drivers/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19 #include "x86_emulate.h"
  20 #include "segment_descriptor.h"
  21
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/gfp.h>
  27 #include <linux/mm.h>
  28 #include <linux/miscdevice.h>
  29 #include <linux/vmalloc.h>
  30 #include <linux/reboot.h>
  31 #include <linux/debugfs.h>
  32 #include <linux/highmem.h>
  33 #include <linux/file.h>
  34 #include <linux/sysdev.h>
  35 #include <linux/cpu.h>
  36 #include <linux/sched.h>
  37 #include <linux/cpumask.h>
  38 #include <linux/smp.h>
  39 #include <linux/anon_inodes.h>
  40
  41 #include <asm/processor.h>
  42 #include <asm/msr.h>
  43 #include <asm/io.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/desc.h>
  46
  47 MODULE_AUTHOR("Qumranet");
  48 MODULE_LICENSE("GPL");
  49
  50 static DEFINE_SPINLOCK(kvm_lock);
  51 static LIST_HEAD(vm_list);
  52
  53 static cpumask_t cpus_hardware_enabled;
  54
  55 struct kvm_arch_ops *kvm_arch_ops;
  56
  57 static __read_mostly struct preempt_ops kvm_preempt_ops;
  58
  59 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  60
  61 static struct kvm_stats_debugfs_item {
  62         const char *name;
  63         int offset;
  64         struct dentry *dentry;
  65 } debugfs_entries[] = {
  66         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  67         { "pf_guest", STAT_OFFSET(pf_guest) },
  68         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  69         { "invlpg", STAT_OFFSET(invlpg) },
  70         { "exits", STAT_OFFSET(exits) },
  71         { "io_exits", STAT_OFFSET(io_exits) },
  72         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  73         { "signal_exits", STAT_OFFSET(signal_exits) },
  74         { "irq_window", STAT_OFFSET(irq_window_exits) },
  75         { "halt_exits", STAT_OFFSET(halt_exits) },
  76         { "request_irq", STAT_OFFSET(request_irq_exits) },
  77         { "irq_exits", STAT_OFFSET(irq_exits) },
  78         { "light_exits", STAT_OFFSET(light_exits) },
  79         { "efer_reload", STAT_OFFSET(efer_reload) },
  80         { NULL }
  81 };
  82
  83 static struct dentry *debugfs_dir;
  84
  85 #define MAX_IO_MSRS 256
  86
  87 #define CR0_RESERVED_BITS                                               \
  88         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  89                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  90                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  91 #define CR4_RESERVED_BITS                                               \
  92         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  93                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
  94                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
  95                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
  96
  97 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
  98 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  99
 100 #ifdef CONFIG_X86_64
 101 // LDT or TSS descriptor in the GDT. 16 bytes.
 102 struct segment_descriptor_64 {
 103         struct segment_descriptor s;
 104         u32 base_higher;
 105         u32 pad_zero;
 106 };
 107
 108 #endif
 109
 110 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 111                            unsigned long arg);
 112
 113 unsigned long segment_base(u16 selector)
 114 {
 115         struct descriptor_table gdt;
 116         struct segment_descriptor *d;
 117         unsigned long table_base;
 118         typedef unsigned long ul;
 119         unsigned long v;
 120
 121         if (selector == 0)
 122                 return 0;
 123
 124         asm ("sgdt %0" : "=m"(gdt));
 125         table_base = gdt.base;
 126
 127         if (selector & 4) {           /* from ldt */
 128                 u16 ldt_selector;
 129
 130                 asm ("sldt %0" : "=g"(ldt_selector));
 131                 table_base = segment_base(ldt_selector);
 132         }
 133         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 134         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 135 #ifdef CONFIG_X86_64
 136         if (d->system == 0
 137             && (d->type == 2 || d->type == 9 || d->type == 11))
 138                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 139 #endif
 140         return v;
 141 }
 142 EXPORT_SYMBOL_GPL(segment_base);
 143
 144 static inline int valid_vcpu(int n)
 145 {
 146         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 147 }
 148
 149 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 150                    void *dest)
 151 {
 152         unsigned char *host_buf = dest;
 153         unsigned long req_size = size;
 154
 155         while (size) {
 156                 hpa_t paddr;
 157                 unsigned now;
 158                 unsigned offset;
 159                 hva_t guest_buf;
 160
 161                 paddr = gva_to_hpa(vcpu, addr);
 162
 163                 if (is_error_hpa(paddr))
 164                         break;
 165
 166                 guest_buf = (hva_t)kmap_atomic(
 167                                         pfn_to_page(paddr >> PAGE_SHIFT),
 168                                         KM_USER0);
 169                 offset = addr & ~PAGE_MASK;
 170                 guest_buf |= offset;
 171                 now = min(size, PAGE_SIZE - offset);
 172                 memcpy(host_buf, (void*)guest_buf, now);
 173                 host_buf += now;
 174                 addr += now;
 175                 size -= now;
 176                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 177         }
 178         return req_size - size;
 179 }
 180 EXPORT_SYMBOL_GPL(kvm_read_guest);
 181
 182 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 183                     void *data)
 184 {
 185         unsigned char *host_buf = data;
 186         unsigned long req_size = size;
 187
 188         while (size) {
 189                 hpa_t paddr;
 190                 unsigned now;
 191                 unsigned offset;
 192                 hva_t guest_buf;
 193                 gfn_t gfn;
 194
 195                 paddr = gva_to_hpa(vcpu, addr);
 196
 197                 if (is_error_hpa(paddr))
 198                         break;
 199
 200                 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
 201                 mark_page_dirty(vcpu->kvm, gfn);
 202                 guest_buf = (hva_t)kmap_atomic(
 203                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
 204                 offset = addr & ~PAGE_MASK;
 205                 guest_buf |= offset;
 206                 now = min(size, PAGE_SIZE - offset);
 207                 memcpy((void*)guest_buf, host_buf, now);
 208                 host_buf += now;
 209                 addr += now;
 210                 size -= now;
 211                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 212         }
 213         return req_size - size;
 214 }
 215 EXPORT_SYMBOL_GPL(kvm_write_guest);
 216
 217 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 218 {
 219         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 220                 return;
 221
 222         vcpu->guest_fpu_loaded = 1;
 223         fx_save(vcpu->host_fx_image);
 224         fx_restore(vcpu->guest_fx_image);
 225 }
 226 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 227
 228 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 229 {
 230         if (!vcpu->guest_fpu_loaded)
 231                 return;
 232
 233         vcpu->guest_fpu_loaded = 0;
 234         fx_save(vcpu->guest_fx_image);
 235         fx_restore(vcpu->host_fx_image);
 236 }
 237 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 238
 239 /*
 240  * Switches to specified vcpu, until a matching vcpu_put()
 241  */
 242 static void vcpu_load(struct kvm_vcpu *vcpu)
 243 {
 244         int cpu;
 245
 246         mutex_lock(&vcpu->mutex);
 247         cpu = get_cpu();
 248         preempt_notifier_register(&vcpu->preempt_notifier);
 249         kvm_arch_ops->vcpu_load(vcpu, cpu);
 250         put_cpu();
 251 }
 252
 253 static void vcpu_put(struct kvm_vcpu *vcpu)
 254 {
 255         preempt_disable();
 256         kvm_arch_ops->vcpu_put(vcpu);
 257         preempt_notifier_unregister(&vcpu->preempt_notifier);
 258         preempt_enable();
 259         mutex_unlock(&vcpu->mutex);
 260 }
 261
 262 static void ack_flush(void *_completed)
 263 {
 264         atomic_t *completed = _completed;
 265
 266         atomic_inc(completed);
 267 }
 268
 269 void kvm_flush_remote_tlbs(struct kvm *kvm)
 270 {
 271         int i, cpu, needed;
 272         cpumask_t cpus;
 273         struct kvm_vcpu *vcpu;
 274         atomic_t completed;
 275
 276         atomic_set(&completed, 0);
 277         cpus_clear(cpus);
 278         needed = 0;
 279         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 280                 vcpu = kvm->vcpus[i];
 281                 if (!vcpu)
 282                         continue;
 283                 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
 284                         continue;
 285                 cpu = vcpu->cpu;
 286                 if (cpu != -1 && cpu != raw_smp_processor_id())
 287                         if (!cpu_isset(cpu, cpus)) {
 288                                 cpu_set(cpu, cpus);
 289                                 ++needed;
 290                         }
 291         }
 292
 293         /*
 294          * We really want smp_call_function_mask() here.  But that's not
 295          * available, so ipi all cpus in parallel and wait for them
 296          * to complete.
 297          */
 298         for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
 299                 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
 300         while (atomic_read(&completed) != needed) {
 301                 cpu_relax();
 302                 barrier();
 303         }
 304 }
 305
 306 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 307 {
 308         struct page *page;
 309         int r;
 310
 311         mutex_init(&vcpu->mutex);
 312         vcpu->cpu = -1;
 313         vcpu->mmu.root_hpa = INVALID_PAGE;
 314         vcpu->kvm = kvm;
 315         vcpu->vcpu_id = id;
 316
 317         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 318         if (!page) {
 319                 r = -ENOMEM;
 320                 goto fail;
 321         }
 322         vcpu->run = page_address(page);
 323
 324         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 325         if (!page) {
 326                 r = -ENOMEM;
 327                 goto fail_free_run;
 328         }
 329         vcpu->pio_data = page_address(page);
 330
 331         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
 332                                            FX_IMAGE_ALIGN);
 333         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
 334
 335         r = kvm_mmu_create(vcpu);
 336         if (r < 0)
 337                 goto fail_free_pio_data;
 338
 339         return 0;
 340
 341 fail_free_pio_data:
 342         free_page((unsigned long)vcpu->pio_data);
 343 fail_free_run:
 344         free_page((unsigned long)vcpu->run);
 345 fail:
 346         return -ENOMEM;
 347 }
 348 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 349
 350 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 351 {
 352         kvm_mmu_destroy(vcpu);
 353         free_page((unsigned long)vcpu->pio_data);
 354         free_page((unsigned long)vcpu->run);
 355 }
 356 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 357
 358 static struct kvm *kvm_create_vm(void)
 359 {
 360         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 361
 362         if (!kvm)
 363                 return ERR_PTR(-ENOMEM);
 364
 365         kvm_io_bus_init(&kvm->pio_bus);
 366         mutex_init(&kvm->lock);
 367         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 368         kvm_io_bus_init(&kvm->mmio_bus);
 369         spin_lock(&kvm_lock);
 370         list_add(&kvm->vm_list, &vm_list);
 371         spin_unlock(&kvm_lock);
 372         return kvm;
 373 }
 374
 375 static int kvm_dev_open(struct inode *inode, struct file *filp)
 376 {
 377         return 0;
 378 }
 379
 380 /*
 381  * Free any memory in @free but not in @dont.
 382  */
 383 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 384                                   struct kvm_memory_slot *dont)
 385 {
 386         int i;
 387
 388         if (!dont || free->phys_mem != dont->phys_mem)
 389                 if (free->phys_mem) {
 390                         for (i = 0; i < free->npages; ++i)
 391                                 if (free->phys_mem[i])
 392                                         __free_page(free->phys_mem[i]);
 393                         vfree(free->phys_mem);
 394                 }
 395
 396         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 397                 vfree(free->dirty_bitmap);
 398
 399         free->phys_mem = NULL;
 400         free->npages = 0;
 401         free->dirty_bitmap = NULL;
 402 }
 403
 404 static void kvm_free_physmem(struct kvm *kvm)
 405 {
 406         int i;
 407
 408         for (i = 0; i < kvm->nmemslots; ++i)
 409                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 410 }
 411
 412 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 413 {
 414         int i;
 415
 416         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
 417                 if (vcpu->pio.guest_pages[i]) {
 418                         __free_page(vcpu->pio.guest_pages[i]);
 419                         vcpu->pio.guest_pages[i] = NULL;
 420                 }
 421 }
 422
 423 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 424 {
 425         vcpu_load(vcpu);
 426         kvm_mmu_unload(vcpu);
 427         vcpu_put(vcpu);
 428 }
 429
 430 static void kvm_free_vcpus(struct kvm *kvm)
 431 {
 432         unsigned int i;
 433
 434         /*
 435          * Unpin any mmu pages first.
 436          */
 437         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 438                 if (kvm->vcpus[i])
 439                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
 440         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 441                 if (kvm->vcpus[i]) {
 442                         kvm_arch_ops->vcpu_free(kvm->vcpus[i]);
 443                         kvm->vcpus[i] = NULL;
 444                 }
 445         }
 446
 447 }
 448
 449 static int kvm_dev_release(struct inode *inode, struct file *filp)
 450 {
 451         return 0;
 452 }
 453
 454 static void kvm_destroy_vm(struct kvm *kvm)
 455 {
 456         spin_lock(&kvm_lock);
 457         list_del(&kvm->vm_list);
 458         spin_unlock(&kvm_lock);
 459         kvm_io_bus_destroy(&kvm->pio_bus);
 460         kvm_io_bus_destroy(&kvm->mmio_bus);
 461         kvm_free_vcpus(kvm);
 462         kvm_free_physmem(kvm);
 463         kfree(kvm);
 464 }
 465
 466 static int kvm_vm_release(struct inode *inode, struct file *filp)
 467 {
 468         struct kvm *kvm = filp->private_data;
 469
 470         kvm_destroy_vm(kvm);
 471         return 0;
 472 }
 473
 474 static void inject_gp(struct kvm_vcpu *vcpu)
 475 {
 476         kvm_arch_ops->inject_gp(vcpu, 0);
 477 }
 478
 479 /*
 480  * Load the pae pdptrs.  Return true is they are all valid.
 481  */
 482 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 483 {
 484         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 485         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 486         int i;
 487         u64 *pdpt;
 488         int ret;
 489         struct page *page;
 490         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 491
 492         mutex_lock(&vcpu->kvm->lock);
 493         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
 494         if (!page) {
 495                 ret = 0;
 496                 goto out;
 497         }
 498
 499         pdpt = kmap_atomic(page, KM_USER0);
 500         memcpy(pdpte, pdpt+offset, sizeof(pdpte));
 501         kunmap_atomic(pdpt, KM_USER0);
 502
 503         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 504                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 505                         ret = 0;
 506                         goto out;
 507                 }
 508         }
 509         ret = 1;
 510
 511         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 512 out:
 513         mutex_unlock(&vcpu->kvm->lock);
 514
 515         return ret;
 516 }
 517
 518 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 519 {
 520         if (cr0 & CR0_RESERVED_BITS) {
 521                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 522                        cr0, vcpu->cr0);
 523                 inject_gp(vcpu);
 524                 return;
 525         }
 526
 527         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 528                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 529                 inject_gp(vcpu);
 530                 return;
 531         }
 532
 533         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 534                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 535                        "and a clear PE flag\n");
 536                 inject_gp(vcpu);
 537                 return;
 538         }
 539
 540         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 541 #ifdef CONFIG_X86_64
 542                 if ((vcpu->shadow_efer & EFER_LME)) {
 543                         int cs_db, cs_l;
 544
 545                         if (!is_pae(vcpu)) {
 546                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 547                                        "in long mode while PAE is disabled\n");
 548                                 inject_gp(vcpu);
 549                                 return;
 550                         }
 551                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 552                         if (cs_l) {
 553                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 554                                        "in long mode while CS.L == 1\n");
 555                                 inject_gp(vcpu);
 556                                 return;
 557
 558                         }
 559                 } else
 560 #endif
 561                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 562                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 563                                "reserved bits\n");
 564                         inject_gp(vcpu);
 565                         return;
 566                 }
 567
 568         }
 569
 570         kvm_arch_ops->set_cr0(vcpu, cr0);
 571         vcpu->cr0 = cr0;
 572
 573         mutex_lock(&vcpu->kvm->lock);
 574         kvm_mmu_reset_context(vcpu);
 575         mutex_unlock(&vcpu->kvm->lock);
 576         return;
 577 }
 578 EXPORT_SYMBOL_GPL(set_cr0);
 579
 580 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 581 {
 582         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 583 }
 584 EXPORT_SYMBOL_GPL(lmsw);
 585
 586 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 587 {
 588         if (cr4 & CR4_RESERVED_BITS) {
 589                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 590                 inject_gp(vcpu);
 591                 return;
 592         }
 593
 594         if (is_long_mode(vcpu)) {
 595                 if (!(cr4 & X86_CR4_PAE)) {
 596                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 597                                "in long mode\n");
 598                         inject_gp(vcpu);
 599                         return;
 600                 }
 601         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 602                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 603                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 604                 inject_gp(vcpu);
 605                 return;
 606         }
 607
 608         if (cr4 & X86_CR4_VMXE) {
 609                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 610                 inject_gp(vcpu);
 611                 return;
 612         }
 613         kvm_arch_ops->set_cr4(vcpu, cr4);
 614         mutex_lock(&vcpu->kvm->lock);
 615         kvm_mmu_reset_context(vcpu);
 616         mutex_unlock(&vcpu->kvm->lock);
 617 }
 618 EXPORT_SYMBOL_GPL(set_cr4);
 619
 620 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 621 {
 622         if (is_long_mode(vcpu)) {
 623                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 624                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 625                         inject_gp(vcpu);
 626                         return;
 627                 }
 628         } else {
 629                 if (is_pae(vcpu)) {
 630                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 631                                 printk(KERN_DEBUG
 632                                        "set_cr3: #GP, reserved bits\n");
 633                                 inject_gp(vcpu);
 634                                 return;
 635                         }
 636                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 637                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 638                                        "reserved bits\n");
 639                                 inject_gp(vcpu);
 640                                 return;
 641                         }
 642                 } else {
 643                         if (cr3 & CR3_NONPAE_RESERVED_BITS) {
 644                                 printk(KERN_DEBUG
 645                                        "set_cr3: #GP, reserved bits\n");
 646                                 inject_gp(vcpu);
 647                                 return;
 648                         }
 649                 }
 650         }
 651
 652         vcpu->cr3 = cr3;
 653         mutex_lock(&vcpu->kvm->lock);
 654         /*
 655          * Does the new cr3 value map to physical memory? (Note, we
 656          * catch an invalid cr3 even in real-mode, because it would
 657          * cause trouble later on when we turn on paging anyway.)
 658          *
 659          * A real CPU would silently accept an invalid cr3 and would
 660          * attempt to use it - with largely undefined (and often hard
 661          * to debug) behavior on the guest side.
 662          */
 663         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 664                 inject_gp(vcpu);
 665         else
 666                 vcpu->mmu.new_cr3(vcpu);
 667         mutex_unlock(&vcpu->kvm->lock);
 668 }
 669 EXPORT_SYMBOL_GPL(set_cr3);
 670
 671 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 672 {
 673         if (cr8 & CR8_RESERVED_BITS) {
 674                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 675                 inject_gp(vcpu);
 676                 return;
 677         }
 678         vcpu->cr8 = cr8;
 679 }
 680 EXPORT_SYMBOL_GPL(set_cr8);
 681
 682 void fx_init(struct kvm_vcpu *vcpu)
 683 {
 684         struct __attribute__ ((__packed__)) fx_image_s {
 685                 u16 control; //fcw
 686                 u16 status; //fsw
 687                 u16 tag; // ftw
 688                 u16 opcode; //fop
 689                 u64 ip; // fpu ip
 690                 u64 operand;// fpu dp
 691                 u32 mxcsr;
 692                 u32 mxcsr_mask;
 693
 694         } *fx_image;
 695
 696         /* Initialize guest FPU by resetting ours and saving into guest's */
 697         preempt_disable();
 698         fx_save(vcpu->host_fx_image);
 699         fpu_init();
 700         fx_save(vcpu->guest_fx_image);
 701         fx_restore(vcpu->host_fx_image);
 702         preempt_enable();
 703
 704         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
 705         fx_image->mxcsr = 0x1f80;
 706         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
 707                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
 708 }
 709 EXPORT_SYMBOL_GPL(fx_init);
 710
 711 /*
 712  * Allocate some memory and give it an address in the guest physical address
 713  * space.
 714  *
 715  * Discontiguous memory is allowed, mostly for framebuffers.
 716  */
 717 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 718                                           struct kvm_memory_region *mem)
 719 {
 720         int r;
 721         gfn_t base_gfn;
 722         unsigned long npages;
 723         unsigned long i;
 724         struct kvm_memory_slot *memslot;
 725         struct kvm_memory_slot old, new;
 726         int memory_config_version;
 727
 728         r = -EINVAL;
 729         /* General sanity checks */
 730         if (mem->memory_size & (PAGE_SIZE - 1))
 731                 goto out;
 732         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 733                 goto out;
 734         if (mem->slot >= KVM_MEMORY_SLOTS)
 735                 goto out;
 736         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 737                 goto out;
 738
 739         memslot = &kvm->memslots[mem->slot];
 740         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 741         npages = mem->memory_size >> PAGE_SHIFT;
 742
 743         if (!npages)
 744                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 745
 746 raced:
 747         mutex_lock(&kvm->lock);
 748
 749         memory_config_version = kvm->memory_config_version;
 750         new = old = *memslot;
 751
 752         new.base_gfn = base_gfn;
 753         new.npages = npages;
 754         new.flags = mem->flags;
 755
 756         /* Disallow changing a memory slot's size. */
 757         r = -EINVAL;
 758         if (npages && old.npages && npages != old.npages)
 759                 goto out_unlock;
 760
 761         /* Check for overlaps */
 762         r = -EEXIST;
 763         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 764                 struct kvm_memory_slot *s = &kvm->memslots[i];
 765
 766                 if (s == memslot)
 767                         continue;
 768                 if (!((base_gfn + npages <= s->base_gfn) ||
 769                       (base_gfn >= s->base_gfn + s->npages)))
 770                         goto out_unlock;
 771         }
 772         /*
 773          * Do memory allocations outside lock.  memory_config_version will
 774          * detect any races.
 775          */
 776         mutex_unlock(&kvm->lock);
 777
 778         /* Deallocate if slot is being removed */
 779         if (!npages)
 780                 new.phys_mem = NULL;
 781
 782         /* Free page dirty bitmap if unneeded */
 783         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 784                 new.dirty_bitmap = NULL;
 785
 786         r = -ENOMEM;
 787
 788         /* Allocate if a slot is being created */
 789         if (npages && !new.phys_mem) {
 790                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 791
 792                 if (!new.phys_mem)
 793                         goto out_free;
 794
 795                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 796                 for (i = 0; i < npages; ++i) {
 797                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 798                                                      | __GFP_ZERO);
 799                         if (!new.phys_mem[i])
 800                                 goto out_free;
 801                         set_page_private(new.phys_mem[i],0);
 802                 }
 803         }
 804
 805         /* Allocate page dirty bitmap if needed */
 806         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 807                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 808
 809                 new.dirty_bitmap = vmalloc(dirty_bytes);
 810                 if (!new.dirty_bitmap)
 811                         goto out_free;
 812                 memset(new.dirty_bitmap, 0, dirty_bytes);
 813         }
 814
 815         mutex_lock(&kvm->lock);
 816
 817         if (memory_config_version != kvm->memory_config_version) {
 818                 mutex_unlock(&kvm->lock);
 819                 kvm_free_physmem_slot(&new, &old);
 820                 goto raced;
 821         }
 822
 823         r = -EAGAIN;
 824         if (kvm->busy)
 825                 goto out_unlock;
 826
 827         if (mem->slot >= kvm->nmemslots)
 828                 kvm->nmemslots = mem->slot + 1;
 829
 830         *memslot = new;
 831         ++kvm->memory_config_version;
 832
 833         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 834         kvm_flush_remote_tlbs(kvm);
 835
 836         mutex_unlock(&kvm->lock);
 837
 838         kvm_free_physmem_slot(&old, &new);
 839         return 0;
 840
 841 out_unlock:
 842         mutex_unlock(&kvm->lock);
 843 out_free:
 844         kvm_free_physmem_slot(&new, &old);
 845 out:
 846         return r;
 847 }
 848
 849 /*
 850  * Get (and clear) the dirty memory log for a memory slot.
 851  */
 852 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 853                                       struct kvm_dirty_log *log)
 854 {
 855         struct kvm_memory_slot *memslot;
 856         int r, i;
 857         int n;
 858         unsigned long any = 0;
 859
 860         mutex_lock(&kvm->lock);
 861
 862         /*
 863          * Prevent changes to guest memory configuration even while the lock
 864          * is not taken.
 865          */
 866         ++kvm->busy;
 867         mutex_unlock(&kvm->lock);
 868         r = -EINVAL;
 869         if (log->slot >= KVM_MEMORY_SLOTS)
 870                 goto out;
 871
 872         memslot = &kvm->memslots[log->slot];
 873         r = -ENOENT;
 874         if (!memslot->dirty_bitmap)
 875                 goto out;
 876
 877         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 878
 879         for (i = 0; !any && i < n/sizeof(long); ++i)
 880                 any = memslot->dirty_bitmap[i];
 881
 882         r = -EFAULT;
 883         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 884                 goto out;
 885
 886         mutex_lock(&kvm->lock);
 887         kvm_mmu_slot_remove_write_access(kvm, log->slot);
 888         kvm_flush_remote_tlbs(kvm);
 889         memset(memslot->dirty_bitmap, 0, n);
 890         mutex_unlock(&kvm->lock);
 891
 892         r = 0;
 893
 894 out:
 895         mutex_lock(&kvm->lock);
 896         --kvm->busy;
 897         mutex_unlock(&kvm->lock);
 898         return r;
 899 }
 900
 901 /*
 902  * Set a new alias region.  Aliases map a portion of physical memory into
 903  * another portion.  This is useful for memory windows, for example the PC
 904  * VGA region.
 905  */
 906 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 907                                          struct kvm_memory_alias *alias)
 908 {
 909         int r, n;
 910         struct kvm_mem_alias *p;
 911
 912         r = -EINVAL;
 913         /* General sanity checks */
 914         if (alias->memory_size & (PAGE_SIZE - 1))
 915                 goto out;
 916         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 917                 goto out;
 918         if (alias->slot >= KVM_ALIAS_SLOTS)
 919                 goto out;
 920         if (alias->guest_phys_addr + alias->memory_size
 921             < alias->guest_phys_addr)
 922                 goto out;
 923         if (alias->target_phys_addr + alias->memory_size
 924             < alias->target_phys_addr)
 925                 goto out;
 926
 927         mutex_lock(&kvm->lock);
 928
 929         p = &kvm->aliases[alias->slot];
 930         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 931         p->npages = alias->memory_size >> PAGE_SHIFT;
 932         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 933
 934         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 935                 if (kvm->aliases[n - 1].npages)
 936                         break;
 937         kvm->naliases = n;
 938
 939         kvm_mmu_zap_all(kvm);
 940
 941         mutex_unlock(&kvm->lock);
 942
 943         return 0;
 944
 945 out:
 946         return r;
 947 }
 948
 949 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 950 {
 951         int i;
 952         struct kvm_mem_alias *alias;
 953
 954         for (i = 0; i < kvm->naliases; ++i) {
 955                 alias = &kvm->aliases[i];
 956                 if (gfn >= alias->base_gfn
 957                     && gfn < alias->base_gfn + alias->npages)
 958                         return alias->target_gfn + gfn - alias->base_gfn;
 959         }
 960         return gfn;
 961 }
 962
 963 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 964 {
 965         int i;
 966
 967         for (i = 0; i < kvm->nmemslots; ++i) {
 968                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 969
 970                 if (gfn >= memslot->base_gfn
 971                     && gfn < memslot->base_gfn + memslot->npages)
 972                         return memslot;
 973         }
 974         return NULL;
 975 }
 976
 977 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 978 {
 979         gfn = unalias_gfn(kvm, gfn);
 980         return __gfn_to_memslot(kvm, gfn);
 981 }
 982
 983 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 984 {
 985         struct kvm_memory_slot *slot;
 986
 987         gfn = unalias_gfn(kvm, gfn);
 988         slot = __gfn_to_memslot(kvm, gfn);
 989         if (!slot)
 990                 return NULL;
 991         return slot->phys_mem[gfn - slot->base_gfn];
 992 }
 993 EXPORT_SYMBOL_GPL(gfn_to_page);
 994
 995 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 996 {
 997         int i;
 998         struct kvm_memory_slot *memslot;
 999         unsigned long rel_gfn;
1000
1001         for (i = 0; i < kvm->nmemslots; ++i) {
1002                 memslot = &kvm->memslots[i];
1003
1004                 if (gfn >= memslot->base_gfn
1005                     && gfn < memslot->base_gfn + memslot->npages) {
1006
1007                         if (!memslot->dirty_bitmap)
1008                                 return;
1009
1010                         rel_gfn = gfn - memslot->base_gfn;
1011
1012                         /* avoid RMW */
1013                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1014                                 set_bit(rel_gfn, memslot->dirty_bitmap);
1015                         return;
1016                 }
1017         }
1018 }
1019
1020 static int emulator_read_std(unsigned long addr,
1021                              void *val,
1022                              unsigned int bytes,
1023                              struct x86_emulate_ctxt *ctxt)
1024 {
1025         struct kvm_vcpu *vcpu = ctxt->vcpu;
1026         void *data = val;
1027
1028         while (bytes) {
1029                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1030                 unsigned offset = addr & (PAGE_SIZE-1);
1031                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1032                 unsigned long pfn;
1033                 struct page *page;
1034                 void *page_virt;
1035
1036                 if (gpa == UNMAPPED_GVA)
1037                         return X86EMUL_PROPAGATE_FAULT;
1038                 pfn = gpa >> PAGE_SHIFT;
1039                 page = gfn_to_page(vcpu->kvm, pfn);
1040                 if (!page)
1041                         return X86EMUL_UNHANDLEABLE;
1042                 page_virt = kmap_atomic(page, KM_USER0);
1043
1044                 memcpy(data, page_virt + offset, tocopy);
1045
1046                 kunmap_atomic(page_virt, KM_USER0);
1047
1048                 bytes -= tocopy;
1049                 data += tocopy;
1050                 addr += tocopy;
1051         }
1052
1053         return X86EMUL_CONTINUE;
1054 }
1055
1056 static int emulator_write_std(unsigned long addr,
1057                               const void *val,
1058                               unsigned int bytes,
1059                               struct x86_emulate_ctxt *ctxt)
1060 {
1061         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
1062                addr, bytes);
1063         return X86EMUL_UNHANDLEABLE;
1064 }
1065
1066 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1067                                                 gpa_t addr)
1068 {
1069         /*
1070          * Note that its important to have this wrapper function because
1071          * in the very near future we will be checking for MMIOs against
1072          * the LAPIC as well as the general MMIO bus
1073          */
1074         return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1075 }
1076
1077 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1078                                                gpa_t addr)
1079 {
1080         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1081 }
1082
1083 static int emulator_read_emulated(unsigned long addr,
1084                                   void *val,
1085                                   unsigned int bytes,
1086                                   struct x86_emulate_ctxt *ctxt)
1087 {
1088         struct kvm_vcpu      *vcpu = ctxt->vcpu;
1089         struct kvm_io_device *mmio_dev;
1090         gpa_t                 gpa;
1091
1092         if (vcpu->mmio_read_completed) {
1093                 memcpy(val, vcpu->mmio_data, bytes);
1094                 vcpu->mmio_read_completed = 0;
1095                 return X86EMUL_CONTINUE;
1096         } else if (emulator_read_std(addr, val, bytes, ctxt)
1097                    == X86EMUL_CONTINUE)
1098                 return X86EMUL_CONTINUE;
1099
1100         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1101         if (gpa == UNMAPPED_GVA)
1102                 return X86EMUL_PROPAGATE_FAULT;
1103
1104         /*
1105          * Is this MMIO handled locally?
1106          */
1107         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1108         if (mmio_dev) {
1109                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1110                 return X86EMUL_CONTINUE;
1111         }
1112
1113         vcpu->mmio_needed = 1;
1114         vcpu->mmio_phys_addr = gpa;
1115         vcpu->mmio_size = bytes;
1116         vcpu->mmio_is_write = 0;
1117
1118         return X86EMUL_UNHANDLEABLE;
1119 }
1120
1121 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1122                                const void *val, int bytes)
1123 {
1124         struct page *page;
1125         void *virt;
1126
1127         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1128                 return 0;
1129         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1130         if (!page)
1131                 return 0;
1132         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1133         virt = kmap_atomic(page, KM_USER0);
1134         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1135         memcpy(virt + offset_in_page(gpa), val, bytes);
1136         kunmap_atomic(virt, KM_USER0);
1137         return 1;
1138 }
1139
1140 static int emulator_write_emulated_onepage(unsigned long addr,
1141                                            const void *val,
1142                                            unsigned int bytes,
1143                                            struct x86_emulate_ctxt *ctxt)
1144 {
1145         struct kvm_vcpu      *vcpu = ctxt->vcpu;
1146         struct kvm_io_device *mmio_dev;
1147         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1148
1149         if (gpa == UNMAPPED_GVA) {
1150                 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1151                 return X86EMUL_PROPAGATE_FAULT;
1152         }
1153
1154         if (emulator_write_phys(vcpu, gpa, val, bytes))
1155                 return X86EMUL_CONTINUE;
1156
1157         /*
1158          * Is this MMIO handled locally?
1159          */
1160         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1161         if (mmio_dev) {
1162                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1163                 return X86EMUL_CONTINUE;
1164         }
1165
1166         vcpu->mmio_needed = 1;
1167         vcpu->mmio_phys_addr = gpa;
1168         vcpu->mmio_size = bytes;
1169         vcpu->mmio_is_write = 1;
1170         memcpy(vcpu->mmio_data, val, bytes);
1171
1172         return X86EMUL_CONTINUE;
1173 }
1174
1175 static int emulator_write_emulated(unsigned long addr,
1176                                    const void *val,
1177                                    unsigned int bytes,
1178                                    struct x86_emulate_ctxt *ctxt)
1179 {
1180         /* Crossing a page boundary? */
1181         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1182                 int rc, now;
1183
1184                 now = -addr & ~PAGE_MASK;
1185                 rc = emulator_write_emulated_onepage(addr, val, now, ctxt);
1186                 if (rc != X86EMUL_CONTINUE)
1187                         return rc;
1188                 addr += now;
1189                 val += now;
1190                 bytes -= now;
1191         }
1192         return emulator_write_emulated_onepage(addr, val, bytes, ctxt);
1193 }
1194
1195 static int emulator_cmpxchg_emulated(unsigned long addr,
1196                                      const void *old,
1197                                      const void *new,
1198                                      unsigned int bytes,
1199                                      struct x86_emulate_ctxt *ctxt)
1200 {
1201         static int reported;
1202
1203         if (!reported) {
1204                 reported = 1;
1205                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1206         }
1207         return emulator_write_emulated(addr, new, bytes, ctxt);
1208 }
1209
1210 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1211 {
1212         return kvm_arch_ops->get_segment_base(vcpu, seg);
1213 }
1214
1215 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1216 {
1217         return X86EMUL_CONTINUE;
1218 }
1219
1220 int emulate_clts(struct kvm_vcpu *vcpu)
1221 {
1222         unsigned long cr0;
1223
1224         cr0 = vcpu->cr0 & ~X86_CR0_TS;
1225         kvm_arch_ops->set_cr0(vcpu, cr0);
1226         return X86EMUL_CONTINUE;
1227 }
1228
1229 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1230 {
1231         struct kvm_vcpu *vcpu = ctxt->vcpu;
1232
1233         switch (dr) {
1234         case 0 ... 3:
1235                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1236                 return X86EMUL_CONTINUE;
1237         default:
1238                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1239                        __FUNCTION__, dr);
1240                 return X86EMUL_UNHANDLEABLE;
1241         }
1242 }
1243
1244 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1245 {
1246         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1247         int exception;
1248
1249         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1250         if (exception) {
1251                 /* FIXME: better handling */
1252                 return X86EMUL_UNHANDLEABLE;
1253         }
1254         return X86EMUL_CONTINUE;
1255 }
1256
1257 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1258 {
1259         static int reported;
1260         u8 opcodes[4];
1261         unsigned long rip = ctxt->vcpu->rip;
1262         unsigned long rip_linear;
1263
1264         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1265
1266         if (reported)
1267                 return;
1268
1269         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1270
1271         printk(KERN_ERR "emulation failed but !mmio_needed?"
1272                " rip %lx %02x %02x %02x %02x\n",
1273                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1274         reported = 1;
1275 }
1276
1277 struct x86_emulate_ops emulate_ops = {
1278         .read_std            = emulator_read_std,
1279         .write_std           = emulator_write_std,
1280         .read_emulated       = emulator_read_emulated,
1281         .write_emulated      = emulator_write_emulated,
1282         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1283 };
1284
1285 int emulate_instruction(struct kvm_vcpu *vcpu,
1286                         struct kvm_run *run,
1287                         unsigned long cr2,
1288                         u16 error_code)
1289 {
1290         struct x86_emulate_ctxt emulate_ctxt;
1291         int r;
1292         int cs_db, cs_l;
1293
1294         vcpu->mmio_fault_cr2 = cr2;
1295         kvm_arch_ops->cache_regs(vcpu);
1296
1297         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1298
1299         emulate_ctxt.vcpu = vcpu;
1300         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1301         emulate_ctxt.cr2 = cr2;
1302         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1303                 ? X86EMUL_MODE_REAL : cs_l
1304                 ? X86EMUL_MODE_PROT64 : cs_db
1305                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1306
1307         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1308                 emulate_ctxt.cs_base = 0;
1309                 emulate_ctxt.ds_base = 0;
1310                 emulate_ctxt.es_base = 0;
1311                 emulate_ctxt.ss_base = 0;
1312         } else {
1313                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1314                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1315                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1316                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1317         }
1318
1319         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1320         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1321
1322         vcpu->mmio_is_write = 0;
1323         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1324
1325         if ((r || vcpu->mmio_is_write) && run) {
1326                 run->exit_reason = KVM_EXIT_MMIO;
1327                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1328                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1329                 run->mmio.len = vcpu->mmio_size;
1330                 run->mmio.is_write = vcpu->mmio_is_write;
1331         }
1332
1333         if (r) {
1334                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1335                         return EMULATE_DONE;
1336                 if (!vcpu->mmio_needed) {
1337                         report_emulation_failure(&emulate_ctxt);
1338                         return EMULATE_FAIL;
1339                 }
1340                 return EMULATE_DO_MMIO;
1341         }
1342
1343         kvm_arch_ops->decache_regs(vcpu);
1344         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1345
1346         if (vcpu->mmio_is_write) {
1347                 vcpu->mmio_needed = 0;
1348                 return EMULATE_DO_MMIO;
1349         }
1350
1351         return EMULATE_DONE;
1352 }
1353 EXPORT_SYMBOL_GPL(emulate_instruction);
1354
1355 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1356 {
1357         if (vcpu->irq_summary)
1358                 return 1;
1359
1360         vcpu->run->exit_reason = KVM_EXIT_HLT;
1361         ++vcpu->stat.halt_exits;
1362         return 0;
1363 }
1364 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1365
1366 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1367 {
1368         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1369
1370         kvm_arch_ops->cache_regs(vcpu);
1371         ret = -KVM_EINVAL;
1372 #ifdef CONFIG_X86_64
1373         if (is_long_mode(vcpu)) {
1374                 nr = vcpu->regs[VCPU_REGS_RAX];
1375                 a0 = vcpu->regs[VCPU_REGS_RDI];
1376                 a1 = vcpu->regs[VCPU_REGS_RSI];
1377                 a2 = vcpu->regs[VCPU_REGS_RDX];
1378                 a3 = vcpu->regs[VCPU_REGS_RCX];
1379                 a4 = vcpu->regs[VCPU_REGS_R8];
1380                 a5 = vcpu->regs[VCPU_REGS_R9];
1381         } else
1382 #endif
1383         {
1384                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1385                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1386                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1387                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1388                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1389                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1390                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1391         }
1392         switch (nr) {
1393         default:
1394                 run->hypercall.nr = nr;
1395                 run->hypercall.args[0] = a0;
1396                 run->hypercall.args[1] = a1;
1397                 run->hypercall.args[2] = a2;
1398                 run->hypercall.args[3] = a3;
1399                 run->hypercall.args[4] = a4;
1400                 run->hypercall.args[5] = a5;
1401                 run->hypercall.ret = ret;
1402                 run->hypercall.longmode = is_long_mode(vcpu);
1403                 kvm_arch_ops->decache_regs(vcpu);
1404                 return 0;
1405         }
1406         vcpu->regs[VCPU_REGS_RAX] = ret;
1407         kvm_arch_ops->decache_regs(vcpu);
1408         return 1;
1409 }
1410 EXPORT_SYMBOL_GPL(kvm_hypercall);
1411
1412 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1413 {
1414         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1415 }
1416
1417 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1418 {
1419         struct descriptor_table dt = { limit, base };
1420
1421         kvm_arch_ops->set_gdt(vcpu, &dt);
1422 }
1423
1424 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1425 {
1426         struct descriptor_table dt = { limit, base };
1427
1428         kvm_arch_ops->set_idt(vcpu, &dt);
1429 }
1430
1431 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1432                    unsigned long *rflags)
1433 {
1434         lmsw(vcpu, msw);
1435         *rflags = kvm_arch_ops->get_rflags(vcpu);
1436 }
1437
1438 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1439 {
1440         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1441         switch (cr) {
1442         case 0:
1443                 return vcpu->cr0;
1444         case 2:
1445                 return vcpu->cr2;
1446         case 3:
1447                 return vcpu->cr3;
1448         case 4:
1449                 return vcpu->cr4;
1450         default:
1451                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1452                 return 0;
1453         }
1454 }
1455
1456 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1457                      unsigned long *rflags)
1458 {
1459         switch (cr) {
1460         case 0:
1461                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1462                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1463                 break;
1464         case 2:
1465                 vcpu->cr2 = val;
1466                 break;
1467         case 3:
1468                 set_cr3(vcpu, val);
1469                 break;
1470         case 4:
1471                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1472                 break;
1473         default:
1474                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1475         }
1476 }
1477
1478 /*
1479  * Register the para guest with the host:
1480  */
1481 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1482 {
1483         struct kvm_vcpu_para_state *para_state;
1484         hpa_t para_state_hpa, hypercall_hpa;
1485         struct page *para_state_page;
1486         unsigned char *hypercall;
1487         gpa_t hypercall_gpa;
1488
1489         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1490         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1491
1492         /*
1493          * Needs to be page aligned:
1494          */
1495         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1496                 goto err_gp;
1497
1498         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1499         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1500         if (is_error_hpa(para_state_hpa))
1501                 goto err_gp;
1502
1503         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1504         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1505         para_state = kmap(para_state_page);
1506
1507         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1508         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1509
1510         para_state->host_version = KVM_PARA_API_VERSION;
1511         /*
1512          * We cannot support guests that try to register themselves
1513          * with a newer API version than the host supports:
1514          */
1515         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1516                 para_state->ret = -KVM_EINVAL;
1517                 goto err_kunmap_skip;
1518         }
1519
1520         hypercall_gpa = para_state->hypercall_gpa;
1521         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1522         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1523         if (is_error_hpa(hypercall_hpa)) {
1524                 para_state->ret = -KVM_EINVAL;
1525                 goto err_kunmap_skip;
1526         }
1527
1528         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1529         vcpu->para_state_page = para_state_page;
1530         vcpu->para_state_gpa = para_state_gpa;
1531         vcpu->hypercall_gpa = hypercall_gpa;
1532
1533         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1534         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1535                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1536         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1537         kunmap_atomic(hypercall, KM_USER1);
1538
1539         para_state->ret = 0;
1540 err_kunmap_skip:
1541         kunmap(para_state_page);
1542         return 0;
1543 err_gp:
1544         return 1;
1545 }
1546
1547 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1548 {
1549         u64 data;
1550
1551         switch (msr) {
1552         case 0xc0010010: /* SYSCFG */
1553         case 0xc0010015: /* HWCR */
1554         case MSR_IA32_PLATFORM_ID:
1555         case MSR_IA32_P5_MC_ADDR:
1556         case MSR_IA32_P5_MC_TYPE:
1557         case MSR_IA32_MC0_CTL:
1558         case MSR_IA32_MCG_STATUS:
1559         case MSR_IA32_MCG_CAP:
1560         case MSR_IA32_MC0_MISC:
1561         case MSR_IA32_MC0_MISC+4:
1562         case MSR_IA32_MC0_MISC+8:
1563         case MSR_IA32_MC0_MISC+12:
1564         case MSR_IA32_MC0_MISC+16:
1565         case MSR_IA32_UCODE_REV:
1566         case MSR_IA32_PERF_STATUS:
1567         case MSR_IA32_EBL_CR_POWERON:
1568                 /* MTRR registers */
1569         case 0xfe:
1570         case 0x200 ... 0x2ff:
1571                 data = 0;
1572                 break;
1573         case 0xcd: /* fsb frequency */
1574                 data = 3;
1575                 break;
1576         case MSR_IA32_APICBASE:
1577                 data = vcpu->apic_base;
1578                 break;
1579         case MSR_IA32_MISC_ENABLE:
1580                 data = vcpu->ia32_misc_enable_msr;
1581                 break;
1582 #ifdef CONFIG_X86_64
1583         case MSR_EFER:
1584                 data = vcpu->shadow_efer;
1585                 break;
1586 #endif
1587         default:
1588                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1589                 return 1;
1590         }
1591         *pdata = data;
1592         return 0;
1593 }
1594 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1595
1596 /*
1597  * Reads an msr value (of 'msr_index') into 'pdata'.
1598  * Returns 0 on success, non-0 otherwise.
1599  * Assumes vcpu_load() was already called.
1600  */
1601 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1602 {
1603         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1604 }
1605
1606 #ifdef CONFIG_X86_64
1607
1608 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1609 {
1610         if (efer & EFER_RESERVED_BITS) {
1611                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1612                        efer);
1613                 inject_gp(vcpu);
1614                 return;
1615         }
1616
1617         if (is_paging(vcpu)
1618             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1619                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1620                 inject_gp(vcpu);
1621                 return;
1622         }
1623
1624         kvm_arch_ops->set_efer(vcpu, efer);
1625
1626         efer &= ~EFER_LMA;
1627         efer |= vcpu->shadow_efer & EFER_LMA;
1628
1629         vcpu->shadow_efer = efer;
1630 }
1631
1632 #endif
1633
1634 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1635 {
1636         switch (msr) {
1637 #ifdef CONFIG_X86_64
1638         case MSR_EFER:
1639                 set_efer(vcpu, data);
1640                 break;
1641 #endif
1642         case MSR_IA32_MC0_STATUS:
1643                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1644                        __FUNCTION__, data);
1645                 break;
1646         case MSR_IA32_MCG_STATUS:
1647                 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1648                         __FUNCTION__, data);
1649                 break;
1650         case MSR_IA32_UCODE_REV:
1651         case MSR_IA32_UCODE_WRITE:
1652         case 0x200 ... 0x2ff: /* MTRRs */
1653                 break;
1654         case MSR_IA32_APICBASE:
1655                 vcpu->apic_base = data;
1656                 break;
1657         case MSR_IA32_MISC_ENABLE:
1658                 vcpu->ia32_misc_enable_msr = data;
1659                 break;
1660         /*
1661          * This is the 'probe whether the host is KVM' logic:
1662          */
1663         case MSR_KVM_API_MAGIC:
1664                 return vcpu_register_para(vcpu, data);
1665
1666         default:
1667                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1668                 return 1;
1669         }
1670         return 0;
1671 }
1672 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1673
1674 /*
1675  * Writes msr value into into the appropriate "register".
1676  * Returns 0 on success, non-0 otherwise.
1677  * Assumes vcpu_load() was already called.
1678  */
1679 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1680 {
1681         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1682 }
1683
1684 void kvm_resched(struct kvm_vcpu *vcpu)
1685 {
1686         if (!need_resched())
1687                 return;
1688         cond_resched();
1689 }
1690 EXPORT_SYMBOL_GPL(kvm_resched);
1691
1692 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1693 {
1694         int i;
1695         u32 function;
1696         struct kvm_cpuid_entry *e, *best;
1697
1698         kvm_arch_ops->cache_regs(vcpu);
1699         function = vcpu->regs[VCPU_REGS_RAX];
1700         vcpu->regs[VCPU_REGS_RAX] = 0;
1701         vcpu->regs[VCPU_REGS_RBX] = 0;
1702         vcpu->regs[VCPU_REGS_RCX] = 0;
1703         vcpu->regs[VCPU_REGS_RDX] = 0;
1704         best = NULL;
1705         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1706                 e = &vcpu->cpuid_entries[i];
1707                 if (e->function == function) {
1708                         best = e;
1709                         break;
1710                 }
1711                 /*
1712                  * Both basic or both extended?
1713                  */
1714                 if (((e->function ^ function) & 0x80000000) == 0)
1715                         if (!best || e->function > best->function)
1716                                 best = e;
1717         }
1718         if (best) {
1719                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1720                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1721                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1722                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1723         }
1724         kvm_arch_ops->decache_regs(vcpu);
1725         kvm_arch_ops->skip_emulated_instruction(vcpu);
1726 }
1727 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1728
1729 static int pio_copy_data(struct kvm_vcpu *vcpu)
1730 {
1731         void *p = vcpu->pio_data;
1732         void *q;
1733         unsigned bytes;
1734         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1735
1736         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1737                  PAGE_KERNEL);
1738         if (!q) {
1739                 free_pio_guest_pages(vcpu);
1740                 return -ENOMEM;
1741         }
1742         q += vcpu->pio.guest_page_offset;
1743         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1744         if (vcpu->pio.in)
1745                 memcpy(q, p, bytes);
1746         else
1747                 memcpy(p, q, bytes);
1748         q -= vcpu->pio.guest_page_offset;
1749         vunmap(q);
1750         free_pio_guest_pages(vcpu);
1751         return 0;
1752 }
1753
1754 static int complete_pio(struct kvm_vcpu *vcpu)
1755 {
1756         struct kvm_pio_request *io = &vcpu->pio;
1757         long delta;
1758         int r;
1759
1760         kvm_arch_ops->cache_regs(vcpu);
1761
1762         if (!io->string) {
1763                 if (io->in)
1764                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1765                                io->size);
1766         } else {
1767                 if (io->in) {
1768                         r = pio_copy_data(vcpu);
1769                         if (r) {
1770                                 kvm_arch_ops->cache_regs(vcpu);
1771                                 return r;
1772                         }
1773                 }
1774
1775                 delta = 1;
1776                 if (io->rep) {
1777                         delta *= io->cur_count;
1778                         /*
1779                          * The size of the register should really depend on
1780                          * current address size.
1781                          */
1782                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1783                 }
1784                 if (io->down)
1785                         delta = -delta;
1786                 delta *= io->size;
1787                 if (io->in)
1788                         vcpu->regs[VCPU_REGS_RDI] += delta;
1789                 else
1790                         vcpu->regs[VCPU_REGS_RSI] += delta;
1791         }
1792
1793         kvm_arch_ops->decache_regs(vcpu);
1794
1795         io->count -= io->cur_count;
1796         io->cur_count = 0;
1797
1798         if (!io->count)
1799                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1800         return 0;
1801 }
1802
1803 static void kernel_pio(struct kvm_io_device *pio_dev,
1804                        struct kvm_vcpu *vcpu,
1805                        void *pd)
1806 {
1807         /* TODO: String I/O for in kernel device */
1808
1809         if (vcpu->pio.in)
1810                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1811                                   vcpu->pio.size,
1812                                   pd);
1813         else
1814                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1815                                    vcpu->pio.size,
1816                                    pd);
1817 }
1818
1819 static void pio_string_write(struct kvm_io_device *pio_dev,
1820                              struct kvm_vcpu *vcpu)
1821 {
1822         struct kvm_pio_request *io = &vcpu->pio;
1823         void *pd = vcpu->pio_data;
1824         int i;
1825
1826         for (i = 0; i < io->cur_count; i++) {
1827                 kvm_iodevice_write(pio_dev, io->port,
1828                                    io->size,
1829                                    pd);
1830                 pd += io->size;
1831         }
1832 }
1833
1834 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1835                   int size, unsigned long count, int string, int down,
1836                   gva_t address, int rep, unsigned port)
1837 {
1838         unsigned now, in_page;
1839         int i, ret = 0;
1840         int nr_pages = 1;
1841         struct page *page;
1842         struct kvm_io_device *pio_dev;
1843
1844         vcpu->run->exit_reason = KVM_EXIT_IO;
1845         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1846         vcpu->run->io.size = size;
1847         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1848         vcpu->run->io.count = count;
1849         vcpu->run->io.port = port;
1850         vcpu->pio.count = count;
1851         vcpu->pio.cur_count = count;
1852         vcpu->pio.size = size;
1853         vcpu->pio.in = in;
1854         vcpu->pio.port = port;
1855         vcpu->pio.string = string;
1856         vcpu->pio.down = down;
1857         vcpu->pio.guest_page_offset = offset_in_page(address);
1858         vcpu->pio.rep = rep;
1859
1860         pio_dev = vcpu_find_pio_dev(vcpu, port);
1861         if (!string) {
1862                 kvm_arch_ops->cache_regs(vcpu);
1863                 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1864                 kvm_arch_ops->decache_regs(vcpu);
1865                 if (pio_dev) {
1866                         kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1867                         complete_pio(vcpu);
1868                         return 1;
1869                 }
1870                 return 0;
1871         }
1872
1873         if (!count) {
1874                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1875                 return 1;
1876         }
1877
1878         now = min(count, PAGE_SIZE / size);
1879
1880         if (!down)
1881                 in_page = PAGE_SIZE - offset_in_page(address);
1882         else
1883                 in_page = offset_in_page(address) + size;
1884         now = min(count, (unsigned long)in_page / size);
1885         if (!now) {
1886                 /*
1887                  * String I/O straddles page boundary.  Pin two guest pages
1888                  * so that we satisfy atomicity constraints.  Do just one
1889                  * transaction to avoid complexity.
1890                  */
1891                 nr_pages = 2;
1892                 now = 1;
1893         }
1894         if (down) {
1895                 /*
1896                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1897                  */
1898                 printk(KERN_ERR "kvm: guest string pio down\n");
1899                 inject_gp(vcpu);
1900                 return 1;
1901         }
1902         vcpu->run->io.count = now;
1903         vcpu->pio.cur_count = now;
1904
1905         for (i = 0; i < nr_pages; ++i) {
1906                 mutex_lock(&vcpu->kvm->lock);
1907                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1908                 if (page)
1909                         get_page(page);
1910                 vcpu->pio.guest_pages[i] = page;
1911                 mutex_unlock(&vcpu->kvm->lock);
1912                 if (!page) {
1913                         inject_gp(vcpu);
1914                         free_pio_guest_pages(vcpu);
1915                         return 1;
1916                 }
1917         }
1918
1919         if (!vcpu->pio.in) {
1920                 /* string PIO write */
1921                 ret = pio_copy_data(vcpu);
1922                 if (ret >= 0 && pio_dev) {
1923                         pio_string_write(pio_dev, vcpu);
1924                         complete_pio(vcpu);
1925                         if (vcpu->pio.count == 0)
1926                                 ret = 1;
1927                 }
1928         } else if (pio_dev)
1929                 printk(KERN_ERR "no string pio read support yet, "
1930                        "port %x size %d count %ld\n",
1931                         port, size, count);
1932
1933         return ret;
1934 }
1935 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1936
1937 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1938 {
1939         int r;
1940         sigset_t sigsaved;
1941
1942         vcpu_load(vcpu);
1943
1944         if (vcpu->sigset_active)
1945                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1946
1947         /* re-sync apic's tpr */
1948         vcpu->cr8 = kvm_run->cr8;
1949
1950         if (vcpu->pio.cur_count) {
1951                 r = complete_pio(vcpu);
1952                 if (r)
1953                         goto out;
1954         }
1955
1956         if (vcpu->mmio_needed) {
1957                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1958                 vcpu->mmio_read_completed = 1;
1959                 vcpu->mmio_needed = 0;
1960                 r = emulate_instruction(vcpu, kvm_run,
1961                                         vcpu->mmio_fault_cr2, 0);
1962                 if (r == EMULATE_DO_MMIO) {
1963                         /*
1964                          * Read-modify-write.  Back to userspace.
1965                          */
1966                         r = 0;
1967                         goto out;
1968                 }
1969         }
1970
1971         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1972                 kvm_arch_ops->cache_regs(vcpu);
1973                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1974                 kvm_arch_ops->decache_regs(vcpu);
1975         }
1976
1977         r = kvm_arch_ops->run(vcpu, kvm_run);
1978
1979 out:
1980         if (vcpu->sigset_active)
1981                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1982
1983         vcpu_put(vcpu);
1984         return r;
1985 }
1986
1987 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1988                                    struct kvm_regs *regs)
1989 {
1990         vcpu_load(vcpu);
1991
1992         kvm_arch_ops->cache_regs(vcpu);
1993
1994         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1995         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1996         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1997         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1998         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1999         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2000         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2001         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2002 #ifdef CONFIG_X86_64
2003         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2004         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2005         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2006         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2007         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2008         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2009         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2010         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2011 #endif
2012
2013         regs->rip = vcpu->rip;
2014         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
2015
2016         /*
2017          * Don't leak debug flags in case they were set for guest debugging
2018          */
2019         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2020                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2021
2022         vcpu_put(vcpu);
2023
2024         return 0;
2025 }
2026
2027 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2028                                    struct kvm_regs *regs)
2029 {
2030         vcpu_load(vcpu);
2031
2032         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2033         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2034         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2035         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2036         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2037         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2038         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2039         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2040 #ifdef CONFIG_X86_64
2041         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2042         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2043         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2044         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2045         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2046         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2047         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2048         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2049 #endif
2050
2051         vcpu->rip = regs->rip;
2052         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
2053
2054         kvm_arch_ops->decache_regs(vcpu);
2055
2056         vcpu_put(vcpu);
2057
2058         return 0;
2059 }
2060
2061 static void get_segment(struct kvm_vcpu *vcpu,
2062                         struct kvm_segment *var, int seg)
2063 {
2064         return kvm_arch_ops->get_segment(vcpu, var, seg);
2065 }
2066
2067 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2068                                     struct kvm_sregs *sregs)
2069 {
2070         struct descriptor_table dt;
2071
2072         vcpu_load(vcpu);
2073
2074         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2075         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2076         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2077         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2078         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2079         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2080
2081         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2082         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2083
2084         kvm_arch_ops->get_idt(vcpu, &dt);
2085         sregs->idt.limit = dt.limit;
2086         sregs->idt.base = dt.base;
2087         kvm_arch_ops->get_gdt(vcpu, &dt);
2088         sregs->gdt.limit = dt.limit;
2089         sregs->gdt.base = dt.base;
2090
2091         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2092         sregs->cr0 = vcpu->cr0;
2093         sregs->cr2 = vcpu->cr2;
2094         sregs->cr3 = vcpu->cr3;
2095         sregs->cr4 = vcpu->cr4;
2096         sregs->cr8 = vcpu->cr8;
2097         sregs->efer = vcpu->shadow_efer;
2098         sregs->apic_base = vcpu->apic_base;
2099
2100         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2101                sizeof sregs->interrupt_bitmap);
2102
2103         vcpu_put(vcpu);
2104
2105         return 0;
2106 }
2107
2108 static void set_segment(struct kvm_vcpu *vcpu,
2109                         struct kvm_segment *var, int seg)
2110 {
2111         return kvm_arch_ops->set_segment(vcpu, var, seg);
2112 }
2113
2114 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2115                                     struct kvm_sregs *sregs)
2116 {
2117         int mmu_reset_needed = 0;
2118         int i;
2119         struct descriptor_table dt;
2120
2121         vcpu_load(vcpu);
2122
2123         dt.limit = sregs->idt.limit;
2124         dt.base = sregs->idt.base;
2125         kvm_arch_ops->set_idt(vcpu, &dt);
2126         dt.limit = sregs->gdt.limit;
2127         dt.base = sregs->gdt.base;
2128         kvm_arch_ops->set_gdt(vcpu, &dt);
2129
2130         vcpu->cr2 = sregs->cr2;
2131         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2132         vcpu->cr3 = sregs->cr3;
2133
2134         vcpu->cr8 = sregs->cr8;
2135
2136         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2137 #ifdef CONFIG_X86_64
2138         kvm_arch_ops->set_efer(vcpu, sregs->efer);
2139 #endif
2140         vcpu->apic_base = sregs->apic_base;
2141
2142         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2143
2144         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2145         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2146
2147         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2148         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2149         if (!is_long_mode(vcpu) && is_pae(vcpu))
2150                 load_pdptrs(vcpu, vcpu->cr3);
2151
2152         if (mmu_reset_needed)
2153                 kvm_mmu_reset_context(vcpu);
2154
2155         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2156                sizeof vcpu->irq_pending);
2157         vcpu->irq_summary = 0;
2158         for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2159                 if (vcpu->irq_pending[i])
2160                         __set_bit(i, &vcpu->irq_summary);
2161
2162         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2163         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2164         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2165         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2166         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2167         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2168
2169         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2170         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2171
2172         vcpu_put(vcpu);
2173
2174         return 0;
2175 }
2176
2177 /*
2178  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2179  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2180  *
2181  * This list is modified at module load time to reflect the
2182  * capabilities of the host cpu.
2183  */
2184 static u32 msrs_to_save[] = {
2185         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2186         MSR_K6_STAR,
2187 #ifdef CONFIG_X86_64
2188         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2189 #endif
2190         MSR_IA32_TIME_STAMP_COUNTER,
2191 };
2192
2193 static unsigned num_msrs_to_save;
2194
2195 static u32 emulated_msrs[] = {
2196         MSR_IA32_MISC_ENABLE,
2197 };
2198
2199 static __init void kvm_init_msr_list(void)
2200 {
2201         u32 dummy[2];
2202         unsigned i, j;
2203
2204         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2205                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2206                         continue;
2207                 if (j < i)
2208                         msrs_to_save[j] = msrs_to_save[i];
2209                 j++;
2210         }
2211         num_msrs_to_save = j;
2212 }
2213
2214 /*
2215  * Adapt set_msr() to msr_io()'s calling convention
2216  */
2217 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2218 {
2219         return kvm_set_msr(vcpu, index, *data);
2220 }
2221
2222 /*
2223  * Read or write a bunch of msrs. All parameters are kernel addresses.
2224  *
2225  * @return number of msrs set successfully.
2226  */
2227 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2228                     struct kvm_msr_entry *entries,
2229                     int (*do_msr)(struct kvm_vcpu *vcpu,
2230                                   unsigned index, u64 *data))
2231 {
2232         int i;
2233
2234         vcpu_load(vcpu);
2235
2236         for (i = 0; i < msrs->nmsrs; ++i)
2237                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2238                         break;
2239
2240         vcpu_put(vcpu);
2241
2242         return i;
2243 }
2244
2245 /*
2246  * Read or write a bunch of msrs. Parameters are user addresses.
2247  *
2248  * @return number of msrs set successfully.
2249  */
2250 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2251                   int (*do_msr)(struct kvm_vcpu *vcpu,
2252                                 unsigned index, u64 *data),
2253                   int writeback)
2254 {
2255         struct kvm_msrs msrs;
2256         struct kvm_msr_entry *entries;
2257         int r, n;
2258         unsigned size;
2259
2260         r = -EFAULT;
2261         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2262                 goto out;
2263
2264         r = -E2BIG;
2265         if (msrs.nmsrs >= MAX_IO_MSRS)
2266                 goto out;
2267
2268         r = -ENOMEM;
2269         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2270         entries = vmalloc(size);
2271         if (!entries)
2272                 goto out;
2273
2274         r = -EFAULT;
2275         if (copy_from_user(entries, user_msrs->entries, size))
2276                 goto out_free;
2277
2278         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2279         if (r < 0)
2280                 goto out_free;
2281
2282         r = -EFAULT;
2283         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2284                 goto out_free;
2285
2286         r = n;
2287
2288 out_free:
2289         vfree(entries);
2290 out:
2291         return r;
2292 }
2293
2294 /*
2295  * Translate a guest virtual address to a guest physical address.
2296  */
2297 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2298                                     struct kvm_translation *tr)
2299 {
2300         unsigned long vaddr = tr->linear_address;
2301         gpa_t gpa;
2302
2303         vcpu_load(vcpu);
2304         mutex_lock(&vcpu->kvm->lock);
2305         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2306         tr->physical_address = gpa;
2307         tr->valid = gpa != UNMAPPED_GVA;
2308         tr->writeable = 1;
2309         tr->usermode = 0;
2310         mutex_unlock(&vcpu->kvm->lock);
2311         vcpu_put(vcpu);
2312
2313         return 0;
2314 }
2315
2316 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2317                                     struct kvm_interrupt *irq)
2318 {
2319         if (irq->irq < 0 || irq->irq >= 256)
2320                 return -EINVAL;
2321         vcpu_load(vcpu);
2322
2323         set_bit(irq->irq, vcpu->irq_pending);
2324         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2325
2326         vcpu_put(vcpu);
2327
2328         return 0;
2329 }
2330
2331 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2332                                       struct kvm_debug_guest *dbg)
2333 {
2334         int r;
2335
2336         vcpu_load(vcpu);
2337
2338         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2339
2340         vcpu_put(vcpu);
2341
2342         return r;
2343 }
2344
2345 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2346                                     unsigned long address,
2347                                     int *type)
2348 {
2349         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2350         unsigned long pgoff;
2351         struct page *page;
2352
2353         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2354         if (pgoff == 0)
2355                 page = virt_to_page(vcpu->run);
2356         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2357                 page = virt_to_page(vcpu->pio_data);
2358         else
2359                 return NOPAGE_SIGBUS;
2360         get_page(page);
2361         if (type != NULL)
2362                 *type = VM_FAULT_MINOR;
2363
2364         return page;
2365 }
2366
2367 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2368         .nopage = kvm_vcpu_nopage,
2369 };
2370
2371 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2372 {
2373         vma->vm_ops = &kvm_vcpu_vm_ops;
2374         return 0;
2375 }
2376
2377 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2378 {
2379         struct kvm_vcpu *vcpu = filp->private_data;
2380
2381         fput(vcpu->kvm->filp);
2382         return 0;
2383 }
2384
2385 static struct file_operations kvm_vcpu_fops = {
2386         .release        = kvm_vcpu_release,
2387         .unlocked_ioctl = kvm_vcpu_ioctl,
2388         .compat_ioctl   = kvm_vcpu_ioctl,
2389         .mmap           = kvm_vcpu_mmap,
2390 };
2391
2392 /*
2393  * Allocates an inode for the vcpu.
2394  */
2395 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2396 {
2397         int fd, r;
2398         struct inode *inode;
2399         struct file *file;
2400
2401         r = anon_inode_getfd(&fd, &inode, &file,
2402                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2403         if (r)
2404                 return r;
2405         atomic_inc(&vcpu->kvm->filp->f_count);
2406         return fd;
2407 }
2408
2409 /*
2410  * Creates some virtual cpus.  Good luck creating more than one.
2411  */
2412 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2413 {
2414         int r;
2415         struct kvm_vcpu *vcpu;
2416
2417         if (!valid_vcpu(n))
2418                 return -EINVAL;
2419
2420         vcpu = kvm_arch_ops->vcpu_create(kvm, n);
2421         if (IS_ERR(vcpu))
2422                 return PTR_ERR(vcpu);
2423
2424         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2425
2426         vcpu_load(vcpu);
2427         r = kvm_mmu_setup(vcpu);
2428         vcpu_put(vcpu);
2429         if (r < 0)
2430                 goto free_vcpu;
2431
2432         mutex_lock(&kvm->lock);
2433         if (kvm->vcpus[n]) {
2434                 r = -EEXIST;
2435                 mutex_unlock(&kvm->lock);
2436                 goto mmu_unload;
2437         }
2438         kvm->vcpus[n] = vcpu;
2439         mutex_unlock(&kvm->lock);
2440
2441         /* Now it's all set up, let userspace reach it */
2442         r = create_vcpu_fd(vcpu);
2443         if (r < 0)
2444                 goto unlink;
2445         return r;
2446
2447 unlink:
2448         mutex_lock(&kvm->lock);
2449         kvm->vcpus[n] = NULL;
2450         mutex_unlock(&kvm->lock);
2451
2452 mmu_unload:
2453         vcpu_load(vcpu);
2454         kvm_mmu_unload(vcpu);
2455         vcpu_put(vcpu);
2456
2457 free_vcpu:
2458         kvm_arch_ops->vcpu_free(vcpu);
2459         return r;
2460 }
2461
2462 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2463 {
2464         u64 efer;
2465         int i;
2466         struct kvm_cpuid_entry *e, *entry;
2467
2468         rdmsrl(MSR_EFER, efer);
2469         entry = NULL;
2470         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2471                 e = &vcpu->cpuid_entries[i];
2472                 if (e->function == 0x80000001) {
2473                         entry = e;
2474                         break;
2475                 }
2476         }
2477         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2478                 entry->edx &= ~(1 << 20);
2479                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2480         }
2481 }
2482
2483 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2484                                     struct kvm_cpuid *cpuid,
2485                                     struct kvm_cpuid_entry __user *entries)
2486 {
2487         int r;
2488
2489         r = -E2BIG;
2490         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2491                 goto out;
2492         r = -EFAULT;
2493         if (copy_from_user(&vcpu->cpuid_entries, entries,
2494                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2495                 goto out;
2496         vcpu->cpuid_nent = cpuid->nent;
2497         cpuid_fix_nx_cap(vcpu);
2498         return 0;
2499
2500 out:
2501         return r;
2502 }
2503
2504 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2505 {
2506         if (sigset) {
2507                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2508                 vcpu->sigset_active = 1;
2509                 vcpu->sigset = *sigset;
2510         } else
2511                 vcpu->sigset_active = 0;
2512         return 0;
2513 }
2514
2515 /*
2516  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2517  * we have asm/x86/processor.h
2518  */
2519 struct fxsave {
2520         u16     cwd;
2521         u16     swd;
2522         u16     twd;
2523         u16     fop;
2524         u64     rip;
2525         u64     rdp;
2526         u32     mxcsr;
2527         u32     mxcsr_mask;
2528         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2529 #ifdef CONFIG_X86_64
2530         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2531 #else
2532         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2533 #endif
2534 };
2535
2536 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2537 {
2538         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2539
2540         vcpu_load(vcpu);
2541
2542         memcpy(fpu->fpr, fxsave->st_space, 128);
2543         fpu->fcw = fxsave->cwd;
2544         fpu->fsw = fxsave->swd;
2545         fpu->ftwx = fxsave->twd;
2546         fpu->last_opcode = fxsave->fop;
2547         fpu->last_ip = fxsave->rip;
2548         fpu->last_dp = fxsave->rdp;
2549         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2550
2551         vcpu_put(vcpu);
2552
2553         return 0;
2554 }
2555
2556 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2557 {
2558         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2559
2560         vcpu_load(vcpu);
2561
2562         memcpy(fxsave->st_space, fpu->fpr, 128);
2563         fxsave->cwd = fpu->fcw;
2564         fxsave->swd = fpu->fsw;
2565         fxsave->twd = fpu->ftwx;
2566         fxsave->fop = fpu->last_opcode;
2567         fxsave->rip = fpu->last_ip;
2568         fxsave->rdp = fpu->last_dp;
2569         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2570
2571         vcpu_put(vcpu);
2572
2573         return 0;
2574 }
2575
2576 static long kvm_vcpu_ioctl(struct file *filp,
2577                            unsigned int ioctl, unsigned long arg)
2578 {
2579         struct kvm_vcpu *vcpu = filp->private_data;
2580         void __user *argp = (void __user *)arg;
2581         int r = -EINVAL;
2582
2583         switch (ioctl) {
2584         case KVM_RUN:
2585                 r = -EINVAL;
2586                 if (arg)
2587                         goto out;
2588                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2589                 break;
2590         case KVM_GET_REGS: {
2591                 struct kvm_regs kvm_regs;
2592
2593                 memset(&kvm_regs, 0, sizeof kvm_regs);
2594                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2595                 if (r)
2596                         goto out;
2597                 r = -EFAULT;
2598                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2599                         goto out;
2600                 r = 0;
2601                 break;
2602         }
2603         case KVM_SET_REGS: {
2604                 struct kvm_regs kvm_regs;
2605
2606                 r = -EFAULT;
2607                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2608                         goto out;
2609                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2610                 if (r)
2611                         goto out;
2612                 r = 0;
2613                 break;
2614         }
2615         case KVM_GET_SREGS: {
2616                 struct kvm_sregs kvm_sregs;
2617
2618                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2619                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2620                 if (r)
2621                         goto out;
2622                 r = -EFAULT;
2623                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2624                         goto out;
2625                 r = 0;
2626                 break;
2627         }
2628         case KVM_SET_SREGS: {
2629                 struct kvm_sregs kvm_sregs;
2630
2631                 r = -EFAULT;
2632                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2633                         goto out;
2634                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2635                 if (r)
2636                         goto out;
2637                 r = 0;
2638                 break;
2639         }
2640         case KVM_TRANSLATE: {
2641                 struct kvm_translation tr;
2642
2643                 r = -EFAULT;
2644                 if (copy_from_user(&tr, argp, sizeof tr))
2645                         goto out;
2646                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2647                 if (r)
2648                         goto out;
2649                 r = -EFAULT;
2650                 if (copy_to_user(argp, &tr, sizeof tr))
2651                         goto out;
2652                 r = 0;
2653                 break;
2654         }
2655         case KVM_INTERRUPT: {
2656                 struct kvm_interrupt irq;
2657
2658                 r = -EFAULT;
2659                 if (copy_from_user(&irq, argp, sizeof irq))
2660                         goto out;
2661                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2662                 if (r)
2663                         goto out;
2664                 r = 0;
2665                 break;
2666         }
2667         case KVM_DEBUG_GUEST: {
2668                 struct kvm_debug_guest dbg;
2669
2670                 r = -EFAULT;
2671                 if (copy_from_user(&dbg, argp, sizeof dbg))
2672                         goto out;
2673                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2674                 if (r)
2675                         goto out;
2676                 r = 0;
2677                 break;
2678         }
2679         case KVM_GET_MSRS:
2680                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2681                 break;
2682         case KVM_SET_MSRS:
2683                 r = msr_io(vcpu, argp, do_set_msr, 0);
2684                 break;
2685         case KVM_SET_CPUID: {
2686                 struct kvm_cpuid __user *cpuid_arg = argp;
2687                 struct kvm_cpuid cpuid;
2688
2689                 r = -EFAULT;
2690                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2691                         goto out;
2692                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2693                 if (r)
2694                         goto out;
2695                 break;
2696         }
2697         case KVM_SET_SIGNAL_MASK: {
2698                 struct kvm_signal_mask __user *sigmask_arg = argp;
2699                 struct kvm_signal_mask kvm_sigmask;
2700                 sigset_t sigset, *p;
2701
2702                 p = NULL;
2703                 if (argp) {
2704                         r = -EFAULT;
2705                         if (copy_from_user(&kvm_sigmask, argp,
2706                                            sizeof kvm_sigmask))
2707                                 goto out;
2708                         r = -EINVAL;
2709                         if (kvm_sigmask.len != sizeof sigset)
2710                                 goto out;
2711                         r = -EFAULT;
2712                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2713                                            sizeof sigset))
2714                                 goto out;
2715                         p = &sigset;
2716                 }
2717                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2718                 break;
2719         }
2720         case KVM_GET_FPU: {
2721                 struct kvm_fpu fpu;
2722
2723                 memset(&fpu, 0, sizeof fpu);
2724                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2725                 if (r)
2726                         goto out;
2727                 r = -EFAULT;
2728                 if (copy_to_user(argp, &fpu, sizeof fpu))
2729                         goto out;
2730                 r = 0;
2731                 break;
2732         }
2733         case KVM_SET_FPU: {
2734                 struct kvm_fpu fpu;
2735
2736                 r = -EFAULT;
2737                 if (copy_from_user(&fpu, argp, sizeof fpu))
2738                         goto out;
2739                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2740                 if (r)
2741                         goto out;
2742                 r = 0;
2743                 break;
2744         }
2745         default:
2746                 ;
2747         }
2748 out:
2749         return r;
2750 }
2751
2752 static long kvm_vm_ioctl(struct file *filp,
2753                            unsigned int ioctl, unsigned long arg)
2754 {
2755         struct kvm *kvm = filp->private_data;
2756         void __user *argp = (void __user *)arg;
2757         int r = -EINVAL;
2758
2759         switch (ioctl) {
2760         case KVM_CREATE_VCPU:
2761                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2762                 if (r < 0)
2763                         goto out;
2764                 break;
2765         case KVM_SET_MEMORY_REGION: {
2766                 struct kvm_memory_region kvm_mem;
2767
2768                 r = -EFAULT;
2769                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2770                         goto out;
2771                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2772                 if (r)
2773                         goto out;
2774                 break;
2775         }
2776         case KVM_GET_DIRTY_LOG: {
2777                 struct kvm_dirty_log log;
2778
2779                 r = -EFAULT;
2780                 if (copy_from_user(&log, argp, sizeof log))
2781                         goto out;
2782                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2783                 if (r)
2784                         goto out;
2785                 break;
2786         }
2787         case KVM_SET_MEMORY_ALIAS: {
2788                 struct kvm_memory_alias alias;
2789
2790                 r = -EFAULT;
2791                 if (copy_from_user(&alias, argp, sizeof alias))
2792                         goto out;
2793                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2794                 if (r)
2795                         goto out;
2796                 break;
2797         }
2798         default:
2799                 ;
2800         }
2801 out:
2802         return r;
2803 }
2804
2805 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2806                                   unsigned long address,
2807                                   int *type)
2808 {
2809         struct kvm *kvm = vma->vm_file->private_data;
2810         unsigned long pgoff;
2811         struct page *page;
2812
2813         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2814         page = gfn_to_page(kvm, pgoff);
2815         if (!page)
2816                 return NOPAGE_SIGBUS;
2817         get_page(page);
2818         if (type != NULL)
2819                 *type = VM_FAULT_MINOR;
2820
2821         return page;
2822 }
2823
2824 static struct vm_operations_struct kvm_vm_vm_ops = {
2825         .nopage = kvm_vm_nopage,
2826 };
2827
2828 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2829 {
2830         vma->vm_ops = &kvm_vm_vm_ops;
2831         return 0;
2832 }
2833
2834 static struct file_operations kvm_vm_fops = {
2835         .release        = kvm_vm_release,
2836         .unlocked_ioctl = kvm_vm_ioctl,
2837         .compat_ioctl   = kvm_vm_ioctl,
2838         .mmap           = kvm_vm_mmap,
2839 };
2840
2841 static int kvm_dev_ioctl_create_vm(void)
2842 {
2843         int fd, r;
2844         struct inode *inode;
2845         struct file *file;
2846         struct kvm *kvm;
2847
2848         kvm = kvm_create_vm();
2849         if (IS_ERR(kvm))
2850                 return PTR_ERR(kvm);
2851         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2852         if (r) {
2853                 kvm_destroy_vm(kvm);
2854                 return r;
2855         }
2856
2857         kvm->filp = file;
2858
2859         return fd;
2860 }
2861
2862 static long kvm_dev_ioctl(struct file *filp,
2863                           unsigned int ioctl, unsigned long arg)
2864 {
2865         void __user *argp = (void __user *)arg;
2866         long r = -EINVAL;
2867
2868         switch (ioctl) {
2869         case KVM_GET_API_VERSION:
2870                 r = -EINVAL;
2871                 if (arg)
2872                         goto out;
2873                 r = KVM_API_VERSION;
2874                 break;
2875         case KVM_CREATE_VM:
2876                 r = -EINVAL;
2877                 if (arg)
2878                         goto out;
2879                 r = kvm_dev_ioctl_create_vm();
2880                 break;
2881         case KVM_GET_MSR_INDEX_LIST: {
2882                 struct kvm_msr_list __user *user_msr_list = argp;
2883                 struct kvm_msr_list msr_list;
2884                 unsigned n;
2885
2886                 r = -EFAULT;
2887                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2888                         goto out;
2889                 n = msr_list.nmsrs;
2890                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2891                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2892                         goto out;
2893                 r = -E2BIG;
2894                 if (n < num_msrs_to_save)
2895                         goto out;
2896                 r = -EFAULT;
2897                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2898                                  num_msrs_to_save * sizeof(u32)))
2899                         goto out;
2900                 if (copy_to_user(user_msr_list->indices
2901                                  + num_msrs_to_save * sizeof(u32),
2902                                  &emulated_msrs,
2903                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2904                         goto out;
2905                 r = 0;
2906                 break;
2907         }
2908         case KVM_CHECK_EXTENSION:
2909                 /*
2910                  * No extensions defined at present.
2911                  */
2912                 r = 0;
2913                 break;
2914         case KVM_GET_VCPU_MMAP_SIZE:
2915                 r = -EINVAL;
2916                 if (arg)
2917                         goto out;
2918                 r = 2 * PAGE_SIZE;
2919                 break;
2920         default:
2921                 ;
2922         }
2923 out:
2924         return r;
2925 }
2926
2927 static struct file_operations kvm_chardev_ops = {
2928         .open           = kvm_dev_open,
2929         .release        = kvm_dev_release,
2930         .unlocked_ioctl = kvm_dev_ioctl,
2931         .compat_ioctl   = kvm_dev_ioctl,
2932 };
2933
2934 static struct miscdevice kvm_dev = {
2935         KVM_MINOR,
2936         "kvm",
2937         &kvm_chardev_ops,
2938 };
2939
2940 /*
2941  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2942  * cached on it.
2943  */
2944 static void decache_vcpus_on_cpu(int cpu)
2945 {
2946         struct kvm *vm;
2947         struct kvm_vcpu *vcpu;
2948         int i;
2949
2950         spin_lock(&kvm_lock);
2951         list_for_each_entry(vm, &vm_list, vm_list)
2952                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2953                         vcpu = vm->vcpus[i];
2954                         if (!vcpu)
2955                                 continue;
2956                         /*
2957                          * If the vcpu is locked, then it is running on some
2958                          * other cpu and therefore it is not cached on the
2959                          * cpu in question.
2960                          *
2961                          * If it's not locked, check the last cpu it executed
2962                          * on.
2963                          */
2964                         if (mutex_trylock(&vcpu->mutex)) {
2965                                 if (vcpu->cpu == cpu) {
2966                                         kvm_arch_ops->vcpu_decache(vcpu);
2967                                         vcpu->cpu = -1;
2968                                 }
2969                                 mutex_unlock(&vcpu->mutex);
2970                         }
2971                 }
2972         spin_unlock(&kvm_lock);
2973 }
2974
2975 static void hardware_enable(void *junk)
2976 {
2977         int cpu = raw_smp_processor_id();
2978
2979         if (cpu_isset(cpu, cpus_hardware_enabled))
2980                 return;
2981         cpu_set(cpu, cpus_hardware_enabled);
2982         kvm_arch_ops->hardware_enable(NULL);
2983 }
2984
2985 static void hardware_disable(void *junk)
2986 {
2987         int cpu = raw_smp_processor_id();
2988
2989         if (!cpu_isset(cpu, cpus_hardware_enabled))
2990                 return;
2991         cpu_clear(cpu, cpus_hardware_enabled);
2992         decache_vcpus_on_cpu(cpu);
2993         kvm_arch_ops->hardware_disable(NULL);
2994 }
2995
2996 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2997                            void *v)
2998 {
2999         int cpu = (long)v;
3000
3001         switch (val) {
3002         case CPU_DYING:
3003         case CPU_DYING_FROZEN:
3004                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3005                        cpu);
3006                 hardware_disable(NULL);
3007                 break;
3008         case CPU_UP_CANCELED:
3009         case CPU_UP_CANCELED_FROZEN:
3010                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3011                        cpu);
3012                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3013                 break;
3014         case CPU_ONLINE:
3015         case CPU_ONLINE_FROZEN:
3016                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3017                        cpu);
3018                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3019                 break;
3020         }
3021         return NOTIFY_OK;
3022 }
3023
3024 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3025                        void *v)
3026 {
3027         if (val == SYS_RESTART) {
3028                 /*
3029                  * Some (well, at least mine) BIOSes hang on reboot if
3030                  * in vmx root mode.
3031                  */
3032                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3033                 on_each_cpu(hardware_disable, NULL, 0, 1);
3034         }
3035         return NOTIFY_OK;
3036 }
3037
3038 static struct notifier_block kvm_reboot_notifier = {
3039         .notifier_call = kvm_reboot,
3040         .priority = 0,
3041 };
3042
3043 void kvm_io_bus_init(struct kvm_io_bus *bus)
3044 {
3045         memset(bus, 0, sizeof(*bus));
3046 }
3047
3048 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3049 {
3050         int i;
3051
3052         for (i = 0; i < bus->dev_count; i++) {
3053                 struct kvm_io_device *pos = bus->devs[i];
3054
3055                 kvm_iodevice_destructor(pos);
3056         }
3057 }
3058
3059 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3060 {
3061         int i;
3062
3063         for (i = 0; i < bus->dev_count; i++) {
3064                 struct kvm_io_device *pos = bus->devs[i];
3065
3066                 if (pos->in_range(pos, addr))
3067                         return pos;
3068         }
3069
3070         return NULL;
3071 }
3072
3073 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3074 {
3075         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3076
3077         bus->devs[bus->dev_count++] = dev;
3078 }
3079
3080 static struct notifier_block kvm_cpu_notifier = {
3081         .notifier_call = kvm_cpu_hotplug,
3082         .priority = 20, /* must be > scheduler priority */
3083 };
3084
3085 static u64 stat_get(void *_offset)
3086 {
3087         unsigned offset = (long)_offset;
3088         u64 total = 0;
3089         struct kvm *kvm;
3090         struct kvm_vcpu *vcpu;
3091         int i;
3092
3093         spin_lock(&kvm_lock);
3094         list_for_each_entry(kvm, &vm_list, vm_list)
3095                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3096                         vcpu = kvm->vcpus[i];
3097                         if (vcpu)
3098                                 total += *(u32 *)((void *)vcpu + offset);
3099                 }
3100         spin_unlock(&kvm_lock);
3101         return total;
3102 }
3103
3104 static void stat_set(void *offset, u64 val)
3105 {
3106 }
3107
3108 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
3109
3110 static __init void kvm_init_debug(void)
3111 {
3112         struct kvm_stats_debugfs_item *p;
3113
3114         debugfs_dir = debugfs_create_dir("kvm", NULL);
3115         for (p = debugfs_entries; p->name; ++p)
3116                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3117                                                 (void *)(long)p->offset,
3118                                                 &stat_fops);
3119 }
3120
3121 static void kvm_exit_debug(void)
3122 {
3123         struct kvm_stats_debugfs_item *p;
3124
3125         for (p = debugfs_entries; p->name; ++p)
3126                 debugfs_remove(p->dentry);
3127         debugfs_remove(debugfs_dir);
3128 }
3129
3130 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3131 {
3132         hardware_disable(NULL);
3133         return 0;
3134 }
3135
3136 static int kvm_resume(struct sys_device *dev)
3137 {
3138         hardware_enable(NULL);
3139         return 0;
3140 }
3141
3142 static struct sysdev_class kvm_sysdev_class = {
3143         set_kset_name("kvm"),
3144         .suspend = kvm_suspend,
3145         .resume = kvm_resume,
3146 };
3147
3148 static struct sys_device kvm_sysdev = {
3149         .id = 0,
3150         .cls = &kvm_sysdev_class,
3151 };
3152
3153 hpa_t bad_page_address;
3154
3155 static inline
3156 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3157 {
3158         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3159 }
3160
3161 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3162 {
3163         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3164
3165         kvm_arch_ops->vcpu_load(vcpu, cpu);
3166 }
3167
3168 static void kvm_sched_out(struct preempt_notifier *pn,
3169                           struct task_struct *next)
3170 {
3171         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3172
3173         kvm_arch_ops->vcpu_put(vcpu);
3174 }
3175
3176 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
3177 {
3178         int r;
3179
3180         if (kvm_arch_ops) {
3181                 printk(KERN_ERR "kvm: already loaded the other module\n");
3182                 return -EEXIST;
3183         }
3184
3185         if (!ops->cpu_has_kvm_support()) {
3186                 printk(KERN_ERR "kvm: no hardware support\n");
3187                 return -EOPNOTSUPP;
3188         }
3189         if (ops->disabled_by_bios()) {
3190                 printk(KERN_ERR "kvm: disabled by bios\n");
3191                 return -EOPNOTSUPP;
3192         }
3193
3194         kvm_arch_ops = ops;
3195
3196         r = kvm_arch_ops->hardware_setup();
3197         if (r < 0)
3198                 goto out;
3199
3200         on_each_cpu(hardware_enable, NULL, 0, 1);
3201         r = register_cpu_notifier(&kvm_cpu_notifier);
3202         if (r)
3203                 goto out_free_1;
3204         register_reboot_notifier(&kvm_reboot_notifier);
3205
3206         r = sysdev_class_register(&kvm_sysdev_class);
3207         if (r)
3208                 goto out_free_2;
3209
3210         r = sysdev_register(&kvm_sysdev);
3211         if (r)
3212                 goto out_free_3;
3213
3214         kvm_chardev_ops.owner = module;
3215
3216         r = misc_register(&kvm_dev);
3217         if (r) {
3218                 printk (KERN_ERR "kvm: misc device register failed\n");
3219                 goto out_free;
3220         }
3221
3222         kvm_preempt_ops.sched_in = kvm_sched_in;
3223         kvm_preempt_ops.sched_out = kvm_sched_out;
3224
3225         return r;
3226
3227 out_free:
3228         sysdev_unregister(&kvm_sysdev);
3229 out_free_3:
3230         sysdev_class_unregister(&kvm_sysdev_class);
3231 out_free_2:
3232         unregister_reboot_notifier(&kvm_reboot_notifier);
3233         unregister_cpu_notifier(&kvm_cpu_notifier);
3234 out_free_1:
3235         on_each_cpu(hardware_disable, NULL, 0, 1);
3236         kvm_arch_ops->hardware_unsetup();
3237 out:
3238         kvm_arch_ops = NULL;
3239         return r;
3240 }
3241
3242 void kvm_exit_arch(void)
3243 {
3244         misc_deregister(&kvm_dev);
3245         sysdev_unregister(&kvm_sysdev);
3246         sysdev_class_unregister(&kvm_sysdev_class);
3247         unregister_reboot_notifier(&kvm_reboot_notifier);
3248         unregister_cpu_notifier(&kvm_cpu_notifier);
3249         on_each_cpu(hardware_disable, NULL, 0, 1);
3250         kvm_arch_ops->hardware_unsetup();
3251         kvm_arch_ops = NULL;
3252 }
3253
3254 static __init int kvm_init(void)
3255 {
3256         static struct page *bad_page;
3257         int r;
3258
3259         r = kvm_mmu_module_init();
3260         if (r)
3261                 goto out4;
3262
3263         kvm_init_debug();
3264
3265         kvm_init_msr_list();
3266
3267         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3268                 r = -ENOMEM;
3269                 goto out;
3270         }
3271
3272         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3273         memset(__va(bad_page_address), 0, PAGE_SIZE);
3274
3275         return 0;
3276
3277 out:
3278         kvm_exit_debug();
3279         kvm_mmu_module_exit();
3280 out4:
3281         return r;
3282 }
3283
3284 static __exit void kvm_exit(void)
3285 {
3286         kvm_exit_debug();
3287         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3288         kvm_mmu_module_exit();
3289 }
3290
3291 module_init(kvm_init)
3292 module_exit(kvm_exit)
3293
3294 EXPORT_SYMBOL_GPL(kvm_init_arch);
3295 EXPORT_SYMBOL_GPL(kvm_exit_arch);