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 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  58
  59 static struct kvm_stats_debugfs_item {
  60         const char *name;
  61         int offset;
  62         struct dentry *dentry;
  63 } debugfs_entries[] = {
  64         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  65         { "pf_guest", STAT_OFFSET(pf_guest) },
  66         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  67         { "invlpg", STAT_OFFSET(invlpg) },
  68         { "exits", STAT_OFFSET(exits) },
  69         { "io_exits", STAT_OFFSET(io_exits) },
  70         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  71         { "signal_exits", STAT_OFFSET(signal_exits) },
  72         { "irq_window", STAT_OFFSET(irq_window_exits) },
  73         { "halt_exits", STAT_OFFSET(halt_exits) },
  74         { "request_irq", STAT_OFFSET(request_irq_exits) },
  75         { "irq_exits", STAT_OFFSET(irq_exits) },
  76         { "light_exits", STAT_OFFSET(light_exits) },
  77         { "efer_reload", STAT_OFFSET(efer_reload) },
  78         { NULL }
  79 };
  80
  81 static struct dentry *debugfs_dir;
  82
  83 #define MAX_IO_MSRS 256
  84
  85 #define CR0_RESERVED_BITS                                               \
  86         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  87                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  88                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  89 #define LMSW_GUEST_MASK 0x0eULL
  90 #define CR4_RESEVED_BITS (~((1ULL << 11) - 1))
  91 #define CR8_RESEVED_BITS (~0x0fULL)
  92 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
  93
  94 #ifdef CONFIG_X86_64
  95 // LDT or TSS descriptor in the GDT. 16 bytes.
  96 struct segment_descriptor_64 {
  97         struct segment_descriptor s;
  98         u32 base_higher;
  99         u32 pad_zero;
 100 };
 101
 102 #endif
 103
 104 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 105                            unsigned long arg);
 106
 107 unsigned long segment_base(u16 selector)
 108 {
 109         struct descriptor_table gdt;
 110         struct segment_descriptor *d;
 111         unsigned long table_base;
 112         typedef unsigned long ul;
 113         unsigned long v;
 114
 115         if (selector == 0)
 116                 return 0;
 117
 118         asm ("sgdt %0" : "=m"(gdt));
 119         table_base = gdt.base;
 120
 121         if (selector & 4) {           /* from ldt */
 122                 u16 ldt_selector;
 123
 124                 asm ("sldt %0" : "=g"(ldt_selector));
 125                 table_base = segment_base(ldt_selector);
 126         }
 127         d = (struct segment_descriptor *)(table_base + (selector & ~7));
 128         v = d->base_low | ((ul)d->base_mid << 16) | ((ul)d->base_high << 24);
 129 #ifdef CONFIG_X86_64
 130         if (d->system == 0
 131             && (d->type == 2 || d->type == 9 || d->type == 11))
 132                 v |= ((ul)((struct segment_descriptor_64 *)d)->base_higher) << 32;
 133 #endif
 134         return v;
 135 }
 136 EXPORT_SYMBOL_GPL(segment_base);
 137
 138 static inline int valid_vcpu(int n)
 139 {
 140         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 141 }
 142
 143 int kvm_read_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 144                    void *dest)
 145 {
 146         unsigned char *host_buf = dest;
 147         unsigned long req_size = size;
 148
 149         while (size) {
 150                 hpa_t paddr;
 151                 unsigned now;
 152                 unsigned offset;
 153                 hva_t guest_buf;
 154
 155                 paddr = gva_to_hpa(vcpu, addr);
 156
 157                 if (is_error_hpa(paddr))
 158                         break;
 159
 160                 guest_buf = (hva_t)kmap_atomic(
 161                                         pfn_to_page(paddr >> PAGE_SHIFT),
 162                                         KM_USER0);
 163                 offset = addr & ~PAGE_MASK;
 164                 guest_buf |= offset;
 165                 now = min(size, PAGE_SIZE - offset);
 166                 memcpy(host_buf, (void*)guest_buf, now);
 167                 host_buf += now;
 168                 addr += now;
 169                 size -= now;
 170                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 171         }
 172         return req_size - size;
 173 }
 174 EXPORT_SYMBOL_GPL(kvm_read_guest);
 175
 176 int kvm_write_guest(struct kvm_vcpu *vcpu, gva_t addr, unsigned long size,
 177                     void *data)
 178 {
 179         unsigned char *host_buf = data;
 180         unsigned long req_size = size;
 181
 182         while (size) {
 183                 hpa_t paddr;
 184                 unsigned now;
 185                 unsigned offset;
 186                 hva_t guest_buf;
 187                 gfn_t gfn;
 188
 189                 paddr = gva_to_hpa(vcpu, addr);
 190
 191                 if (is_error_hpa(paddr))
 192                         break;
 193
 194                 gfn = vcpu->mmu.gva_to_gpa(vcpu, addr) >> PAGE_SHIFT;
 195                 mark_page_dirty(vcpu->kvm, gfn);
 196                 guest_buf = (hva_t)kmap_atomic(
 197                                 pfn_to_page(paddr >> PAGE_SHIFT), KM_USER0);
 198                 offset = addr & ~PAGE_MASK;
 199                 guest_buf |= offset;
 200                 now = min(size, PAGE_SIZE - offset);
 201                 memcpy((void*)guest_buf, host_buf, now);
 202                 host_buf += now;
 203                 addr += now;
 204                 size -= now;
 205                 kunmap_atomic((void *)(guest_buf & PAGE_MASK), KM_USER0);
 206         }
 207         return req_size - size;
 208 }
 209 EXPORT_SYMBOL_GPL(kvm_write_guest);
 210
 211 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 212 {
 213         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 214                 return;
 215
 216         vcpu->guest_fpu_loaded = 1;
 217         fx_save(vcpu->host_fx_image);
 218         fx_restore(vcpu->guest_fx_image);
 219 }
 220 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 221
 222 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 223 {
 224         if (!vcpu->guest_fpu_loaded)
 225                 return;
 226
 227         vcpu->guest_fpu_loaded = 0;
 228         fx_save(vcpu->guest_fx_image);
 229         fx_restore(vcpu->host_fx_image);
 230 }
 231 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 232
 233 /*
 234  * Switches to specified vcpu, until a matching vcpu_put()
 235  */
 236 static void vcpu_load(struct kvm_vcpu *vcpu)
 237 {
 238         mutex_lock(&vcpu->mutex);
 239         kvm_arch_ops->vcpu_load(vcpu);
 240 }
 241
 242 static void vcpu_put(struct kvm_vcpu *vcpu)
 243 {
 244         kvm_arch_ops->vcpu_put(vcpu);
 245         mutex_unlock(&vcpu->mutex);
 246 }
 247
 248 static void ack_flush(void *_completed)
 249 {
 250         atomic_t *completed = _completed;
 251
 252         atomic_inc(completed);
 253 }
 254
 255 void kvm_flush_remote_tlbs(struct kvm *kvm)
 256 {
 257         int i, cpu, needed;
 258         cpumask_t cpus;
 259         struct kvm_vcpu *vcpu;
 260         atomic_t completed;
 261
 262         atomic_set(&completed, 0);
 263         cpus_clear(cpus);
 264         needed = 0;
 265         for (i = 0; i < kvm->nvcpus; ++i) {
 266                 vcpu = &kvm->vcpus[i];
 267                 if (test_and_set_bit(KVM_TLB_FLUSH, &vcpu->requests))
 268                         continue;
 269                 cpu = vcpu->cpu;
 270                 if (cpu != -1 && cpu != raw_smp_processor_id())
 271                         if (!cpu_isset(cpu, cpus)) {
 272                                 cpu_set(cpu, cpus);
 273                                 ++needed;
 274                         }
 275         }
 276
 277         /*
 278          * We really want smp_call_function_mask() here.  But that's not
 279          * available, so ipi all cpus in parallel and wait for them
 280          * to complete.
 281          */
 282         for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
 283                 smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
 284         while (atomic_read(&completed) != needed) {
 285                 cpu_relax();
 286                 barrier();
 287         }
 288 }
 289
 290 static struct kvm *kvm_create_vm(void)
 291 {
 292         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 293         int i;
 294
 295         if (!kvm)
 296                 return ERR_PTR(-ENOMEM);
 297
 298         kvm_io_bus_init(&kvm->pio_bus);
 299         spin_lock_init(&kvm->lock);
 300         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 301         kvm_io_bus_init(&kvm->mmio_bus);
 302         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 303                 struct kvm_vcpu *vcpu = &kvm->vcpus[i];
 304
 305                 mutex_init(&vcpu->mutex);
 306                 vcpu->cpu = -1;
 307                 vcpu->kvm = kvm;
 308                 vcpu->mmu.root_hpa = INVALID_PAGE;
 309         }
 310         spin_lock(&kvm_lock);
 311         list_add(&kvm->vm_list, &vm_list);
 312         spin_unlock(&kvm_lock);
 313         return kvm;
 314 }
 315
 316 static int kvm_dev_open(struct inode *inode, struct file *filp)
 317 {
 318         return 0;
 319 }
 320
 321 /*
 322  * Free any memory in @free but not in @dont.
 323  */
 324 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 325                                   struct kvm_memory_slot *dont)
 326 {
 327         int i;
 328
 329         if (!dont || free->phys_mem != dont->phys_mem)
 330                 if (free->phys_mem) {
 331                         for (i = 0; i < free->npages; ++i)
 332                                 if (free->phys_mem[i])
 333                                         __free_page(free->phys_mem[i]);
 334                         vfree(free->phys_mem);
 335                 }
 336
 337         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 338                 vfree(free->dirty_bitmap);
 339
 340         free->phys_mem = NULL;
 341         free->npages = 0;
 342         free->dirty_bitmap = NULL;
 343 }
 344
 345 static void kvm_free_physmem(struct kvm *kvm)
 346 {
 347         int i;
 348
 349         for (i = 0; i < kvm->nmemslots; ++i)
 350                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 351 }
 352
 353 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 354 {
 355         int i;
 356
 357         for (i = 0; i < 2; ++i)
 358                 if (vcpu->pio.guest_pages[i]) {
 359                         __free_page(vcpu->pio.guest_pages[i]);
 360                         vcpu->pio.guest_pages[i] = NULL;
 361                 }
 362 }
 363
 364 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 365 {
 366         if (!vcpu->vmcs)
 367                 return;
 368
 369         vcpu_load(vcpu);
 370         kvm_mmu_unload(vcpu);
 371         vcpu_put(vcpu);
 372 }
 373
 374 static void kvm_free_vcpu(struct kvm_vcpu *vcpu)
 375 {
 376         if (!vcpu->vmcs)
 377                 return;
 378
 379         vcpu_load(vcpu);
 380         kvm_mmu_destroy(vcpu);
 381         vcpu_put(vcpu);
 382         kvm_arch_ops->vcpu_free(vcpu);
 383         free_page((unsigned long)vcpu->run);
 384         vcpu->run = NULL;
 385         free_page((unsigned long)vcpu->pio_data);
 386         vcpu->pio_data = NULL;
 387         free_pio_guest_pages(vcpu);
 388 }
 389
 390 static void kvm_free_vcpus(struct kvm *kvm)
 391 {
 392         unsigned int i;
 393
 394         /*
 395          * Unpin any mmu pages first.
 396          */
 397         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 398                 kvm_unload_vcpu_mmu(&kvm->vcpus[i]);
 399         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 400                 kvm_free_vcpu(&kvm->vcpus[i]);
 401 }
 402
 403 static int kvm_dev_release(struct inode *inode, struct file *filp)
 404 {
 405         return 0;
 406 }
 407
 408 static void kvm_destroy_vm(struct kvm *kvm)
 409 {
 410         spin_lock(&kvm_lock);
 411         list_del(&kvm->vm_list);
 412         spin_unlock(&kvm_lock);
 413         kvm_io_bus_destroy(&kvm->pio_bus);
 414         kvm_io_bus_destroy(&kvm->mmio_bus);
 415         kvm_free_vcpus(kvm);
 416         kvm_free_physmem(kvm);
 417         kfree(kvm);
 418 }
 419
 420 static int kvm_vm_release(struct inode *inode, struct file *filp)
 421 {
 422         struct kvm *kvm = filp->private_data;
 423
 424         kvm_destroy_vm(kvm);
 425         return 0;
 426 }
 427
 428 static void inject_gp(struct kvm_vcpu *vcpu)
 429 {
 430         kvm_arch_ops->inject_gp(vcpu, 0);
 431 }
 432
 433 /*
 434  * Load the pae pdptrs.  Return true is they are all valid.
 435  */
 436 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 437 {
 438         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 439         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 440         int i;
 441         u64 pdpte;
 442         u64 *pdpt;
 443         int ret;
 444         struct page *page;
 445
 446         spin_lock(&vcpu->kvm->lock);
 447         page = gfn_to_page(vcpu->kvm, pdpt_gfn);
 448         /* FIXME: !page - emulate? 0xff? */
 449         pdpt = kmap_atomic(page, KM_USER0);
 450
 451         ret = 1;
 452         for (i = 0; i < 4; ++i) {
 453                 pdpte = pdpt[offset + i];
 454                 if ((pdpte & 1) && (pdpte & 0xfffffff0000001e6ull)) {
 455                         ret = 0;
 456                         goto out;
 457                 }
 458         }
 459
 460         for (i = 0; i < 4; ++i)
 461                 vcpu->pdptrs[i] = pdpt[offset + i];
 462
 463 out:
 464         kunmap_atomic(pdpt, KM_USER0);
 465         spin_unlock(&vcpu->kvm->lock);
 466
 467         return ret;
 468 }
 469
 470 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 471 {
 472         if (cr0 & CR0_RESERVED_BITS) {
 473                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 474                        cr0, vcpu->cr0);
 475                 inject_gp(vcpu);
 476                 return;
 477         }
 478
 479         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 480                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 481                 inject_gp(vcpu);
 482                 return;
 483         }
 484
 485         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 486                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 487                        "and a clear PE flag\n");
 488                 inject_gp(vcpu);
 489                 return;
 490         }
 491
 492         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 493 #ifdef CONFIG_X86_64
 494                 if ((vcpu->shadow_efer & EFER_LME)) {
 495                         int cs_db, cs_l;
 496
 497                         if (!is_pae(vcpu)) {
 498                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 499                                        "in long mode while PAE is disabled\n");
 500                                 inject_gp(vcpu);
 501                                 return;
 502                         }
 503                         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 504                         if (cs_l) {
 505                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 506                                        "in long mode while CS.L == 1\n");
 507                                 inject_gp(vcpu);
 508                                 return;
 509
 510                         }
 511                 } else
 512 #endif
 513                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 514                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 515                                "reserved bits\n");
 516                         inject_gp(vcpu);
 517                         return;
 518                 }
 519
 520         }
 521
 522         kvm_arch_ops->set_cr0(vcpu, cr0);
 523         vcpu->cr0 = cr0;
 524
 525         spin_lock(&vcpu->kvm->lock);
 526         kvm_mmu_reset_context(vcpu);
 527         spin_unlock(&vcpu->kvm->lock);
 528         return;
 529 }
 530 EXPORT_SYMBOL_GPL(set_cr0);
 531
 532 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 533 {
 534         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 535 }
 536 EXPORT_SYMBOL_GPL(lmsw);
 537
 538 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 539 {
 540         if (cr4 & CR4_RESEVED_BITS) {
 541                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 542                 inject_gp(vcpu);
 543                 return;
 544         }
 545
 546         if (is_long_mode(vcpu)) {
 547                 if (!(cr4 & CR4_PAE_MASK)) {
 548                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 549                                "in long mode\n");
 550                         inject_gp(vcpu);
 551                         return;
 552                 }
 553         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & CR4_PAE_MASK)
 554                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 555                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 556                 inject_gp(vcpu);
 557         }
 558
 559         if (cr4 & CR4_VMXE_MASK) {
 560                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 561                 inject_gp(vcpu);
 562                 return;
 563         }
 564         kvm_arch_ops->set_cr4(vcpu, cr4);
 565         spin_lock(&vcpu->kvm->lock);
 566         kvm_mmu_reset_context(vcpu);
 567         spin_unlock(&vcpu->kvm->lock);
 568 }
 569 EXPORT_SYMBOL_GPL(set_cr4);
 570
 571 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 572 {
 573         if (is_long_mode(vcpu)) {
 574                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 575                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 576                         inject_gp(vcpu);
 577                         return;
 578                 }
 579         } else {
 580                 if (is_pae(vcpu)) {
 581                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 582                                 printk(KERN_DEBUG
 583                                        "set_cr3: #GP, reserved bits\n");
 584                                 inject_gp(vcpu);
 585                                 return;
 586                         }
 587                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 588                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 589                                        "reserved bits\n");
 590                                 inject_gp(vcpu);
 591                                 return;
 592                         }
 593                 } else {
 594                         if (cr3 & CR3_NONPAE_RESERVED_BITS) {
 595                                 printk(KERN_DEBUG
 596                                        "set_cr3: #GP, reserved bits\n");
 597                                 inject_gp(vcpu);
 598                                 return;
 599                         }
 600                 }
 601         }
 602
 603         vcpu->cr3 = cr3;
 604         spin_lock(&vcpu->kvm->lock);
 605         /*
 606          * Does the new cr3 value map to physical memory? (Note, we
 607          * catch an invalid cr3 even in real-mode, because it would
 608          * cause trouble later on when we turn on paging anyway.)
 609          *
 610          * A real CPU would silently accept an invalid cr3 and would
 611          * attempt to use it - with largely undefined (and often hard
 612          * to debug) behavior on the guest side.
 613          */
 614         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 615                 inject_gp(vcpu);
 616         else
 617                 vcpu->mmu.new_cr3(vcpu);
 618         spin_unlock(&vcpu->kvm->lock);
 619 }
 620 EXPORT_SYMBOL_GPL(set_cr3);
 621
 622 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 623 {
 624         if ( cr8 & CR8_RESEVED_BITS) {
 625                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 626                 inject_gp(vcpu);
 627                 return;
 628         }
 629         vcpu->cr8 = cr8;
 630 }
 631 EXPORT_SYMBOL_GPL(set_cr8);
 632
 633 void fx_init(struct kvm_vcpu *vcpu)
 634 {
 635         struct __attribute__ ((__packed__)) fx_image_s {
 636                 u16 control; //fcw
 637                 u16 status; //fsw
 638                 u16 tag; // ftw
 639                 u16 opcode; //fop
 640                 u64 ip; // fpu ip
 641                 u64 operand;// fpu dp
 642                 u32 mxcsr;
 643                 u32 mxcsr_mask;
 644
 645         } *fx_image;
 646
 647         fx_save(vcpu->host_fx_image);
 648         fpu_init();
 649         fx_save(vcpu->guest_fx_image);
 650         fx_restore(vcpu->host_fx_image);
 651
 652         fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
 653         fx_image->mxcsr = 0x1f80;
 654         memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
 655                0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
 656 }
 657 EXPORT_SYMBOL_GPL(fx_init);
 658
 659 /*
 660  * Allocate some memory and give it an address in the guest physical address
 661  * space.
 662  *
 663  * Discontiguous memory is allowed, mostly for framebuffers.
 664  */
 665 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 666                                           struct kvm_memory_region *mem)
 667 {
 668         int r;
 669         gfn_t base_gfn;
 670         unsigned long npages;
 671         unsigned long i;
 672         struct kvm_memory_slot *memslot;
 673         struct kvm_memory_slot old, new;
 674         int memory_config_version;
 675
 676         r = -EINVAL;
 677         /* General sanity checks */
 678         if (mem->memory_size & (PAGE_SIZE - 1))
 679                 goto out;
 680         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 681                 goto out;
 682         if (mem->slot >= KVM_MEMORY_SLOTS)
 683                 goto out;
 684         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 685                 goto out;
 686
 687         memslot = &kvm->memslots[mem->slot];
 688         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 689         npages = mem->memory_size >> PAGE_SHIFT;
 690
 691         if (!npages)
 692                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 693
 694 raced:
 695         spin_lock(&kvm->lock);
 696
 697         memory_config_version = kvm->memory_config_version;
 698         new = old = *memslot;
 699
 700         new.base_gfn = base_gfn;
 701         new.npages = npages;
 702         new.flags = mem->flags;
 703
 704         /* Disallow changing a memory slot's size. */
 705         r = -EINVAL;
 706         if (npages && old.npages && npages != old.npages)
 707                 goto out_unlock;
 708
 709         /* Check for overlaps */
 710         r = -EEXIST;
 711         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 712                 struct kvm_memory_slot *s = &kvm->memslots[i];
 713
 714                 if (s == memslot)
 715                         continue;
 716                 if (!((base_gfn + npages <= s->base_gfn) ||
 717                       (base_gfn >= s->base_gfn + s->npages)))
 718                         goto out_unlock;
 719         }
 720         /*
 721          * Do memory allocations outside lock.  memory_config_version will
 722          * detect any races.
 723          */
 724         spin_unlock(&kvm->lock);
 725
 726         /* Deallocate if slot is being removed */
 727         if (!npages)
 728                 new.phys_mem = NULL;
 729
 730         /* Free page dirty bitmap if unneeded */
 731         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 732                 new.dirty_bitmap = NULL;
 733
 734         r = -ENOMEM;
 735
 736         /* Allocate if a slot is being created */
 737         if (npages && !new.phys_mem) {
 738                 new.phys_mem = vmalloc(npages * sizeof(struct page *));
 739
 740                 if (!new.phys_mem)
 741                         goto out_free;
 742
 743                 memset(new.phys_mem, 0, npages * sizeof(struct page *));
 744                 for (i = 0; i < npages; ++i) {
 745                         new.phys_mem[i] = alloc_page(GFP_HIGHUSER
 746                                                      | __GFP_ZERO);
 747                         if (!new.phys_mem[i])
 748                                 goto out_free;
 749                         set_page_private(new.phys_mem[i],0);
 750                 }
 751         }
 752
 753         /* Allocate page dirty bitmap if needed */
 754         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 755                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 756
 757                 new.dirty_bitmap = vmalloc(dirty_bytes);
 758                 if (!new.dirty_bitmap)
 759                         goto out_free;
 760                 memset(new.dirty_bitmap, 0, dirty_bytes);
 761         }
 762
 763         spin_lock(&kvm->lock);
 764
 765         if (memory_config_version != kvm->memory_config_version) {
 766                 spin_unlock(&kvm->lock);
 767                 kvm_free_physmem_slot(&new, &old);
 768                 goto raced;
 769         }
 770
 771         r = -EAGAIN;
 772         if (kvm->busy)
 773                 goto out_unlock;
 774
 775         if (mem->slot >= kvm->nmemslots)
 776                 kvm->nmemslots = mem->slot + 1;
 777
 778         *memslot = new;
 779         ++kvm->memory_config_version;
 780
 781         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 782         kvm_flush_remote_tlbs(kvm);
 783
 784         spin_unlock(&kvm->lock);
 785
 786         kvm_free_physmem_slot(&old, &new);
 787         return 0;
 788
 789 out_unlock:
 790         spin_unlock(&kvm->lock);
 791 out_free:
 792         kvm_free_physmem_slot(&new, &old);
 793 out:
 794         return r;
 795 }
 796
 797 /*
 798  * Get (and clear) the dirty memory log for a memory slot.
 799  */
 800 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 801                                       struct kvm_dirty_log *log)
 802 {
 803         struct kvm_memory_slot *memslot;
 804         int r, i;
 805         int n;
 806         unsigned long any = 0;
 807
 808         spin_lock(&kvm->lock);
 809
 810         /*
 811          * Prevent changes to guest memory configuration even while the lock
 812          * is not taken.
 813          */
 814         ++kvm->busy;
 815         spin_unlock(&kvm->lock);
 816         r = -EINVAL;
 817         if (log->slot >= KVM_MEMORY_SLOTS)
 818                 goto out;
 819
 820         memslot = &kvm->memslots[log->slot];
 821         r = -ENOENT;
 822         if (!memslot->dirty_bitmap)
 823                 goto out;
 824
 825         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 826
 827         for (i = 0; !any && i < n/sizeof(long); ++i)
 828                 any = memslot->dirty_bitmap[i];
 829
 830         r = -EFAULT;
 831         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 832                 goto out;
 833
 834         spin_lock(&kvm->lock);
 835         kvm_mmu_slot_remove_write_access(kvm, log->slot);
 836         kvm_flush_remote_tlbs(kvm);
 837         memset(memslot->dirty_bitmap, 0, n);
 838         spin_unlock(&kvm->lock);
 839
 840         r = 0;
 841
 842 out:
 843         spin_lock(&kvm->lock);
 844         --kvm->busy;
 845         spin_unlock(&kvm->lock);
 846         return r;
 847 }
 848
 849 /*
 850  * Set a new alias region.  Aliases map a portion of physical memory into
 851  * another portion.  This is useful for memory windows, for example the PC
 852  * VGA region.
 853  */
 854 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
 855                                          struct kvm_memory_alias *alias)
 856 {
 857         int r, n;
 858         struct kvm_mem_alias *p;
 859
 860         r = -EINVAL;
 861         /* General sanity checks */
 862         if (alias->memory_size & (PAGE_SIZE - 1))
 863                 goto out;
 864         if (alias->guest_phys_addr & (PAGE_SIZE - 1))
 865                 goto out;
 866         if (alias->slot >= KVM_ALIAS_SLOTS)
 867                 goto out;
 868         if (alias->guest_phys_addr + alias->memory_size
 869             < alias->guest_phys_addr)
 870                 goto out;
 871         if (alias->target_phys_addr + alias->memory_size
 872             < alias->target_phys_addr)
 873                 goto out;
 874
 875         spin_lock(&kvm->lock);
 876
 877         p = &kvm->aliases[alias->slot];
 878         p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
 879         p->npages = alias->memory_size >> PAGE_SHIFT;
 880         p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
 881
 882         for (n = KVM_ALIAS_SLOTS; n > 0; --n)
 883                 if (kvm->aliases[n - 1].npages)
 884                         break;
 885         kvm->naliases = n;
 886
 887         kvm_mmu_zap_all(kvm);
 888
 889         spin_unlock(&kvm->lock);
 890
 891         return 0;
 892
 893 out:
 894         return r;
 895 }
 896
 897 static gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 898 {
 899         int i;
 900         struct kvm_mem_alias *alias;
 901
 902         for (i = 0; i < kvm->naliases; ++i) {
 903                 alias = &kvm->aliases[i];
 904                 if (gfn >= alias->base_gfn
 905                     && gfn < alias->base_gfn + alias->npages)
 906                         return alias->target_gfn + gfn - alias->base_gfn;
 907         }
 908         return gfn;
 909 }
 910
 911 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 912 {
 913         int i;
 914
 915         for (i = 0; i < kvm->nmemslots; ++i) {
 916                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 917
 918                 if (gfn >= memslot->base_gfn
 919                     && gfn < memslot->base_gfn + memslot->npages)
 920                         return memslot;
 921         }
 922         return NULL;
 923 }
 924
 925 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 926 {
 927         gfn = unalias_gfn(kvm, gfn);
 928         return __gfn_to_memslot(kvm, gfn);
 929 }
 930
 931 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 932 {
 933         struct kvm_memory_slot *slot;
 934
 935         gfn = unalias_gfn(kvm, gfn);
 936         slot = __gfn_to_memslot(kvm, gfn);
 937         if (!slot)
 938                 return NULL;
 939         return slot->phys_mem[gfn - slot->base_gfn];
 940 }
 941 EXPORT_SYMBOL_GPL(gfn_to_page);
 942
 943 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 944 {
 945         int i;
 946         struct kvm_memory_slot *memslot;
 947         unsigned long rel_gfn;
 948
 949         for (i = 0; i < kvm->nmemslots; ++i) {
 950                 memslot = &kvm->memslots[i];
 951
 952                 if (gfn >= memslot->base_gfn
 953                     && gfn < memslot->base_gfn + memslot->npages) {
 954
 955                         if (!memslot->dirty_bitmap)
 956                                 return;
 957
 958                         rel_gfn = gfn - memslot->base_gfn;
 959
 960                         /* avoid RMW */
 961                         if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 962                                 set_bit(rel_gfn, memslot->dirty_bitmap);
 963                         return;
 964                 }
 965         }
 966 }
 967
 968 static int emulator_read_std(unsigned long addr,
 969                              void *val,
 970                              unsigned int bytes,
 971                              struct x86_emulate_ctxt *ctxt)
 972 {
 973         struct kvm_vcpu *vcpu = ctxt->vcpu;
 974         void *data = val;
 975
 976         while (bytes) {
 977                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
 978                 unsigned offset = addr & (PAGE_SIZE-1);
 979                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
 980                 unsigned long pfn;
 981                 struct page *page;
 982                 void *page_virt;
 983
 984                 if (gpa == UNMAPPED_GVA)
 985                         return X86EMUL_PROPAGATE_FAULT;
 986                 pfn = gpa >> PAGE_SHIFT;
 987                 page = gfn_to_page(vcpu->kvm, pfn);
 988                 if (!page)
 989                         return X86EMUL_UNHANDLEABLE;
 990                 page_virt = kmap_atomic(page, KM_USER0);
 991
 992                 memcpy(data, page_virt + offset, tocopy);
 993
 994                 kunmap_atomic(page_virt, KM_USER0);
 995
 996                 bytes -= tocopy;
 997                 data += tocopy;
 998                 addr += tocopy;
 999         }
1000
1001         return X86EMUL_CONTINUE;
1002 }
1003
1004 static int emulator_write_std(unsigned long addr,
1005                               const void *val,
1006                               unsigned int bytes,
1007                               struct x86_emulate_ctxt *ctxt)
1008 {
1009         printk(KERN_ERR "emulator_write_std: addr %lx n %d\n",
1010                addr, bytes);
1011         return X86EMUL_UNHANDLEABLE;
1012 }
1013
1014 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1015                                                 gpa_t addr)
1016 {
1017         /*
1018          * Note that its important to have this wrapper function because
1019          * in the very near future we will be checking for MMIOs against
1020          * the LAPIC as well as the general MMIO bus
1021          */
1022         return kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1023 }
1024
1025 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1026                                                gpa_t addr)
1027 {
1028         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1029 }
1030
1031 static int emulator_read_emulated(unsigned long addr,
1032                                   void *val,
1033                                   unsigned int bytes,
1034                                   struct x86_emulate_ctxt *ctxt)
1035 {
1036         struct kvm_vcpu      *vcpu = ctxt->vcpu;
1037         struct kvm_io_device *mmio_dev;
1038         gpa_t                 gpa;
1039
1040         if (vcpu->mmio_read_completed) {
1041                 memcpy(val, vcpu->mmio_data, bytes);
1042                 vcpu->mmio_read_completed = 0;
1043                 return X86EMUL_CONTINUE;
1044         } else if (emulator_read_std(addr, val, bytes, ctxt)
1045                    == X86EMUL_CONTINUE)
1046                 return X86EMUL_CONTINUE;
1047
1048         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1049         if (gpa == UNMAPPED_GVA)
1050                 return X86EMUL_PROPAGATE_FAULT;
1051
1052         /*
1053          * Is this MMIO handled locally?
1054          */
1055         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1056         if (mmio_dev) {
1057                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1058                 return X86EMUL_CONTINUE;
1059         }
1060
1061         vcpu->mmio_needed = 1;
1062         vcpu->mmio_phys_addr = gpa;
1063         vcpu->mmio_size = bytes;
1064         vcpu->mmio_is_write = 0;
1065
1066         return X86EMUL_UNHANDLEABLE;
1067 }
1068
1069 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1070                                const void *val, int bytes)
1071 {
1072         struct page *page;
1073         void *virt;
1074         unsigned offset = offset_in_page(gpa);
1075
1076         if (((gpa + bytes - 1) >> PAGE_SHIFT) != (gpa >> PAGE_SHIFT))
1077                 return 0;
1078         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1079         if (!page)
1080                 return 0;
1081         mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
1082         virt = kmap_atomic(page, KM_USER0);
1083         kvm_mmu_pte_write(vcpu, gpa, virt + offset, val, bytes);
1084         memcpy(virt + offset_in_page(gpa), val, bytes);
1085         kunmap_atomic(virt, KM_USER0);
1086         return 1;
1087 }
1088
1089 static int emulator_write_emulated_onepage(unsigned long addr,
1090                                            const void *val,
1091                                            unsigned int bytes,
1092                                            struct x86_emulate_ctxt *ctxt)
1093 {
1094         struct kvm_vcpu      *vcpu = ctxt->vcpu;
1095         struct kvm_io_device *mmio_dev;
1096         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1097
1098         if (gpa == UNMAPPED_GVA) {
1099                 kvm_arch_ops->inject_page_fault(vcpu, addr, 2);
1100                 return X86EMUL_PROPAGATE_FAULT;
1101         }
1102
1103         if (emulator_write_phys(vcpu, gpa, val, bytes))
1104                 return X86EMUL_CONTINUE;
1105
1106         /*
1107          * Is this MMIO handled locally?
1108          */
1109         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1110         if (mmio_dev) {
1111                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1112                 return X86EMUL_CONTINUE;
1113         }
1114
1115         vcpu->mmio_needed = 1;
1116         vcpu->mmio_phys_addr = gpa;
1117         vcpu->mmio_size = bytes;
1118         vcpu->mmio_is_write = 1;
1119         memcpy(vcpu->mmio_data, val, bytes);
1120
1121         return X86EMUL_CONTINUE;
1122 }
1123
1124 static int emulator_write_emulated(unsigned long addr,
1125                                    const void *val,
1126                                    unsigned int bytes,
1127                                    struct x86_emulate_ctxt *ctxt)
1128 {
1129         /* Crossing a page boundary? */
1130         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1131                 int rc, now;
1132
1133                 now = -addr & ~PAGE_MASK;
1134                 rc = emulator_write_emulated_onepage(addr, val, now, ctxt);
1135                 if (rc != X86EMUL_CONTINUE)
1136                         return rc;
1137                 addr += now;
1138                 val += now;
1139                 bytes -= now;
1140         }
1141         return emulator_write_emulated_onepage(addr, val, bytes, ctxt);
1142 }
1143
1144 static int emulator_cmpxchg_emulated(unsigned long addr,
1145                                      const void *old,
1146                                      const void *new,
1147                                      unsigned int bytes,
1148                                      struct x86_emulate_ctxt *ctxt)
1149 {
1150         static int reported;
1151
1152         if (!reported) {
1153                 reported = 1;
1154                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1155         }
1156         return emulator_write_emulated(addr, new, bytes, ctxt);
1157 }
1158
1159 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1160 {
1161         return kvm_arch_ops->get_segment_base(vcpu, seg);
1162 }
1163
1164 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1165 {
1166         return X86EMUL_CONTINUE;
1167 }
1168
1169 int emulate_clts(struct kvm_vcpu *vcpu)
1170 {
1171         unsigned long cr0;
1172
1173         cr0 = vcpu->cr0 & ~X86_CR0_TS;
1174         kvm_arch_ops->set_cr0(vcpu, cr0);
1175         return X86EMUL_CONTINUE;
1176 }
1177
1178 int emulator_get_dr(struct x86_emulate_ctxt* ctxt, int dr, unsigned long *dest)
1179 {
1180         struct kvm_vcpu *vcpu = ctxt->vcpu;
1181
1182         switch (dr) {
1183         case 0 ... 3:
1184                 *dest = kvm_arch_ops->get_dr(vcpu, dr);
1185                 return X86EMUL_CONTINUE;
1186         default:
1187                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1188                        __FUNCTION__, dr);
1189                 return X86EMUL_UNHANDLEABLE;
1190         }
1191 }
1192
1193 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1194 {
1195         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1196         int exception;
1197
1198         kvm_arch_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1199         if (exception) {
1200                 /* FIXME: better handling */
1201                 return X86EMUL_UNHANDLEABLE;
1202         }
1203         return X86EMUL_CONTINUE;
1204 }
1205
1206 static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
1207 {
1208         static int reported;
1209         u8 opcodes[4];
1210         unsigned long rip = ctxt->vcpu->rip;
1211         unsigned long rip_linear;
1212
1213         rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
1214
1215         if (reported)
1216                 return;
1217
1218         emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt);
1219
1220         printk(KERN_ERR "emulation failed but !mmio_needed?"
1221                " rip %lx %02x %02x %02x %02x\n",
1222                rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1223         reported = 1;
1224 }
1225
1226 struct x86_emulate_ops emulate_ops = {
1227         .read_std            = emulator_read_std,
1228         .write_std           = emulator_write_std,
1229         .read_emulated       = emulator_read_emulated,
1230         .write_emulated      = emulator_write_emulated,
1231         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1232 };
1233
1234 int emulate_instruction(struct kvm_vcpu *vcpu,
1235                         struct kvm_run *run,
1236                         unsigned long cr2,
1237                         u16 error_code)
1238 {
1239         struct x86_emulate_ctxt emulate_ctxt;
1240         int r;
1241         int cs_db, cs_l;
1242
1243         vcpu->mmio_fault_cr2 = cr2;
1244         kvm_arch_ops->cache_regs(vcpu);
1245
1246         kvm_arch_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1247
1248         emulate_ctxt.vcpu = vcpu;
1249         emulate_ctxt.eflags = kvm_arch_ops->get_rflags(vcpu);
1250         emulate_ctxt.cr2 = cr2;
1251         emulate_ctxt.mode = (emulate_ctxt.eflags & X86_EFLAGS_VM)
1252                 ? X86EMUL_MODE_REAL : cs_l
1253                 ? X86EMUL_MODE_PROT64 : cs_db
1254                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1255
1256         if (emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1257                 emulate_ctxt.cs_base = 0;
1258                 emulate_ctxt.ds_base = 0;
1259                 emulate_ctxt.es_base = 0;
1260                 emulate_ctxt.ss_base = 0;
1261         } else {
1262                 emulate_ctxt.cs_base = get_segment_base(vcpu, VCPU_SREG_CS);
1263                 emulate_ctxt.ds_base = get_segment_base(vcpu, VCPU_SREG_DS);
1264                 emulate_ctxt.es_base = get_segment_base(vcpu, VCPU_SREG_ES);
1265                 emulate_ctxt.ss_base = get_segment_base(vcpu, VCPU_SREG_SS);
1266         }
1267
1268         emulate_ctxt.gs_base = get_segment_base(vcpu, VCPU_SREG_GS);
1269         emulate_ctxt.fs_base = get_segment_base(vcpu, VCPU_SREG_FS);
1270
1271         vcpu->mmio_is_write = 0;
1272         r = x86_emulate_memop(&emulate_ctxt, &emulate_ops);
1273
1274         if ((r || vcpu->mmio_is_write) && run) {
1275                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1276                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1277                 run->mmio.len = vcpu->mmio_size;
1278                 run->mmio.is_write = vcpu->mmio_is_write;
1279         }
1280
1281         if (r) {
1282                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1283                         return EMULATE_DONE;
1284                 if (!vcpu->mmio_needed) {
1285                         report_emulation_failure(&emulate_ctxt);
1286                         return EMULATE_FAIL;
1287                 }
1288                 return EMULATE_DO_MMIO;
1289         }
1290
1291         kvm_arch_ops->decache_regs(vcpu);
1292         kvm_arch_ops->set_rflags(vcpu, emulate_ctxt.eflags);
1293
1294         if (vcpu->mmio_is_write) {
1295                 vcpu->mmio_needed = 0;
1296                 return EMULATE_DO_MMIO;
1297         }
1298
1299         return EMULATE_DONE;
1300 }
1301 EXPORT_SYMBOL_GPL(emulate_instruction);
1302
1303 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1304 {
1305         if (vcpu->irq_summary)
1306                 return 1;
1307
1308         vcpu->run->exit_reason = KVM_EXIT_HLT;
1309         ++vcpu->stat.halt_exits;
1310         return 0;
1311 }
1312 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1313
1314 int kvm_hypercall(struct kvm_vcpu *vcpu, struct kvm_run *run)
1315 {
1316         unsigned long nr, a0, a1, a2, a3, a4, a5, ret;
1317
1318         kvm_arch_ops->cache_regs(vcpu);
1319         ret = -KVM_EINVAL;
1320 #ifdef CONFIG_X86_64
1321         if (is_long_mode(vcpu)) {
1322                 nr = vcpu->regs[VCPU_REGS_RAX];
1323                 a0 = vcpu->regs[VCPU_REGS_RDI];
1324                 a1 = vcpu->regs[VCPU_REGS_RSI];
1325                 a2 = vcpu->regs[VCPU_REGS_RDX];
1326                 a3 = vcpu->regs[VCPU_REGS_RCX];
1327                 a4 = vcpu->regs[VCPU_REGS_R8];
1328                 a5 = vcpu->regs[VCPU_REGS_R9];
1329         } else
1330 #endif
1331         {
1332                 nr = vcpu->regs[VCPU_REGS_RBX] & -1u;
1333                 a0 = vcpu->regs[VCPU_REGS_RAX] & -1u;
1334                 a1 = vcpu->regs[VCPU_REGS_RCX] & -1u;
1335                 a2 = vcpu->regs[VCPU_REGS_RDX] & -1u;
1336                 a3 = vcpu->regs[VCPU_REGS_RSI] & -1u;
1337                 a4 = vcpu->regs[VCPU_REGS_RDI] & -1u;
1338                 a5 = vcpu->regs[VCPU_REGS_RBP] & -1u;
1339         }
1340         switch (nr) {
1341         default:
1342                 run->hypercall.args[0] = a0;
1343                 run->hypercall.args[1] = a1;
1344                 run->hypercall.args[2] = a2;
1345                 run->hypercall.args[3] = a3;
1346                 run->hypercall.args[4] = a4;
1347                 run->hypercall.args[5] = a5;
1348                 run->hypercall.ret = ret;
1349                 run->hypercall.longmode = is_long_mode(vcpu);
1350                 kvm_arch_ops->decache_regs(vcpu);
1351                 return 0;
1352         }
1353         vcpu->regs[VCPU_REGS_RAX] = ret;
1354         kvm_arch_ops->decache_regs(vcpu);
1355         return 1;
1356 }
1357 EXPORT_SYMBOL_GPL(kvm_hypercall);
1358
1359 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1360 {
1361         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1362 }
1363
1364 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1365 {
1366         struct descriptor_table dt = { limit, base };
1367
1368         kvm_arch_ops->set_gdt(vcpu, &dt);
1369 }
1370
1371 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1372 {
1373         struct descriptor_table dt = { limit, base };
1374
1375         kvm_arch_ops->set_idt(vcpu, &dt);
1376 }
1377
1378 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1379                    unsigned long *rflags)
1380 {
1381         lmsw(vcpu, msw);
1382         *rflags = kvm_arch_ops->get_rflags(vcpu);
1383 }
1384
1385 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1386 {
1387         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
1388         switch (cr) {
1389         case 0:
1390                 return vcpu->cr0;
1391         case 2:
1392                 return vcpu->cr2;
1393         case 3:
1394                 return vcpu->cr3;
1395         case 4:
1396                 return vcpu->cr4;
1397         default:
1398                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1399                 return 0;
1400         }
1401 }
1402
1403 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1404                      unsigned long *rflags)
1405 {
1406         switch (cr) {
1407         case 0:
1408                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1409                 *rflags = kvm_arch_ops->get_rflags(vcpu);
1410                 break;
1411         case 2:
1412                 vcpu->cr2 = val;
1413                 break;
1414         case 3:
1415                 set_cr3(vcpu, val);
1416                 break;
1417         case 4:
1418                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1419                 break;
1420         default:
1421                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1422         }
1423 }
1424
1425 /*
1426  * Register the para guest with the host:
1427  */
1428 static int vcpu_register_para(struct kvm_vcpu *vcpu, gpa_t para_state_gpa)
1429 {
1430         struct kvm_vcpu_para_state *para_state;
1431         hpa_t para_state_hpa, hypercall_hpa;
1432         struct page *para_state_page;
1433         unsigned char *hypercall;
1434         gpa_t hypercall_gpa;
1435
1436         printk(KERN_DEBUG "kvm: guest trying to enter paravirtual mode\n");
1437         printk(KERN_DEBUG ".... para_state_gpa: %08Lx\n", para_state_gpa);
1438
1439         /*
1440          * Needs to be page aligned:
1441          */
1442         if (para_state_gpa != PAGE_ALIGN(para_state_gpa))
1443                 goto err_gp;
1444
1445         para_state_hpa = gpa_to_hpa(vcpu, para_state_gpa);
1446         printk(KERN_DEBUG ".... para_state_hpa: %08Lx\n", para_state_hpa);
1447         if (is_error_hpa(para_state_hpa))
1448                 goto err_gp;
1449
1450         mark_page_dirty(vcpu->kvm, para_state_gpa >> PAGE_SHIFT);
1451         para_state_page = pfn_to_page(para_state_hpa >> PAGE_SHIFT);
1452         para_state = kmap_atomic(para_state_page, KM_USER0);
1453
1454         printk(KERN_DEBUG "....  guest version: %d\n", para_state->guest_version);
1455         printk(KERN_DEBUG "....           size: %d\n", para_state->size);
1456
1457         para_state->host_version = KVM_PARA_API_VERSION;
1458         /*
1459          * We cannot support guests that try to register themselves
1460          * with a newer API version than the host supports:
1461          */
1462         if (para_state->guest_version > KVM_PARA_API_VERSION) {
1463                 para_state->ret = -KVM_EINVAL;
1464                 goto err_kunmap_skip;
1465         }
1466
1467         hypercall_gpa = para_state->hypercall_gpa;
1468         hypercall_hpa = gpa_to_hpa(vcpu, hypercall_gpa);
1469         printk(KERN_DEBUG ".... hypercall_hpa: %08Lx\n", hypercall_hpa);
1470         if (is_error_hpa(hypercall_hpa)) {
1471                 para_state->ret = -KVM_EINVAL;
1472                 goto err_kunmap_skip;
1473         }
1474
1475         printk(KERN_DEBUG "kvm: para guest successfully registered.\n");
1476         vcpu->para_state_page = para_state_page;
1477         vcpu->para_state_gpa = para_state_gpa;
1478         vcpu->hypercall_gpa = hypercall_gpa;
1479
1480         mark_page_dirty(vcpu->kvm, hypercall_gpa >> PAGE_SHIFT);
1481         hypercall = kmap_atomic(pfn_to_page(hypercall_hpa >> PAGE_SHIFT),
1482                                 KM_USER1) + (hypercall_hpa & ~PAGE_MASK);
1483         kvm_arch_ops->patch_hypercall(vcpu, hypercall);
1484         kunmap_atomic(hypercall, KM_USER1);
1485
1486         para_state->ret = 0;
1487 err_kunmap_skip:
1488         kunmap_atomic(para_state, KM_USER0);
1489         return 0;
1490 err_gp:
1491         return 1;
1492 }
1493
1494 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1495 {
1496         u64 data;
1497
1498         switch (msr) {
1499         case 0xc0010010: /* SYSCFG */
1500         case 0xc0010015: /* HWCR */
1501         case MSR_IA32_PLATFORM_ID:
1502         case MSR_IA32_P5_MC_ADDR:
1503         case MSR_IA32_P5_MC_TYPE:
1504         case MSR_IA32_MC0_CTL:
1505         case MSR_IA32_MCG_STATUS:
1506         case MSR_IA32_MCG_CAP:
1507         case MSR_IA32_MC0_MISC:
1508         case MSR_IA32_MC0_MISC+4:
1509         case MSR_IA32_MC0_MISC+8:
1510         case MSR_IA32_MC0_MISC+12:
1511         case MSR_IA32_MC0_MISC+16:
1512         case MSR_IA32_UCODE_REV:
1513         case MSR_IA32_PERF_STATUS:
1514         case MSR_IA32_EBL_CR_POWERON:
1515                 /* MTRR registers */
1516         case 0xfe:
1517         case 0x200 ... 0x2ff:
1518                 data = 0;
1519                 break;
1520         case 0xcd: /* fsb frequency */
1521                 data = 3;
1522                 break;
1523         case MSR_IA32_APICBASE:
1524                 data = vcpu->apic_base;
1525                 break;
1526         case MSR_IA32_MISC_ENABLE:
1527                 data = vcpu->ia32_misc_enable_msr;
1528                 break;
1529 #ifdef CONFIG_X86_64
1530         case MSR_EFER:
1531                 data = vcpu->shadow_efer;
1532                 break;
1533 #endif
1534         default:
1535                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", msr);
1536                 return 1;
1537         }
1538         *pdata = data;
1539         return 0;
1540 }
1541 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1542
1543 /*
1544  * Reads an msr value (of 'msr_index') into 'pdata'.
1545  * Returns 0 on success, non-0 otherwise.
1546  * Assumes vcpu_load() was already called.
1547  */
1548 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1549 {
1550         return kvm_arch_ops->get_msr(vcpu, msr_index, pdata);
1551 }
1552
1553 #ifdef CONFIG_X86_64
1554
1555 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1556 {
1557         if (efer & EFER_RESERVED_BITS) {
1558                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1559                        efer);
1560                 inject_gp(vcpu);
1561                 return;
1562         }
1563
1564         if (is_paging(vcpu)
1565             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1566                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1567                 inject_gp(vcpu);
1568                 return;
1569         }
1570
1571         kvm_arch_ops->set_efer(vcpu, efer);
1572
1573         efer &= ~EFER_LMA;
1574         efer |= vcpu->shadow_efer & EFER_LMA;
1575
1576         vcpu->shadow_efer = efer;
1577 }
1578
1579 #endif
1580
1581 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1582 {
1583         switch (msr) {
1584 #ifdef CONFIG_X86_64
1585         case MSR_EFER:
1586                 set_efer(vcpu, data);
1587                 break;
1588 #endif
1589         case MSR_IA32_MC0_STATUS:
1590                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1591                        __FUNCTION__, data);
1592                 break;
1593         case MSR_IA32_MCG_STATUS:
1594                 printk(KERN_WARNING "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1595                         __FUNCTION__, data);
1596                 break;
1597         case MSR_IA32_UCODE_REV:
1598         case MSR_IA32_UCODE_WRITE:
1599         case 0x200 ... 0x2ff: /* MTRRs */
1600                 break;
1601         case MSR_IA32_APICBASE:
1602                 vcpu->apic_base = data;
1603                 break;
1604         case MSR_IA32_MISC_ENABLE:
1605                 vcpu->ia32_misc_enable_msr = data;
1606                 break;
1607         /*
1608          * This is the 'probe whether the host is KVM' logic:
1609          */
1610         case MSR_KVM_API_MAGIC:
1611                 return vcpu_register_para(vcpu, data);
1612
1613         default:
1614                 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr);
1615                 return 1;
1616         }
1617         return 0;
1618 }
1619 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1620
1621 /*
1622  * Writes msr value into into the appropriate "register".
1623  * Returns 0 on success, non-0 otherwise.
1624  * Assumes vcpu_load() was already called.
1625  */
1626 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1627 {
1628         return kvm_arch_ops->set_msr(vcpu, msr_index, data);
1629 }
1630
1631 void kvm_resched(struct kvm_vcpu *vcpu)
1632 {
1633         if (!need_resched())
1634                 return;
1635         vcpu_put(vcpu);
1636         cond_resched();
1637         vcpu_load(vcpu);
1638 }
1639 EXPORT_SYMBOL_GPL(kvm_resched);
1640
1641 void load_msrs(struct vmx_msr_entry *e, int n)
1642 {
1643         int i;
1644
1645         for (i = 0; i < n; ++i)
1646                 wrmsrl(e[i].index, e[i].data);
1647 }
1648 EXPORT_SYMBOL_GPL(load_msrs);
1649
1650 void save_msrs(struct vmx_msr_entry *e, int n)
1651 {
1652         int i;
1653
1654         for (i = 0; i < n; ++i)
1655                 rdmsrl(e[i].index, e[i].data);
1656 }
1657 EXPORT_SYMBOL_GPL(save_msrs);
1658
1659 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1660 {
1661         int i;
1662         u32 function;
1663         struct kvm_cpuid_entry *e, *best;
1664
1665         kvm_arch_ops->cache_regs(vcpu);
1666         function = vcpu->regs[VCPU_REGS_RAX];
1667         vcpu->regs[VCPU_REGS_RAX] = 0;
1668         vcpu->regs[VCPU_REGS_RBX] = 0;
1669         vcpu->regs[VCPU_REGS_RCX] = 0;
1670         vcpu->regs[VCPU_REGS_RDX] = 0;
1671         best = NULL;
1672         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1673                 e = &vcpu->cpuid_entries[i];
1674                 if (e->function == function) {
1675                         best = e;
1676                         break;
1677                 }
1678                 /*
1679                  * Both basic or both extended?
1680                  */
1681                 if (((e->function ^ function) & 0x80000000) == 0)
1682                         if (!best || e->function > best->function)
1683                                 best = e;
1684         }
1685         if (best) {
1686                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1687                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1688                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1689                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1690         }
1691         kvm_arch_ops->decache_regs(vcpu);
1692         kvm_arch_ops->skip_emulated_instruction(vcpu);
1693 }
1694 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1695
1696 static int pio_copy_data(struct kvm_vcpu *vcpu)
1697 {
1698         void *p = vcpu->pio_data;
1699         void *q;
1700         unsigned bytes;
1701         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1702
1703         kvm_arch_ops->vcpu_put(vcpu);
1704         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1705                  PAGE_KERNEL);
1706         if (!q) {
1707                 kvm_arch_ops->vcpu_load(vcpu);
1708                 free_pio_guest_pages(vcpu);
1709                 return -ENOMEM;
1710         }
1711         q += vcpu->pio.guest_page_offset;
1712         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1713         if (vcpu->pio.in)
1714                 memcpy(q, p, bytes);
1715         else
1716                 memcpy(p, q, bytes);
1717         q -= vcpu->pio.guest_page_offset;
1718         vunmap(q);
1719         kvm_arch_ops->vcpu_load(vcpu);
1720         free_pio_guest_pages(vcpu);
1721         return 0;
1722 }
1723
1724 static int complete_pio(struct kvm_vcpu *vcpu)
1725 {
1726         struct kvm_pio_request *io = &vcpu->pio;
1727         long delta;
1728         int r;
1729
1730         kvm_arch_ops->cache_regs(vcpu);
1731
1732         if (!io->string) {
1733                 if (io->in)
1734                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1735                                io->size);
1736         } else {
1737                 if (io->in) {
1738                         r = pio_copy_data(vcpu);
1739                         if (r) {
1740                                 kvm_arch_ops->cache_regs(vcpu);
1741                                 return r;
1742                         }
1743                 }
1744
1745                 delta = 1;
1746                 if (io->rep) {
1747                         delta *= io->cur_count;
1748                         /*
1749                          * The size of the register should really depend on
1750                          * current address size.
1751                          */
1752                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1753                 }
1754                 if (io->down)
1755                         delta = -delta;
1756                 delta *= io->size;
1757                 if (io->in)
1758                         vcpu->regs[VCPU_REGS_RDI] += delta;
1759                 else
1760                         vcpu->regs[VCPU_REGS_RSI] += delta;
1761         }
1762
1763         kvm_arch_ops->decache_regs(vcpu);
1764
1765         io->count -= io->cur_count;
1766         io->cur_count = 0;
1767
1768         if (!io->count)
1769                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1770         return 0;
1771 }
1772
1773 static void kernel_pio(struct kvm_io_device *pio_dev,
1774                        struct kvm_vcpu *vcpu,
1775                        void *pd)
1776 {
1777         /* TODO: String I/O for in kernel device */
1778
1779         if (vcpu->pio.in)
1780                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1781                                   vcpu->pio.size,
1782                                   pd);
1783         else
1784                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1785                                    vcpu->pio.size,
1786                                    pd);
1787 }
1788
1789 static void pio_string_write(struct kvm_io_device *pio_dev,
1790                              struct kvm_vcpu *vcpu)
1791 {
1792         struct kvm_pio_request *io = &vcpu->pio;
1793         void *pd = vcpu->pio_data;
1794         int i;
1795
1796         for (i = 0; i < io->cur_count; i++) {
1797                 kvm_iodevice_write(pio_dev, io->port,
1798                                    io->size,
1799                                    pd);
1800                 pd += io->size;
1801         }
1802 }
1803
1804 int kvm_setup_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1805                   int size, unsigned long count, int string, int down,
1806                   gva_t address, int rep, unsigned port)
1807 {
1808         unsigned now, in_page;
1809         int i, ret = 0;
1810         int nr_pages = 1;
1811         struct page *page;
1812         struct kvm_io_device *pio_dev;
1813
1814         vcpu->run->exit_reason = KVM_EXIT_IO;
1815         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1816         vcpu->run->io.size = size;
1817         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1818         vcpu->run->io.count = count;
1819         vcpu->run->io.port = port;
1820         vcpu->pio.count = count;
1821         vcpu->pio.cur_count = count;
1822         vcpu->pio.size = size;
1823         vcpu->pio.in = in;
1824         vcpu->pio.port = port;
1825         vcpu->pio.string = string;
1826         vcpu->pio.down = down;
1827         vcpu->pio.guest_page_offset = offset_in_page(address);
1828         vcpu->pio.rep = rep;
1829
1830         pio_dev = vcpu_find_pio_dev(vcpu, port);
1831         if (!string) {
1832                 kvm_arch_ops->cache_regs(vcpu);
1833                 memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1834                 kvm_arch_ops->decache_regs(vcpu);
1835                 if (pio_dev) {
1836                         kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1837                         complete_pio(vcpu);
1838                         return 1;
1839                 }
1840                 return 0;
1841         }
1842
1843         if (!count) {
1844                 kvm_arch_ops->skip_emulated_instruction(vcpu);
1845                 return 1;
1846         }
1847
1848         now = min(count, PAGE_SIZE / size);
1849
1850         if (!down)
1851                 in_page = PAGE_SIZE - offset_in_page(address);
1852         else
1853                 in_page = offset_in_page(address) + size;
1854         now = min(count, (unsigned long)in_page / size);
1855         if (!now) {
1856                 /*
1857                  * String I/O straddles page boundary.  Pin two guest pages
1858                  * so that we satisfy atomicity constraints.  Do just one
1859                  * transaction to avoid complexity.
1860                  */
1861                 nr_pages = 2;
1862                 now = 1;
1863         }
1864         if (down) {
1865                 /*
1866                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1867                  */
1868                 printk(KERN_ERR "kvm: guest string pio down\n");
1869                 inject_gp(vcpu);
1870                 return 1;
1871         }
1872         vcpu->run->io.count = now;
1873         vcpu->pio.cur_count = now;
1874
1875         for (i = 0; i < nr_pages; ++i) {
1876                 spin_lock(&vcpu->kvm->lock);
1877                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1878                 if (page)
1879                         get_page(page);
1880                 vcpu->pio.guest_pages[i] = page;
1881                 spin_unlock(&vcpu->kvm->lock);
1882                 if (!page) {
1883                         inject_gp(vcpu);
1884                         free_pio_guest_pages(vcpu);
1885                         return 1;
1886                 }
1887         }
1888
1889         if (!vcpu->pio.in) {
1890                 /* string PIO write */
1891                 ret = pio_copy_data(vcpu);
1892                 if (ret >= 0 && pio_dev) {
1893                         pio_string_write(pio_dev, vcpu);
1894                         complete_pio(vcpu);
1895                         if (vcpu->pio.count == 0)
1896                                 ret = 1;
1897                 }
1898         } else if (pio_dev)
1899                 printk(KERN_ERR "no string pio read support yet, "
1900                        "port %x size %d count %ld\n",
1901                         port, size, count);
1902
1903         return ret;
1904 }
1905 EXPORT_SYMBOL_GPL(kvm_setup_pio);
1906
1907 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1908 {
1909         int r;
1910         sigset_t sigsaved;
1911
1912         vcpu_load(vcpu);
1913
1914         if (vcpu->sigset_active)
1915                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1916
1917         /* re-sync apic's tpr */
1918         vcpu->cr8 = kvm_run->cr8;
1919
1920         if (vcpu->pio.cur_count) {
1921                 r = complete_pio(vcpu);
1922                 if (r)
1923                         goto out;
1924         }
1925
1926         if (vcpu->mmio_needed) {
1927                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1928                 vcpu->mmio_read_completed = 1;
1929                 vcpu->mmio_needed = 0;
1930                 r = emulate_instruction(vcpu, kvm_run,
1931                                         vcpu->mmio_fault_cr2, 0);
1932                 if (r == EMULATE_DO_MMIO) {
1933                         /*
1934                          * Read-modify-write.  Back to userspace.
1935                          */
1936                         kvm_run->exit_reason = KVM_EXIT_MMIO;
1937                         r = 0;
1938                         goto out;
1939                 }
1940         }
1941
1942         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1943                 kvm_arch_ops->cache_regs(vcpu);
1944                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1945                 kvm_arch_ops->decache_regs(vcpu);
1946         }
1947
1948         r = kvm_arch_ops->run(vcpu, kvm_run);
1949
1950 out:
1951         if (vcpu->sigset_active)
1952                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1953
1954         vcpu_put(vcpu);
1955         return r;
1956 }
1957
1958 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
1959                                    struct kvm_regs *regs)
1960 {
1961         vcpu_load(vcpu);
1962
1963         kvm_arch_ops->cache_regs(vcpu);
1964
1965         regs->rax = vcpu->regs[VCPU_REGS_RAX];
1966         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1967         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1968         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1969         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1970         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1971         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1972         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1973 #ifdef CONFIG_X86_64
1974         regs->r8 = vcpu->regs[VCPU_REGS_R8];
1975         regs->r9 = vcpu->regs[VCPU_REGS_R9];
1976         regs->r10 = vcpu->regs[VCPU_REGS_R10];
1977         regs->r11 = vcpu->regs[VCPU_REGS_R11];
1978         regs->r12 = vcpu->regs[VCPU_REGS_R12];
1979         regs->r13 = vcpu->regs[VCPU_REGS_R13];
1980         regs->r14 = vcpu->regs[VCPU_REGS_R14];
1981         regs->r15 = vcpu->regs[VCPU_REGS_R15];
1982 #endif
1983
1984         regs->rip = vcpu->rip;
1985         regs->rflags = kvm_arch_ops->get_rflags(vcpu);
1986
1987         /*
1988          * Don't leak debug flags in case they were set for guest debugging
1989          */
1990         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1991                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1992
1993         vcpu_put(vcpu);
1994
1995         return 0;
1996 }
1997
1998 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1999                                    struct kvm_regs *regs)
2000 {
2001         vcpu_load(vcpu);
2002
2003         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2004         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2005         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2006         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2007         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2008         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2009         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2010         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2011 #ifdef CONFIG_X86_64
2012         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2013         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2014         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2015         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2016         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2017         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2018         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2019         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2020 #endif
2021
2022         vcpu->rip = regs->rip;
2023         kvm_arch_ops->set_rflags(vcpu, regs->rflags);
2024
2025         kvm_arch_ops->decache_regs(vcpu);
2026
2027         vcpu_put(vcpu);
2028
2029         return 0;
2030 }
2031
2032 static void get_segment(struct kvm_vcpu *vcpu,
2033                         struct kvm_segment *var, int seg)
2034 {
2035         return kvm_arch_ops->get_segment(vcpu, var, seg);
2036 }
2037
2038 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2039                                     struct kvm_sregs *sregs)
2040 {
2041         struct descriptor_table dt;
2042
2043         vcpu_load(vcpu);
2044
2045         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2046         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2047         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2048         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2049         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2050         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2051
2052         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2053         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2054
2055         kvm_arch_ops->get_idt(vcpu, &dt);
2056         sregs->idt.limit = dt.limit;
2057         sregs->idt.base = dt.base;
2058         kvm_arch_ops->get_gdt(vcpu, &dt);
2059         sregs->gdt.limit = dt.limit;
2060         sregs->gdt.base = dt.base;
2061
2062         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2063         sregs->cr0 = vcpu->cr0;
2064         sregs->cr2 = vcpu->cr2;
2065         sregs->cr3 = vcpu->cr3;
2066         sregs->cr4 = vcpu->cr4;
2067         sregs->cr8 = vcpu->cr8;
2068         sregs->efer = vcpu->shadow_efer;
2069         sregs->apic_base = vcpu->apic_base;
2070
2071         memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2072                sizeof sregs->interrupt_bitmap);
2073
2074         vcpu_put(vcpu);
2075
2076         return 0;
2077 }
2078
2079 static void set_segment(struct kvm_vcpu *vcpu,
2080                         struct kvm_segment *var, int seg)
2081 {
2082         return kvm_arch_ops->set_segment(vcpu, var, seg);
2083 }
2084
2085 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2086                                     struct kvm_sregs *sregs)
2087 {
2088         int mmu_reset_needed = 0;
2089         int i;
2090         struct descriptor_table dt;
2091
2092         vcpu_load(vcpu);
2093
2094         dt.limit = sregs->idt.limit;
2095         dt.base = sregs->idt.base;
2096         kvm_arch_ops->set_idt(vcpu, &dt);
2097         dt.limit = sregs->gdt.limit;
2098         dt.base = sregs->gdt.base;
2099         kvm_arch_ops->set_gdt(vcpu, &dt);
2100
2101         vcpu->cr2 = sregs->cr2;
2102         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2103         vcpu->cr3 = sregs->cr3;
2104
2105         vcpu->cr8 = sregs->cr8;
2106
2107         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2108 #ifdef CONFIG_X86_64
2109         kvm_arch_ops->set_efer(vcpu, sregs->efer);
2110 #endif
2111         vcpu->apic_base = sregs->apic_base;
2112
2113         kvm_arch_ops->decache_cr4_guest_bits(vcpu);
2114
2115         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2116         kvm_arch_ops->set_cr0(vcpu, sregs->cr0);
2117
2118         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2119         kvm_arch_ops->set_cr4(vcpu, sregs->cr4);
2120         if (!is_long_mode(vcpu) && is_pae(vcpu))
2121                 load_pdptrs(vcpu, vcpu->cr3);
2122
2123         if (mmu_reset_needed)
2124                 kvm_mmu_reset_context(vcpu);
2125
2126         memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2127                sizeof vcpu->irq_pending);
2128         vcpu->irq_summary = 0;
2129         for (i = 0; i < NR_IRQ_WORDS; ++i)
2130                 if (vcpu->irq_pending[i])
2131                         __set_bit(i, &vcpu->irq_summary);
2132
2133         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2134         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2135         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2136         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2137         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2138         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2139
2140         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2141         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2142
2143         vcpu_put(vcpu);
2144
2145         return 0;
2146 }
2147
2148 /*
2149  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
2150  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
2151  *
2152  * This list is modified at module load time to reflect the
2153  * capabilities of the host cpu.
2154  */
2155 static u32 msrs_to_save[] = {
2156         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
2157         MSR_K6_STAR,
2158 #ifdef CONFIG_X86_64
2159         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
2160 #endif
2161         MSR_IA32_TIME_STAMP_COUNTER,
2162 };
2163
2164 static unsigned num_msrs_to_save;
2165
2166 static u32 emulated_msrs[] = {
2167         MSR_IA32_MISC_ENABLE,
2168 };
2169
2170 static __init void kvm_init_msr_list(void)
2171 {
2172         u32 dummy[2];
2173         unsigned i, j;
2174
2175         for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2176                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2177                         continue;
2178                 if (j < i)
2179                         msrs_to_save[j] = msrs_to_save[i];
2180                 j++;
2181         }
2182         num_msrs_to_save = j;
2183 }
2184
2185 /*
2186  * Adapt set_msr() to msr_io()'s calling convention
2187  */
2188 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2189 {
2190         return kvm_set_msr(vcpu, index, *data);
2191 }
2192
2193 /*
2194  * Read or write a bunch of msrs. All parameters are kernel addresses.
2195  *
2196  * @return number of msrs set successfully.
2197  */
2198 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2199                     struct kvm_msr_entry *entries,
2200                     int (*do_msr)(struct kvm_vcpu *vcpu,
2201                                   unsigned index, u64 *data))
2202 {
2203         int i;
2204
2205         vcpu_load(vcpu);
2206
2207         for (i = 0; i < msrs->nmsrs; ++i)
2208                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2209                         break;
2210
2211         vcpu_put(vcpu);
2212
2213         return i;
2214 }
2215
2216 /*
2217  * Read or write a bunch of msrs. Parameters are user addresses.
2218  *
2219  * @return number of msrs set successfully.
2220  */
2221 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2222                   int (*do_msr)(struct kvm_vcpu *vcpu,
2223                                 unsigned index, u64 *data),
2224                   int writeback)
2225 {
2226         struct kvm_msrs msrs;
2227         struct kvm_msr_entry *entries;
2228         int r, n;
2229         unsigned size;
2230
2231         r = -EFAULT;
2232         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2233                 goto out;
2234
2235         r = -E2BIG;
2236         if (msrs.nmsrs >= MAX_IO_MSRS)
2237                 goto out;
2238
2239         r = -ENOMEM;
2240         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2241         entries = vmalloc(size);
2242         if (!entries)
2243                 goto out;
2244
2245         r = -EFAULT;
2246         if (copy_from_user(entries, user_msrs->entries, size))
2247                 goto out_free;
2248
2249         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2250         if (r < 0)
2251                 goto out_free;
2252
2253         r = -EFAULT;
2254         if (writeback && copy_to_user(user_msrs->entries, entries, size))
2255                 goto out_free;
2256
2257         r = n;
2258
2259 out_free:
2260         vfree(entries);
2261 out:
2262         return r;
2263 }
2264
2265 /*
2266  * Translate a guest virtual address to a guest physical address.
2267  */
2268 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2269                                     struct kvm_translation *tr)
2270 {
2271         unsigned long vaddr = tr->linear_address;
2272         gpa_t gpa;
2273
2274         vcpu_load(vcpu);
2275         spin_lock(&vcpu->kvm->lock);
2276         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2277         tr->physical_address = gpa;
2278         tr->valid = gpa != UNMAPPED_GVA;
2279         tr->writeable = 1;
2280         tr->usermode = 0;
2281         spin_unlock(&vcpu->kvm->lock);
2282         vcpu_put(vcpu);
2283
2284         return 0;
2285 }
2286
2287 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2288                                     struct kvm_interrupt *irq)
2289 {
2290         if (irq->irq < 0 || irq->irq >= 256)
2291                 return -EINVAL;
2292         vcpu_load(vcpu);
2293
2294         set_bit(irq->irq, vcpu->irq_pending);
2295         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2296
2297         vcpu_put(vcpu);
2298
2299         return 0;
2300 }
2301
2302 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2303                                       struct kvm_debug_guest *dbg)
2304 {
2305         int r;
2306
2307         vcpu_load(vcpu);
2308
2309         r = kvm_arch_ops->set_guest_debug(vcpu, dbg);
2310
2311         vcpu_put(vcpu);
2312
2313         return r;
2314 }
2315
2316 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2317                                     unsigned long address,
2318                                     int *type)
2319 {
2320         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2321         unsigned long pgoff;
2322         struct page *page;
2323
2324         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2325         if (pgoff == 0)
2326                 page = virt_to_page(vcpu->run);
2327         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2328                 page = virt_to_page(vcpu->pio_data);
2329         else
2330                 return NOPAGE_SIGBUS;
2331         get_page(page);
2332         if (type != NULL)
2333                 *type = VM_FAULT_MINOR;
2334
2335         return page;
2336 }
2337
2338 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2339         .nopage = kvm_vcpu_nopage,
2340 };
2341
2342 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2343 {
2344         vma->vm_ops = &kvm_vcpu_vm_ops;
2345         return 0;
2346 }
2347
2348 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2349 {
2350         struct kvm_vcpu *vcpu = filp->private_data;
2351
2352         fput(vcpu->kvm->filp);
2353         return 0;
2354 }
2355
2356 static struct file_operations kvm_vcpu_fops = {
2357         .release        = kvm_vcpu_release,
2358         .unlocked_ioctl = kvm_vcpu_ioctl,
2359         .compat_ioctl   = kvm_vcpu_ioctl,
2360         .mmap           = kvm_vcpu_mmap,
2361 };
2362
2363 /*
2364  * Allocates an inode for the vcpu.
2365  */
2366 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2367 {
2368         int fd, r;
2369         struct inode *inode;
2370         struct file *file;
2371
2372         r = anon_inode_getfd(&fd, &inode, &file,
2373                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2374         if (r)
2375                 return r;
2376         atomic_inc(&vcpu->kvm->filp->f_count);
2377         return fd;
2378 }
2379
2380 /*
2381  * Creates some virtual cpus.  Good luck creating more than one.
2382  */
2383 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2384 {
2385         int r;
2386         struct kvm_vcpu *vcpu;
2387         struct page *page;
2388
2389         r = -EINVAL;
2390         if (!valid_vcpu(n))
2391                 goto out;
2392
2393         vcpu = &kvm->vcpus[n];
2394         vcpu->vcpu_id = n;
2395
2396         mutex_lock(&vcpu->mutex);
2397
2398         if (vcpu->vmcs) {
2399                 mutex_unlock(&vcpu->mutex);
2400                 return -EEXIST;
2401         }
2402
2403         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2404         r = -ENOMEM;
2405         if (!page)
2406                 goto out_unlock;
2407         vcpu->run = page_address(page);
2408
2409         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2410         r = -ENOMEM;
2411         if (!page)
2412                 goto out_free_run;
2413         vcpu->pio_data = page_address(page);
2414
2415         vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
2416                                            FX_IMAGE_ALIGN);
2417         vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
2418         vcpu->cr0 = 0x10;
2419
2420         r = kvm_arch_ops->vcpu_create(vcpu);
2421         if (r < 0)
2422                 goto out_free_vcpus;
2423
2424         r = kvm_mmu_create(vcpu);
2425         if (r < 0)
2426                 goto out_free_vcpus;
2427
2428         kvm_arch_ops->vcpu_load(vcpu);
2429         r = kvm_mmu_setup(vcpu);
2430         if (r >= 0)
2431                 r = kvm_arch_ops->vcpu_setup(vcpu);
2432         vcpu_put(vcpu);
2433
2434         if (r < 0)
2435                 goto out_free_vcpus;
2436
2437         r = create_vcpu_fd(vcpu);
2438         if (r < 0)
2439                 goto out_free_vcpus;
2440
2441         spin_lock(&kvm_lock);
2442         if (n >= kvm->nvcpus)
2443                 kvm->nvcpus = n + 1;
2444         spin_unlock(&kvm_lock);
2445
2446         return r;
2447
2448 out_free_vcpus:
2449         kvm_free_vcpu(vcpu);
2450 out_free_run:
2451         free_page((unsigned long)vcpu->run);
2452         vcpu->run = NULL;
2453 out_unlock:
2454         mutex_unlock(&vcpu->mutex);
2455 out:
2456         return r;
2457 }
2458
2459 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2460 {
2461         u64 efer;
2462         int i;
2463         struct kvm_cpuid_entry *e, *entry;
2464
2465         rdmsrl(MSR_EFER, efer);
2466         entry = NULL;
2467         for (i = 0; i < vcpu->cpuid_nent; ++i) {
2468                 e = &vcpu->cpuid_entries[i];
2469                 if (e->function == 0x80000001) {
2470                         entry = e;
2471                         break;
2472                 }
2473         }
2474         if (entry && (entry->edx & (1 << 20)) && !(efer & EFER_NX)) {
2475                 entry->edx &= ~(1 << 20);
2476                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2477         }
2478 }
2479
2480 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2481                                     struct kvm_cpuid *cpuid,
2482                                     struct kvm_cpuid_entry __user *entries)
2483 {
2484         int r;
2485
2486         r = -E2BIG;
2487         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2488                 goto out;
2489         r = -EFAULT;
2490         if (copy_from_user(&vcpu->cpuid_entries, entries,
2491                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2492                 goto out;
2493         vcpu->cpuid_nent = cpuid->nent;
2494         cpuid_fix_nx_cap(vcpu);
2495         return 0;
2496
2497 out:
2498         return r;
2499 }
2500
2501 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2502 {
2503         if (sigset) {
2504                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2505                 vcpu->sigset_active = 1;
2506                 vcpu->sigset = *sigset;
2507         } else
2508                 vcpu->sigset_active = 0;
2509         return 0;
2510 }
2511
2512 /*
2513  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2514  * we have asm/x86/processor.h
2515  */
2516 struct fxsave {
2517         u16     cwd;
2518         u16     swd;
2519         u16     twd;
2520         u16     fop;
2521         u64     rip;
2522         u64     rdp;
2523         u32     mxcsr;
2524         u32     mxcsr_mask;
2525         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2526 #ifdef CONFIG_X86_64
2527         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2528 #else
2529         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2530 #endif
2531 };
2532
2533 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2534 {
2535         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2536
2537         vcpu_load(vcpu);
2538
2539         memcpy(fpu->fpr, fxsave->st_space, 128);
2540         fpu->fcw = fxsave->cwd;
2541         fpu->fsw = fxsave->swd;
2542         fpu->ftwx = fxsave->twd;
2543         fpu->last_opcode = fxsave->fop;
2544         fpu->last_ip = fxsave->rip;
2545         fpu->last_dp = fxsave->rdp;
2546         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2547
2548         vcpu_put(vcpu);
2549
2550         return 0;
2551 }
2552
2553 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2554 {
2555         struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
2556
2557         vcpu_load(vcpu);
2558
2559         memcpy(fxsave->st_space, fpu->fpr, 128);
2560         fxsave->cwd = fpu->fcw;
2561         fxsave->swd = fpu->fsw;
2562         fxsave->twd = fpu->ftwx;
2563         fxsave->fop = fpu->last_opcode;
2564         fxsave->rip = fpu->last_ip;
2565         fxsave->rdp = fpu->last_dp;
2566         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2567
2568         vcpu_put(vcpu);
2569
2570         return 0;
2571 }
2572
2573 static long kvm_vcpu_ioctl(struct file *filp,
2574                            unsigned int ioctl, unsigned long arg)
2575 {
2576         struct kvm_vcpu *vcpu = filp->private_data;
2577         void __user *argp = (void __user *)arg;
2578         int r = -EINVAL;
2579
2580         switch (ioctl) {
2581         case KVM_RUN:
2582                 r = -EINVAL;
2583                 if (arg)
2584                         goto out;
2585                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2586                 break;
2587         case KVM_GET_REGS: {
2588                 struct kvm_regs kvm_regs;
2589
2590                 memset(&kvm_regs, 0, sizeof kvm_regs);
2591                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2592                 if (r)
2593                         goto out;
2594                 r = -EFAULT;
2595                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2596                         goto out;
2597                 r = 0;
2598                 break;
2599         }
2600         case KVM_SET_REGS: {
2601                 struct kvm_regs kvm_regs;
2602
2603                 r = -EFAULT;
2604                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2605                         goto out;
2606                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2607                 if (r)
2608                         goto out;
2609                 r = 0;
2610                 break;
2611         }
2612         case KVM_GET_SREGS: {
2613                 struct kvm_sregs kvm_sregs;
2614
2615                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2616                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2617                 if (r)
2618                         goto out;
2619                 r = -EFAULT;
2620                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2621                         goto out;
2622                 r = 0;
2623                 break;
2624         }
2625         case KVM_SET_SREGS: {
2626                 struct kvm_sregs kvm_sregs;
2627
2628                 r = -EFAULT;
2629                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2630                         goto out;
2631                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2632                 if (r)
2633                         goto out;
2634                 r = 0;
2635                 break;
2636         }
2637         case KVM_TRANSLATE: {
2638                 struct kvm_translation tr;
2639
2640                 r = -EFAULT;
2641                 if (copy_from_user(&tr, argp, sizeof tr))
2642                         goto out;
2643                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2644                 if (r)
2645                         goto out;
2646                 r = -EFAULT;
2647                 if (copy_to_user(argp, &tr, sizeof tr))
2648                         goto out;
2649                 r = 0;
2650                 break;
2651         }
2652         case KVM_INTERRUPT: {
2653                 struct kvm_interrupt irq;
2654
2655                 r = -EFAULT;
2656                 if (copy_from_user(&irq, argp, sizeof irq))
2657                         goto out;
2658                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2659                 if (r)
2660                         goto out;
2661                 r = 0;
2662                 break;
2663         }
2664         case KVM_DEBUG_GUEST: {
2665                 struct kvm_debug_guest dbg;
2666
2667                 r = -EFAULT;
2668                 if (copy_from_user(&dbg, argp, sizeof dbg))
2669                         goto out;
2670                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2671                 if (r)
2672                         goto out;
2673                 r = 0;
2674                 break;
2675         }
2676         case KVM_GET_MSRS:
2677                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2678                 break;
2679         case KVM_SET_MSRS:
2680                 r = msr_io(vcpu, argp, do_set_msr, 0);
2681                 break;
2682         case KVM_SET_CPUID: {
2683                 struct kvm_cpuid __user *cpuid_arg = argp;
2684                 struct kvm_cpuid cpuid;
2685
2686                 r = -EFAULT;
2687                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2688                         goto out;
2689                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2690                 if (r)
2691                         goto out;
2692                 break;
2693         }
2694         case KVM_SET_SIGNAL_MASK: {
2695                 struct kvm_signal_mask __user *sigmask_arg = argp;
2696                 struct kvm_signal_mask kvm_sigmask;
2697                 sigset_t sigset, *p;
2698
2699                 p = NULL;
2700                 if (argp) {
2701                         r = -EFAULT;
2702                         if (copy_from_user(&kvm_sigmask, argp,
2703                                            sizeof kvm_sigmask))
2704                                 goto out;
2705                         r = -EINVAL;
2706                         if (kvm_sigmask.len != sizeof sigset)
2707                                 goto out;
2708                         r = -EFAULT;
2709                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2710                                            sizeof sigset))
2711                                 goto out;
2712                         p = &sigset;
2713                 }
2714                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2715                 break;
2716         }
2717         case KVM_GET_FPU: {
2718                 struct kvm_fpu fpu;
2719
2720                 memset(&fpu, 0, sizeof fpu);
2721                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2722                 if (r)
2723                         goto out;
2724                 r = -EFAULT;
2725                 if (copy_to_user(argp, &fpu, sizeof fpu))
2726                         goto out;
2727                 r = 0;
2728                 break;
2729         }
2730         case KVM_SET_FPU: {
2731                 struct kvm_fpu fpu;
2732
2733                 r = -EFAULT;
2734                 if (copy_from_user(&fpu, argp, sizeof fpu))
2735                         goto out;
2736                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2737                 if (r)
2738                         goto out;
2739                 r = 0;
2740                 break;
2741         }
2742         default:
2743                 ;
2744         }
2745 out:
2746         return r;
2747 }
2748
2749 static long kvm_vm_ioctl(struct file *filp,
2750                            unsigned int ioctl, unsigned long arg)
2751 {
2752         struct kvm *kvm = filp->private_data;
2753         void __user *argp = (void __user *)arg;
2754         int r = -EINVAL;
2755
2756         switch (ioctl) {
2757         case KVM_CREATE_VCPU:
2758                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2759                 if (r < 0)
2760                         goto out;
2761                 break;
2762         case KVM_SET_MEMORY_REGION: {
2763                 struct kvm_memory_region kvm_mem;
2764
2765                 r = -EFAULT;
2766                 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2767                         goto out;
2768                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_mem);
2769                 if (r)
2770                         goto out;
2771                 break;
2772         }
2773         case KVM_GET_DIRTY_LOG: {
2774                 struct kvm_dirty_log log;
2775
2776                 r = -EFAULT;
2777                 if (copy_from_user(&log, argp, sizeof log))
2778                         goto out;
2779                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2780                 if (r)
2781                         goto out;
2782                 break;
2783         }
2784         case KVM_SET_MEMORY_ALIAS: {
2785                 struct kvm_memory_alias alias;
2786
2787                 r = -EFAULT;
2788                 if (copy_from_user(&alias, argp, sizeof alias))
2789                         goto out;
2790                 r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
2791                 if (r)
2792                         goto out;
2793                 break;
2794         }
2795         default:
2796                 ;
2797         }
2798 out:
2799         return r;
2800 }
2801
2802 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2803                                   unsigned long address,
2804                                   int *type)
2805 {
2806         struct kvm *kvm = vma->vm_file->private_data;
2807         unsigned long pgoff;
2808         struct page *page;
2809
2810         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2811         page = gfn_to_page(kvm, pgoff);
2812         if (!page)
2813                 return NOPAGE_SIGBUS;
2814         get_page(page);
2815         if (type != NULL)
2816                 *type = VM_FAULT_MINOR;
2817
2818         return page;
2819 }
2820
2821 static struct vm_operations_struct kvm_vm_vm_ops = {
2822         .nopage = kvm_vm_nopage,
2823 };
2824
2825 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2826 {
2827         vma->vm_ops = &kvm_vm_vm_ops;
2828         return 0;
2829 }
2830
2831 static struct file_operations kvm_vm_fops = {
2832         .release        = kvm_vm_release,
2833         .unlocked_ioctl = kvm_vm_ioctl,
2834         .compat_ioctl   = kvm_vm_ioctl,
2835         .mmap           = kvm_vm_mmap,
2836 };
2837
2838 static int kvm_dev_ioctl_create_vm(void)
2839 {
2840         int fd, r;
2841         struct inode *inode;
2842         struct file *file;
2843         struct kvm *kvm;
2844
2845         kvm = kvm_create_vm();
2846         if (IS_ERR(kvm))
2847                 return PTR_ERR(kvm);
2848         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2849         if (r) {
2850                 kvm_destroy_vm(kvm);
2851                 return r;
2852         }
2853
2854         kvm->filp = file;
2855
2856         return fd;
2857 }
2858
2859 static long kvm_dev_ioctl(struct file *filp,
2860                           unsigned int ioctl, unsigned long arg)
2861 {
2862         void __user *argp = (void __user *)arg;
2863         long r = -EINVAL;
2864
2865         switch (ioctl) {
2866         case KVM_GET_API_VERSION:
2867                 r = -EINVAL;
2868                 if (arg)
2869                         goto out;
2870                 r = KVM_API_VERSION;
2871                 break;
2872         case KVM_CREATE_VM:
2873                 r = -EINVAL;
2874                 if (arg)
2875                         goto out;
2876                 r = kvm_dev_ioctl_create_vm();
2877                 break;
2878         case KVM_GET_MSR_INDEX_LIST: {
2879                 struct kvm_msr_list __user *user_msr_list = argp;
2880                 struct kvm_msr_list msr_list;
2881                 unsigned n;
2882
2883                 r = -EFAULT;
2884                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2885                         goto out;
2886                 n = msr_list.nmsrs;
2887                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2888                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2889                         goto out;
2890                 r = -E2BIG;
2891                 if (n < num_msrs_to_save)
2892                         goto out;
2893                 r = -EFAULT;
2894                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2895                                  num_msrs_to_save * sizeof(u32)))
2896                         goto out;
2897                 if (copy_to_user(user_msr_list->indices
2898                                  + num_msrs_to_save * sizeof(u32),
2899                                  &emulated_msrs,
2900                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2901                         goto out;
2902                 r = 0;
2903                 break;
2904         }
2905         case KVM_CHECK_EXTENSION:
2906                 /*
2907                  * No extensions defined at present.
2908                  */
2909                 r = 0;
2910                 break;
2911         case KVM_GET_VCPU_MMAP_SIZE:
2912                 r = -EINVAL;
2913                 if (arg)
2914                         goto out;
2915                 r = 2 * PAGE_SIZE;
2916                 break;
2917         default:
2918                 ;
2919         }
2920 out:
2921         return r;
2922 }
2923
2924 static struct file_operations kvm_chardev_ops = {
2925         .open           = kvm_dev_open,
2926         .release        = kvm_dev_release,
2927         .unlocked_ioctl = kvm_dev_ioctl,
2928         .compat_ioctl   = kvm_dev_ioctl,
2929 };
2930
2931 static struct miscdevice kvm_dev = {
2932         KVM_MINOR,
2933         "kvm",
2934         &kvm_chardev_ops,
2935 };
2936
2937 /*
2938  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2939  * cached on it.
2940  */
2941 static void decache_vcpus_on_cpu(int cpu)
2942 {
2943         struct kvm *vm;
2944         struct kvm_vcpu *vcpu;
2945         int i;
2946
2947         spin_lock(&kvm_lock);
2948         list_for_each_entry(vm, &vm_list, vm_list)
2949                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2950                         vcpu = &vm->vcpus[i];
2951                         /*
2952                          * If the vcpu is locked, then it is running on some
2953                          * other cpu and therefore it is not cached on the
2954                          * cpu in question.
2955                          *
2956                          * If it's not locked, check the last cpu it executed
2957                          * on.
2958                          */
2959                         if (mutex_trylock(&vcpu->mutex)) {
2960                                 if (vcpu->cpu == cpu) {
2961                                         kvm_arch_ops->vcpu_decache(vcpu);
2962                                         vcpu->cpu = -1;
2963                                 }
2964                                 mutex_unlock(&vcpu->mutex);
2965                         }
2966                 }
2967         spin_unlock(&kvm_lock);
2968 }
2969
2970 static void hardware_enable(void *junk)
2971 {
2972         int cpu = raw_smp_processor_id();
2973
2974         if (cpu_isset(cpu, cpus_hardware_enabled))
2975                 return;
2976         cpu_set(cpu, cpus_hardware_enabled);
2977         kvm_arch_ops->hardware_enable(NULL);
2978 }
2979
2980 static void hardware_disable(void *junk)
2981 {
2982         int cpu = raw_smp_processor_id();
2983
2984         if (!cpu_isset(cpu, cpus_hardware_enabled))
2985                 return;
2986         cpu_clear(cpu, cpus_hardware_enabled);
2987         decache_vcpus_on_cpu(cpu);
2988         kvm_arch_ops->hardware_disable(NULL);
2989 }
2990
2991 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2992                            void *v)
2993 {
2994         int cpu = (long)v;
2995
2996         switch (val) {
2997         case CPU_DYING:
2998         case CPU_DYING_FROZEN:
2999                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3000                        cpu);
3001                 hardware_disable(NULL);
3002                 break;
3003         case CPU_UP_CANCELED:
3004         case CPU_UP_CANCELED_FROZEN:
3005                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3006                        cpu);
3007                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3008                 break;
3009         case CPU_ONLINE:
3010         case CPU_ONLINE_FROZEN:
3011                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3012                        cpu);
3013                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3014                 break;
3015         }
3016         return NOTIFY_OK;
3017 }
3018
3019 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3020                        void *v)
3021 {
3022         if (val == SYS_RESTART) {
3023                 /*
3024                  * Some (well, at least mine) BIOSes hang on reboot if
3025                  * in vmx root mode.
3026                  */
3027                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3028                 on_each_cpu(hardware_disable, NULL, 0, 1);
3029         }
3030         return NOTIFY_OK;
3031 }
3032
3033 static struct notifier_block kvm_reboot_notifier = {
3034         .notifier_call = kvm_reboot,
3035         .priority = 0,
3036 };
3037
3038 void kvm_io_bus_init(struct kvm_io_bus *bus)
3039 {
3040         memset(bus, 0, sizeof(*bus));
3041 }
3042
3043 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3044 {
3045         int i;
3046
3047         for (i = 0; i < bus->dev_count; i++) {
3048                 struct kvm_io_device *pos = bus->devs[i];
3049
3050                 kvm_iodevice_destructor(pos);
3051         }
3052 }
3053
3054 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3055 {
3056         int i;
3057
3058         for (i = 0; i < bus->dev_count; i++) {
3059                 struct kvm_io_device *pos = bus->devs[i];
3060
3061                 if (pos->in_range(pos, addr))
3062                         return pos;
3063         }
3064
3065         return NULL;
3066 }
3067
3068 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3069 {
3070         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3071
3072         bus->devs[bus->dev_count++] = dev;
3073 }
3074
3075 static struct notifier_block kvm_cpu_notifier = {
3076         .notifier_call = kvm_cpu_hotplug,
3077         .priority = 20, /* must be > scheduler priority */
3078 };
3079
3080 static u64 stat_get(void *_offset)
3081 {
3082         unsigned offset = (long)_offset;
3083         u64 total = 0;
3084         struct kvm *kvm;
3085         struct kvm_vcpu *vcpu;
3086         int i;
3087
3088         spin_lock(&kvm_lock);
3089         list_for_each_entry(kvm, &vm_list, vm_list)
3090                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3091                         vcpu = &kvm->vcpus[i];
3092                         total += *(u32 *)((void *)vcpu + offset);
3093                 }
3094         spin_unlock(&kvm_lock);
3095         return total;
3096 }
3097
3098 static void stat_set(void *offset, u64 val)
3099 {
3100 }
3101
3102 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, stat_set, "%llu\n");
3103
3104 static __init void kvm_init_debug(void)
3105 {
3106         struct kvm_stats_debugfs_item *p;
3107
3108         debugfs_dir = debugfs_create_dir("kvm", NULL);
3109         for (p = debugfs_entries; p->name; ++p)
3110                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3111                                                 (void *)(long)p->offset,
3112                                                 &stat_fops);
3113 }
3114
3115 static void kvm_exit_debug(void)
3116 {
3117         struct kvm_stats_debugfs_item *p;
3118
3119         for (p = debugfs_entries; p->name; ++p)
3120                 debugfs_remove(p->dentry);
3121         debugfs_remove(debugfs_dir);
3122 }
3123
3124 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3125 {
3126         hardware_disable(NULL);
3127         return 0;
3128 }
3129
3130 static int kvm_resume(struct sys_device *dev)
3131 {
3132         hardware_enable(NULL);
3133         return 0;
3134 }
3135
3136 static struct sysdev_class kvm_sysdev_class = {
3137         set_kset_name("kvm"),
3138         .suspend = kvm_suspend,
3139         .resume = kvm_resume,
3140 };
3141
3142 static struct sys_device kvm_sysdev = {
3143         .id = 0,
3144         .cls = &kvm_sysdev_class,
3145 };
3146
3147 hpa_t bad_page_address;
3148
3149 int kvm_init_arch(struct kvm_arch_ops *ops, struct module *module)
3150 {
3151         int r;
3152
3153         if (kvm_arch_ops) {
3154                 printk(KERN_ERR "kvm: already loaded the other module\n");
3155                 return -EEXIST;
3156         }
3157
3158         if (!ops->cpu_has_kvm_support()) {
3159                 printk(KERN_ERR "kvm: no hardware support\n");
3160                 return -EOPNOTSUPP;
3161         }
3162         if (ops->disabled_by_bios()) {
3163                 printk(KERN_ERR "kvm: disabled by bios\n");
3164                 return -EOPNOTSUPP;
3165         }
3166
3167         kvm_arch_ops = ops;
3168
3169         r = kvm_arch_ops->hardware_setup();
3170         if (r < 0)
3171                 goto out;
3172
3173         on_each_cpu(hardware_enable, NULL, 0, 1);
3174         r = register_cpu_notifier(&kvm_cpu_notifier);
3175         if (r)
3176                 goto out_free_1;
3177         register_reboot_notifier(&kvm_reboot_notifier);
3178
3179         r = sysdev_class_register(&kvm_sysdev_class);
3180         if (r)
3181                 goto out_free_2;
3182
3183         r = sysdev_register(&kvm_sysdev);
3184         if (r)
3185                 goto out_free_3;
3186
3187         kvm_chardev_ops.owner = module;
3188
3189         r = misc_register(&kvm_dev);
3190         if (r) {
3191                 printk (KERN_ERR "kvm: misc device register failed\n");
3192                 goto out_free;
3193         }
3194
3195         return r;
3196
3197 out_free:
3198         sysdev_unregister(&kvm_sysdev);
3199 out_free_3:
3200         sysdev_class_unregister(&kvm_sysdev_class);
3201 out_free_2:
3202         unregister_reboot_notifier(&kvm_reboot_notifier);
3203         unregister_cpu_notifier(&kvm_cpu_notifier);
3204 out_free_1:
3205         on_each_cpu(hardware_disable, NULL, 0, 1);
3206         kvm_arch_ops->hardware_unsetup();
3207 out:
3208         kvm_arch_ops = NULL;
3209         return r;
3210 }
3211
3212 void kvm_exit_arch(void)
3213 {
3214         misc_deregister(&kvm_dev);
3215         sysdev_unregister(&kvm_sysdev);
3216         sysdev_class_unregister(&kvm_sysdev_class);
3217         unregister_reboot_notifier(&kvm_reboot_notifier);
3218         unregister_cpu_notifier(&kvm_cpu_notifier);
3219         on_each_cpu(hardware_disable, NULL, 0, 1);
3220         kvm_arch_ops->hardware_unsetup();
3221         kvm_arch_ops = NULL;
3222 }
3223
3224 static __init int kvm_init(void)
3225 {
3226         static struct page *bad_page;
3227         int r;
3228
3229         r = kvm_mmu_module_init();
3230         if (r)
3231                 goto out4;
3232
3233         kvm_init_debug();
3234
3235         kvm_init_msr_list();
3236
3237         if ((bad_page = alloc_page(GFP_KERNEL)) == NULL) {
3238                 r = -ENOMEM;
3239                 goto out;
3240         }
3241
3242         bad_page_address = page_to_pfn(bad_page) << PAGE_SHIFT;
3243         memset(__va(bad_page_address), 0, PAGE_SIZE);
3244
3245         return 0;
3246
3247 out:
3248         kvm_exit_debug();
3249         kvm_mmu_module_exit();
3250 out4:
3251         return r;
3252 }
3253
3254 static __exit void kvm_exit(void)
3255 {
3256         kvm_exit_debug();
3257         __free_page(pfn_to_page(bad_page_address >> PAGE_SHIFT));
3258         kvm_mmu_module_exit();
3259 }
3260
3261 module_init(kvm_init)
3262 module_exit(kvm_exit)
3263
3264 EXPORT_SYMBOL_GPL(kvm_init_arch);
3265 EXPORT_SYMBOL_GPL(kvm_exit_arch);