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.h"
  20 #include "x86_emulate.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 #include <linux/profile.h>
  42 #include <linux/kvm_para.h>
  43 #include <linux/pagemap.h>
  44 #include <linux/mman.h>
  45
  46 #include <asm/processor.h>
  47 #include <asm/msr.h>
  48 #include <asm/io.h>
  49 #include <asm/uaccess.h>
  50 #include <asm/desc.h>
  51
  52 MODULE_AUTHOR("Qumranet");
  53 MODULE_LICENSE("GPL");
  54
  55 static DEFINE_SPINLOCK(kvm_lock);
  56 static LIST_HEAD(vm_list);
  57
  58 static cpumask_t cpus_hardware_enabled;
  59
  60 struct kvm_x86_ops *kvm_x86_ops;
  61 struct kmem_cache *kvm_vcpu_cache;
  62 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  63
  64 static __read_mostly struct preempt_ops kvm_preempt_ops;
  65
  66 #define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
  67
  68 static struct kvm_stats_debugfs_item {
  69         const char *name;
  70         int offset;
  71         struct dentry *dentry;
  72 } debugfs_entries[] = {
  73         { "pf_fixed", STAT_OFFSET(pf_fixed) },
  74         { "pf_guest", STAT_OFFSET(pf_guest) },
  75         { "tlb_flush", STAT_OFFSET(tlb_flush) },
  76         { "invlpg", STAT_OFFSET(invlpg) },
  77         { "exits", STAT_OFFSET(exits) },
  78         { "io_exits", STAT_OFFSET(io_exits) },
  79         { "mmio_exits", STAT_OFFSET(mmio_exits) },
  80         { "signal_exits", STAT_OFFSET(signal_exits) },
  81         { "irq_window", STAT_OFFSET(irq_window_exits) },
  82         { "halt_exits", STAT_OFFSET(halt_exits) },
  83         { "halt_wakeup", STAT_OFFSET(halt_wakeup) },
  84         { "request_irq", STAT_OFFSET(request_irq_exits) },
  85         { "irq_exits", STAT_OFFSET(irq_exits) },
  86         { "light_exits", STAT_OFFSET(light_exits) },
  87         { "efer_reload", STAT_OFFSET(efer_reload) },
  88         { NULL }
  89 };
  90
  91 static struct dentry *debugfs_dir;
  92
  93 #define CR0_RESERVED_BITS                                               \
  94         (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
  95                           | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
  96                           | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
  97 #define CR4_RESERVED_BITS                                               \
  98         (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
  99                           | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE     \
 100                           | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR  \
 101                           | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
 102
 103 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
 104 #define EFER_RESERVED_BITS 0xfffffffffffff2fe
 105
 106 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 107                            unsigned long arg);
 108
 109 static inline int valid_vcpu(int n)
 110 {
 111         return likely(n >= 0 && n < KVM_MAX_VCPUS);
 112 }
 113
 114 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 115 {
 116         if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
 117                 return;
 118
 119         vcpu->guest_fpu_loaded = 1;
 120         fx_save(&vcpu->host_fx_image);
 121         fx_restore(&vcpu->guest_fx_image);
 122 }
 123 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 124
 125 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 126 {
 127         if (!vcpu->guest_fpu_loaded)
 128                 return;
 129
 130         vcpu->guest_fpu_loaded = 0;
 131         fx_save(&vcpu->guest_fx_image);
 132         fx_restore(&vcpu->host_fx_image);
 133 }
 134 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 135
 136 /*
 137  * Switches to specified vcpu, until a matching vcpu_put()
 138  */
 139 void vcpu_load(struct kvm_vcpu *vcpu)
 140 {
 141         int cpu;
 142
 143         mutex_lock(&vcpu->mutex);
 144         cpu = get_cpu();
 145         preempt_notifier_register(&vcpu->preempt_notifier);
 146         kvm_arch_vcpu_load(vcpu, cpu);
 147         put_cpu();
 148 }
 149
 150 void vcpu_put(struct kvm_vcpu *vcpu)
 151 {
 152         preempt_disable();
 153         kvm_arch_vcpu_put(vcpu);
 154         preempt_notifier_unregister(&vcpu->preempt_notifier);
 155         preempt_enable();
 156         mutex_unlock(&vcpu->mutex);
 157 }
 158
 159 static void ack_flush(void *_completed)
 160 {
 161 }
 162
 163 void kvm_flush_remote_tlbs(struct kvm *kvm)
 164 {
 165         int i, cpu;
 166         cpumask_t cpus;
 167         struct kvm_vcpu *vcpu;
 168
 169         cpus_clear(cpus);
 170         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 171                 vcpu = kvm->vcpus[i];
 172                 if (!vcpu)
 173                         continue;
 174                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 175                         continue;
 176                 cpu = vcpu->cpu;
 177                 if (cpu != -1 && cpu != raw_smp_processor_id())
 178                         cpu_set(cpu, cpus);
 179         }
 180         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 181 }
 182
 183 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 184 {
 185         struct page *page;
 186         int r;
 187
 188         mutex_init(&vcpu->mutex);
 189         vcpu->cpu = -1;
 190         vcpu->mmu.root_hpa = INVALID_PAGE;
 191         vcpu->kvm = kvm;
 192         vcpu->vcpu_id = id;
 193         if (!irqchip_in_kernel(kvm) || id == 0)
 194                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
 195         else
 196                 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
 197         init_waitqueue_head(&vcpu->wq);
 198
 199         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 200         if (!page) {
 201                 r = -ENOMEM;
 202                 goto fail;
 203         }
 204         vcpu->run = page_address(page);
 205
 206         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 207         if (!page) {
 208                 r = -ENOMEM;
 209                 goto fail_free_run;
 210         }
 211         vcpu->pio_data = page_address(page);
 212
 213         r = kvm_mmu_create(vcpu);
 214         if (r < 0)
 215                 goto fail_free_pio_data;
 216
 217         if (irqchip_in_kernel(kvm)) {
 218                 r = kvm_create_lapic(vcpu);
 219                 if (r < 0)
 220                         goto fail_mmu_destroy;
 221         }
 222
 223         return 0;
 224
 225 fail_mmu_destroy:
 226         kvm_mmu_destroy(vcpu);
 227 fail_free_pio_data:
 228         free_page((unsigned long)vcpu->pio_data);
 229 fail_free_run:
 230         free_page((unsigned long)vcpu->run);
 231 fail:
 232         return r;
 233 }
 234 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 235
 236 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 237 {
 238         kvm_free_lapic(vcpu);
 239         kvm_mmu_destroy(vcpu);
 240         free_page((unsigned long)vcpu->pio_data);
 241         free_page((unsigned long)vcpu->run);
 242 }
 243 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 244
 245 static struct kvm *kvm_create_vm(void)
 246 {
 247         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 248
 249         if (!kvm)
 250                 return ERR_PTR(-ENOMEM);
 251
 252         kvm_io_bus_init(&kvm->pio_bus);
 253         mutex_init(&kvm->lock);
 254         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 255         kvm_io_bus_init(&kvm->mmio_bus);
 256         spin_lock(&kvm_lock);
 257         list_add(&kvm->vm_list, &vm_list);
 258         spin_unlock(&kvm_lock);
 259         return kvm;
 260 }
 261
 262 /*
 263  * Free any memory in @free but not in @dont.
 264  */
 265 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 266                                   struct kvm_memory_slot *dont)
 267 {
 268         if (!dont || free->rmap != dont->rmap)
 269                 vfree(free->rmap);
 270
 271         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 272                 vfree(free->dirty_bitmap);
 273
 274         free->npages = 0;
 275         free->dirty_bitmap = NULL;
 276         free->rmap = NULL;
 277 }
 278
 279 static void kvm_free_physmem(struct kvm *kvm)
 280 {
 281         int i;
 282
 283         for (i = 0; i < kvm->nmemslots; ++i)
 284                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 285 }
 286
 287 static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
 288 {
 289         int i;
 290
 291         for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
 292                 if (vcpu->pio.guest_pages[i]) {
 293                         kvm_release_page(vcpu->pio.guest_pages[i]);
 294                         vcpu->pio.guest_pages[i] = NULL;
 295                 }
 296 }
 297
 298 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 299 {
 300         vcpu_load(vcpu);
 301         kvm_mmu_unload(vcpu);
 302         vcpu_put(vcpu);
 303 }
 304
 305 static void kvm_free_vcpus(struct kvm *kvm)
 306 {
 307         unsigned int i;
 308
 309         /*
 310          * Unpin any mmu pages first.
 311          */
 312         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 313                 if (kvm->vcpus[i])
 314                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
 315         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 316                 if (kvm->vcpus[i]) {
 317                         kvm_x86_ops->vcpu_free(kvm->vcpus[i]);
 318                         kvm->vcpus[i] = NULL;
 319                 }
 320         }
 321
 322 }
 323
 324 static void kvm_destroy_vm(struct kvm *kvm)
 325 {
 326         spin_lock(&kvm_lock);
 327         list_del(&kvm->vm_list);
 328         spin_unlock(&kvm_lock);
 329         kvm_io_bus_destroy(&kvm->pio_bus);
 330         kvm_io_bus_destroy(&kvm->mmio_bus);
 331         kfree(kvm->vpic);
 332         kfree(kvm->vioapic);
 333         kvm_free_vcpus(kvm);
 334         kvm_free_physmem(kvm);
 335         kfree(kvm);
 336 }
 337
 338 static int kvm_vm_release(struct inode *inode, struct file *filp)
 339 {
 340         struct kvm *kvm = filp->private_data;
 341
 342         kvm_destroy_vm(kvm);
 343         return 0;
 344 }
 345
 346 static void inject_gp(struct kvm_vcpu *vcpu)
 347 {
 348         kvm_x86_ops->inject_gp(vcpu, 0);
 349 }
 350
 351 /*
 352  * Load the pae pdptrs.  Return true is they are all valid.
 353  */
 354 static int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 355 {
 356         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
 357         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
 358         int i;
 359         int ret;
 360         u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
 361
 362         mutex_lock(&vcpu->kvm->lock);
 363         ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
 364                                   offset * sizeof(u64), sizeof(pdpte));
 365         if (ret < 0) {
 366                 ret = 0;
 367                 goto out;
 368         }
 369         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
 370                 if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
 371                         ret = 0;
 372                         goto out;
 373                 }
 374         }
 375         ret = 1;
 376
 377         memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
 378 out:
 379         mutex_unlock(&vcpu->kvm->lock);
 380
 381         return ret;
 382 }
 383
 384 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 385 {
 386         if (cr0 & CR0_RESERVED_BITS) {
 387                 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
 388                        cr0, vcpu->cr0);
 389                 inject_gp(vcpu);
 390                 return;
 391         }
 392
 393         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
 394                 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
 395                 inject_gp(vcpu);
 396                 return;
 397         }
 398
 399         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
 400                 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
 401                        "and a clear PE flag\n");
 402                 inject_gp(vcpu);
 403                 return;
 404         }
 405
 406         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 407 #ifdef CONFIG_X86_64
 408                 if ((vcpu->shadow_efer & EFER_LME)) {
 409                         int cs_db, cs_l;
 410
 411                         if (!is_pae(vcpu)) {
 412                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 413                                        "in long mode while PAE is disabled\n");
 414                                 inject_gp(vcpu);
 415                                 return;
 416                         }
 417                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 418                         if (cs_l) {
 419                                 printk(KERN_DEBUG "set_cr0: #GP, start paging "
 420                                        "in long mode while CS.L == 1\n");
 421                                 inject_gp(vcpu);
 422                                 return;
 423
 424                         }
 425                 } else
 426 #endif
 427                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
 428                         printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
 429                                "reserved bits\n");
 430                         inject_gp(vcpu);
 431                         return;
 432                 }
 433
 434         }
 435
 436         kvm_x86_ops->set_cr0(vcpu, cr0);
 437         vcpu->cr0 = cr0;
 438
 439         mutex_lock(&vcpu->kvm->lock);
 440         kvm_mmu_reset_context(vcpu);
 441         mutex_unlock(&vcpu->kvm->lock);
 442         return;
 443 }
 444 EXPORT_SYMBOL_GPL(set_cr0);
 445
 446 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
 447 {
 448         set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
 449 }
 450 EXPORT_SYMBOL_GPL(lmsw);
 451
 452 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 453 {
 454         if (cr4 & CR4_RESERVED_BITS) {
 455                 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
 456                 inject_gp(vcpu);
 457                 return;
 458         }
 459
 460         if (is_long_mode(vcpu)) {
 461                 if (!(cr4 & X86_CR4_PAE)) {
 462                         printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
 463                                "in long mode\n");
 464                         inject_gp(vcpu);
 465                         return;
 466                 }
 467         } else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
 468                    && !load_pdptrs(vcpu, vcpu->cr3)) {
 469                 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
 470                 inject_gp(vcpu);
 471                 return;
 472         }
 473
 474         if (cr4 & X86_CR4_VMXE) {
 475                 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
 476                 inject_gp(vcpu);
 477                 return;
 478         }
 479         kvm_x86_ops->set_cr4(vcpu, cr4);
 480         vcpu->cr4 = cr4;
 481         mutex_lock(&vcpu->kvm->lock);
 482         kvm_mmu_reset_context(vcpu);
 483         mutex_unlock(&vcpu->kvm->lock);
 484 }
 485 EXPORT_SYMBOL_GPL(set_cr4);
 486
 487 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
 488 {
 489         if (is_long_mode(vcpu)) {
 490                 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
 491                         printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
 492                         inject_gp(vcpu);
 493                         return;
 494                 }
 495         } else {
 496                 if (is_pae(vcpu)) {
 497                         if (cr3 & CR3_PAE_RESERVED_BITS) {
 498                                 printk(KERN_DEBUG
 499                                        "set_cr3: #GP, reserved bits\n");
 500                                 inject_gp(vcpu);
 501                                 return;
 502                         }
 503                         if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
 504                                 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
 505                                        "reserved bits\n");
 506                                 inject_gp(vcpu);
 507                                 return;
 508                         }
 509                 }
 510                 /*
 511                  * We don't check reserved bits in nonpae mode, because
 512                  * this isn't enforced, and VMware depends on this.
 513                  */
 514         }
 515
 516         mutex_lock(&vcpu->kvm->lock);
 517         /*
 518          * Does the new cr3 value map to physical memory? (Note, we
 519          * catch an invalid cr3 even in real-mode, because it would
 520          * cause trouble later on when we turn on paging anyway.)
 521          *
 522          * A real CPU would silently accept an invalid cr3 and would
 523          * attempt to use it - with largely undefined (and often hard
 524          * to debug) behavior on the guest side.
 525          */
 526         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
 527                 inject_gp(vcpu);
 528         else {
 529                 vcpu->cr3 = cr3;
 530                 vcpu->mmu.new_cr3(vcpu);
 531         }
 532         mutex_unlock(&vcpu->kvm->lock);
 533 }
 534 EXPORT_SYMBOL_GPL(set_cr3);
 535
 536 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
 537 {
 538         if (cr8 & CR8_RESERVED_BITS) {
 539                 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
 540                 inject_gp(vcpu);
 541                 return;
 542         }
 543         if (irqchip_in_kernel(vcpu->kvm))
 544                 kvm_lapic_set_tpr(vcpu, cr8);
 545         else
 546                 vcpu->cr8 = cr8;
 547 }
 548 EXPORT_SYMBOL_GPL(set_cr8);
 549
 550 unsigned long get_cr8(struct kvm_vcpu *vcpu)
 551 {
 552         if (irqchip_in_kernel(vcpu->kvm))
 553                 return kvm_lapic_get_cr8(vcpu);
 554         else
 555                 return vcpu->cr8;
 556 }
 557 EXPORT_SYMBOL_GPL(get_cr8);
 558
 559 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
 560 {
 561         if (irqchip_in_kernel(vcpu->kvm))
 562                 return vcpu->apic_base;
 563         else
 564                 return vcpu->apic_base;
 565 }
 566 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
 567
 568 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
 569 {
 570         /* TODO: reserve bits check */
 571         if (irqchip_in_kernel(vcpu->kvm))
 572                 kvm_lapic_set_base(vcpu, data);
 573         else
 574                 vcpu->apic_base = data;
 575 }
 576 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
 577
 578 void fx_init(struct kvm_vcpu *vcpu)
 579 {
 580         unsigned after_mxcsr_mask;
 581
 582         /* Initialize guest FPU by resetting ours and saving into guest's */
 583         preempt_disable();
 584         fx_save(&vcpu->host_fx_image);
 585         fpu_init();
 586         fx_save(&vcpu->guest_fx_image);
 587         fx_restore(&vcpu->host_fx_image);
 588         preempt_enable();
 589
 590         vcpu->cr0 |= X86_CR0_ET;
 591         after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
 592         vcpu->guest_fx_image.mxcsr = 0x1f80;
 593         memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
 594                0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
 595 }
 596 EXPORT_SYMBOL_GPL(fx_init);
 597
 598 /*
 599  * Allocate some memory and give it an address in the guest physical address
 600  * space.
 601  *
 602  * Discontiguous memory is allowed, mostly for framebuffers.
 603  */
 604 int kvm_set_memory_region(struct kvm *kvm,
 605                           struct kvm_userspace_memory_region *mem,
 606                           int user_alloc)
 607 {
 608         int r;
 609         gfn_t base_gfn;
 610         unsigned long npages;
 611         unsigned long i;
 612         struct kvm_memory_slot *memslot;
 613         struct kvm_memory_slot old, new;
 614
 615         r = -EINVAL;
 616         /* General sanity checks */
 617         if (mem->memory_size & (PAGE_SIZE - 1))
 618                 goto out;
 619         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 620                 goto out;
 621         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 622                 goto out;
 623         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 624                 goto out;
 625
 626         memslot = &kvm->memslots[mem->slot];
 627         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 628         npages = mem->memory_size >> PAGE_SHIFT;
 629
 630         if (!npages)
 631                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 632
 633         mutex_lock(&kvm->lock);
 634
 635         new = old = *memslot;
 636
 637         new.base_gfn = base_gfn;
 638         new.npages = npages;
 639         new.flags = mem->flags;
 640
 641         /* Disallow changing a memory slot's size. */
 642         r = -EINVAL;
 643         if (npages && old.npages && npages != old.npages)
 644                 goto out_unlock;
 645
 646         /* Check for overlaps */
 647         r = -EEXIST;
 648         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 649                 struct kvm_memory_slot *s = &kvm->memslots[i];
 650
 651                 if (s == memslot)
 652                         continue;
 653                 if (!((base_gfn + npages <= s->base_gfn) ||
 654                       (base_gfn >= s->base_gfn + s->npages)))
 655                         goto out_unlock;
 656         }
 657
 658         /* Free page dirty bitmap if unneeded */
 659         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 660                 new.dirty_bitmap = NULL;
 661
 662         r = -ENOMEM;
 663
 664         /* Allocate if a slot is being created */
 665         if (npages && !new.rmap) {
 666                 new.rmap = vmalloc(npages * sizeof(struct page *));
 667
 668                 if (!new.rmap)
 669                         goto out_unlock;
 670
 671                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 672
 673                 new.user_alloc = user_alloc;
 674                 if (user_alloc)
 675                         new.userspace_addr = mem->userspace_addr;
 676                 else {
 677                         down_write(&current->mm->mmap_sem);
 678                         new.userspace_addr = do_mmap(NULL, 0,
 679                                                      npages * PAGE_SIZE,
 680                                                      PROT_READ | PROT_WRITE,
 681                                                      MAP_SHARED | MAP_ANONYMOUS,
 682                                                      0);
 683                         up_write(&current->mm->mmap_sem);
 684
 685                         if (IS_ERR((void *)new.userspace_addr))
 686                                 goto out_unlock;
 687                 }
 688         } else {
 689                 if (!old.user_alloc && old.rmap) {
 690                         int ret;
 691
 692                         down_write(&current->mm->mmap_sem);
 693                         ret = do_munmap(current->mm, old.userspace_addr,
 694                                         old.npages * PAGE_SIZE);
 695                         up_write(&current->mm->mmap_sem);
 696                         if (ret < 0)
 697                                 printk(KERN_WARNING
 698                                        "kvm_vm_ioctl_set_memory_region: "
 699                                        "failed to munmap memory\n");
 700                 }
 701         }
 702
 703         /* Allocate page dirty bitmap if needed */
 704         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 705                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 706
 707                 new.dirty_bitmap = vmalloc(dirty_bytes);
 708                 if (!new.dirty_bitmap)
 709                         goto out_unlock;
 710                 memset(new.dirty_bitmap, 0, dirty_bytes);
 711         }
 712
 713         if (mem->slot >= kvm->nmemslots)
 714                 kvm->nmemslots = mem->slot + 1;
 715
 716         if (!kvm->n_requested_mmu_pages) {
 717                 unsigned int n_pages;
 718
 719                 if (npages) {
 720                         n_pages = npages * KVM_PERMILLE_MMU_PAGES / 1000;
 721                         kvm_mmu_change_mmu_pages(kvm, kvm->n_alloc_mmu_pages +
 722                                                  n_pages);
 723                 } else {
 724                         unsigned int nr_mmu_pages;
 725
 726                         n_pages = old.npages * KVM_PERMILLE_MMU_PAGES / 1000;
 727                         nr_mmu_pages = kvm->n_alloc_mmu_pages - n_pages;
 728                         nr_mmu_pages = max(nr_mmu_pages,
 729                                         (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
 730                         kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
 731                 }
 732         }
 733
 734         *memslot = new;
 735
 736         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 737         kvm_flush_remote_tlbs(kvm);
 738
 739         mutex_unlock(&kvm->lock);
 740
 741         kvm_free_physmem_slot(&old, &new);
 742         return 0;
 743
 744 out_unlock:
 745         mutex_unlock(&kvm->lock);
 746         kvm_free_physmem_slot(&new, &old);
 747 out:
 748         return r;
 749
 750 }
 751 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 752
 753 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 754                                    struct
 755                                    kvm_userspace_memory_region *mem,
 756                                    int user_alloc)
 757 {
 758         if (mem->slot >= KVM_MEMORY_SLOTS)
 759                 return -EINVAL;
 760         return kvm_set_memory_region(kvm, mem, user_alloc);
 761 }
 762
 763 /*
 764  * Get (and clear) the dirty memory log for a memory slot.
 765  */
 766 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 767                                       struct kvm_dirty_log *log)
 768 {
 769         struct kvm_memory_slot *memslot;
 770         int r, i;
 771         int n;
 772         unsigned long any = 0;
 773
 774         mutex_lock(&kvm->lock);
 775
 776         r = -EINVAL;
 777         if (log->slot >= KVM_MEMORY_SLOTS)
 778                 goto out;
 779
 780         memslot = &kvm->memslots[log->slot];
 781         r = -ENOENT;
 782         if (!memslot->dirty_bitmap)
 783                 goto out;
 784
 785         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 786
 787         for (i = 0; !any && i < n/sizeof(long); ++i)
 788                 any = memslot->dirty_bitmap[i];
 789
 790         r = -EFAULT;
 791         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 792                 goto out;
 793
 794         /* If nothing is dirty, don't bother messing with page tables. */
 795         if (any) {
 796                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 797                 kvm_flush_remote_tlbs(kvm);
 798                 memset(memslot->dirty_bitmap, 0, n);
 799         }
 800
 801         r = 0;
 802
 803 out:
 804         mutex_unlock(&kvm->lock);
 805         return r;
 806 }
 807
 808 int is_error_page(struct page *page)
 809 {
 810         return page == bad_page;
 811 }
 812 EXPORT_SYMBOL_GPL(is_error_page);
 813
 814 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 815 {
 816         int i;
 817         struct kvm_mem_alias *alias;
 818
 819         for (i = 0; i < kvm->naliases; ++i) {
 820                 alias = &kvm->aliases[i];
 821                 if (gfn >= alias->base_gfn
 822                     && gfn < alias->base_gfn + alias->npages)
 823                         return alias->target_gfn + gfn - alias->base_gfn;
 824         }
 825         return gfn;
 826 }
 827
 828 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 829 {
 830         int i;
 831
 832         for (i = 0; i < kvm->nmemslots; ++i) {
 833                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 834
 835                 if (gfn >= memslot->base_gfn
 836                     && gfn < memslot->base_gfn + memslot->npages)
 837                         return memslot;
 838         }
 839         return NULL;
 840 }
 841
 842 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 843 {
 844         gfn = unalias_gfn(kvm, gfn);
 845         return __gfn_to_memslot(kvm, gfn);
 846 }
 847
 848 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 849 {
 850         int i;
 851
 852         gfn = unalias_gfn(kvm, gfn);
 853         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 854                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 855
 856                 if (gfn >= memslot->base_gfn
 857                     && gfn < memslot->base_gfn + memslot->npages)
 858                         return 1;
 859         }
 860         return 0;
 861 }
 862 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 863
 864 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 865 {
 866         struct kvm_memory_slot *slot;
 867         struct page *page[1];
 868         int npages;
 869
 870         might_sleep();
 871
 872         gfn = unalias_gfn(kvm, gfn);
 873         slot = __gfn_to_memslot(kvm, gfn);
 874         if (!slot) {
 875                 get_page(bad_page);
 876                 return bad_page;
 877         }
 878
 879         down_read(&current->mm->mmap_sem);
 880         npages = get_user_pages(current, current->mm,
 881                                 slot->userspace_addr
 882                                 + (gfn - slot->base_gfn) * PAGE_SIZE, 1,
 883                                 1, 1, page, NULL);
 884         up_read(&current->mm->mmap_sem);
 885         if (npages != 1) {
 886                 get_page(bad_page);
 887                 return bad_page;
 888         }
 889
 890         return page[0];
 891 }
 892 EXPORT_SYMBOL_GPL(gfn_to_page);
 893
 894 void kvm_release_page(struct page *page)
 895 {
 896         if (!PageReserved(page))
 897                 SetPageDirty(page);
 898         put_page(page);
 899 }
 900 EXPORT_SYMBOL_GPL(kvm_release_page);
 901
 902 static int next_segment(unsigned long len, int offset)
 903 {
 904         if (len > PAGE_SIZE - offset)
 905                 return PAGE_SIZE - offset;
 906         else
 907                 return len;
 908 }
 909
 910 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 911                         int len)
 912 {
 913         void *page_virt;
 914         struct page *page;
 915
 916         page = gfn_to_page(kvm, gfn);
 917         if (is_error_page(page)) {
 918                 kvm_release_page(page);
 919                 return -EFAULT;
 920         }
 921         page_virt = kmap_atomic(page, KM_USER0);
 922
 923         memcpy(data, page_virt + offset, len);
 924
 925         kunmap_atomic(page_virt, KM_USER0);
 926         kvm_release_page(page);
 927         return 0;
 928 }
 929 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 930
 931 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 932 {
 933         gfn_t gfn = gpa >> PAGE_SHIFT;
 934         int seg;
 935         int offset = offset_in_page(gpa);
 936         int ret;
 937
 938         while ((seg = next_segment(len, offset)) != 0) {
 939                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 940                 if (ret < 0)
 941                         return ret;
 942                 offset = 0;
 943                 len -= seg;
 944                 data += seg;
 945                 ++gfn;
 946         }
 947         return 0;
 948 }
 949 EXPORT_SYMBOL_GPL(kvm_read_guest);
 950
 951 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
 952                          int offset, int len)
 953 {
 954         void *page_virt;
 955         struct page *page;
 956
 957         page = gfn_to_page(kvm, gfn);
 958         if (is_error_page(page)) {
 959                 kvm_release_page(page);
 960                 return -EFAULT;
 961         }
 962         page_virt = kmap_atomic(page, KM_USER0);
 963
 964         memcpy(page_virt + offset, data, len);
 965
 966         kunmap_atomic(page_virt, KM_USER0);
 967         mark_page_dirty(kvm, gfn);
 968         kvm_release_page(page);
 969         return 0;
 970 }
 971 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
 972
 973 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
 974                     unsigned long len)
 975 {
 976         gfn_t gfn = gpa >> PAGE_SHIFT;
 977         int seg;
 978         int offset = offset_in_page(gpa);
 979         int ret;
 980
 981         while ((seg = next_segment(len, offset)) != 0) {
 982                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
 983                 if (ret < 0)
 984                         return ret;
 985                 offset = 0;
 986                 len -= seg;
 987                 data += seg;
 988                 ++gfn;
 989         }
 990         return 0;
 991 }
 992
 993 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
 994 {
 995         void *page_virt;
 996         struct page *page;
 997
 998         page = gfn_to_page(kvm, gfn);
 999         if (is_error_page(page)) {
1000                 kvm_release_page(page);
1001                 return -EFAULT;
1002         }
1003         page_virt = kmap_atomic(page, KM_USER0);
1004
1005         memset(page_virt + offset, 0, len);
1006
1007         kunmap_atomic(page_virt, KM_USER0);
1008         kvm_release_page(page);
1009         return 0;
1010 }
1011 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1012
1013 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1014 {
1015         gfn_t gfn = gpa >> PAGE_SHIFT;
1016         int seg;
1017         int offset = offset_in_page(gpa);
1018         int ret;
1019
1020         while ((seg = next_segment(len, offset)) != 0) {
1021                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1022                 if (ret < 0)
1023                         return ret;
1024                 offset = 0;
1025                 len -= seg;
1026                 ++gfn;
1027         }
1028         return 0;
1029 }
1030 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1031
1032 /* WARNING: Does not work on aliased pages. */
1033 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1034 {
1035         struct kvm_memory_slot *memslot;
1036
1037         memslot = __gfn_to_memslot(kvm, gfn);
1038         if (memslot && memslot->dirty_bitmap) {
1039                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1040
1041                 /* avoid RMW */
1042                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1043                         set_bit(rel_gfn, memslot->dirty_bitmap);
1044         }
1045 }
1046
1047 int emulator_read_std(unsigned long addr,
1048                              void *val,
1049                              unsigned int bytes,
1050                              struct kvm_vcpu *vcpu)
1051 {
1052         void *data = val;
1053
1054         while (bytes) {
1055                 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1056                 unsigned offset = addr & (PAGE_SIZE-1);
1057                 unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1058                 int ret;
1059
1060                 if (gpa == UNMAPPED_GVA)
1061                         return X86EMUL_PROPAGATE_FAULT;
1062                 ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1063                 if (ret < 0)
1064                         return X86EMUL_UNHANDLEABLE;
1065
1066                 bytes -= tocopy;
1067                 data += tocopy;
1068                 addr += tocopy;
1069         }
1070
1071         return X86EMUL_CONTINUE;
1072 }
1073 EXPORT_SYMBOL_GPL(emulator_read_std);
1074
1075 static int emulator_write_std(unsigned long addr,
1076                               const void *val,
1077                               unsigned int bytes,
1078                               struct kvm_vcpu *vcpu)
1079 {
1080         pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1081         return X86EMUL_UNHANDLEABLE;
1082 }
1083
1084 /*
1085  * Only apic need an MMIO device hook, so shortcut now..
1086  */
1087 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1088                                                 gpa_t addr)
1089 {
1090         struct kvm_io_device *dev;
1091
1092         if (vcpu->apic) {
1093                 dev = &vcpu->apic->dev;
1094                 if (dev->in_range(dev, addr))
1095                         return dev;
1096         }
1097         return NULL;
1098 }
1099
1100 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1101                                                 gpa_t addr)
1102 {
1103         struct kvm_io_device *dev;
1104
1105         dev = vcpu_find_pervcpu_dev(vcpu, addr);
1106         if (dev == NULL)
1107                 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1108         return dev;
1109 }
1110
1111 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1112                                                gpa_t addr)
1113 {
1114         return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1115 }
1116
1117 static int emulator_read_emulated(unsigned long addr,
1118                                   void *val,
1119                                   unsigned int bytes,
1120                                   struct kvm_vcpu *vcpu)
1121 {
1122         struct kvm_io_device *mmio_dev;
1123         gpa_t                 gpa;
1124
1125         if (vcpu->mmio_read_completed) {
1126                 memcpy(val, vcpu->mmio_data, bytes);
1127                 vcpu->mmio_read_completed = 0;
1128                 return X86EMUL_CONTINUE;
1129         } else if (emulator_read_std(addr, val, bytes, vcpu)
1130                    == X86EMUL_CONTINUE)
1131                 return X86EMUL_CONTINUE;
1132
1133         gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1134         if (gpa == UNMAPPED_GVA)
1135                 return X86EMUL_PROPAGATE_FAULT;
1136
1137         /*
1138          * Is this MMIO handled locally?
1139          */
1140         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1141         if (mmio_dev) {
1142                 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1143                 return X86EMUL_CONTINUE;
1144         }
1145
1146         vcpu->mmio_needed = 1;
1147         vcpu->mmio_phys_addr = gpa;
1148         vcpu->mmio_size = bytes;
1149         vcpu->mmio_is_write = 0;
1150
1151         return X86EMUL_UNHANDLEABLE;
1152 }
1153
1154 static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1155                                const void *val, int bytes)
1156 {
1157         int ret;
1158
1159         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1160         if (ret < 0)
1161                 return 0;
1162         kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1163         return 1;
1164 }
1165
1166 static int emulator_write_emulated_onepage(unsigned long addr,
1167                                            const void *val,
1168                                            unsigned int bytes,
1169                                            struct kvm_vcpu *vcpu)
1170 {
1171         struct kvm_io_device *mmio_dev;
1172         gpa_t                 gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1173
1174         if (gpa == UNMAPPED_GVA) {
1175                 kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1176                 return X86EMUL_PROPAGATE_FAULT;
1177         }
1178
1179         if (emulator_write_phys(vcpu, gpa, val, bytes))
1180                 return X86EMUL_CONTINUE;
1181
1182         /*
1183          * Is this MMIO handled locally?
1184          */
1185         mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1186         if (mmio_dev) {
1187                 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1188                 return X86EMUL_CONTINUE;
1189         }
1190
1191         vcpu->mmio_needed = 1;
1192         vcpu->mmio_phys_addr = gpa;
1193         vcpu->mmio_size = bytes;
1194         vcpu->mmio_is_write = 1;
1195         memcpy(vcpu->mmio_data, val, bytes);
1196
1197         return X86EMUL_CONTINUE;
1198 }
1199
1200 int emulator_write_emulated(unsigned long addr,
1201                                    const void *val,
1202                                    unsigned int bytes,
1203                                    struct kvm_vcpu *vcpu)
1204 {
1205         /* Crossing a page boundary? */
1206         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1207                 int rc, now;
1208
1209                 now = -addr & ~PAGE_MASK;
1210                 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1211                 if (rc != X86EMUL_CONTINUE)
1212                         return rc;
1213                 addr += now;
1214                 val += now;
1215                 bytes -= now;
1216         }
1217         return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1218 }
1219 EXPORT_SYMBOL_GPL(emulator_write_emulated);
1220
1221 static int emulator_cmpxchg_emulated(unsigned long addr,
1222                                      const void *old,
1223                                      const void *new,
1224                                      unsigned int bytes,
1225                                      struct kvm_vcpu *vcpu)
1226 {
1227         static int reported;
1228
1229         if (!reported) {
1230                 reported = 1;
1231                 printk(KERN_WARNING "kvm: emulating exchange as write\n");
1232         }
1233         return emulator_write_emulated(addr, new, bytes, vcpu);
1234 }
1235
1236 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1237 {
1238         return kvm_x86_ops->get_segment_base(vcpu, seg);
1239 }
1240
1241 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1242 {
1243         return X86EMUL_CONTINUE;
1244 }
1245
1246 int emulate_clts(struct kvm_vcpu *vcpu)
1247 {
1248         kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
1249         return X86EMUL_CONTINUE;
1250 }
1251
1252 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1253 {
1254         struct kvm_vcpu *vcpu = ctxt->vcpu;
1255
1256         switch (dr) {
1257         case 0 ... 3:
1258                 *dest = kvm_x86_ops->get_dr(vcpu, dr);
1259                 return X86EMUL_CONTINUE;
1260         default:
1261                 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1262                 return X86EMUL_UNHANDLEABLE;
1263         }
1264 }
1265
1266 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1267 {
1268         unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1269         int exception;
1270
1271         kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1272         if (exception) {
1273                 /* FIXME: better handling */
1274                 return X86EMUL_UNHANDLEABLE;
1275         }
1276         return X86EMUL_CONTINUE;
1277 }
1278
1279 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1280 {
1281         static int reported;
1282         u8 opcodes[4];
1283         unsigned long rip = vcpu->rip;
1284         unsigned long rip_linear;
1285
1286         rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1287
1288         if (reported)
1289                 return;
1290
1291         emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1292
1293         printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1294                context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1295         reported = 1;
1296 }
1297 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1298
1299 struct x86_emulate_ops emulate_ops = {
1300         .read_std            = emulator_read_std,
1301         .write_std           = emulator_write_std,
1302         .read_emulated       = emulator_read_emulated,
1303         .write_emulated      = emulator_write_emulated,
1304         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
1305 };
1306
1307 int emulate_instruction(struct kvm_vcpu *vcpu,
1308                         struct kvm_run *run,
1309                         unsigned long cr2,
1310                         u16 error_code,
1311                         int no_decode)
1312 {
1313         int r;
1314
1315         vcpu->mmio_fault_cr2 = cr2;
1316         kvm_x86_ops->cache_regs(vcpu);
1317
1318         vcpu->mmio_is_write = 0;
1319         vcpu->pio.string = 0;
1320
1321         if (!no_decode) {
1322                 int cs_db, cs_l;
1323                 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1324
1325                 vcpu->emulate_ctxt.vcpu = vcpu;
1326                 vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1327                 vcpu->emulate_ctxt.cr2 = cr2;
1328                 vcpu->emulate_ctxt.mode =
1329                         (vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1330                         ? X86EMUL_MODE_REAL : cs_l
1331                         ? X86EMUL_MODE_PROT64 : cs_db
1332                         ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1333
1334                 if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1335                         vcpu->emulate_ctxt.cs_base = 0;
1336                         vcpu->emulate_ctxt.ds_base = 0;
1337                         vcpu->emulate_ctxt.es_base = 0;
1338                         vcpu->emulate_ctxt.ss_base = 0;
1339                 } else {
1340                         vcpu->emulate_ctxt.cs_base =
1341                                         get_segment_base(vcpu, VCPU_SREG_CS);
1342                         vcpu->emulate_ctxt.ds_base =
1343                                         get_segment_base(vcpu, VCPU_SREG_DS);
1344                         vcpu->emulate_ctxt.es_base =
1345                                         get_segment_base(vcpu, VCPU_SREG_ES);
1346                         vcpu->emulate_ctxt.ss_base =
1347                                         get_segment_base(vcpu, VCPU_SREG_SS);
1348                 }
1349
1350                 vcpu->emulate_ctxt.gs_base =
1351                                         get_segment_base(vcpu, VCPU_SREG_GS);
1352                 vcpu->emulate_ctxt.fs_base =
1353                                         get_segment_base(vcpu, VCPU_SREG_FS);
1354
1355                 r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1356                 if (r)  {
1357                         if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1358                                 return EMULATE_DONE;
1359                         return EMULATE_FAIL;
1360                 }
1361         }
1362
1363         r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1364
1365         if (vcpu->pio.string)
1366                 return EMULATE_DO_MMIO;
1367
1368         if ((r || vcpu->mmio_is_write) && run) {
1369                 run->exit_reason = KVM_EXIT_MMIO;
1370                 run->mmio.phys_addr = vcpu->mmio_phys_addr;
1371                 memcpy(run->mmio.data, vcpu->mmio_data, 8);
1372                 run->mmio.len = vcpu->mmio_size;
1373                 run->mmio.is_write = vcpu->mmio_is_write;
1374         }
1375
1376         if (r) {
1377                 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1378                         return EMULATE_DONE;
1379                 if (!vcpu->mmio_needed) {
1380                         kvm_report_emulation_failure(vcpu, "mmio");
1381                         return EMULATE_FAIL;
1382                 }
1383                 return EMULATE_DO_MMIO;
1384         }
1385
1386         kvm_x86_ops->decache_regs(vcpu);
1387         kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1388
1389         if (vcpu->mmio_is_write) {
1390                 vcpu->mmio_needed = 0;
1391                 return EMULATE_DO_MMIO;
1392         }
1393
1394         return EMULATE_DONE;
1395 }
1396 EXPORT_SYMBOL_GPL(emulate_instruction);
1397
1398 /*
1399  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1400  */
1401 static void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1402 {
1403         DECLARE_WAITQUEUE(wait, current);
1404
1405         add_wait_queue(&vcpu->wq, &wait);
1406
1407         /*
1408          * We will block until either an interrupt or a signal wakes us up
1409          */
1410         while (!kvm_cpu_has_interrupt(vcpu)
1411                && !signal_pending(current)
1412                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
1413                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
1414                 set_current_state(TASK_INTERRUPTIBLE);
1415                 vcpu_put(vcpu);
1416                 schedule();
1417                 vcpu_load(vcpu);
1418         }
1419
1420         __set_current_state(TASK_RUNNING);
1421         remove_wait_queue(&vcpu->wq, &wait);
1422 }
1423
1424 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1425 {
1426         ++vcpu->stat.halt_exits;
1427         if (irqchip_in_kernel(vcpu->kvm)) {
1428                 vcpu->mp_state = VCPU_MP_STATE_HALTED;
1429                 kvm_vcpu_block(vcpu);
1430                 if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1431                         return -EINTR;
1432                 return 1;
1433         } else {
1434                 vcpu->run->exit_reason = KVM_EXIT_HLT;
1435                 return 0;
1436         }
1437 }
1438 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1439
1440 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1441 {
1442         unsigned long nr, a0, a1, a2, a3, ret;
1443
1444         kvm_x86_ops->cache_regs(vcpu);
1445
1446         nr = vcpu->regs[VCPU_REGS_RAX];
1447         a0 = vcpu->regs[VCPU_REGS_RBX];
1448         a1 = vcpu->regs[VCPU_REGS_RCX];
1449         a2 = vcpu->regs[VCPU_REGS_RDX];
1450         a3 = vcpu->regs[VCPU_REGS_RSI];
1451
1452         if (!is_long_mode(vcpu)) {
1453                 nr &= 0xFFFFFFFF;
1454                 a0 &= 0xFFFFFFFF;
1455                 a1 &= 0xFFFFFFFF;
1456                 a2 &= 0xFFFFFFFF;
1457                 a3 &= 0xFFFFFFFF;
1458         }
1459
1460         switch (nr) {
1461         default:
1462                 ret = -KVM_ENOSYS;
1463                 break;
1464         }
1465         vcpu->regs[VCPU_REGS_RAX] = ret;
1466         kvm_x86_ops->decache_regs(vcpu);
1467         return 0;
1468 }
1469 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1470
1471 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1472 {
1473         char instruction[3];
1474         int ret = 0;
1475
1476         mutex_lock(&vcpu->kvm->lock);
1477
1478         /*
1479          * Blow out the MMU to ensure that no other VCPU has an active mapping
1480          * to ensure that the updated hypercall appears atomically across all
1481          * VCPUs.
1482          */
1483         kvm_mmu_zap_all(vcpu->kvm);
1484
1485         kvm_x86_ops->cache_regs(vcpu);
1486         kvm_x86_ops->patch_hypercall(vcpu, instruction);
1487         if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1488             != X86EMUL_CONTINUE)
1489                 ret = -EFAULT;
1490
1491         mutex_unlock(&vcpu->kvm->lock);
1492
1493         return ret;
1494 }
1495
1496 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1497 {
1498         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1499 }
1500
1501 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1502 {
1503         struct descriptor_table dt = { limit, base };
1504
1505         kvm_x86_ops->set_gdt(vcpu, &dt);
1506 }
1507
1508 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1509 {
1510         struct descriptor_table dt = { limit, base };
1511
1512         kvm_x86_ops->set_idt(vcpu, &dt);
1513 }
1514
1515 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1516                    unsigned long *rflags)
1517 {
1518         lmsw(vcpu, msw);
1519         *rflags = kvm_x86_ops->get_rflags(vcpu);
1520 }
1521
1522 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1523 {
1524         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1525         switch (cr) {
1526         case 0:
1527                 return vcpu->cr0;
1528         case 2:
1529                 return vcpu->cr2;
1530         case 3:
1531                 return vcpu->cr3;
1532         case 4:
1533                 return vcpu->cr4;
1534         default:
1535                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1536                 return 0;
1537         }
1538 }
1539
1540 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1541                      unsigned long *rflags)
1542 {
1543         switch (cr) {
1544         case 0:
1545                 set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1546                 *rflags = kvm_x86_ops->get_rflags(vcpu);
1547                 break;
1548         case 2:
1549                 vcpu->cr2 = val;
1550                 break;
1551         case 3:
1552                 set_cr3(vcpu, val);
1553                 break;
1554         case 4:
1555                 set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1556                 break;
1557         default:
1558                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1559         }
1560 }
1561
1562 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1563 {
1564         u64 data;
1565
1566         switch (msr) {
1567         case 0xc0010010: /* SYSCFG */
1568         case 0xc0010015: /* HWCR */
1569         case MSR_IA32_PLATFORM_ID:
1570         case MSR_IA32_P5_MC_ADDR:
1571         case MSR_IA32_P5_MC_TYPE:
1572         case MSR_IA32_MC0_CTL:
1573         case MSR_IA32_MCG_STATUS:
1574         case MSR_IA32_MCG_CAP:
1575         case MSR_IA32_MC0_MISC:
1576         case MSR_IA32_MC0_MISC+4:
1577         case MSR_IA32_MC0_MISC+8:
1578         case MSR_IA32_MC0_MISC+12:
1579         case MSR_IA32_MC0_MISC+16:
1580         case MSR_IA32_UCODE_REV:
1581         case MSR_IA32_PERF_STATUS:
1582         case MSR_IA32_EBL_CR_POWERON:
1583                 /* MTRR registers */
1584         case 0xfe:
1585         case 0x200 ... 0x2ff:
1586                 data = 0;
1587                 break;
1588         case 0xcd: /* fsb frequency */
1589                 data = 3;
1590                 break;
1591         case MSR_IA32_APICBASE:
1592                 data = kvm_get_apic_base(vcpu);
1593                 break;
1594         case MSR_IA32_MISC_ENABLE:
1595                 data = vcpu->ia32_misc_enable_msr;
1596                 break;
1597 #ifdef CONFIG_X86_64
1598         case MSR_EFER:
1599                 data = vcpu->shadow_efer;
1600                 break;
1601 #endif
1602         default:
1603                 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1604                 return 1;
1605         }
1606         *pdata = data;
1607         return 0;
1608 }
1609 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1610
1611 /*
1612  * Reads an msr value (of 'msr_index') into 'pdata'.
1613  * Returns 0 on success, non-0 otherwise.
1614  * Assumes vcpu_load() was already called.
1615  */
1616 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1617 {
1618         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1619 }
1620
1621 #ifdef CONFIG_X86_64
1622
1623 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
1624 {
1625         if (efer & EFER_RESERVED_BITS) {
1626                 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
1627                        efer);
1628                 inject_gp(vcpu);
1629                 return;
1630         }
1631
1632         if (is_paging(vcpu)
1633             && (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
1634                 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
1635                 inject_gp(vcpu);
1636                 return;
1637         }
1638
1639         kvm_x86_ops->set_efer(vcpu, efer);
1640
1641         efer &= ~EFER_LMA;
1642         efer |= vcpu->shadow_efer & EFER_LMA;
1643
1644         vcpu->shadow_efer = efer;
1645 }
1646
1647 #endif
1648
1649 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1650 {
1651         switch (msr) {
1652 #ifdef CONFIG_X86_64
1653         case MSR_EFER:
1654                 set_efer(vcpu, data);
1655                 break;
1656 #endif
1657         case MSR_IA32_MC0_STATUS:
1658                 pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
1659                        __FUNCTION__, data);
1660                 break;
1661         case MSR_IA32_MCG_STATUS:
1662                 pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
1663                         __FUNCTION__, data);
1664                 break;
1665         case MSR_IA32_UCODE_REV:
1666         case MSR_IA32_UCODE_WRITE:
1667         case 0x200 ... 0x2ff: /* MTRRs */
1668                 break;
1669         case MSR_IA32_APICBASE:
1670                 kvm_set_apic_base(vcpu, data);
1671                 break;
1672         case MSR_IA32_MISC_ENABLE:
1673                 vcpu->ia32_misc_enable_msr = data;
1674                 break;
1675         default:
1676                 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
1677                 return 1;
1678         }
1679         return 0;
1680 }
1681 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1682
1683 /*
1684  * Writes msr value into into the appropriate "register".
1685  * Returns 0 on success, non-0 otherwise.
1686  * Assumes vcpu_load() was already called.
1687  */
1688 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1689 {
1690         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
1691 }
1692
1693 void kvm_resched(struct kvm_vcpu *vcpu)
1694 {
1695         if (!need_resched())
1696                 return;
1697         cond_resched();
1698 }
1699 EXPORT_SYMBOL_GPL(kvm_resched);
1700
1701 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1702 {
1703         int i;
1704         u32 function;
1705         struct kvm_cpuid_entry *e, *best;
1706
1707         kvm_x86_ops->cache_regs(vcpu);
1708         function = vcpu->regs[VCPU_REGS_RAX];
1709         vcpu->regs[VCPU_REGS_RAX] = 0;
1710         vcpu->regs[VCPU_REGS_RBX] = 0;
1711         vcpu->regs[VCPU_REGS_RCX] = 0;
1712         vcpu->regs[VCPU_REGS_RDX] = 0;
1713         best = NULL;
1714         for (i = 0; i < vcpu->cpuid_nent; ++i) {
1715                 e = &vcpu->cpuid_entries[i];
1716                 if (e->function == function) {
1717                         best = e;
1718                         break;
1719                 }
1720                 /*
1721                  * Both basic or both extended?
1722                  */
1723                 if (((e->function ^ function) & 0x80000000) == 0)
1724                         if (!best || e->function > best->function)
1725                                 best = e;
1726         }
1727         if (best) {
1728                 vcpu->regs[VCPU_REGS_RAX] = best->eax;
1729                 vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1730                 vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1731                 vcpu->regs[VCPU_REGS_RDX] = best->edx;
1732         }
1733         kvm_x86_ops->decache_regs(vcpu);
1734         kvm_x86_ops->skip_emulated_instruction(vcpu);
1735 }
1736 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1737
1738 static int pio_copy_data(struct kvm_vcpu *vcpu)
1739 {
1740         void *p = vcpu->pio_data;
1741         void *q;
1742         unsigned bytes;
1743         int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1744
1745         q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1746                  PAGE_KERNEL);
1747         if (!q) {
1748                 free_pio_guest_pages(vcpu);
1749                 return -ENOMEM;
1750         }
1751         q += vcpu->pio.guest_page_offset;
1752         bytes = vcpu->pio.size * vcpu->pio.cur_count;
1753         if (vcpu->pio.in)
1754                 memcpy(q, p, bytes);
1755         else
1756                 memcpy(p, q, bytes);
1757         q -= vcpu->pio.guest_page_offset;
1758         vunmap(q);
1759         free_pio_guest_pages(vcpu);
1760         return 0;
1761 }
1762
1763 static int complete_pio(struct kvm_vcpu *vcpu)
1764 {
1765         struct kvm_pio_request *io = &vcpu->pio;
1766         long delta;
1767         int r;
1768
1769         kvm_x86_ops->cache_regs(vcpu);
1770
1771         if (!io->string) {
1772                 if (io->in)
1773                         memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1774                                io->size);
1775         } else {
1776                 if (io->in) {
1777                         r = pio_copy_data(vcpu);
1778                         if (r) {
1779                                 kvm_x86_ops->cache_regs(vcpu);
1780                                 return r;
1781                         }
1782                 }
1783
1784                 delta = 1;
1785                 if (io->rep) {
1786                         delta *= io->cur_count;
1787                         /*
1788                          * The size of the register should really depend on
1789                          * current address size.
1790                          */
1791                         vcpu->regs[VCPU_REGS_RCX] -= delta;
1792                 }
1793                 if (io->down)
1794                         delta = -delta;
1795                 delta *= io->size;
1796                 if (io->in)
1797                         vcpu->regs[VCPU_REGS_RDI] += delta;
1798                 else
1799                         vcpu->regs[VCPU_REGS_RSI] += delta;
1800         }
1801
1802         kvm_x86_ops->decache_regs(vcpu);
1803
1804         io->count -= io->cur_count;
1805         io->cur_count = 0;
1806
1807         return 0;
1808 }
1809
1810 static void kernel_pio(struct kvm_io_device *pio_dev,
1811                        struct kvm_vcpu *vcpu,
1812                        void *pd)
1813 {
1814         /* TODO: String I/O for in kernel device */
1815
1816         mutex_lock(&vcpu->kvm->lock);
1817         if (vcpu->pio.in)
1818                 kvm_iodevice_read(pio_dev, vcpu->pio.port,
1819                                   vcpu->pio.size,
1820                                   pd);
1821         else
1822                 kvm_iodevice_write(pio_dev, vcpu->pio.port,
1823                                    vcpu->pio.size,
1824                                    pd);
1825         mutex_unlock(&vcpu->kvm->lock);
1826 }
1827
1828 static void pio_string_write(struct kvm_io_device *pio_dev,
1829                              struct kvm_vcpu *vcpu)
1830 {
1831         struct kvm_pio_request *io = &vcpu->pio;
1832         void *pd = vcpu->pio_data;
1833         int i;
1834
1835         mutex_lock(&vcpu->kvm->lock);
1836         for (i = 0; i < io->cur_count; i++) {
1837                 kvm_iodevice_write(pio_dev, io->port,
1838                                    io->size,
1839                                    pd);
1840                 pd += io->size;
1841         }
1842         mutex_unlock(&vcpu->kvm->lock);
1843 }
1844
1845 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1846                   int size, unsigned port)
1847 {
1848         struct kvm_io_device *pio_dev;
1849
1850         vcpu->run->exit_reason = KVM_EXIT_IO;
1851         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1852         vcpu->run->io.size = vcpu->pio.size = size;
1853         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1854         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1855         vcpu->run->io.port = vcpu->pio.port = port;
1856         vcpu->pio.in = in;
1857         vcpu->pio.string = 0;
1858         vcpu->pio.down = 0;
1859         vcpu->pio.guest_page_offset = 0;
1860         vcpu->pio.rep = 0;
1861
1862         kvm_x86_ops->cache_regs(vcpu);
1863         memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1864         kvm_x86_ops->decache_regs(vcpu);
1865
1866         kvm_x86_ops->skip_emulated_instruction(vcpu);
1867
1868         pio_dev = vcpu_find_pio_dev(vcpu, port);
1869         if (pio_dev) {
1870                 kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1871                 complete_pio(vcpu);
1872                 return 1;
1873         }
1874         return 0;
1875 }
1876 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1877
1878 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1879                   int size, unsigned long count, int down,
1880                   gva_t address, int rep, unsigned port)
1881 {
1882         unsigned now, in_page;
1883         int i, ret = 0;
1884         int nr_pages = 1;
1885         struct page *page;
1886         struct kvm_io_device *pio_dev;
1887
1888         vcpu->run->exit_reason = KVM_EXIT_IO;
1889         vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1890         vcpu->run->io.size = vcpu->pio.size = size;
1891         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1892         vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1893         vcpu->run->io.port = vcpu->pio.port = port;
1894         vcpu->pio.in = in;
1895         vcpu->pio.string = 1;
1896         vcpu->pio.down = down;
1897         vcpu->pio.guest_page_offset = offset_in_page(address);
1898         vcpu->pio.rep = rep;
1899
1900         if (!count) {
1901                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1902                 return 1;
1903         }
1904
1905         if (!down)
1906                 in_page = PAGE_SIZE - offset_in_page(address);
1907         else
1908                 in_page = offset_in_page(address) + size;
1909         now = min(count, (unsigned long)in_page / size);
1910         if (!now) {
1911                 /*
1912                  * String I/O straddles page boundary.  Pin two guest pages
1913                  * so that we satisfy atomicity constraints.  Do just one
1914                  * transaction to avoid complexity.
1915                  */
1916                 nr_pages = 2;
1917                 now = 1;
1918         }
1919         if (down) {
1920                 /*
1921                  * String I/O in reverse.  Yuck.  Kill the guest, fix later.
1922                  */
1923                 pr_unimpl(vcpu, "guest string pio down\n");
1924                 inject_gp(vcpu);
1925                 return 1;
1926         }
1927         vcpu->run->io.count = now;
1928         vcpu->pio.cur_count = now;
1929
1930         if (vcpu->pio.cur_count == vcpu->pio.count)
1931                 kvm_x86_ops->skip_emulated_instruction(vcpu);
1932
1933         for (i = 0; i < nr_pages; ++i) {
1934                 mutex_lock(&vcpu->kvm->lock);
1935                 page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1936                 vcpu->pio.guest_pages[i] = page;
1937                 mutex_unlock(&vcpu->kvm->lock);
1938                 if (!page) {
1939                         inject_gp(vcpu);
1940                         free_pio_guest_pages(vcpu);
1941                         return 1;
1942                 }
1943         }
1944
1945         pio_dev = vcpu_find_pio_dev(vcpu, port);
1946         if (!vcpu->pio.in) {
1947                 /* string PIO write */
1948                 ret = pio_copy_data(vcpu);
1949                 if (ret >= 0 && pio_dev) {
1950                         pio_string_write(pio_dev, vcpu);
1951                         complete_pio(vcpu);
1952                         if (vcpu->pio.count == 0)
1953                                 ret = 1;
1954                 }
1955         } else if (pio_dev)
1956                 pr_unimpl(vcpu, "no string pio read support yet, "
1957                        "port %x size %d count %ld\n",
1958                         port, size, count);
1959
1960         return ret;
1961 }
1962 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1963
1964 /*
1965  * Check if userspace requested an interrupt window, and that the
1966  * interrupt window is open.
1967  *
1968  * No need to exit to userspace if we already have an interrupt queued.
1969  */
1970 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1971                                           struct kvm_run *kvm_run)
1972 {
1973         return (!vcpu->irq_summary &&
1974                 kvm_run->request_interrupt_window &&
1975                 vcpu->interrupt_window_open &&
1976                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1977 }
1978
1979 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1980                               struct kvm_run *kvm_run)
1981 {
1982         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1983         kvm_run->cr8 = get_cr8(vcpu);
1984         kvm_run->apic_base = kvm_get_apic_base(vcpu);
1985         if (irqchip_in_kernel(vcpu->kvm))
1986                 kvm_run->ready_for_interrupt_injection = 1;
1987         else
1988                 kvm_run->ready_for_interrupt_injection =
1989                                         (vcpu->interrupt_window_open &&
1990                                          vcpu->irq_summary == 0);
1991 }
1992
1993 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1994 {
1995         int r;
1996
1997         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1998                 pr_debug("vcpu %d received sipi with vector # %x\n",
1999                        vcpu->vcpu_id, vcpu->sipi_vector);
2000                 kvm_lapic_reset(vcpu);
2001                 r = kvm_x86_ops->vcpu_reset(vcpu);
2002                 if (r)
2003                         return r;
2004                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
2005         }
2006
2007 preempted:
2008         if (vcpu->guest_debug.enabled)
2009                 kvm_x86_ops->guest_debug_pre(vcpu);
2010
2011 again:
2012         r = kvm_mmu_reload(vcpu);
2013         if (unlikely(r))
2014                 goto out;
2015
2016         kvm_inject_pending_timer_irqs(vcpu);
2017
2018         preempt_disable();
2019
2020         kvm_x86_ops->prepare_guest_switch(vcpu);
2021         kvm_load_guest_fpu(vcpu);
2022
2023         local_irq_disable();
2024
2025         if (signal_pending(current)) {
2026                 local_irq_enable();
2027                 preempt_enable();
2028                 r = -EINTR;
2029                 kvm_run->exit_reason = KVM_EXIT_INTR;
2030                 ++vcpu->stat.signal_exits;
2031                 goto out;
2032         }
2033
2034         if (irqchip_in_kernel(vcpu->kvm))
2035                 kvm_x86_ops->inject_pending_irq(vcpu);
2036         else if (!vcpu->mmio_read_completed)
2037                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
2038
2039         vcpu->guest_mode = 1;
2040         kvm_guest_enter();
2041
2042         if (vcpu->requests)
2043                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
2044                         kvm_x86_ops->tlb_flush(vcpu);
2045
2046         kvm_x86_ops->run(vcpu, kvm_run);
2047
2048         vcpu->guest_mode = 0;
2049         local_irq_enable();
2050
2051         ++vcpu->stat.exits;
2052
2053         /*
2054          * We must have an instruction between local_irq_enable() and
2055          * kvm_guest_exit(), so the timer interrupt isn't delayed by
2056          * the interrupt shadow.  The stat.exits increment will do nicely.
2057          * But we need to prevent reordering, hence this barrier():
2058          */
2059         barrier();
2060
2061         kvm_guest_exit();
2062
2063         preempt_enable();
2064
2065         /*
2066          * Profile KVM exit RIPs:
2067          */
2068         if (unlikely(prof_on == KVM_PROFILING)) {
2069                 kvm_x86_ops->cache_regs(vcpu);
2070                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
2071         }
2072
2073         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
2074
2075         if (r > 0) {
2076                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2077                         r = -EINTR;
2078                         kvm_run->exit_reason = KVM_EXIT_INTR;
2079                         ++vcpu->stat.request_irq_exits;
2080                         goto out;
2081                 }
2082                 if (!need_resched()) {
2083                         ++vcpu->stat.light_exits;
2084                         goto again;
2085                 }
2086         }
2087
2088 out:
2089         if (r > 0) {
2090                 kvm_resched(vcpu);
2091                 goto preempted;
2092         }
2093
2094         post_kvm_run_save(vcpu, kvm_run);
2095
2096         return r;
2097 }
2098
2099
2100 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2101 {
2102         int r;
2103         sigset_t sigsaved;
2104
2105         vcpu_load(vcpu);
2106
2107         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
2108                 kvm_vcpu_block(vcpu);
2109                 vcpu_put(vcpu);
2110                 return -EAGAIN;
2111         }
2112
2113         if (vcpu->sigset_active)
2114                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
2115
2116         /* re-sync apic's tpr */
2117         if (!irqchip_in_kernel(vcpu->kvm))
2118                 set_cr8(vcpu, kvm_run->cr8);
2119
2120         if (vcpu->pio.cur_count) {
2121                 r = complete_pio(vcpu);
2122                 if (r)
2123                         goto out;
2124         }
2125 #if CONFIG_HAS_IOMEM
2126         if (vcpu->mmio_needed) {
2127                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
2128                 vcpu->mmio_read_completed = 1;
2129                 vcpu->mmio_needed = 0;
2130                 r = emulate_instruction(vcpu, kvm_run,
2131                                         vcpu->mmio_fault_cr2, 0, 1);
2132                 if (r == EMULATE_DO_MMIO) {
2133                         /*
2134                          * Read-modify-write.  Back to userspace.
2135                          */
2136                         r = 0;
2137                         goto out;
2138                 }
2139         }
2140 #endif
2141         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
2142                 kvm_x86_ops->cache_regs(vcpu);
2143                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
2144                 kvm_x86_ops->decache_regs(vcpu);
2145         }
2146
2147         r = __vcpu_run(vcpu, kvm_run);
2148
2149 out:
2150         if (vcpu->sigset_active)
2151                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
2152
2153         vcpu_put(vcpu);
2154         return r;
2155 }
2156
2157 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
2158                                    struct kvm_regs *regs)
2159 {
2160         vcpu_load(vcpu);
2161
2162         kvm_x86_ops->cache_regs(vcpu);
2163
2164         regs->rax = vcpu->regs[VCPU_REGS_RAX];
2165         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
2166         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
2167         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
2168         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
2169         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
2170         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
2171         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
2172 #ifdef CONFIG_X86_64
2173         regs->r8 = vcpu->regs[VCPU_REGS_R8];
2174         regs->r9 = vcpu->regs[VCPU_REGS_R9];
2175         regs->r10 = vcpu->regs[VCPU_REGS_R10];
2176         regs->r11 = vcpu->regs[VCPU_REGS_R11];
2177         regs->r12 = vcpu->regs[VCPU_REGS_R12];
2178         regs->r13 = vcpu->regs[VCPU_REGS_R13];
2179         regs->r14 = vcpu->regs[VCPU_REGS_R14];
2180         regs->r15 = vcpu->regs[VCPU_REGS_R15];
2181 #endif
2182
2183         regs->rip = vcpu->rip;
2184         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2185
2186         /*
2187          * Don't leak debug flags in case they were set for guest debugging
2188          */
2189         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2190                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2191
2192         vcpu_put(vcpu);
2193
2194         return 0;
2195 }
2196
2197 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
2198                                    struct kvm_regs *regs)
2199 {
2200         vcpu_load(vcpu);
2201
2202         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2203         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2204         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2205         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2206         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2207         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2208         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2209         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2210 #ifdef CONFIG_X86_64
2211         vcpu->regs[VCPU_REGS_R8] = regs->r8;
2212         vcpu->regs[VCPU_REGS_R9] = regs->r9;
2213         vcpu->regs[VCPU_REGS_R10] = regs->r10;
2214         vcpu->regs[VCPU_REGS_R11] = regs->r11;
2215         vcpu->regs[VCPU_REGS_R12] = regs->r12;
2216         vcpu->regs[VCPU_REGS_R13] = regs->r13;
2217         vcpu->regs[VCPU_REGS_R14] = regs->r14;
2218         vcpu->regs[VCPU_REGS_R15] = regs->r15;
2219 #endif
2220
2221         vcpu->rip = regs->rip;
2222         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2223
2224         kvm_x86_ops->decache_regs(vcpu);
2225
2226         vcpu_put(vcpu);
2227
2228         return 0;
2229 }
2230
2231 static void get_segment(struct kvm_vcpu *vcpu,
2232                         struct kvm_segment *var, int seg)
2233 {
2234         return kvm_x86_ops->get_segment(vcpu, var, seg);
2235 }
2236
2237 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2238                                     struct kvm_sregs *sregs)
2239 {
2240         struct descriptor_table dt;
2241         int pending_vec;
2242
2243         vcpu_load(vcpu);
2244
2245         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2246         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2247         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2248         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2249         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2250         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2251
2252         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2253         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2254
2255         kvm_x86_ops->get_idt(vcpu, &dt);
2256         sregs->idt.limit = dt.limit;
2257         sregs->idt.base = dt.base;
2258         kvm_x86_ops->get_gdt(vcpu, &dt);
2259         sregs->gdt.limit = dt.limit;
2260         sregs->gdt.base = dt.base;
2261
2262         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2263         sregs->cr0 = vcpu->cr0;
2264         sregs->cr2 = vcpu->cr2;
2265         sregs->cr3 = vcpu->cr3;
2266         sregs->cr4 = vcpu->cr4;
2267         sregs->cr8 = get_cr8(vcpu);
2268         sregs->efer = vcpu->shadow_efer;
2269         sregs->apic_base = kvm_get_apic_base(vcpu);
2270
2271         if (irqchip_in_kernel(vcpu->kvm)) {
2272                 memset(sregs->interrupt_bitmap, 0,
2273                        sizeof sregs->interrupt_bitmap);
2274                 pending_vec = kvm_x86_ops->get_irq(vcpu);
2275                 if (pending_vec >= 0)
2276                         set_bit(pending_vec,
2277                                 (unsigned long *)sregs->interrupt_bitmap);
2278         } else
2279                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2280                        sizeof sregs->interrupt_bitmap);
2281
2282         vcpu_put(vcpu);
2283
2284         return 0;
2285 }
2286
2287 static void set_segment(struct kvm_vcpu *vcpu,
2288                         struct kvm_segment *var, int seg)
2289 {
2290         return kvm_x86_ops->set_segment(vcpu, var, seg);
2291 }
2292
2293 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2294                                     struct kvm_sregs *sregs)
2295 {
2296         int mmu_reset_needed = 0;
2297         int i, pending_vec, max_bits;
2298         struct descriptor_table dt;
2299
2300         vcpu_load(vcpu);
2301
2302         dt.limit = sregs->idt.limit;
2303         dt.base = sregs->idt.base;
2304         kvm_x86_ops->set_idt(vcpu, &dt);
2305         dt.limit = sregs->gdt.limit;
2306         dt.base = sregs->gdt.base;
2307         kvm_x86_ops->set_gdt(vcpu, &dt);
2308
2309         vcpu->cr2 = sregs->cr2;
2310         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2311         vcpu->cr3 = sregs->cr3;
2312
2313         set_cr8(vcpu, sregs->cr8);
2314
2315         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2316 #ifdef CONFIG_X86_64
2317         kvm_x86_ops->set_efer(vcpu, sregs->efer);
2318 #endif
2319         kvm_set_apic_base(vcpu, sregs->apic_base);
2320
2321         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2322
2323         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2324         vcpu->cr0 = sregs->cr0;
2325         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2326
2327         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2328         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2329         if (!is_long_mode(vcpu) && is_pae(vcpu))
2330                 load_pdptrs(vcpu, vcpu->cr3);
2331
2332         if (mmu_reset_needed)
2333                 kvm_mmu_reset_context(vcpu);
2334
2335         if (!irqchip_in_kernel(vcpu->kvm)) {
2336                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2337                        sizeof vcpu->irq_pending);
2338                 vcpu->irq_summary = 0;
2339                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2340                         if (vcpu->irq_pending[i])
2341                                 __set_bit(i, &vcpu->irq_summary);
2342         } else {
2343                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2344                 pending_vec = find_first_bit(
2345                         (const unsigned long *)sregs->interrupt_bitmap,
2346                         max_bits);
2347                 /* Only pending external irq is handled here */
2348                 if (pending_vec < max_bits) {
2349                         kvm_x86_ops->set_irq(vcpu, pending_vec);
2350                         pr_debug("Set back pending irq %d\n",
2351                                  pending_vec);
2352                 }
2353         }
2354
2355         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2356         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2357         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2358         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2359         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2360         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2361
2362         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2363         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2364
2365         vcpu_put(vcpu);
2366
2367         return 0;
2368 }
2369
2370 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2371 {
2372         struct kvm_segment cs;
2373
2374         get_segment(vcpu, &cs, VCPU_SREG_CS);
2375         *db = cs.db;
2376         *l = cs.l;
2377 }
2378 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2379
2380 /*
2381  * Translate a guest virtual address to a guest physical address.
2382  */
2383 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
2384                                     struct kvm_translation *tr)
2385 {
2386         unsigned long vaddr = tr->linear_address;
2387         gpa_t gpa;
2388
2389         vcpu_load(vcpu);
2390         mutex_lock(&vcpu->kvm->lock);
2391         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
2392         tr->physical_address = gpa;
2393         tr->valid = gpa != UNMAPPED_GVA;
2394         tr->writeable = 1;
2395         tr->usermode = 0;
2396         mutex_unlock(&vcpu->kvm->lock);
2397         vcpu_put(vcpu);
2398
2399         return 0;
2400 }
2401
2402 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2403                                     struct kvm_interrupt *irq)
2404 {
2405         if (irq->irq < 0 || irq->irq >= 256)
2406                 return -EINVAL;
2407         if (irqchip_in_kernel(vcpu->kvm))
2408                 return -ENXIO;
2409         vcpu_load(vcpu);
2410
2411         set_bit(irq->irq, vcpu->irq_pending);
2412         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
2413
2414         vcpu_put(vcpu);
2415
2416         return 0;
2417 }
2418
2419 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2420                                       struct kvm_debug_guest *dbg)
2421 {
2422         int r;
2423
2424         vcpu_load(vcpu);
2425
2426         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2427
2428         vcpu_put(vcpu);
2429
2430         return r;
2431 }
2432
2433 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
2434                                     unsigned long address,
2435                                     int *type)
2436 {
2437         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2438         unsigned long pgoff;
2439         struct page *page;
2440
2441         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2442         if (pgoff == 0)
2443                 page = virt_to_page(vcpu->run);
2444         else if (pgoff == KVM_PIO_PAGE_OFFSET)
2445                 page = virt_to_page(vcpu->pio_data);
2446         else
2447                 return NOPAGE_SIGBUS;
2448         get_page(page);
2449         if (type != NULL)
2450                 *type = VM_FAULT_MINOR;
2451
2452         return page;
2453 }
2454
2455 static struct vm_operations_struct kvm_vcpu_vm_ops = {
2456         .nopage = kvm_vcpu_nopage,
2457 };
2458
2459 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2460 {
2461         vma->vm_ops = &kvm_vcpu_vm_ops;
2462         return 0;
2463 }
2464
2465 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2466 {
2467         struct kvm_vcpu *vcpu = filp->private_data;
2468
2469         fput(vcpu->kvm->filp);
2470         return 0;
2471 }
2472
2473 static struct file_operations kvm_vcpu_fops = {
2474         .release        = kvm_vcpu_release,
2475         .unlocked_ioctl = kvm_vcpu_ioctl,
2476         .compat_ioctl   = kvm_vcpu_ioctl,
2477         .mmap           = kvm_vcpu_mmap,
2478 };
2479
2480 /*
2481  * Allocates an inode for the vcpu.
2482  */
2483 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2484 {
2485         int fd, r;
2486         struct inode *inode;
2487         struct file *file;
2488
2489         r = anon_inode_getfd(&fd, &inode, &file,
2490                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
2491         if (r)
2492                 return r;
2493         atomic_inc(&vcpu->kvm->filp->f_count);
2494         return fd;
2495 }
2496
2497 /*
2498  * Creates some virtual cpus.  Good luck creating more than one.
2499  */
2500 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
2501 {
2502         int r;
2503         struct kvm_vcpu *vcpu;
2504
2505         if (!valid_vcpu(n))
2506                 return -EINVAL;
2507
2508         vcpu = kvm_x86_ops->vcpu_create(kvm, n);
2509         if (IS_ERR(vcpu))
2510                 return PTR_ERR(vcpu);
2511
2512         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2513
2514         /* We do fxsave: this must be aligned. */
2515         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
2516
2517         vcpu_load(vcpu);
2518         r = kvm_x86_ops->vcpu_reset(vcpu);
2519         if (r == 0)
2520                 r = kvm_mmu_setup(vcpu);
2521         vcpu_put(vcpu);
2522         if (r < 0)
2523                 goto free_vcpu;
2524
2525         mutex_lock(&kvm->lock);
2526         if (kvm->vcpus[n]) {
2527                 r = -EEXIST;
2528                 mutex_unlock(&kvm->lock);
2529                 goto mmu_unload;
2530         }
2531         kvm->vcpus[n] = vcpu;
2532         mutex_unlock(&kvm->lock);
2533
2534         /* Now it's all set up, let userspace reach it */
2535         r = create_vcpu_fd(vcpu);
2536         if (r < 0)
2537                 goto unlink;
2538         return r;
2539
2540 unlink:
2541         mutex_lock(&kvm->lock);
2542         kvm->vcpus[n] = NULL;
2543         mutex_unlock(&kvm->lock);
2544
2545 mmu_unload:
2546         vcpu_load(vcpu);
2547         kvm_mmu_unload(vcpu);
2548         vcpu_put(vcpu);
2549
2550 free_vcpu:
2551         kvm_x86_ops->vcpu_free(vcpu);
2552         return r;
2553 }
2554
2555 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2556 {
2557         if (sigset) {
2558                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2559                 vcpu->sigset_active = 1;
2560                 vcpu->sigset = *sigset;
2561         } else
2562                 vcpu->sigset_active = 0;
2563         return 0;
2564 }
2565
2566 /*
2567  * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
2568  * we have asm/x86/processor.h
2569  */
2570 struct fxsave {
2571         u16     cwd;
2572         u16     swd;
2573         u16     twd;
2574         u16     fop;
2575         u64     rip;
2576         u64     rdp;
2577         u32     mxcsr;
2578         u32     mxcsr_mask;
2579         u32     st_space[32];   /* 8*16 bytes for each FP-reg = 128 bytes */
2580 #ifdef CONFIG_X86_64
2581         u32     xmm_space[64];  /* 16*16 bytes for each XMM-reg = 256 bytes */
2582 #else
2583         u32     xmm_space[32];  /* 8*16 bytes for each XMM-reg = 128 bytes */
2584 #endif
2585 };
2586
2587 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2588 {
2589         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2590
2591         vcpu_load(vcpu);
2592
2593         memcpy(fpu->fpr, fxsave->st_space, 128);
2594         fpu->fcw = fxsave->cwd;
2595         fpu->fsw = fxsave->swd;
2596         fpu->ftwx = fxsave->twd;
2597         fpu->last_opcode = fxsave->fop;
2598         fpu->last_ip = fxsave->rip;
2599         fpu->last_dp = fxsave->rdp;
2600         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2601
2602         vcpu_put(vcpu);
2603
2604         return 0;
2605 }
2606
2607 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2608 {
2609         struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2610
2611         vcpu_load(vcpu);
2612
2613         memcpy(fxsave->st_space, fpu->fpr, 128);
2614         fxsave->cwd = fpu->fcw;
2615         fxsave->swd = fpu->fsw;
2616         fxsave->twd = fpu->ftwx;
2617         fxsave->fop = fpu->last_opcode;
2618         fxsave->rip = fpu->last_ip;
2619         fxsave->rdp = fpu->last_dp;
2620         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2621
2622         vcpu_put(vcpu);
2623
2624         return 0;
2625 }
2626
2627 static long kvm_vcpu_ioctl(struct file *filp,
2628                            unsigned int ioctl, unsigned long arg)
2629 {
2630         struct kvm_vcpu *vcpu = filp->private_data;
2631         void __user *argp = (void __user *)arg;
2632         int r;
2633
2634         switch (ioctl) {
2635         case KVM_RUN:
2636                 r = -EINVAL;
2637                 if (arg)
2638                         goto out;
2639                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
2640                 break;
2641         case KVM_GET_REGS: {
2642                 struct kvm_regs kvm_regs;
2643
2644                 memset(&kvm_regs, 0, sizeof kvm_regs);
2645                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
2646                 if (r)
2647                         goto out;
2648                 r = -EFAULT;
2649                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
2650                         goto out;
2651                 r = 0;
2652                 break;
2653         }
2654         case KVM_SET_REGS: {
2655                 struct kvm_regs kvm_regs;
2656
2657                 r = -EFAULT;
2658                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
2659                         goto out;
2660                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
2661                 if (r)
2662                         goto out;
2663                 r = 0;
2664                 break;
2665         }
2666         case KVM_GET_SREGS: {
2667                 struct kvm_sregs kvm_sregs;
2668
2669                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
2670                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
2671                 if (r)
2672                         goto out;
2673                 r = -EFAULT;
2674                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
2675                         goto out;
2676                 r = 0;
2677                 break;
2678         }
2679         case KVM_SET_SREGS: {
2680                 struct kvm_sregs kvm_sregs;
2681
2682                 r = -EFAULT;
2683                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
2684                         goto out;
2685                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
2686                 if (r)
2687                         goto out;
2688                 r = 0;
2689                 break;
2690         }
2691         case KVM_TRANSLATE: {
2692                 struct kvm_translation tr;
2693
2694                 r = -EFAULT;
2695                 if (copy_from_user(&tr, argp, sizeof tr))
2696                         goto out;
2697                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
2698                 if (r)
2699                         goto out;
2700                 r = -EFAULT;
2701                 if (copy_to_user(argp, &tr, sizeof tr))
2702                         goto out;
2703                 r = 0;
2704                 break;
2705         }
2706         case KVM_INTERRUPT: {
2707                 struct kvm_interrupt irq;
2708
2709                 r = -EFAULT;
2710                 if (copy_from_user(&irq, argp, sizeof irq))
2711                         goto out;
2712                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2713                 if (r)
2714                         goto out;
2715                 r = 0;
2716                 break;
2717         }
2718         case KVM_DEBUG_GUEST: {
2719                 struct kvm_debug_guest dbg;
2720
2721                 r = -EFAULT;
2722                 if (copy_from_user(&dbg, argp, sizeof dbg))
2723                         goto out;
2724                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
2725                 if (r)
2726                         goto out;
2727                 r = 0;
2728                 break;
2729         }
2730         case KVM_SET_SIGNAL_MASK: {
2731                 struct kvm_signal_mask __user *sigmask_arg = argp;
2732                 struct kvm_signal_mask kvm_sigmask;
2733                 sigset_t sigset, *p;
2734
2735                 p = NULL;
2736                 if (argp) {
2737                         r = -EFAULT;
2738                         if (copy_from_user(&kvm_sigmask, argp,
2739                                            sizeof kvm_sigmask))
2740                                 goto out;
2741                         r = -EINVAL;
2742                         if (kvm_sigmask.len != sizeof sigset)
2743                                 goto out;
2744                         r = -EFAULT;
2745                         if (copy_from_user(&sigset, sigmask_arg->sigset,
2746                                            sizeof sigset))
2747                                 goto out;
2748                         p = &sigset;
2749                 }
2750                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2751                 break;
2752         }
2753         case KVM_GET_FPU: {
2754                 struct kvm_fpu fpu;
2755
2756                 memset(&fpu, 0, sizeof fpu);
2757                 r = kvm_vcpu_ioctl_get_fpu(vcpu, &fpu);
2758                 if (r)
2759                         goto out;
2760                 r = -EFAULT;
2761                 if (copy_to_user(argp, &fpu, sizeof fpu))
2762                         goto out;
2763                 r = 0;
2764                 break;
2765         }
2766         case KVM_SET_FPU: {
2767                 struct kvm_fpu fpu;
2768
2769                 r = -EFAULT;
2770                 if (copy_from_user(&fpu, argp, sizeof fpu))
2771                         goto out;
2772                 r = kvm_vcpu_ioctl_set_fpu(vcpu, &fpu);
2773                 if (r)
2774                         goto out;
2775                 r = 0;
2776                 break;
2777         }
2778         default:
2779                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2780         }
2781 out:
2782         return r;
2783 }
2784
2785 static long kvm_vm_ioctl(struct file *filp,
2786                            unsigned int ioctl, unsigned long arg)
2787 {
2788         struct kvm *kvm = filp->private_data;
2789         void __user *argp = (void __user *)arg;
2790         int r;
2791
2792         switch (ioctl) {
2793         case KVM_CREATE_VCPU:
2794                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2795                 if (r < 0)
2796                         goto out;
2797                 break;
2798         case KVM_SET_USER_MEMORY_REGION: {
2799                 struct kvm_userspace_memory_region kvm_userspace_mem;
2800
2801                 r = -EFAULT;
2802                 if (copy_from_user(&kvm_userspace_mem, argp,
2803                                                 sizeof kvm_userspace_mem))
2804                         goto out;
2805
2806                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2807                 if (r)
2808                         goto out;
2809                 break;
2810         }
2811         case KVM_GET_DIRTY_LOG: {
2812                 struct kvm_dirty_log log;
2813
2814                 r = -EFAULT;
2815                 if (copy_from_user(&log, argp, sizeof log))
2816                         goto out;
2817                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2818                 if (r)
2819                         goto out;
2820                 break;
2821         }
2822         default:
2823                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2824         }
2825 out:
2826         return r;
2827 }
2828
2829 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
2830                                   unsigned long address,
2831                                   int *type)
2832 {
2833         struct kvm *kvm = vma->vm_file->private_data;
2834         unsigned long pgoff;
2835         struct page *page;
2836
2837         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2838         if (!kvm_is_visible_gfn(kvm, pgoff))
2839                 return NOPAGE_SIGBUS;
2840         page = gfn_to_page(kvm, pgoff);
2841         if (is_error_page(page)) {
2842                 kvm_release_page(page);
2843                 return NOPAGE_SIGBUS;
2844         }
2845         if (type != NULL)
2846                 *type = VM_FAULT_MINOR;
2847
2848         return page;
2849 }
2850
2851 static struct vm_operations_struct kvm_vm_vm_ops = {
2852         .nopage = kvm_vm_nopage,
2853 };
2854
2855 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2856 {
2857         vma->vm_ops = &kvm_vm_vm_ops;
2858         return 0;
2859 }
2860
2861 static struct file_operations kvm_vm_fops = {
2862         .release        = kvm_vm_release,
2863         .unlocked_ioctl = kvm_vm_ioctl,
2864         .compat_ioctl   = kvm_vm_ioctl,
2865         .mmap           = kvm_vm_mmap,
2866 };
2867
2868 static int kvm_dev_ioctl_create_vm(void)
2869 {
2870         int fd, r;
2871         struct inode *inode;
2872         struct file *file;
2873         struct kvm *kvm;
2874
2875         kvm = kvm_create_vm();
2876         if (IS_ERR(kvm))
2877                 return PTR_ERR(kvm);
2878         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
2879         if (r) {
2880                 kvm_destroy_vm(kvm);
2881                 return r;
2882         }
2883
2884         kvm->filp = file;
2885
2886         return fd;
2887 }
2888
2889 static long kvm_dev_ioctl(struct file *filp,
2890                           unsigned int ioctl, unsigned long arg)
2891 {
2892         void __user *argp = (void __user *)arg;
2893         long r = -EINVAL;
2894
2895         switch (ioctl) {
2896         case KVM_GET_API_VERSION:
2897                 r = -EINVAL;
2898                 if (arg)
2899                         goto out;
2900                 r = KVM_API_VERSION;
2901                 break;
2902         case KVM_CREATE_VM:
2903                 r = -EINVAL;
2904                 if (arg)
2905                         goto out;
2906                 r = kvm_dev_ioctl_create_vm();
2907                 break;
2908         case KVM_CHECK_EXTENSION: {
2909                 int ext = (long)argp;
2910
2911                 switch (ext) {
2912                 case KVM_CAP_IRQCHIP:
2913                 case KVM_CAP_HLT:
2914                 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2915                 case KVM_CAP_USER_MEMORY:
2916                 case KVM_CAP_SET_TSS_ADDR:
2917                         r = 1;
2918                         break;
2919                 default:
2920                         r = 0;
2921                         break;
2922                 }
2923                 break;
2924         }
2925         case KVM_GET_VCPU_MMAP_SIZE:
2926                 r = -EINVAL;
2927                 if (arg)
2928                         goto out;
2929                 r = 2 * PAGE_SIZE;
2930                 break;
2931         default:
2932                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2933         }
2934 out:
2935         return r;
2936 }
2937
2938 static struct file_operations kvm_chardev_ops = {
2939         .unlocked_ioctl = kvm_dev_ioctl,
2940         .compat_ioctl   = kvm_dev_ioctl,
2941 };
2942
2943 static struct miscdevice kvm_dev = {
2944         KVM_MINOR,
2945         "kvm",
2946         &kvm_chardev_ops,
2947 };
2948
2949 /*
2950  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
2951  * cached on it.
2952  */
2953 static void decache_vcpus_on_cpu(int cpu)
2954 {
2955         struct kvm *vm;
2956         struct kvm_vcpu *vcpu;
2957         int i;
2958
2959         spin_lock(&kvm_lock);
2960         list_for_each_entry(vm, &vm_list, vm_list)
2961                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
2962                         vcpu = vm->vcpus[i];
2963                         if (!vcpu)
2964                                 continue;
2965                         /*
2966                          * If the vcpu is locked, then it is running on some
2967                          * other cpu and therefore it is not cached on the
2968                          * cpu in question.
2969                          *
2970                          * If it's not locked, check the last cpu it executed
2971                          * on.
2972                          */
2973                         if (mutex_trylock(&vcpu->mutex)) {
2974                                 if (vcpu->cpu == cpu) {
2975                                         kvm_x86_ops->vcpu_decache(vcpu);
2976                                         vcpu->cpu = -1;
2977                                 }
2978                                 mutex_unlock(&vcpu->mutex);
2979                         }
2980                 }
2981         spin_unlock(&kvm_lock);
2982 }
2983
2984 static void hardware_enable(void *junk)
2985 {
2986         int cpu = raw_smp_processor_id();
2987
2988         if (cpu_isset(cpu, cpus_hardware_enabled))
2989                 return;
2990         cpu_set(cpu, cpus_hardware_enabled);
2991         kvm_x86_ops->hardware_enable(NULL);
2992 }
2993
2994 static void hardware_disable(void *junk)
2995 {
2996         int cpu = raw_smp_processor_id();
2997
2998         if (!cpu_isset(cpu, cpus_hardware_enabled))
2999                 return;
3000         cpu_clear(cpu, cpus_hardware_enabled);
3001         decache_vcpus_on_cpu(cpu);
3002         kvm_x86_ops->hardware_disable(NULL);
3003 }
3004
3005 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3006                            void *v)
3007 {
3008         int cpu = (long)v;
3009
3010         switch (val) {
3011         case CPU_DYING:
3012         case CPU_DYING_FROZEN:
3013                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3014                        cpu);
3015                 hardware_disable(NULL);
3016                 break;
3017         case CPU_UP_CANCELED:
3018         case CPU_UP_CANCELED_FROZEN:
3019                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
3020                        cpu);
3021                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
3022                 break;
3023         case CPU_ONLINE:
3024         case CPU_ONLINE_FROZEN:
3025                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
3026                        cpu);
3027                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
3028                 break;
3029         }
3030         return NOTIFY_OK;
3031 }
3032
3033 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3034                       void *v)
3035 {
3036         if (val == SYS_RESTART) {
3037                 /*
3038                  * Some (well, at least mine) BIOSes hang on reboot if
3039                  * in vmx root mode.
3040                  */
3041                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
3042                 on_each_cpu(hardware_disable, NULL, 0, 1);
3043         }
3044         return NOTIFY_OK;
3045 }
3046
3047 static struct notifier_block kvm_reboot_notifier = {
3048         .notifier_call = kvm_reboot,
3049         .priority = 0,
3050 };
3051
3052 void kvm_io_bus_init(struct kvm_io_bus *bus)
3053 {
3054         memset(bus, 0, sizeof(*bus));
3055 }
3056
3057 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3058 {
3059         int i;
3060
3061         for (i = 0; i < bus->dev_count; i++) {
3062                 struct kvm_io_device *pos = bus->devs[i];
3063
3064                 kvm_iodevice_destructor(pos);
3065         }
3066 }
3067
3068 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
3069 {
3070         int i;
3071
3072         for (i = 0; i < bus->dev_count; i++) {
3073                 struct kvm_io_device *pos = bus->devs[i];
3074
3075                 if (pos->in_range(pos, addr))
3076                         return pos;
3077         }
3078
3079         return NULL;
3080 }
3081
3082 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
3083 {
3084         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
3085
3086         bus->devs[bus->dev_count++] = dev;
3087 }
3088
3089 static struct notifier_block kvm_cpu_notifier = {
3090         .notifier_call = kvm_cpu_hotplug,
3091         .priority = 20, /* must be > scheduler priority */
3092 };
3093
3094 static u64 stat_get(void *_offset)
3095 {
3096         unsigned offset = (long)_offset;
3097         u64 total = 0;
3098         struct kvm *kvm;
3099         struct kvm_vcpu *vcpu;
3100         int i;
3101
3102         spin_lock(&kvm_lock);
3103         list_for_each_entry(kvm, &vm_list, vm_list)
3104                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
3105                         vcpu = kvm->vcpus[i];
3106                         if (vcpu)
3107                                 total += *(u32 *)((void *)vcpu + offset);
3108                 }
3109         spin_unlock(&kvm_lock);
3110         return total;
3111 }
3112
3113 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
3114
3115 static __init void kvm_init_debug(void)
3116 {
3117         struct kvm_stats_debugfs_item *p;
3118
3119         debugfs_dir = debugfs_create_dir("kvm", NULL);
3120         for (p = debugfs_entries; p->name; ++p)
3121                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
3122                                                 (void *)(long)p->offset,
3123                                                 &stat_fops);
3124 }
3125
3126 static void kvm_exit_debug(void)
3127 {
3128         struct kvm_stats_debugfs_item *p;
3129
3130         for (p = debugfs_entries; p->name; ++p)
3131                 debugfs_remove(p->dentry);
3132         debugfs_remove(debugfs_dir);
3133 }
3134
3135 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
3136 {
3137         hardware_disable(NULL);
3138         return 0;
3139 }
3140
3141 static int kvm_resume(struct sys_device *dev)
3142 {
3143         hardware_enable(NULL);
3144         return 0;
3145 }
3146
3147 static struct sysdev_class kvm_sysdev_class = {
3148         .name = "kvm",
3149         .suspend = kvm_suspend,
3150         .resume = kvm_resume,
3151 };
3152
3153 static struct sys_device kvm_sysdev = {
3154         .id = 0,
3155         .cls = &kvm_sysdev_class,
3156 };
3157
3158 struct page *bad_page;
3159
3160 static inline
3161 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3162 {
3163         return container_of(pn, struct kvm_vcpu, preempt_notifier);
3164 }
3165
3166 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3167 {
3168         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3169
3170         kvm_x86_ops->vcpu_load(vcpu, cpu);
3171 }
3172
3173 static void kvm_sched_out(struct preempt_notifier *pn,
3174                           struct task_struct *next)
3175 {
3176         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3177
3178         kvm_x86_ops->vcpu_put(vcpu);
3179 }
3180
3181 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
3182                   struct module *module)
3183 {
3184         int r;
3185         int cpu;
3186
3187         if (kvm_x86_ops) {
3188                 printk(KERN_ERR "kvm: already loaded the other module\n");
3189                 return -EEXIST;
3190         }
3191
3192         if (!ops->cpu_has_kvm_support()) {
3193                 printk(KERN_ERR "kvm: no hardware support\n");
3194                 return -EOPNOTSUPP;
3195         }
3196         if (ops->disabled_by_bios()) {
3197                 printk(KERN_ERR "kvm: disabled by bios\n");
3198                 return -EOPNOTSUPP;
3199         }
3200
3201         kvm_x86_ops = ops;
3202
3203         r = kvm_x86_ops->hardware_setup();
3204         if (r < 0)
3205                 goto out;
3206
3207         for_each_online_cpu(cpu) {
3208                 smp_call_function_single(cpu,
3209                                 kvm_x86_ops->check_processor_compatibility,
3210                                 &r, 0, 1);
3211                 if (r < 0)
3212                         goto out_free_0;
3213         }
3214
3215         on_each_cpu(hardware_enable, NULL, 0, 1);
3216         r = register_cpu_notifier(&kvm_cpu_notifier);
3217         if (r)
3218                 goto out_free_1;
3219         register_reboot_notifier(&kvm_reboot_notifier);
3220
3221         r = sysdev_class_register(&kvm_sysdev_class);
3222         if (r)
3223                 goto out_free_2;
3224
3225         r = sysdev_register(&kvm_sysdev);
3226         if (r)
3227                 goto out_free_3;
3228
3229         /* A kmem cache lets us meet the alignment requirements of fx_save. */
3230         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
3231                                            __alignof__(struct kvm_vcpu), 0, 0);
3232         if (!kvm_vcpu_cache) {
3233                 r = -ENOMEM;
3234                 goto out_free_4;
3235         }
3236
3237         kvm_chardev_ops.owner = module;
3238
3239         r = misc_register(&kvm_dev);
3240         if (r) {
3241                 printk(KERN_ERR "kvm: misc device register failed\n");
3242                 goto out_free;
3243         }
3244
3245         kvm_preempt_ops.sched_in = kvm_sched_in;
3246         kvm_preempt_ops.sched_out = kvm_sched_out;
3247
3248         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3249
3250         return 0;
3251
3252 out_free:
3253         kmem_cache_destroy(kvm_vcpu_cache);
3254 out_free_4:
3255         sysdev_unregister(&kvm_sysdev);
3256 out_free_3:
3257         sysdev_class_unregister(&kvm_sysdev_class);
3258 out_free_2:
3259         unregister_reboot_notifier(&kvm_reboot_notifier);
3260         unregister_cpu_notifier(&kvm_cpu_notifier);
3261 out_free_1:
3262         on_each_cpu(hardware_disable, NULL, 0, 1);
3263 out_free_0:
3264         kvm_x86_ops->hardware_unsetup();
3265 out:
3266         kvm_x86_ops = NULL;
3267         return r;
3268 }
3269 EXPORT_SYMBOL_GPL(kvm_init_x86);
3270
3271 void kvm_exit_x86(void)
3272 {
3273         misc_deregister(&kvm_dev);
3274         kmem_cache_destroy(kvm_vcpu_cache);
3275         sysdev_unregister(&kvm_sysdev);
3276         sysdev_class_unregister(&kvm_sysdev_class);
3277         unregister_reboot_notifier(&kvm_reboot_notifier);
3278         unregister_cpu_notifier(&kvm_cpu_notifier);
3279         on_each_cpu(hardware_disable, NULL, 0, 1);
3280         kvm_x86_ops->hardware_unsetup();
3281         kvm_x86_ops = NULL;
3282 }
3283 EXPORT_SYMBOL_GPL(kvm_exit_x86);
3284
3285 static __init int kvm_init(void)
3286 {
3287         int r;
3288
3289         r = kvm_mmu_module_init();
3290         if (r)
3291                 goto out4;
3292
3293         kvm_init_debug();
3294
3295         kvm_arch_init();
3296
3297         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
3298
3299         if (bad_page == NULL) {
3300                 r = -ENOMEM;
3301                 goto out;
3302         }
3303
3304         return 0;
3305
3306 out:
3307         kvm_exit_debug();
3308         kvm_mmu_module_exit();
3309 out4:
3310         return r;
3311 }
3312
3313 static __exit void kvm_exit(void)
3314 {
3315         kvm_exit_debug();
3316         __free_page(bad_page);
3317         kvm_mmu_module_exit();
3318 }
3319
3320 module_init(kvm_init)
3321 module_exit(kvm_exit)