err.no Git - linux-2.6/blob - 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 void hardware_disable(void *ignored);
  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_RESEVED_BITS 0xffffffff1ffaffc0ULL
  88 #define LMSW_GUEST_MASK 0x0eULL
  89 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
  90 #define CR8_RESEVED_BITS (~0x0fULL)
  91 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  92
  93 #ifdef CONFIG_X86_64
  94 // LDT or TSS descriptor in the GDT. 16 bytes.
  95 struct segment_descriptor_64 {
  96         struct segment_descriptor s;
  97         u32 base_higher;
  98         u32 pad_zero;
  99 };
 100
 101 #endif
 102
 103 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 104                            unsigned long arg);
 105
 106 unsigned long segment_base(u16 selector)
 107 {
 108         struct descriptor_table gdt;
 109         struct segment_descriptor *d;
 110         unsigned long table_base;
 111         typedef unsigned long ul;
 112         unsigned long v;
 113
 114         if (selector == 0)
 115                 return 0;
 116
 117         asm ("sgdt %0" : "=m"(gdt));
 118         table_base = gdt.base;
 119
 120         if (selector & 4) {           /* from ldt */
 121                 u16 ldt_selector;
 122
 123                 asm ("sldt %0" : "=g"(ldt_selector));
 124                 table_base = segment_base(ldt_selector);
 125         }
 126         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 127         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 128 #ifdef CONFIG_X86_64
 129         if (d->system == 0
 130             && (d->type == 2 || d->type == 9 || d->type == 11))
 131                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 132 #endif
 133         return v;
 134 }
 135 EXPORT_SYMBOL_GPL(segment_base);
 136
 137 static inline int valid_vcpu(int n)
 138 {
 139         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 140 }
 141
 142 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 143                    void *dest)
 144 {
 145         unsigned char *host_buf = dest;
 146         unsigned long req_size = size;
 147
 148         while (size) {
 149                 hpa_t paddr;
 150                 unsigned now;
 151                 unsigned offset;
 152                 hva_t guest_buf;
 153
 154                 paddr = gva_to_hpa(vcpu, addr);
 155
 156                 if (is_error_hpa(paddr))
 157                         break;
 158
 159                 guest_buf = (hva_t)kmap_atomic(
 160                                         pfn_to_page(paddr >> PAGE_SHIFT),
 161                                         KM_USER0);
 162                 offset = addr & ~PAGE_MASK;
 163                 guest_buf |= offset;
 164                 now = min(size, PAGE_SIZE - offset);
 165                 memcpy(host_buf, (void*)guest_buf, now);
 166                 host_buf += now;
 167                 addr += now;
 168                 size -= now;
 169                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 170         }
 171         return req_size - size;
 172 }
 173 EXPORT_SYMBOL_GPL(kvm_read_guest);
 174
 175 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 176                     void *data)
 177 {
 178         unsigned char *host_buf = data;
 179         unsigned long req_size = size;
 180
 181         while (size) {
 182                 hpa_t paddr;
 183                 unsigned now;
 184                 unsigned offset;
 185                 hva_t guest_buf;
 186                 gfn_t gfn;
 187
 188                 paddr = gva_to_hpa(vcpu, addr);
 189
 190                 if (is_error_hpa(paddr))
 191                         break;
 192
 193                 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
 194                 mark_page_dirty(vcpu->kvm, gfn);
 195                 guest_buf = (hva_t)kmap_atomic(
 196                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
 197                 offset = addr & ~PAGE_MASK;
 198                 guest_buf |= offset;
 199                 now = min(size, PAGE_SIZE - offset);
 200                 memcpy((void*)guest_buf, host_buf, now);
 201                 host_buf += now;
 202                 addr += now;
 203                 size -= now;
 204                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 205         }
 206         return req_size - size;
 207 }
 208 EXPORT_SYMBOL_GPL(kvm_write_guest);
 209
 210 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 211 {
 212         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 213                 return;
 214
 215         vcpu->guest_fpu_loaded = 1;
 216         fx_save(vcpu->host_fx_image);
 217         fx_restore(vcpu->guest_fx_image);
 218 }
 219 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 220
 221 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 222 {
 223         if (!vcpu->guest_fpu_loaded)
 224                 return;
 225
 226         vcpu->guest_fpu_loaded = 0;
 227         fx_save(vcpu->guest_fx_image);
 228         fx_restore(vcpu->host_fx_image);
 229 }
 230 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 231
 232 /*
 233  * Switches to specified vcpu, until a matching vcpu_put()
 234  */
 235 static void vcpu_load(struct kvm_vcpu *vcpu)
 236 {
 237         mutex_lock(&vcpu->mutex);
 238         kvm_arch_ops->vcpu_load(vcpu);
 239 }
 240
 241 static void vcpu_put(struct kvm_vcpu *vcpu)
 242 {
 243         kvm_arch_ops->vcpu_put(vcpu);
 244         mutex_unlock(&vcpu->mutex);
 245 }
 246
 247 static void ack_flush(void *_completed)
 248 {
 249         atomic_t *completed = _completed;
 250
 251         atomic_inc(completed);
 252 }
 253
 254 void kvm_flush_remote_tlbs(struct kvm *kvm)
 255 {
 256         int i, cpu, needed;
 257         cpumask_t cpus;
 258         struct kvm_vcpu *vcpu;
 259         atomic_t completed;
 260
 261         atomic_set(&completed, 0);
 262         cpus_clear(cpus);
 263         needed = 0;
 264         for (i = 0; i < kvm->nvcpus; ++i) {
 265                 vcpu = &kvm->vcpus[i];
 266                 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
 267                         continue;
 268                 cpu = vcpu->cpu;
 269                 if (cpu != -1 && cpu != raw_smp_processor_id())
 270                         if (!cpu_isset(cpu, cpus)) {
 271                                 cpu_set(cpu, cpus);
 272                                 ++needed;
 273                         }
 274         }
 275
 276         /*
 277          * We really want smp_call_function_mask() here.  But that's not
 278          * available, so ipi all cpus in parallel and wait for them
 279          * to complete.
 280          */
 281         for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
 282                 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
 283         while (atomic_read(&completed) != needed) {
 284                 cpu_relax();
 285                 barrier();
 286         }
 287 }
 288
 289 static struct kvm *kvm_create_vm(void)
 290 {
 291         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 292         int i;
 293
 294         if (!kvm)
 295                 return ERR_PTR(-ENOMEM);
 296
 297         kvm_io_bus_init(&kvm->pio_bus);
 298         spin_lock_init(&kvm->lock);
 299         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 300         kvm_io_bus_init(&kvm->mmio_bus);
 301         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 302                 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
 303
 304                 mutex_init(&vcpu->mutex);
 305                 vcpu->cpu = -1;
 306                 vcpu->kvm = kvm;
 307                 vcpu->mmu.root_hpa = INVALID_PAGE;
 308         }
 309         spin_lock(&kvm_lock);
 310         list_add(&kvm->vm_list, &vm_list);
 311         spin_unlock(&kvm_lock);
 312         return kvm;
 313 }
 314
 315 static int kvm_dev_open(struct inode *inode, struct file *filp)
 316 {
 317         return 0;
 318 }
 319
 320 /*
 321  * Free any memory in @free but not in @dont.
 322  */
 323 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 324                                   struct kvm_memory_slot *dont)
 325 {
 326         int i;
 327
 328         if (!dont || free->phys_mem != dont->phys_mem)
 329                 if (free->phys_mem) {
 330                         for (i = 0; i < free->npages; ++i)
 331                                 if (free->phys_mem[i])
 332                                         __free_page(free->phys_mem[i]);
 333                         vfree(free->phys_mem);
 334                 }
 335
 336         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 337                 vfree(free->dirty_bitmap);
 338
 339         free->phys_mem = NULL;
 340         free->npages = 0;
 341         free->dirty_bitmap = NULL;
 342 }
 343
 344 static void kvm_free_physmem(struct kvm *kvm)
 345 {
 346         int i;
 347
 348         for (i = 0; i < kvm->nmemslots; ++i)
 349                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 350 }
 351
 352 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 353 {
 354         int i;
 355
 356         for (i = 0; i < 2; ++i)
 357                 if (vcpu->pio.guest_pages[i]) {
 358                         __free_page(vcpu->pio.guest_pages[i]);
 359                         vcpu->pio.guest_pages[i] = NULL;
 360                 }
 361 }
 362
 363 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 364 {
 365         if (!vcpu->vmcs)
 366                 return;
 367
 368         vcpu_load(vcpu);
 369         kvm_mmu_unload(vcpu);
 370         vcpu_put(vcpu);
 371 }
 372
 373 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
 374 {
 375         if (!vcpu->vmcs)
 376                 return;
 377
 378         vcpu_load(vcpu);
 379         kvm_mmu_destroy(vcpu);
 380         vcpu_put(vcpu);
 381         kvm_arch_ops->vcpu_free(vcpu);
 382         free_page((unsigned long)vcpu->run);
 383         vcpu->run = NULL;
 384         free_page((unsigned long)vcpu->pio_data);
 385         vcpu->pio_data = NULL;
 386         free_pio_guest_pages(vcpu);
 387 }
 388
 389 static void kvm_free_vcpus(struct kvm *kvm)
 390 {
 391         unsigned int i;
 392
 393         /*
 394          * Unpin any mmu pages first.
 395          */
 396         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 397                 kvm_unload_vcpu_mmu(&kvm->vcpus[i]);
 398         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 399                 kvm_free_vcpu(&kvm->vcpus[i]);
 400 }
 401
 402 static int kvm_dev_release(struct inode *inode, struct file *filp)
 403 {
 404         return 0;
 405 }
 406
 407 static void kvm_destroy_vm(struct kvm *kvm)
 408 {
 409         spin_lock(&kvm_lock);
 410         list_del(&kvm->vm_list);
 411         spin_unlock(&kvm_lock);
 412         kvm_io_bus_destroy(&kvm->pio_bus);
 413         kvm_io_bus_destroy(&kvm->mmio_bus);
 414         kvm_free_vcpus(kvm);
 415         kvm_free_physmem(kvm);
 416         kfree(kvm);
 417 }
 418
 419 static int kvm_vm_release(struct inode *inode, struct file *filp)
 420 {
 421         struct kvm *kvm = filp->private_data;
 422
 423         kvm_destroy_vm(kvm);
 424         return 0;
 425 }
 426
 427 static void inject_gp(struct kvm_vcpu *vcpu)
 428 {
 429         kvm_arch_ops->inject_gp(vcpu, 0);
 430 }
 431
 432 /*
 433  * Load the pae pdptrs.  Return true is they are all valid.
 434  */
 435 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 436 {
 437         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 438         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 439         int i;
 440         u64 pdpte;
 441         u64 *pdpt;
 442         int ret;
 443         struct page *page;
 444
 445         spin_lock(&vcpu->kvm->lock);
 446         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
 447         /* FIXME: !page - emulate? 0xff? */
 448         pdpt = kmap_atomic(page, KM_USER0);
 449
 450         ret = 1;
 451         for (i = 0; i < 4; ++i) {
 452                 pdpte = pdpt[offset + i];
 453                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
 454                         ret = 0;
 455                         goto out;
 456                 }
 457         }
 458
 459         for (i = 0; i < 4; ++i)
 460                 vcpu->pdptrs[i] = pdpt[offset + i];
 461
 462 out:
 463         kunmap_atomic(pdpt, KM_USER0);
 464         spin_unlock(&vcpu->kvm->lock);
 465
 466         return ret;
 467 }
 468
 469 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 470 {
 471         if (cr0 & CR0_RESEVED_BITS) {
 472                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 473                        cr0, vcpu->cr0);
 474                 inject_gp(vcpu);
 475                 return;
 476         }
 477
 478         if ((cr0 & CR0_NW_MASK) && !(cr0 & CR0_CD_MASK)) {
 479                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 480                 inject_gp(vcpu);
 481                 return;
 482         }
 483
 484         if ((cr0 & CR0_PG_MASK) && !(cr0 & CR0_PE_MASK)) {
 485                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 486                        "and a clear PE flag\n");
 487                 inject_gp(vcpu);
 488                 return;
 489         }
 490
 491         if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
 492 #ifdef CONFIG_X86_64
 493                 if ((vcpu->shadow_efer & EFER_LME)) {
 494                         int cs_db, cs_l;
 495
 496                         if (!is_pae(vcpu)) {
 497                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 498                                        "in long mode while PAE is disabled\n");
 499                                 inject_gp(vcpu);
 500                                 return;
 501                         }
 502                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 503                         if (cs_l) {
 504                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 505                                        "in long mode while CS.L == 1\n");
 506                                 inject_gp(vcpu);
 507                                 return;
 508
 509                         }
 510                 } else
 511 #endif
 512                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 513                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 514                                "reserved bits\n");
 515                         inject_gp(vcpu);
 516                         return;
 517                 }
 518
 519         }
 520
 521         kvm_arch_ops->set_cr0(vcpu, cr0);
 522         vcpu->cr0 = cr0;
 523
 524         spin_lock(&vcpu->kvm->lock);
 525         kvm_mmu_reset_context(vcpu);
 526         spin_unlock(&vcpu->kvm->lock);
 527         return;
 528 }
 529 EXPORT_SYMBOL_GPL(set_cr0);
 530
 531 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 532 {
 533         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 534 }
 535 EXPORT_SYMBOL_GPL(lmsw);
 536
 537 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 538 {
 539         if (cr4 & CR4_RESEVED_BITS) {
 540                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 541                 inject_gp(vcpu);
 542                 return;
 543         }
 544
 545         if (is_long_mode(vcpu)) {
 546                 if (!(cr4 & CR4_PAE_MASK)) {
 547                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 548                                "in long mode\n");
 549                         inject_gp(vcpu);
 550                         return;
 551                 }
 552         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
 553                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 554                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 555                 inject_gp(vcpu);
 556         }
 557
 558         if (cr4 & CR4_VMXE_MASK) {
 559                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 560                 inject_gp(vcpu);
 561                 return;
 562         }
 563         kvm_arch_ops->set_cr4(vcpu, cr4);
 564         spin_lock(&vcpu->kvm->lock);
 565         kvm_mmu_reset_context(vcpu);
 566         spin_unlock(&vcpu->kvm->lock);
 567 }
 568 EXPORT_SYMBOL_GPL(set_cr4);
 569
 570 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 571 {
 572         if (is_long_mode(vcpu)) {
 573                 if (cr3 & CR3_L_MODE_RESEVED_BITS) {
 574                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 575                         inject_gp(vcpu);
 576                         return;
 577                 }
 578         } else {
 579                 if (cr3 & CR3_RESEVED_BITS) {
 580                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 581                         inject_gp(vcpu);
 582                         return;
 583                 }
 584                 if (is_paging(vcpu) && is_pae(vcpu) &&
 585                     !load_pdptrs(vcpu, cr3)) {
 586                         printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 587                                "reserved bits\n");
 588                         inject_gp(vcpu);
 589                         return;
 590                 }
 591         }
 592
 593         vcpu->cr3 = cr3;
 594         spin_lock(&vcpu->kvm->lock);
 595         /*
 596          * Does the new cr3 value map to physical memory? (Note, we
 597          * catch an invalid cr3 even in real-mode, because it would
 598          * cause trouble later on when we turn on paging anyway.)
 599          *
 600          * A real CPU would silently accept an invalid cr3 and would
 601          * attempt to use it - with largely undefined (and often hard
 602          * to debug) behavior on the guest side.
 603          */
 604         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 605                 inject_gp(vcpu);
 606         else
 607                 vcpu->mmu.new_cr3(vcpu);
 608         spin_unlock(&vcpu->kvm->lock);
 609 }
 610 EXPORT_SYMBOL_GPL(set_cr3);
 611
 612 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 613 {
 614         if ( cr8 & CR8_RESEVED_BITS) {
 615                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 616                 inject_gp(vcpu);
 617                 return;
 618         }
 619         vcpu->cr8 = cr8;
 620 }
 621 EXPORT_SYMBOL_GPL(set_cr8);
 622
 623 void fx_init(struct kvm_vcpu *vcpu)
 624 {
 625         struct __attribute__ ((__packed__)) fx_image_s {
 626                 u16 control; //fcw
 627                 u16 status; //fsw
 628                 u16 tag; // ftw
 629                 u16 opcode; //fop
 630                 u64 ip; // fpu ip
 631                 u64 operand;// fpu dp
 632                 u32 mxcsr;
 633                 u32 mxcsr_mask;
 634
 635         } *fx_image;
 636
 637         fx_save(vcpu->host_fx_image);
 638         fpu_init();
 639         fx_save(vcpu->guest_fx_image);
 640         fx_restore(vcpu->host_fx_image);
 641
 642         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
 643         fx_image->mxcsr = 0x1f80;
 644         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
 645                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
 646 }
 647 EXPORT_SYMBOL_GPL(fx_init);
 648
 649 /*
 650  * Allocate some memory and give it an address in the guest physical address
 651  * space.
 652  *
 653  * Discontiguous memory is allowed, mostly for framebuffers.
 654  */
 655 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 656                                           struct kvm_memory_region *mem)
 657 {
 658         int r;
 659         gfn_t base_gfn;
 660         unsigned long npages;
 661         unsigned long i;
 662         struct kvm_memory_slot *memslot;
 663         struct kvm_memory_slot old, new;
 664         int memory_config_version;
 665
 666         r = -EINVAL;
 667         /* General sanity checks */
 668         if (mem->memory_size & (PAGE_SIZE - 1))
 669                 goto out;
 670         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 671                 goto out;
 672         if (mem->slot >= KVM_MEMORY_SLOTS)
 673                 goto out;
 674         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 675                 goto out;
 676
 677         memslot = &kvm->memslots[mem->slot];
 678         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 679         npages = mem->memory_size >> PAGE_SHIFT;
 680
 681         if (!npages)
 682                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 683
 684 raced:
 685         spin_lock(&kvm->lock);
 686
 687         memory_config_version = kvm->memory_config_version;
 688         new = old = *memslot;
 689
 690         new.base_gfn = base_gfn;
 691         new.npages = npages;
 692         new.flags = mem->flags;
 693
 694         /* Disallow changing a memory slot's size. */
 695         r = -EINVAL;
 696         if (npages && old.npages && npages != old.npages)
 697                 goto out_unlock;
 698
 699         /* Check for overlaps */
 700         r = -EEXIST;
 701         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 702                 struct kvm_memory_slot *s = &kvm->memslots[i];
 703
 704                 if (s == memslot)
 705                         continue;
 706                 if (!((base_gfn + npages <= s->base_gfn) ||
 707                       (base_gfn >= s->base_gfn + s->npages)))
 708                         goto out_unlock;
 709         }
 710         /*
 711          * Do memory allocations outside lock.  memory_config_version will
 712          * detect any races.
 713          */
 714         spin_unlock(&kvm->lock);
 715
 716         /* Deallocate if slot is being removed */
 717         if (!npages)
 718                 new.phys_mem = NULL;
 719
 720         /* Free page dirty bitmap if unneeded */
 721         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 722                 new.dirty_bitmap = NULL;
 723
 724         r = -ENOMEM;
 725
 726         /* Allocate if a slot is being created */
 727         if (npages && !new.phys_mem) {
 728                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 729
 730                 if (!new.phys_mem)
 731                         goto out_free;
 732
 733                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 734                 for (i = 0; i < npages; ++i) {
 735                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 736                                                      | __GFP_ZERO);
 737                         if (!new.phys_mem[i])
 738                                 goto out_free;
 739                         set_page_private(new.phys_mem[i],0);
 740                 }
 741         }
 742
 743         /* Allocate page dirty bitmap if needed */
 744         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 745                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 746
 747                 new.dirty_bitmap = vmalloc(dirty_bytes);
 748                 if (!new.dirty_bitmap)
 749                         goto out_free;
 750                 memset(new.dirty_bitmap, 0, dirty_bytes);
 751         }
 752
 753         spin_lock(&kvm->lock);
 754
 755         if (memory_config_version != kvm->memory_config_version) {
 756                 spin_unlock(&kvm->lock);
 757                 kvm_free_physmem_slot(&new, &old);
 758                 goto raced;
 759         }
 760
 761         r = -EAGAIN;
 762         if (kvm->busy)
 763                 goto out_unlock;
 764
 765         if (mem->slot >= kvm->nmemslots)
 766                 kvm->nmemslots = mem->slot + 1;
 767
 768         *memslot = new;
 769         ++kvm->memory_config_version;
 770
 771         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 772         kvm_flush_remote_tlbs(kvm);
 773
 774         spin_unlock(&kvm->lock);
 775
 776         kvm_free_physmem_slot(&old, &new);
 777         return 0;
 778
 779 out_unlock:
 780         spin_unlock(&kvm->lock);
 781 out_free:
 782         kvm_free_physmem_slot(&new, &old);
 783 out:
 784         return r;
 785 }
 786
 787 /*
 788  * Get (and clear) the dirty memory log for a memory slot.
 789  */
 790 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 791                                       struct kvm_dirty_log *log)
 792 {
 793         struct kvm_memory_slot *memslot;
 794         int r, i;
 795         int n;
 796         unsigned long any = 0;
 797
 798         spin_lock(&kvm->lock);
 799
 800         /*
 801          * Prevent changes to guest memory configuration even while the lock
 802          * is not taken.
 803          */
 804         ++kvm->busy;
 805         spin_unlock(&kvm->lock);
 806         r = -EINVAL;
 807         if (log->slot >= KVM_MEMORY_SLOTS)
 808                 goto out;
 809
 810         memslot = &kvm->memslots[log->slot];
 811         r = -ENOENT;
 812         if (!memslot->dirty_bitmap)
 813                 goto out;
 814
 815         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 816
 817         for (i = 0; !any && i < n/sizeof(long); ++i)
 818                 any = memslot->dirty_bitmap[i];
 819
 820         r = -EFAULT;
 821         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 822                 goto out;
 823
 824         spin_lock(&kvm->lock);
 825         kvm_mmu_slot_remove_write_access(kvm, log->slot);
 826         kvm_flush_remote_tlbs(kvm);
 827         memset(memslot->dirty_bitmap, 0, n);
 828         spin_unlock(&kvm->lock);
 829
 830         r = 0;
 831
 832 out:
 833         spin_lock(&kvm->lock);
 834         --kvm->busy;
 835         spin_unlock(&kvm->lock);
 836         return r;
 837 }
 838
 839 /*
 840  * Set a new alias region.  Aliases map a portion of physical memory into
 841  * another portion.  This is useful for memory windows, for example the PC
 842  * VGA region.
 843  */
 844 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 845                                          struct kvm_memory_alias *alias)
 846 {
 847         int r, n;
 848         struct kvm_mem_alias *p;
 849
 850         r = -EINVAL;
 851         /* General sanity checks */
 852         if (alias->memory_size & (PAGE_SIZE - 1))
 853                 goto out;
 854         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 855                 goto out;
 856         if (alias->slot >= KVM_ALIAS_SLOTS)
 857                 goto out;
 858         if (alias->guest_phys_addr + alias->memory_size
 859             < alias->guest_phys_addr)
 860                 goto out;
 861         if (alias->target_phys_addr + alias->memory_size
 862             < alias->target_phys_addr)
 863                 goto out;
 864
 865         spin_lock(&kvm->lock);
 866
 867         p = &kvm->aliases[alias->slot];
 868         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 869         p->npages = alias->memory_size >> PAGE_SHIFT;
 870         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 871
 872         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 873                 if (kvm->aliases[n - 1].npages)
 874                         break;
 875         kvm->naliases = n;
 876
 877         kvm_mmu_zap_all(kvm);
 878
 879         spin_unlock(&kvm->lock);
 880
 881         return 0;
 882
 883 out:
 884         return r;
 885 }
 886
 887 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 888 {
 889         int i;
 890         struct kvm_mem_alias *alias;
 891
 892         for (i = 0; i < kvm->naliases; ++i) {
 893                 alias = &kvm->aliases[i];
 894                 if (gfn >= alias->base_gfn
 895                     && gfn < alias->base_gfn + alias->npages)
 896                         return alias->target_gfn + gfn - alias->base_gfn;
 897         }
 898         return gfn;
 899 }
 900
 901 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 902 {
 903         int i;
 904
 905         for (i = 0; i < kvm->nmemslots; ++i) {
 906                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 907
 908                 if (gfn >= memslot->base_gfn
 909                     && gfn < memslot->base_gfn + memslot->npages)
 910                         return memslot;
 911         }
 912         return NULL;
 913 }
 914
 915 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 916 {
 917         gfn = unalias_gfn(kvm, gfn);
 918         return __gfn_to_memslot(kvm, gfn);
 919 }
 920
 921 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 922 {
 923         struct kvm_memory_slot *slot;
 924
 925         gfn = unalias_gfn(kvm, gfn);
 926         slot = __gfn_to_memslot(kvm, gfn);
 927         if (!slot)
 928                 return NULL;
 929         return slot->phys_mem[gfn - slot->base_gfn];
 930 }
 931 EXPORT_SYMBOL_GPL(gfn_to_page);
 932
 933 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 934 {
 935         int i;
 936         struct kvm_memory_slot *memslot;
 937         unsigned long rel_gfn;
 938
 939         for (i = 0; i < kvm->nmemslots; ++i) {
 940                 memslot = &kvm->memslots[i];
 941
 942                 if (gfn >= memslot->base_gfn
 943                     && gfn < memslot->base_gfn + memslot->npages) {
 944
 945                         if (!memslot->dirty_bitmap)
 946                                 return;
 947
 948                         rel_gfn = gfn - memslot->base_gfn;
 949
 950                         /* avoid RMW */
 951                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 952                                 set_bit(rel_gfn, memslot->dirty_bitmap);
 953                         return;
 954                 }
 955         }
 956 }
 957
 958 static int emulator_read_std(unsigned long addr,
 959                              void *val,
 960                              unsigned int bytes,
 961                              struct x86_emulate_ctxt *ctxt)
 962 {
 963         struct kvm_vcpu *vcpu = ctxt->vcpu;
 964         void *data = val;
 965
 966         while (bytes) {
 967                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 968                 unsigned offset = addr & (PAGE_SIZE-1);
 969                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 970                 unsigned long pfn;
 971                 struct page *page;
 972                 void *page_virt;
 973
 974                 if (gpa == UNMAPPED_GVA)
 975                         return X86EMUL_PROPAGATE_FAULT;
 976                 pfn = gpa >> PAGE_SHIFT;
 977                 page = gfn_to_page(vcpu->kvm, pfn);
 978                 if (!page)
 979                         return X86EMUL_UNHANDLEABLE;
 980                 page_virt = kmap_atomic(page, KM_USER0);
 981
 982                 memcpy(data, page_virt + offset, tocopy);
 983
 984                 kunmap_atomic(page_virt, KM_USER0);
 985
 986                 bytes -= tocopy;
 987                 data += tocopy;
 988                 addr += tocopy;
 989         }
 990
 991         return X86EMUL_CONTINUE;
 992 }
 993
 994 static int emulator_write_std(unsigned long addr,
 995                               const void *val,
 996                               unsigned int bytes,
 997                               struct x86_emulate_ctxt *ctxt)
 998 {
 999         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
1000                addr, bytes);
1001         return X86EMUL_UNHANDLEABLE;
1002 }
1003
1004 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1005                                                 gpa_t addr)
1006 {
1007         /*
1008          * Note that its important to have this wrapper function because
1009          * in the very near future we will be checking for MMIOs against
1010          * the LAPIC as well as the general MMIO bus
1011          */
1012         return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1013 }
1014
1015 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1016                                                gpa_t addr)
1017 {
1018         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1019 }
1020
1021 static int emulator_read_emulated(unsigned long addr,
1022                                   void *val,
1023                                   unsigned int bytes,
1024                                   struct x86_emulate_ctxt *ctxt)
1025 {
1026         struct kvm_vcpu      *vcpu = ctxt->vcpu;
1027         struct kvm_io_device *mmio_dev;
1028         gpa_t                 gpa;
1029
1030         if (vcpu->mmio_read_completed) {
1031                 memcpy(val, vcpu->mmio_data, bytes);
1032                 vcpu->mmio_read_completed = 0;
1033                 return X86EMUL_CONTINUE;
1034         } else if (emulator_read_std(addr, val, bytes, ctxt)
1035                    == X86EMUL_CONTINUE)
1036                 return X86EMUL_CONTINUE;
1037
1038         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1039         if (gpa == UNMAPPED_GVA)
1040                 return X86EMUL_PROPAGATE_FAULT;
1041
1042         /*
1043          * Is this MMIO handled locally?
1044          */
1045         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1046         if (mmio_dev) {
1047                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1048                 return X86EMUL_CONTINUE;
1049         }
1050
1051         vcpu->mmio_needed = 1;
1052         vcpu->mmio_phys_addr = gpa;
1053         vcpu->mmio_size = bytes;
1054         vcpu->mmio_is_write = 0;
1055
1056         return X86EMUL_UNHANDLEABLE;
1057 }
1058
1059 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1060                                const void *val, int bytes)
1061 {
1062         struct page *page;
1063         void *virt;
1064         unsigned offset = offset_in_page(gpa);
1065
1066         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1067                 return 0;
1068         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1069         if (!page)
1070                 return 0;
1071         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1072         virt = kmap_atomic(page, KM_USER0);
1073         kvm_mmu_pte_write(vcpu, gpa, virt + offset, val, bytes);
1074         memcpy(virt + offset_in_page(gpa), val, bytes);
1075         kunmap_atomic(virt, KM_USER0);
1076         return 1;
1077 }
1078
1079 static int emulator_write_emulated_onepage(unsigned long addr,
1080                                            const void *val,
1081                                            unsigned int bytes,
1082                                            struct x86_emulate_ctxt *ctxt)
1083 {
1084         struct kvm_vcpu      *vcpu = ctxt->vcpu;
1085         struct kvm_io_device *mmio_dev;
1086         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1087
1088         if (gpa == UNMAPPED_GVA) {
1089                 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1090                 return X86EMUL_PROPAGATE_FAULT;
1091         }
1092
1093         if (emulator_write_phys(vcpu, gpa, val, bytes))
1094                 return X86EMUL_CONTINUE;
1095
1096         /*
1097          * Is this MMIO handled locally?
1098          */
1099         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1100         if (mmio_dev) {
1101                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1102                 return X86EMUL_CONTINUE;
1103         }
1104
1105         vcpu->mmio_needed = 1;
1106         vcpu->mmio_phys_addr = gpa;
1107         vcpu->mmio_size = bytes;
1108         vcpu->mmio_is_write = 1;
1109         memcpy(vcpu->mmio_data, val, bytes);
1110
1111         return X86EMUL_CONTINUE;
1112 }
1113
1114 static int emulator_write_emulated(unsigned long addr,
1115                                    const void *val,
1116                                    unsigned int bytes,
1117                                    struct x86_emulate_ctxt *ctxt)
1118 {
1119         /* Crossing a page boundary? */
1120         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1121                 int rc, now;
1122
1123                 now = -addr & ~PAGE_MASK;
1124                 rc = emulator_write_emulated_onepage(addr, val, now, ctxt);
1125                 if (rc != X86EMUL_CONTINUE)
1126                         return rc;
1127                 addr += now;
1128                 val += now;
1129                 bytes -= now;
1130         }
1131         return emulator_write_emulated_onepage(addr, val, bytes, ctxt);
1132 }
1133
1134 static int emulator_cmpxchg_emulated(unsigned long addr,
1135                                      const void *old,
1136                                      const void *new,
1137                                      unsigned int bytes,
1138                                      struct x86_emulate_ctxt *ctxt)
1139 {
1140         static int reported;
1141
1142         if (!reported) {
1143                 reported = 1;
1144                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1145         }
1146         return emulator_write_emulated(addr, new, bytes, ctxt);
1147 }
1148
1149 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1150 {
1151         return kvm_arch_ops->get_segment_base(vcpu, seg);
1152 }
1153
1154 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1155 {
1156         return X86EMUL_CONTINUE;
1157 }
1158
1159 int emulate_clts(struct kvm_vcpu *vcpu)
1160 {
1161         unsigned long cr0;
1162
1163         cr0 = vcpu->cr0 & ~CR0_TS_MASK;
1164         kvm_arch_ops->set_cr0(vcpu, cr0);
1165         return X86EMUL_CONTINUE;
1166 }
1167
1168 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1169 {
1170         struct kvm_vcpu *vcpu = ctxt->vcpu;
1171
1172         switch (dr) {
1173         case 0 ... 3:
1174                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1175                 return X86EMUL_CONTINUE;
1176         default:
1177                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1178                        __FUNCTION__, dr);
1179                 return X86EMUL_UNHANDLEABLE;
1180         }
1181 }
1182
1183 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1184 {
1185         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1186         int exception;
1187
1188         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1189         if (exception) {
1190                 /* FIXME: better handling */
1191                 return X86EMUL_UNHANDLEABLE;
1192         }
1193         return X86EMUL_CONTINUE;
1194 }
1195
1196 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1197 {
1198         static int reported;
1199         u8 opcodes[4];
1200         unsigned long rip = ctxt->vcpu->rip;
1201         unsigned long rip_linear;
1202
1203         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1204
1205         if (reported)
1206                 return;
1207
1208         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1209
1210         printk(KERN_ERR "emulation failed but !mmio_needed?"
1211                " rip %lx %02x %02x %02x %02x\n",
1212                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1213         reported = 1;
1214 }
1215
1216 struct x86_emulate_ops emulate_ops = {
1217         .read_std            = emulator_read_std,
1218         .write_std           = emulator_write_std,
1219         .read_emulated       = emulator_read_emulated,
1220         .write_emulated      = emulator_write_emulated,
1221         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1222 };
1223
1224 int emulate_instruction(struct kvm_vcpu *vcpu,
1225                         struct kvm_run *run,
1226                         unsigned long cr2,
1227                         u16 error_code)
1228 {
1229         struct x86_emulate_ctxt emulate_ctxt;
1230         int r;
1231         int cs_db, cs_l;
1232
1233         vcpu->mmio_fault_cr2 = cr2;
1234         kvm_arch_ops->cache_regs(vcpu);
1235
1236         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1237
1238         emulate_ctxt.vcpu = vcpu;
1239         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1240         emulate_ctxt.cr2 = cr2;
1241         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1242                 ? X86EMUL_MODE_REAL : cs_l
1243                 ? X86EMUL_MODE_PROT64 : cs_db
1244                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1245
1246         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1247                 emulate_ctxt.cs_base = 0;
1248                 emulate_ctxt.ds_base = 0;
1249                 emulate_ctxt.es_base = 0;
1250                 emulate_ctxt.ss_base = 0;
1251         } else {
1252                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1253                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1254                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1255                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1256         }
1257
1258         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1259         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1260
1261         vcpu->mmio_is_write = 0;
1262         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1263
1264         if ((r || vcpu->mmio_is_write) && run) {
1265                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1266                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1267                 run->mmio.len = vcpu->mmio_size;
1268                 run->mmio.is_write = vcpu->mmio_is_write;
1269         }
1270
1271         if (r) {
1272                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1273                         return EMULATE_DONE;
1274                 if (!vcpu->mmio_needed) {
1275                         report_emulation_failure(&emulate_ctxt);
1276                         return EMULATE_FAIL;
1277                 }
1278                 return EMULATE_DO_MMIO;
1279         }
1280
1281         kvm_arch_ops->decache_regs(vcpu);
1282         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1283
1284         if (vcpu->mmio_is_write) {
1285                 vcpu->mmio_needed = 0;
1286                 return EMULATE_DO_MMIO;
1287         }
1288
1289         return EMULATE_DONE;
1290 }
1291 EXPORT_SYMBOL_GPL(emulate_instruction);
1292
1293 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1294 {
1295         if (vcpu->irq_summary)
1296                 return 1;
1297
1298         vcpu->run->exit_reason = KVM_EXIT_HLT;
1299         ++vcpu->stat.halt_exits;
1300         return 0;
1301 }
1302 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1303
1304 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1305 {
1306         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1307
1308         kvm_arch_ops->cache_regs(vcpu);
1309         ret = -KVM_EINVAL;
1310 #ifdef CONFIG_X86_64
1311         if (is_long_mode(vcpu)) {
1312                 nr = vcpu->regs[VCPU_REGS_RAX];
1313                 a0 = vcpu->regs[VCPU_REGS_RDI];
1314                 a1 = vcpu->regs[VCPU_REGS_RSI];
1315                 a2 = vcpu->regs[VCPU_REGS_RDX];
1316                 a3 = vcpu->regs[VCPU_REGS_RCX];
1317                 a4 = vcpu->regs[VCPU_REGS_R8];
1318                 a5 = vcpu->regs[VCPU_REGS_R9];
1319         } else
1320 #endif
1321         {
1322                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1323                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1324                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1325                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1326                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1327                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1328                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1329         }
1330         switch (nr) {
1331         default:
1332                 run->hypercall.args[0] = a0;
1333                 run->hypercall.args[1] = a1;
1334                 run->hypercall.args[2] = a2;
1335                 run->hypercall.args[3] = a3;
1336                 run->hypercall.args[4] = a4;
1337                 run->hypercall.args[5] = a5;
1338                 run->hypercall.ret = ret;
1339                 run->hypercall.longmode = is_long_mode(vcpu);
1340                 kvm_arch_ops->decache_regs(vcpu);
1341                 return 0;
1342         }
1343         vcpu->regs[VCPU_REGS_RAX] = ret;
1344         kvm_arch_ops->decache_regs(vcpu);
1345         return 1;
1346 }
1347 EXPORT_SYMBOL_GPL(kvm_hypercall);
1348
1349 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1350 {
1351         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1352 }
1353
1354 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1355 {
1356         struct descriptor_table dt = { limit, base };
1357
1358         kvm_arch_ops->set_gdt(vcpu, &dt);
1359 }
1360
1361 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1362 {
1363         struct descriptor_table dt = { limit, base };
1364
1365         kvm_arch_ops->set_idt(vcpu, &dt);
1366 }
1367
1368 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1369                    unsigned long *rflags)
1370 {
1371         lmsw(vcpu, msw);
1372         *rflags = kvm_arch_ops->get_rflags(vcpu);
1373 }
1374
1375 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1376 {
1377         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1378         switch (cr) {
1379         case 0:
1380                 return vcpu->cr0;
1381         case 2:
1382                 return vcpu->cr2;
1383         case 3:
1384                 return vcpu->cr3;
1385         case 4:
1386                 return vcpu->cr4;
1387         default:
1388                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1389                 return 0;
1390         }
1391 }
1392
1393 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1394                      unsigned long *rflags)
1395 {
1396         switch (cr) {
1397         case 0:
1398                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1399                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1400                 break;
1401         case 2:
1402                 vcpu->cr2 = val;
1403                 break;
1404         case 3:
1405                 set_cr3(vcpu, val);
1406                 break;
1407         case 4:
1408                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1409                 break;
1410         default:
1411                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1412         }
1413 }
1414
1415 /*
1416  * Register the para guest with the host:
1417  */
1418 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1419 {
1420         struct kvm_vcpu_para_state *para_state;
1421         hpa_t para_state_hpa, hypercall_hpa;
1422         struct page *para_state_page;
1423         unsigned char *hypercall;
1424         gpa_t hypercall_gpa;
1425
1426         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1427         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1428
1429         /*
1430          * Needs to be page aligned:
1431          */
1432         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1433                 goto err_gp;
1434
1435         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1436         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1437         if (is_error_hpa(para_state_hpa))
1438                 goto err_gp;
1439
1440         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1441         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1442         para_state = kmap_atomic(para_state_page, KM_USER0);
1443
1444         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1445         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1446
1447         para_state->host_version = KVM_PARA_API_VERSION;
1448         /*
1449          * We cannot support guests that try to register themselves
1450          * with a newer API version than the host supports:
1451          */
1452         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1453                 para_state->ret = -KVM_EINVAL;
1454                 goto err_kunmap_skip;
1455         }
1456
1457         hypercall_gpa = para_state->hypercall_gpa;
1458         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1459         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1460         if (is_error_hpa(hypercall_hpa)) {
1461                 para_state->ret = -KVM_EINVAL;
1462                 goto err_kunmap_skip;
1463         }
1464
1465         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1466         vcpu->para_state_page = para_state_page;
1467         vcpu->para_state_gpa = para_state_gpa;
1468         vcpu->hypercall_gpa = hypercall_gpa;
1469
1470         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1471         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1472                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1473         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1474         kunmap_atomic(hypercall, KM_USER1);
1475
1476         para_state->ret = 0;
1477 err_kunmap_skip:
1478         kunmap_atomic(para_state, KM_USER0);
1479         return 0;
1480 err_gp:
1481         return 1;
1482 }
1483
1484 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1485 {
1486         u64 data;
1487
1488         switch (msr) {
1489         case 0xc0010010: /* SYSCFG */
1490         case 0xc0010015: /* HWCR */
1491         case MSR_IA32_PLATFORM_ID:
1492         case MSR_IA32_P5_MC_ADDR:
1493         case MSR_IA32_P5_MC_TYPE:
1494         case MSR_IA32_MC0_CTL:
1495         case MSR_IA32_MCG_STATUS:
1496         case MSR_IA32_MCG_CAP:
1497         case MSR_IA32_MC0_MISC:
1498         case MSR_IA32_MC0_MISC+4:
1499         case MSR_IA32_MC0_MISC+8:
1500         case MSR_IA32_MC0_MISC+12:
1501         case MSR_IA32_MC0_MISC+16:
1502         case MSR_IA32_UCODE_REV:
1503         case MSR_IA32_PERF_STATUS:
1504         case MSR_IA32_EBL_CR_POWERON:
1505                 /* MTRR registers */
1506         case 0xfe:
1507         case 0x200 ... 0x2ff:
1508                 data = 0;
1509                 break;
1510         case 0xcd: /* fsb frequency */
1511                 data = 3;
1512                 break;
1513         case MSR_IA32_APICBASE:
1514                 data = vcpu->apic_base;
1515                 break;
1516         case MSR_IA32_MISC_ENABLE:
1517                 data = vcpu->ia32_misc_enable_msr;
1518                 break;
1519 #ifdef CONFIG_X86_64
1520         case MSR_EFER:
1521                 data = vcpu->shadow_efer;
1522                 break;
1523 #endif
1524         default:
1525                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1526                 return 1;
1527         }
1528         *pdata = data;
1529         return 0;
1530 }
1531 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1532
1533 /*
1534  * Reads an msr value (of 'msr_index') into 'pdata'.
1535  * Returns 0 on success, non-0 otherwise.
1536  * Assumes vcpu_load() was already called.
1537  */
1538 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1539 {
1540         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1541 }
1542
1543 #ifdef CONFIG_X86_64
1544
1545 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1546 {
1547         if (efer & EFER_RESERVED_BITS) {
1548                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1549                        efer);
1550                 inject_gp(vcpu);
1551                 return;
1552         }
1553
1554         if (is_paging(vcpu)
1555             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1556                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1557                 inject_gp(vcpu);
1558                 return;
1559         }
1560
1561         kvm_arch_ops->set_efer(vcpu, efer);
1562
1563         efer &= ~EFER_LMA;
1564         efer |= vcpu->shadow_efer & EFER_LMA;
1565
1566         vcpu->shadow_efer = efer;
1567 }
1568
1569 #endif
1570
1571 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1572 {
1573         switch (msr) {
1574 #ifdef CONFIG_X86_64
1575         case MSR_EFER:
1576                 set_efer(vcpu, data);
1577                 break;
1578 #endif
1579         case MSR_IA32_MC0_STATUS:
1580                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1581                        __FUNCTION__, data);
1582                 break;
1583         case MSR_IA32_MCG_STATUS:
1584                 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1585                         __FUNCTION__, data);
1586                 break;
1587         case MSR_IA32_UCODE_REV:
1588         case MSR_IA32_UCODE_WRITE:
1589         case 0x200 ... 0x2ff: /* MTRRs */
1590                 break;
1591         case MSR_IA32_APICBASE:
1592                 vcpu->apic_base = data;
1593                 break;
1594         case MSR_IA32_MISC_ENABLE:
1595                 vcpu->ia32_misc_enable_msr = data;
1596                 break;
1597         /*
1598          * This is the 'probe whether the host is KVM' logic:
1599          */
1600         case MSR_KVM_API_MAGIC:
1601                 return vcpu_register_para(vcpu, data);
1602
1603         default:
1604                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1605                 return 1;
1606         }
1607         return 0;
1608 }
1609 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1610
1611 /*
1612  * Writes msr value into into the appropriate "register".
1613  * Returns 0 on success, non-0 otherwise.
1614  * Assumes vcpu_load() was already called.
1615  */
1616 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1617 {
1618         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1619 }
1620
1621 void kvm_resched(struct kvm_vcpu *vcpu)
1622 {
1623         if (!need_resched())
1624                 return;
1625         vcpu_put(vcpu);
1626         cond_resched();
1627         vcpu_load(vcpu);
1628 }
1629 EXPORT_SYMBOL_GPL(kvm_resched);
1630
1631 void load_msrs(struct vmx_msr_entry *e, int n)
1632 {
1633         int i;
1634
1635         for (i = 0; i < n; ++i)
1636                 wrmsrl(e[i].index, e[i].data);
1637 }
1638 EXPORT_SYMBOL_GPL(load_msrs);
1639
1640 void save_msrs(struct vmx_msr_entry *e, int n)
1641 {
1642         int i;
1643
1644         for (i = 0; i < n; ++i)
1645                 rdmsrl(e[i].index, e[i].data);
1646 }
1647 EXPORT_SYMBOL_GPL(save_msrs);
1648
1649 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1650 {
1651         int i;
1652         u32 function;
1653         struct kvm_cpuid_entry *e, *best;
1654
1655         kvm_arch_ops->cache_regs(vcpu);
1656         function = vcpu->regs[VCPU_REGS_RAX];
1657         vcpu->regs[VCPU_REGS_RAX] = 0;
1658         vcpu->regs[VCPU_REGS_RBX] = 0;
1659         vcpu->regs[VCPU_REGS_RCX] = 0;
1660         vcpu->regs[VCPU_REGS_RDX] = 0;
1661         best = NULL;
1662         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1663                 e = &vcpu->cpuid_entries[i];
1664                 if (e->function == function) {
1665                         best = e;
1666                         break;
1667                 }
1668                 /*
1669                  * Both basic or both extended?
1670                  */
1671                 if (((e->function ^ function) & 0x80000000) == 0)
1672                         if (!best || e->function > best->function)
1673                                 best = e;
1674         }
1675         if (best) {
1676                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1677                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1678                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1679                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1680         }
1681         kvm_arch_ops->decache_regs(vcpu);
1682         kvm_arch_ops->skip_emulated_instruction(vcpu);
1683 }
1684 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1685
1686 static int pio_copy_data(struct kvm_vcpu *vcpu)
1687 {
1688         void *p = vcpu->pio_data;
1689         void *q;
1690         unsigned bytes;
1691         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1692
1693         kvm_arch_ops->vcpu_put(vcpu);
1694         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1695                  PAGE_KERNEL);
1696         if (!q) {
1697                 kvm_arch_ops->vcpu_load(vcpu);
1698                 free_pio_guest_pages(vcpu);
1699                 return -ENOMEM;
1700         }
1701         q += vcpu->pio.guest_page_offset;
1702         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1703         if (vcpu->pio.in)
1704                 memcpy(q, p, bytes);
1705         else
1706                 memcpy(p, q, bytes);
1707         q -= vcpu->pio.guest_page_offset;
1708         vunmap(q);
1709         kvm_arch_ops->vcpu_load(vcpu);
1710         free_pio_guest_pages(vcpu);
1711         return 0;
1712 }
1713
1714 static int complete_pio(struct kvm_vcpu *vcpu)
1715 {
1716         struct kvm_pio_request *io = &vcpu->pio;
1717         long delta;
1718         int r;
1719
1720         kvm_arch_ops->cache_regs(vcpu);
1721
1722         if (!io->string) {
1723                 if (io->in)
1724                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1725                                io->size);
1726         } else {
1727                 if (io->in) {
1728                         r = pio_copy_data(vcpu);
1729                         if (r) {
1730                                 kvm_arch_ops->cache_regs(vcpu);
1731                                 return r;
1732                         }
1733                 }
1734
1735                 delta = 1;
1736                 if (io->rep) {
1737                         delta *= io->cur_count;
1738                         /*
1739                          * The size of the register should really depend on
1740                          * current address size.
1741                          */
1742                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1743                 }
1744                 if (io->down)
1745                         delta = -delta;
1746                 delta *= io->size;
1747                 if (io->in)
1748                         vcpu->regs[VCPU_REGS_RDI] += delta;
1749                 else
1750                         vcpu->regs[VCPU_REGS_RSI] += delta;
1751         }
1752
1753         kvm_arch_ops->decache_regs(vcpu);
1754
1755         io->count -= io->cur_count;
1756         io->cur_count = 0;
1757
1758         if (!io->count)
1759                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1760         return 0;
1761 }
1762
1763 static void kernel_pio(struct kvm_io_device *pio_dev,
1764                        struct kvm_vcpu *vcpu,
1765                        void *pd)
1766 {
1767         /* TODO: String I/O for in kernel device */
1768
1769         if (vcpu->pio.in)
1770                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1771                                   vcpu->pio.size,
1772                                   pd);
1773         else
1774                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1775                                    vcpu->pio.size,
1776                                    pd);
1777 }
1778
1779 static void pio_string_write(struct kvm_io_device *pio_dev,
1780                              struct kvm_vcpu *vcpu)
1781 {
1782         struct kvm_pio_request *io = &vcpu->pio;
1783         void *pd = vcpu->pio_data;
1784         int i;
1785
1786         for (i = 0; i < io->cur_count; i++) {
1787                 kvm_iodevice_write(pio_dev, io->port,
1788                                    io->size,
1789                                    pd);
1790                 pd += io->size;
1791         }
1792 }
1793
1794 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1795                   int size, unsigned long count, int string, int down,
1796                   gva_t address, int rep, unsigned port)
1797 {
1798         unsigned now, in_page;
1799         int i, ret = 0;
1800         int nr_pages = 1;
1801         struct page *page;
1802         struct kvm_io_device *pio_dev;
1803
1804         vcpu->run->exit_reason = KVM_EXIT_IO;
1805         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1806         vcpu->run->io.size = size;
1807         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1808         vcpu->run->io.count = count;
1809         vcpu->run->io.port = port;
1810         vcpu->pio.count = count;
1811         vcpu->pio.cur_count = count;
1812         vcpu->pio.size = size;
1813         vcpu->pio.in = in;
1814         vcpu->pio.port = port;
1815         vcpu->pio.string = string;
1816         vcpu->pio.down = down;
1817         vcpu->pio.guest_page_offset = offset_in_page(address);
1818         vcpu->pio.rep = rep;
1819
1820         pio_dev = vcpu_find_pio_dev(vcpu, port);
1821         if (!string) {
1822                 kvm_arch_ops->cache_regs(vcpu);
1823                 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1824                 kvm_arch_ops->decache_regs(vcpu);
1825                 if (pio_dev) {
1826                         kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1827                         complete_pio(vcpu);
1828                         return 1;
1829                 }
1830                 return 0;
1831         }
1832
1833         if (!count) {
1834                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1835                 return 1;
1836         }
1837
1838         now = min(count, PAGE_SIZE / size);
1839
1840         if (!down)
1841                 in_page = PAGE_SIZE - offset_in_page(address);
1842         else
1843                 in_page = offset_in_page(address) + size;
1844         now = min(count, (unsigned long)in_page / size);
1845         if (!now) {
1846                 /*
1847                  * String I/O straddles page boundary.  Pin two guest pages
1848                  * so that we satisfy atomicity constraints.  Do just one
1849                  * transaction to avoid complexity.
1850                  */
1851                 nr_pages = 2;
1852                 now = 1;
1853         }
1854         if (down) {
1855                 /*
1856                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1857                  */
1858                 printk(KERN_ERR "kvm: guest string pio down\n");
1859                 inject_gp(vcpu);
1860                 return 1;
1861         }
1862         vcpu->run->io.count = now;
1863         vcpu->pio.cur_count = now;
1864
1865         for (i = 0; i < nr_pages; ++i) {
1866                 spin_lock(&vcpu->kvm->lock);
1867                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1868                 if (page)
1869                         get_page(page);
1870                 vcpu->pio.guest_pages[i] = page;
1871                 spin_unlock(&vcpu->kvm->lock);
1872                 if (!page) {
1873                         inject_gp(vcpu);
1874                         free_pio_guest_pages(vcpu);
1875                         return 1;
1876                 }
1877         }
1878
1879         if (!vcpu->pio.in) {
1880                 /* string PIO write */
1881                 ret = pio_copy_data(vcpu);
1882                 if (ret >= 0 && pio_dev) {
1883                         pio_string_write(pio_dev, vcpu);
1884                         complete_pio(vcpu);
1885                         if (vcpu->pio.count == 0)
1886                                 ret = 1;
1887                 }
1888         } else if (pio_dev)
1889                 printk(KERN_ERR "no string pio read support yet, "
1890                        "port %x size %d count %ld\n",
1891                         port, size, count);
1892
1893         return ret;
1894 }
1895 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1896
1897 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1898 {
1899         int r;
1900         sigset_t sigsaved;
1901
1902         vcpu_load(vcpu);
1903
1904         if (vcpu->sigset_active)
1905                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1906
1907         /* re-sync apic's tpr */
1908         vcpu->cr8 = kvm_run->cr8;
1909
1910         if (vcpu->pio.cur_count) {
1911                 r = complete_pio(vcpu);
1912                 if (r)
1913                         goto out;
1914         }
1915
1916         if (vcpu->mmio_needed) {
1917                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1918                 vcpu->mmio_read_completed = 1;
1919                 vcpu->mmio_needed = 0;
1920                 r = emulate_instruction(vcpu, kvm_run,
1921                                         vcpu->mmio_fault_cr2, 0);
1922                 if (r == EMULATE_DO_MMIO) {
1923                         /*
1924                          * Read-modify-write.  Back to userspace.
1925                          */
1926                         kvm_run->exit_reason = KVM_EXIT_MMIO;
1927                         r = 0;
1928                         goto out;
1929                 }
1930         }
1931
1932         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1933                 kvm_arch_ops->cache_regs(vcpu);
1934                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1935                 kvm_arch_ops->decache_regs(vcpu);
1936         }
1937
1938         r = kvm_arch_ops->run(vcpu, kvm_run);
1939
1940 out:
1941         if (vcpu->sigset_active)
1942                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1943
1944         vcpu_put(vcpu);
1945         return r;
1946 }
1947
1948 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1949                                    struct kvm_regs *regs)
1950 {
1951         vcpu_load(vcpu);
1952
1953         kvm_arch_ops->cache_regs(vcpu);
1954
1955         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1956         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1957         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1958         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1959         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1960         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1961         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1962         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1963 #ifdef CONFIG_X86_64
1964         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1965         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1966         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1967         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1968         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1969         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1970         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1971         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1972 #endif
1973
1974         regs->rip = vcpu->rip;
1975         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1976
1977         /*
1978          * Don't leak debug flags in case they were set for guest debugging
1979          */
1980         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1981                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1982
1983         vcpu_put(vcpu);
1984
1985         return 0;
1986 }
1987
1988 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1989                                    struct kvm_regs *regs)
1990 {
1991         vcpu_load(vcpu);
1992
1993         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1994         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1995         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1996         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1997         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1998         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1999         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2000         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2001 #ifdef CONFIG_X86_64
2002         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2003         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2004         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2005         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2006         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2007         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2008         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2009         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2010 #endif
2011
2012         vcpu->rip = regs->rip;
2013         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
2014
2015         kvm_arch_ops->decache_regs(vcpu);
2016
2017         vcpu_put(vcpu);
2018
2019         return 0;
2020 }
2021
2022 static void get_segment(struct kvm_vcpu *vcpu,
2023                         struct kvm_segment *var, int seg)
2024 {
2025         return kvm_arch_ops->get_segment(vcpu, var, seg);
2026 }
2027
2028 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2029                                     struct kvm_sregs *sregs)
2030 {
2031         struct descriptor_table dt;
2032
2033         vcpu_load(vcpu);
2034
2035         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2036         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2037         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2038         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2039         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2040         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2041
2042         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2043         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2044
2045         kvm_arch_ops->get_idt(vcpu, &dt);
2046         sregs->idt.limit = dt.limit;
2047         sregs->idt.base = dt.base;
2048         kvm_arch_ops->get_gdt(vcpu, &dt);
2049         sregs->gdt.limit = dt.limit;
2050         sregs->gdt.base = dt.base;
2051
2052         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2053         sregs->cr0 = vcpu->cr0;
2054         sregs->cr2 = vcpu->cr2;
2055         sregs->cr3 = vcpu->cr3;
2056         sregs->cr4 = vcpu->cr4;
2057         sregs->cr8 = vcpu->cr8;
2058         sregs->efer = vcpu->shadow_efer;
2059         sregs->apic_base = vcpu->apic_base;
2060
2061         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2062                sizeof sregs->interrupt_bitmap);
2063
2064         vcpu_put(vcpu);
2065
2066         return 0;
2067 }
2068
2069 static void set_segment(struct kvm_vcpu *vcpu,
2070                         struct kvm_segment *var, int seg)
2071 {
2072         return kvm_arch_ops->set_segment(vcpu, var, seg);
2073 }
2074
2075 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2076                                     struct kvm_sregs *sregs)
2077 {
2078         int mmu_reset_needed = 0;
2079         int i;
2080         struct descriptor_table dt;
2081
2082         vcpu_load(vcpu);
2083
2084         dt.limit = sregs->idt.limit;
2085         dt.base = sregs->idt.base;
2086         kvm_arch_ops->set_idt(vcpu, &dt);
2087         dt.limit = sregs->gdt.limit;
2088         dt.base = sregs->gdt.base;
2089         kvm_arch_ops->set_gdt(vcpu, &dt);
2090
2091         vcpu->cr2 = sregs->cr2;
2092         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2093         vcpu->cr3 = sregs->cr3;
2094
2095         vcpu->cr8 = sregs->cr8;
2096
2097         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2098 #ifdef CONFIG_X86_64
2099         kvm_arch_ops->set_efer(vcpu, sregs->efer);
2100 #endif
2101         vcpu->apic_base = sregs->apic_base;
2102
2103         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2104
2105         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2106         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2107
2108         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2109         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2110         if (!is_long_mode(vcpu) && is_pae(vcpu))
2111                 load_pdptrs(vcpu, vcpu->cr3);
2112
2113         if (mmu_reset_needed)
2114                 kvm_mmu_reset_context(vcpu);
2115
2116         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2117                sizeof vcpu->irq_pending);
2118         vcpu->irq_summary = 0;
2119         for (i = 0; i < NR_IRQ_WORDS; ++i)
2120                 if (vcpu->irq_pending[i])
2121                         __set_bit(i, &vcpu->irq_summary);
2122
2123         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2124         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2125         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2126         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2127         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2128         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2129
2130         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2131         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2132
2133         vcpu_put(vcpu);
2134
2135         return 0;
2136 }
2137
2138 /*
2139  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2140  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2141  *
2142  * This list is modified at module load time to reflect the
2143  * capabilities of the host cpu.
2144  */
2145 static u32 msrs_to_save[] = {
2146         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2147         MSR_K6_STAR,
2148 #ifdef CONFIG_X86_64
2149         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2150 #endif
2151         MSR_IA32_TIME_STAMP_COUNTER,
2152 };
2153
2154 static unsigned num_msrs_to_save;
2155
2156 static u32 emulated_msrs[] = {
2157         MSR_IA32_MISC_ENABLE,
2158 };
2159
2160 static __init void kvm_init_msr_list(void)
2161 {
2162         u32 dummy[2];
2163         unsigned i, j;
2164
2165         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2166                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2167                         continue;
2168                 if (j < i)
2169                         msrs_to_save[j] = msrs_to_save[i];
2170                 j++;
2171         }
2172         num_msrs_to_save = j;
2173 }
2174
2175 /*
2176  * Adapt set_msr() to msr_io()'s calling convention
2177  */
2178 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2179 {
2180         return kvm_set_msr(vcpu, index, *data);
2181 }
2182
2183 /*
2184  * Read or write a bunch of msrs. All parameters are kernel addresses.
2185  *
2186  * @return number of msrs set successfully.
2187  */
2188 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2189                     struct kvm_msr_entry *entries,
2190                     int (*do_msr)(struct kvm_vcpu *vcpu,
2191                                   unsigned index, u64 *data))
2192 {
2193         int i;
2194
2195         vcpu_load(vcpu);
2196
2197         for (i = 0; i < msrs->nmsrs; ++i)
2198                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2199                         break;
2200
2201         vcpu_put(vcpu);
2202
2203         return i;
2204 }
2205
2206 /*
2207  * Read or write a bunch of msrs. Parameters are user addresses.
2208  *
2209  * @return number of msrs set successfully.
2210  */
2211 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2212                   int (*do_msr)(struct kvm_vcpu *vcpu,
2213                                 unsigned index, u64 *data),
2214                   int writeback)
2215 {
2216         struct kvm_msrs msrs;
2217         struct kvm_msr_entry *entries;
2218         int r, n;
2219         unsigned size;
2220
2221         r = -EFAULT;
2222         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2223                 goto out;
2224
2225         r = -E2BIG;
2226         if (msrs.nmsrs >= MAX_IO_MSRS)
2227                 goto out;
2228
2229         r = -ENOMEM;
2230         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2231         entries = vmalloc(size);
2232         if (!entries)
2233                 goto out;
2234
2235         r = -EFAULT;
2236         if (copy_from_user(entries, user_msrs->entries, size))
2237                 goto out_free;
2238
2239         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2240         if (r < 0)
2241                 goto out_free;
2242
2243         r = -EFAULT;
2244         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2245                 goto out_free;
2246
2247         r = n;
2248
2249 out_free:
2250         vfree(entries);
2251 out:
2252         return r;
2253 }
2254
2255 /*
2256  * Translate a guest virtual address to a guest physical address.
2257  */
2258 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2259                                     struct kvm_translation *tr)
2260 {
2261         unsigned long vaddr = tr->linear_address;
2262         gpa_t gpa;
2263
2264         vcpu_load(vcpu);
2265         spin_lock(&vcpu->kvm->lock);
2266         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2267         tr->physical_address = gpa;
2268         tr->valid = gpa != UNMAPPED_GVA;
2269         tr->writeable = 1;
2270         tr->usermode = 0;
2271         spin_unlock(&vcpu->kvm->lock);
2272         vcpu_put(vcpu);
2273
2274         return 0;
2275 }
2276
2277 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2278                                     struct kvm_interrupt *irq)
2279 {
2280         if (irq->irq < 0 || irq->irq >= 256)
2281                 return -EINVAL;
2282         vcpu_load(vcpu);
2283
2284         set_bit(irq->irq, vcpu->irq_pending);
2285         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2286
2287         vcpu_put(vcpu);
2288
2289         return 0;
2290 }
2291
2292 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2293                                       struct kvm_debug_guest *dbg)
2294 {
2295         int r;
2296
2297         vcpu_load(vcpu);
2298
2299         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2300
2301         vcpu_put(vcpu);
2302
2303         return r;
2304 }
2305
2306 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2307                                     unsigned long address,
2308                                     int *type)
2309 {
2310         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2311         unsigned long pgoff;
2312         struct page *page;
2313
2314         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2315         if (pgoff == 0)
2316                 page = virt_to_page(vcpu->run);
2317         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2318                 page = virt_to_page(vcpu->pio_data);
2319         else
2320                 return NOPAGE_SIGBUS;
2321         get_page(page);
2322         if (type != NULL)
2323                 *type = VM_FAULT_MINOR;
2324
2325         return page;
2326 }
2327
2328 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2329         .nopage = kvm_vcpu_nopage,
2330 };
2331
2332 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2333 {
2334         vma->vm_ops = &kvm_vcpu_vm_ops;
2335         return 0;
2336 }
2337
2338 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2339 {
2340         struct kvm_vcpu *vcpu = filp->private_data;
2341
2342         fput(vcpu->kvm->filp);
2343         return 0;
2344 }
2345
2346 static struct file_operations kvm_vcpu_fops = {
2347         .release        = kvm_vcpu_release,
2348         .unlocked_ioctl = kvm_vcpu_ioctl,
2349         .compat_ioctl   = kvm_vcpu_ioctl,
2350         .mmap           = kvm_vcpu_mmap,
2351 };
2352
2353 /*
2354  * Allocates an inode for the vcpu.
2355  */
2356 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2357 {
2358         int fd, r;
2359         struct inode *inode;
2360         struct file *file;
2361
2362         r = anon_inode_getfd(&fd, &inode, &file,
2363                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2364         if (r)
2365                 return r;
2366         atomic_inc(&vcpu->kvm->filp->f_count);
2367         return fd;
2368 }
2369
2370 /*
2371  * Creates some virtual cpus.  Good luck creating more than one.
2372  */
2373 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2374 {
2375         int r;
2376         struct kvm_vcpu *vcpu;
2377         struct page *page;
2378
2379         r = -EINVAL;
2380         if (!valid_vcpu(n))
2381                 goto out;
2382
2383         vcpu = &kvm->vcpus[n];
2384         vcpu->vcpu_id = n;
2385
2386         mutex_lock(&vcpu->mutex);
2387
2388         if (vcpu->vmcs) {
2389                 mutex_unlock(&vcpu->mutex);
2390                 return -EEXIST;
2391         }
2392
2393         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2394         r = -ENOMEM;
2395         if (!page)
2396                 goto out_unlock;
2397         vcpu->run = page_address(page);
2398
2399         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2400         r = -ENOMEM;
2401         if (!page)
2402                 goto out_free_run;
2403         vcpu->pio_data = page_address(page);
2404
2405         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2406                                            FX_IMAGE_ALIGN);
2407         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2408         vcpu->cr0 = 0x10;
2409
2410         r = kvm_arch_ops->vcpu_create(vcpu);
2411         if (r < 0)
2412                 goto out_free_vcpus;
2413
2414         r = kvm_mmu_create(vcpu);
2415         if (r < 0)
2416                 goto out_free_vcpus;
2417
2418         kvm_arch_ops->vcpu_load(vcpu);
2419         r = kvm_mmu_setup(vcpu);
2420         if (r >= 0)
2421                 r = kvm_arch_ops->vcpu_setup(vcpu);
2422         vcpu_put(vcpu);
2423
2424         if (r < 0)
2425                 goto out_free_vcpus;
2426
2427         r = create_vcpu_fd(vcpu);
2428         if (r < 0)
2429                 goto out_free_vcpus;
2430
2431         spin_lock(&kvm_lock);
2432         if (n >= kvm->nvcpus)
2433                 kvm->nvcpus = n + 1;
2434         spin_unlock(&kvm_lock);
2435
2436         return r;
2437
2438 out_free_vcpus:
2439         kvm_free_vcpu(vcpu);
2440 out_free_run:
2441         free_page((unsigned long)vcpu->run);
2442         vcpu->run = NULL;
2443 out_unlock:
2444         mutex_unlock(&vcpu->mutex);
2445 out:
2446         return r;
2447 }
2448
2449 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2450 {
2451         u64 efer;
2452         int i;
2453         struct kvm_cpuid_entry *e, *entry;
2454
2455         rdmsrl(MSR_EFER, efer);
2456         entry = NULL;
2457         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2458                 e = &vcpu->cpuid_entries[i];
2459                 if (e->function == 0x80000001) {
2460                         entry = e;
2461                         break;
2462                 }
2463         }
2464         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2465                 entry->edx &= ~(1 << 20);
2466                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2467         }
2468 }
2469
2470 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2471                                     struct kvm_cpuid *cpuid,
2472                                     struct kvm_cpuid_entry __user *entries)
2473 {
2474         int r;
2475
2476         r = -E2BIG;
2477         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2478                 goto out;
2479         r = -EFAULT;
2480         if (copy_from_user(&vcpu->cpuid_entries, entries,
2481                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2482                 goto out;
2483         vcpu->cpuid_nent = cpuid->nent;
2484         cpuid_fix_nx_cap(vcpu);
2485         return 0;
2486
2487 out:
2488         return r;
2489 }
2490
2491 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2492 {
2493         if (sigset) {
2494                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2495                 vcpu->sigset_active = 1;
2496                 vcpu->sigset = *sigset;
2497         } else
2498                 vcpu->sigset_active = 0;
2499         return 0;
2500 }
2501
2502 /*
2503  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2504  * we have asm/x86/processor.h
2505  */
2506 struct fxsave {
2507         u16     cwd;
2508         u16     swd;
2509         u16     twd;
2510         u16     fop;
2511         u64     rip;
2512         u64     rdp;
2513         u32     mxcsr;
2514         u32     mxcsr_mask;
2515         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2516 #ifdef CONFIG_X86_64
2517         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2518 #else
2519         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2520 #endif
2521 };
2522
2523 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2524 {
2525         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2526
2527         vcpu_load(vcpu);
2528
2529         memcpy(fpu->fpr, fxsave->st_space, 128);
2530         fpu->fcw = fxsave->cwd;
2531         fpu->fsw = fxsave->swd;
2532         fpu->ftwx = fxsave->twd;
2533         fpu->last_opcode = fxsave->fop;
2534         fpu->last_ip = fxsave->rip;
2535         fpu->last_dp = fxsave->rdp;
2536         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2537
2538         vcpu_put(vcpu);
2539
2540         return 0;
2541 }
2542
2543 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2544 {
2545         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2546
2547         vcpu_load(vcpu);
2548
2549         memcpy(fxsave->st_space, fpu->fpr, 128);
2550         fxsave->cwd = fpu->fcw;
2551         fxsave->swd = fpu->fsw;
2552         fxsave->twd = fpu->ftwx;
2553         fxsave->fop = fpu->last_opcode;
2554         fxsave->rip = fpu->last_ip;
2555         fxsave->rdp = fpu->last_dp;
2556         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2557
2558         vcpu_put(vcpu);
2559
2560         return 0;
2561 }
2562
2563 static long kvm_vcpu_ioctl(struct file *filp,
2564                            unsigned int ioctl, unsigned long arg)
2565 {
2566         struct kvm_vcpu *vcpu = filp->private_data;
2567         void __user *argp = (void __user *)arg;
2568         int r = -EINVAL;
2569
2570         switch (ioctl) {
2571         case KVM_RUN:
2572                 r = -EINVAL;
2573                 if (arg)
2574                         goto out;
2575                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2576                 break;
2577         case KVM_GET_REGS: {
2578                 struct kvm_regs kvm_regs;
2579
2580                 memset(&kvm_regs, 0, sizeof kvm_regs);
2581                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2582                 if (r)
2583                         goto out;
2584                 r = -EFAULT;
2585                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2586                         goto out;
2587                 r = 0;
2588                 break;
2589         }
2590         case KVM_SET_REGS: {
2591                 struct kvm_regs kvm_regs;
2592
2593                 r = -EFAULT;
2594                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2595                         goto out;
2596                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2597                 if (r)
2598                         goto out;
2599                 r = 0;
2600                 break;
2601         }
2602         case KVM_GET_SREGS: {
2603                 struct kvm_sregs kvm_sregs;
2604
2605                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2606                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2607                 if (r)
2608                         goto out;
2609                 r = -EFAULT;
2610                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2611                         goto out;
2612                 r = 0;
2613                 break;
2614         }
2615         case KVM_SET_SREGS: {
2616                 struct kvm_sregs kvm_sregs;
2617
2618                 r = -EFAULT;
2619                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2620                         goto out;
2621                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2622                 if (r)
2623                         goto out;
2624                 r = 0;
2625                 break;
2626         }
2627         case KVM_TRANSLATE: {
2628                 struct kvm_translation tr;
2629
2630                 r = -EFAULT;
2631                 if (copy_from_user(&tr, argp, sizeof tr))
2632                         goto out;
2633                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2634                 if (r)
2635                         goto out;
2636                 r = -EFAULT;
2637                 if (copy_to_user(argp, &tr, sizeof tr))
2638                         goto out;
2639                 r = 0;
2640                 break;
2641         }
2642         case KVM_INTERRUPT: {
2643                 struct kvm_interrupt irq;
2644
2645                 r = -EFAULT;
2646                 if (copy_from_user(&irq, argp, sizeof irq))
2647                         goto out;
2648                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2649                 if (r)
2650                         goto out;
2651                 r = 0;
2652                 break;
2653         }
2654         case KVM_DEBUG_GUEST: {
2655                 struct kvm_debug_guest dbg;
2656
2657                 r = -EFAULT;
2658                 if (copy_from_user(&dbg, argp, sizeof dbg))
2659                         goto out;
2660                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2661                 if (r)
2662                         goto out;
2663                 r = 0;
2664                 break;
2665         }
2666         case KVM_GET_MSRS:
2667                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2668                 break;
2669         case KVM_SET_MSRS:
2670                 r = msr_io(vcpu, argp, do_set_msr, 0);
2671                 break;
2672         case KVM_SET_CPUID: {
2673                 struct kvm_cpuid __user *cpuid_arg = argp;
2674                 struct kvm_cpuid cpuid;
2675
2676                 r = -EFAULT;
2677                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2678                         goto out;
2679                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2680                 if (r)
2681                         goto out;
2682                 break;
2683         }
2684         case KVM_SET_SIGNAL_MASK: {
2685                 struct kvm_signal_mask __user *sigmask_arg = argp;
2686                 struct kvm_signal_mask kvm_sigmask;
2687                 sigset_t sigset, *p;
2688
2689                 p = NULL;
2690                 if (argp) {
2691                         r = -EFAULT;
2692                         if (copy_from_user(&kvm_sigmask, argp,
2693                                            sizeof kvm_sigmask))
2694                                 goto out;
2695                         r = -EINVAL;
2696                         if (kvm_sigmask.len != sizeof sigset)
2697                                 goto out;
2698                         r = -EFAULT;
2699                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2700                                            sizeof sigset))
2701                                 goto out;
2702                         p = &sigset;
2703                 }
2704                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2705                 break;
2706         }
2707         case KVM_GET_FPU: {
2708                 struct kvm_fpu fpu;
2709
2710                 memset(&fpu, 0, sizeof fpu);
2711                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2712                 if (r)
2713                         goto out;
2714                 r = -EFAULT;
2715                 if (copy_to_user(argp, &fpu, sizeof fpu))
2716                         goto out;
2717                 r = 0;
2718                 break;
2719         }
2720         case KVM_SET_FPU: {
2721                 struct kvm_fpu fpu;
2722
2723                 r = -EFAULT;
2724                 if (copy_from_user(&fpu, argp, sizeof fpu))
2725                         goto out;
2726                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2727                 if (r)
2728                         goto out;
2729                 r = 0;
2730                 break;
2731         }
2732         default:
2733                 ;
2734         }
2735 out:
2736         return r;
2737 }
2738
2739 static long kvm_vm_ioctl(struct file *filp,
2740                            unsigned int ioctl, unsigned long arg)
2741 {
2742         struct kvm *kvm = filp->private_data;
2743         void __user *argp = (void __user *)arg;
2744         int r = -EINVAL;
2745
2746         switch (ioctl) {
2747         case KVM_CREATE_VCPU:
2748                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2749                 if (r < 0)
2750                         goto out;
2751                 break;
2752         case KVM_SET_MEMORY_REGION: {
2753                 struct kvm_memory_region kvm_mem;
2754
2755                 r = -EFAULT;
2756                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2757                         goto out;
2758                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2759                 if (r)
2760                         goto out;
2761                 break;
2762         }
2763         case KVM_GET_DIRTY_LOG: {
2764                 struct kvm_dirty_log log;
2765
2766                 r = -EFAULT;
2767                 if (copy_from_user(&log, argp, sizeof log))
2768                         goto out;
2769                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2770                 if (r)
2771                         goto out;
2772                 break;
2773         }
2774         case KVM_SET_MEMORY_ALIAS: {
2775                 struct kvm_memory_alias alias;
2776
2777                 r = -EFAULT;
2778                 if (copy_from_user(&alias, argp, sizeof alias))
2779                         goto out;
2780                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2781                 if (r)
2782                         goto out;
2783                 break;
2784         }
2785         default:
2786                 ;
2787         }
2788 out:
2789         return r;
2790 }
2791
2792 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2793                                   unsigned long address,
2794                                   int *type)
2795 {
2796         struct kvm *kvm = vma->vm_file->private_data;
2797         unsigned long pgoff;
2798         struct page *page;
2799
2800         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2801         page = gfn_to_page(kvm, pgoff);
2802         if (!page)
2803                 return NOPAGE_SIGBUS;
2804         get_page(page);
2805         if (type != NULL)
2806                 *type = VM_FAULT_MINOR;
2807
2808         return page;
2809 }
2810
2811 static struct vm_operations_struct kvm_vm_vm_ops = {
2812         .nopage = kvm_vm_nopage,
2813 };
2814
2815 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2816 {
2817         vma->vm_ops = &kvm_vm_vm_ops;
2818         return 0;
2819 }
2820
2821 static struct file_operations kvm_vm_fops = {
2822         .release        = kvm_vm_release,
2823         .unlocked_ioctl = kvm_vm_ioctl,
2824         .compat_ioctl   = kvm_vm_ioctl,
2825         .mmap           = kvm_vm_mmap,
2826 };
2827
2828 static int kvm_dev_ioctl_create_vm(void)
2829 {
2830         int fd, r;
2831         struct inode *inode;
2832         struct file *file;
2833         struct kvm *kvm;
2834
2835         kvm = kvm_create_vm();
2836         if (IS_ERR(kvm))
2837                 return PTR_ERR(kvm);
2838         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2839         if (r) {
2840                 kvm_destroy_vm(kvm);
2841                 return r;
2842         }
2843
2844         kvm->filp = file;
2845
2846         return fd;
2847 }
2848
2849 static long kvm_dev_ioctl(struct file *filp,
2850                           unsigned int ioctl, unsigned long arg)
2851 {
2852         void __user *argp = (void __user *)arg;
2853         long r = -EINVAL;
2854
2855         switch (ioctl) {
2856         case KVM_GET_API_VERSION:
2857                 r = -EINVAL;
2858                 if (arg)
2859                         goto out;
2860                 r = KVM_API_VERSION;
2861                 break;
2862         case KVM_CREATE_VM:
2863                 r = -EINVAL;
2864                 if (arg)
2865                         goto out;
2866                 r = kvm_dev_ioctl_create_vm();
2867                 break;
2868         case KVM_GET_MSR_INDEX_LIST: {
2869                 struct kvm_msr_list __user *user_msr_list = argp;
2870                 struct kvm_msr_list msr_list;
2871                 unsigned n;
2872
2873                 r = -EFAULT;
2874                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2875                         goto out;
2876                 n = msr_list.nmsrs;
2877                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2878                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2879                         goto out;
2880                 r = -E2BIG;
2881                 if (n < num_msrs_to_save)
2882                         goto out;
2883                 r = -EFAULT;
2884                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2885                                  num_msrs_to_save * sizeof(u32)))
2886                         goto out;
2887                 if (copy_to_user(user_msr_list->indices
2888                                  + num_msrs_to_save * sizeof(u32),
2889                                  &emulated_msrs,
2890                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2891                         goto out;
2892                 r = 0;
2893                 break;
2894         }
2895         case KVM_CHECK_EXTENSION:
2896                 /*
2897                  * No extensions defined at present.
2898                  */
2899                 r = 0;
2900                 break;
2901         case KVM_GET_VCPU_MMAP_SIZE:
2902                 r = -EINVAL;
2903                 if (arg)
2904                         goto out;
2905                 r = 2 * PAGE_SIZE;
2906                 break;
2907         default:
2908                 ;
2909         }
2910 out:
2911         return r;
2912 }
2913
2914 static struct file_operations kvm_chardev_ops = {
2915         .open           = kvm_dev_open,
2916         .release        = kvm_dev_release,
2917         .unlocked_ioctl = kvm_dev_ioctl,
2918         .compat_ioctl   = kvm_dev_ioctl,
2919 };
2920
2921 static struct miscdevice kvm_dev = {
2922         KVM_MINOR,
2923         "kvm",
2924         &kvm_chardev_ops,
2925 };
2926
2927 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2928                        void *v)
2929 {
2930         if (val == SYS_RESTART) {
2931                 /*
2932                  * Some (well, at least mine) BIOSes hang on reboot if
2933                  * in vmx root mode.
2934                  */
2935                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2936                 on_each_cpu(hardware_disable, NULL, 0, 1);
2937         }
2938         return NOTIFY_OK;
2939 }
2940
2941 static struct notifier_block kvm_reboot_notifier = {
2942         .notifier_call = kvm_reboot,
2943         .priority = 0,
2944 };
2945
2946 /*
2947  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2948  * cached on it.
2949  */
2950 static void decache_vcpus_on_cpu(int cpu)
2951 {
2952         struct kvm *vm;
2953         struct kvm_vcpu *vcpu;
2954         int i;
2955
2956         spin_lock(&kvm_lock);
2957         list_for_each_entry(vm, &vm_list, vm_list)
2958                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2959                         vcpu = &vm->vcpus[i];
2960                         /*
2961                          * If the vcpu is locked, then it is running on some
2962                          * other cpu and therefore it is not cached on the
2963                          * cpu in question.
2964                          *
2965                          * If it's not locked, check the last cpu it executed
2966                          * on.
2967                          */
2968                         if (mutex_trylock(&vcpu->mutex)) {
2969                                 if (vcpu->cpu == cpu) {
2970                                         kvm_arch_ops->vcpu_decache(vcpu);
2971                                         vcpu->cpu = -1;
2972                                 }
2973                                 mutex_unlock(&vcpu->mutex);
2974                         }
2975                 }
2976         spin_unlock(&kvm_lock);
2977 }
2978
2979 static void hardware_enable(void *junk)
2980 {
2981         int cpu = raw_smp_processor_id();
2982
2983         if (cpu_isset(cpu, cpus_hardware_enabled))
2984                 return;
2985         cpu_set(cpu, cpus_hardware_enabled);
2986         kvm_arch_ops->hardware_enable(NULL);
2987 }
2988
2989 static void hardware_disable(void *junk)
2990 {
2991         int cpu = raw_smp_processor_id();
2992
2993         if (!cpu_isset(cpu, cpus_hardware_enabled))
2994                 return;
2995         cpu_clear(cpu, cpus_hardware_enabled);
2996         decache_vcpus_on_cpu(cpu);
2997         kvm_arch_ops->hardware_disable(NULL);
2998 }
2999
3000 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3001                            void *v)
3002 {
3003         int cpu = (long)v;
3004
3005         switch (val) {
3006         case CPU_DYING:
3007         case CPU_DYING_FROZEN:
3008                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3009                        cpu);
3010                 hardware_disable(NULL);
3011                 break;
3012         case CPU_UP_CANCELED:
3013         case CPU_UP_CANCELED_FROZEN:
3014                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3015                        cpu);
3016                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3017                 break;
3018         case CPU_ONLINE:
3019         case CPU_ONLINE_FROZEN:
3020                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3021                        cpu);
3022                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3023                 break;
3024         }
3025         return NOTIFY_OK;
3026 }
3027
3028 void kvm_io_bus_init(struct kvm_io_bus *bus)
3029 {
3030         memset(bus, 0, sizeof(*bus));
3031 }
3032
3033 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3034 {
3035         int i;
3036
3037         for (i = 0; i < bus->dev_count; i++) {
3038                 struct kvm_io_device *pos = bus->devs[i];
3039
3040                 kvm_iodevice_destructor(pos);
3041         }
3042 }
3043
3044 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3045 {
3046         int i;
3047
3048         for (i = 0; i < bus->dev_count; i++) {
3049                 struct kvm_io_device *pos = bus->devs[i];
3050
3051                 if (pos->in_range(pos, addr))
3052                         return pos;
3053         }
3054
3055         return NULL;
3056 }
3057
3058 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3059 {
3060         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3061
3062         bus->devs[bus->dev_count++] = dev;
3063 }
3064
3065 static struct notifier_block kvm_cpu_notifier = {
3066         .notifier_call = kvm_cpu_hotplug,
3067         .priority = 20, /* must be > scheduler priority */
3068 };
3069
3070 static u64 stat_get(void *_offset)
3071 {
3072         unsigned offset = (long)_offset;
3073         u64 total = 0;
3074         struct kvm *kvm;
3075         struct kvm_vcpu *vcpu;
3076         int i;
3077
3078         spin_lock(&kvm_lock);
3079         list_for_each_entry(kvm, &vm_list, vm_list)
3080                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3081                         vcpu = &kvm->vcpus[i];
3082                         total += *(u32 *)((void *)vcpu + offset);
3083                 }
3084         spin_unlock(&kvm_lock);
3085         return total;
3086 }
3087
3088 static void stat_set(void *offset, u64 val)
3089 {
3090 }
3091
3092 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
3093
3094 static __init void kvm_init_debug(void)
3095 {
3096         struct kvm_stats_debugfs_item *p;
3097
3098         debugfs_dir = debugfs_create_dir("kvm", NULL);
3099         for (p = debugfs_entries; p->name; ++p)
3100                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3101                                                 (void *)(long)p->offset,
3102                                                 &stat_fops);
3103 }
3104
3105 static void kvm_exit_debug(void)
3106 {
3107         struct kvm_stats_debugfs_item *p;
3108
3109         for (p = debugfs_entries; p->name; ++p)
3110                 debugfs_remove(p->dentry);
3111         debugfs_remove(debugfs_dir);
3112 }
3113
3114 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3115 {
3116         hardware_disable(NULL);
3117         return 0;
3118 }
3119
3120 static int kvm_resume(struct sys_device *dev)
3121 {
3122         hardware_enable(NULL);
3123         return 0;
3124 }
3125
3126 static struct sysdev_class kvm_sysdev_class = {
3127         set_kset_name("kvm"),
3128         .suspend = kvm_suspend,
3129         .resume = kvm_resume,
3130 };
3131
3132 static struct sys_device kvm_sysdev = {
3133         .id = 0,
3134         .cls = &kvm_sysdev_class,
3135 };
3136
3137 hpa_t bad_page_address;
3138
3139 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
3140 {
3141         int r;
3142
3143         if (kvm_arch_ops) {
3144                 printk(KERN_ERR "kvm: already loaded the other module\n");
3145                 return -EEXIST;
3146         }
3147
3148         if (!ops->cpu_has_kvm_support()) {
3149                 printk(KERN_ERR "kvm: no hardware support\n");
3150                 return -EOPNOTSUPP;
3151         }
3152         if (ops->disabled_by_bios()) {
3153                 printk(KERN_ERR "kvm: disabled by bios\n");
3154                 return -EOPNOTSUPP;
3155         }
3156
3157         kvm_arch_ops = ops;
3158
3159         r = kvm_arch_ops->hardware_setup();
3160         if (r < 0)
3161                 goto out;
3162
3163         on_each_cpu(hardware_enable, NULL, 0, 1);
3164         r = register_cpu_notifier(&kvm_cpu_notifier);
3165         if (r)
3166                 goto out_free_1;
3167         register_reboot_notifier(&kvm_reboot_notifier);
3168
3169         r = sysdev_class_register(&kvm_sysdev_class);
3170         if (r)
3171                 goto out_free_2;
3172
3173         r = sysdev_register(&kvm_sysdev);
3174         if (r)
3175                 goto out_free_3;
3176
3177         kvm_chardev_ops.owner = module;
3178
3179         r = misc_register(&kvm_dev);
3180         if (r) {
3181                 printk (KERN_ERR "kvm: misc device register failed\n");
3182                 goto out_free;
3183         }
3184
3185         return r;
3186
3187 out_free:
3188         sysdev_unregister(&kvm_sysdev);
3189 out_free_3:
3190         sysdev_class_unregister(&kvm_sysdev_class);
3191 out_free_2:
3192         unregister_reboot_notifier(&kvm_reboot_notifier);
3193         unregister_cpu_notifier(&kvm_cpu_notifier);
3194 out_free_1:
3195         on_each_cpu(hardware_disable, NULL, 0, 1);
3196         kvm_arch_ops->hardware_unsetup();
3197 out:
3198         kvm_arch_ops = NULL;
3199         return r;
3200 }
3201
3202 void kvm_exit_arch(void)
3203 {
3204         misc_deregister(&kvm_dev);
3205         sysdev_unregister(&kvm_sysdev);
3206         sysdev_class_unregister(&kvm_sysdev_class);
3207         unregister_reboot_notifier(&kvm_reboot_notifier);
3208         unregister_cpu_notifier(&kvm_cpu_notifier);
3209         on_each_cpu(hardware_disable, NULL, 0, 1);
3210         kvm_arch_ops->hardware_unsetup();
3211         kvm_arch_ops = NULL;
3212 }
3213
3214 static __init int kvm_init(void)
3215 {
3216         static struct page *bad_page;
3217         int r;
3218
3219         r = kvm_mmu_module_init();
3220         if (r)
3221                 goto out4;
3222
3223         kvm_init_debug();
3224
3225         kvm_init_msr_list();
3226
3227         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3228                 r = -ENOMEM;
3229                 goto out;
3230         }
3231
3232         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3233         memset(__va(bad_page_address), 0, PAGE_SIZE);
3234
3235         return 0;
3236
3237 out:
3238         kvm_exit_debug();
3239         kvm_mmu_module_exit();
3240 out4:
3241         return r;
3242 }
3243
3244 static __exit void kvm_exit(void)
3245 {
3246         kvm_exit_debug();
3247         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3248         kvm_mmu_module_exit();
3249 }
3250
3251 module_init(kvm_init)
3252 module_exit(kvm_exit)
3253
3254 EXPORT_SYMBOL_GPL(kvm_init_arch);
3255 EXPORT_SYMBOL_GPL(kvm_exit_arch);