm_arch_ops->set_idt(vcpu, &dt);
1413 }
1414
1415 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1416                    unsigned long *rflags)
1417 {
1418         lmsw(vcpu, msw);
1419         *rflags = kvm_arch_ops->get_rflags(vcpu);
1420 }
1421
1422 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1423 {
1424         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1425         switch (cr) {
1426         case 0:
1427                 return vcpu->cr0;
1428         case 2:
1429                 return vcpu->cr2;
1430         case 3:
1431                 return vcpu->cr3;
1432         case 4:
1433                 return vcpu->cr4;
1434         default:
1435                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1436                 return 0;
1437         }
1438 }
1439
1440 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1441                      unsigned long *rflags)
1442 {
1443         switch (cr) {
1444         case 0:
1445                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1446                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1447                 break;
1448         case 2:
1449                 vcpu->cr2 = val;
1450                 break;
1451         case 3:
1452                 set_cr3(vcpu, val);
1453                 break;
1454         case 4:
1455                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1456                 break;
1457         default:
1458                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1459         }
1460 }
1461
1462 /*
1463  * Register the para guest with the host:
1464  */
1465 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1466 {
1467         struct kvm_vcpu_para_state *para_state;
1468         hpa_t para_state_hpa, hypercall_hpa;
1469         struct page *para_state_page;
1470         unsigned char *hypercall;
1471         gpa_t hypercall_gpa;
1472
1473         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1474         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1475
1476         /*
1477          * Needs to be page aligned:
1478          */
1479         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1480                 goto err_gp;
1481
1482         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1483         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1484         if (is_error_hpa(para_state_hpa))
1485                 goto err_gp;
1486
1487         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1488         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1489         para_state = kmap(para_state_page);
1490
1491         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1492         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1493
1494         para_state->host_version = KVM_PARA_API_VERSION;
1495         /*
1496          * We cannot support guests that try to register themselves
1497          * with a newer API version than the host supports:
1498          */
1499         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1500                 para_state->ret = -KVM_EINVAL;
1501                 goto err_kunmap_skip;
1502         }
1503
1504         hypercall_gpa = para_state->hypercall_gpa;
1505         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1506         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1507         if (is_error_hpa(hypercall_hpa)) {
1508                 para_state->ret = -KVM_EINVAL;
1509                 goto err_kunmap_skip;
1510         }
1511
1512         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1513         vcpu->para_state_page = para_state_page;
1514         vcpu->para_state_gpa = para_state_gpa;
1515         vcpu->hypercall_gpa = hypercall_gpa;
1516
1517         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1518         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1519                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1520         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1521         kunmap_atomic(hypercall, KM_USER1);
1522
1523         para_state->ret = 0;
1524 err_kunmap_skip:
1525         kunmap(para_state_page);
1526         return 0;
1527 err_gp:
1528         return 1;
1529 }
1530
1531 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1532 {
1533         u64 data;
1534
1535         switch (msr) {
1536         case 0xc0010010: /* SYSCFG */
1537         case 0xc0010015: /* HWCR */
1538         case MSR_IA32_PLATFORM_ID:
1539         case MSR_IA32_P5_MC_ADDR:
1540         case MSR_IA32_P5_MC_TYPE:
1541         case MSR_IA32_MC0_CTL:
1542         case MSR_IA32_MCG_STATUS:
1543         case MSR_IA32_MCG_CAP:
1544         case MSR_IA32_MC0_MISC:
1545         case MSR_IA32_MC0_MISC+4:
1546         case MSR_IA32_MC0_MISC+8:
1547         case MSR_IA32_MC0_MISC+12:
1548         case MSR_IA32_MC0_MISC+16:
1549         case MSR_IA32_UCODE_REV:
1550         case MSR_IA32_PERF_STATUS:
1551         case MSR_IA32_EBL_CR_POWERON:
1552                 /* MTRR registers */
1553         case 0xfe:
1554         case 0x200 ... 0x2ff:
1555                 data = 0;
1556                 break;
1557         case 0xcd: /* fsb frequency */
1558                 data = 3;
1559                 break;
1560         case MSR_IA32_APICBASE:
1561                 data = kvm_get_apic_base(vcpu);
1562                 break;
1563         case MSR_IA32_MISC_ENABLE:
1564                 data = vcpu->ia32_misc_enable_msr;
1565                 break;
1566 #ifdef CONFIG_X86_64
1567         case MSR_EFER:
1568                 data = vcpu->shadow_efer;
1569                 break;
1570 #endif
1571         default:
1572                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1573                 return 1;
1574         }
1575         *pdata = data;
1576         return 0;
1577 }
1578 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1579
1580 /*
1581  * Reads an msr value (of 'msr_index') into 'pdata'.
1582  * Returns 0 on success, non-0 otherwise.
1583  * Assumes vcpu_load() was already called.
1584  */
1585 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1586 {
1587         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1588 }
1589
1590 #ifdef CONFIG_X86_64
1591
1592 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1593 {
1594         if (efer & EFER_RESERVED_BITS) {
1595                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1596                        efer);
1597                 inject_gp(vcpu);
1598                 return;
1599         }
1600
1601         if (is_paging(vcpu)
1602             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1603                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1604                 inject_gp(vcpu);
1605                 return;
1606         }
1607
1608         kvm_arch_ops->set_efer(vcpu, efer);
1609
1610         efer &= ~EFER_LMA;
1611         efer |= vcpu->shadow_efer & EFER_LMA;
1612
1613         vcpu->shadow_efer = efer;
1614 }
1615
1616 #endif
1617
1618 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1619 {
1620         switch (msr) {
1621 #ifdef CONFIG_X86_64
1622         case MSR_EFER:
1623                 set_efer(vcpu, data);
1624                 break;
1625 #endif
1626         case MSR_IA32_MC0_STATUS:
1627                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1628                        __FUNCTION__, data);
1629                 break;
1630         case MSR_IA32_MCG_STATUS:
1631                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1632                         __FUNCTION__, data);
1633                 break;
1634         case MSR_IA32_UCODE_REV:
1635         case MSR_IA32_UCODE_WRITE:
1636         case 0x200 ... 0x2ff: /* MTRRs */
1637                 break;
1638         case MSR_IA32_APICBASE:
1639                 kvm_set_apic_base(vcpu, data);
1640                 break;
1641         case MSR_IA32_MISC_ENABLE:
1642                 vcpu->ia32_misc_enable_msr = data;
1643                 break;
1644         /*
1645          * This is the 'probe whether the host is KVM' logic:
1646          */
1647         case MSR_KVM_API_MAGIC:
1648                 return vcpu_register_para(vcpu, data);
1649
1650         default:
1651                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1652                 return 1;
1653         }
1654         return 0;
1655 }
1656 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1657
1658 /*
1659  * Writes msr value into into the appropriate "register".
1660  * Returns 0 on success, non-0 otherwise.
1661  * Assumes vcpu_load() was already called.
1662  */
1663 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1664 {
1665         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1666 }
1667
1668 void kvm_resched(struct kvm_vcpu *vcpu)
1669 {
1670         if (!need_resched())
1671                 return;
1672         cond_resched();
1673 }
1674 EXPORT_SYMBOL_GPL(kvm_resched);
1675
1676 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1677 {
1678         int i;
1679         u32 function;
1680         struct kvm_cpuid_entry *e, *best;
1681
1682         kvm_arch_ops->cache_regs(vcpu);
1683         function = vcpu->regs[VCPU_REGS_RAX];
1684         vcpu->regs[VCPU_REGS_RAX] = 0;
1685         vcpu->regs[VCPU_REGS_RBX] = 0;
1686         vcpu->regs[VCPU_REGS_RCX] = 0;
1687         vcpu->regs[VCPU_REGS_RDX] = 0;
1688         best = NULL;
1689         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1690                 e = &vcpu->cpuid_entries[i];
1691                 if (e->function == function) {
1692                         best = e;
1693                         break;
1694                 }
1695                 /*
1696                  * Both basic or both extended?
1697                  */
1698                 if (((e->function ^ function) & 0x80000000) == 0)
1699                         if (!best || e->function > best->function)
1700                                 best = e;
1701         }
1702         if (best) {
1703                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1704                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1705                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1706                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1707         }
1708         kvm_arch_ops->decache_regs(vcpu);
1709         kvm_arch_ops->skip_emulated_instruction(vcpu);
1710 }
1711 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1712
1713 static int pio_copy_data(struct kvm_vcpu *vcpu)
1714 {
1715         void *p = vcpu->pio_data;
1716         void *q;
1717         unsigned bytes;
1718         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1719
1720         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1721                  PAGE_KERNEL);
1722         if (!q) {
1723                 free_pio_guest_pages(vcpu);
1724                 return -ENOMEM;
1725         }
1726         q += vcpu->pio.guest_page_offset;
1727         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1728         if (vcpu->pio.in)
1729                 memcpy(q, p, bytes);
1730         else
1731                 memcpy(p, q, bytes);
1732         q -= vcpu->pio.guest_page_offset;
1733         vunmap(q);
1734         free_pio_guest_pages(vcpu);
1735         return 0;
1736 }
1737
1738 static int complete_pio(struct kvm_vcpu *vcpu)
1739 {
1740         struct kvm_pio_request *io = &vcpu->pio;
1741         long delta;
1742         int r;
1743
1744         kvm_arch_ops->cache_regs(vcpu);
1745
1746         if (!io->string) {
1747                 if (io->in)
1748                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1749                                io->size);
1750         } else {
1751                 if (io->in) {
1752                         r = pio_copy_data(vcpu);
1753                         if (r) {
1754                                 kvm_arch_ops->cache_regs(vcpu);
1755                                 return r;
1756                         }
1757                 }
1758
1759                 delta = 1;
1760                 if (io->rep) {
1761                         delta *= io->cur_count;
1762                         /*
1763                          * The size of the register should really depend on
1764                          * current address size.
1765                          */
1766                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1767                 }
1768                 if (io->down)
1769                         delta = -delta;
1770                 delta *= io->size;
1771                 if (io->in)
1772                         vcpu->regs[VCPU_REGS_RDI] += delta;
1773                 else
1774                         vcpu->regs[VCPU_REGS_RSI] += delta;
1775         }
1776
1777         kvm_arch_ops->decache_regs(vcpu);
1778
1779         io->count -= io->cur_count;
1780         io->cur_count = 0;
1781
1782         if (!io->count)
1783                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1784         return 0;
1785 }
1786
1787 static void kernel_pio(struct kvm_io_device *pio_dev,
1788                        struct kvm_vcpu *vcpu,
1789                        void *pd)
1790 {
1791         /* TODO: String I/O for in kernel device */
1792
1793         if (vcpu->pio.in)
1794                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1795                                   vcpu->pio.size,
1796                                   pd);
1797         else
1798                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1799                                    vcpu->pio.size,
1800                                    pd);
1801 }
1802
1803 static void pio_string_write(struct kvm_io_device *pio_dev,
1804                              struct kvm_vcpu *vcpu)
1805 {
1806         struct kvm_pio_request *io = &vcpu->pio;
1807         void *pd = vcpu->pio_data;
1808         int i;
1809
1810         for (i = 0; i < io->cur_count; i++) {
1811                 kvm_iodevice_write(pio_dev, io->port,
1812                                    io->size,
1813                                    pd);
1814                 pd += io->size;
1815         }
1816 }
1817
1818 int kvm_emulate_pio (struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1819                   int size, unsigned port)
1820 {
1821         struct kvm_io_device *pio_dev;
1822
1823         vcpu->run->exit_reason = KVM_EXIT_IO;
1824         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1825         vcpu->run->io.size = vcpu->pio.size = size;
1826         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1827         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1828         vcpu->run->io.port = vcpu->pio.port = port;
1829         vcpu->pio.in = in;
1830         vcpu->pio.string = 0;
1831         vcpu->pio.down = 0;
1832         vcpu->pio.guest_page_offset = 0;
1833         vcpu->pio.rep = 0;
1834
1835         kvm_arch_ops->cache_regs(vcpu);
1836         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1837         kvm_arch_ops->decache_regs(vcpu);
1838
1839         pio_dev = vcpu_find_pio_dev(vcpu, port);
1840         if (pio_dev) {
1841                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1842                 complete_pio(vcpu);
1843                 return 1;
1844         }
1845         return 0;
1846 }
1847 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1848
1849 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1850                   int size, unsigned long count, int down,
1851                   gva_t address, int rep, unsigned port)
1852 {
1853         unsigned now, in_page;
1854         int i, ret = 0;
1855         int nr_pages = 1;
1856         struct page *page;
1857         struct kvm_io_device *pio_dev;
1858
1859         vcpu->run->exit_reason = KVM_EXIT_IO;
1860         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1861         vcpu->run->io.size = vcpu->pio.size = size;
1862         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1863         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1864         vcpu->run->io.port = vcpu->pio.port = port;
1865         vcpu->pio.in = in;
1866         vcpu->pio.string = 1;
1867         vcpu->pio.down = down;
1868         vcpu->pio.guest_page_offset = offset_in_page(address);
1869         vcpu->pio.rep = rep;
1870
1871         if (!count) {
1872                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1873                 return 1;
1874         }
1875
1876         if (!down)
1877                 in_page = PAGE_SIZE - offset_in_page(address);
1878         else
1879                 in_page = offset_in_page(address) + size;
1880         now = min(count, (unsigned long)in_page / size);
1881         if (!now) {
1882                 /*
1883                  * String I/O straddles page boundary.  Pin two guest pages
1884                  * so that we satisfy atomicity constraints.  Do just one
1885                  * transaction to avoid complexity.
1886                  */
1887                 nr_pages = 2;
1888                 now = 1;
1889         }
1890         if (down) {
1891                 /*
1892                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1893                  */
1894                 pr_unimpl(vcpu, "guest string pio down\n");
1895                 inject_gp(vcpu);
1896                 return 1;
1897         }
1898         vcpu->run->io.count = now;
1899         vcpu->pio.cur_count = now;
1900
1901         for (i = 0; i < nr_pages; ++i) {
1902                 mutex_lock(&vcpu->kvm->lock);
1903                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1904                 if (page)
1905                         get_page(page);
1906                 vcpu->pio.guest_pages[i] = page;
1907                 mutex_unlock(&vcpu->kvm->lock);
1908                 if (!page) {
1909                         inject_gp(vcpu);
1910                         free_pio_guest_pages(vcpu);
1911                         return 1;
1912                 }
1913         }
1914
1915         pio_dev = vcpu_find_pio_dev(vcpu, port);
1916         if (!vcpu->pio.in) {
1917                 /* string PIO write */
1918                 ret = pio_copy_data(vcpu);
1919                 if (ret >= 0 && pio_dev) {
1920                         pio_string_write(pio_dev, vcpu);
1921                         complete_pio(vcpu);
1922                         if (vcpu->pio.count == 0)
1923                                 ret = 1;
1924                 }
1925         } else if (pio_dev)
1926                 pr_unimpl(vcpu, "no string pio read support yet, "
1927                        "port %x size %d count %ld\n",
1928                         port, size, count);
1929
1930         return ret;
1931 }
1932 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1933
1934 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1935 {
1936         int r;
1937         sigset_t sigsaved;
1938
1939         vcpu_load(vcpu);
1940
1941         if (vcpu->sigset_active)
1942                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1943
1944         /* re-sync apic's tpr */
1945         set_cr8(vcpu, kvm_run->cr8);
1946
1947         if (vcpu->pio.cur_count) {
1948                 r = complete_pio(vcpu);
1949                 if (r)
1950                         goto out;
1951         }
1952
1953         if (vcpu->mmio_needed) {
1954                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1955                 vcpu->mmio_read_completed = 1;
1956                 vcpu->mmio_needed = 0;
1957                 r = emulate_instruction(vcpu, kvm_run,
1958                                         vcpu->mmio_fault_cr2, 0);
1959                 if (r == EMULATE_DO_MMIO) {
1960                         /*
1961                          * Read-modify-write.  Back to userspace.
1962                          */
1963                         r = 0;
1964                         goto out;
1965                 }
1966         }
1967
1968         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1969                 kvm_arch_ops->cache_regs(vcpu);
1970                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1971                 kvm_arch_ops->decache_regs(vcpu);
1972         }
1973
1974         r = kvm_arch_ops->run(vcpu, kvm_run);
1975
1976 out:
1977         if (vcpu->sigset_active)
1978                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1979
1980         vcpu_put(vcpu);
1981         return r;
1982 }
1983
1984 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1985                                    struct kvm_regs *regs)
1986 {
1987         vcpu_load(vcpu);
1988
1989         kvm_arch_ops->cache_regs(vcpu);
1990
1991         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1992         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1993         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1994         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1995         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1996         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1997         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1998         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1999 #ifdef CONFIG_X86_64
2000         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2001         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2002         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2003         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2004         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2005         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2006         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2007         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2008 #endif
2009
2010         regs->rip = vcpu->rip;
2011         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
2012
2013         /*
2014          * Don't leak debug flags in case they were set for guest debugging
2015          */
2016         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2017                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2018
2019         vcpu_put(vcpu);
2020
2021         return 0;
2022 }
2023
2024 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2025                                    struct kvm_regs *regs)
2026 {
2027         vcpu_load(vcpu);
2028
2029         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2030         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2031         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2032         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2033         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2034         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2035         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2036         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2037 #ifdef CONFIG_X86_64
2038         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2039         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2040         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2041         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2042         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2043         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2044         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2045         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2046 #endif
2047
2048         vcpu->rip = regs->rip;
2049         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
2050
2051         kvm_arch_ops->decache_regs(vcpu);
2052
2053         vcpu_put(vcpu);
2054
2055         return 0;
2056 }
2057
2058 static void get_segment(struct kvm_vcpu *vcpu,
2059                         struct kvm_segment *var, int seg)
2060 {
2061         return kvm_arch_ops->get_segment(vcpu, var, seg);
2062 }
2063
2064 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2065                                     struct kvm_sregs *sregs)
2066 {
2067         struct descriptor_table dt;
2068
2069         vcpu_load(vcpu);
2070
2071         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2072         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2073         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2074         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2075         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2076         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2077
2078         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2079         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2080
2081         kvm_arch_ops->get_idt(vcpu, &dt);
2082         sregs->idt.limit = dt.limit;
2083         sregs->idt.base = dt.base;
2084         kvm_arch_ops->get_gdt(vcpu, &dt);
2085         sregs->gdt.limit = dt.limit;
2086         sregs->gdt.base = dt.base;
2087
2088         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2089         sregs->cr0 = vcpu->cr0;
2090         sregs->cr2 = vcpu->cr2;
2091         sregs->cr3 = vcpu->cr3;
2092         sregs->cr4 = vcpu->cr4;
2093         sregs->cr8 = get_cr8(vcpu);
2094         sregs->efer = vcpu->shadow_efer;
2095         sregs->apic_base = kvm_get_apic_base(vcpu);
2096
2097         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2098                sizeof sregs->interrupt_bitmap);
2099
2100         vcpu_put(vcpu);
2101
2102         return 0;
2103 }
2104
2105 static void set_segment(struct kvm_vcpu *vcpu,
2106                         struct kvm_segment *var, int seg)
2107 {
2108         return kvm_arch_ops->set_segment(vcpu, var, seg);
2109 }
2110
2111 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2112                                     struct kvm_sregs *sregs)
2113 {
2114         int mmu_reset_needed = 0;
2115         int i;
2116         struct descriptor_table dt;
2117
2118         vcpu_load(vcpu);
2119
2120         dt.limit = sregs->idt.limit;
2121         dt.base = sregs->idt.base;
2122         kvm_arch_ops->set_idt(vcpu, &dt);
2123         dt.limit = sregs->gdt.limit;
2124         dt.base = sregs->gdt.base;
2125         kvm_arch_ops->set_gdt(vcpu, &dt);
2126
2127         vcpu->cr2 = sregs->cr2;
2128         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2129         vcpu->cr3 = sregs->cr3;
2130
2131         set_cr8(vcpu, sregs->cr8);
2132
2133         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2134 #ifdef CONFIG_X86_64
2135         kvm_arch_ops->set_efer(vcpu, sregs->efer);
2136 #endif
2137         kvm_set_apic_base(vcpu, sregs->apic_base);
2138
2139         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2140
2141         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2142         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2143
2144         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2145         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2146         if (!is_long_mode(vcpu) && is_pae(vcpu))
2147                 load_pdptrs(vcpu, vcpu->cr3);
2148
2149         if (mmu_reset_needed)
2150                 kvm_mmu_reset_context(vcpu);
2151
2152         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2153                sizeof vcpu->irq_pending);
2154         vcpu->irq_summary = 0;
2155         for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2156                 if (vcpu->irq_pending[i])
2157                         __set_bit(i, &vcpu->irq_summary);
2158
2159         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2160         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2161         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2162         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2163         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2164         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2165
2166         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2167         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2168
2169         vcpu_put(vcpu);
2170
2171         return 0;
2172 }
2173
2174 /*
2175  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2176  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2177  *
2178  * This list is modified at module load time to reflect the
2179  * capabilities of the host cpu.
2180  */
2181 static u32 msrs_to_save[] = {
2182         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2183         MSR_K6_STAR,
2184 #ifdef CONFIG_X86_64
2185         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2186 #endif
2187         MSR_IA32_TIME_STAMP_COUNTER,
2188 };
2189
2190 static unsigned num_msrs_to_save;
2191
2192 static u32 emulated_msrs[] = {
2193         MSR_IA32_MISC_ENABLE,
2194 };
2195
2196 static __init void kvm_init_msr_list(void)
2197 {
2198         u32 dummy[2];
2199         unsigned i, j;
2200
2201         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2202                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2203                         continue;
2204                 if (j < i)
2205                         msrs_to_save[j] = msrs_to_save[i];
2206                 j++;
2207         }
2208         num_msrs_to_save = j;
2209 }
2210
2211 /*
2212  * Adapt set_msr() to msr_io()'s calling convention
2213  */
2214 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2215 {
2216         return kvm_set_msr(vcpu, index, *data);
2217 }
2218
2219 /*
2220  * Read or write a bunch of msrs. All parameters are kernel addresses.
2221  *
2222  * @return number of msrs set successfully.
2223  */
2224 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2225                     struct kvm_msr_entry *entries,
2226                     int (*do_msr)(struct kvm_vcpu *vcpu,
2227                                   unsigned index, u64 *data))
2228 {
2229         int i;
2230
2231         vcpu_load(vcpu);
2232
2233         for (i = 0; i < msrs->nmsrs; ++i)
2234                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2235                         break;
2236
2237         vcpu_put(vcpu);
2238
2239         return i;
2240 }
2241
2242 /*
2243  * Read or write a bunch of msrs. Parameters are user addresses.
2244  *
2245  * @return number of msrs set successfully.
2246  */
2247 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2248                   int (*do_msr)(struct kvm_vcpu *vcpu,
2249                                 unsigned index, u64 *data),
2250                   int writeback)
2251 {
2252         struct kvm_msrs msrs;
2253         struct kvm_msr_entry *entries;
2254         int r, n;
2255         unsigned size;
2256
2257         r = -EFAULT;
2258         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2259                 goto out;
2260
2261         r = -E2BIG;
2262         if (msrs.nmsrs >= MAX_IO_MSRS)
2263                 goto out;
2264
2265         r = -ENOMEM;
2266         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2267         entries = vmalloc(size);
2268         if (!entries)
2269                 goto out;
2270
2271         r = -EFAULT;
2272         if (copy_from_user(entries, user_msrs->entries, size))
2273                 goto out_free;
2274
2275         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2276         if (r < 0)
2277                 goto out_free;
2278
2279         r = -EFAULT;
2280         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2281                 goto out_free;
2282
2283         r = n;
2284
2285 out_free:
2286         vfree(entries);
2287 out:
2288         return r;
2289 }
2290
2291 /*
2292  * Translate a guest virtual address to a guest physical address.
2293  */
2294 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2295                                     struct kvm_translation *tr)
2296 {
2297         unsigned long vaddr = tr->linear_address;
2298         gpa_t gpa;
2299
2300         vcpu_load(vcpu);
2301         mutex_lock(&vcpu->kvm->lock);
2302         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2303         tr->physical_address = gpa;
2304         tr->valid = gpa != UNMAPPED_GVA;
2305         tr->writeable = 1;
2306         tr->usermode = 0;
2307         mutex_unlock(&vcpu->kvm->lock);
2308         vcpu_put(vcpu);
2309
2310         return 0;
2311 }
2312
2313 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2314                                     struct kvm_interrupt *irq)
2315 {
2316         if (irq->irq < 0 || irq->irq >= 256)
2317                 return -EINVAL;
2318         if (irqchip_in_kernel(vcpu->kvm))
2319                 return -ENXIO;
2320         vcpu_load(vcpu);
2321
2322         set_bit(irq->irq, vcpu->irq_pending);
2323         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2324
2325         vcpu_put(vcpu);
2326
2327         return 0;
2328 }
2329
2330 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2331                                       struct kvm_debug_guest *dbg)
2332 {
2333         int r;
2334
2335         vcpu_load(vcpu);
2336
2337         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2338
2339         vcpu_put(vcpu);
2340
2341         return r;
2342 }
2343
2344 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2345                                     unsigned long address,
2346                                     int *type)
2347 {
2348         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2349         unsigned long pgoff;
2350         struct page *page;
2351
2352         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2353         if (pgoff == 0)
2354                 page = virt_to_page(vcpu->run);
2355         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2356                 page = virt_to_page(vcpu->pio_data);
2357         else
2358                 return NOPAGE_SIGBUS;
2359         get_page(page);
2360         if (type != NULL)
2361                 *type = VM_FAULT_MINOR;
2362
2363         return page;
2364 }
2365
2366 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2367         .nopage = kvm_vcpu_nopage,
2368 };
2369
2370 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2371 {
2372         vma->vm_ops = &kvm_vcpu_vm_ops;
2373         return 0;
2374 }
2375
2376 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2377 {
2378         struct kvm_vcpu *vcpu = filp->private_data;
2379
2380         fput(vcpu->kvm->filp);
2381         return 0;
2382 }
2383
2384 static struct file_operations kvm_vcpu_fops = {
2385         .release        = kvm_vcpu_release,
2386         .unlocked_ioctl = kvm_vcpu_ioctl,
2387         .compat_ioctl   = kvm_vcpu_ioctl,
2388         .mmap           = kvm_vcpu_mmap,
2389 };
2390
2391 /*
2392  * Allocates an inode for the vcpu.
2393  */
2394 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2395 {
2396         int fd, r;
2397         struct inode *inode;
2398         struct file *file;
2399
2400         r = anon_inode_getfd(&fd, &inode, &file,
2401                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2402         if (r)
2403                 return r;
2404         atomic_inc(&vcpu->kvm->filp->f_count);
2405         return fd;
2406 }
2407
2408 /*
2409  * Creates some virtual cpus.  Good luck creating more than one.
2410  */
2411 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2412 {
2413         int r;
2414         struct kvm_vcpu *vcpu;
2415
2416         if (!valid_vcpu(n))
2417                 return -EINVAL;
2418
2419         vcpu = kvm_arch_ops->vcpu_create(kvm, n);
2420         if (IS_ERR(vcpu))
2421                 return PTR_ERR(vcpu);
2422
2423         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2424
2425         /* We do fxsave: this must be aligned. */
2426         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2427
2428         vcpu_load(vcpu);
2429         r = kvm_mmu_setup(vcpu);
2430         vcpu_put(vcpu);
2431         if (r < 0)
2432                 goto free_vcpu;
2433
2434         mutex_lock(&kvm->lock);
2435         if (kvm->vcpus[n]) {
2436                 r = -EEXIST;
2437                 mutex_unlock(&kvm->lock);
2438                 goto mmu_unload;
2439         }
2440         kvm->vcpus[n] = vcpu;
2441         mutex_unlock(&kvm->lock);
2442
2443         /* Now it's all set up, let userspace reach it */
2444         r = create_vcpu_fd(vcpu);
2445         if (r < 0)
2446                 goto unlink;
2447         return r;
2448
2449 unlink:
2450         mutex_lock(&kvm->lock);
2451         kvm->vcpus[n] = NULL;
2452         mutex_unlock(&kvm->lock);
2453
2454 mmu_unload:
2455         vcpu_load(vcpu);
2456         kvm_mmu_unload(vcpu);
2457         vcpu_put(vcpu);
2458
2459 free_vcpu:
2460         kvm_arch_ops->vcpu_free(vcpu);
2461         return r;
2462 }
2463
2464 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2465 {
2466         u64 efer;
2467         int i;
2468         struct kvm_cpuid_entry *e, *entry;
2469
2470         rdmsrl(MSR_EFER, efer);
2471         entry = NULL;
2472         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2473                 e = &vcpu->cpuid_entries[i];
2474                 if (e->function == 0x80000001) {
2475                         entry = e;
2476                         break;
2477                 }
2478         }
2479         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2480                 entry->edx &= ~(1 << 20);
2481                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2482         }
2483 }
2484
2485 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2486                                     struct kvm_cpuid *cpuid,
2487                                     struct kvm_cpuid_entry __user *entries)
2488 {
2489         int r;
2490
2491         r = -E2BIG;
2492         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2493                 goto out;
2494         r = -EFAULT;
2495         if (copy_from_user(&vcpu->cpuid_entries, entries,
2496                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2497                 goto out;
2498         vcpu->cpuid_nent = cpuid->nent;
2499         cpuid_fix_nx_cap(vcpu);
2500         return 0;
2501
2502 out:
2503         return r;
2504 }
2505
2506 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2507 {
2508         if (sigset) {
2509                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2510                 vcpu->sigset_active = 1;
2511                 vcpu->sigset = *sigset;
2512         } else
2513                 vcpu->sigset_active = 0;
2514         return 0;
2515 }
2516
2517 /*
2518  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2519  * we have asm/x86/processor.h
2520  */
2521 struct fxsave {
2522         u16     cwd;
2523         u16     swd;
2524         u16     twd;
2525         u16     fop;
2526         u64     rip;
2527         u64     rdp;
2528         u32     mxcsr;
2529         u32     mxcsr_mask;
2530         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2531 #ifdef CONFIG_X86_64
2532         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2533 #else
2534         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2535 #endif
2536 };
2537
2538 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2539 {
2540         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2541
2542         vcpu_load(vcpu);
2543
2544         memcpy(fpu->fpr, fxsave->st_space, 128);
2545         fpu->fcw = fxsave->cwd;
2546         fpu->fsw = fxsave->swd;
2547         fpu->ftwx = fxsave->twd;
2548         fpu->last_opcode = fxsave->fop;
2549         fpu->last_ip = fxsave->rip;
2550         fpu->last_dp = fxsave->rdp;
2551         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2552
2553         vcpu_put(vcpu);
2554
2555         return 0;
2556 }
2557
2558 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2559 {
2560         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2561
2562         vcpu_load(vcpu);
2563
2564         memcpy(fxsave->st_space, fpu->fpr, 128);
2565         fxsave->cwd = fpu->fcw;
2566         fxsave->swd = fpu->fsw;
2567         fxsave->twd = fpu->ftwx;
2568         fxsave->fop = fpu->last_opcode;
2569         fxsave->rip = fpu->last_ip;
2570         fxsave->rdp = fpu->last_dp;
2571         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2572
2573         vcpu_put(vcpu);
2574
2575         return 0;
2576 }
2577
2578 static long kvm_vcpu_ioctl(struct file *filp,
2579                            unsigned int ioctl, unsigned long arg)
2580 {
2581         struct kvm_vcpu *vcpu = filp->private_data;
2582         void __user *argp = (void __user *)arg;
2583         int r = -EINVAL;
2584
2585         switch (ioctl) {
2586         case KVM_RUN:
2587                 r = -EINVAL;
2588                 if (arg)
2589                         goto out;
2590                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2591                 break;
2592         case KVM_GET_REGS: {
2593                 struct kvm_regs kvm_regs;
2594
2595                 memset(&kvm_regs, 0, sizeof kvm_regs);
2596                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2597                 if (r)
2598                         goto out;
2599                 r = -EFAULT;
2600                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2601                         goto out;
2602                 r = 0;
2603                 break;
2604         }
2605         case KVM_SET_REGS: {
2606                 struct kvm_regs kvm_regs;
2607
2608                 r = -EFAULT;
2609                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2610                         goto out;
2611                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2612                 if (r)
2613                         goto out;
2614                 r = 0;
2615                 break;
2616         }
2617         case KVM_GET_SREGS: {
2618                 struct kvm_sregs kvm_sregs;
2619
2620                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2621                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2622                 if (r)
2623                         goto out;
2624                 r = -EFAULT;
2625                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2626                         goto out;
2627                 r = 0;
2628                 break;
2629         }
2630         case KVM_SET_SREGS: {
2631                 struct kvm_sregs kvm_sregs;
2632
2633                 r = -EFAULT;
2634                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2635                         goto out;
2636                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2637                 if (r)
2638                         goto out;
2639                 r = 0;
2640                 break;
2641         }
2642         case KVM_TRANSLATE: {
2643                 struct kvm_translation tr;
2644
2645                 r = -EFAULT;
2646                 if (copy_from_user(&tr, argp, sizeof tr))
2647                         goto out;
2648                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2649                 if (r)
2650                         goto out;
2651                 r = -EFAULT;
2652                 if (copy_to_user(argp, &tr, sizeof tr))
2653                         goto out;
2654                 r = 0;
2655                 break;
2656         }
2657         case KVM_INTERRUPT: {
2658                 struct kvm_interrupt irq;
2659
2660                 r = -EFAULT;
2661                 if (copy_from_user(&irq, argp, sizeof irq))
2662                         goto out;
2663                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2664                 if (r)
2665                         goto out;
2666                 r = 0;
2667                 break;
2668         }
2669         case KVM_DEBUG_GUEST: {
2670                 struct kvm_debug_guest dbg;
2671
2672                 r = -EFAULT;
2673                 if (copy_from_user(&dbg, argp, sizeof dbg))
2674                         goto out;
2675                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2676                 if (r)
2677                         goto out;
2678                 r = 0;
2679                 break;
2680         }
2681         case KVM_GET_MSRS:
2682                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2683                 break;
2684         case KVM_SET_MSRS:
2685                 r = msr_io(vcpu, argp, do_set_msr, 0);
2686                 break;
2687         case KVM_SET_CPUID: {
2688                 struct kvm_cpuid __user *cpuid_arg = argp;
2689                 struct kvm_cpuid cpuid;
2690
2691                 r = -EFAULT;
2692                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2693                         goto out;
2694                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2695                 if (r)
2696                         goto out;
2697                 break;
2698         }
2699         case KVM_SET_SIGNAL_MASK: {
2700                 struct kvm_signal_mask __user *sigmask_arg = argp;
2701                 struct kvm_signal_mask kvm_sigmask;
2702                 sigset_t sigset, *p;
2703
2704                 p = NULL;
2705                 if (argp) {
2706                         r = -EFAULT;
2707                         if (copy_from_user(&kvm_sigmask, argp,
2708                                            sizeof kvm_sigmask))
2709                                 goto out;
2710                         r = -EINVAL;
2711                         if (kvm_sigmask.len != sizeof sigset)
2712                                 goto out;
2713                         r = -EFAULT;
2714                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2715                                            sizeof sigset))
2716                                 goto out;
2717                         p = &sigset;
2718                 }
2719                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2720                 break;
2721         }
2722         case KVM_GET_FPU: {
2723                 struct kvm_fpu fpu;
2724
2725                 memset(&fpu, 0, sizeof fpu);
2726                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2727                 if (r)
2728                         goto out;
2729                 r = -EFAULT;
2730                 if (copy_to_user(argp, &fpu, sizeof fpu))
2731                         goto out;
2732                 r = 0;
2733                 break;
2734         }
2735         case KVM_SET_FPU: {
2736                 struct kvm_fpu fpu;
2737
2738                 r = -EFAULT;
2739                 if (copy_from_user(&fpu, argp, sizeof fpu))
2740                         goto out;
2741                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2742                 if (r)
2743                         goto out;
2744                 r = 0;
2745                 break;
2746         }
2747         default:
2748                 ;
2749         }
2750 out:
2751         return r;
2752 }
2753
2754 static long kvm_vm_ioctl(struct file *filp,
2755                            unsigned int ioctl, unsigned long arg)
2756 {
2757         struct kvm *kvm = filp->private_data;
2758         void __user *argp = (void __user *)arg;
2759         int r = -EINVAL;
2760
2761         switch (ioctl) {
2762         case KVM_CREATE_VCPU:
2763                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2764                 if (r < 0)
2765                         goto out;
2766                 break;
2767         case KVM_SET_MEMORY_REGION: {
2768                 struct kvm_memory_region kvm_mem;
2769
2770                 r = -EFAULT;
2771                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2772                         goto out;
2773                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2774                 if (r)
2775                         goto out;
2776                 break;
2777         }
2778         case KVM_GET_DIRTY_LOG: {
2779                 struct kvm_dirty_log log;
2780
2781                 r = -EFAULT;
2782                 if (copy_from_user(&log, argp, sizeof log))
2783                         goto out;
2784                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2785                 if (r)
2786                         goto out;
2787                 break;
2788         }
2789         case KVM_SET_MEMORY_ALIAS: {
2790                 struct kvm_memory_alias alias;
2791
2792                 r = -EFAULT;
2793                 if (copy_from_user(&alias, argp, sizeof alias))
2794                         goto out;
2795                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2796                 if (r)
2797                         goto out;
2798                 break;
2799         }
2800         case KVM_CREATE_IRQCHIP:
2801                 r = -ENOMEM;
2802                 kvm->vpic = kvm_create_pic(kvm);
2803                 if (kvm->vpic) {
2804                         r = kvm_ioapic_init(kvm);
2805                         if (r) {
2806                                 kfree(kvm->vpic);
2807                                 kvm->vpic = NULL;
2808                                 goto out;
2809                         }
2810                 }
2811                 else
2812                         goto out;
2813                 break;
2814         case KVM_IRQ_LINE: {
2815                 struct kvm_irq_level irq_event;
2816
2817                 r = -EFAULT;
2818                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2819                         goto out;
2820                 if (irqchip_in_kernel(kvm)) {
2821                         if (irq_event.irq < 16)
2822                                 kvm_pic_set_irq(pic_irqchip(kvm),
2823                                         irq_event.irq,
2824                                         irq_event.level);
2825                         kvm_ioapic_set_irq(kvm->vioapic,
2826                                         irq_event.irq,
2827                                         irq_event.level);
2828                         r = 0;
2829                 }
2830                 break;
2831         }
2832         default:
2833                 ;
2834         }
2835 out:
2836         return r;
2837 }
2838
2839 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2840                                   unsigned long address,
2841                                   int *type)
2842 {
2843         struct kvm *kvm = vma->vm_file->private_data;
2844         unsigned long pgoff;
2845         struct page *page;
2846
2847         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2848         page = gfn_to_page(kvm, pgoff);
2849         if (!page)
2850                 return NOPAGE_SIGBUS;
2851         get_page(page);
2852         if (type != NULL)
2853                 *type = VM_FAULT_MINOR;
2854
2855         return page;
2856 }
2857
2858 static struct vm_operations_struct kvm_vm_vm_ops = {
2859         .nopage = kvm_vm_nopage,
2860 };
2861
2862 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2863 {
2864         vma->vm_ops = &kvm_vm_vm_ops;
2865         return 0;
2866 }
2867
2868 static struct file_operations kvm_vm_fops = {
2869         .release        = kvm_vm_release,
2870         .unlocked_ioctl = kvm_vm_ioctl,
2871         .compat_ioctl   = kvm_vm_ioctl,
2872         .mmap           = kvm_vm_mmap,
2873 };
2874
2875 static int kvm_dev_ioctl_create_vm(void)
2876 {
2877         int fd, r;
2878         struct inode *inode;
2879         struct file *file;
2880         struct kvm *kvm;
2881
2882         kvm = kvm_create_vm();
2883         if (IS_ERR(kvm))
2884                 return PTR_ERR(kvm);
2885         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2886         if (r) {
2887                 kvm_destroy_vm(kvm);
2888                 return r;
2889         }
2890
2891         kvm->filp = file;
2892
2893         return fd;
2894 }
2895
2896 static long kvm_dev_ioctl(struct file *filp,
2897                           unsigned int ioctl, unsigned long arg)
2898 {
2899         void __user *argp = (void __user *)arg;
2900         long r = -EINVAL;
2901
2902         switch (ioctl) {
2903         case KVM_GET_API_VERSION:
2904                 r = -EINVAL;
2905                 if (arg)
2906                         goto out;
2907                 r = KVM_API_VERSION;
2908                 break;
2909         case KVM_CREATE_VM:
2910                 r = -EINVAL;
2911                 if (arg)
2912                         goto out;
2913                 r = kvm_dev_ioctl_create_vm();
2914                 break;
2915         case KVM_GET_MSR_INDEX_LIST: {
2916                 struct kvm_msr_list __user *user_msr_list = argp;
2917                 struct kvm_msr_list msr_list;
2918                 unsigned n;
2919
2920                 r = -EFAULT;
2921                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2922                         goto out;
2923                 n = msr_list.nmsrs;
2924                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2925                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2926                         goto out;
2927                 r = -E2BIG;
2928                 if (n < num_msrs_to_save)
2929                         goto out;
2930                 r = -EFAULT;
2931                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2932                                  num_msrs_to_save * sizeof(u32)))
2933                         goto out;
2934                 if (copy_to_user(user_msr_list->indices
2935                                  + num_msrs_to_save * sizeof(u32),
2936                                  &emulated_msrs,
2937                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2938                         goto out;
2939                 r = 0;
2940                 break;
2941         }
2942         case KVM_CHECK_EXTENSION: {
2943                 int ext = (long)argp;
2944
2945                 switch (ext) {
2946                 case KVM_CAP_IRQCHIP:
2947                 case KVM_CAP_HLT:
2948                         r = 1;
2949                         break;
2950                 default:
2951                         r = 0;
2952                         break;
2953                 }
2954                 break;
2955         }
2956         case KVM_GET_VCPU_MMAP_SIZE:
2957                 r = -EINVAL;
2958                 if (arg)
2959                         goto out;
2960                 r = 2 * PAGE_SIZE;
2961                 break;
2962         default:
2963                 ;
2964         }
2965 out:
2966         return r;
2967 }
2968
2969 static struct file_operations kvm_chardev_ops = {
2970         .unlocked_ioctl = kvm_dev_ioctl,
2971         .compat_ioctl   = kvm_dev_ioctl,
2972 };
2973
2974 static struct miscdevice kvm_dev = {
2975         KVM_MINOR,
2976         "kvm",
2977         &kvm_chardev_ops,
2978 };
2979
2980 /*
2981  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2982  * cached on it.
2983  */
2984 static void decache_vcpus_on_cpu(int cpu)
2985 {
2986         struct kvm *vm;
2987         struct kvm_vcpu *vcpu;
2988         int i;
2989
2990         spin_lock(&kvm_lock);
2991         list_for_each_entry(vm, &vm_list, vm_list)
2992                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2993                         vcpu = vm->vcpus[i];
2994                         if (!vcpu)
2995                                 continue;
2996                         /*
2997                          * If the vcpu is locked, then it is running on some
2998                          * other cpu and therefore it is not cached on the
2999                          * cpu in question.
3000                          *
3001                          * If it's not locked, check the last cpu it executed
3002                          * on.
3003                          */
3004                         if (mutex_trylock(&vcpu->mutex)) {
3005                                 if (vcpu->cpu == cpu) {
3006                                         kvm_arch_ops->vcpu_decache(vcpu);
3007                                         vcpu->cpu = -1;
3008                                 }
3009                                 mutex_unlock(&vcpu->mutex);
3010                         }
3011                 }
3012         spin_unlock(&kvm_lock);
3013 }
3014
3015 static void hardware_enable(void *junk)
3016 {
3017         int cpu = raw_smp_processor_id();
3018
3019         if (cpu_isset(cpu, cpus_hardware_enabled))
3020                 return;
3021         cpu_set(cpu, cpus_hardware_enabled);
3022         kvm_arch_ops->hardware_enable(NULL);
3023 }
3024
3025 static void hardware_disable(void *junk)
3026 {
3027         int cpu = raw_smp_processor_id();
3028
3029         if (!cpu_isset(cpu, cpus_hardware_enabled))
3030                 return;
3031         cpu_clear(cpu, cpus_hardware_enabled);
3032         decache_vcpus_on_cpu(cpu);
3033         kvm_arch_ops->hardware_disable(NULL);
3034 }
3035
3036 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3037                            void *v)
3038 {
3039         int cpu = (long)v;
3040
3041         switch (val) {
3042         case CPU_DYING:
3043         case CPU_DYING_FROZEN:
3044                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3045                        cpu);
3046                 hardware_disable(NULL);
3047                 break;
3048         case CPU_UP_CANCELED:
3049         case CPU_UP_CANCELED_FROZEN:
3050                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3051                        cpu);
3052                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3053                 break;
3054         case CPU_ONLINE:
3055         case CPU_ONLINE_FROZEN:
3056                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3057                        cpu);
3058                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3059                 break;
3060         }
3061         return NOTIFY_OK;
3062 }
3063
3064 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3065                        void *v)
3066 {
3067         if (val == SYS_RESTART) {
3068                 /*
3069                  * Some (well, at least mine) BIOSes hang on reboot if
3070                  * in vmx root mode.
3071                  */
3072                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3073                 on_each_cpu(hardware_disable, NULL, 0, 1);
3074         }
3075         return NOTIFY_OK;
3076 }
3077
3078 static struct notifier_block kvm_reboot_notifier = {
3079         .notifier_call = kvm_reboot,
3080         .priority = 0,
3081 };
3082
3083 void kvm_io_bus_init(struct kvm_io_bus *bus)
3084 {
3085         memset(bus, 0, sizeof(*bus));
3086 }
3087
3088 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3089 {
3090         int i;
3091
3092         for (i = 0; i < bus->dev_count; i++) {
3093                 struct kvm_io_device *pos = bus->devs[i];
3094
3095                 kvm_iodevice_destructor(pos);
3096         }
3097 }
3098
3099 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3100 {
3101         int i;
3102
3103         for (i = 0; i < bus->dev_count; i++) {
3104                 struct kvm_io_device *pos = bus->devs[i];
3105
3106                 if (pos->in_range(pos, addr))
3107                         return pos;
3108         }
3109
3110         return NULL;
3111 }
3112
3113 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3114 {
3115         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3116
3117         bus->devs[bus->dev_count++] = dev;
3118 }
3119
3120 static struct notifier_block kvm_cpu_notifier = {
3121         .notifier_call = kvm_cpu_hotplug,
3122         .priority = 20, /* must be > scheduler priority */
3123 };
3124
3125 static u64 stat_get(void *_offset)
3126 {
3127         unsigned offset = (long)_offset;
3128         u64 total = 0;
3129         struct kvm *kvm;
3130         struct kvm_vcpu *vcpu;
3131         int i;
3132
3133         spin_lock(&kvm_lock);
3134         list_for_each_entry(kvm, &vm_list, vm_list)
3135                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3136                         vcpu = kvm->vcpus[i];
3137                         if (vcpu)
3138                                 total += *(u32 *)((void *)vcpu + offset);
3139                 }
3140         spin_unlock(&kvm_lock);
3141         return total;
3142 }
3143
3144 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3145
3146 static __init void kvm_init_debug(void)
3147 {
3148         struct kvm_stats_debugfs_item *p;
3149
3150         debugfs_dir = debugfs_create_dir("kvm", NULL);
3151         for (p = debugfs_entries; p->name; ++p)
3152                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3153                                                 (void *)(long)p->offset,
3154                                                 &stat_fops);
3155 }
3156
3157 static void kvm_exit_debug(void)
3158 {
3159         struct kvm_stats_debugfs_item *p;
3160
3161         for (p = debugfs_entries; p->name; ++p)
3162                 debugfs_remove(p->dentry);
3163         debugfs_remove(debugfs_dir);
3164 }
3165
3166 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3167 {
3168         hardware_disable(NULL);
3169         return 0;
3170 }
3171
3172 static int kvm_resume(struct sys_device *dev)
3173 {
3174         hardware_enable(NULL);
3175         return 0;
3176 }
3177
3178 static struct sysdev_class kvm_sysdev_class = {
3179         set_kset_name("kvm"),
3180         .suspend = kvm_suspend,
3181         .resume = kvm_resume,
3182 };
3183
3184 static struct sys_device kvm_sysdev = {
3185         .id = 0,
3186         .cls = &kvm_sysdev_class,
3187 };
3188
3189 hpa_t bad_page_address;
3190
3191 static inline
3192 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3193 {
3194         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3195 }
3196
3197 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3198 {
3199         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3200
3201         kvm_arch_ops->vcpu_load(vcpu, cpu);
3202 }
3203
3204 static void kvm_sched_out(struct preempt_notifier *pn,
3205                           struct task_struct *next)
3206 {
3207         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3208
3209         kvm_arch_ops->vcpu_put(vcpu);
3210 }
3211
3212 int kvm_init_arch(struct kvm_arch_ops *ops, unsigned int vcpu_size,
3213                   struct module *module)
3214 {
3215         int r;
3216         int cpu;
3217
3218         if (kvm_arch_ops) {
3219                 printk(KERN_ERR "kvm: already loaded the other module\n");
3220                 return -EEXIST;
3221         }
3222
3223         if (!ops->cpu_has_kvm_support()) {
3224                 printk(KERN_ERR "kvm: no hardware support\n");
3225                 return -EOPNOTSUPP;
3226         }
3227         if (ops->disabled_by_bios()) {
3228                 printk(KERN_ERR "kvm: disabled by bios\n");
3229                 return -EOPNOTSUPP;
3230         }
3231
3232         kvm_arch_ops = ops;
3233
3234         r = kvm_arch_ops->hardware_setup();
3235         if (r < 0)
3236                 goto out;
3237
3238         for_each_online_cpu(cpu) {
3239                 smp_call_function_single(cpu,
3240                                 kvm_arch_ops->check_processor_compatibility,
3241                                 &r, 0, 1);
3242                 if (r < 0)
3243                         goto out_free_0;
3244         }
3245
3246         on_each_cpu(hardware_enable, NULL, 0, 1);
3247         r = register_cpu_notifier(&kvm_cpu_notifier);
3248         if (r)
3249                 goto out_free_1;
3250         register_reboot_notifier(&kvm_reboot_notifier);
3251
3252         r = sysdev_class_register(&kvm_sysdev_class);
3253         if (r)
3254                 goto out_free_2;
3255
3256         r = sysdev_register(&kvm_sysdev);
3257         if (r)
3258                 goto out_free_3;
3259
3260         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3261         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3262                                            __alignof__(struct kvm_vcpu), 0, 0);
3263         if (!kvm_vcpu_cache) {
3264                 r = -ENOMEM;
3265                 goto out_free_4;
3266         }
3267
3268         kvm_chardev_ops.owner = module;
3269
3270         r = misc_register(&kvm_dev);
3271         if (r) {
3272                 printk (KERN_ERR "kvm: misc device register failed\n");
3273                 goto out_free;
3274         }
3275
3276         kvm_preempt_ops.sched_in = kvm_sched_in;
3277         kvm_preempt_ops.sched_out = kvm_sched_out;
3278
3279         return r;
3280
3281 out_free:
3282         kmem_cache_destroy(kvm_vcpu_cache);
3283 out_free_4:
3284         sysdev_unregister(&kvm_sysdev);
3285 out_free_3:
3286         sysdev_class_unregister(&kvm_sysdev_class);
3287 out_free_2:
3288         unregister_reboot_notifier(&kvm_reboot_notifier);
3289         unregister_cpu_notifier(&kvm_cpu_notifier);
3290 out_free_1:
3291         on_each_cpu(hardware_disable, NULL, 0, 1);
3292 out_free_0:
3293         kvm_arch_ops->hardware_unsetup();
3294 out:
3295         kvm_arch_ops = NULL;
3296         return r;
3297 }
3298
3299 void kvm_exit_arch(void)
3300 {
3301         misc_deregister(&kvm_dev);
3302         kmem_cache_destroy(kvm_vcpu_cache);
3303         sysdev_unregister(&kvm_sysdev);
3304         sysdev_class_unregister(&kvm_sysdev_class);
3305         unregister_reboot_notifier(&kvm_reboot_notifier);
3306         unregister_cpu_notifier(&kvm_cpu_notifier);
3307         on_each_cpu(hardware_disable, NULL, 0, 1);
3308         kvm_arch_ops->hardware_unsetup();
3309         kvm_arch_ops = NULL;
3310 }
3311
3312 static __init int kvm_init(void)
3313 {
3314         static struct page *bad_page;
3315         int r;
3316
3317         r = kvm_mmu_module_init();
3318         if (r)
3319                 goto out4;
3320
3321         kvm_init_debug();
3322
3323         kvm_init_msr_list();
3324
3325         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3326                 r = -ENOMEM;
3327                 goto out;
3328         }
3329
3330         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3331         memset(__va(bad_page_address), 0, PAGE_SIZE);
3332
3333         return 0;
3334
3335 out:
3336         kvm_exit_debug();
3337         kvm_mmu_module_exit();
3338 out4:
3339         return r;
3340 }
3341
3342 static __exit void kvm_exit(void)
3343 {
3344         kvm_exit_debug();
3345         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3346         kvm_mmu_module_exit();
3347 }
3348
3349 module_init(kvm_init)
3350 module_exit(kvm_exit)
3351
3352 EXPORT_SYMBOL_GPL(kvm_init_arch);
3353 EXPORT_SYMBOL_GPL(kvm_exit_arch);