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